vvEPA
United States
Environmental Protection
Agency
Risk Reduction
Engineering Laboratory
Cincinnati OH 45268
EPA/600/2-88/068a
November 1988
Research and Development
Evaluation of
Volumetric Leak
Detection Methods for
Underground Fuel
Storage Tanks
Volume I
-------
EPA/600/2-88/068a
November 1988
EVALUATION OF VOLUMETRIC LEAK DETECTION METHODS FOR
UNDERGROUND FUEL STORAGE TANKS
Volume I
by
Robert D. Roach, James W. Starr and Joseph W. Maresca, Jr.
Vista Research, Inc.
Mountain View, California 94042
Contract No. 68-03-3409
Project Officer
John S. Farlow, Chief
Releases Control Branch
Risk Reduction Engineering Laboratory
Edison, New Jersey 08837
RISK REDUCTION ENGINEERING LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U. S. ENVIRONMENTAL PROTECTION AGENCY
CINCINNATI, OHIO 45268
U.S. Environmental Protection Agency
legion 5, Library (5PL-16)
230 S. Dearborn St-eet, Iro >u IG?0
Chicago, -iL 60S04
-------
Disclaimer
The information in this document, prepared by Vista Research, Inc., has been funded
wholly or in part by the United States Environmental Protection Agency under Contract No.
68-03-3409 to CDM Federal Programs Corporation. It has been subjected to the Agency's peer
and administrative review, and has been approved for publication. Mention of trade names or
commercial products does not constitute an endorsement or recommendation for use.
n
-------
Foreword
Today's rapidly developing and changing technologies and industrial products and
practices frequently carry with them the increased generation of materials that, if improperly
dealt with, can threaten both public health and the environment. The U.S. Environmental
Protection Agency (EPA) is charged by Congress with protecting the Nation's land, air, and
water resources. Under a mandate of national environmental laws, the agency strives to
fonrmlate and implement actions leading to a compatible balance between human activities and
the ability of natural systems to support and nurture life. These laws direct the EPA to perform
research to define our environmental problems, measure the impact, and search for solutions.
The Risk Reduction Engineering Laboratory is responsible for planning, implementing,
and managing research, development, and demonstration programs to provide an authoritative,
defensible basis in support of the policies, programs, and regulations of the EPA with respect to
drinking water, waste water, pesticides, toxic substances, solid and hazardous wastes, and
Superfund-related activities. This publication is one of the products of that research, and
provides a vital communication link between the researcher and the user community.
This document summarizes the results of an evaluation of 25 commercial, quantitative,
in-tank volumetric methods for detecting small leaks in underground storage tank systems
containing petroleum-derived fuels. The evaluation results and the basic information generated
from this program will be useful to both government and industry personnel concerned with this
aspect of preventing groundwater pollution.
For further information, please contact the Superfund Technology Demonstration Division
of the Risk Reduction Engineering Laboratory.
E. Timothy Oppelt, Director
Risk Reduction Engineering Laboratory
111
-------
Preface
This report presents the results of an evaluation by the U.S. Environmental Protection
Agency (EPA) of the performance of 25 volumetric methods for detecting leaks in underground
storage tanks containing motor fuels. The report is divided into two volumes. Volume I
summarizes the evaluation results and the experimental work supporting the results, and
Volume n presents the results of the evaluation performed for each method.
The field testing was done at EPA's Underground Storage Tank Test Apparatus located at
the Risk Reduction Engineering Laboratory (RREL) (formerly the Hazardous Waste Engineering
Research Laboratory) in Edison, New Jersey. With one exception, each method tested was used
to perform a precision or tank tightness test. The performance of each test method is presented
in terms of leak rate, probability of detection, and probability of false alarm. The results suggest
that the performance claim of 0.19 L/h (0.05 gal/h) commonly made by the majority of the
manufacturers is not being reliably met. The user of this report should be careful not to simply
rank the methods according to the quantitative estimate presented, because with simple
modifications, many of the methods can dramatically improve performance. Several methods
have the capability to meet the EPA release detection regulations for tank tightness tests.
Recommendations for improvement are made for each method. After improvements are made, it
is anticipated that many more of the methods evaluated will meet the regulations.
IV
-------
Abstract
This report summarizes the results of the United States Environmental Protection Agency
(EPA) research program to evaluate the current performance of commercially available
volumetric test methods for the detection of small leaks in underground gasoline storage tanks.
The evaluations were performed by the EPA at its Risk Reduction Engineering Laboratory
(RREL) Underground Storage Tank (UST) Test Apparatus located in Edison, New Jersey.
The objectives of the program were to:
o produce the technical data necessary to support the development of release detection
regulations
o define the current practice of commercially available systems
o make specific recommendations to improve the current practice
o provide technical information that will help users select suitable methods for testing
the integrity of underground storage tanks
A unique approach to conducting the evaluation has made it possible to determine and
resolve the technological and engineering issues associated with volumetric leak detection, as
well as to define the current practice of commercially available test methods. The approach uses
experimentally validated models of the important sources of ambient noise that effect volume
changes in nonleaking and leaking tanks, a large database of product-temperature changes that
result from the delivery of product to a tank at a different temperature than the product in the
tank, and a mathematical model of each test method to estimate the performance of that method.
The test-method model includes the instrumentation noise, the configuration of the sensors, the
test protocol, the data analysis algorithms, and the detection criterion. This study and the
ambient noise experiments contributed to a better understanding of the environmental factors that
inhibit detection (temperature, structural deformation, trapped vapor, evaporation and
condensation, and waves). These factors are now not only better understood but also better
quantified. This knowledge is expected to lead to significant improvement in the performance of
current and developing methods.
Twenty-five commercially available volumetric leak detection systems were evaluated.
An estimate of the performance of each method was made in terms of the probability of detection
and probability of false alarm against a 0.38-L/h (0.10-gal/h) leak rate using the detection
threshold employed by each method at the time of the evaluation. Another set of performance
estimates was made for each method in terms of the smallest leak rate that could be detected and
still maintain a probability of detection and a probability of false alarm of 0.95 and 0.05, and
0.99 and 0.01. This performance estimate does not employ the manufacturer's detection
-------
threshold; instead a threshold was selected which yields a probability of false alarm of 0.05 and
0.01, respectively. The leak rate measurable by these systems ranged from 0.26 to 6.97 L/h (0.07
to 1.84 gal/h), with a probability of detection of 95% and probability of false alarm of 5%. Five
of the methods achieved a performance that was better than 0.57 L/h (0.15 gal/h), and a total of
eight methods had a performance that was better than 0.95 L/h (0.25 gal/h). The leak rate
measurable by these systems ranged from 0.47 to 12.95 L/h (0.12 to 3.42 gal/h) when the
probability of detection increased to 99% and the probability of false alarm decreased to 1%.
Only one of the methods achieved a performance better than 0.56 L/h, but five methods achieved
a performance between 0.56 L/h and 0.95 L/h. The measured performance was considerably
poorer than the often claimed 0.19 L/h (0.05 gal/h); this was probably a consequence of two
factors: (1) the requirement that the detectable leak rate be framed in terms of a probability of
detection (the probability that a test will result in the declaration of a leak when the tank is
indeed leaking) and a probability of false alarm (the probability that a test will result in the
declaration of a leak when the tank is tight); and (2) the fact that, in many instances, these
measurements represented the first systematic evaluation of the test method (because, hitherto,
no suitable test apparatus existed).
It is important to realize that, by and large, the performance of the methods evaluated here
was limited by current protocol and practice rather than by hardware. Such limitations can be
overcome by rather modest modifications to testing practice rather than by major system
redesign. The results of tests at the UST Test Apparatus show that attention should be given to
the following aspects of testing practice: waiting period after product delivery or adjustments to
product level (e.g., topping); vapor pocket removal (in tests on overfilled tanks); adequate spatial
sampling of the vertical temperature profiles of the product in the tank; length of tests; data
acquisition, processing, and analysis; and maintaining constant product level. In the foregoing
list of performance limitations, two of the items might require modifications to the measurement
system itself. Modifications of the temperature and product-level measurement systems could
require a more extensive development effort. As part of this study, an estimate was made of the
potential performance that could be achieved by the various volumetric methods evaluated. The
results showed that, with modifications, over 60% of the methods should be able to achieve a
performance of between 0.19 L/h (0.05 gal/h) and 0.56 L/h (0.15 gal/h), and 100% should be
able to achieve a performance of approximately 0.80 L/h (0.20 gal/h), given that a probability of
detection of 99% and a probability of false alarm of 1% is to be achieved.
Leak detection technology is rapidly evolving. With the passage of federal legislation in
1984, the drafting of a proposed regulation by the United States Environmental Protection
Agency in 1987, the development of numerous state and local UST programs, and the initiation
vi
-------
of the Edison evaluation effort, the economic motivation for system improvements and
technological discipline is increasing. Because the performance of individual test methods can
be improved with minor modification, it is likely that most (including some of the
worst-performing) methods will raise their performance to the level of the best-performing
methods in the immediate future. These improvements are already known to be in progress.
Consequently, by the time this report is published, its ranking of test methods will be obsolete,
and the current ranking should not be used by itself for the selection of volumetric leak detection
methods. The state of the art is not sufficiently developed for the ranking to be definitive,
because simple modifications (many already under way) can change performance by a factor of 2
to 10, shifting the rank of a method from the bottom third to the top third. Selection of a method
based simply on its ranking in this report, or worse yet, eliminating a method from consideration
based on its current ranking, would be a mistake.
The current practice of precision leak testing is labor-intensive. The evaluations presented
in this report reflect the performance of systems and crews that represent the best that the
manufacturer has to offer. The performance of these systems can be significantly degraded by
careless or inattentive application of the technology by operators. Complicated underground
storage systems, such as those featuring tilted tanks, manifolded tanks, complicated piping, and
tanks with vapor recovery systems, are commonly encountered in the field. If the operator does
not discover these complications and take them into account the results can be further degraded.
The dynamic technological environment makes the establishment of discipline in the practice
more difficult. In this regard, the lettered appendices to this report (Volume II) should be of
significant help to both the regulators and the users. Each appendix, describing the protocol and
practice of a given method, represents the manufacturer's description of how a test is to be
conducted. Operators should adhere to protocol, or, when deviating, should be sure that the
change is such as to demonstrably improve performance (perhaps by following the specific
suggestions given in the appendices). Increasing the waiting period after adding product to the
tank is a typical example of such a change.
The regulatory and user communities will he besieged by new claims of improved
performance from manufacturers. This report can be of value in estimating the impact of system
improvements, but any new claims of improved performance, to be completely credible, must be
supported, in most cases, by additional experimental and analytic evidence.
This report was submitted by Vista Research, Inc. in partial fulfillment of Contract No.
68-03-3409 with COM Federal Programs Corporation under the sponsorship of the U.S.
Environmental Protection Agency. This report covers a period from September 1986 to July
1987, and work was completed as of June 1988.
vii
-------
Authors and Contributors
A multi-company project team has contributed to this Work Assignment. The Vista
Research team includes Work Assignment Manager Dr. Joseph W. Maresca, Jr. (Vice President),
Analysis Director Mr. Robert D. Roach (Senior Software Engineer), and Experiment Director
Mr. James W. Starr (Manager of Edison Operations). Messrs. Roach and Starr and Dr. Maresca
are the authors of this report, and they provided the technical leadership throughout the program.
Mr. David F. Doyle, Senior Vice President, Camp Dresser and McKee, Inc. (CDM) and Dr.
Myron S. Rosenberg (CDM) were the Project and Deputy Project Directors, respectively, for
EPA's prime contractor.
The following individuals and organizations contributed to the project:
Thomas J. Bahr, Research Analyst, Vista Research, Inc.
Kyle D. Barnes, Research Analyst, Vista Research, Inc.
Dr. Alan A. Bums, Staff Scientist, Vista Research, Inc.
James Z. Butkowski, Senior Technician, Vista Research, Inc.
Art G. Caceras, Research Analyst, Vista Research, Inc.
Joseph A. Cervantes, Research Engineer, Vista Research, Inc.
Isaac A. Chang, Research Analyst, Vista Research, Inc.
Noel L. Chang, Jr., Research Engineer, Vista Research, Inc.
Robert B. Cooper, Senior Technician, Vista Research, Inc.
Joseph Gibney, Research Analyst, Camp Dresser and McKee, Inc.
Dr. Alan K. Koide, Research Engineer, Vista Research, Inc.
Cynthia J. Komorowski, Senior Secretary, Vista Research, Inc.
Maria P. Mac Arthur, Research Engineer, Vista Research, Inc.
Joseph M. Makwinski, Engineer, Camp Dresser and McKee, Inc.
John Mazza, Engineer, CDM Federal Programs Corporation
Daniel Naar, Research Engineer, Vista Research, Inc.
Tuyet-Hahn Nguyen, Research Analyst, Vista Research, Inc.
Scott Reed, Research Analyst, Vista Research, Inc.
Angela Regalia, Research Analyst, Vista Research, Inc.
Monique Seibel, Technical Writer and Illustrator, Vista Research, Inc.
Robert M. Smedfjeld, Research Engineer, Vista Research, Inc.
Mark Thomas, Research Analyst, Camp Dresser and McKee, Inc.
William Volchok, Research Engineer, Vista Research, Inc.
Dr. Christopher P. Wilson, Director of Research, Vista Research, Inc.
Morton N. Winter, Research Engineer, Vista Research, Inc.
Richard F. Wise, Research Engineer, Vista Research, Inc.
Lisa A. Yurfest, Word Processor Operator, Vista Research, Inc.
Vlll
-------
Table of Contents
Volume I
Disclaimer u
Foreword m
Preface iv
Abstract v
Authors and Contributors viii
List of Figures xii
List of Tables xiv
Acknowledgments xvi
List of Abbreviations xix
List of Equivalent Units xxi
1 Introduction 1-1
1.1 Objectives 1-1
1.2 Regulatory Needs 1-1
1.3 Volumetric Test Methods 1-2
1.4 Definition of Test Method Performance 1-2
1.5 Evaluation Approach 1-3
1.6 Underground Storage Tank Test Apparatus 1-6
1.7 Industry Participation 1-6
1.8 How to Use This Report 1-10
2 Conclusions 2-1
3 Recommendations 3-1
4 Overview of Volumetric Precision Tank Testing 4-1
4.1 Product-Level Fluctuations in an Underground Storage Tank 4-2
4.2 Flow Rate Measurements during a Tank Test 4-5
4.3 Classification of Volumetric Test Methods 4-6
4.4 Major Features of a Tank Test 4-6
4.5 Time Required to Complete a Test 4-12
4.6 Test Accuracy 4-12
4.7 Groundwater Level Affects Magnitude of the Leak Rate 4-15
5 Performance of a Test Method 5-1
5.1 Signal 5-3
5.2 Noise Histogram 5-4
5.3 Signal-Plus-Noise Histogram 5-9
5.4 Performance Model 5-12
5.5 Multiple Test Strategy 5-15
5.6 Summary 5-17
6 Performance Simulation 6-1
6.1 Experimental Data 6-2
6.2 Database Management 6-3
6.3 Temperature Database 6-4
6.4 Product-Level Database 6-7
ix
-------
6.5 Test Method Models 6-26
6.6 Tank-Test Simulator 6-27
6.7 Performance Analysis 6-27
7 UST Test Apparatus and Quality Control of Measurements 7-1
7.1 Test Apparatus 7-1
7.2 Data Quality Objectives 7-4
7.3 Development of Data Quality Specifications 7-6
7.4 Instrumentation 7-11
7.5 Underground Storage Tanks 7-21
8 Ambient Noise Experiments 8-1
8.1 Product-Temperature Changes 8-1
8.2 Trapped Vapor 8-10
8.3 Structural Defomiation 8-22
8.4 Evaporation and Condensation 8-27
8.5 Surface Waves 8-28
8.6 Internal Waves 8-28
8.7 Summary of Results from Ambient Noise Experiments 8-29
9 Evaluation Protocol 9-1
9.1 Test Method Description 9-3
9.2 Field Verification Tests 9-4
9.3 Test Method Model Validation 9-5
9.4 Evaluation Results 9-11
10 Performance of Canonical Test Methods 10-1
10.1 Test Method Model Description 10-1
10.2 Thermal Expansion and Contraction of the Product 10-3
10.3 Trapped Vapor 10-8
10.4 Structural Deformation 10-9
11 Performance of Commercial Test Methods 11-1
11.1 How to Interpret Performance Results 11-2
1 1.2 Performance Results 11-4
11.3 Instrumentation 11-18
1 1.4 Test Operations 11-18
1 1.5 Interpretation of the Evaluation Results 11-20
11.6 Expected Performance after Implementing Improvements 11-21
11.7 Selection of a Test Method 11-21
12 Performance Improvements 12-1
12.1 Instrumentation 12-1
12.2 Protocol 12-2
12.3 Data Analysis 12-9
12.4 Detection Criterion 12-10
12.5 Operator Influence 12-10
12.6 Multiple-Test Strategy 12-11
13 References 13-1
Appendix: Descriptions of the 25 Volumetric Methods Evaluated in the EPA Study
and Suggestions for Improving the Performance of Each A-l
-------
Volume II
Appendix A:
Appendix B:
Appendix C:
Appendix D:
Appendix E:
Appendix F:
Appendix G:
Appendix H:
Appendix I:
Appendix J:
Appendix K:
Appendix L:
Appendix M:
Appendix N:
Appendix O:
Appendix P:
Appendix Q:
Appendix R:
Appendix S:
Appendix T:
Appendix U:
Appendix V:
Appendix W:
Appendix X:
Appendix Y:
Evaluation of the
Evaluation of the
Evaluation of the
(AUTAMAT)
AES/Brockman Leak Detecting System A-l
Ainlay Tank Tegrity Tester B-l
Exxon Automatic Tank Monitor and Tester
C-l
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Evaluation of the
Computerized VPLT Tank Leak Testing System D-l
DWY Leak Sensor E-1
Horner EZY CHEK Leak Detection System F-1
Gasoline Tank Monitor (GTM) System G-l
GiJharco Tank Monitor System H-l
Inductive Leak Detector 3100 System 1-1
EASIINSTA-TEST Leak Detection System J-l
Tank Audit Leak Computer K-l
Leak-O-Meter Leak Detection System L-l
LiquidManager Leak Detection System M-l
Pneumercator LMS-750 Leak Detection System N-l
MCG-1100 Leak Detection System O-l
Mooney Leak Detection System P-l
OTEC Leak Sensor Q-l
PACE Leak Tester System R-l
Heath Petro The Leak Detection System S-l
TankTech Portable Small Leak Detector (PSLD) T-l
S.M.A.R.T Leak Detection System U-l
Tank Auditor Leak Detection System V-l
Pandel Tank Monitoring Device (TMD-1) W-l
Tank Sentry II Leak Detection System X-l
Veeder-Root TLS-250 Tank Level Sensing System.... Y-l
XI
-------
List of Figures
4.1 Cross-sectional view of an underground storage tank 4-1
4.2 Overview of volumetric tank test procedure 4-7
4.3 UST geometry showing hydrostatic pressure relationships 4-17
5.1 Typical measurement output for a sequence of tank tests 5-3
5.2 Normal probability model overlaid on noise histogram for a test method with a
zero mean 5-5
5.3 The probability of false alarm 5-6
5.4 Noise histogram with a positive bias 5-7
5.5 Illustration of the histogram of the bias produced by deformation-induced vol-
ume changes 5-7
5.6 TVR and VR histograms 5-10
5.7 Comparison of the signal-plus-noise histogram for a -1.2-L/h leak rate and the
noise histogram under the same product temperature conditions after temperature
compensation 5-11
5.8 Comparison of the signal-plus-noise histogram for a -1.2-L/h leak rate and the
noise histogram under the same product temperature conditions before tempera-
ture compensation 5-11
5.9 Statistical model to estimate the accuracy of constant-head leak detection system 5-12
5.10 Statistical model to estimate the accuracy of a variable-head test 5-14
6.1 Flowchart of the performance simulation 6-2
6.2 Histogram of temperature volume changes over a 1-h period 12 h after product
was added to the tank 6-6
6.3 Histogram of the temperature changes in the circulated product temperature
database 6-6
6.4 Comparison of the Fill-Tube Dynamics Model predictions 6-22
6.5 Examples of performance curves for a generic leak detection method 6-30
7.1 UST Tank Test Apparatus 7-2
7.2 Nominal resistance-versus-temperature curve for a 30,000-£1 thermistor 7-17
7.3 Experimental and theoretical height-to-volume conversion curves near the top of
the steel tank 7-22
8.1 Temperature volume time series 8-3
8.2 Differences in residual fluctuations in mis temperature volume between Arrays
2 and 3 8-4
8.3 Coherence between two thermistors (18 and 19), located near the middle of the
tank on Array 2 of the Test Apparatus 8-6
8.4 Residual fluctuations in volume after compensating for thermally induced
product-volume changes 8-7
xu
-------
8.5 Effects of topping the tank with colder product, as represented by 11 thermistors
located on an array at the fill tube of the Test Apparatus's Fiberglass Tank 8-8
8.6 Effects of topping the tank with warmer product, as represented by 11 thermis-
tors located on an array at the fill tube of the Test Apparatus's Fiberglass Tank ... 8-9
8.7 Experimental configuration for trapped vapor tests in a well-bled fiberglass tank 8-11
8.8 Experimental configuration for the trapped vapor tests in the fiberglass tank 8-11
8.9 Experimental apparatus to estimate n 8-14
8.10 Time series of the product-level changes in the fill tube of an overfilled fiber-
glass tank on 6 June 1987, produced by inserting and removing four different-
size bars (625. 953, 1551, and 2477 ml, respectively) 8-16
8.11 AcM computed from the volume of the cylindrical bar, Vbar and the measured
product-level change data, Ah, obtained on 6 June 1987 8-17
8.12 Time series of the product-level changes in the sleeve of an underfilled fiberglass
tank on 5 June 1987; product-level changes were produced by inserting four
different-size bars (625, 1551, 2477, 5071 ml, respectively) 8-18
8.13 Acff computed from the volume of the cylindrical bar, Vbar, and the measured
product-level change data, Ah, obtained on 5 June 1987 8-19
8.14 Estimate of Aisd from data collected on 28-30 May and 6 June 1987 8-20
8.15 Comparison of the measured and predicted volume time series for data collected
on 3 May 1988 in the fill tube of the steel tank at the UST Test Apparatus 8-23
9.1 Flowchart of the evaluation protocol 9-1
10.1 The histogram (A) and cumulative distribution function (B) for the five-
thermistor method with a test duration of 1 h 10-5
10.2 Histograms of the noise compiled from overfilled, constant-head tank tests that
compensate for thermal expansion and contraction of the product using (A) one
temperature sensor located at the midpoint of the tank, (B) five equally spaced
temperature sensors that are weighted volumetrically, and (C) eleven equally
spaced temperature sensors that are weighted volumetrically 10-6
10.3 Histograms of the noise generated from overfilled, variable-head tank tests
conducted immediately after topping the tank (1-m product-level addition) for
(A) Tc = 0.75 h, K = 120 cm2 and (B) Tc = 3 h, K =120 cm2 10-11
10.4 Histograms of the noise generated from overfilled, variable-head tank tests
conducted 3 time constants after topping the tank (1-m product-level addition)
for (A) Tc = 0.75 h, K = 120 cm2 and (B) Tc = 3 h, K =120 cm2 10-13
10.5 Histograms of the noise in overfilled, variable-head tank tests produced by a
uniform distribution of (A) K (30, 60, 90, and 120 cm2) with Tc= 0.75 h and (B)
Tc(0.5, 1,2, and 3 h) with K= 12()cm2 10-14
xui
-------
List of Tables
1.1 List of Participants in the EPA Volumetric Test Method Evaluation Program 1-8
1.2 Commonly Used Quantities in English Units and SI Equivalents 1-12
5.1 Possible Outcomes of a Leak Detection Test 5-2
5.2 Multiple-Test Performance Calculations for Two Tests 5-17
6.1 Conditions Used to Create Temperature Database for the Steel Tank 6-5
6.2 Nominal Run Conditions in the Circulated Product Temperature Database 6-7
6.3 Gas Phase Densities 6-14
6.4 Estimates of How Tf1l Changes with Aell and K 6-19
7.1 Specifications of the Apparatus Instrumentation 7-6
7.2 Precision of Each Sensor Required to Satisfy Criterion A in a 30,000-L
(8,000-GaDTank 7-7
7.3 Height Sensor Precision Required to Satisfy Criterion B in a 30,000-L Tank
Given That Precision of the Temperature Sensor is 0.001°C 7-8
7.4 Uncertainty of the Temperature-Compensated Volume Rate Estimated Using the
Temperature and Height Precision Specifications for a 1-h Measurement Period
and a Rate of 1 Sample/Min 7-8
7.5 Estimate of the Volume Surrounding Thermistors in a 30,000-L Steel Tank 2.43
m in Diameter 7-15
8.1 Thermal Volume Changes Estimated from a 1-h Block of Data as a Function of 8-2
Time After Delivery
8.2 Average Slope Differences for Three Thermistor Arrays 8-5
8.3 Results of Laboratory Experiments to Estimate n (Bottle Volume = 803 ml) 8-14
8.4 Summary of the Results of the 6 June 1988 Overfilled Fiberglass Tank Test
(Product Depth 335.3 cm) 8-15
8.5 Summary of the Trapped Vapor Tests Conducted in the Fiberglass Tank on
5 June 1988 8-18
8.6 Summary of the Experimental Data Used to Estimte the Magnitude of the
Instantaneous Deformation, Alsd 8-20
8.7 Error in Estimating Product Volume from Product-Level Meausrements If the
Measured Height-to-Volume Calibration Factor, A,,,,Is Not Used 8-21
8.8 Estimates of Acll, T, Tcll, and K made from the Product-level Measurements in
the 10-cm Diameter Fill Tube of the Steel and Fiberglass Tanks at the UST Test
Apparatus 8-22
8.9 Effects of the Fill Tube Diameter on the Model Predictions 8-25
8.10 Effects of Tank Elasticity on the Model Predictions 8-25
8.11 Time Constant of the Volume Time Series 8-26
XJV
-------
10.1 Temperature Compensation Schemes 10-3
10.2 Standard Deviations of Temperature-Compensated Volume Rate Histogram 10-4
10.3 Smallest Detectable Leak Rates with, a PD = 0.99 and PFA = 0.01 10-4
10.4 Standard Deviations for One-Thermistor Case at Different Heights 10-7
10.5 Smallest Detectable Leak Rates for One-Thennistor Case at Different Heights
(Pn = 0.99 and P,.A = 0.01) 10-7
10.6 Standard Deviations for Nonleaking Tank Immediately after Topping
(T = 0.75 h,Aetl = 82.0 cm2) 10-11
10.7 Standard Deviations for Nonleaking Tank Immediately after Topping
(K= 12()cm2, Acll = 82.0 cm2) 10-12
10.S Standard Deviations for Nonleaking Tank 3 Time Constants after Topping
(A,,, = 82.0 cm2) 10-12
II.I Performance of Overfilled-Tank Test Methods at Variable Head 11-6
11.2 Factors That Affect the Performance of Overfilled Tank Tests Conducted at
Variable Head 11-8
11.3 Performance of Overfilled-Tank Test Methods at Constant Head 11-12
11.4 Factors That Affect the Performance of Overfilled-Tank Tests Conducted at
Constant Head 11-13
11.5 Performance of Partially-Filled-Tank Test Methods 11-14
11.6 Factors That Affect the Performance of Partially-Filled-Tank Tests 11-16
11.7 Estimates of Test Method Performance In Terms of PD and PFA for the Detection
of a Leak Rate of 0.38 L/h (0.1 gal/h) Using the Manufacturer's Detection
Threshold 11-17
11.8 Estimate of Test Method Performance in Terms of Leak Rate for Two Different
Sets of Pn and Pr.^ Using the Detection Threshold Established by the PFA 11-17
11.M Estimate of the Performance of Volumetric Test Methods Evaluated at the UST
Test Apparatus After Two Levels of Modifications; Expressed in Terms of the
Smallest Leak Rate That Can Be Detected with a PD =0.99 and a PFA = 0.01 11-21
XV
-------
Acknowledgments
This report was prepared for the U. S. Environmental Protection Agency (EPA) by Vista
Research, Inc. under subcontract to CDM Federal Programs Corporation (CDM/FPC). Vista
Research has performed this work under subcontract to and in conjunction with technical staff
provided by CDM Federal Programs Corporation from February 1987 to the present, and by
Enviresponse, Inc. prior to February 1987. The original program plan, developed by Vista
Research, was completed in December 1984 under subcontract to IT Corporation. This work
was performed under the auspices of the Risk Reduction Engineering Laboratory (RREL) of the
Office of Research and Development (ORD) and of the Office of Underground Storage Tanks
(OUST). Dr. Thomas R. Hauser, Director of RREL during this program, and Mr. Ronald Brand,
Director of OUST, have actively supported this program.
Many individuals must be acknowledged for their energetic support and technical and
administrative efforts in completing this complex project. This work would not have been
possible without the sound technical guidance and technical review, enthusiastic encouragement,
and vigorous support provided by Mr. John S. Farlow, Technical Project Monitor, and Chief of
the Releases Control Branch (RGB), Superfund Technology Demonstration Division (STDD),
RREL, and his staff. Mr. Robert Hillger contributed directly to this project in a variety of
capacities. Mr. Anthony Tafuri served as Contract Project Officer. Dr. John E. Brugger,
Chi-Yuan Fan, Richard Field, and Richard Griffiths also supported and reviewed this work.
Mr. Ira Wilder, prior Chief of the RGB, and Mr. Ronald Hill, Director of the STDD, are
acknowledged for their program support.
Mr. Ronald Brand and his technical staff at the Office of Underground Storage Tanks have
been active participants and reviewers of the work. The efforts of David O'Brien, Michael
Kalinoski, Steven Nacht, Steven Way, and Thomas Young are appreciated. Louise Wise and
James McCormick have provided program direction and support during this work. In particular,
the contribution made by Mr. Young must be singled out. He has been involved in an ongoing
program of technical review, and has provided invaluable suggestions to focus the work, to
improve it technically, and to better present it.
The Office of Environmental Engineering and Technology Demonstration of ORD has
made valuable contributions to this effort. The contributions that have been made by Dr. John
XVI
-------
Skinner, Director, and his staff, including Jay Benforado, David Berg, Alfred W. Lindsey,
Charles Mitchell, and Burnell Vincent, are acknowledged. In particular, the efforts of Mr. Berg
have been extremely important, and his contributions are gratefully acknowledged.
At the inception of this project a peer review of the proposed evaluation approach was
conducted by the EPA and an external panel consisting of William E. 8 lain, Assistant Engineer,
New York State Department of Environmental Conservation; Gerald L. Anderson, President,
American Gas and Chemical Company, Ltd.; and Dr. Robert W. Coughlin, Professor and Head
of the Department of Chemical Engineering, University of Connecticut. Their helpful comments
are acknowledged.
The American Petroleum Institute (API), the Leak Detection Technology Association, and
the Petroleum Equipment Institute, and their member organizations are gratefully acknowledged
for their review and recommendations, particularly during the planning stage prior to the conduct
of the evaluations. Many of their recommendations with regard to the Test Apparatus were
implemented. In particular, Dr. Rudy White and Marc J. Meteyer of the API are acknowledged
for their support and encouragement.
The authors and contributors to this report would like to acknowledge the project
management and the administrative and technical support provided by CDM Federal Programs
Corporation and Camp Dresser and McKee, Inc. (CDM). The development and experimental
validation of the performance simulation, the conduct of the evaluations of the 25 commercial
test methods, and the preparation of the final report were completed as part of the CDM/FPC
contract. The Project Office on the contract was led by the Project Director, Mr. David F. Dolye,
Senior Vice President of CDM; Dr. Myron Rosenberg, the Deputy Project Director, and Julia M.
Nault of CDM, and Mr. William M. Kaschak of CDM/FPC actively supported the project team.
The project team wishes to thank these individuals for their support throughout the conduct of
this multi-company project. The support and encouragement of senior management,
Mr. Thomas F. Cheyer, Senior Vice President of CDM, and Mr. William H. Parker HI, Senior
Vice President of CDM/FPC, are gratefully acknowledged. A special acknowledgment is made
to Rosemary J. Ellersick of CDM/FPC and David Johnson, a member of her staff, for Quality
Control and Quality Assurance support.
The authors and contributors would also like to acknowledge the contributions made by the
management and technical staff of Enviresponse, Inc. The preliminary experiments conducted to
design the UST Test Apparatus, the final design and construction of the UST Test Apparatus, the
development and verification of the equipment and instrumentation, and the early work on the
xvu
-------
performance simulation were completed before February 1987. In particular, the contributions
made by Mitchell D. Garber, Dr. A. C. Gangadharan, Stephen Sawyer, and Paul Faccone are
acknowledged.
The preliminary design for the UST Test Apparatus was performed by IT Corporation.
Messrs. John A. Broscious and Shahzad Niaki are acknowledged for thek contributions during
the formative stages of the project.
Dr. Harold Guthart, President of Vista Research, has been senior technical advisor for the
program and Project Supervisor for all of the Vista Research work. The successful completion
of this project would not have been possible without his support, encouragement, and technical
input.
Preparation of this report represented a significant effort. Monique Seibel led the
publication team, contributed to the technical writing of the report, and edited, illustrated, and
organized it. Lisa A. Yurfest and Cynthia Komorowski did the word processing for the report.
They were supported by Vivian A. Solis and George Koukoutsakis. Thek efforts are sincerely
appreciated.
The last and most important acknowledgment is extended to the participants in the
evaluation program. The management, technical staff, and test crews who participated in this
voluntary program were highly professional and cooperative throughout. Without thek support,
committment, and cooperation, this program would not have been possible. Thank you.
Dr. Joseph W. Maresca, Jr.
Vice President
Vista Research, Inc.
XVlil
-------
List of Abbreviations
A = Cross-sectional area of the product surface
Aeff = Effective cross-sectional area of the product surface
ALR = Cross-sectional area of a hole in a tank
atm = Atmospheres
C = Volume rate
°C = Degrees Celsius
cm = Centimeters
ft = Feet
°F = Degrees Fahrenheit
g = Grams
gal = Gallons
h = Hours
Hz = Hertz
in. = Inches
k = P = Fraction of true volume change
K = Elasticity constant
°K = Degrees Kelvin
kg = Kilograms
kHz = Kilohertz
L = Liters
Ib = Pounds
LR = Leak rate (flow rate)
m = Meters
mb = Millibars
mg = Milligrams
MHz = Megahertz
ml = Milliliters
mm = Millimeters
ms = Milliseconds
(O.A = Microamperes
Jim = Microns (micrometers)
(J.S = Microseconds
Q = Ohms
PD = Probability of detection
xix
-------
pdf = Probability density function
PFA = Probability of false alarm
rrns = Root mean square
s = Seconds
t = time
t0 = start time
T = Temperature
Tc = Time constant
Teff = Effective time constant
Th = Threshold
TV = Temperature volume
API = American Petroleum Institute
ASTM = American Society of Testing Materials
ATGS = Automatic tank gauging system
EPA = Environmental Protection Agency
HWERL = Hazardous Waste Engineering Research Laboratory
LDTA = Leak Detection Technology Association
NBS = National Bureau of Standards
NFPA = National Fire Protection Association
PEI = Petroleum Equipment Institute
RREL = Risk Reduction Engineering Laboratory
UST = Underground storage tank(s)
XX
-------
List of Equivalent Units
1 millimeter = 0.04 inches 1 inch = 25.40 millimeters
1 centimeter = 0.39 inches 1 inch = 2.54 centimeters
1 centimeter = 0.033 feet 1 foot = 30.48 centimeters
1 meter = 3.28 feet 1 foot = 0.30 meters
1 gram = 0.035 ounces 1 ounce = 28.35 grams
1 kilogram = 2.21 pounds 1 pound = 0.45 kilograms
1 liter = 0.26 gallons (U.S.) 1 fluid ounce = 29.57 milliliters
1 milliliters = 0.03 fluid ounces 1 gallon (U.S.) = 3.79 liters
(degrees Celsius) 9/5 + 32 = degrees Fahrenheit
(degrees Fahrenheit -32) x 5/9 = degrees Centigrade
XXI
-------
1 Introduction
This report summarizes the results of the United States Environmental Protection Agency
(EPA) research program to evaluate the current performance of commercially available
volumetric test methods that attempt to detect small leaks in underground gasoline storage tanks
(UST).
1.1 Objectives
The specific objectives of the program were to produce the technical data necessary to
support the development of release detection regulations; to define the current practice of
commercially available systems; to make specific recommendations to improve the current
practice; and to provide technical information that will help users select suitable methods for
testing the integrity of underground storage tanks.
1.2 Regulatory Needs
Leaking underground storage tank systems represent a serious environmental threat.
Estimates of the fraction of UST systems that are leaking range from 10 to 25% [1, 2]. Records
from past release incidents indicate that, without the use of release detection, a release can
become substantial before it is detected [3]. The 1984 Hazardous and Solid Waste Amendments
to the Resource Conservation and Recovery Act of 1976 have charged the EPA with developing
regulations for the detection of releases from UST. The final version of the technical regulation
was published in the Federal Register on 23 September 1988 [4]. The performance standard for
volumetric testing requires that a method must be capable of detecting a leak rate of 0.38 L/h
(0.10 gal/h) with a probability of detection of 0.95 and a probability of false alarm of 0.05.
Development of technically sound and defensible regulations requires that both the threat
to the environment and the technological limits of release detection be known. The threat to the
environment is extremely difficult to define because the transport, fate, and amount of petroleum
that is hazardous to the environment are site-specific.
A performance standard that is based on the current technology will minimize the
uncontrolled release of petroleum product. Unfortunately, the data required to formulate a
realistic regulatory policy were incomplete or nonexistent before this study was undertaken.
While many leak detection methods are available and can be used to detect small releases, the
performance of these methods was unknown. Almost all of the volumetric test methods claim to
meet the 190-ml/h (0.05-gal/h) practice recommended in the National Fire Protection Agency
Pamphlet NFPA 329 [5], but very little evidence, theoretical or experimental, had been provided
1-1
-------
by the manufacturers to support these claims. The limited evidence available prior to the Edison
evaluations [2, 6-8] suggested that the methods are not reliably meeting these claims, a fact that
has been confirmed by this study. However, the study has also shown that, with relatively minor
changes, many leak detection methods can meet the new regulatory requirements.
1.3 Volumetric Test Methods
Many types of commercial systems are available to detect leaks in underground storage
tanks. In 1986, the EPA published a survey of available methods [1], grouping them into four
categories: volumetric, nonvolumetric, inventory control, and leak-effects monitoring. For all
the methods in the first three categories, tests are conducted inside the tank, while leak-effects
monitoring is performed outside the tank.
Volumetric test methods were the first to be selected for evaluation because (1) these
methods have the potential for detecting small leaks, (2) the measured quantity can be directly
related to leak rate, (3) the main sources of testing errors were believed to be well known, even
though the empirical data necessary to quantify and correct many of these errors did not exist,
and (4) the technology was commercially available and widely accepted in industry. Volumetric
methods that claim to meet the NFPA practice of 0.19 L/h (0.05 gal/h) are commonly known as
"precision tests," "tank tightness tests," or "tank integrity tests." Some methods test in a partially
filled tank, and others test in a tank overfilled into the fill tube or an above-grade standpipe,
A volumetric method measures the change in product volume that results from a leak in the
tank; a leak can represent either the release of product from the tank or an inflow of groundwater
into the tank. Most methods measure product level and product temperature. The product-level
data are converted to product volumes, which are then temperature-compensated. Next, flow
rate is calculated using one or more different analysis schemes. The flow rate is then compared
to a threshold flow rate to determine whether the tank is leaking. If the flow rate exceeds the
threshold, the tank is declared leaking. If not, the tank is declared tight. While the details of the
actual instrumentation, measurement protocols, data reduction and analysis algorithms, and
detection criteria differ from method to method, the testing approach is essentially the same for
all methods.
1.4 Definition of Test Method Performance
The confident detection of very small releases represents a considerable technical
challenge. Release detection is, by its nature, a statistical process. The uncertainty in release
detection is a consequence of environmental factors, operational practice, and
instrumentation precision and accuracy. Testing errors are manifested in one of two ways:
1-2
-------
missed detections (leaking tanks declared to be tight) that result in the undiscovered release of
product into the ground, or false alarms (tight tanks declared to be leaking) that lead to additional
testing and may result in the needless and considerable expense of tank repair investigations
and/or replacement.
A complete specification of system performance requires a description of the probability of
detection (PD) and probability of false alarm (PFA) at a defined leak rate and also requires an
estimate of the uncertainty of the PD and PFA. If, in addition, a frequency of testing is specified,
then the limits of the threat to the environment, the confidence with which these limits can be
met, and the costs associated with mistakes in testing can be defined.
The performance of each method evaluated is expressed as a leak rate at the product level
at which the test was conducted. Some consideration was given to normalizing all results to the
same hydrostatic head relative to a leak; this is impractical, because the hydrostatic pressure
depends on the level of the product in the tank, the level of the groundwater outside the tank, and
the unknown location(s) of the leak(s) in the tank. There is a set of conditions for each test
method in which the hydrostatic pressure produced by the product in the tank and the
groundwater outside the tank will prevent flow into or out of a tank, even if a hole is present.
Proper interpretation of the test result is the responsibility of the test operator.
1.5 Evaluation Approach
The approach was designed to satisfy all four of the objectives listed above (Section 1.1).
1.5.1 Data Quality Objectives
To address the program objectives, a set of data quality objectives was established at the
beginning of the program and was adhered to throughout the data collection. The data quality
objectives were developed to evaluate the methods claiming to meet the 0.19-L/h (0.05-gal/h)
practice recommended by the NFPA. The precision and accuracy of the product-level,
product-temperature, and leak rate data collected at the UST Test Apparatus were sufficient to
evaluate the performance of each test method at a leak rate of 0.19 L/h with the probability of
detection of 0.99 and the probability of false alann of 0.01 called for in the proposed UST
regulations [3]. This level of performance requires that the one-standard-deviation uncertainty in
the histogram of the volume rate results compiled from many tests of one or more nonleaking
tanks be 0.04 L/h (0.01 gal/h) or less. The UST Test Apparatus instrumentation, the calibration
procedures, and the data quality analyses after each test were designed to verify that the data
were meeting these objectives (see Section 7).
1-3
-------
1.5.2 Evaluation vs. Validation
An important distinction is made between evaluation and validation. The EPA evaluation
program was not meant to validate the performance of a test method. Rather, it was intended to
estimate the performance of a given system under the tank conditions selected for the evaluation.
Ideally, the performance of each method submitted for evaluation has been validated by the
manufacturer over the range of testing conditions for the method. Because the EPA recognized
that most manufacturers participating in the program had not systematically evaluated or
validated their methods, the test conditions selected for the evaluation were fairly
comprehensive. The evaluation reported here was designed to verify the manufacturer's
performance claim over a limited set of test conditions. A thorough discussion of the test
conditions is presented in this report in order to help the reader interpret the applicability of the
results.
1.5.3 Approach
The performance of a leak detection system was determined from the histograms of the
noise (developed from the volume-rate fluctuations in a nonleaking tank compiled for all
conditions under which a test had been conducted), and of the signal-plus-noise (developed from
the relationship between leak rate and these volume-rate fluctuations). If the evaluation had
included only a few test methods, each manufacturer could have been requested to perform a
standard tank test for each ambient condition in the test matrix, and a histogram could have been
generated from all of the volume rates measured. However, because both the test matrix and the
number of methods to be tested were large, this approach would have been too time-consuming
and too costly to implement. In addition, this direct approach would not have provided any
useful information, either to assess the limits of the technology in general, or to improve the
performance of a given method. Instead, a unique approach, which also provides this
information, was developed to perform the evaluation; this latter approach takes advantage of the
common methodology of the majority of the volumetric test methods. This approach was first
formulated in [9] and is summarized in [10]. A detailed description of the evaluation protocol is
given in [11].
A three-step procedure was used to conduct the evaluations. The first step was to develop
and experimentally confirm models of the important sources of noise that control the
performance of each test method. If the total noise field is accurately modeled, the sum of the
volume contributions from each noise source will be equal to the product-level changes in a
nonleaking tank. As part of the modeling effort, a large database, reflecting the different product
1-4
-------
temperature conditions which could be experienced during field testing, was obtained to simulate
a test performed after a delivery of approximately 15,000 L (4,000 gal) of product at one
temperature to a 30,000-L storage tank half-filled with product at another temperature.
The second step was to develop and validate, for each leak detection method, a model that
mathematically described it. The test-method model includes the precision and accuracy of the
instruments, the test protocol, the data collection, analysis and compensation algorithms, and the
detection criterion. The model, in turn, was validated in two steps. First, each manufacturer was
required to review the model for accuracy, make corrections to the model as necessary, and
finally to concur that it accurately represented the method submitted for evaluation; and second,
the manufacturer was required to participate in the Field Verification Tests, a three-day program
of tank-test and calibration experiments at the UST Test Apparatus. The manufacturer used his
own testing crews and test equipment for the three days of testing. Methods that were not
operational at the time of the tests, or that were different from those with which their respective
manufacturers had concurred, were not evaluated.
Finally, a performance estimate for each method was made using the test-method model
approved by the manufacturer, the product-level measurements estimated from the noise models,
and the temperature database. The performance of all test methods but one was evaluated
against the same database of temperature conditions encompassing over 500 h of data. A special
database of over 180 h was developed to evaluate the one method that continuously circulated
the product in the tank during the test. Operational effects and deviations from the prescribed
protocols during the Field Verification Tests were also examined and discussed.
1.5.4 Presentation of Results
The performance results are presented and discussed in four categories: underfilled-tank
tests, overfilled-tank tests conducted at a nearly constant product level, overfilled-tank tests
conducted at a variable product level, and tests for which no performance estimate was or could
be made. The performance estimate was made for a 2.43-m (8-ft)-diameter, 30,000-L
(8,000-gal) tank containing unleaded gasoline and assumes that the test procedure was followed
precisely as specified by the manufacturer.
The performance of each method was obtained by combining several different
calculations. First, a performance estimate was made for a single tank without any trapped
vapor; it included the instrumentation noise, a wide range of temperature conditions, and one set
of tank deformation characteristics. The performance of each method is presented in terms of the
PDand PFAfor leak rates between 190 ml/h (0.05 gal/h) and 5,000 ml/h (1.3 gal/h).
1-5
-------
Second, the effects of structural deformation and trapped vapor were examined. For a
wide range of petroleum storage facilities, neither the range of tank and backfill properties
affecting the structural deformation of storage tanks nor the distribution of the volume of trapped
vapor is known. For this reason, the effects of structural deformation and trapped vapor are
discussed separately if they have a direct influence on the performance of a given method. An
arbitrary selection of these conditions could have resulted, unfairly, in anomalously poor
performance for many methods.
Third, the impact of the following variables was quantified: (1) operational effects such as
topping off the tank before a test, which can impair effective temperature compensation and
significantly increase structural deformation of the tank, and (2) protocol deviations such as
waiting periods that are longer or shorter than specified for starting or ending a test, which can
significantly change the deformation effects. Methods whose performance would be dominated
by trapped vapor, deformation, and operational effects are identified in the report so that the
performance, which is based primarily on the ability to compensate for product-temperature
changes, can be properly interpreted. These effects are very important, since they are a prime
cause of false alarms and missed detections.
1.6 Underground Storage Tank Test Apparatus
The evaluations were performed by the RREL at the EPA's UST Test Apparatus located in
Edison, New Jersey. The Test Apparatus is environmentally safe, and was designed and built to
evaluate the performance of in-tank leak detection systems; construction was completed in
August 1986. The Test Apparatus consists of two 2.43-m (8-ft)-diameter, 30,000-L (8,000-gal)
underground storage tanks installed in a pea-gravel backfill material; one is a steel tank coated
with plastic, and the other is a fiberglass tank (see Section 7). With this combined apparatus,
different product temperatures, product levels, and leak rates can be generated and accurately
measured. Thus, a wide range of tank testing conditions can be simulated.
1.7 Industry Participation
The only methods evaluated were those in-tank test methods that were capable of
quantifying the flow rate produced by a leak in the tank. The EPA openly solicited all
manufacturers of commercially available volumetric test methods to participate in the program.
This was accomplished by public announcement in the Commerce Business Daily in July 1986
and with assistance from the American Petroleum Institute (API), the Petroleum Equipment
1-6
-------
Institute (PEI), and the Leak Detection Technology Association (LDTA). Admission to this
program was not closed until 1 April 1987, one month after the start of the Field Verification
Tests.
Forty-three manufacturers and vendors of test methods originally indicated their desire to
participate in this voluntary program. Twenty-five manufacturers completed the requirements
necessary for evaluation. In fifteen of the methods evaluated, tests are conducted in an overfilled
tank. In three of these methods, a constant head is maintained during testing. Of the remaining
ten methods in which tests are conducted in an underfilled tank, the majority were automatic tank
gauging systems (ATGS) for which a special protocol had been developed for this program. As
a consequence, the performance of the ATGS evaluated as precision tests in this program may be
different from the performance achieved when the same systems are operated as ATGS. A list of
the twenty-five test methods and manufacturers that participated throughout the entire program is
presented in Table 1.1.
Eighteen chose not to participate in the evaluation program. The reader should not assume
that nonparticipation in the program is equivalent to poor performance. Manufacturers were not
required to give reasons for not completing the program. About half of the 18 manufacturers
who chose not to participate had been active at the beginning of the program, but dropped out
because they were unable to complete the development and/or testing of their systems prior to
the start of the Field Verification Tests. Several of the manufacturers in question missed their
scheduled test times at the UST Test Apparatus because of development problems. Additional
opportunities were provided, on a time-available basis, to accommodate the manufacturers'
development efforts and still meet the EPA's deadline for completion.
Participation in the program involved a number of steps. First, manufacturers were
required to execute a letter stating their desire to participate in the evaluation and their agreement
to abide by its rules. Each company was required to designate a single contact for the exchange
of technical and administrative information.
Second, each manufacturer was requested to answer a questionnaire that described specific
aspects of the test method. This technical information was needed to describe the test method
quantitatively; it included a claim of the precision and accuracy of the instrumentation, the
configuration of the equipment deployed during a test, the specific procedure for conducting a
test, the detailed data collection and data analysis procedures, and the criteria for determining
whether a test was validly conducted and for declaring whether a tank is leaking.
1-7
-------
Table 1.1. Participants Completing the EPA Volumetric Test Method Evaluation Program
Test Method Name
AES/Brockman Leak Detecting System
Ainlay Tank Tegrity Tester™
Automatic Tank Monitor and Tester
(AUTAMAT)
Computerized VPLT Tank Leak Testing
System
DWY Leak Sensor
EZY CHEK
Gasoline Tank Monitor (GTM)
Gilbarco Tank Monitor
Inductive Leak Detector 3100
INSTA-TEST
Leak Computer
Leak-O-Meter
LiquidManager
LMS-750
Test Method Manufacturer
Associated Environmental Systems
Mr. R. Brockman (805)325-2212
Soiltest, Inc.
Mr. F.R. Kin (312)869-5500
Exxon Research and Engineering Co.
Mr. D.B. Bolland (201) 765-3786
NDE Technology, Inc.
Dr. J.R. Mastandrea (213)212-5244
DWY Corp.
Mr. J.W. Hamblen (715) 735-9520
Homer Creative Products, Inc.
Mr. J. Homer (517) 684-7180
Tidel Systems
Mr. M. Gregory (214) 416-8222
Gilbarco, Inc.
Mr. M. Black (919) 292-301 1
Sarasota Automations, Inc.
Mr. R.P. Piccone (813)366-8770
EASI, Inc.
Mr. C. Scafidi (219) 239-7003
Tank Audit, Inc.
Mr. D.E. Hasselmann (619) 693-8277
Fluid Components, Inc.
Mr. M.M. McQueen (6 1 9) 744-6950
Colt Industries
Mr. C. Wohlers (813)882-0663
PNEUMERCATOR Co., Inc.
Mr. K.Slovak (516)293-8450
1-8
-------
Table 1.1 (concluded). Participants Completing the EPA Volumetric Test Method Evaluation
Program
Test Method Name
Test Method Manufacturer
MCG-1100
Mooney Leak Detection System
OTEC Leak Sensor
PACE Leak Tester
Petro Tite
Portable Small Leak Detector (PSLD)
S.M.A.R.T.
Tank Auditor
Tank Monitoring Device (TMD-1)
Tank Sentry II
TLS-250 Tank Level Sensing System
L & J Engineering, Inc.
Mr. L. Jannotta
(312)396-2600
The Mooney Equipment Co., Inc.
(504) 282-6959
OTEC, Inc.
Mr. J.W. Hamblen
(715)735-9520
PACE
(Petroleum Association for Conservation of
the Canadian Environment)
Mr. P. Casson (416) 298-1144
Heath Consultants, Inc.
Mr. G. Lomax
TankTech, Inc.
Mr. J.A. Carlin
(617) 344-1400
(303) 757-7876
Michael & Associates of Columbia, Inc.
Mr. M. Diimmler (803) 786-4192
Leak Detection Systems, Inc.
Mr. W.E. Baird (617) 740-1717
Pandel Instruments, Inc.
Mr. P. Lagergren (214) 660-1106
Core Laboratories, Inc.
Mr. M. Sullivan
Veeder-Root Co.
Mr. D. Fleischer
(512)289-2673
(203) 527-7201
After technical interaction with the manufacturer's representative, a report describing each
method as a mathematical model was generated; this model is a logical sequence of
mathematical steps that can be (and were) implemented on a computer. The report is referred to
as the "mathematical modeling report" or simply the "model report." A written concurrence with
the test-method description, as presented in the mathematical modeling report, was required of
each manufacturer. Some mathematical modeling reports contained information designated as
proprietary, trade-secret, or company-confidential by the manufacturer. So as to prevent the
1-9
-------
unauthorized release of information, these reports are not available for public distribution by the
EPA. Those nonproprietary aspects of the test methods necessary to interpret and understand the
results of the evaluation are summarized in the appendix in Volume I and also in Volume n.
Each manufacturer was then invited to participate in the Field Verification Tests at the
UST Test Apparatus in Edison, New Jersey. The Field Verification Tests consisted of two parts.
First, a series of product temperature conditions was established; under each condition, a leak
was simulated by withdrawing product from the Test Apparatus tank at a constant rate. Each
manufacturer was asked to test the Test Apparatus tank for leaks by following his standard test
protocol, using his own test crew and equipment. Second, the manufacturer's measurement
system was calibrated to derive an estimate of its precision and accuracy. These tests and
calibrations were used to validate the test-method model.
An evaluation report was prepared for every manufacturer who participated in the Field
Verification Tests at the UST Test Apparatus. Each evaluation report is included as an appendix
in Volume II of this report.
Finally, the manufacturers were asked to provide a written technical review of their
respective appendices and to discuss these with the EPA and its contractor before publication of
the final report. The manufacturer had three opportunities to review his evaluation while the
final report was being prepared. Valid technical comments were incorporated in the final report.
The evaluation was conducted at no cost to the participants; however, travel and other
expenses incurred by the participants during the program were not reimbursed by the EPA.
1.8 How to Use This Report
The body of this report summarizes the results of the evaluation. It contains a description
of the approach used for the evaluations, a description of the performance model used to present
the results, a summary of the performance currently being achieved by commercial systems, a
quantitative description of the important sources of testing errors, and specific recommendations
on how to minimize these errors and improve performance. Sections 2 and 3 summarize the
main conclusions and recommendations of this research program. Section 4 gives an overview
of volumetric tank testing, and provides basic information necessary to understand the other
sections of the report. The performance of a test method is described in Section 5. Section 6
describes the simulation used to estimate the performance of the methods evaluated in this
program. (Section 6.4 is highly technical and describes the ambient-noise-source models that
were used in the simulation and that control the performance of a volumetric method.) Section 7
describes the UST Test Apparatus located in Edison, New Jersey, the data quality objectives, and
1-10
-------
the quality assurance and quality control procedures used during the collection and analysis of
the experimental data. Section 8 presents important experimental and theoretical findings about
the sources of ambient noise. Section 9 summarizes the protocol used to evaluate the
performance of the 25 volumetric methods that participated in the program. In Section 10,
several calculations are performed for generic test methods to illustrate how the performance of a
volumetric method is affected by the sources of ambient noise, including how the performance of
a method changes with the number of temperature sensors used to compensate for thermal
expansion or contraction of the product. Section 11 summarizes the evaluation results presented
for each method in Volume II of this report. Section 12 discusses how test methods can improve
their performance.
The evaluation results for each method are presented in the lettered appendices to this
report (Volume II). Each appendix is sufficiently detailed that the manufacturer, or the user, of
the method can assess the performance results that are presented. Sections 1 and 2 of each
appendix gives the name of the method and the manufacturer, and a description of the method
that was evaluated. Section 3 presents the results of some limited field testing with the
manufacturer's equipment and test crew, and Section 4 describes the validation of the
performance model. Section 5 presents the performance results and gives suggestions for
improving performance, in sufficient detail, it is hoped, for the user to determine whether the
version of the test method being offered has been upgraded. The description of each method and
the suggestions for improving performance are also found in the appendix in Volume I.
The report was prepared assuming the performance standard for tank tightness tests in the
proposed rule [3], i.e., a method must be capable of detecting a leak rate of 0.38 L/h (0.10 gal/h)
with a probability of detection of 0.99 and a probability of false alarm of 0.01. The tables
summarizing performance in each manufacturer's appendix in Volume II of the report are
presented for these values of PD and PFA. The final regulation [4] is addressed in the body of the
report. The performance results summarized in the body of the report (Volume I) are given in
terms of the PDand PFA achieved by each method against a leak rate of 0.38 L/h (0.10 gal/h) using
the manufacturer's detection threshold employed at the time of the evaluation, and of the
smallest leak rates that could be detected and still maintain a PD of 0.95 and PFAof 0.05, and a PD
of 0.99 and PFAof 0.01. The latter estimates were made using thresholds that were different than
those employed by the manufacturer in the evaluation.
The evaluation results are also summarized in [12, 13]. The first reference is a report
being prepared by EPA's Center for Environmental Research Information (CERI), and the
second reference is a professional society paper.
1-11
-------
Both the body of the report and the appendices should be used to interpret the performance
and the performance limitations of a particular test method. The description of the method given
in each appendix in Volume II is an extremely important part of the evaluation. Any changes to
the manufacturer's method may dramatically alter the performance results given in the appendix.
In accordance with the EPA guidelines applicable to ORD scientific and technical
documents, all quantities in the report are presented in metric units and follow the standard
format of the International System of Units (SI). Because it is common practice in the leak
detection industry to present quantities in English units (e.g., gal/h) instead of SI units (e.g., L/h),
three steps have been taken to help the reader to readily perform the conversions and to interpret
and relate the quantities in SI units to the common practice. First, a List of Units and
Conversions is presented in the front of the report. For more detail, the reader is referred to the
Metric Practice Guide, ASTM [14]. Second, to facilitate the reading of the report, the SI
quantities are followed by the English quantities in parentheses when they are first introduced in
each chapter. Third, Table 1.2 presents the numbers most commonly used in this report in SI and
English units.
Table 1.2. Commonly Used Quantities in English Units and SI Equivalents
English Units
0.05 gal/h
0.10 gal/h
0.20 gal/h
Igal
8,000 gal
10,000 gal
0.001T
SI Equivalents
().19L/h(190orl89ml/h)
0.38 L/h
0.76 L/h
3.8 L
30,000 L
38,000 L
0.0006°C
1-12
-------
2 Conclusions
An important EPA-sponsored research program has been completed that has evaluated and
made estimates of the performance of commercially available volumetric leak detection methods
as they existed in the period March through July 1987. For each method evaluated,
recommendations were made, as required, to improve perfonnance. This two-year project has
determined and resolved key technological and engineering issues associated with this type of
leak detection. The following objectives were accomplished: (1) evaluation of the performance
of 25 currently available volumetric systems for detection of leaks in underground gasoline
storage tanks; (2) development of technical information important in the development of EPA's
underground storage tank regulations; (3) development of specific recommendations that will
allow manufacturers to improve the current practice of each method; and (4) development of
basic information to assist the test users in selecting a method that meets EPA's new regulatory
requirements for underground storage tanks. A summary of the key conclusions of this research
project are provided below.
After minor modifications, most methods should meet EPA performance requirements.
By and large, the leak detection systems evaluated were limited by protocol and current practice
rather than by hardware. In general, such limitations can be overcome by rather modest
modifications to testing practices; major equipment redesign is not necessarily required. As part
of this study, an estimate was made of the potential perfonnance that could be achieved by the
various precision test methods evaluated. The results show that with modifications, 12 of the 19
methods (over 60%) that completed the evaluations should be able to achieve a perfonnance
between 0.19 L/h (0.05 gal/h) and 0.57 L/h (0.15 gal/h), and all 25 methods (100%) evaluated
should be able to achieve a performance of approximately 0.76 L/h (0.20 gal/h). Some
manufacturers are already using the results of this evaluation to improve practices and equipment
to achieve the above perfonnance levels, and are in the process of quantifying the performance
actually achieved by the modified systems.
Presentation of evaluation results in terms ofPD and PFA gives a quantitative estimate
of performance. Twenty-five commercially available volumetric leak detection systems were
evaluated. An estimate of the perfonnance of each method was made in terms of the probability
of detection and probability of false alarm against a 0.38-L/h (0.10-gal/h) leak rate using the
detection threshold employed by each method at the time of the evaluation. Another set of
performance estimates was made for each method in temis of the smallest leak rate that could be
detected and still maintain probabilities of detection and false alarm of 0.95 and 0.05, and 0.99
and 0.01. This performance estimate does not employ the manufacturer's detection threshold;
2-1
-------
instead a threshold was selected which yields a probability of false alarm of 0.05 and 0.01,
respectively. The leak rate measurable by these systems ranged from 0.26 to 6.97 L/h (0.07 to
1.84 gal/h), with a probability of detection of 0.95 and probability of false alarm of 0.05. Five of
the methods achieved a performance that was better than 0.57 L/h (0.15 gal/h), and a total of
eight methods had a performance that was better than 0.95 L/h (0.25 gal/h). The leak rate
measurable by these systems ranged from 0.47 to 12.95 L/h (0.12 to 3.42 gal/h) when the
probability of detection increased to 0.99 and the probability of false alarm decreased to 0.01.
Only one of the methods achieved a performance better than 0.57 L/h, but five methods achieved
a performance between 0.57 L/h and 0.95 L/h. While these results are less than what is generally
claimed by the manufacturers, the phenomena that degrade performance have been identified,
and in most instances, the problems can be easily fixed. Six systems could not be evaluated
under the conditions of this evaluation either because the manufacturer was unable to
successfully conduct a tank test during a scheduled three-day testing period, because the
measurement systems did not perform as described by the manufacturer, or (in one instance)
because the Test Apparatus had not been properly configured for the tests. These six appeared to
be either new systems or systems whose basic principles of operation were not yet fully
understood by the manufacturer.
Tank testing is complex, but a high level of performance can be achieved if several key
principles are followed. To avoid serious degradation of performance, several key factors must
be accounted for when using any of the volumetric test methods. Those systems that did well in
the evaluation had adequate spatial sampling of the vertical temperature profiles of the product in
the tank; they incorporated adequate waiting periods after product delivery and/or topping the
tank (in tests that overfilled tanks) to allow the spatial inhomogeneities in the product
temperature field and the tank deformation to become negligible; they maintained a nearly
constant hydrostatic pressure head during the test; they used an experimental estimate of the
height-to-volume conversion factor; and they used sound data analysis algorithms and detection
criteria. Performance of a test method suffered significantly whenever one of these aspects of
testing was ignored or poorly implemented. In general, any method will perform poorly and
provide results that are difficult to interpret if it: (1) fails to maintain a nearly constant
hydrostatic head during the test; (2) does not accurately estimate the height-to-volume
conversion factor; (3) tops the tank and begins to test almost immediately, or (4) waits an
insufficient period of time after product delivery before beginning the test. Most manufacturers
recognized the need to wait after a product delivery, but they did not appear to fully appreciate
the magnitude of the degradation that occurs when the waiting period after topping (in methods
that overfill the tank) is not long enough.
2-2
-------
Current performance is significantly less than what is claimed by most manufacturers.
Of the 25 commercially-available volumetric leak detection systems evaluated, most presently
perform at a level that is considerably poorer than the common industry practice of 0.19 L/h
(0.05 gal/h). This discrepancy between vendor claims and actual performance appears to be due
to two reasons. First, in almost all instances, the measurements made by EPA under this project
appeared to be the first systematic evaluation of each test method. Second, the performance
estimates were formed in terms of a probability of detection and a probability of false alann, a
presentation that most manufacturers have not previously used to quantitatively describe
performance.
Removal of vapor pockets is important to the performance of overfilled-tank test
methods. Operationally, achieving a high level of performance with methods that overfill the
tank requires the removal of trapped vapor before conducting a test; this is a necessary skill that,
during this evaluation, was best demonstrated by the most experienced and best-trained test
crews. Test methods designed for use in partially filled tanks are not subject to the effects of
trapped vapor but can be affected by evaporation from the product surface and evaporation and
condensation from the tank walls. Analysis of the evaporation and condensation data is
incomplete. Based on a qualitative inspection of the test results, it is observed that when
temperature conditions in the vapor space are relatively stable, the impact of evaporation and
condensation on test performance is relatively small.
Constant-level testing is important with overjilled-tank test methods. A serious testing
flaw was discovered in all methods which overfill the tank into a fill tube or standpipe and
measure product-level changes. The error associated with this flaw was theoretically described
and experimentally verified in this program. The flaw is easily eliminated by conducting the test
at a nearly constant hydrostatic head. This can be accomplished by releveling the product in the
fill tube and measuring volume directly, or by significantly increasing the cross-sectional area of
the fill tube. The essence of the flaw is that the volume changes measured in the fill tube, after
the waiting period designed for the observed deformation changes to subside, are reduced to an
unknown fraction of the true volume changes. The flaw is a result of the fact that the tank
structurally deforms continuously with any product-level changes in the fill tube. The
contributions from all sources of volume change, including operator-induced and ambient
product-level changes, are coupled and interact dynamically in a complex way to deform the
tank. The volume changes due to leaks are similarly affected. The magnitude of the error
depends on the cross-sectional area of the product surface, the elasticity properties of the
tank-backfill-soil system, and the volume of trapped vapor.
2-3
-------
Reliable tank testing takes time. The total time required for the methods evaluated at the
UST Test Apparatus to complete a reliable tank test, from delivery of product to removal of the
equipment from the testing site, is generally 12 to 24 h. The total duration of the test is
controlled by the waiting periods after product delivery or topping the tank. The waiting periods
can be shortened by incorporating data analysis algorithms into the test protocol. In this way the
duration of the waiting periods can be estimated individually for each test.
2-4
-------
3 Recommendations
Leak detection technology is rapidly evolving. With the passage of federal legislation in
1984, the drafting of national UST rules by EPA, the establishment of numerous state and local
UST programs over the past several years, and the completion of this RREL research effort, the
economic motivation for system improvements and technological discipline has been
established. However, it is also clear that all the methods evaluated can be improved and that
many of the methods need such improvement in order to satisfy the new EPA regulatory
requirements given in [4]. The following recommendations are a direct outcome of the
evaluation effort and are provided so that manufacturers, test users, and state and local regulators
will be clearly informed as to what steps must he taken to make the necessary improvements.
Acceptable means are needed for evaluating anticipated improvements in performance.
The results of this evaluation indicate that many manufacturers will have to make improvements
to their systems in order to be able to meet the new EPA regulatory requirements. Others will
make improvements to achieve a higher level of performance. The majority of these will be
straightforward changes in how the method collects, analyzes, and interprets test data,
compensates for temperature, accounts for tank deformation, and ensures that adequate waiting
periods have elapsed. After these changes are made, many of the methods will meet the
regulatory standard. However, this does not eliminate the need for verifying that the required
performance improvements have been achieved.
To estimate the performance of a detection system, the histograms of the noise and
signal-plus-noise need to be characterized. A complete performance evaluation includes the
histogram of the noise, the conditions used to define the noise, the histogram of the
signal-plus-noise (usually developed from the relationship between the signal and the noise), the
performance in terms of the probability of detection, the probability of false alarm and the leak
rate, and the uncertainty of the probability of detection and the probability of false alarm (i.e., the
number of independent test conditions).
There are many different but viable approaches to estimating the performance of detection
systems. For some of the methods evaluated an estimate of the performance achieved after
improvements have been implemented can be calculated using information provided in this
report. For the majority of methods, however, the need exists for experimental evidence before
any new claims of improved performance can be fully supported. The necessary experimental
data can be gathered from operational data collected at retail stations or test data collected on
specially instrumented tanks. There are many acceptable approaches to evaluation that
incorporate the principles used by the EPA at the Test Apparatus in Edison without the need for
3-1
-------
such a sophisticated facility. Acceptable methods of evaluation should be and are being
developed by government, industry, and professional societies as guidance to manufacturers,
users, and regulators whose responsibility it is to evaluate or interpret performance claims. Until
such evaluation procedures are completed, it is strongly recommended that users employ only
those test methods that have (at the minimum) adopted the procedural and equipment
recommendations made in this report.
Two of the anticipated ramifications of this research project are rapid improvements in
performance made by manufacturers and and increase in test users' and regulators' expectations
concerning verification of future performance claims for all volumetric methods. Manufacturers
of many of the methods that were evaluated by RREL have already begun to make the changes
necessary to improve their systems' performance and to verify the new performance claims.
This is a very encouraging development.
All volumetric test methods must account for and control the key sources of noise that
impact performance. Reliably detecting small leaks in underground storage tanks is a
technically challenging task. To achieve a high level of performance against leaks as small as
0.38 L/h (0.10 gal/h) requires:
o Sufficient waiting period after product delivery. Spatial inhomogeneities in the
temperature field make the conventional approach to temperature compensation
unacceptable for a period of up to 4 to 6 h after delivery. Tank deformation effects,
however, may continue even longer. As with product delivery, a waiting period which
minimizes these effects should be incorporated into the test protocol. The waiting period
should be at least 6 h long unless validated analysis algorithms designed to minimize the
waiting period are incorporated into the test protocol.
o Sufficient waiting period after topping. Several overfilled-tank test methods top the tank
to a specified level immediately before testing. Although not much product is added
compared to a product delivery, the resulting temperature and tank deformation effects can
be significant. The horizontal temperature gradients that develop from topping last 2 to
3 h. The deformation effects can be large for a considerable period. A waiting period
which minimizes these effects should be incorporated into the test protocol. The waiting
period should be at least 3 h long unless validated analysis algorithms designed to
minimize the waiting period are incorporated into the test protocol.
3-2
-------
o Sufficient waiting period after disturbing the vapor space in a partially filled tank. In
some instances, it was observed that after a tank was opened to the environment the effects
of evaporation and condensation of product remained large for 4 to 6 h. Additional
analysis is required before a minimum waiting period can be recommended.
o Adequate temperature compensation. Leaks smaller than 3.8 L/h (1.0 gal/h) cannot be
reliably detected without temperature compensation. To reliably detect 0.38 L/h
(0.10 gal/h) leaks, a temperature array with at least five temperature sensors (or an
averaging sensor with equivalent spatial coverage) appears to be necessary to adequately
characterize the vertical temperature profile in the tank (for methods which do not circulate
the product).
o Testing at a nearly constant hydrostatic head. All tests should be conducted at a nearly
constant hydrostatic pressure. Tests in partially-filled tanks are conducted at a nearly
constant hydrostatic head but this is not true for overfilled-tank tests which measure
product-level changes in a fill rube or standpipe. For accurate testing, these overfilled-tank
tests should (1) measure volume directly by periodically releveling the product in the fill
tube or standpipe, (2) enlarge the cross-sectional area of the measurement container, or
(3) measure the elastic properties of the tank-backfill-soil system and incorporate them into
the data analysis.
o Approaching data collection and data analysis quantitatively. Reliable leak detection can
be achieved by using simple, standard textbook data collection, reduction, and analysis
algorithms. The effect of the data collection and data analysis algorithms needs to be
understood, particularly when statistical tests are peifonned on correlated or aliased data.
o Testing with calibrated measurement systems. To achieve a high level of performance in
the field, the leak detection systems must be well maintained, properly installed, and
regularly and accurately calibrated. Futheimore, it is highly desirable to include diagnostic
calibration checks of the instrumentation as part of the test procedure.
Test users and regulators should require that the protocols developed for individual test
methods be followed closely in the field by test operators. The evaluation of the volumetric test
methods has shown that test operators must follow their own test method's protocol to achieve
the level of performance that the method is capable of. Any deviations from the protocol will
affect the performance of the method. It is recommended that once a test protocol is established
and evaluated, it should be carefully followed for each test. It is particularly important to
-------
observe the specified waiting periods that are required to minimize the effects of temperature
change and deformation that result when product is added to the tank. Moreover, any arbitrary
judgments about the conduct of the test should be avoided.
Users of test methods should request from the manufacturer a written copy of the
important features of the test method protocol. The user should also take time to observe
whether those specifications are being following. Regulators should examine written protocols
of methods that have been approved for use in their jurisdictions. If the method has been
evaluated by the EPA, they should determine whether equipment and protocol changes have
been made since the evaluation and should review the new performance claims.
Volumetric test methods should not be selected based solely on the ranking contained in
this report. Because it has been shown that the performance of many of the individual test
methods can be significantly improved with only minor modifications, some of the
worst-performing methods are expected to raise their perfomnance to the level (or beyond) of the
best-performing methods in the immediate future. As a consequence, and because many
methods have already incorporated many of the recommended changes for improvement, the
current ranking of test method performance implied in this report is not particularly significant.
Manufacturers of volumetric test methods must closely heed the findings of this report to
assure future compliance with the new regulatory requirements. Three of the methods were
able to detect leaks of 0.38 L/h (0.1 gal/h) with a probability of detection of 0.95 and a
probability of false alarm of 0.05. That means some additional improvements to many of the
volumetric methods evaluated by RREL will eventually be needed in order for them to meet the
new regulatory requirements for periodic tank tightness testing. Even the methods that meet the
regulatory requirements may need improvement, because effects like topping the tank, which
would degrade the performance of these methods, were not included in the performance
estimates. These recommended improvements are often very simple to make but have been
found to be critical to improving performance. Test users and state and local regulators will
undoubtedly expect the improvements to be made. Under the new federal regulations,
manufacturers of volumetric methods will have up to two years to demonstrate they have
achieved the performance levels (specifically the Pn and PFA values) required by the EPA.
3-4
-------
4 Overview of Volumetric Precision Tank Testing
A volumetric leak detection system is designed to make an estimate of the total volume of
product that is lost due to a leak over the duration of a test. For most tests, the volume lost is
based on a change in product level. Tests may be conducted in a partially filled tank or in a tank
overfilled into a fill tube that is connected to the tank or to an above-grade standpipe
(see Figure 4.1). Two types of overfilled-tank tests are used. The most common one measures
the product-level changes in the fill tube or standpipe, and the other periodically adds or removes
product to maintain a constant product level during the test. Product level is measured using a
variety of sensor systems.
Unfortunately, product-level fluctuations in a tank (the fundamental measurement) are also
produced by numerous effects not related to a leak. These ambient fluctuations
(i.e., ambient noise) can be as large as or larger than the smallest leaks to be detected. They may
be caused by long waves in the tank which periodically move the product back and forth,
primarily along the long axis of the tank (i.e., seiching); by volume changes produced by thermal
expansion or contraction of the product; by evaporation or condensation at the product surface;
by changes in the volume of the trapped vapor produced by temperature and pressure changes;
and by changes, produced by structural deformation, in the volume of the tank itself. The
accuracy, or "performance," achieved by a method usually depends on the magnitude of these
fluctuations and on how well that method can minimize or compensate for these non-leak-related
volume changes.
Grade
Partially Filled Tank
Fill Tube
\
Backfill
Standpipe
Tank Overfilled into Fill Tube
Grade Fill Tube
B
Product
Level
Tank
Native
Soil
Excavation
Native
Soil
Excavation
Figure 4.1. Cross-sectional view of an underground storage tank that is (A) partially filled and
(B) overfilled into the fill tube.
The basic strategy in any volumetric test is as follows: (1) before taking measurements,
observe appropriate waiting periods so as to avoid the problems associated with structural
deformation of the tank and with the thermal instabilities that develop after product delivery or
topping off the tank; (2) measure product-level change over the duration of the test; (3) measure
4-1
-------
product temperature over the same period; (4) properly sample and average the data so as to
remove the effects of both surface and internal waves; (5) convert measurements to volume
changes; (6) subtract out the thermally induced volume contribution; and (7) declare the tank
leaking if the residual exceeds a specific threshold value. The basic assumptions are that vapor
pockets have been removed, evaporation-condensation effects are small, and the groundwater
table is below the bottom of the tank. Interpretation of the test result (i.e., the size of the leak) is
discussed below.
4.1 Product-Level Fluctuations in an Underground Storage Tank
Product-level fluctuations in an underground storage tank can result from a leak as well as
from a wide variety of other physical changes in the tank, product, or product surface.
4.1.1 Flow Rate Produced by a Leak
A leak may occur at any breach, hole, or loose fitting in an underground storage tank.
Leaks near the top of the tank may occur because of missing bungs, or loose or broken joints
where the piping is attached to the tank. A breach caused by corrosion or a crack in the tank
material may occur at any location in the tank.
The performance of a test method is based simply on the ability of the method to measure
the flow rate that occurs during a test. The flow rate through a hole depends, in general, on the
hydrostatic pressure exerted on the hole, on the physical properties of the hole, and on sediment
particulates in the tank and the backfill material outside the tank that may retard flow through the
hole. The hydrostatic pressure is, in turn, a function of the level and density of product in the
tank, the level and density of the groundwater table outside the tank, and the location of the hole.
Flow through a hole could, under a different hydrostatic pressure, represent a larger or smaller
leak rate than that measured during the test. Interpretation problems can be minimized by
ensuring that the water table is below the bottom of the tank, as was the case during this
evaluation.
A more detailed description of how the flow rate is affected by different combinations of
product level and groundwater level is given in Section 4.7.
4.1.2 Apparent Flow Rate Produced by Non-Leak-Related Volume Fluctuations
Product-level fluctuations can affect the performance of a test method because they may be
mistaken for (or interpreted as) a leak. The sources of these product-level fluctuations are:
o changes in the volume of the product and of air or vapor trapped in the tank
4-2
-------
o changes in the volume of the tank itself
o product-level changes caused by surface and internal waves in the product
4.1.2.1 Changes in the Volume of Product and Trapped Vapor
Changes in the volume of product and trapped vapor are caused by any of several
phenomena.
o Expansion and contraction of the product due to temperature changes. Large volume
changes result from thermal expansion and contraction because of three factors: the
temperature changes in the tank may be large; the volume of the product in the tank is
large; and the coefficient of thermal expansion of petroleum products is also large
(nearly 4 to 7 times larger than that of water).
o Expansion and contraction of vapor trapped in the tank. Product-level changes also
result from the expansion and contraction of any trapped vapor and are a consequence
of changes in vapor temperature, vapor pressure, atmospheric pressure, and product
level. The magnitude of these volume changes is strongly dependent on the total
volume of the vapor trapped in the tank. These changes occur only if the tank is
overfilled for the test.
o Evaporation and condensation of product at the product surface and along the tank
walls. The volume changes due to this phenomenon can be important for tests
conducted in a partially filled tank.
4.1.2.2 Changes in Tank Volume
The tank itself may undergo a change in volume during the test. The most important and
least understood cause of such a change is structural deformation of the tank. Deformation,
which occurs whenever the hydrostatic pressure relative to the bottom of the tank changes,
consists of an instantaneous change in product level and a time-dependent relaxation of the
tank-backfill-soil system (see Figure 8.15). Deformation can result in volume changes large
enough to mask a leak, and, under some test conditions, can also affect the magnitude of the
measured leak rate. For this reason, an understanding of the effects of structural deformation is
essential.
A change in volume can also result from thermal expansion or contraction of the tank
walls. This effect is not large unless the temperature changes are large. A simple calculation for
a steel tank, assuming that the tank is a right regular cylinder, suggests that a 1°C change in
4-3
-------
temperature would result in a 0.30-L (0.08-gal) change in volume. This effect is most significant
immediately after a delivery, when the temperature difference between the product in the tank
and the backfill and in situ soil is largest.
4.1.2.3 Waves
Periodic product surface undulations can have several sources. A disturbance of the
surface causes a seiching motion in the tank, meaning that product moves back and forth as a
standing wave along either horizontal axis of the tank. The dominant seiche occurs along the
long axis of the tank. Seiches, which can be quite large, occur mainly during partially-filled-tank
tests, but can also occur during overfilled-tank tests if two product surfaces exist in the tank (e.g.,
fill tube and vent tube). The period of the standing waves in a partially filled tank may range
from 2 to 10s, depending on the geometry of the product surface. The presence of other wave
motions in the tank (e.g., progressive waves) does not appear to be an important source of error.
Periodic surface undulations can occasionally be caused by internal waves, which are
subsurface waves that can occur on a density gradient in the product. Since, for a given product,
density is dependent mainly on temperature, these waves are typically found on temperature
gradients. However, they may also occur at the boundary separating two products of different
composition. Internal waves, which typically have a period of 5 to 20 min or longer in a
30,000-L (8,000-gal) tank, do not generally have a significant influence on the surface, but this is
possible under some conditions. The presence of internal waves is more likely to impact
temperature estimation. Undersampling the temperature data either spatially in the vertical field
or temporally can lead to significant errors in estimating the thermal expansion of the product in
the tank.
Waves should not affect the performance of a test method unless the data are
undersampled, a situation which can lead to aliasing of the results. This error can be avoided if
the data are sampled at least as often as one-half the period of the waves'. It is generally
recommended that the data be sampled at one-third of the period. Thus, a 6-s period should be
sampled at 2 s. Similarly, a 10-min internal wave would require that both the temperature and
product-level measurements be sampled at 3.3 min. To accommodate both wave effects, a
scheme that sampled the data at 2 s and averaged the samples to 3 min would suffice.
1. Waves in a half-filled 4,000-L (1,000-gal), 30,000-L (8,000-gal), and 80,000-L (20,000-gal) tanks typically have
periods of approximately 3, 5, and 6 s, respectively.
4-4
-------
4.2 Flow Rate Measurements during a Tank Test
An underlying assumption of a volumetric test is that product-level changes in a tank,
whether produced by a leak or by any other volume change (such as the thermal expansion or
contraction of the product), can be interpreted as product-volume changes. An accurate estimate
of the height-to-volume conversion factor is required in order to convert the product-level
changes to product-volume changes. The height-to-volume conversion factor is defined by
AV .. ,.
Aeff = -rr (4-1)
Ah
where A^f is the effective cross-sectional area of the product surface and AV is the volume
change produced by a product-level change, Ah. A,.ff may or may not be equal to the actual or
geometrical cross-section of the product surface, A. If, in an overfilled-tank test, the
instantaneous deformation of the tank is negligible, and if no vapor is trapped in the tank, then
Aeff is the actual cross-sectional area of the product surface. If the tank deforms instantaneously,
and if trapped vapor exists in the tank, then A,,ff can be defined as the sum of the individual
contributions to the effective surface area.
Aeff = A+Avp+Aisd (4.2)
where A^ and Aisd are the volume changes per unit of product-level change produced by the
compressibility of trapped vapor and the instantaneous structural deformation, respectively.
Typically, A,sd and Avp cannot be measured separately unless the volume of the trapped vapor is
known. Eq. (4.2) indicates that Aeff can be interpreted as the sum of the individual contributions
to Aeff. For partially-filled-tank tests, Acff is usually equal to A and can be determined from the
geometry of the tank, if it is well known. Because of the uncertainties in the tank geometry,
particularly in the upper 10% of the tank, it is more accurate to measure A rather than to compute
A from the manufacturer's tank chart relating the nominal volume of the tank with product level
in the tank. For both overfilled-tank and partially-filled-tank tests, Aeff can be estimated from a
simple measurement of the product-level change produced by inserting and removing a solid of
known volume, e.g., a cylindrical bar.
In an overfilled-tank test, A^ must be measured in order to account for the instantaneous
structural deformation of the tank and the volume changes in the trapped vapor. A geometrical
estimate will frequently lead to erroneous test results. Section 6 discusses the dynamics of
product-level fluctuations within the fill tube or an above-grade standpipe and the impact of this
phenomenon on system performance. For variable-head overfilled-tank tests, it has been shown
4-5
-------
that \ff cannot be used to convert product-level changes to product-volume changes, and that the
effective height-to-volume calibration factor is defined by A,.ff + K, where K is the elasticity
constant of the tank-backfill-soil system.
In a partially-filled-tank test (conducted with product level in the upper 10% of the tank)
Acff must also be measured, for two reasons. First, the geometry of the upper portion of the tank,
which may be distorted during installation, is not known. Errors of 100% in estimating A^ in
the upper 5% of the two 30,000-L UST Test Apparatus tanks were typical, even though the tanks
had been installed horizontally and the inside dimensions had been accurately measured.
Second, it can be shown that a stick measurement of the mean product level required in order to
use the tank chart to estimate A is not accurate enough if small leak rates are to be reliably
detected.
4.3 Classification of Volumetric Test Methods
For the purpose of evaluating performance, volumetric test methods may be divided into
two groups. The first group includes those methods in which the hydrostatic pressure is
approximately constant throughout the test. For these tests, product-volume measurements can
be readily interpreted from the product-level measurements. In this group are those methods that
test in a partially filled tank, as well as those that test in an overfilled tank and periodically
relevel the product at short intervals. A partially-filled-tank test is considered a constant-head
test, because even large volume changes cause only small product-level changes in the tank. For
example, a 0.76-L (0.20-gal) product-volume change will cause a change of only 0.013 cm
(0.005 in.) in a partially filled tank, but the same volume addition in an overfilled tank will cause
a product-level change of 10 cm (4 in.) in a 10-cm-diameter fill tube if the tank is rigid and
contains no trapped vapor.
In the second group are tests which allow the product level in the fill tube to change; these
do not maintain a constant hydrostatic pressure, and product volume cannot be interpreted from
the product-level measurements unless the deformation characteristics of the tank-backfill-soil
system are known.
4.4 Major Features of a Tank Test
A volumetric tank test has several major features that affect the performance of a given
method. These features are presented in Figure 4.2. Not all methods include or adequately treat
all of the features, and as a result performance is sacrificed. Some manufacturers have operating
manuals with detailed instructions on how to conduct a test, but many do not.
4-6
-------
Gross Adjustment of Product Level
(Product Delivery)
Gwait for Temperature
nd Structural Deformation
to Become Small
Equipment Setup
Fine Adjustment of Product Level
(Topping the Tank)
Wait
Product Level and Temperature Measurement
Data Analysis
Figure 4.2. Overview of volumetric tank test procedure.
4.4.1 Product Delivery
The majority of volumetric test methods require that the test be conducted at a specified
product level. Product is usually added in two steps to attain the required level. The first step is a
gross adjustment and usually involves the delivery of a large amount of product to the tank,
typically done the night before the test is to be conducted. Thousands of gallons of product may
be added to the tank. The final level required for the conduct of a test is almost never attained in
this first delivery. This is particularly true of tests that are conducted when the tank is overfilled
because the person delivering product to the tank will not usually overfill it. Whether the level
specified for a partially-filled-tank test is accurately attained as a result of a delivery depends on
4-7
-------
the accuracy of the predicted level of the product in the tank at the time the product is delivered,
on the accuracy of the tank chart used to calculate the amount of product to be delivered to fill
the tank, and on the accuracy of the volume estimate of product actually delivered. The second
step is the fine adjustment required to attain the specified test level. This is usually done the next
day, immediately before starting the test. The amount of product required to reach the test level
may be several tens of liters. Both additions could be done at the time of the delivery if the test
crew were present at that time and remained there to fine-tune the product level during the time
required for the major temperature changes to dissipate. Such an approach would significantly
decrease the time required to perform a test.
The addition of product has two effects on a test. First, the added product is likely to be at
a temperature different from the original product in the tank and from the backfill and soil
around the tank. This temperature difference results in large temperature changes in the product
immediately after the delivery. The horizontal temperature gradients that develop during the
period immediately following a delivery are large enough to prevent accurate temperature
compensation. Second, the tank will deform as the product level in the tank changes. The
product level (which affects the hydrostatic pressure in the tank) could change as much as 2.43 m
(8 ft) in a tank 2.43 m in diameter. More typically, the product level will change 1 to 2 m. The
structural deformation, which occurs nearly instantaneously in response to the addition of
product, may be large (e.g., tens of liters). This deflection (what many people think of when
structural deformation is discussed) is not what typically produces testing errors. The errors
really result from the time-dependent relaxation of the tank as it tries to reach a geometric
equilibrium with the backfill and soil surrounding it.
4.4.2 Waiting Period after Product Delivery
Most test methods, but not all, require a minimum waiting period after the product delivery
and before the start of data collection. This waiting period is included in the test protocol to
allow the temperature field in the tank to equilibrate and to allow the deformation effects to
become small enough to be negligible. In general, most test methods require a waiting period of
at least 4 to 6 h.
The performance of the methods presented in this evaluation depends on the
manufacturer's observance of the minimum waiting period specified by the test protocol. If this
period is too short to begin with, or if it is long enough in the protocol but arbitrarily shortened in
the field, the performance of the method will decrease.
4-8
-------
4.4.3 Equipment Setup
The time required to set up equipment varies among manufacturers. In general, it takes
longer to prepare the tank for an overfilled-tank test than for a partially-filled-tank test because
of the time required to completely seal the tank and bleed it of trapped vapor. The criterion for
determining whether the tank is free of vapor is at best qualitative, and thus depends on the
judgment of the test crew. The manufacturers and test crews with the most training and
commercial tank testing experience were more proficient at removing trapped vapor than those
crews with less training and experience.
4.4.4 Topping the Tank
The product level must be adjusted to the level required for the test. These adjustments are
normally required for the overfilled-tank tests. Usually, the product level in the tank after a
delivery is near the top of the tank (e.g., 10 cm), but is not in the fill tube. In comparison to the
delivery the night before, only a small amount of product is needed to overfill or top the tank.
Once product is at the level of the fill tube, the addition of 4 L (1.1 gal) of product could change
the level up to 0.5 m (1.6 ft).
Note that large changes in the hydrostatic pressure occur when product level is being
adjusted to the height specified for starting a test. The product-level changes during topping can
be larger than those during the delivery process. In addition, the temperature changes following
the topping can be as severe as those that follow a delivery. This addition of product disturbs the
temperature field in the tank as a whole, as well as in the local area around the temperature
measurement system located at the fill hole of the tank.
4.4.5 Waiting Period after Topping the Tank
Many manufacturers of overfilled-tank tests also specify that the test cannot begin
immediately after the tank has been topped; this allows the structural deformation and
inhomogeneities in the product-temperature field to become negligible and is usually the reason
that a protocol may stipulate a waiting period. Those methods in which the tank is partially filled
may also require a waiting period when the level is adjusted. Typically, a period no longer than
1 h is required by the tester.
The topping of the tank may be the single most influential factor affecting the performance
of the test method. How long it takes to top the tank, as well as the waiting time after topping, is
important. The evaluations conducted during this program were in strict accord with the
operational protocol of the manufacturer. Because of the complexity of the deformation in an
4-9
-------
overfilled tank, any changes to the test protocol affect the results of a test. It is particularly
important that the waiting time not be any shorter than the minimum specified by the
manufacturer's test protocol.
4.4.6 Leak Detection Test
The test consists of collecting and storing the product-level and temperature data for
analysis. Additional measurements are sometimes made by the manufacturer to interpret the test
data.
4.4.6.1 Data Collection
Most of the manufacturers make a product-level measurement and a temperature
measurement. A few of the methods measure change in volume or product mass. The leak
detection tests vary from 1 to 12 h, although the data required to compute a leak rate may be as
short as 15 min. The actual test duration is sometimes fixed by the test protocol and sometimes
dependent on the collected data. The data sampling interval varies from less than 1 s to 12 h.
Those manufacturers who do not specify a fixed test tune usually terminate the test based on a
measurement of the product-level and temperature fluctuations. Some of the criteria in the
protocol are physically impossible to meet, while others are so stringent that a test could never be
completed in any practical length of time. In these cases, the operator's decision to stop the test
is often subjective. Most of the data are collected manually by reading an instrument or making
a measurement, and the results are recorded on a data sheet. Some manufacturers have
automated the data collection with the use of a computer.
4.4.6.2 Supplementary Measurements
There are two additional measurements commonly made to more accurately analyze the
data. First, the majority of methods make a field estimate of the coefficient of thermal expansion
of the product by making an API gravity measurement [15]. An average value of the coefficient
for the type of product being tested (e.g., gasoline, diesel) is assumed if no measurements are
made. Few of the manufacturers make a laboratory estimate of the coefficient. Although the
error in estimating the coefficient could be 5 to 10%, this is usually not significant as compared
to other errors presently found in the current practice.
Second, some manufacturers attempt to measure the height-to-volume conversion factor
directly using a bar of known volume to displace the product level. If no measurement is made,
4-10
-------
the conversion factor is calculated from an estimate of the exposed product surface. This is done
using a tank chart in the case of a partially-filled-tank test, and by measurement of the fill tube,
vent, and other openings in the case of an overfilled-tank test.
4.4.7 Data Analysis
All of the manufacturers have well defined procedures for reducing the data, doing
temperature compensation, and computing a volume rate. The data reduction may require
averaging the data or resampling them at a different rate. In some instances, the temperature data
are not sampled at the same rate as the product-level measurements.
Product-level changes are converted to product volume using a height-to-volume
conversion factor. Some of the overfilled-tank methods that relevel the product make a direct
measurement of the volume added to or removed from the fill tube.
The means for estimating thermal volume changes are more or less common to all
methods. The algorithm estimates the mean rate of change of product temperature and converts
the temperature change to a volume change. The majority of the manufacturers estimate the
volume of the product in the tank from a tank chart, measure or assume a value of the coefficient
of thermal expansion, and measure the rate of change of temperature.
The temperature-induced volume changes are then subtracted from the the product volume
changes. These temperature-compensated volume fluctuations are then used to estimate the
temperature-compensated volume rate. If the tank is leaking, the temperature-compensated
volume is equal to the leak rate.
Some manufacturers do not use all of the data in a volume rate analysis. Some algorithms
calculate a volume rate by subtracting end-of-test data from start-of-test data, and divide by the
time that has elapsed between the two. Other algorithms add the cumulative volume changes to
find the total volume change, and then divide by the test duration to get a rate. Finally, some
manufacturers use a least-squares estimation approach to determine volume rate.
4.4.8 Detection Criterion
Most manufacturers use a single-threshold criterion for determining whether the tank
should be declared leaking. If the temperature-compensated volume rate, when compared to the
threshold volume rate, is greater than the threshold, the tank is declared leaking. The most
common threshold is 0.19 L/h (0.05 gal/h) [5].
4-11
-------
In a few cases, the detection criterion is very complicated and, in some instances, appears
to be rather arbitrary. For some test methods, the criterion for declaring a leak is ambiguous
when the temperature-compensated volume indicates that (a) the flow is into the tank, and (b) the
groundwater table is below the bottom of the tank (or unknown but assumed to be below the
bottom of the tank). In some instances the tank is declared tight, even though the threshold is
exceeded. At other times, the data collection period is extended so that, in effect, another test is
conducted.
In actual practice, the detection criterion may not be observed if the
temperature-compensated volume rate is close to but exceeds the threshold, or if an extremely
large value is obtained, suggesting that the test was faulty. Many test crews conduct a second
test if the first exceeds the threshold by only a small amount; they declare the tank to be tight if
the threshold is not exceeded in the second test. This particular multiple testing approach lowers
the probability of false alarm but also lowers the probability of detection. If this unplanned
retesting approach is repeated several times, the detection performance of a method having a
high probability of detection from a single test would be lowered to an unacceptable level. The
performance of the method depends on how many tests are conducted, and how the results from
each test are used.
4.5 Time Required to Complete a Test
The time required to complete a test was defined as the total time that had elapsed between
the arrival of the test crew at the storage facility and the final removal of the equipment from the
tank. Delays in setting up the equipment, topping off the tank, and extending a test tend to
increase the test time. On the average, the time required to set up the equipment and conduct a
leak detection test is approximately 8 h; the time required for product delivery and the waiting
period after the delivery will increase this time. This estimate is based on testing experience at
the Test Apparatus.
4.6 Test Accuracy
The accuracy of a volumetric precision tank test can be altered significantly if the operator
modifies the test procedure.
4.6.1 Performance of the Method
The performance of each method represents the capability of the method to detect a flow
rate produced by a leak located somewhere in the tank. The probability of detection and
probability of false alarm are determined for each flow rate. The measured flow rate cannot be
4-12
-------
properly interpreted in terms of potential environmental damage unless the hydrostatic pressure
on the leak is known. The estimates made in this evaluation assume that the test crew follows
the test procedures exactly as outlined by the manufacturer. If this is not done, the performance
will be altered. In some instances, performance may actually improve as a result of delays in
setting up and conducting a test. In most cases, the operator can degrade performance.
Sometimes the errors are easily identified in a quality control/quality assurance audit, while in
other cases, the effects are not known.
4.6.2 Operator Influences
Of the 25 test methods evaluated in the program, 19 involve the use of an operator to
conduct a test. Tests with the other six methods are performed automatically once the equipment
has been installed. For some methods, the role of the operator is simply to implement a
well-structured test protocol. In other instances, the operator is allowed, or even required, to
alter the protocol based on his own judgment; this is the more common situation. During the
development of the mathematical model for each method, any operator judgments required to
start or stop a test, or to interpret the results of the test, were quantified by the manufacturer. In
most instances, the manufacturers had a great deal of difficulty doing this, or seeing the need to
do it. Without quantitative judgments, however, the test method could not be evaluated.
The operator errors can be divided into four categories:
o improper setup of the test equipment or improper preparation of the tank before the
test
o mistakes or carelessness in following the test protocol
o mistakes in recording the data or in performing the analysis
o deliberate changes to the test protocol based on operator judgments about the quality
of the data or on real-time operator interpretation of the results
4.6.2.1 Improper Setup
The performance estimates made in this evaluation assume that the tank has been properly
prepared for the test and that the equipment is properly set up. While there was difficulty with
setting up equipment for some of the newer methods, this was, in general, not a major problem.
Only for one of the methods was the crew unable to set up and conduct a test. The crew for
another method was able to conduct one test, but did not believe that it was a valid one. Several
other crews had instrumentation difficulties, but were able to correct the deficiencies and conduct
tests.
4-13
-------
Because trapped vapor can be a significant source of error in overfilled-tank tests,
satisfactory removal of all trapped vapor is essential for an accurate test (see Section 8.2). It is
the operator's responsibility to set up the test equipment without trapping additional vapor, and
to properly remove from the tank, before a test is begun, any vapor that was already present. All
the manufacturers claim that this can be satisfactorily done. They also claim to be able to
recognize the presence of trapped vapor, even though they do not use a quantitative procedure.
Tests conducted at the Test Apparatus also suggest that the effects of trapped vapor cannot be
differentiated from the instantaneous tank deformation. If the tank did not deform
instantaneously, it would be possible to measure the volume of trapped vapor, but since the tank
does deform, any estimates of trapped vapor also include deformation effects.
4.6.2.2 Failure to Carefully Follow Test Protocol
The performance estimates assume that the test method is followed exactly. Failing to
follow a certain aspect of the test protocol can seriously alter the expected performance of a test.
The most common infraction of the protocol is the failure to observe specified waiting periods.
The resulting error is subtle, and involves the time history of product-level changes in preparing
for a test. Since future product-level changes depend on past ones, delays in testing, or not
following the same procedure to add or remove product from the tank, will produce different
results. (See Section 6.4). In some instances, the performance may improve as a result of delays
in initiating testing.
4.6.2.3 Data Tabulation and Analysis Errors
The performance estimates assume that no errors occur in recording and analyzing the
data. Many of the methods require the operator to read and manually record data. The
calculations to estimate the temperature-compensated volume rate are often performed manually.
Many of the test methods have detailed analysis sheets to help keep the operator from making
mistakes.
The majority of the data recording errors and all of the calculational errors can be
identified in a post-test data quality review. Whether a thorough data quality and analysis review
is performed after the completion of a test is not known.
4.6.2.4 Arbitrary Changes to the Test Protocol Made by the Test Operator
As indicated in Section 4.6.2.2, any change to the test protocol will affect the perfonnance
of the method. While the operator should be required to make certain judgments (for example,
whether the equipment is properly installed), any arbitrary judgments concerning the quality of
4-14
-------
the data as they are being collected, any changes in the length of the waiting periods, or any
changes in the detection criterion should be avoided. These operator judgments are usually made
because something about the test does not look right. However, in a validated system, these
so-called problems have already been included in the performance estimate. If the problems are
important to solve, the procedure should be included in the protocol and quantified as pan of the
validation. If not, the performance of the method will be affected, usually in an unknown way.
4.7 Groundwater Level Affects the Magnitude of the Leak Rate
The performance of a test method is based simply on the ability of the method to measure
the flow rate that occurs during a test. Flow through the hole in the tank, whether this hole is
large or small, can, under different hydrostatic pressures, represent a larger or smaller leak rate
than that measured during the test. Interpretation of the environmental effects of a leak in the
tank (i.e., the loss of product from the tank) requires more information than is known at the time
of the test. The magnitude of the flow rate produced by a leak or hole in the tank is discussed
below.
The relationship between the leak rate and the hydrostatic pressure is not well known. We
could assume that the leak rate, LR, from a hole in the tank can be computed using the following
relationship developed for flow from a free orifice:
LR = K,ALR(2gAh)1/2 (4.3)
where K, is a flow coefficient, ALR is the cross-sectional area of the hole, g is the acceleration
due to gravity, and Ah is the hydrostatic pressure head. The flow rate varies as the square root of
the hydrostatic pressure head. This relationship is valid if the leak is large and the backfill does
not inhibit flow through the hole in the tank. For small leaks or in cases when the flow is not
free, as when improper backfill containing clay restricts the flow, a linear relationship might best
describe the flow rate. In this case, the flow rate can be estimated by
LR = K2ALRgAh (4.4)
where K2 is a flow coefficient. Both relationships are probably valid for some tests. The actual
flow rate will depend on the magnitude of the measured constant. However, no experimental
evidence exists to validate either relationship. A 50% change in hydrostatic head will result in a
flow rate that is 70% of the original flow rate for the free-orifice model, and 50% for the linear
model. Regardless, the larger the hydrostatic pressure, the larger the leak rate.
4-15
-------
The hydrostatic pressure, Ah, (i.e., the relative internal and external pressure at the location
of the hole) has to be known in order to correctly interpret the flow rate measured during a test.
Since the location of the leak is not typically known during an actual test, a quantitative estimate
of Ah cannot be made. The test result that is simplest to interpret occurs when the groundwater
level is below the bottom of the tank. If the tank has a leak, the flow rate measured during a test
will at least be representative of the hydrostatic pressure exerted by the product; the flow rate
measured at a lower product level will be less. This is not true if the groundwater level is above
the bottom of the tank. It is also possible for the hydrostatic pressure exerted by the product
inside the tank to equal the hydrostatic pressure exerted by the groundwater table outside the
tank, producing a no-flow condition at the hole. This can occur because there is a density
difference between the two fluids, even though the product and groundwater table are at different
levels relative to the hole. For gasoline fuels, the no-flow condition occurs when the ratio of the
groundwater level to product level is approximately 0.7, the ratio of the specific gravity or
density of the two fluids. If the hydrostatic pressure inside the tank is greater than that outside
the tank, product will flow out. Conversely, if the hydrostatic pressure is greater outside the
tank, then water will flow into the tank. Thus, the magnitude of the measured leak rate during
any one test might not be a good indication of the actual size of the hole in the tank or of the total
loss of product that might occur during normal operations.
The underground storage tank geometry presented in Figure 4.3 can be used to derive the
hydrostatic pressure exerted on a hole in the tank and to interpret the results of a test on a leaking
tank. The hydrostatic pressure will depend on whether the hole is located above or below the
product level, or above or below the groundwater level. When the hole is below the product
level and groundwater level (Figures 4.3(a) and 4.3(b)), the hydrostatic pressure, expressed in
terms of the product in the tank, is described by the following relationship:
= hPL-h'GWL = hPL-(hGWL/SPL) (4.5)
PL
where SPL and SOWL are the specific gravities of the product and the water, respectively, where h
and howLare the levels of the product and groundwater, respectively, and h'OWL= hGWL/SPLis the
equivalent pressure that is exerted by the groundwater in terms of product head when SGWLis
assumed to be 1.0. When h'GWL> hPL, the direction of the flow is into the tank, and when
hpL> h'owu the direction of the flow is out of the tank. A no-flow condition exists when
h'awL= "PL- When the groundwater level is below the hole (Figure 4.3(c)), the hydrostatic
pressure, expressed in terms of the product in the tank, is described by
= hPL-hL (4.6)
4-16
-------
Finally, when the groundwater is above the hole and the product level is below it (Figure 4.3
(d)), the hydrostatic pressure, expressed in terms of the groundwater, is described by
= hGWL-hL (4.7)
hL, hPL> hL, hPL> hQWL
B
GWL
hL , hPL > hL , hPL < h
GWL
Figure 4.3. UST geometry showing hydrostatic-pressure relationships when (A, B) hole is
below product and groundwater levels, (C) hole is above groundwater level and below product
level, and (D) hole is above product level and below groundwater level.
4-17
-------
IQWL
hL, hPL> hL
D
hL, hPL < hL
Figure 4.3 (concluded). UST geometry showing hydrostatic-pressure relationships when (A, B)
hole is below product and groundwater levels, (C) hole is above groundwater level and below
product level, and (D) hole is above product level and below groundwater level.
4-18
-------
As a consequence, the performance of each test method is specified relative to the product level
required to perform the test, and all flow rate measurements are referenced to this level. This
approach yields a quantitative estimate of the method's ability to detect the flow rate. If the flow
rate produced by the hydrostatic pressure is small, the leak may not be detected, even though the
hole may be large.
Some methods attempt to conduct a test at two product levels, but this approach is not
successful unless a large change in the hydrostatic pressure is induced. A 10% change in product
level will produce a change in the flow rate of only 5 to 10%; in some cases, the flow rate will
change by less than half of these percentages. In general, two-level tests have the disadvantage
of either being long or being susceptible to other errors. The flow rate as measured by a test can
not be easily interpreted in terms of the amount of product that could be lost to the environment,
since it is impossible to predict the hydrostatic pressure on the hole in the tank unless the product
level, groundwater table and hole location are known, as well as the relationship between leak
rate and hydrostatic pressure.
Consideration was given to the idea of normalizing the results of each method to a standard
product level so that the results of different methods could be compared directly. This idea was
rejected because of the uncertainty in estimating the hydrostatic pressure exerted on the hole
during an actual leak detection test and the relationship required to perform the normalization.
Two methods with the same performance that are tested at different product levels can be subject
to different flow rates from the same hole. Which method will work better depends on the
hydrostatic pressure on the hole. In general, the high-product-level tests have larger flow rates
than the low-level tests if the groundwater level is below the tank, or if the hole is below both the
groundwater and product levels. As indicated before, the latter cannot be determined during a
test because the location of the hole is generally unknown. If both the groundwater level and
product level are above the hole, it is also possible that the hydrostatic pressure produced by a
low-product-level test can be higher than that of a high-level test.
4-19
-------
5 Performance of a Test Method
Detection of leaks in underground storage tank systems is an example of the classical
statistical problem of finding a signal in a background of noise. In storage tank testing, the
signal is the product-level or product-volume changes produced by a leak, and the noise is the
sum of the product-level changes produced by the measurement system itself, by the
environment, and by the operational practice. In a properly designed system, the noise
introduced by the measurement system and the operational practices should be small compared
to the environmental or ambient noise. The measurement system noise is easily controlled and
should be designed to be smaller by at least a factor of 5 than the minimum signal to be detected.
For some methods, the operational practice may significantly affect the magnitude of the ambient
noise field. The impact of the operational practice can be minimized by proper test design.
The solution to the leak detection problem is straightforward and is accepted by the
scientific and engineering communities. The same method used to evaluate the performance of a
radar or sonar system, for example, can be (and is here) used to evaluate the performance of a
leak detection system for an underground storage tank. Numerous descriptions of the statistical
models used to analyze the data can be found in the scientific literature [e.g., 16-18].
Application of these models to underground storage tank leak detection systems was first
described in [19, 20].
The performance of a test method is presented in terms of its probability of false alarm, or
PFA (the probability that a test will result in the declaration of a leak when the tank is tight), and
its probability of detection, or PD (the probability that a test will result in the declaration of a leak
when the tank is indeed leaking), for a prescribed leak rate, LR.
The dominant sources of environmental or ambient noise are:
o product temperature (changes in product volume produced by temperature changes in
the product)
o vapor pockets (changes in the volume of trapped vapor produced by temperature
changes in that vapor due in turn to atmospheric and hydrostatic pressure changes,
and by evaporation and condensation within the trapped vapor)
o evaporation and condensation (changes in product volume produced by evaporation
from the product surface and condensation from the tank walls, and manifested as
losses or gains of product)
o structural deformation (changes in the volume of the tank produced by changes in
hydrostatic pressure on the tank prior to and/or during testing, e.g., product-level
changes)
5-1
-------
o surface waves (periodic product-level fluctuations that are unrelated to volume
changes)
o internal waves (periodic subsurface temperature and/or product-level fluctuations that
are unrelated to volume changes)
The first four sources of noise have temporal characteristics similar to those of the signal
(i.e., linear changes). However, they may also have temporal fluctuations on scales different
from those of the signal. Unless the tests are short, the temporal fluctuations of the surface and
internal waves generally can be removed by appropriately sampling and filtering the data.
The ability to detect a signal is limited by that portion of the noise energy with the same
characteristics as the signal (i.e., that portion which could be confused with the signal). For tank
leaks, the signal in its simplest form is assumed to be a linear change of product level or product
volume with time. The essential noise, therefore, is that which also leads to a linear change of
volume with time. The portion of the noise that does not exhibit a linear change with time can be
removed by averaging the data appropriately. A linear least-squares fit of volume as a function
of time is one such method. The measurement made during a test may be either noise or the
signal-plus-noise; unfortunately, it is not possible to separate the signal from the noise in a
volume measurement alone, although measurements of other quantities may provide a
separation.
Detection is usually accomplished by selecting a threshold level at the output of the
measurement system. When the output exceeds the threshold, a signal is presumed to be present.
The four possible outcomes of a leak detection test are presented in Table 5.1, where the test
declaration is given on the vertical axis and the actual state of the tank is given on the horizontal
Table 5.1. Possible Outcomes of a Leak Detection Test
Measured Conditions
Leak
No Leak
Actual Conditions
Leak
Correct Declaration
(Leak)
Incorrect Declaration
(Missed Detection)
No Leak
Incorrect Declaration
(False Alarm)
Correct Declaration
(Tight)
5-2
-------
axis. The two correct cieclaralions arc- that the tank is accurately declared to be leaking or
nonleaking. The two incorrect declarations are that a tight tank is declared to be leaking (i.e., a
false alarm) and that a leaking tank is declared to be tight (i.e., a missed detection). For
illustration, a measurement system's output for many tank tests is shown in Figure 5.1. The
output fluctuates because of the random nature of the noise. If the signal (i.e., the leak) is much
larger than the threshold, as it is for Tank C, it is not difficult to decide that the signal is present.
But consider the measurements for Tanks A and B, which are leaking at the same rate. The noise
fluctuation at A is large enough that the combination of signal-plus-noise exceeds the threshold.
At B, the noise fluctuation is negative, and the resultant signal-plus-noise does not exceed the
threshold. Thus, the presence of noise can sometimes enhance the detection of weak signals, but
it may also cause the loss of a signal that would otherwise be detected, i.e., a missed detection.
Leak Out
Threshold
Level
Measured
Leak
Rate
DO
B
O
o
o
o
C)
o
O Q O
O
A Leaking Tank
O Nonleaking Tank
Leak In
Figure 5.1. Typical measurement output for a sequence of tank tests. A and B are tanks leaking
at the threshold rate. A is declared leaking and B is declared nonleaking. Tank C is leaking at a
rate well above the threshold level, so there is no ambiguity in declaring it leaking. The open
dots represent nonleaking tanks; one of them (D) has a noise fluctuation exceeding the threshold,
producing a false alarm.
5.1 Signal
The signal is the rate of change of product level or product volume produced by a leak.
The flow rate produced by a leak is assumed to be a constant (i.e., to be a linear change in
5-3
-------
volume with time) during the tests. For tests conducted at a constant hydrostatic pressure, the
signal will be equal to the leak rate. Tests in partially filled tanks, or tests in overfilled tanks in
which the product level is kept at a constant level by adding or removing a measured volume, are
examples of constant-head tests. However, as is shown in Section 6.4.4, it is not true that the
signal is equal to the leak rate if the product level is allowed to fluctuate during a test. Tests
conducted in a tank overfilled into the fill tube or an above-grade standpipe, in which the product
level changes, are examples of tests conducted under variable hydrostatic pressure. For these
tests, the signal will be only a fraction, k, of the actual product level or product volume expected
to be produced by a given leak rate. For constant-head tests the measured leak rate, LR, is equal
to the actual leak rate, LR^,^,. For variable-head tests, LR = k LR^,^. The volume changes
caused by noise will be similarly affected, such that VR = k VR^,^, where VR,,,.,,,,,, is a volume
rate due to any noise source.
The specification of the weakest detectable signal is sometimes difficult because the
criterion for deciding whether or not the signal is present (i.e., the threshold) may be hard to
define. For example, weak signals, such as at B, would not be lost if the threshold were lower.
However, too low a threshold increases the likelihood that a noise fluctuation alone will rise
above the threshold and be mistaken for a real signal, such as the noise fluctuation at D, a false
alarm. Conversely, too high a threshold means that signals might be missed. The selection of
the proper threshold level is a compromise that depends on the relative importance of avoiding
missed detections as opposed to the importance of avoiding false alarms.
5.2 Noise Histogram
The performance of a detection system can only be determined once the fluctuation level
(product-level or product-volume changes) at the output of the measurement system is known
with and without the signal present. For any test method, the statistical fluctuation of the noise is
observed in the histogram of the volume-rate results created by plotting the measured volume
rates from a large number of tests conducted (1) over a wide range of conditions, (2) with many
systems on one or more nonleaking tanks, and (3) by many different operators. The histogram
indicates the probability that a particular volume rate will result from a test on a nonleaking tank.
The perfonnance analysis requires that the test sample consist of random, independent events. If
there are no systematic errors in the measurement, the mean will not be statistically different
from zero, and the standard deviation will reflect the uncertainty of the test method. It is usually
assumed that the data are stationary (i.e., that the histogram of the noise does not change with
5-4
-------
time) and spatially homogeneous (i.e., that the histogram of the noise obtained from one tank is
not statistically different from the histogram obtained at another tank). This is not always true,
and as a consequence, estimating performance is more complicated.
The noise histogram can be described empirically or by a probability density function
(pdf), both of which describe the fraction of the total number of occurrences of an event that
appear in a defined interval. The likelihood of exceeding a specified noise level is described by
the integral of the pdf (i.e., the area under the curve beyond the specified noise level1). Thus, if
the area beyond some noise level is 0.25, for example, the probability of exceeding that noise
level is 25%.
Frequency
fes
p
///
I
2Z
Volume Rate - L/h
Figure 5.2. Normal probability model overlaid on noise histogram for a test method with a zero
mean.
If the test method has no systematic errors, the mean of the histogram of the noise will not
be statistically different from zero. This is illustrated in Figure 5.2, where it is assumed that the
histogram follows a normal probability model. The probability of false alarm is determined from
the noise histogram once the threshold leak rate has been selected. The PFA is the portion of time
that a noise fluctuation will exceed the threshold. This is computed as the area under the
histogram beyond the threshold (Th), and is represented by the hatched area under the normal
pdf shown in Figure 5.3. Mathematically, the probability of false alarm, also known as a false
positive or a Type I error, is defined by
1. The entire area under the pdf is defined as unity.
5-5
-------
FA
JIT
f(x)dx
(5.1)
For a given method's noise histogram, the threshold and PFA are directly related. Once the
threshold has been selected, the PFA is automatically defined, and vice versa.
Frequency
Volume Rate • Uh
Figure 5.3. The probability of false alarm. The PFA is the fractional time that a noise fluctuation
will exceed the threshold (Th).
If the test method has a bias (i.e., if it is controlled by systematic errors), the histogram has
a mean displacement, as shown in Figure 5.4. If the bias is large, it will generally control the
performance of the method. A bias will result, for example, if a tank test is routinely conducted
immediately after raising the level in the tank, fill tube, or standpipe without waiting for the
product-volume changes produced by structural deformation to become small. This is
particularly true if the time constant of the tank is several hours. In this case, the noise histogram
will be shifted to negative volume rates (i.e., flow out of the tank). While the reason for the bias
is known, it is difficult to quantify it because of the large number of tank combinations, backfill
materials, and /'// siru soil conditions. If the bias cannot be quantified, the histogram may not be
representative of the noise, and an accurate estimate of performance cannot be made. Many of
the test methods evaluated in this study have a bias.
The effect of deformation on the overfilled-tank tests is illustrated in Figure 5.5. If no
deformation were occurring, the histogram would be centered around 0. The bias is produced
when the measurement is made during the steepest portion of the product-level fluctuations
produced by the deformation (rather than later, after they have flattened out).
5-6
-------
Frequency
•Th
PFA
Volume Rate - L/h
Figure 5.4. Noise histogram with a positive bias. The PFA for a positive threshold (+Th) is
larger than for a negative threshold (-Th).
Standard Deviation: 502 ml
unction
c
o
•a
1
b
1?
ol
0.12
0.10
0.08
0.06
0.04
0.02
0.00
-40
"
—
00
/\
/x
-3500
-3000 -2500
Flow Rate - ml/h
-2000
-1500
-1000
Figure 5.5. Illustration of the histogram of the bias produced by deformation-induced volume
changes. The changes were produced during an overfilled-tank test and with no waiting period.
5.2.1 Overlllled-Tank Tests With Releveling/Parlially-Filled-Tank Tests
In tests conducted with a constant hydrostatic pressure, the histogram of the noise is
different from the histogram generated from variable-head tests, even though the product
temperature and tank conditions are identical. In constant-head tests, the measured noise
histogram, compiled from the temperature-compensated volume rates on one or more nonleaking
tanks, is a good estimate of the actual noise histogram. This is not true if a test is conducted with
a variable head in a tank that is subject to deformation.
5-7
-------
5.2.2 Overfilled-Tank Tests with Variable Hydrostatic Pressure
In variable-head tests, the measured volume rate, VR, is only a fraction, k, of the actual
volume rate, VR^^. Thus, VR = k VR^^, where k = 1 for constant-head tests and 0 < k < 1 for
variable-head tests. The mean and the standard deviation of noise histograms, compiled from the
temperature-compensated volume rate for a nonleaking tank or tanks, are different for
variable-head tests than for constant-head tests. The accuracy in measuring volume changes in a
variable-head test depends on the deformation characteristics of the tank-backfill-soil system;
these are described in Section 6.4.4.
The effect on the noise histogram, as well as the signal-plus-noise histogram, is illustrated
below, assuming that (1) the characteristics of the noise histogram are described by its mean and
standard deviation, and (2) only two sources of noise are present: thermal expansion and
contraction of the product and one other source of noise. This mean rate of change of volume is
defined by
(5.2)
where a bar over a quantity indicates a mean, where TVR is the rate of change of volume
produced by thermal expansion or contraction of the product, and where AVR is the rate of
change of volume produced by any other (or all other) source(s) of noise. The standard
deviation, S, is given by
— \ ^TVR + ^>AVR (5.3)
where S-J-VR and SAVR are the standard deviations of the thermally induced volume changes and of
other noise sources, respectively. The histogram of the thermally induced volume changes
(estimated from temperature measurements) and of the measured volume changes, assuming that
thermal changes are the only source of noise, is shown in Figure 5.6.
The histogram of the thermal changes (i.e., TVR) was simulated by assuming a normal
distribution with TVR = 0 and STVR = 850 ml/h, similar to the empirical histogram used in the
performance estimates of the commercial methods. The mean of both histograms is zero since
the mean of the TVR histogram is zero.
In principle, temperature compensation should reduce the standard deviation of the
measured VR noise histogram and not change the mean. The mean and standard deviation of the
temperature-compensated volume rate, TCVR, are given by
= VR-TVR = (k-l)TVR+kAVR (5.4)
5-8
-------
and
S-rrvR= V STVR + SAVR — STVR (5.5)
For constant-head tests, k = 1, and STCVR= SAVR, but this is not true when deformation is
important and k < 1 . The temperature-compensated volume rate histogram for k = 0.6, assuming
no other sources of noise, is shown in Figure 5.6.
5.3 Signal-Plus-Noise Histogram
In an actual test, the signal cannot be measured independently of the noise.
5.3.1 Overfillecl-Tank Tests with Releveling/Partially-Filled-Tank Tests
For tests conducted with a constant hydrostatic pressure, the signal is assumed to be equal
to the leak rate, and to be constant, independent, and additive with the noise. For a deterministic
signal, the standard deviation of the signal-plus-noise is equal to the standard deviation of the
noise. As a consequence, the performance can be estimated without making detailed
measurements of the signal-plus-noise. The signal-plus-noise histogram has the same shape as
the histogram of the noise but is displaced by the leak rate (e.g., Figure 5.7).
The magnitude of the leak signal is a function of the hydrostatic pressure on the hole
(see Section 4). The level of the ground water affects the size of the signal, but unless this level
changes during the test, the noise is not affected. Changes in the groundwater level might be a
problem during a heavy rainfall just before or during a test in a highly permeable soil or close to
sea level near a tidal shore. However, during overfilled-tank tests in which the product level is
not kept approximately constant, and the deformation is large, the groundwater level may indeed
affect the size of the noise field as well.
5.3.2 OveiTilled-Tank Tests with Variable Pressure
If a test is conducted with a variable head, the signal-plus-noise histogram must account
for the deformation-induced changes. The volume change due to a constant leak rate affects only
the histogram means derived in the analysis in Section 5.2.2. Eqs. (5.2) and (5.4), with a leak
rate, LR, included, become
VR = k(LR+TVR + AVR) (5.6)
TCVR = kLR + (k-l)TVR+kAVR (5.7)
5-9
-------
0.10
o
c
0.05 -
0
0.10
0.05 -
0.12
0.06
Mean = 0 ml/h
Std. Dev. . 850 ml/h.
Temperature Volume Rate - ml/h
Mean = 0 ml/h
Std. Dev. = 472 ml/h
iiiiij
iitj
1111
Volume Rate - ml/h
Mean = 0 ml/h
Std. Dev. = 375 ml/h
-3,000 -2.000 -1.000 0 1.000 2.000 3.000
Temperature-Compensated Volume Rate - ml/h
Figure 5.6. (A) TVR histogram estimated from product temperature measurements (the same
for k = 1); (B) VR histogram for k = 0.6 without temperature compensation; and (C) VR
histogram for k = 0.6 with temperature compensation. It is assumed that the only source of noise
is the thermal expansion and contraction of the product, and that the mean of the thermal
fluctuations is zero.
5-10
-------
0.10
o
c
0)
13
CT
Q)
0.05 -
Mean = -674 ml/h
Std. Dev. = 473 ml/h
0 '
-3.000 -2,000 -1.000 0 1.000 2.000 3.000
Volume Rate - ml/h
Figure 5.7. Comparison of the signal-plus-noise histogram for a -1.2-L/h leak rate and the noise
histogram under the same product temperature conditions after temperature compensation. It is
assumed that the only source of noise is thermal expansion or contraction of the product.
0.12 -
O
C
§j 0.06 -
CT
Q)
LL
n -
! ;;
n" ' '
J::
nfTfH
Mean = -671 ml/h
Std. Dev. = 375 m!7h
' '[]
•
• • - •
]
•••• tfln T
-3.000 -2.000 -1.000 0 1.000 2.000 3.000
Temperature-Compensated Volume Rate - ml/h
Figure 5.8. Comparison of the signal-plus-noise histogram for a -1.2-L/h leak rate and the noise
histogram under the same product temperature conditions before temperature compensation. It is
assumed that the only source of noise is thermal expansion or contraction of the product.
5-11
-------
An example of the signal-plus-noise histogram for a -1.2-L/h leak with k = 0.6
(variable-head test), assuming the entire contribution to the noise is from thermally induced
product-volume changes, is presented, before and after thermal compensation, in Figures 5.7 and
5.8. If the noise has a nonzero mean, the mean of the signal-plus-noise histograms will be
different before and after temperature compensation.
For comparison, the histogram of the signal-plus-noise with k = 1.0 (constant-headtest)
after temperature compensation would be identical to the leak rate (i.e., TCVR = LR and
STCVR = 0)- Before temperature compensation, TCVR = LR and STCVR = SJVR.
5.4 Performance Model
The performance of a volumetric test method is determined from the histograms of the
signal-plus-noise and the noise. For tests conducted under a constant hydrostatic pressure (e.g.,
partially-filled tank tests, or overfilled-tank tests which relevel), performance is estimated from
the model shown in Figure 5.9. This model assumes that the data are stationary and spatially
homogeneous, that the noise histogram has a zero mean, and that the signal is constant, equal to
the leak rate, independent and additive with the noise. This model applies to all volumetric tests,
providing that the systematic errors are small.
Noise
(Non-Leaking Tank)
Signal-Plus-Noise
(Leaking Tank)
Volume Rate - L/h
Figure 5.9. Statistical model to estimate the accuracy of a constant-head leak detection system.
5-12
-------
Figure 5.9 shows the PD( PFA, Th, and the leak rate, LR, for the constant-head test (k = 1).
Once Th has been selected, the PFA is determined. The PD is defined as the fractional time the
signal-plus-noise fluctuation will exceed the threshold, and is represented by the hatched area.
Clearly, if the threshold does not change, the PD will be higher for larger leak rates. For tests
conducted in an overfilled tank in which the product level fluctuates during the test, performance
is estimated from the model shown in Figure 5.10. In this case, the leak rate determined by the
signal-plus-noise histogram is not equal to the actual leak rate and the temperature-compensated
noise histogram has a larger spread than it does for a constant-head test. As discussed in
Sections 5.2.2 and 5.3.2, the noise and signal-plus-noise histograms, after temperature
compensation, are impacted by the structural deformation of the tank. Since the noise histogram
derived after temperature compensation has a larger standard deviation for variable-head tests
(k < 1) than for constant-head tests (k = 1), the PDwill decrease and the PFAwill increase in
comparison.
It is important to understand that the PD,PFA,Th, and LR are all interrelated; changing one
parameter affects the value of one or more of the other parameters. If any two parameters are
known, the others are fixed by the model. The choice of parameters affects the conclusions to be
drawn from leak detection tests (i.e., the reliability of the test result). For a given leak, LR,
choosing a high threshold, Th, gives an extremely small PFAbut results in a reduced PD(i.e., the
number of false declarations will be small, but the number of missed leaking tanks will be large).
A low value for Th yields extremely large values for both the PD, and PFA (i.e., the number of
leaking tanks detected will be large, but the number of false declarations will also be large). The
benefits and costs associated with reducing the number of missed detections and false alarms can
be balanced by judicious selection of the threshold.
The standard deviation of the noise and the signal-plus-noise is a measure of the spread of
the data and is directly proportional to performance (e.g., Figure 5.9). The smaller the standard
deviation, the better the performance. The performance can be directly calculated from the
standard deviation if (1) the noise is stationary, spatially homogeneous, and normally distributed
with a zero mean, and (2) the signal is constant, equal to the leak rate, independent, and additive
with the noise. For such conditions, the leak rate that can be detected with a PD = 0.99 and a
PFA= 0.01 is 4.67 times the standard deviation. The threshold, established by the selection of a
PFA = 0.01, is 2.33 times the standard deviation. In this particular instance, the threshold is one
half the leak rate because the PFA plus the PD is equal to 1. (However, this is not true in general.)
Thus, a standard deviation of 0.163 L/h (0.043 gal/h) will result in the detection of a 0.76-L/h
5-13
-------
Signal-Plus-Noise
PD
Volume Rate - L/h
Figure 5.10. Statistical model to estimate the performance of a variable-head test. The same
conditions apply for this model as for the constant-head model in Figure 5.9.
(0.20-gal/h) leak rate with a PD= 0.99 and a PFA= 0.01 for a threshold of ± 0.38 L/h (0.10 gal/h).
The PD is greater than 0.99 for leak rates of more than 0.76 L/h and smaller than 0.99 for leak
rates less than 0.76 L/h.
In most of the methods evaluated here, the threshold has been set so as to be equal either to
(1) the smallest leak rate detectable by the system or (2) the leak rate specified by regulatory
policy or industrial practice. When the threshold is equal to the leak rate (common practice in
the industry), the probability that a leak will be detected is only 50%. Typically, the methods
evaluated here claim a high performance against leak rates of 0.19 L/h (0.05 gal/h), and yet they
use a 0.19-L/h threshold to declare a leak. The threshold would have to be set to a value less
than the minimum detectable leak rate in order for the probability of detection to be better than
50%.
Measurements of the ambient noise associated with underground storage tanks show
fluctuations much larger than the smallest leaks to be detected (i.e., less than 0.76 L/h
(0.20 gal/h)). In order to achieve satisfactory detection performance, the ambient noise
fluctuation must be substantially reduced. The two approaches to noise reduction are
(1) incoherent averaging, and (2) coherent cancellation or compensation. The goal of both
approaches is to reduce the mean and standard deviation of the noise to zero.
Incoherent averaging reduces noise only if successive measurements contain substantially
different noise fluctuations. The need to make many measurements spaced many hours apart
makes incoherent averaging impractical as an approach to noise reduction.
5-14
-------
Coherent noise cancellation can be effective and efficient, but its use requires an
understanding (i.e., models) of the principal sources of noise as well as measurements of these
sources of noise. In this process, the auxiliary measurements of the noise source are input to the
model to predict volume changes which can then be removed from the measured volume change.
The use of temperature sensors to estimate the thermal expansion or contraction of the product
volume is an example of such a measurement. If the model is not known accurately, the
procedure can be made data-adaptive, with the data themselves determining how the auxiliary
measurements are to be manipulated before removal. However, this must be done with great
care lest the data-adaptive procedure remove the signal along with the noise. Coherent noise
cancellation, whether predictive or data-adaptive, can be accomplished readily with real-time
algorithms and microprocessor hardware. Both the accuracy and adequacy of the noise-source
measurements detennine the effectiveness of the process; very large noise reductions are
possible.
5.5 Multiple-Test Strategy
A properly designed multiple-test strategy can Improve the performance of a volumetric
test method in terms of PD, PFA, or both, over the performance obtained for a single test. The
actual improvement depends on how many tests are conducted, how the data are analyzed, and
whether the tests are independent. In the methods evaluated in this study, two types of analysis
were used to implement a multiple-test strategy. Both attempt to reduce the statistical fluctuation
of the histogram of the noise. The more common uses some form of an m-out-of-n approach,
where n is the total number of tests and m is the number of times that the threshold is exceeded.
The other calls for averaging the data from two or more tests before applying the detection
criteria. If the multiple-test strategy is not properly designed, the performance achieved can be
drastically different from what is expected or desired.
Substantial benefits can be achieved if the tests are independent and random. Because
most multiple-testing strategies require that additional data be collected at the end or as part of a
test, the results are not independent. Studies have shown that the statistical uncertainty obtained
in a single test is nor equal to the statistical uncertainty obtained from many different tests [8].
Uncompensated sources of noise result in systematic trends in the data which are not reduced by
averaging. For example, if the temperature is only partially compensated for, the measured
volume rate in a nonleaking tank will contain the effects of thermal expansion or contraction of
the product in the tank. Two tests that are performed in close enough succession that the
temperature field has not changed would result in approximately the same measured value. The
second test would contain the same temperature-induced error as the first. No substantial
5-15
-------
benefits are gained from performing multiple tests if the errors are systematic and are large
enough to dominate the random error. Multiple testing reduces only the random error, not the
systematic error. Design of a multiple-testing strategy requires a thorough understanding of the
testing system and the ambient noise associated with it.
The simplest multiple-test approach and one of the best is to average the measured volume
rate results from one or more independent tests and use the average to determine whether the
tank is leaking. Averaging will reduce the spread in the histogram of the data by the square root
of the number of independent tests. If the tests are not independent (i.e., if the tests are
correlated because of a systematic error), the reduction of the standard deviation will be less than
the square root of the number of tests. Averaging will improve both the PD and the PFAover
those obtained with a single test. Substantial improvements usually require that a large number
of tests be conducted, even if the tests are independent.
Another tactic is to use an m-out-of-n approach. In this approach, two or more tests are
conducted and a leak is declared if one, several, or all of the tests exceed the threshold. The
m-out-of-n detection approach is more mathematically complex than the averaging approach, but
is as easy to use. If the number of tests conducted is large, the m-out-of-n approach can result in
substantial improvements in both the PD and the PFA If the number of tests is small,
improvement may be seen in either the Pu or the PFA, but not in both. An increase in the PD over
that obtained with a single test can be achieved, but at the expense of an increase in the PFA.
Likewise, the PFA can be reduced below that obtained with a single test, but usually at the
expense of a lower PD.
The averaging and the m-out-of-n approaches are compared in Table 5.2 for a multiple-test
strategy using two tests. In this example, it is assumed that the noise histogram compiled from
many independent tests is normal, with a mean of zero and a standard deviation of S. Table 5.2
presents the the results of a detection system against a leak rate that is 2.33 times larger than the
standard deviation by conducting two independent tests and declaring a leak if one, two or the
average of the two tests exceeds the threshold. The threshold used to declare a leak is presented
in terms of the standard deviation of the probability density function. The three thresholds
selected result in a probability of false alarm of 0.1, 0.05, and 0.01 for a single test. For a
standard deviation of S = 0.19 L/h (0.05 gal/h), for example, Table 5.2 gives the performance
achieved by the three approaches against a 0.45-L/h (0.12-gal/h) leak rate using thresholds of
0.24, 0.31, and 0.45 L/h (0.064, 0.092, and 0.12 gal/h), respectively. The averaging improves
both the probability of false alarm and probability of detection over that of a single test. The
probability of detection is increased (i.e., it is greater than the PD obtained with a single test, or
that obtained by averaging two tests when a leak is declared as a result of either test exceeding
5-16
-------
the threshold), but at the expense of an increased probability of false alarm. Likewise, the
probability of false alarm is less than the PFA obtained with either a single test or by averaging
two tests when both tests must exceed the threshold to declare a leak, but at the expense of a
reduced probability of detection.
Table 5.2. Multiple-Test Performance Calculations for Two Tests*
Threshold
Single Test
(m = n= 1)
PFA PD
m-out-of-n
m = 1 , n = 2
PFA PD
m-out-of-n
m = 2, n = 2
PFA PD
Average
of Two Tests
PFA PD
1.28 S 0.10 0.85 0.19 0.98 0.01 0.72 0.04 0.93
1.64 S 0.05 0.75 0.10 0.94 0.003 0.56 0.01 0.83
2.33 S 0.01 0.50 0.02 0.75 0.0001 0.25 0.0005 0.50
The results are expressed in terms of the standard deviation S. The leak rate is 2.33 S.
(Source: [21])
The m-out-of-n testing approach is less efficient than the averaging approach; more tests
are required to achieve the same performance. While the m-out-of-n detection strategy is not as
efficient as the averaging approach, one anomalously large test result will have less of an impact
with the m-out-of-n approach than with the averaging approach. Averaging four independent
tests will result in the same detection performance against a leak rate that is one half as large as
one that can be detected in a single test. For a probability of detection of 0.99 and a probability
of false alarm of 0.01, a test criterion which requires that the threshold be exceeded in three out
of five tests reduces the standard deviation by a factor of 0.54.
5.6 Summary
The performance of a leak detection system governed by the performance model
(Figures 5.9 and 5.10) can be determined from the histogram of the noise and from the
quantitative relationship between the signal and the noise. The model assumptions about the
signal and signal-plus-noise histograms have been verified using simulated leaks at the Test
Apparatus. Using the commercial method's measurement system and following its prescribed
test protocol, the histogram of its noise can be determined by compiling the volume rate results
from a large number of tests in nonleaking underground storage tank systems over the range of
conditions to which the leak detection system will be applied. The analysis was done not
assuming a normal model. The measured data were used to estimate performance. Instead of
5-17
-------
repeating this extensive series of tests for each method, the data were collected once to create a
temperature database. Then the response of each commercial method was determined using the
same temperature database.
Throughout the report, performance is discussed in terms of the standard deviation of the
noise. A large standard deviation reflects poor performance. This description was used for
convenience to avoid displaying histograms and performance curves. It is particularly
meaningful if the noise histogram is normally distributed.
5-18
-------
6 Performance Simulation
In order to estimate the performance of a test method, it is necessary to generate
histograms of noise and signal-plus-noise; these depend on the elements of the test method. For
detection of leaks in underground storage tanks, the performance of a volumetric method can be
estimated once the noise histogram has been characterized. This is possible because the signal
can be assumed to be an additive linear change in volume with time, and because the
product-level changes produced in a tank by the signal are predictable for both variable- and
constant-hydrostatic pressure volumetric tests. Characterizing the noise histogram requires a
substantial data collection effort. The noise is derived from the temperature-compensated
volume rate estimates made from a volumetric test over a wide range of ambient test conditions.
A simulator, which mimics a test in an underground storage tank, was developed to
estimate the performance of each volumetric test method against a given set of tank and
temperature conditions. The performance of a test method was evaluated by repeatedly
simulating the conduct of a tank test in order to develop a histogram of the noise. When
necessary (i.e., when data-dependent analysis and detection criteria were employed to declare a
leak), the simulator was also used to develop a signal-plus-noise histogram.
To estimate performance, the simulation used an empirically derived database of
product-level and temperature data developed for a 30,000-L (8,000-gal) underground storage
tank backfilled with pea gravel and filled with a blend of unleaded gasoline. The test method
was modeled in terms of the precision and accuracy of the instrumentation, test protocol (data
collection procedure), data analysis, and detection criterion. The approach to temperature
compensation was incorporated in the test protocol and data analysis. Many tests were then
simulated, using the empirically derived database and the test method model to develop a noise
and a signal-plus-noise histogram.
The performance simulation (summarized in Figure 6.1) has eight basic elements:
o Experimental Data
o Database Management
o Temperature Database
o Product-Level Database
o Test-Method Model
o Tank-Test Simulator
o Noise and Signal-Plus-Noise Histograms
o Performance Analysis
6-1
-------
Test Method
Model
^
Experimental Conditions
1
Ambient
Noise
I
Product Level
Database
1
I
Temperature
Database
Simulator
i
Database Manager
Histograms
zn
Performance
Estimate
Figure 6.1. Flowchart of the performance simulation.
6.1 Experimental Data
Experimental data were used to generate a representative set of environmental conditions
found in underground storage tanks. The temperature database was developed directly from a
special set of experiments designed to emulate the delivery of product to a tank. The
product-level database was developed from models of the ambient noise, the tank-backfill-soil
conditions, and the instrumentation, measurement configuration, and protocol of the method.
The models of the product-level or product-volume changes produced by each source of ambient
noise that were used in the simulation were developed and validated with the data collected in a
special set of ambient-noise experiments. The data were stored and manipulated using the
database manager. The performance simulator specifies the test condition and accesses both the
database manager and the test-method model to simulate a test. Histograms of the noise and
signal-plus-noise were derived from the temperature-compensated volume-rate test results
simulated for many tests and used to analyze performance.
6-2
-------
The accuracy of the performance estimates is dependent on the accuracy of the test-method
model, the accuracy of the product-level and temperature databases used to drive the simulation,
and the number of tests used to derive the noise and signal-plus-noise histograms. Except for
those methods which are affected by the ambient environment outside the tank, or which
vertically integrate temperature using non-point-source sensors, the test-method model is simply
a logical description of the instrumentation, test protocol, data analysis, and detection criterion.
The temperature database was measured with thermistors that are calibrated to within 0.001 °C.
The product-level database was generated separately for each method using the applicable noise
models, which predict the product-level or volume fluctuations in the tank. The ambient noise
models used to predict product-level changes were validated during the ambient-noise
experiments conducted at the Test Apparatus; the test-method model and, in particular, the
temperature measurement approaches used by each method were validated during the Field
Verification Tests.
6.2 Database Management
The temperature and product-level data were organized by means of a database
management program that was used to store and manipulate data files for each individual
tank-test condition. This database management program creates a separate binary file for each
test run. The binary formatted run files provide efficient storage and rapid access.
Each binary run file, which has up to 64 channels of data, consists of a header record
followed by the data records. Information describing the test run was recorded in the header,
which also contains channel identification information describing the data type. The majority of
the runs were 24 h in duration. The experimental data (e.g., product temperature), in engineering
units, were added to the run file from ASCII LOTUS 1-2-3 importable files. The product-level
and product-volume files were created in the program once the necessary experimental data had
been added (e.g., temperature data are required in order to estimate the thermally induced
volume changes). The test-method model was used to specify the geometry and tank
deformation characteristics.
A variety of special utilities was developed for channel manipulation, perusal, and
analysis. These utilities permit new channels to be created as functions of existing channels
within the run file. They permit the data to be graphically displayed in either the time or
frequency domain. In addition, it is possible to use these utilities to generate the data necessary
for the evaluation or to conduct a special-purpose analysis. The noise-model realization software
6-3
-------
modules are a special class of the data manipulation routines. These modules are used to create
the product-level or product-volume changes from the ambient noise models and the test-method
model.
A database of product-level changes encountered during a wide range of test conditions
was developed from the ambient noise models. The magnitude of the product-level fluctuations
encountered during a tank test depends on several factors: the transfer of heat between the air,
the ground, and the tank; the level and volume of product in the tank; the change in level of the
product or groundwater immediately before or during the test; the elasticity and time constants of
the backfill-soil surrounding the tank; and the volume of the vapor pockets trapped in the tank.
6.3 Temperature Database
To generate a useful temperature database, the volume fluctuations caused by thermal
expansion of the product must be representative of the annual range of product-temperature
changes that occur in the United States. As there was no existing information, a study was
undertaken to estimate the range of temperature conditions that would be encountered throughout
the 50 states [11]. It was determined from this study that the difference between the temperature
of product to be added to a tank and the temperature of product in a tank is randomly distributed
by geographical location, and that the range of temperature differences is between +10°C.
Two temperature conditions were developed, one corresponding to tests that require the
delivery of product to the tank and allow tune for the temperature to come into equilibrium with
the surrounding backfill, and a second one for tests that mix or circulate the product in the tank.
This latter database was developed using a Petro The circulation pump, and was used to evaluate
the Petro Tite method. Both conditions were developed by adding, to a half-filled 30,000-L tank,
approximately 15,000 L of product that was either warmer or cooler than the resident product
and the surrounding backfill. With each simulated delivery, the temperature field was measured
over a period of 24 to 48 h, and temperature data were collected for the range of temperature
differences of+10°C between new and existing product. The temperature database that was used
in the evaluations was selected so as to give a normal distribution of the rate of change of volume
over a 1-h period. The nominal conditions used to develop the temperature database are
summarized in Table 6.1, and the 1-h thermally induced volume changes are shown in
Figure 6.2.
6-4
-------
Table 6.1. Conditions Used to Create Temperature Database
for the Steel Tank
Dule
9/13/86
9/20/86
9/22/86
9/27/86
10/09/86
10/10/86
10/11/86
10/13/86
10/14/86
10/17/86
10/18/86
10/19/86
10/25/86
10/28/86
10/29/86
11/01/86
1 1/02/86
1 1/04/86
1 1/05/86
1 1/08/86
1 1/30/86
1/23/87
4/24/87
7/27/87
7/28/87
7/29/87
7/30/87
A Temperature (°C)
Ambient
8.3
2.8
-5.6
-11.1
Ambient
5.6
-2.8
Ambient
5.6
Ambient
2.8
-3.3
-5.6
Ambient
5.6
Ambient
8.0
Ambient
2.8
Ambient
Ambient
Ambient
-8.4
Ambient
-2.7
-2.2
The database contains over 500 h of temperature and product-level data. The temperature
conditions identified as ambient denote that data collection was started a minimum of 24 h after
the last product addition. The standard deviation of the data in Figure 6.2 is 848 ml/h, or
0.023°C/h in a 30,000-L tank. The temperature changes can be significantly larger in methods
6-5
-------
0.16
c 0.14
o
£ 0.12
c
£ 0.10
c
o
'•5 0.08
:G
^ 0.06
b
£• 0.04
:=
jo 0.02
o
—
—
—
.„.».£*
i o.oo
-3000 -2000
\
n
x/
v
X
v
X
v
\ .
v
\
\ .
7-
><
J>
\
/
^7
X
x'
-1000
7\
-A-
/x
\-
^
x ^
\^
/ •'
•t~
V
<
^
«,
V
\
\x
<
x
s
x"
/
x
X,
x
x^
>x
>
•-a
y
X
0
~s
x
v
-7*
s<
s<
X
><
>
<
<
>
J>
s
X
<
\/
,A.
X
X
—
X
-------
Table 6.2. Nominal Run Conditions in the Circulated Product Temperature Database
Date
5/10/87
5/1 1/87
5/11/87
5/12/87
5/10/87
5/11/87
5/11/87
5/12/87
5/14/87
5/15/87
5/16/87
5/18/87
5/19/87
5/20/87
6/23/87
6/24/87
6/25/87
6/26/87
6/27/87
6/28/87
6/30/87
6/30/87
7/1/87
7/2/87
7/2/87
7/3/87
Tank
Steel
Steel
Steel
Steel
Fiberglass
Fiberglass
Fiberglass
Fiberglass
Fiberglass
Fiberglass
Fiberglass
Steel
Steel
Steel
Fiberglass
Fiberglass
Fiberglass
Fiberglass
Fiberglass
Fiberglass
Steel
Steel
Steel
Steel
Steel
Steel
A Temperature (°C)
Ambient
-4.6
8.2
-4.6
Ambient
-3.6
7.2
-2.9
-14.4
Ambient
7.78
Ambient
Ambient
Ambient
Ambient
-2.78
2.78
Ambient
2.78
2.78
2.78
-2.78
-5.56
5.56
-2.78
2.78
6.4 Product-Level Database
Product-level fluctuations depend on the ambient conditions in the tank, the
tank-backfill-soil characteristics, and the test method. Using the test-method model and the
ambient noise models, a product-level database was developed corresponding to each method.
The test-method model specifies the system noise, the location and geometry of the product-level
6-7
-------
container, and the test protocol. The tank conditions were specified by experimental data. The
ambient noise models describe the induced product-level fluctuations corresponding to each test
method.
The physics of the ambient noise models are readily described mathematically. Because
the product-temperature fluctuations that drive the models are difficult to predict theoretically,
product-temperature fluctuations were measured.
It is possible to estimate the values of the tank-backfill-soil elasticity characteristics from
the Test Apparatus data that is necessary to determine the effects of structural deformation. It is
also straightforward to estimate trapped vapor pocket volume changes produced by pressure and
temperature fluctuations when the volume of trapped vapor is known. Unfortunately, the range
and distribution of values of tank elasticity and vapor pocket volume for installed tanks in the
United States are not known.
The ambient noise models described below were used to estimate product-level and
product-volume changes for calculation in the simulation. All of these models were validated
with data collected during the ambient noise experiments at the UST Test Apparatus. Three of
the sources of ambient noise result in product-volume changes: thermal expansion or contraction
of the product, expansion or contraction of the trapped vapor, and surface losses or gains of
product through evaporation and condensation. Changes in the volume of the container are
described by structural deformation. Finally, surface and internal waves can induce estimation
errors. The following sections describe the models developed for the simulation.
6.4.1 Thermal Expansion and Contraction of the Product
Those volume changes that occur as a result of thermal expansion and contraction of the
product can be calculated as the sum of volume changes produced in each individual thermistor
cell /', i.e.,
AV = IAV,.= ICV.AT, (6.1)
where AV is the total change in volume caused by temperature changes; AV, is the volume
change experienced in cell /; C is the coefficient of thermal expansion for the product; V, is the
volume of product in cell;'; AT, is the change in temperature in cell /; and n is the number of
cells.
6-8
-------
The temperature-volume time series is calculated using the data from three vertical
thermistor arrays. The thermistors were spaced 20.3 cm (8 in.) apart, the lowest thermistor being
12.7 cm (5 in.) from the tank bottom. A thermistor cell is taken to be the longitudinal slab
starting 10.2 cm (4 in.) below each thermistor and extending 10.2 cm above it. Of course, the
lowest slab starts at the bottom of the tank, and the topmost slab ends at the gasoline surface.
The change in volume for each slab was calculated by multiplying the first difference of the
thermistor time series (or of the average of the three thermistors located at this slab level) by the
coefficient of thermal expansion and by the volume of the slab. The temperature volume is thus
represented by the sum of the changes in volume of all of the thermal cells.
The volume of a cell was computed by multiplying the length of the tank by the
cross-sectional area between the top and bottom of the cell. By solving the integral equation
representing this area, it can be shown that the formula for finding the area that extends from the
bottom of the tank to a level, d, is:
Area = cV2dr-d2+2[arcsin[- + -)Jr2 (6.2)
where Area is the cross-sectional area below the product level, r is the radius of the tank, and
c = d - r. Eq. (6.2) was supplemented by empirical measurements in the upper 15 cm of the Test
Apparatus tanks.
6.4.2 Trapped Vapor
Trapped vapor can be a significant source of error in a volumetric test conducted in an
overfilled tank. Air and vapor can be trapped in pipes, manways, and at the top of the tank,
either because the tank is tilted or because it has an uneven top surface. The expansion and
contraction of this trapped vapor, which are induced by changes in temperature, product level, or
atmospheric pressure, will, in turn, induce changes in the product level. An estimate of the
volume changes is made assuming that the air trapped by overfilling the tank is saturated with
vapor. If dry air is mixed with product vapor, large pressure changes, and therefore large volume
changes, will occur until the partial pressure of the trapped vapor mixture has reached a saturated
state for the pressure and temperature in the vapor space. Because the partial pressure of
gasoline is much larger than the partial pressure of air, even a small volume of trapped vapor can
produce a volume change which is larger than the smallest leaks to be detected.
6-9
-------
6.4.2.1 Expansion and Contraction of Trapped Vapor
To calculate the changes in volume of trapped vapor as a function of the temperature and
pressure in the pocket, the air partial pressure, PA, and the vapor partial pressure, Pv, are
calculated as
VVP
(6.3)
Pv = P0e kT (6.4)
where N is the number of molecules, k is Boltzmann's constant, and E is the thermal energy.
Then, the total pressure, PT, is given by
(6.5)
For small changes of PT, Eq. (6.5) gives
lk~TM ^6'6^
After mathematical manipulations,
APT_( pv( E | AT ( PV)AVVP
p~ = I1+p~\VT~ [ I~T~~r~p~ ~v (6J)
IT ^ r T ^ K1 Jj 1 \ r-j- I V vp
The CRC Handbook of Chemistry and Physics [22] gives PV(T) for various C7H,6and CRH18
compounds. For T = 20°C, Pv is approximately 25 mm, and thus, for an extreme range of
f -4500 ±300° K,
k
AV VP AT APT
—— = (1.52±.03) —-1.03—- (6.8)
v yp 1 *"T
Eq. (6.8), then, describes the vapor pocket volume changes in terms of temperature and pressure
for typical conditions at the UST Test Apparatus. Eq. (6.8) is not valid if the vapor is not
saturated.
These same results can be obtained from the pressure volume and temperature volume for
a perfect-gas polytropic process defined by
6-10
-------
(6-9)
2
and
T. fv.Y-'
(6.10)
where V, is the initial volume of the trapped vapor; V2 is the final volume; P, is the initial
absolute pressure and P2 is the final absolute pressure; and T, is the initial temperature in °K and
T2 is the final temperature in °K. For a constant-temperature process (i.e., n = 1) Eq. (6.9)
reduces to the perfect gas law. For a constant-pressure process (i.e., n = 0), Eq. (6.10) also
reduces to the perfect gas law. The value of n is 1.4 for a reversible adiabatic change of a
diatomic gas.
It is safe to assume that all overfilled tanks trap vapor. Eq. (6.8) was used to estimate the
amount of vapor that will prevent the detection of a 0.19-L/h (0.05-gal/h) leak with a PD = 0.99
and aPFA = 0.01. To achieve detection performance requires that the change in volume of this
vapor be less than 0.04 L/h (0.01 gal/h). Most temperature changes in an underground tank are
less than O.rC/h. A 72-L (19-gal) vapor pocket would be required to produce a 0.04-L/h
volume change. If the temperature changes are influenced by the ambient air (e.g., vapor trapped
in an underground pipe located near the surface), vapor pockets that are much smaller in size can
degrade detection performance because much larger temperature changes may occur. An
atmospheric pressure change of 1 mb/h will produce a 0.04-L/h (0.01-gal) volume change in a
37.2-L (9.8-gal) vapor pocket. The atmospheric pressure calculations suggest that tests in
overfilled tanks should not be conducted if the atmospheric pressure is changing rapidly.
Against a 0.38-L/h (0.10-gal/h) leak rate, the allowable volume of trapped vapor would be twice
as large as these estimates.
Small product-level changes in a fill tube or a standpipe will also produce large volume
changes. A 2.5-cm/h (l-in./h) product-level change, which results in a 0.21-L/h (0.054-gal/h)
change in volume in a tank that does not deform, will produce a change of 0.04 L/h if the volume
of trapped vapor is 15.4-L (4.1 gal). For comparison, a 2.5-cm/h product-level change
corresponds to a product-temperature change of only 0.005°C/h in a 30,000-L tank. Fortunately,
these product-level-induced pressure changes can be entirely compensated for by using an
experimentally measured height-to-volume conversion factor, A^, to interpret the product-level
changes.
6-11
-------
Vapor pocket volumes should be minimized. For reliable detection of a 0.19-L/h leak rate,
it is recommended that the volume of trapped vapor be less than 50 L. Even an amount much
smaller than 50 L can have significant consequences if tests are conducted before the partial
pressure of the vapor has come into equilibrium with the tank environment.
6.4.2.2 Vapor Pocket Volume
Because the volume and location of the trapped vapor are usually unknown, it is nearly
impossible to compensate for or even to estimate accurately the magnitude of the product-level
(volume) changes produced by expansion and contraction of trapped vapor. Even if the location
and the volume of the trapped vapor were known, it would be difficult to measure accurately the
temperature and/or pressure of the trapped vapor. What is needed, then, instead of
compensation, is a method of estimating the volume of trapped vapor. In this way, if the volume
of trapped vapor were determined to be too large, an effort could be made to remove it, or a
decision made not to conduct a test.
For a well-bled, overfilled tank containing small amounts of trapped vapor, Eq. (6.9)
reduces to
fP.+n
-------
6.4.3.1 Equilibrium Conditions
The equilibrium condition exists when the partial pressure of gasoline in the air space
equals the vapor pressure for the temperature at that point. The temperature is allowed to vary,
though the prospect of maintaining the equilibrium condition with variable temperature is
somewhat unrealistic. In practice, equilibrium will only be achieved if the saturated air is in
close proximity to liquid at the same temperature, i.e., at the product surface or on tank walls wet
with condensate. Recall also that gasoline is not a single compound, but a mixture of
compounds.
First, consider the total volume of gasoline held in saturated vapor, which is one of the
major factors in achieving the equilibrium condition. (Throughout this discussion, the quantity
of gasoline is measured in equivalent condensed liquid volume, even though the gasoline may be
in a gaseous state.) The density of gasoline in saturated vapor is derived from the gas law
PV = NkT (6.13)
The density is
p = ~t- (6.14)
P A0V
where \i - molecular weight = 114 g/mole and A0 = 6.02x10" /mole. Substituting Eqs. (6.4) and
(6.13) into Eq. (6.14) gives
2.13 xlO5 -^ , ,
= - ^ - f g/ml
for E/k = 4500'K and P0 = 1.55x10'" N/m:, which are mean values for several C7H16 and CgH,g
compounds for which PV(T) data are given in the CRC Handbook [22]. This compares to a
liquid density p, = 0,746 g/ml.
To calculate the density of saturated air, we must add the unsaturated-air density.
Assuming the total pressure remains constant at PA = 1 atm = 1 .OlxlO5 N/m2,
(6-16)
where (IA = 29 g/mole. Table 6.3 gives these values for a range of temperatures.
6-13
-------
For typical temperatures, saturated air contains 0.1 to 0.3 L of gasoline for each 1,000 L of
air. A representative condition of 15,000 L (4,000 gal) of air (a half-filled tank) corresponds to
roughly 3.8 L (1 gal) of liquid in the vapor space. More importantly, this value changes by about
0.01 L/°C for each 1,000 L of air. The vapor space temperature is assumed to vary sinusoidally
over a 24-h period, with an extreme-case magnitude of i"C (as inferred from observations of
vapor space temperatures in the Test Apparatus). This is equivalent to a vapor space temperature
change of approximately 0.25°C/h.. Combining these factors results in an apparent volume rate
of about 0.04 L/h (0.01 gal/h), which is negligible. Thus, if the vapor stays in equilibrium, the
amount and variation of gasoline held in the vapor phase should not be an important inhibition
on leak detection. This condition is not apt to be experienced after product addition.
Table 6.3. Gas Phase Densities
Temperature (°C)
0
5
10
15
20
25
30
35
Density
of Vapor (pv) of
0.054
0.072
0.094
0.121
0.155
0.198
0.249
0.312
(g/L)
Saturated Air (pA)
1.333
1.323
1.317
1.316
1.321
1.332
1.351
1.379
pv/pL(xl<)4)
0.72
0.97
1.26
1.62
2.08
2.65
3.34
4.18
Second, consider the condensate clinging to the walls of the tank; this liquid is the other
major factor in achieving the equilibrium condition. The amount can be highly variable. At one
extreme, the tank walls will be dry if the wall temperature is greater than the product temperature
(of course, it is the product temperature at the surface that is relevant). At the other extreme,
with the tank walls cooler than the product, there will be condensation, eventually reaching an
equilibrium with liquid dripping off at the same rate that vapor is condensing. For a tank 6.5 m
long, with a 1.2-m radius, there will be 29 m2 of wall area when the tank is half full. As an
estimate, the maximum amount of liquid held on the walls would be approximately 0.2 mm thick
(e.g., droplets 1 mm thick covering 20% of the surface). This number, which might easily be off
by a factor of 2, yields approximately 6 L of liquid on the walls. The contribution of liquid from
the tank walls to the equilibrium condition, then, could range from 0 to 6 L. The equilibrium
state should be independent of temperature (or time) unless the temperature difference between
6-14
-------
the wall and the product changes sign. Note that if the walls are holding the maximum (constant)
amount of liquid, there can still be substantial evaporation from the product surface, vapor
transport to the walls, condensation on the walls, and dripping back into the tank.
6.4.3.2 Nonequilihrium Conditions
If the vapor space is out of equilibrium, either because dry air has been introduced to the
tank or because the product temperature has changed with a new delivery, a significant volume
change may result from the shifting amount of gasoline in the gaseous state. Depending on how
extreme the starting condition is, there may well be a change of a liter or more in product volume
as the vapor space comes into equilibrium. The time required for equilibration to take place is
hard to predict; hence, volume rates are also hard to predict.
The temperature at the product surface is normally greater than or equal to the ground
temperature. If the tank has been sitting for some time, the product temperature is observed to be
the same as that of the ground. Recently added cool gasoline sinks to the bottom of the tank and
does not influence evaporation and condensation, while recently added warm gasoline floats on
top and is important. Consequently, attaining equilibrium usually means that product is lost to
the vapor space rather than vice versa.
Because the upper tank walls respond slightly to diurnal temperature variations, they may
become slightly warmer than the product at mean ground temperature. If the wall temperature
exceeds the minimum or maximum product temperature during the day, the walls may be dry at
one point and may carry the maximum condensate load at another, moving back and forth
between these two extremes. Volume changes of several liters could result.
If recently added product is warmer than the ground, so that even with diurnal variations
the wall temperature remains below the product temperature, the walls will maintain a maximum
condensate load with no net effect on product volume. The evaporation from the product
surface, condensation on the walls, and dripping back into the tank are part of the thermal
equilibration process that moves heat from the warm product to the cooler tank walls.
One of the reasons that predicting the temporal evolution of the equilibration process is so
difficult is that the density of the saturated air has a peculiar behavior with regard to temperature.
The exact temperature depends on the vapor pressure curve for a particular product. When
temperature increases, the air density drops, but the vapor pressure increases, replacing light air
molecules with heavy gasoline molecules. At low temperatures, the vapor pressure is
sufficiently low that the air dominates, but at higher temperatures, the gasoline dominates. This
results in some unconventional effects. For example, if the product temperature is 13°C, there
6-15
-------
will be convection regardless of whether the walls are warmer than the product (and thus wet,
allowing evaporation to occur) or cooler than the product. However, because the density of
saturated air varies only slightly with temperature, convection will be relatively weak.
The above assumes that the air remains in saturated equilibrium as it moves from the
region near the product surface to the wall. In actuality, if the air is moving into a warmer
region, liquid may not be available for evaporation. Thus, while warm wet walls can cause
convection, warm dry walls will lead to a stable, nonconvective vapor space.
Predicting the detailed process of equilibrium is very complicated. Even such a
fundamental question as whether convection is occurring may not be knowable a priori, as it
depends on the initial condition of condensate on the walls.
6.4.3.3 Summary
Gasoline in the gaseous state does not appear to present a problem except for the initial
transients that appear when warm product has just been added. After these initial transients, the
amount of condensate on the walls should remain stable. When the product surface temperature
is near the tank wall temperature, the amount of condensate carried on the walls can vary by
several liters depending on small diurnal temperature changes. This rate is effectively
unpredictable.
6.4.4 Structural Deformation
Whenever the product level in the tank changes, the change in hydrostatic pressure causes
the volume of the tank to change and therefore, the product level to change further. This
expansion and contraction of the tank volume is strongly dependent on the compressibility of the
backfill in the excavation and on the native soil surrounding the excavation. The deformation
effects are particularly complex for tests conducted while product level in the fill tube or
standpipe of an overfilled tank.
6.4.4.1 Product-Level Measurements in Overfilled Tanks under Variable Pressure
When product is brought into the fill tube (or into an above-grade standpipe) of an
underground storage tank, the resultant increase in hydrostatic pressure causes the tank to
expand. The increase in tank volume produced by this expansion lowers the product level in the
fill tube. This drop in level decreases the hydrostatic pressure, and the tank will now relax
towards an equilibrium level, one that is changing. Similarly, when product is removed from the
fill tube, the decrease in hydrostatic pressure will cause the tank to contract, and the product level
in the fill tube will rise.
6-16
-------
An exponential relaxation model, referred to as the Fill-Tube Dynamics Model, is
hypothesized to describe the volume and product-level changes produced by tank deformation in
an overfilled tank. The model is mathematically derived first for the specific case of
instantaneously adding a known amount of product, AVP, into the fill tube of an initially
overfilled tank, where the volume of the product in the tank, fill tube and standpipe, VP(t), is
assumed to be greater than the volume of the tank, V(t). This model describes the product-level
changes produced by topping the tank, a procedure used by most methods in order to reach the
product level required to start a tank test. (The model formulation does not include an estimate
of the change in the leak rate produced by a change in the hydrostatic head.) This addition of
product results in an instantaneous rise, Ah, in the product level in the fill tube that is
mathematically described by dh/dt = Ah 5(t-t0), where t0 is the tune at which the product level is
changed and 8(t-tJ is the Dirac delta function. The model assumes an exponential increase in
the volume of the tank itself. This is described by
V(t) = V,,+KAh(i _f-"-'-l'T'-) (6.17)
for t > t0 where V0 is the volume of the tank at time tn; K is the equilibrium elasticity of the
tank-backfill-soil system; and Tcis the hydrostatic-pressure relaxation time constant of the
tank-backfill-soil system.
The rate of change of the volume of the tank with respect to time for t > t0 is given by
anddV(t)/dt = Ofort V(t), and where Aeff is the effective cross-sectional area of the product surface.
(See Section 4.2 and [23] for more detailed discussion.)
It is assumed that the volume changes produced by all product-level changes will add by
linear superposition. This means that the total volume change can be computed by adding all of
the individual volume changes produced by each instantaneous product-level change. This is
given mathematically by the convolution integral
6-17
-------
i__ K_ f'dh(T
~TcJ dt
dV(t) K f'dh(T) .
= — e dT (6.2U)
dt "" '
where T is the time variable of integration. Eq. (6.20) can be expressed as
dV(t) Kfdh(t). ^ .._
-- = — -- *xj/ (6.21)
dt Tc^ dt VJ
where * indicates the convolution between dh(t)/dt and \|/, with vy = 0 for t < 0 and
Y = exp(-t/Tc) fort > 0. Eq. (6.21) is substituted into the differentiation of Eq. (6.19) to obtain
the rate of change of product level represented by
dt A,,, dt A,,,TC dt
where
dVp(t)
dt
= AVP6(t-t0) (6.23)
Eq. (6.22) can be solved via Fourier Transfonn methods and integrated from —<» to time t to give
the product-level time series for t > t0,
AV.Y K- ^, _
(6.24)
where the effective time constant of the tank is defined by
The product-level changes can be converted to product-volume changes by multiplying by
the effective cross-sectional area, Ad(. The first term in Eq. (6.24) is the product-level change
caused by the sum of the effects of the instantaneous deformation of the tank and of any trapped
vapor. The magnitude of the instantaneous deformation is detennined by the magnitude of the
effective cross-sectional area of the fill tube. The second term is the relaxation due to tank
deformation. The time it takes for the tank to deform is determined by Tcff , the effective time
6-18
-------
constant of the tank. Eq. (6.25) suggests that the effective tune constant is always shorter than
the time constant in a constant-head test (i.e., time constant of the tank itself), and is a function of
Acff and the unknown elasticity constant, K, of the tank-backfill-soil system.
Table 6.4 presents the values of the effective time constant, Teff, for different values of Aeff
and K, where it can be assumed that A,.ff is mainly a function of the fill tube cross-sectional area,
A. The values of Teff and K are within the range of values estimated experimentally for the steel
and fiberglass tanks at the UST Test Apparatus. For fill tubes of small diameter, the deformation
occurs very quickly in comparison to those with large diameters. The calculation suggests that a
fill tube with a diameter of 81 cm (32 in.) has a time constant approximately equal to that of the
tank.
Table 6.4. Estimate of How Teff Changes with A^ and K*
Effective Fill Tube Diameter
(cm)
10.2
10.2
10.2
10.2
5.1
20.3
40.6
81.2
Aeff
(cm2)
81.1
81.1
81.1
81.1
20.3
324.3
1297.2
5188.7
K
(cm2)
120
60
30
15
120
120
120
120
T*
(h)
0.40
0.57
0.73
0.84
0.14
0.73
0.92
0.98
Eq. (6.25) is used to make the predictions, using T = 1 h for a 1-m rise in product level prior
to starting a tank test.
The product level at infinity (i.e., many hours after the initial deformation) is given by
The total height change in the fill tube is interpreted in terms of the long-term height-to-volume
conversion factor, (Aeff + K), rather than the short-term height-to-volume conversion factor, Aeff.
An estimate of h(°°) cannot be made unless K, the elasticity constant of the tank system, is
known. The total drop in product in the fill tube is
AVP AVD
h(t0)-h(oo) = —!'——*- (6.27)
6-19
-------
For small K, [h(t()) - h(<*>) ] is approximately zero.
Eq. (6.22) can be generalized to predict the product-level changes produced by any type of
product-volume change as opposed to just the instantaneous volume change used to derive
Eq. (6.24). The product-level changes that result are given by
-"T"dT (6.28)
The product-volume changes in the fill tube can be estimated by multiplying Eq. (6.28) by A,.ff.
Eq. (6.28) indicates that the product-level changes in the tank at any point in tune are
affected by the product-volume (or product-level) changes that occurred in the past. Thus, in
order to interpret the current product-level changes in terms of volume, the past product-level
changes must be known and included. It is also clear from Eq. (6.28) that the procedure for
topping the tank will affect the product-level changes. The deformation of the tank and the
resulting product-level changes will be different even if the same volume of product is added to
the tank, but at a different rate or in different amounts. While in this case the product-level and
product-volume changes were defined as zero for the period before the product was added, this is
not necessarily true in other situations. It is important to follow the same topping procedure to
maintain a predictable performance.
If the rate of change of volume of the product in the tank is a constant, C, then it can be
shown that the rate of change of the product level in the tank is given by
dh(t) C
where C is the flow rate produced by a leak or any other product-volume change. C is equal to
the leak rate, LR, if all other volume changes are equal to zero. A negative leak rate or volume
rate is defined as an outflow or a decrease in the volume of the product in the tank. In the fill
tube, a leak which produces a constant rate of change of product volume produces a
product-level change that cannot be estimated from Aeff unless K is zero. To estimate C using
Eq. (6.29), an estimate of the long-term height-to-volume conversion factor, A^f{+ K, is required.
Since K is generally unknown during an actual test, C cannot be accurately estimated. If K is
equal in magnitude to Aeff, the measured product-level change and, therefore, the product-volume
change obtained by multiplying the product-level changes by the measured height-to-volume
6-20
-------
conversion factor may be too small by a factor of 2 or more. In a partially filled tank, the
product-level changes can be directly related to volume by the height-to-volume estimated from
a tank chart because (Aeff = A) » K. For a partially filled tank, Eq. (6.29) reduces to
(6.30)
dt A
Eq. (6.29) also demonstrates that the effects of deformation are important and will occur even if
there are no deliberate, operator-initiated product-level changes as assumed in the derivation of
Eq. (6.24). Any change of volume, whether it be a leak or thermally induced expansion or
contraction of the product, will affect the measured product-level changes similarly.
Finally, consider the case in which the product level is instantaneously raised in the fill
tube when the rate of change of volume in the tank is a constant. This is described
mathematically by
Eq. (6.31) predicts the product-level changes that would occur in a leaking tank, or a tank in
which the product volume is changing, when a volumetric test is initiated by topping the tank.
The product-level change consists of a large exponential change in the product level, followed by
a linear rate of change. The linear rate of change predicted by Eq. (6.31) after several time
constants, Tefr, have elapsed is identical to that of Eq. (6.29).
Eq. (6.31) was used to predict the product-volume changes in a 10-cm (4-in.)-diameter fill
tube that are produced (1 ) by a 1-m product-level rise resulting from topping the tank
immediately before the start of a test and (2) by a - 1 .2 L/h rate of change of volume during the
test. The product-volume changes are shown in Figure 6.4. The product-level changes are
converted to volume changes using Aeff. The volume change is assumed to be produced by a
leak, by thermal contraction of the product, or by a combination of these. No other volume
changes are considered. Values of K = 120 cnr, T = 3 h, Teff= 1.5 h, and Aeff= 125 cm2, typical
of the steel tank at the UST Test Apparatus, were used in the calculations. The predicted
product-volume change, after several tune constants, Teff, have elapsed, is -0.61 L/h, only 51% of
the actual -1.2 L/h volume rate. The dashed curve illustrates a -1.2-L/h volume rate. For
comparison, the product-level changes that are the result of deformation only are also shown in
this prediction; the rate of change of volume is zero. This could represent a large error in either
the leak rate, the temperature-induced volume changes, or a linear combination of both. If the
6-21
-------
tank is not leaking, and the volume changes are produced only by temperature changes, the
temperature-compensated volume rate obtained by subtracting the measured volume rate of
-0.61 L/h from the actual volume rate of -1.2 L/h would be +0.59 L/h instead of zero.
14
o>
ZJ
T3
O
0
-14
-~^ Actual Leak Rate
--1.2Uh
No Leak
Leak
-0.61 L/h
0
8
10
Time - h
A: Aeff = 125 cm2, K = 120 cm2, T = 3 h, Crredi,C(1 = CKtuat = 0.0 L/h.
B: \ff = 125 cm2, K = 120 cm2, T = 3 h, Cpredll,ed = -0.61 L/h, C^ = -1.2 L/h.
Figure 6.4. Comparison of the Fill-Tube Dynamics Model predictions using Eq. (6.31) for
different combinations of A^,, K, and C. In a small-diameter fill tube, only a fraction of the
actual leak rate is predicted. Cpn.dicled is estimated 3 or more time constants after the start of the
test.
The two effects which occur after the tank has been topped are significant. First, if a test is
conducted too soon after a product-level change, the measurement is typically dominated by the
exponential volume change of the tank. Many methods attempt, before starting a test, to wait
until the large exponential decay has occurred. Second, it is mistakenly assumed that once the
large decrease in product level becomes constant, an accurate test of the tank's integrity can be
conducted. This is not true if K is approximately equal in magnitude to Aeff, because the linear
product-level changes that occur after the exponential changes have occurred are only a fraction
6-22
-------
of the actual product-level changes that would occur if the tank were rigid and did not deform.
The effect is particularly severe when the diameter of the fill tube is small, because in this case
even large leak rates produce only small product-level changes.
6.4.4.1.1 Product-Level Measurements in Overfilled Tanks under Constant Pressure
Some methods measure volume changes directly by periodically adding or removing a
measured amount of product to maintain a constant product level in the fill tube. Because
product level is kept approximately constant, no significant additional deformation of the tank
occurs during the test. The measured volume changes (obtained by using A^ to convert to
product-level changes) represent the actual volume changes occurring in the tank.
The model was extended to include the effects of periodically releveling the product in the
fill tube every t' minutes in the presence of a constant volume change, C, where C is the sum of a
constant leak rate, LR, or any other constant product-volume change not related to a leak. The
model includes the topping effect whereby a known amount of product is instantaneously added
to the fill tube to attain a specified level for the test. The volume at t = nt' required to bring the
product level to zero is given by
(6.32)
The derivation of Eq. (6.32) is mathematically complex and only the result is given here. The
analytical solution was compared to the numerical solution generated using Eq. (6.28) for
verification. The total amount of product added to maintain a constant level in the fill tube is
obtained by summing the Vn calculated using Eq. (6.32) for all n. This is equal to the
product-volume drop in a fill tube of very large diameter; the drop would be estimated for the
same parameters using Eq. (6.31). If the releveling period is very much smaller than the
effective time constant, Eq. (6.32) is approximated by
(6.33)
In practice, Vp(0) is unknown because the time history of past volume changes is unknown.
It is assumed that the product level, -h,,, is constant (i.e., that there are no volume changes), for
all time t < 0, and that it is then raised a constant amount, h^ at t = 0. It can then be shown that
6-23
-------
Vn(0) = -Kh0 (6.34)
p
Thus, the volume change of the product at t = 0 is dependent on the value of K, which is not
known under most testing conditions.
If the product in the fill tube is releveled continuously, the volume changes will be equal to
the leak rate after the deformation produced by an initial addition of product to the tank has
ceased. The constant-hydrostatic-pressure time constant, Tc, governs the deformation. The
volume changes predicted by Eq. (6.32) are approximately equal to those shown in Figure 6.4
when Eq. (6.31) for the calculation made with Aeff = 4560 cm2» K = 120 cm2, providing that the
product is releveled almost continuously. If the tank is releveled every 15 min, the time constant
of the volume changes is approximately 1.14 Tc. This was obtained by solving Eq. (6.32) and
computing the time constant dkectly. Thus, the penalty for not releveling continuously is a small
increase in the time requked to test a tank.
6.4.4.1.2 Summary
Any product-level change will deform the tank. Tests which allow the product level in the
fill tube or in a standpipe to change freely may be subject to serious errors, because
product-volume changes are not accurately estimated from the product-level changes in the fill
tube using Acffwhen K » 0 cm2 or when K » Acff. While Tc is also unknown, its exponential
behavior is known, and an algorithm can be developed to minimize this effect by waiting for the
volume changes to become negligible. The actual or true volume changes can be obtained by
multiplying by the long-term volume-to-height conversion factor (i.e., Aeff + K) once the
exponential changes induced by product-level adjustments necessary to start a test have become
small. The magnitude of the error will depend on the elasticity of the tank. Since K is unknown,
accurate estimates of the actual volume rate are difficult to make. If A,.ff is not measured, then
the error is larger yet. If K happens to be small compared to A,.ff, the error is also small, but the
tester has no way of knowing this.
The product-level fluctuations that occur in a volumetric test having a nearly constant
hydrostatic pressure can be estimated by a simple sum of the individual contributions of each
noise source. Product-level fluctuations can be converted to volume using the effective
cross-sectional area of the product surface, A,,ff.
6.4.4.2 Product-Level Measurements in Partially-Filled-Tank Tests
In partially-filled-tank tests conducted at a constant hydrostatic pressure, any change in
product level causes the tank to deform. This deformation can be estimated using Eq. (6.35),
6-24
-------
except that the effective time constant of the tank is equal to Tc and the effective cross-sectional
area can be estimated from the geometry of the tank and does not have to be measured. The
deformation of a partially filled tank is estimated by
= KAh(i_ t,,.
Except after product additions, structural deformation in partially-filled-tank tests is
negligible. The height changes produced by temperature fluctuations, waves, or evaporation are
very small when the tank is less than 90% filled.
6.4.5 Surface Waves
Changes in height caused by surface waves in the tank are sinusoidal, and can be modeled
by
i
Ah(t) = Bc 'wcos(tL>t-Kt>) (6.36)
where Ah is the change in height, B is the initial magnitude of the height change, t is time, Tw is
the decay constant, to is the frequency of the surface wave, and cp is the phase at t = 0. In a tank,
surface waves caused by external vibrations typically have a period on the order of 2 to 10s.
Surface waves have an impact on product-level data acquisition requirements.
6.4.6 Internal Waves
Large subsurface internal waves with periods on the order of several minutes to an hour or
more propagating along a temperature (density) gradient are recorded as apparent fluctuations in
the temperature of the product measured at any point affected by the wave. In some instances,
these waves may result in surface disturbances. This noise source can be represented by a
sinusoidal curve similar to the one described for surface waves. The period and amplitude are
difficult to predict, and depend on the geometry of the tank, the density gradients in the tank and
the generating force. The effect of internal waves in the simulation was incorporated as
measured in the temperature database. Internal waves have an impact on thermal data
acquisition requirements.
6-25
-------
6.5 Test-Method Models
A mathematical model of each test method was developed from information provided by
the manufacturer. A report summarizing the method, and including a description of the
instrumentation, test protocol, data analysis, and detection criterion, was prepared and was given
to the manufacturer for review and written concurrence.
With the ambient noise models and the experimental database, the test-method model
generated the product-level database. A familiarity with the test configuration, system noise, and
the test protocol was required in order to generate the product-level database, which was then
input to the database manager. Finally, the product-level database was analyzed, using the
method-specific data analysis algorithms to compute a temperature-compensated volume rate.
6.5.1 Instrumentation
The instrumentation used in each leak detection method was modeled with experimentally
measured estimates of the resolution, precision and accuracy of the product-level and
temperature measurement systems. The model reflects the number, the configuration, and the
locations of the sensors. The system noise was added to the temperature and ambient noise
database for use in the performance simulation.
6.5.2 Test Protocol
The method-specific test protocol reflects the conduct of the entire test, starting with
product delivery and ending with the data necessary to compute the temperature-compensated
flow rate. The test protocol includes the waiting periods after adding product to the tank, the
data sampling, the test duration, and number of test periods required by the method (Figure 4.3).
The test protocol also incorporates nonunifonn and interactive sampling schemes, as well as test
durations that are nonunifonn or that are dynamically determined from the data.
6.5.3 Data Analysis
The data analysis algorithm for each leak detection method evaluated was identical to that
specified by the manufacturer. The data analysis included the preparation of product-volume and
temperature-volume time series, temperature-volume compensation schemes, and calculation of
the temperature-compensated volume rate.
6.5.4 Detection Criterion
For each leak detection method, the method-specific detection criterion used to declare a
leak was incorporated into its performance model.
6-26
-------
6.6 Tank-Test Simulator
The tank-test simulator was used to mimic the conduct of a test of an underground storage
tank. The simulator accesses the temperature and product-level databases, as well as the
test-method model. The operator can specify which test method to implement, which part of the
database to use, and how many tests can be conducted for each run file in the database.
The product-level time series is defined, in general, by
(I-T)
Tefrdi (6.37)
V(t) = VSN(t) + V^t) + VL(t) + Vprotocol(t) (6.38)
where V(t) is the volume change associated with product-level changes, h(t); VSN(t) is the
volume change associated with the product-level sensor system; V^t) is the product-level
volume change associated with each of the six noise sources ; VL(t) represents the volume
change associated with a leak; and Vproux.ol(t) represents product volume changes called for in the
test protocol.
In a nonleaking tank, VL(t)= 0. However, in general, V(t) will not equal 0, because the
combined system and ambient noise are statistically independent and will not sum to 0. The
noise-compensated volume time series, Vc(t), is defined by
Vc(t) = V(t) - VNC(t) = VSN(t) + V^t) + VL(t) + Vprolocol(t) - VNC(t) (6.39)
where VNC(t)are the volume change estimates inferred by the test method to compensate for the
noise-induced volume changes.
The simulator then generates a histogram of the volume-compensated noise and, if desired,
the signal-plus-noise. For the majority of the tests, only the noise histograms were generated.
6.7 Performance Analysis
The temperature-compensated volume rate data obtained for each simulated test were
stored cumulatively in a file, ready for performance analysis. The histograms and cumulative
distribution function of each test method were generated and displayed graphically.
The performance curves generated in this report were based on the measured noise and
signal-plus-noise cumulative distribution functions. For high levels of perfonnance, the
probability of false alarm and the probability of detection are estimated from the tails of the
cumulative distribution function. With limited data, good estimates of the PD and PFA are
6-27
-------
sometimes difficult to make. In this study, the performance estimates were typically based on 50
to 200 independent realizations of the manufacturer's test. For a few methods, the performance
estimate was based on less than 50 test.
With this modest amount of data, it was difficult to make reliable estimates of the false
alarm rate at 0.01 and the probability of detection at 0.99. The total number of realizations
depended on the duration of the test. A model of the data can be used to estimate the cumulative
distribution function more accurately.
An attempt was made to model the cumulative distribution functions with a normal model,
which was calculated using the sample mean and standard deviation of the simulation results. In
many instances, the model was in good agreement with the measured results. When there was
poor agreement, it occurred because the tails of the measured histogram had not been modeled
well by a normal distribution. In general, the normal model does not accurately describe the tails
of the distribution. This is not unusual; a normal model usually predicts false alann rates which
are much lower than those observed.
An exponential function was fit to the data to more reliably estimate the tails of the
cumulative distribution function at both the 0.99 and 0.01 and the 0.95 and 0.05 cumulative
frequencies. This yields a better estimate than can be obtained by extrapolating the
temperature-compensated volume rate with only one or two points. The model could also have
been used to make performance estimates beyond the range of the data, but this was not
generally necessary, nor was it desired. Several other functional forms were fit to the data, but
the exponential curve provided the best fit.
In general, 20% of the data were used to fit the tails, 10% for each tail. The exponential
model is given by
'-aU' (6.40)
where C(x) is the cumulative distribution function, x is the temperature-compensated volume
rate, D0 is the value of the cumulative distribution function at 0.90 and 0.10, D, is the amplitude
of the exponential change, and a(x) is the rate of change of the cumulative distribution function.
In some cases, less than 10% of the data were used to model the tails, and D0 changes
appropriately.
A least-squares technique was used to estimate the uncertainty in the probability of
detection and the probability of false alann at 0.99 (or 0.95) and 0.01 (or 0.05) from the curve-fit
coefficients. The resulting one-standard-deviation rms error of the PD and PFA was calculated and
tabulated for each method.
6-28
-------
The performance is presented in three displays, the first of which is a plot of the
probability of detection versus detection threshold for a family of leak rates with flow into and
out of the tank (positive and negative volume rates). The second display is a plot of the
probability of false alarm versus threshold. The third display shows the probability of detection
versus the probability of false alarm for a family of leak rates. The mean of the noise histogram
was removed for the third display (i.e., the bias was assumed to be zero), but the bias was
included in the first two displays.
Three test results are summarized in the body of the report: (1) the performance of the
method using the manufacturer's standard detection criterion (i.e., threshold) against a 0.38-L/h
(0.10 gal/h) leak rate in terms of the PD and Pr.A, (2) the potenital performance that could be
achieved with the method in terms of leak rate for a PD of 0.99 and PFA of 0.01, and a PD of 0.95
and PFA Of 0.05 after using a threshold defined by the PFA. A more thorough presentation of
results in given in the appendices in Volume El.
Examples of the performance curves illustrative of the output generated for each test
method evaluated are shown in Figures 6.5 for a hypothetical test method which tests when
product is near the top of the tank, using an array of five equally spaced, volumetrically weighted
thermistors. The noise histogram is shown in Figure 10.1. It was assumed that the temperature
and product-level sensors used by this method had sufficient precision to measure ambient
product-volume changes that were less than 0.04 L/h (0.01 gal/h). The data were sampled once
per minute and the duration of the test was 1 h. The only source of noise considered in the
simulation was thermal expansion or contraction of the product.
6-29
-------
1.00
0.00
-3000 -2000 -1000 0 1000 2000 3000
Threshold - ml/h
0.20
0.15
75 0.10
005
0.00
-3000 -2000 -1000
Threshold - ml/h
B
1000 2000
3000
Figure 6.5. Examples of performance curves for a generic leak detection method. (A) PDvs.
Threshold, (B) PFAvs. Threshold, (C) PD vs. PPAfor flow out of the tank and (D) PD vs. PrA for
flow into the tank.
6-30
-------
o
0)
«
O
•5
1.00
0.80
0.60
0.40
CO
.a
2 0.20
0.00
-0.19
0.00 0.05 0.10 0.15
Probability of False Alarm
0.20
1.00
0.80 —
_
t> 0.60
o
Q
0.40
,? 0.20
0.00
0.10
0.00 0.05 0.10 0.15
Probability of False Alarm
0.20
Figure 6.5 .(concluded). Examples of performance curves for a generic leak detection method.
(A) PD vs. Threshold, (B) PFA vs. Threshold, (C) P,, vs. PnA for flow out of the tank and (D) PD vs.
PFA for flow into the tank.
6-31
-------
7 UST Test Apparatus and Quality Control of Measurements
All of the experimental data used to estimate the performance of the volumetric test
methods were collected at the UST Test Apparatus. With this apparatus, full-scale tank tests can
be conducted under controlled field conditions. The Test Apparatus can simulate the field
delivery of product at temperatures different from that of the product stored underground. Leaks
up to 8 L/h (2 gal/h) can be simulated in either the tank or its associated piping with an accuracy
better than 0.04 L/h (0.01 gal/h). This is accomplished using a variable-speed pump. The
apparatus can also evaluate the effects of subsurface water on test-method responses. In
addition, vapor pockets of known size can be created in the tanks or in the associated piping such
that, when the tank is filled, vapor is trapped at the top of the tank.
To address the overall project objective, a set of data quality objectives was established at
the beginning of the program and was adhered to throughout the data collection [24]. The UST
Test Apparatus instrumentation, calibration procedures, and data quality analyses after each test
were designed to verify that the data were meeting the data quality objectives.
The majority of the commercially available test methods claim to detect leak rates as small
as 0.19 L/h (0.05 gal/h). This claim satisfies the 0.05-gal/h recommended practice specified by
the National Fire Protection Association in its Pamphlet 329 [5], which for many years was the
only guidance available and thus became the nationally accepted standard for any method used to
make a final determination as to whether a tank is leaking or not. The data quality objectives for
this project were established to evaluate this common performance claim. Unfortunately, the
process was not straightforward, because both the recommended practice and the performance
claims are ambiguously stated. Because of the statistical nature of tank testing, a complete
specification also requires that the probability of false alarm and the probability of detection and
the uncertainty of each be included for the leak rate specified in the performance claim.
7.1 Test Apparatus
The UST Test Apparatus allows control to be exercised, as far as is practical, over the
major factors that affect the accuracy of volumetric leak detection methods. The Test Apparatus
is environmentally safe and permits the establishment of known test conditions in full-scale
underground storage tanks. The Test Apparatus, shown in Figure 7.1, consists of the following
major components:
o (l)30,000-L(8,000-gal) steel tank
o (1) 30,000-L (8,000-gal) fiberglass tank
7-1
-------
Figure 7.1. usTTanJc Test Apparatus.
7-2
-------
o Separate, synthetic-membrane pit liners to provide secondary containment for each
underground tank and its associated underground piping
o (1)11,500-L (3,000-gal) tank with heat exchange coils
o (1) 23,000-L (6,000-gal) tank with heat exchange coils
o (1) 154,000-Btu/h electric heater
o (l)60,000-Btu/h chiller
o (2) 1,500-L/min (400-gal/min) transfer/circulation pumps
o Suction piping and pressure piping
o Monitoring wells both inside and outside the secondary containment of each tank
The two underground tanks adjacent to the above-ground equipment permit easy access for
experimentation. Numerous tank penetrations, including manways, accommodate extensive
instrumentation and permit maximum flexibility in experimental design. These penetrations also
enable any of the commercial volumetric test methods to be used with either tank.
The secondary containment (synthetic membrane pit liner) surrounding each of the
underground tanks serves three main purposes. It protects the environment from any unforeseen
product release; in the event of a release, it allows the Test Apparatus personnel to determine
whether the leak is in the tank or in the associated piping; and it permits, for test purposes,
control of the water table surrounding the tank.
The field delivery of product at temperatures different from those of the product stored
underground in situ can be simulated by extracting product from one of the underground tanks,
heating or cooling it in one of the two above-ground tanks (in conjunction with the heater and
chiller), and subsequently draining it back into the underground tanks before testing. The
resulting internal temperature distributions produce a good test of thermal compensation
techniques for leak detection methods.
The effects of subsurface water on test-method responses also can be evaluated in the
Test Apparatus. The underground secondary containment liner can be deliberately flooded or
drained at controlled rates that are representative of actual field situations. As a result, the
effects of the water table elevation change on the structural deformation of the tank and of the
thermal responses of the tank and the contained product can be evaluated.
Vapor pockets of known size can be created in the tanks or in the associated piping. The
effects of these pockets on methods requiring tank overfilling before testing can be quantified
over a wide range of vapor pocket sizes.
7-3
-------
Other features of the Test Apparatus include the ability to withdraw product from the
underground tanks by either above-grade (suction) or submerged (positive pressure) pumps. In
addition, leaks can be induced in either of the tanks or in their associated piping at known,
controlled rates over a wide range. These features facilitate the evaluation of each volumetric
leak detection method's ability to differentiate piping leaks from tank leaks and allow
verification tests to be conducted with a known leak rate.
The Test Apparatus design contains features that not only are directly applicable to the
current evaluation program but also provide for considerable flexibility in anticipation of future
possible areas of investigation. Among the areas considered are type of product; non-volumetric
leak detection methods; and tank size, type of construction, and materials.
The Test Apparatus is situated to enable additional underground tanks to be added qu ickly
(if needed), with minimal disruption of the existing equipment. It is anticipated that, at the
minimum, a different tank capacity and a different type of construction (e.g., double-walled
tanks) may be evaluated. The numerous monitoring wells, located both within and exterior to
the secondary containments of each underground tank, and the numerous penetrations on each
tank facilitate access so that additional test methods (such as those that identify leaks by other
than volumetric means) can be evaluated readily.
7.2 Data Quality Objectives
The precision and accuracy of the product-level and temperature data collected at the UST
Test Apparatus were specified so as to evaluate the performance of each test method at a leak
rate of 0.19 L/h (0.05 gal/h) with a probability of detection of 0.95 and a probability of false
alarm of 0.001 [11]. An estimate of the precision and accuracy of the measurement systems was
made using the performance model shown in Figure 5.9. Assuming that the histograms of the
noise and the signal-plus-noise are normally distributed, an estimate of the one standard
deviation of the temperature-compensated volume rate needed to meet this performance
specification was made. The standard deviation must be 4.76 times smaller than the leak rate to
be detected in order to meet the probability of detection and probability of false alarm specified
above. This corresponds to 0.04 L/h (0.01 gal/h) for a leak rate of 0.19 L/h. This means that the
precision of the instruments used to measure temperature and product level, the accuracy of the
constants used to convert temperature and product level to volume, and the number and spatial
distribution of the sensors used to measure the temperature-compensated volume must have a
total uncertainty of less than 0.04 L/h.
7-4
-------
Precision is defined as the one-standard-deviation uncertainty in each successive
measurement. The precision of the temperature and product-level data must be defined so that
the rate of change of temperature and product level, after conversion to the rate of change of
volume, is less than 0.04 L/h. To meet this objective, the temperature measurement system and
the temperature calibrations were designed and implemented to ensure a precision of 0.001°C.
For product-level measurements, a precision of 0.005 m for measurements made in a 10.2-cm
(4-in.)-diameter fill tube and a precision of 0.0000013 m for measurements made in a half-filled,
30,000-L (8,000-gal) tank were required.
Accuracy (bias) is defined as the difference between the measured and the true value of a
given quantity. The accuracy of the product-level measurement depends primarily on the
magnitude of the error in estimating the volume of product in the tank and in estimating the
height-to-volume conversion factor using a perfect tank chart (i.e., a table relating product depth
to product volume). The degree of accuracy in using the product-level measurements to estimate
a height-to-volume conversion factor from a tank chart is low, except when measurements are
made when product level is in the upper 10 to 20% of the tank, where accuracy can be as exact
as the precision requirements. Further errors can result from errors in the tank chart itself. A
tank chart is usually developed for the nominal geometry of the tank. Any differences between
nominal and actual dimensions that occur during fabrication of the tank, or any tilting of the tank
or geometric distortion near the top or bottom that occurs during installation or use of the tank,
will result in errors in the tank chart. In the upper 5 to 10% of the tank, an error of a factor of 2
can occur due to distortion. Because of errors in the tank chart, especially near the top of the
tank, and because of the high degree of accuracy that is required when a tank chart is used to
convert product-level measurements to volume, the height-to-volume conversion factor was
measured directly in all experiments. The decision to measure the height-to-volume conversion
factor eliminated the need to establish unusually stringent accuracy requirements that would be
similar in magnitude to the precision requirements. These would have been difficult to achieve.
The accuracy of the product-level measurements was based upon the nominal accuracy
specified by the manufacturer of the pressure sensor, viz., 5 mm (0.2 in.), within the 0.04-L/h
uncertainty necessary to detect a 0.19-L/h leak rate at the prescribed probabilities of detection
and false alarm. There is no specific accuracy requirement for the temperature measurements. A
value of 0.55°C (1°F) was selected for the air and ground temperatures. A value of 0.05°C was
selected for the vapor and product temperatures so that profiles could be plotted without
excessive deviations.
7-5
-------
The precision and accuracy requirements, which are summarized in Table 7.1, were
reevaluated in early January 1987 after a probability of detection of 0.99 and a probability of
false alarm of 0.01 had been selected as the minimum acceptable testing errors in the EPA's
proposed underground storage tank regulations [3]. The selection of a high probability of
detection was based upon the EPA's desire to identify as many leaking tanks as possible. Since a
false alarm does not result in environmental damage, the probability of false alarm was selected
to maximize the number of test methods that could meet the performance standard and still
provide a measure of protection from excessive testing errors to the owner or operator of an
underground storage tank. The precision needed to satisfy this performance standard require that
the standard deviation of the noise histogram be 4.67 times smaller than the leak rate to be
detected. This is less stringent than originally estimated. Thus, the precision and accuracy
requirements remained essentially the same, but the number of tests required to estimate a
probability of false alarm decreased significantly.
Table 7.1. Specifications of the Apparatus Instrumentation
Sensor
Ak Temperature
Vapor Temperature
Ground Temperature
Product Temperature
Product Level below Tank Top
Product Level above Tank Top
Accuracy
0.55°C
0.55°C
0.55°C
0.05'C
0.005 m
0.005 m
Precision
0.55°C
0.55°C
0.55°C
o.oo rc
0.0000013 m
0.005 m
In its final version, the regulation stipulated a PD of 0.95 and a PFA of 0.05 [4]. The
standard deviation of the noise histogram required to meet the final EPA regulation is 3.92 times
smaller than the leak rate to be detected. The precision requirements for making temperature and
product-level measurements are at their practical technological limits. The data quality objective
represents the most stringent requirements on the data, because it assumes that the methods to be
evaluated will meet the 0.19-L/h performance claim. In retrospect, neither the EPA regulation
nor the performance achieved by the manufacturers required the data to meet this objective. For
the majority of the evaluations performed in this program, the precision requirements specified
here are 2 to 20 times more stringent than necessary.
7.3 Development of Data Quality Specifications
The data quality specifications are developed below.
7-6
-------
7.3.1 Precision
The precision specifications are based on two criteria, referred to as A and B.
Criterion A. The uncertainty in the rate of change of the total volume times series
(generated in the simulation by summing up to four ambient-noise volume time series collected
at 1 sample/min over a 1-h period) is less than or equal to 0.04 L/h (0.01 gal/h) at any level in the
tank.
Criterion B. The uncertainty in the rate of change of the temperature-compensated time
series (generated from a two-point measurement of height and temperature over a 1-h period by
the test method being evaluated) is less than or equal to 0.04 L/h at any level in the tank.
These criteria were selected because they represented a realistic range of noise and
test-method conditions. If both criteria are satisfied, each test method can be evaluated, using the
data collected in the apparatus, to within 0.04 L/h. The sensor precision required to satisfy
Criteria A and B is presented in Tables 7.2 and 7.3, respectively.
Table 7.2. Precision of Each Sensor Required to Satisfy Criterion A in a 30,000-L (8,000-Gal)
Tank (An instrument with an uncertainty of 0.0424 L (0.0112 gal) would result in an uncertainty
of 0.000314 L/h (0.000083 gal/h) over a 1-h period.)*
Product Level
Half-full 30,000-L tank
Full 30,000-L tank
(product level in 0.102-m (4-in.)
diameter fill tube)
c
Llpr«ls!on
Height
0.0000027 m
(0.0424 L)
0.052196m
(0.0424 L)
Temperature
0.002 rc
(0.0424 L)
o.ooi rc
(0.0424 L)
*Note: Using a 60-min time series wherein data are collected at 1 sample/min, an S hionof
0.0848 L (0.0224 gal) is required for a volume time series to satisfy the Srale= 0.04 L/h
(0.01 gal/h) requirement. Since there are up to four time series that may be added
together, the required uncertainty in S isionfor any one sensor is 0.0424 L (0.0112 gal).
If the precision of the temperature ana height sensors is specified as O.OOTC and
0.0000025 m, respectively, then the uncertainty in rate is less than 0.04 L/h (0.01 gal/h)
at any product level.
The standard deviation of the volume rate can be estimated from the standard deviation of
the slope, Sral(:, of a least-squares line fit to a volume time series using
S2 precision
= -~
n(It2)-(It)2
(7.1)
7-7
-------
where Spreclsionis the standard deviation of the ordinate (volume) about the regression line, n is the
number of independent points, and t is the time of each volume sample. Eq. (7.1) assumes that
the system noise of the height and temperature sensors is white. The precision (Spn.cision) can be
estimated for each criterion using Eq. (7.1).
Table 7.3. Height Sensor Precision Required to Satisfy Criterion Bin a 30,000-L Tank Given
That Precision of the Temperature Sensor is 0.001°C*
Product Level
Half-full 30,000-L tank
Full 30,000-L tank
(product level in 0.102-m (4-in.)
diameter fill tube)
c
'-'precision
Height
0.0000013 m
(0.0201 L)
0.002540 m
(0.0201 L)*
Temperature
o.oo rc
(0.0178 L)
o.oo rc
(0.0178 L)
* Calculated for a temperature-compensated volume rate of 0.0541 L/h (0.0143 gal/h)
Table 7.4. Uncertainty of the Temperature-Compensated Volume Rate Estimated Using the
Temperature and Height Precision Specifications for a 1-h Measurement Period and a Rate of
1 Sample/Min
Height
Half-full 30,000-L tank
Full 30,000-L tank (product
level in 0.102-m (4-in.)
diameter fill tube)
Uncertainly (L/h)
One
Thermistor
0.0000135
0.0000178
One Thermistor Three Thermistor
Array Arrays
(12 sensors) (36 sensors)
0.0000038
0.0000051
0.0000023
0.0000030
The precision specification is the most crucial. Based on the two criteria, a temperature
precision of O.OOrC and a height precision of 0.0000013 m are required. With this precision,
the uncertainty in the temperature-compensated volume rate derived from a 1-h measurement of
data collected at a rate of 1 sample/min is significantly better than 0.04 L/h (0.01 gal/h), even if
only one temperature sensor (e.g., a thermistor) is used for compensation. Since 36 temperature
sensors were used, the uncertainty of the rate was six times smaller. This is shown in Table 7.4,
which presents the uncertainty in the temperature-compensated volume for measurements made
in both half-full and full 30,000-L tanks using a single thermistor, one array of thermistors, and
three arrays of thermistors.
7-8
-------
7.3.2 Accuracy
The accuracy specifications are based on Criterion C, which states that the uncertainty in
the product-level measurement required to estimate the product volume from a tank chart, and
used in making an estimate of the thermally induced expansion or contraction of the product,
results in a volume error less than or equal to 0.04 L/h due to a temperature change of 0.1 °C/h.
The largest volume changes per unit of product-level change occur in a half-filled tank
(where the surface area of the free product is largest in a horizontal right-regular cylindrical
tank). This corresponds to a 157-L/cm (106-gal/in.) volume change in a half-filled 30,000-L
tank containing gasoline. At this level, a 303-L (80-gal) error in the product volume results in a
0.04-L/h error due to a temperature change of 0.1 °C/h. This corresponds to a product-level
change of 1.3 cm (0.51 in.). The geometry of the UST Test Apparatus tanks was determined to
an accuracy of 3.2 mm (0.125 in.) using a tape. A 3.2-mm error in tank dimensions results in an
error of 95 L (25 gal), within the 303-L allowable error. Criterion C does not place any severe
constraints on accuracy.
In general, the errors inherent in the tank chart itself can be large. The accuracy criterion,
which was established assuming that the tank chart is accurate, was not required in order to
estimate the height-to-volume conversion factor in the upper portion of the tank.
7.3.3 Representativeness
The performance estimates depend strongly upon the product-temperature changes that
occur in a storage tank. Before the evaluations were started, a study was conducted to estimate
the range of difference between the temperature of the product delivered to the tank and that of
the in situ product and of the surrounding ground [11]. In all, 77 cities in the United States,
including Hawaii and Alaska, were represented. The mean temperature difference during
January was +3°C and during July -8.5°C. Based on this study, it was concluded that
product-temperature changes caused by the addition of product at least 10.5°C warmer or cooler
than the product in the tank would cover approximately 95% of the possible test conditions
found nationally over the course of a year. During any one month, the 95% confidence intervals
on the temperature differences were ± 5°C.
Product temperature conditions must be representative of conditions that would be found
in the field during actual testing. Therefore, the data from which the temperature database was
developed were collected at the Test Apparatus, and were generated assuming a uniform
7-9
-------
distribution of temperature differences between +10°C so that the thermal changes expressed as a
volume rate were approximately normally distributed with a standard deviation of 850 ml/h
(0.225 gal/h).
Both of the 30,000-L tanks were used to collect the data at the Test Apparatus. In theory,
the performance of a method should improve as tank capacity decreases, because, for example,
the magnitude of the thermal expansion or contraction of the product also was shown to decrease
as product volume decreased. Thus, the results of the evaluation are applicable to
smaller-capacity tanks as well as to other tanks with similar diameters. The mean capacity of
tanks in the United States is less than 23,000 L (6,000 gal) [2].
An extremely important source of noise is the deformation of the tank that may occur when
product level is changed during test preparation and setup. The steel and fiberglass tanks were
installed according to their manufacturers' specifications, and represent tanks which are
relatively new (approximately one year old during the evaluations) and correctly installed. Since
there are several hundred different types of tanks, different methods of installation, and different
soils and backfill materials, the deformation characteristics of these particular tanks represent
only one of a range of conditions found in the field. Available information [7, 25] suggests that
the time constant of the tanks is similar to other measurements of time constants. Furthermore,
since it is the backfill and in situ soil that seem to control the deformation of the tank walls, it is
believed that the UST Test Apparatus tanks are at least typical of other new tank systems
installed with a pea-gravel backfill.
The distribution of the volume of vapor trapped in an overfilled-tank test is also unknown.
To obtain an accurate estimate of the trapped vapor distribution would require, as a minimum,
basic information on the geometry of the tank as installed and feedback on how effectively the
tank-test operators remove the vapor during the overfilling process. As a consequence, precise
estimates are not available for incorporation in the perfonnance evaluation. Qualitative
estimates of vapor pocket effects, however, were developed and discussed based upon vapor
bound estimates of the maximum size of vapor pockets which could be expected to exist.
The making of a complete performance estimate is hampered by the lack of information on
the volume of trapped vapor and the time constant and tank elasticity constant for a
representative national population of tanks. This is an obvious constraint that was known and
planned for at the beginning of the program.
It should be noted that the above-ground piping system installed in the Test Apparatus is
not typical of retail service stations. This was known and was also pointed out by test-method
manufacturers, representatives of the petroleum industry and trade associations, and the EPA
7-10
-------
peer review panel. To avoid potential complications, none of the product transfer piping was
included in the tests. The tank being tested was isolated from the above-ground tanks for the
duration of the evaluations. All other aspects of the Test Apparatus were judged to be
representative of a 30,000-L tank system that might be encountered in the field.
7.3.4 Completeness
The amount of temperature and product-level data required to validate the noise models
and perform the evaluation was sufficient to meet the data quality objectives of the program.
7.3.5 Comparability
All data collected during the program were obtained at the UST Test Apparatus using
automated data acquisition systems for data collection, calibration, and reduction. The data set
that was collected is unique. For three of the methods, performance evaluations with these data
were compared to field test evaluations of these same methods employed with different testing
protocols. Agreement was within the experimental errors of the ambient noise field and the
protocol differences [6-8, 21].
7.4 Instrumentation
Instrumentation used to gather data during the ambient noise experiments and to conduct
the Field Verification Tests consists of several integrated systems designed to acquire highly
precise measurements of product level and temperature in the underground tanks. A laser
interferometer with two channels obtains detailed measurements of product-level changes.
Absolute product level and atmospheric pressure are measured by separate pressure sensors. In
addition, product temperatures are measured by means of 36 to 42 thermistors distributed
vertically at three locations in each tank, seven thermistors distributed vertically in the ground
near each tank, and one air thermistor. The temperature of the concrete pad and laser stand are
also measured, as required, to support specific experiments.
The instrumentation was designed and calibrated so that the uncertainty in the volume
fluctuations produced by the instrumentation itself was low enough to meet the data quality
objectives. All instrumentation met the design specifications. Any individual sensor (e.g., a
thermistor) which failed to meet specifications was not used in the analysis and was replaced at
the next calibration. All instruments were calibrated on a regular schedule (usually monthly)
according to the Quality Assurance Project Plan (QAPP) [16].
7-11
-------
7.4.1 Laser Interferometer
7.4.1.1 Description
The laser interferometer measurement system is a highly precise instrument used to
measure changes in the height of the product surface in a UST. The laser interferometer system
consists of two elements: a laser interferometer mounted on a rugged adjustable stand, and two
brass tubes inside of which are positioned floats containing corner-cube beam reflectors. The
signal from the two interferometer channels is shown on two local displays which are in turn
connected to a data acquisition computer so that long-term testing can be conducted. In
operation, the laser measures changes in float position with respect to the interferometers
mounted on the stand. The reference standard for these distance measurements is the wavelength
of the laser light, which is very tightly regulated.
Product-level changes are measured simultaneously in both tubes, with one opened and
one closed to the product in the tank. Product-level changes in the tube open to the tank are
identical to the product-level changes in the tank, and thus are affected by all sources of noise
and simulated leaks. Product-level changes in the closed tube are theoretically affected only by
the thermal expansion or contraction of the product and by evaporation or condensation of the
product.
Because the heat generated by the laser head and electronics is separated from the
interferometers, the height measurements are very precise. The resolution and precision of the
laser interferometer (according to its manufacturer's specification) are 0.025 and 0.043 (im,
respectively. The laser is re-zeroed each time the power is turned on or off, or each time the
laser display units are reset. As a consequence, the system is used to measure product-level
changes but is not used to measure the depth of product in the tank. The product-level data are
sampled at 200 Hz (200 samples/s), with a special-purpose data acquisition program (LIDAS)
[26], written in Hewlett-Packard (HP) Basic, on an HP 9836 computer system that collects,
downsamples, stores, and displays the data in real time. The data are transferred between the
laser display units and the computer system via two binary-coded decimal (BCD) interface cards
and two specifically designed 64-pin connecting cables.
7.4.1.2 Calibration
The performance of the interferometer system was verified in four separate tests, and was
regularly checked during all experiments. First, the precision of the laser height measurement
was detennined to be 0.075 (am. This was determined from the white, high-frequency region of
measured height change spectra obtained at a sample rate of 5 Hz with the laser.
7-12
-------
Second, the float/tube assembly was checked for calibration by inducing height changes in
the tube itself with precisely known volumes. This was accomplished by connecting the tubes so
that, as measured by the laser, a cylinder of known volume inserted in one tube effects a
displacement of one-half the volume in the other tube. The results of these tests suggested that,
based upon the known geometry of the displacement volumes and of the brass tubes, the
indicated volume change was well within the range of the experimental error of the expected
responses.
Third, a similar calibration of the float/tube assembly conducted as part of each test
consisted of repeatedly immersing a cylinder of known volume into the tank. The height change
indicated by the laser system was then compared to the height change that would be expected
based upon the previously measured tank geometry and measured absolute liquid level in the
tank. Three different known changes in product height were induced in the half-filled steel tank
and were measured by the interferometer to within 2 u,m.
A fourth type of test was conducted periodically with both tubes open to the tank
environment for 24 h. A high degree of coherence between the two channels was observed.
Differential height changes observed between the two interferometers were caused by thennal
expansion of the aluminum stand separating them and were compensated for by making use of
the measured temperature of the laser stand.
7.4.1.3 System Checks Between Calibrations
The test of the laser interferometer height measurement system was made for each
experimental run to determine whether the float/tube system was functioning properly. This was
accomplished by inserting into the tank, and then removing, a solid object of known volume and
measuring the resulting height change. The average height change for multiple in-and-out cycles
was computed and compared to the tank calibration data for the mean product level in the tank.
Typically, a 1,544-ml cylinder was used for the calibration check. The object was completely
immersed for 90 s and then removed entirely for 90 s. Up to 10 cycles were made for each test
run.
7.4.2 Pressure Measurement System
7.4.2.1 Description
Two pressure sensors were employed in the Test Apparatus. One sensor, located at the
bottom of the tank being tested, was used to determine the mean level of product in the tank.
This level was then converted, by means of a tank chart, to a mean volume of the product stored
7-13
-------
in the UST. The other sensor, located on the surface of the Test Apparatus, provided a detailed
record of changes in atmospheric pressure. The piezo-resistive pressure transducers that were
used to make the pressure measurements were selected because of their high resolution,
long-term stability and stainless steel contact surface. Pressure and temperature were measured
using two HP 3421A Data Acquisition/Control Units and an IBM AT computer system. Each
HP 3421A unit scanned up to 30 channels of two-wire sensor data serially with a 5-1/2 digit
multimeter. The data were transferred from the HP 3421A to the computer via an interface bus
(HPIB). A modified version of the HP 3055S data acquisition software (TPDAS) [27] was
developed for acquiring data with two HP 3421A units, and an IBM AT-compatible system was
used for data collection.
7.4.2.2 Calibration
Calibration of the pressure sensor was accomplished monthly by putting a known pressure
on the sensor and measuring the resistance. A known pressure was produced by inserting the
pressure sensor into a long tube filled to a known level with liquid. The calibrations were
performed over the entire measurement range (approximately 4 m for product and 1 m for
atmospheric pressure) in either 15-cm (6-in.) or 30 cm (12-in.) increments, depending on the
sensor being tested. The liquid levels were read visually to the nearest 0.5 mm, and the pressure
readings were automatically recorded to the nearest 0.1 Q. At each pressure, data were collected
at 1 sample/min for a period of 7 min. Calibrations were performed using water. The gasoline
calibration values were generated from those for water calibration, using the density
measurements of the gasoline.
The precision of the pressure sensor was determined by the standard deviation of the
ordinate about the regression line developed from 5-min averages of data at each pressure level.
The accuracy, which depends on the accuracy of the density measurement and the accuracy of
water levels in the calibration tube, was within 1 mm at the completion of the calibration.
7.4.2.3 System Checks Between Calibrations
Before and/or after each experimental run, a gauge-rod measurement of the mean product
level in the tank was made and was compared to the pressure sensor measurements. These data
were used to check the accuracy of the pressure sensor and to refine the calibration for each test,
if necessary.
7-14
-------
7.4.3 Temperature Measurement System
7.4.3.1 Description
The temperature-measurement system, which consisted of up to 51 two-wire, 30,000-£i
thermistors, two HP 3421A Data Acquisition/Control units, and an IBM AT-compatible
computer system, measured product, ground, and air temperature. Two ultra-stable resistors
were connected to each of the HP 3421A units to monitor the electronics.
Product temperature was measured using three vertical arrays of thennistors separated
horizontally, permitting accurate measurement of the average rate of change of temperature of
the product and vapor in the tank. The individual thermistors were spaced vertically at intervals
of 20 cm (8 in.), with the bottom thennistor located 10 cm from the bottom of the tank. For each
array, two thermistors were placed in the 10-cm-diameter fill tube, 10 cm and 46 cm,
respectively, above the top of the tank. Up to 36 thermistors were used to measure the average
rate of change of temperature of the product.
With three arrays it is possible to measure of the horizontal gradients in the temperature
field along the long axis of the tank. Each array measures the temperature of approximately one
third of the product volume. The average volume surrounding each thennistor level in a
30,000-L steel tank is given in Table 7.5. Thus, if the precision of the thermistor is 0.00 TC, the
uncertainty in the product-volume change caused by a temperature change in the vicinity of a
thennistor, even one located at the middle of the tank, is small. The uncertainty for a thermistor
located at the midpoint of a gasoline tank is 0.0019 L (0.0005 gal).
Table 7.5. Estimate of the Volume Surrounding Thennistor Levels in a 30,000-L Steel Tank
2.43 m in Diameter
Thermistor
Height from
Thermistor
Number
1
2
3
4
5
6
7
8
9
10
11
12
Bottom
(m)
0.102
0.305
0.508
0.711
0.914
1.118
1.321
1.524
1.727
1.930
2.134
2.337
(in.)
4
12
20
28
36
44
52
60
68
76
84
92
Volume
Around Each
Ther
(L)
1,204
2,112
2,605
2,915
3,108
3,199
3,199
3,108
2,915
2,605
2,112
1,204
mistor
(gal)
318
558
688
770
821
845
845
821
770
688
558
318
7-15
-------
An array of seven thermistors was used to measure the ground temperature. These data
were used primarily to specify a test condition. The thermistors were placed in a steel tube so
that they could be replaced if they failed. The tube was filled with sand and its top insulated
from surface temperature changes.
Air temperature was measured using a single thermistor. The thermistor was located in a
protected area, so that the sun did not shine directly on it.
A 30,000-Q, two-wire thermistor was used to make all temperature measurements (ground,
air and product). The thermistor, a thermally sensitive resistor, was designed to resolve
temperature changes of 0.001°C and to begin sensing ambient temperature changes of 0.001°C
(as opposed to sensing the temperature changes produced by the electronics), using the HP
3421A. This was accomplished by driving the themiistor with a constant current of only 10 uA.
At this current, a resolution of 1 Q was obtained with the HP 3421A. At 25°C, this corresponds
to a resolution of 0.0008°C. The resolution increases for lower temperatures and decreases for
higher temperatures (Figure 7.2). Each thermistor was placed in a stainless steel sheath, 0.038 m
(1.5 in.) in length, which was then filled with epoxy. The time constant of the thermistors was
only a few seconds. The data were collected at 1-min intervals, using the HP 3421A driven by
an IBM AT compatible computer.
7.4.3.2 Calibration
The characteristics of the thermistors and the accuracy and precision of the system that
measures temperature were estimated during calibration. Calibration of the thermistors includes:
o developing a mathematical curve to define the resistance-versus-temperature
characteristics
o determining the precision and accuracy of the system that measures temperature
All calibration measurements were made using the entire array of thermistors, the two HP
3421A Data Acquisition/Control units, and the IBM AT compatible computer system.
Calibration curves were developed to convert the measured resistance to temperature over
the expected range of temperatures to be encountered during the experiments (0 to 30°C).
Figure 7.2 presents the nominal curve of resistance versus temperature for a 30,000-Q
thermistor. The relationship between temperature and resistance is highly nonlinear. The
equation,
R)2 + a3(lnR)' (7.2)
7-16
-------
where T is in degrees Kelvin and R is in ohms, has been determined to accurately represent the
mathematical relationship between temperature and resistance for a
negative-temperature-coefficient thermistor. The coefficients (a^) were used to convert the
resistance measurements made in ohms to temperature in degrees Celsius.
The calibration curves were developed by comparing the resistance measurements made by
each thermistor to a reference thermometer, which was referenced to the triple point of water.
Repeated measurements of the triple point of water can be made to within 0.0002"C. The
reference thermometer, an HP Quartz thermometer, has a nominal accuracy of 0.04°C and a
precision of O.OOOrC. The thermistor measurements were also referenced to two ultra-stable
resistors, which indicates the presence of any drift in the electronics over time and permits the
calibration curves of the individual thermistors to be adjusted accordingly. Resistors stable to
0.5 fi or less are required; resistors stable to within 0.1 Q were used.
150,000
E
.c
O 100,000
8
OJ
CD
QC
"ro
- 50,000-
~~i i i i i i i r~i i i i i i i i i—i—i—i—i—i—i—i—r~i—i—i—i—i—i—i—i—i—i—:—i—r
-5 0 5 10 15 20 25 30 35
Temperature - deg C
Figure 7.2. Nominal resistance-versus-temperature curve for a 30,000-ft thermistor (YSI
44008).
7-17
-------
The calibration procedure was performed monthly over the duration of the program.
During the life of this project, three procedures were used for calibration. All involved placing
all of the thermistors and the reference thermometer into an insulated constant-temperature water
bath. The methods used to change the temperature of the bath evolved during the course of the
program. In the first method, simultaneous measurements of resistance and bath water
temperature were made for periods of 15 to 20 min at approximately 8 to 12 distinct bath
temperatures. The bath temperature was changed or "stepped" in increments of approximately
2.5 to 5°C over the range of 0 to 30°C. The bath temperature was changed by adding warmer or
cooler water to the existing bath. At each water bath, the temperature was allowed to reach
thermal stability prior to collecting calibration data. This was accomplished by mixing the water
thoroughly for 15 to 20 min before the calibration data were collected. The coefficients in
Eq. (7.2) were estimated using a least-squares fit to the mean temperature and mean resistance at
each nominal bath temperature.
An estimate of the accuracy, the inter-thermistor variability, was obtained by computing
the standard deviation of the temperatures measured by all the thermistors for each temperature
bath.
An estimate of the precision (or repeatability) of each thermistor was made for each bath
temperature from the standard deviation of the ordinate (i.e., temperature) of a linear regression
curve fit to the temperature and resistance data.
The procedure was improved in March 1987. Instead of sampling at 8 to 12 distinct
temperature baths, an attempt was made to sample evenly throughout the temperature range.
This was done primarily to reduce the manual input required to perform the calibrations and to
reduce the risk of underdetermining the resultant curve fit. The procedure involved placing the
thermistors and reference thermometer into an insulated ice-water bath and allowing a drift to
ambient temperature to take place via conductive heat transfer. As the temperature of the water
bath approached the ambient air temperature, the rate of temperature drift began to approach
zero. At this point, it was necessary to add hot water (that is, to "step" the temperature bath as
was done in the previously described calibration method) in order to collect data through the
required temperature range. The resulting data set was then downsampled to produce a set
containing approximately 250 data points. The coefficients of Eq. (7.2) were determined by a
least-squares solution using the downsampled resistance-temperature data set.
The third method of thermistor calibration employs a microprocessor-controlled
temperature bath that maintains a constant rate of change and that was first used in
October 1987. The thermistors were placed, along with the reference thermometer, into an
7-18
-------
insulated water bath, which, in turn, was placed into a thermally controlled water bath. The
insulated water bath drifted at a controlled rate of approximately
0.01°C/min through a range of 2.5 to 27.5°C. Once again, the resulting data set was
downsampled to produce a set consisting of approximately 250 data points, and the coefficients
of Eq. (7.2) were determined by a least-squares solution using the downsampled
resistance-temperature data set.
When using either of the latter two calibration methods, an estimate of the precision of a
thermistor is given by the standard deviation of the ordinate (i.e., temperature) of the regression
curve fit of Eq. (7.2) to the temperature and resistance data calculated for each contiguous 0.1'C
block of data over the entire temperature range. Two estimates of accuracy were made. To
show the inter-thermistor variability, the first estimate was made, as before, by calculating the
standard deviation of the temperatures measured by all of the thermistors at each of the 250
temperature points. The second estimate was made to determine the accuracy of each thermistor.
This was achieved by computing the difference in the mean temperature between each 0.1'C
segment of data and the calibration curve (offset).
Any thermistors that did not meet the precision and accuracy specifications over the entire
temperature range (excluding calibration artifacts) were either physically replaced or were not
used in the analysis. Typically, the thermistors that needed replacement had been identified
before the monthly calibration. These thermistors, in addition to any that were suspected of poor
performance, were replaced before the calibration. Thermistors that developed problems
between calibrations were not used in the analysis, and their data were replaced by averaging the
data from the thermistors vertically or horizontally bracketing them.
As a post-calibration analysis, the drift of the thermistors from the previous month's values
was calculated for a nominal value of 25,000 Q. (nominal temperature of 25*C). This analysis
gave an indication of how the performance of the thermistors had changed between calibrations.
7.4.3.3 System Checks Between Calibrations
Two checks of the temperature measurement system were made for each test. First,
resistance measurements of the ultra-stable resistors during the thermistor calibration check were
made throughout the test to determine whether there were electronic problems. If resistance
changes of approximately 2 Q occurred during any 1-h period, it was determined that the
electronics were generating enough heat to affect the measurement of temperature of 0.001'C. If
this occurred, the HP 3421A was replaced. In general, the electronics were robust and did not
7-19
-------
affect the temperature measurements. Second, a visual check of the time series of all of the
thermistors was made after each test, also to identify any obvious problems. A more in-depth
analysis was performed for thermistors which exhibited anomalous behavior.
7.4.4 Leak Inducement System
7.4.4.1 Description
Leaks were generated by withdrawing liquid from the tank (or adding it) using a
variable-speed metering pump. The product was weighed and the leak rate calculated using the
calibration curve. Leak rates can be resolved to 0.001 L/h or better. The magnitude of the leak
was given by the slope obtained through least-squares regression analysis.
7.4.4.2 Calibration
Calibration of the leak inducement system was performed initially prior to installation, and
was checked periodically during the test program. The initial calibration of the scale was
performed by its manufacturer, using NBS traceable reference weights. The entire assembly
(including scale, pumps, piping, and product-receiver tank) was calibrated upon its completion.
NBS traceable weights were again used for this calibration. Readings were taken at increments
covering the complete range of operating conditions. A calibration curve was then generated by
means of a least-squares curve fit, and the precision and accuracy of the assembly were
estimated.
The variable-speed metering pump, used to withdraw product from different locations in
the apparatus, was initially calibrated with gasoline. The pump was operated at eight different
speeds covering the entire anticipated operating range. For each speed, the discharge liquid was
collected in a graduated cylinder for a period of 3 min. Five separate readings were made at each
operating speed, and the results averaged to produce a single value. A least-squares curve fit to
these data allowed the desired leak rate to be established subsequently during a particular test,
while simultaneously allowing an estimate of the variance associated with pump performance to
be made.
The accuracy of the pump's flow rate was checked by means of a high-precision industrial
scale capable of resolving weight changes as small as 0.0001 kg (0.0002 Ib). For the current set
of tests, induced leak rates of unleaded gasoline ranging from 0.04 L/h to 7.6 L/h (from
0.01 gal/h to 2.0 gal/h) were anticipated. This corresponds to weight changes ranging from
approximately 0.028 kg/h (0.062 Ib/h) to 5.63 kg/h (12.4 Ib/h). Based upon these flow rates, the
most demanding precision required for the weighing apparatus occurs at the low flow rate of
7-20
-------
0.04 L/h. Ideally, it is desirable to establish the flow rate within a 15-min measuring period. As
a consequence, a precision of the weighing apparatus of approximately 0.0014 kg (0.003 Ib) was
required.
7.4.4.3 System Checks Between Calibrations
Calibration checks, also using NBS traceable reference weights, were performed
periodically in order to ensure that the scale was functioning properly. Checks were made by
placing known weights on top of the receiver tank and subsequently recording the scale
response. Deviations from the known value of the indicated weight were used to identify
potential scale problems, and to enable rapid corrective action to be taken.
7.5 Underground Storage Tanks
In order to convert product-level measurements to corresponding volume changes reliably,
calibration was required for each tank. For the current tests, two means of determining the tank
charts were devised. First, the inside dimensions of both the steel and fiberglass tanks were
measured in detail after installation but before adding product. These measurements were used
to calculate the volume of the tank as a function of height from the tank bottom. The results of
these calculations were then compared with the manufacturer's tabulated values. Four
measurements of diameter at 45° intervals around the longitudinal axis of the tank were made at
each often axial locations. The measurements were made to the nearest 0.32 cm (0.125 in.),
using a steel tape. A theoretical tank chart was generated to compute tank volume at any depth
in the tank, assuming that the tank was a right regular cylinder. For the steel tank, for example, a
diameter of 242.89 cm (95.625 in.) and a length of 649 cm (21 ft 3.5 in.) were used to generate
the chart. The total capacity of the tank was estimated to be 30,107.8 L (7944 gal). An error of
± 0.32 cm (0.125 in.) in diameter would result in disparate tank volume estimates of 30,028.2 L
(7923 gal) and 30,183.6 L (7964 gal), respectively. An error of ± 0.32 cm (0.125 in.) in tank
length would result in tank volume estimates of 30,092.6 L (7940 gal) and 30,119.1 L (7947 gal),
respectively. The maximum error in estimating tank capacity from these measurements is 95 L
(25 gal).
Second (and in conjunction with these physical measurements), the laser interferometer
was employed to determine a height-to-volume conversion factor for product at any level in the
tank. This was achieved by measuring the gross level in the tank (either with the product
pressure sensor or a dipstick) and then alternately inserting and removing a cylinder of known
volume and noting the corresponding product-level changes. The largest uncertainty in the
height-to-volume calibration factor occurs when product level is between 231.14 and 242.89 cm
7-21
-------
(91 and 95.625 in.) in the tank. An extensive set of measurements at increments of 1.27 cm
(0.5 in.) was made in order to generate an experimental curve. The theoretical and experimental
height-to-volume curves for the steel tank are shown in Figure 7.3.
A height-to-volume calibration was performed routinely after each test run and was
compared to the calibration factor that would be expected based upon the measured tank
geometry and gross product level; this was done to verify that the laser interferometer
measurement system was functioning properly and to obtain the actual height-to-volume
conversion factor for that test run.
80
-------
8 Ambient Noise Experiments
The objective of the ambient noise experiments conducted at the UST Test Apparatus was
to support the test method performance evaluation and to validate the ambient noise models.
Four of the ambient noise sources are, in one way or another, dependent on a temperature
fluctuation. Temperature fluctuations in the product or in the vapor space are driven by a
differential between the temperature of the backfill and soil and that of the product; this
differential is incurred immediately after product addition. Neither surface-wave effects nor the
most important structural deformation effects are not temperature-driven, and, after model
validation, their impact can be assessed by exercise of the simulation. As a consequence, the
Test Apparatus experiments were designed to sample the product temperature and level under a
variety of conditions. The insights gained during the modeling and measurement program
represent a significant contribution toward improving the performance of volumetric tank tests.
The emphasis of the ambient noise experiments was placed on the effects of
product-temperature changes (including those that occur after product deliveries and topping the
tank), structural deformation, and trapped vapor. Some of the effects of product temperature
have been described previously in [28, 29], and some of the effects of tank deformation have
been described in [23].
8.1 Product-Temperature Changes
Thermal expansion and contraction of the product in the tank can be inferred from a direct
measurement of temperature using three thermistor arrays. In order to find the volume
fluctuations produced by thermal fluctuations, it is necessary to make an accurate measurement
of the product density. The coefficient of thermal expansion is determined in the laboratory from
product density measurements as a function of temperature. The range of thermal
product-volume fluctuations experienced by UST systems in the United States can be
approximated by conducting experiments where the differential temperature between added and
existing product is varied. The magnitude of the modeling error is estimated by overfilling the
tank, removing any vapor pockets that may be present, waiting until structural deformation
effects have become small, and comparing the measured volume changes to the modeled volume
changes.
An analysis of the temperature fluctuation that occurs in an overfilled tank during the first
24 h after a delivery was performed using temperature data collected in the steel tank of the Test
8-1
-------
Apparatus. To conduct each run, 15,000 L (4,000 gal) of product, at temperatures that were 0 to
10°C cooler or wanner than the temperature of the ground and of the in situ product, was added
to the half-filled tank.
Thirteen 24- to 48-h tests were analyzed in order to characterize the temperature field and
volume changes generated by temperature fluctuations. All tests were conducted with the fluid
level between 233.7 and 238.8 cm (92 and 94 in.) in the 242.9-cm (95.6-in.)-diameter tank.
Testing at this level permits the approximation of temperature changes in a full tank while
simultaneously avoiding the incidental trapping of vapor pockets.
8.1.1 Product Temperature Analysis
Three analyses were performed. The first estimated and tabulated the magnitude of the
thermal volume changes as a function of time after delivery. Temperature-volume changes were
computed using the volumetrically weighted measurements from all submerged thermistors on
the three vertical arrays. Results for data collected on 28 October 1986 are presented in
Table 8.1 and Figure 8.1.
Table 8.1. Thermal Volume Changes Estimated from a 1-h
Block of Data as a Function of Time after Delivery
(28 October 1986)
Time
(h)
3
6
9
15
21
Temperature Volume
Change (L/h)
2.532
1.900
1.459
1.026
0.841
This analysis, which included all 13 tests, indicated that thermally induced volume changes
are large even 24 h after product has been added to the tank, and that temperature compensation
is necessary in order to conduct an accurate tank test. In the course of a 24-h test,
uncompensated volume rates of 0.8 to 2.5 L/h were observed, as illustrated in Table 8.1.
The second analysis characterized the horizontal spatial inhomogeneities of the
temperature field. The fundamental issue was whether one thermistor array is sufficient to
characterize the temperature field of the whole tank. The analysis was conducted by:
8-2
-------
40
30 —
CD
E
I
I
CD
Q.
CD
20 —
10 ~
0 5 10 15 20
Time after Delivery - h
Figure 8.1. Temperature volume time series (28 October 1986).
o generating a thermal volume fluctuation time series sampled once per minute for each
thermistor array as well as for the average of the three arrays
o subtracting the average thermal volume time series from each array's volume time
series
o calculating the slope (i.e., volume rate) of the thermal volume time series by fitting a
least squares line to 1-h blocks of data updated every minute
o differencing the volume rate (slope) time series of each array from the average of all
three arrays
Results suggest that when testing is begun at least 4 to 6 h after product delivery, a single
vertical array of thermistors having a vertical spacing of 20 cm is sufficient to characterize the
temperature field of the whole 30,000-L tank. In the first 4 to 6 h, large differences in
temperature between the three horizontally spaced arrays are evident. In this interval, then, three
arrays are not sufficient to characterize the temperature field. After 6 h, the differences in the
rate of change of temperature between arrays, expressed as a volume, is small. This is illustrated
in Figure 8.2 in a time series plot of residual fluctuations, or differences, in the rate of change of
temperature volume between Arrays 2 and 3. (The added product was 5.6°C cooler than the
in situ product). During the first 6 h, volume-rate differences larger than 0.5 L/h and smaller
8-3
-------
than 0.1 L/h were observed. Conducting a test during this period may lead to erroneous results.
The temperature field is stable after 6 h, that is, after the horizontal temperature gradients along
the long axis of the tank become small.
1.0
0.8
0.6 —
(D
13
O 0.4
0.2
0.0
-5
0 5 10 15
Time after Delivery - h
20
25
Figure 8.2. Differences in residual fluctuations in rms temperature volume between Arrays 2
and 3 (28 October 1986).
For this test, the average difference in slope for each array, after 6 h have elapsed, is
summarized in Table 8.2. This analysis suggests that 6 h after product addition a single array
can measure the temperature changes to within 0.04 L/h (0.01 gal/h), the precision required to
reliably detect leak rates of 0.19 L/h (0.05 gal/h).
8-4
-------
Table 8.2. Average Slope Differences for Three Thermistor Arrays
Time(h)
6
8
12
16
20
Array 1
0.0099
0.0047
0.0089
0.0266
0.0083
Average Slope Differences
Array 2
0.0055
0.0078
0.0269
0.0114
0.0118
Array 3
0.0154
0.0031
0.0181
0.0102
0.0201
A third analysis was undertaken to determine the adequacy of the 20-cm vertical separation
of thermistors. The mathematical coherence between thermistor pairs was computed and the
coherence measured to unity for frequencies less than 1/2 cycle per hour. The result is illustrated
in Figure 8.3, which shows the data from two thermistors located near the middle of the tank.
Because of the very high level of coherence for all fluctuations with frequencies lower than 0.02
cycles/min (periods longer than 5 min), it was concluded that a 20-cm spacing between
thermistors provides an adequate characterization of the temperature field. Similar results were
often observed for a 40-cm spacing. Even a 20-min spacing is not adequate when very steep
gradients in the temperature field occur, or at boundaries between regions where temperature is
increasing or decreasing.
These results have significant implications for tank testing. First, temperature
compensation is essential to the conduct of an accurate tank test; without compensation, volume
changes of 1- to 3-L/h were observed. Second, independent of structural deformation effects, a
waiting period of at least 4 to 6 h after topping a partially filled tank is required before testing
can begin. Third, for moderately severe initial thermal conditions, a single array of thermistors,
at the fill hole, is sufficient for temperature compensation.
8-5
-------
d
TJ
'c
o
2
I
o
O
Time: 18.0 to 42.1
9 Point Smoothing
IOE-04 10E-03 10E-02 10E-01 I06.OO
Frequency - cycles/min
Figure 8.3. Coherence between two thermistors (18 and 19), located near the middle of the tank
on Array 2 of the Test Apparatus. Twenty-four hours of data were used in the analysis. There
are 7 degrees of freedom in the estimate.
8.1.2 Model Validation
To determine the thermally induced volume fluctuations in a tank, accurate estimates are
necessary for the mean change of the temperatures, the coefficient of thennal expansion, and the
total volume of product in the tank. The uncertainty in the computed thennal volume change
depends on the spatial homogeneity of the product temperature and on the errors in measuring
the aforementioned quantities. In a full 30,000-L (8,000-gal) tank, with thermistor precision
equal to O.OOTC, the error in the thermal expansion coefficient is 0.000016/°C, and the error in
volume is 95 L. The maximum error resulting from a 0.05°C/h change in temperature would be
the 0.067-L/h (0.018-gal/h) error in the estimate of product-volume change.
The ability to compensate for thermally induced product-volume changes at the Test
Apparatus was estimated from the residual fluctuations in the volume after the thennally induced
volume changes had been subtracted. The product temperature experiments demonstrated that a
single, vertical array of thermistors at the fill hole was sufficient to estimate the mean rate of
change of temperature to within 0.04 L/li (0.01 gal/h). Figure 8.4 is an example of the residual
8-6
-------
fluctuations that remain after temperature compensation; these are approximately 0.04 L/h. The
smallest residuals were obtained during overnight experiments. The residuals were usually
manifested as linear trends.
18.40
18.20
18.00
17.80
17.60
10
15
20
25
30
35
Time - h
Figure 8.4. Residual fluctuations in volume after compensating for thermally induced
product-volume changes.
8.1.3 Topping
In all methods in which the tank is overfilled, the filling process has two steps. The first is
to fill the tank with sufficient product to raise the level as close to the top as possible, without
letting the product rise into the fill tube. The exact level is variable depending on safety
considerations and on the accuracy of the tank chart used to estimate the required volume of
product. The second step is to top the tank to the product level required for the test. As little as
10 to 15 L (3 to 4 gal) or as much as 175 to 350 L (50 to 100 gal) may need to be added.
Because the temperature of the added product is almost always different from the temperature of
the product in the tank, the product added in the second step will affect the accuracy of the
product temperature estimates required for temperature compensation as discussed below. The
impact of topping the tank was observed frequently during the Field Verification Tests.
8-7
-------
Experiments were conducted in the fiberglass tank to investigate the effects of topping.
Three vertical thermistor arrays were deployed to monitor the temperature field in the vicinity of
the fill tube where the product was added. One array was inserted in the fill tube and the other
two arrays were inserted on either side of it, approximately 75 cm (30 in.) away. This array
configuration accounts for any horizontal gradients that might develop after topping. The tank
was initially overfilled to a level within the fill tube more than 24 h before the start of the test.
Approximately 19 L (5 gal) of product, either 7°C cooler or warmer than the mean temperature
of the product in the tank, was added to raise the level an additional 60 cm. This product
addition would effect a mean change in temperature of 0.004°C for the product in the tank. It
should not, however, affect the mean rate of change of product temperature being driven by the
mean temperature of the product in the tank and the backfill-soil.
The results of adding cold and warm product to the tank are presented in Figures 8.5 and
8.6, respectively. Both plots display the temperature measured by 11 of the thermistors located
on a single array at the fill tube for at least 0.5 h before topping and 4.5 h after topping. These
thermistors are located 11 cm, 31 cm, 52 cm, 72 cm, 92 cm, 113 cm, 133 cm, 153 cm, 174 cm,
194 cm and 214 cm from the bottom of the tank. The mean temperature of the layer at each
thermistor location was removed and plotted with an offset of 0.02°C greater than the layer
below. This allows for the change in temperature at each layer to be compared to the others.
O
CD
O)
c
O 0.20 H
0)
Q_
E
CD
0.00
Time - h
Figure 8.5. Effects of topping the tank with colder product, as represented by 11 thennistors
located on an array at the fill tube of the Test Apparatus's fiberglass tank. The numbers next to
each line indicate the height of the thermistors, in centimeters, from the bottom of the tank.
8-8
-------
0.35
0>
o.oo
Figure 8.6 Effects of topping the tank with warmer product, as represented by 11 thermistors
located on an array at the fill tube of the Test Apparatus's fiberglass tank. The numbers next to
each line indicate the height of the thermistors, in centimeters, from the bottom of the tank.
Several observations about both tests are noteworthy. First, the temperature field before
topping exhibited a small but stable change in temperature. After the product had been added,
the temperature field became highly disturbed for 2 to 3 h before re-approaching the pretest
temperature conditions. The temperature-volume fluctuations inferred by the two outer arrays
were found to be similar to one another but different from those measured by the fill tube array.
The symmetry was not perfect, probably because the addition of product sets up a flow in one
direction or another. Third, the data suggested that an accurate estimate of the mean rate of
temperature throughout the tank, required for thermal compensation, cannot be made until after
the temperature field has stabilized. Fourth, if the temperature field is vertically undersampled, it
is difficult to identify the time at which the horizontal and vertical temperature gradients have
dissipated. A minimum of five thermistors would be required in order to do this.
The temperature field changes with the addition of product that is cooler or warmer than
the product in the tank. For both tests, the magnitude of the high-frequency fluctuations
increases. The colder product (Figure 8.5) sinks and spreads out spatially at a level where the
buoyancy forces balance the gravity forces. It took approximately 9 min after adding colder
product for the bottom layer of the tank to be affected, as shown by the reaction of the thennistor
8-9
-------
located 11 cm from the bottom of the tank. The addition of colder product sets up a steep
thermal gradient in the upper layers, probably from the entrainment of warmer product near the
top of the tank down to layers which are cooler.
In Figure 8.6, it is observed that the force generated during the addition of warm product
initially moves this warm product toward the bottom of the tank, but that warm product returns
toward the top, where it heats the surface layers for a short period of time.
8.2 Trapped Vapor
A set of experiments was designed and conducted at the UST Test Apparatus to determine
whether Eq. (6.8) is a satisfactory equation of state both for predicting pressure-related changes
of volume in trapped vapor and for determining whether a method of estimating the volume of
trapped vapor could be devised using the equation of state. The same basic experimental
approach was used for both of the above objectives. The experiments were designed to estimate
the volume of a known amount of trapped vapor in an overfilled tank by varying the pressure in a
predetermined way. The experimental design, as well as interpretation of the data, were
dependent on the magnitude of the instantaneous structural deformation of the tank. It was
initially assumed that the instantaneous structural deformation was small. When it was found
that this was not the case, many additional experiments were designed and conducted to estimate
the magnitude of the instantaneous deformation in terms of A^.
8.2.1 Experimental Approach
A set of iterative experiments was designed and conducted to validate Eq. (6.11) and to
develop and test a method of estimating the total volume of the trapped vapor in the tank. The
approach was straightforward, but the experimental implementation and interpretation were not.
Two types of experimental configurations (Figures 8.7 and 8.8) were used to conduct the
experiments. In the first type, the tank was overfilled to a level within the fill tube, and an effort
was made to remove all trapped vapor, as if a volumetric test were to be conducted. In the
second, a sleeve was inserted into the fill tube of the tank to trap a known volume of vapor. For
both configurations, a pressure change was produced by rapidly inserting a cylindrical bar of
known volume into the tank. The height-to-volume conversion factor, Acff, was computed
directly from the bar volume, Vbar, and the measured product-level change, Ah. Validation of the
model and accurate estimates of the volume of trapped vapor in the tank require that Ajsd « Avp
in Eq. (4.2) or AVlsd« AVvpin Eq. (6.12). If the volume changes produced by the instantaneous
structural deformation of the tank are large, neither the validation nor the method of making
volume estimates will be accurate.
8-10
-------
Fill Tube
Final Level
(No Trapped Vapor Prasant)
Final Level
(Trapped Vapor Prasant)
Initial Level
Cylindrical Bar of
Known Volume (V bar
Fiberglass Tank
}AV =Vbar - A Ah
Ah
Vbar
A
Trapped Vapor
Product
Figure 8.7. Experimental configuration for trapped vapor tests in a well-bled fiberglass tank. A
cylindrical bar of known volume, Vbaf, was used to raise the level of product.
Sleeve
Final Level
(No Trappad Vapor Prasant)
Final Level
(Trappad Vapor Prasant)
Valve
Cylindrical Bar of
Known Volume (V bar )
Fiberglass Tank
Fill Tube
(saalad at junclura of slaava)
AVyp
ir *
Ah
'bar
Vapor
Initial
Level
Product
Figure 8.8. Experimental configuration for the trapped vapor tests in the fiberglass tank. A
known amount of vapor was trapped using a sleeve that was extended into the tank. The volume
of trapped vapor was estimated from an experimentally determined tank chart. It is assumed that
the change in product level in the tank after a bar of known volume has been inserted is
negligible. The valve is opened and closed to the atmosphere immediately before starting the
test.
8-11
-------
8.2.1.1 Procedure
Product-level changes produced by vapor pocket fluctuations can be measured by using the
following procedure:
1) Fill the tank to a specified level so that the desired volume of vapor can be trapped in
the tank.
2) Observe a waiting period that is long enough to ensure that the structural deformation
effects resulting from filling the tank have become negligible.
3) For the configuration shown in Figure 8.7, overfill the tank with product to a level
within the fill tube, while once again observing an adequate waiting period for
structural deformation effects to subside.
4) For the configuration shown in Figure 8.8, insert a sleeve in the fill tube to trap the
vapor.
5) For the configuration shown in Figure 8.8 only, open a bleed valve to allow the tank
to come to atmospheric pressure; then close the valve and raise the product level in
the sleeve.
6) For both configurations, insert a cylindrical bar of known volume into the tank
through the fill tube to change the product level; then measure the product-level
change.
In order to minimize the effects of the large product-level fluctuations and the exponential
deformation of the tank that occur upon insertion of the bar, the following procedure was used:
measure the product level; insert the bar over a 45-s period starting 15s after the measurement;
measure the product level 15s after the bar has been inserted; measure the product level 2.5 min
after the last measurement; wait 15s and remove the bar slowly over the next 45 s; measure the
product level; and vent the tank to the atmosphere in preparation for another cycle.
The tank was modified for these experiments so as to allow control of the size of the vapor
pocket. A sleeve, extending 61 cm (24 in.) into the body of the tank, was inserted into the fill
tube, which was then sealed at the juncture of the sleeve. The tank was then filled to a
predetermined height below the top of the tank but above the bottom of the sleeve. Product
levels for these experiments ranged from 200 to 232 cm (79 to 91 in.).
8.2.1.2 Experiments
In order to validate the model described by Eq. (6.11), the experimental configuration
shown in Figure 8.8 was used. The experiment, which was designed to predict a known volume
of vapor trapped in the tank, required that the values of n and Vvp be known, and that the volume
change produced by the instantaneous structural deformation be small compared to the volume
change of the vapor (i.e., A,,d« Avp). Experiments were conducted to estimate the volume in
8-12
-------
the upper 15 cm of the tank; this was done in 1-cm intervals to account for any tilting or
distortion of the tank geometry. Laboratory experiments were conducted to estimate the value of
n for the product in the UST Test Apparatus tanks. To minimize the effects of the instantaneous
structural deformation, a large vapor pocket was trapped for the experiment. The experiment
was designed to limit the effects of Ajsd to less than 5 to 10% of Avp To determine what size
vapor pocket was required, a set of preliminary experiments was performed in a well-bled,
overfilled tank using the configuration shown in Figure 8.7. The results of the experiment gave
an upper bound on the value of AjSd or Avp. Different bar sizes and different vapor pocket
volumes were used. It was assumed that A,,,, was a constant for small changes in the initial
product level. This assumption was verified by inserting, at the same initial level, three or four
bars of different volume to produce a product-level change. Eq. (6.1 1) was solved using a value
of n = 1 and the volume of vapor estimated from the tank geometry. The volume change
produced by the structural deformation was calculated by
(8.1)
8.2.2 Results
Although many vapor pocket experiments were conducted in both the fiberglass and steel
tanks, the only results reported here are those from a limited number of experiments in the
fiberglass tank.
8.2.2.1 Estimate of the (ias Constant
It was initially hypothesized that the volume changes produced by rapidly inserting the bar
into the fill tube would occur adiabatically and that n (see Eq. (6.1 1)) would be equal to 1.4. A
set of experiments was conducted in a rigid glass bottle to determine n. Figure 8.9 shows the
experimental setup. The experiments were first conducted with water and then with the product
used in the Test Apparatus tanks. A known volume of liquid was placed into a bottle of known
volume. The liquid was then raised into the tube of known diameter by suction, while the vapor
space was kept at atmospheric pressure. The level of the liquid was measured, and the suction
pressure was released after the opening to atmospheric pressure had been sealed. The liquid in
the tube fluctuated for several seconds until it reached a stable level. The level was then
measured, and an estimate of n was made using Eq. (6. 1 1 ). The results for water and gasoline
are given in Table 8.3. The value of n = 0.99 ± 0.06.
8-13
-------
L,
0.35-mm (0.25-in.) O.D. tube
Product Level
(valve open)
Product Level
(valve closed)
Valve
(open or closed to atmospheric pressure)
Stopper
Vapor Pocket, V
vp
Liquid
Bottle
Figure 8.9. Experimental apparatus to estimate n. The initial pressure on the vapor pocket was
atmospheric pressure, PA. The change in the volume of the vapor pocket was estimated from the
drop in product within the tube with cross-sectional area A,ube.
Table 8.3. Results of Laboratory Experiments to Estimate n (Bottle Volume = 803 ml)
Product
Water
Water
Gasoline
vvp
(ml)
179
330
260
Number of
Tests
6
3
8
Mean Standard Deviation
n
1.017
1.039
0.990
n
0.021
0.004
0.057
The results clearly indicate that n is closer to 1.0 than to 1.4, and that the process was not
adiabatic. Also, the oscillating behavior after the suction pressure on the liquid in the tube had
been released suggested potential measurement problems in full scale experiments. Furthermore,
the product level in the tube began to rise as the partial pressure of the air-gasoline vapor
increased.
8-14
-------
8.2.2.2 Tests in Well-Bled Overfilled Tanks
Many overfilled-tank tests were conducted to determine the bound on the instantaneous
volume changes induced by vapor pockets and tank deformation. A special effort was made to
remove all sources of vapor from the tank and associated piping by means of bleed valves that
were placed at critical locations in the system. The best that could be done was to reduce the
vapor pocket to 10 L. However, the impact of this residual was determined to be negligible, as
discussed in Section 6.4.2.
Table 8.4 summarizes the results of the tests in a well-bled, overfilled tank. This
experimental configuration is shown in Figure 8.7. Figure 8.10 illustrates the product-level
changes produced during a set of experiments conducted on 6 June 1987 by inserting and
removing bars of different size (625, 953,1551, and 2477 ml). The product-level measurements
were made with a ruler to the nearest 3 mm. The value of A,.rf as a function of bar volume is
shown in Figure 8.11.
Table 8.4. Summary of the Results of the 6 June 1988 Overfilled Fiberglass Tank Tests
(Product depth 335.3 cm)
Bar
Volume
(ml)
625
953
1551
2477
Ah
(cm)
5.20
7.77
12.67
20.40
(cm2)
120.2
122.7
122.4
121.4
A
(cm2)
81.1
81.1
81.1
81.1
(cm2)
39.0
41.6
41.3
40.3
Vvpfor
(L)
55
58
58
57
Avp for
Vvp=l()L
(cm2)
7.0
7.1
7.1
7.0
Aijd for
VTp=l()L
(cm2)
32.0
34.5
34.2
33.3
Several observations about the data are noteworthy. First, product-level changes induced
by the exponential deformation of the tank can be seen where the two largest bars were inserted
(Figure 8.10). Second, it can be seen from Figure 8.11 that the values of Aeff and A,,,, + Avp are
constant over the range of product-level changes. Since it can be shown from Eq. (6.11) that Avp
is a constant over a wide range of pressure changes, it can be concluded that Ajsd is also constant.
The uncertainty in the measurement is indicated by the spread in the individual values of Aeff for
each bar.
The results shown in Table 8.4 indicate that the maximum amount of vapor that can be
trapped in the top of the tank is approximately 60 L; this assumes that the instantaneous
8-15
-------
structural deformation is zero. Based on geometric considerations, it is estimated that the tank
should contain no more than 10 L of vapor. If it is assumed that the quantity of trapped vapor is
10 L, Ai3d = 35 cm2.
The causes of trapped vapor were minimized in the Test Apparatus by paying careful
attention to certain details. All manways were installed so as to prevent the trapping of vapor.
All lips from fill tubes extending down into the tank were eliminated. Since the tank was
installed horizontally, the only remaining variable that could be responsible for trapping vapor
was the undulation of fluid in the top of the tank. A 1.25-cm (0.5-in.) layer of vapor along the
surface would account for 19 L of vapor, but because of the irregularity of the surface, it is
reasonable to assume that the amount of vapor actually trapped was less than 50% of this figure.
The steel tank affords a more reliable estimate than does the fiberglass tank. In the former, water
marks left near the top of the tank as a result of the water tests (conducted before the tanks were
filled with product) are still visible; these show the exact location and quantity of the vapor
pockets. Calculations in both the steel and fiberglass tanks suggest that it is possible that less
than 10 L could be trapped.
38
E
O
S
•o
O
36 -
34
32 -
30 -
28 -
26
22 -
20 -
18 -
16 -
14
9.0
9.2
9.4
—i—
9.6
9.8 10.0
Time - h
10.2
10.4
10.6
Figure 8.10. Time series of the product-level changes in the fill tube of an overfilled fiberglass
tank on 6 June 1987, produced by inserting and removing four different- size bars (625, 953,
1551, and 2477 ml). The measurements were made with a ruler to the nearest 3 mm. The tank
was well bled prior to testing in order to minimize the total volume of trapped vapor in the tank
system. The initial product level was 335.3 cm above the bottom of the tank. The time interval
between points is approximately 3 min.
8-16
-------
CJ
E
o
i
^
127 -
126 -
125 -
124 -
123 -
122 -
121 -
120 -
119 -
118 -
•M7 -
" .
•
" .
'
--r— 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 I " ™!
600
1200 1800
Bar Volume - ml
2400
Figure 8.11. \ff computed from the volume of the cylindrical bar, Vbar, and the measured
product-level change data, Ah, obtained on 6 June 1987. The tank was well bled in order to
minimize the total volume of trapped vapor in the tank system. The initial product level in the
fill tube was 335.3 cm above the bottom of the tank.
If the maximum quantity (60 L) were trapped, the location of the vapor pocket would have
to be in the upper 3 cm of the tank, and this, given the geometric considerations, is impossible.
Similarly, to trap 60 L of vapor in a pipe would require that the status of a length of pipe 7.4 m
long and 10 cm in diameter be unknown. Because all pipes have been filled with product and
bled of trapped vapor through valves, it is unlikely that the piping conditions would be unknown
to such a large extent.
After considerable experimentation, it was concluded that large changes in product level,
since they are evidently not the result of trapped vapor, are due to the instantaneous deformation
of the tank. Because the effects of trapped vapor and deformation are so similar, experiments
were designed and conducted to estimate both.
8.2.2.3 Trapped Vapor Tests
An 821-L vapor pocket was trapped in the top of the tank using the sleeve shown in
Figure 8.8. Table 8.5 summarizes the results of the tests. Figure 8.12 illustrates the
product-level changes produced by inserting and removing bars of different size (625, 1551,
2477, and 5071 ml). The value of Aeff as a function of bar volume is shown in Figure 8.13.
8-17
-------
Table 8.5. Summary of the Trapped Vapor Tests Conducted in the Fiberglass Tank on
5 June 1988*
Bar
Volume
(ml)
625
1551
2477
5071
Ah
(cm)
0.82
2.30
3.65
7.52
AeIT
(cm2)
765.0
674.4
678.6
674.6
A
(cm2)
71.4
71.4
71.4
71.4
A-Af A
hd T r\vp
(cm2)
693.6
603.0
607.0
603.2
VVP
(L)
963 **
838
845
842
* The calculations were done by using Eq. (6.11) with n = 1 and assuming atmospheric pressure
was 13.90 m of gasoline. The product depth was 218.1 cm. A 821-L vapor pocket was trapped
in the tank and fill tubes.
** Values not used in the analysis
39
-a
o
38 -
37 -
36
35 -
34
33 -
32
31
30
AAA
17.4
—i 1 1 1 j 1 1 1 1 1—
17.6 17.8 18 182 184 186
1 !•
18.8 19.0
Time - h
Figure 8.12. Time series of the product-level changes in the sleeve of an underfilled fiberglass
tank on 5 June 1987; product-level changes were produced by inserting four different-size bars
(625, 1551, 2477, 5071 ml). The measurements were made with a ruler to the nearest 3 mm. A
vapor pocket of 821 L was trapped in the top of the tank for the measurements. The initial
product level was 218.1 cm above the bottom of the tank. The time interval between points is
approximately 3 min.
The observations made for the data shown in Figures 8.12 and 8.13 can also be made for
the data from the 6 June 1987 overfilled-tank tests (Figures 8.10 and 8.11), except that the
exponential deformation of the tank and/or vapor is more pronounced.
8-18
-------
The data relevant to the smallest bar were removed from the data set, because the mean
value of these data was different from that of the other three. The difference was probably due to
the 35% relative error in measuring product-level changes (i.e., 0.3-cm error in measuring an
0.8-cm product-level change).
The three estimates of the 821-L vapor pocket are given in Table 8.5. The agreement is
within the experimental error of the measurement. The two largest sources of error are the
uncertainty in the volume of the trapped vapor and the contribution of the instantaneous
deformation. Assuming an error of 0.63 cm (0.25 in.) in measuring the absolute level of the
product in the tank, there is a 50-L uncertainty in the volume estimate at 218 cm. An error of
approximately 50 L, due to the instantaneous deformation of the tank, is estimated for
AjSd= 35 cm2.
800
750 -
, 700 -
650 -
600
500
1500
2500
3500
4500
5500
Bar Volume - ml
Figure 8.13. Aeff computed from the volume of the cylindrical bar, Vbar, and the measured
product-level change data, Ah, obtained on 5 June 1987. An 821-L vapor pocket was trapped in
the fiberglass tank. The initial product level was in the sleeve, 218.1 cm above the bottom of the
tank.
8.2.2.4 Instantaneous Structural Deformation
An estimate of the instantaneous structural deformation was made, using the four days of
experimental data presented in Table 8.6 (where h0 indicates initial product level) and assuming
that the volume of the vapor pocket was known. Estimates of AVjsd were made for each vapor
pocket and bar. Figure 8.14, a plot of the data, shows that instantaneous structural deformation is
directly proportional to change in pressure (product height). Ajsd is estimated, from the slope of
the least-squares line, as 35 cm2. The value of A,,,,, estimated from all experiments is 36 cm2.
This is very consistent with the value of A^sd calculated in an overfilled tank in which the vapor
pocket estimate was 10 L.
8-19
-------
Table 8.6. Summary of the Experimental Data Used to Estimate the Magnitude of the
Instantaneous Deformation, Ajsd
Date
5/28
5/28
5/29
5/29
5/30
5/30
6/5
6/5
6/5
h»
(cm)
228.8
228.8
229.2
229.2
229.6
229.6
218.1
218.1
218.1
vbar
(ml)
4021
5071
953
1544
952
2477
5071
2477
1551
Ah
(cm)
15.90
20.14
3.93
6.14
4.25
10.85
7.52
3.65
2.30
^eff
(cm2)
252.9
251.8
242.8
251.2
224.2
228.3
674.6
674.4
674.3
Predicted
VVP
(U
254
253
238
250
212
218
842
845
838
Measured
Vvp
(L)
207
207
194
184
184
184
821
821
821
AV¥p
(ml)
2337
2951
546
853
561
1426
4416
2150
1356
AVisd
(ml)
533
662
123
246
84
266
111
63
28
700
600 -
500 -
400 -
I
300
]
200
100 -
A, . =35.1 cm
isd
—i 1 1 1 1—
8 10 12
14
16
18 20
A h - cm
Figure 8.14. Estimate of A^from data collected on 28-30 May and 6 June 1987. The slope of
the line is Ajsd= 35.1 cm2.
8.2.3 Evaporation and Condensation
Because of the large pressure changes that occur as the vapor approaches partial-pressure
equilibrium, large product-level changes can occur in the fill tube of an overfilled tank if the
vapor is unsaturated. This effect was observed in field experiments conducted prior to the UST
8-20
-------
Test Apparatus experiments [25], and in the laboratory experiments. An accurate volumetric test
cannot be conducted unless the trapped vapor has approached an equilibrium condition.
Available data suggest that this occurs in tens of minutes.
8.2.4 Experimental Estimates of AefT
The presence of vapor affects the magnitude of Aeff. Table 8.7 gives the error that occurs
in estimating product volume from product-level measurements when using the theoretical A
rather than the measured A^ height-to-volume calibration. The presence of even small amounts
of trapped vapor can cause large errors in the volume estimates.
Table 8.7. Error in Estimating Product Volume from Product-Level Measurements If the
Measured Height-to-Volume Calibration Factor, A,.ff, Is Not Used (A 1000-ml bar was used to
raise the product level in the fill tube.)
Trapped
Vapor
Volume
(L)
10
25
50
75
100
150
200
A
(cm2)
81.1
81.1
81.1
81.1
81.1
81.1
81.1
Aisd
(cm5)
0
0
0
0
0
0
0
AefT
(cm2)
88.1
98.9
116.8
134.7
152.6
188.5
224.4
Ayp
(cm5)
7.1
17.9
35.8
53.7
71.6
107.5
143.4
Height-to-Volume
Error
(%)
8.1
18.1
30.7
39.9
46.9
57.0
63.9
8.2.5 Summary
The experiments at the UST Test Apparatus suggest that unless it is very large, the volume
of trapped vapor is difficult to estimate, because the effects of the instantaneous deformation of
the tank can be large in comparison to the volume of trapped vapor. Using Eq. (6.11), it was
estimated that the tanks at the Test Apparatus contained a maximum of 60 L of vapor,
approximately 50 L more than were believed to be present on the basis of geometrical
considerations and experimental analysis. Even though the volume of the trapped vapor cannot
be accurately estimated with Eq. (6.11), such an approach is a good method of identifying a
potential problem, because it yields an upper bound on the volume of trapped vapor. If the
volume predicted by Eq. (6.11) is greater than 100 L, the presence of trapped vapor should be
suspected.
The experiments also showed that a value of n = 1 should be used with Eq. (6.11) to
estimate volume changes effected by pressure changes.
8-21
-------
8.3 Structural Deformation
A series of experiments was conducted to validate the model of the product-level changes
described by Eq. (6.24). The tank was overfilled to a level within the fill tube, and all known
vapor was removed from the tank. The resulting product-level changes in the tank were
monitored over the course of the subsequent 24-h period, until the effects of the exponential
deformation and of the product-temperature changes could be assumed to be small. A bar was
the inserted into the fill tube, instantaneously raising the product level; product-level and
product-temperature changes were measured over the next 6 h. The experiment was designed so
that all sources of noise except thermal volume changes could be made negligible.
8.3.1 Model Validation Results
An experiment was conducted in the steel tank on 3 May 1988 to estimate K, Tc, T,.ff, and
Aeff. These data were selected for analysis because the thermally induced volume changes,
which were less than 0.04 L/h, as well as all other product volume changes, were small enough
to be negligible. The initial product level in the fill tube, before the start of the tests, was 30 cm
above the top of the tank. A 5.045-L bar was used to displace this product. The rise and drop in
the product level was approximately 38 cm (15 in.), only 59% of the expected 62-cm change
based on geometrical considerations. A^ was estimated, from the 38-cm displacement produced
by the 5.045-L bar, to be 132.7 cm2. The product-level changes in a 10-cm (4-in.)-diameter fill
tube were converted to product-volume changes using the measured volume of Acff. The
product-volume data were detrended, and the model described by Eq. (6.24) was fit to the data
using a least-squares technique. The results are given in Table 8.8 and Figure 8.15.
Table 8.8. Estimates of A^f, Tc, Teff, and K made from the Product-level Measurements in the
10-cm-Diameter Fill Tube of the Steel and Fiberglass Tanks at the UST Test Apparatus
Tank
Steel
Fiberglass
A err
(cm2)
132.7
125.1
Tc
(h)
3.0
2.6
Terr
(h)
1.6
1.6
K
(cm2)
117
75
Many different types of experiments were conducted to verify the theoretical relationships
in Section 6.4.4, including releveling experiments. As expected, Teff decreased as the diameter of
the fill tube, A, decreased. Teff was also shown to be a fraction of the actual time constant of the
8-22
-------
0)
|o
-2
-4
Modeled Product Volume
Product Volume
8
10 12
14 16
Time -h
18 20 22 24
Figure 8.15. Comparison of the measured and predicted volume time series for data collected
on 3 May 1988 in the fill tube of the steel tank at the UST Test Apparatus. A 5.045-L cylinder
was used to raise and lower the product level in the fill tube instantaneously. The instantaneous
product-level change was approximately 38 cm, only 59% of the expected 64-cm change.
tank, Te, which controls the deformation for a constant head test. Finally, it was shown that the
product-level changes predicted by Eq. (6.29) were controlled by a height-to-volume conversion
factor, A,,ff + K, that was dependent on the elasticity properties of the tank-backfill-soil system.
The experiments showed that K tended to decrease as product height above the bottom of
the tank increased. The experimental estimates made for the steel tank showed that K was less
than 15 to 30 cm2 for product-level measurements made when product was 1.3 m above grade,
but greater than 100 cm2 for those made when product was in the fill tube but below grade. The
data are not sufficient to generalize that the effects of deformation are small for tests that are
conducted above grade. Neither are the data sufficient to generalize K and Tc for the wide range
of tank, backfill, and soil conditions found nationally.
The experiments also showed that K changed as a function of the time of year.
Experiments conducted 6 to 12 months after the initial ones resulted in different values of K, a
finding which is consistent with the notion that K is strongly dependent on the elasticity
properties of the backfill and in situ soil. The properties of the backfill and the native soil,
particularly the consolidation of the soil, would be expected to change seasonally, as the
temperature and water content changed.
8-23
-------
8.3.2 Model Predictions of Product-Level Changes in the Fill Tube
The Fill-Tube Dynamics Model described by Eq. (6.31) and Eq. (6.32) was exercised to
illustrate how the product level changed in the fill tube as a function of Tc, K, and Aeff. The
constants used in the analysis cover the range of conditions measured in the steel and fiberglass
tank experiments at the UST Test Apparatus.
8.3.2.1 Deformation Produced by Topping the Tank
Many overfilled-tank tests require that the tank be topped off before the test is started.
This may involve adding tens of gallons of product to the tank to elevate the product into the fill
tube and an additional gallon or two to attain a specified level within the fill tube. In addition,
many tests require adjustments of the product level in the fill tube immediately prior to or during
a test. While the volume changes are typically small, the resulting change in the hydrostatic
pressure head may be large.
Eq. (6.31) was solved for a range of values of K and A,.ffto illustrate the product-volume
changes that are measured after topping, when the rate of change of volume in the tank is -1.2
L/h. The deformation of the tank associated with a 1-m increase in the product level within the
fill tube was modeled. The initial product level was assumed to be above the top of the tank and
was also assumed to be constant for the entire time prior to the simulation of a product addition
to the tank (i.e., Vp(t) = 0 for t < t0). The rate of change of volume estimated from the
product-level changes after the large exponential changes have stabilized is presented in
Tables 8.9 and 8.10.
8-24
-------
Table 8.9. Effects of the Fill Tube Diameter on the Model Predictions*
Effective
Diameter
(cm)
2.5
5.1
7.6
10.2
15.2
20.3
25.4
50.8
76.2
Aefr
(cm2)
5.1
20.3
45.6
81.1
182.4
324.3
506.7
2026.8
4560.4
Te(T
(h')
0.04
0.14
0.28
0.40
0.60
0.73
0.81
0.94
0.97
Predicted
Volume
Rate
(L/h)
-0.05
-0.17
-0.33
-0.48
-0.72
-0.88
-0.97
-1.13
-1.17
Actual
Volume
Rate
(L/h)
-1.2
-1.2
-1.2
-1.2
-1.2
-1.2
-1.2
-1.2
-1.2
Predicted:
Actual
0.04
0.14
0.28
0.40
0.60
0.73
0.81
0.94
0.97
Temperature-
Compensated
Volume Rate
(L/h)
1.15
1.03
0.87
0.72
0.48
0.32
0.23
0.07
0.03
Estimates of how A^ = A (i.e., the fill tube diameter) affects the product-volume changes
predicted by Eq. (6.31), which are produced by a 1-m product-level rise resulting from
topping the tank immediately before starting a test, and by a -1.2-L/h volume change
resulting either from a leak or from thermal contraction of the product. The volume change
and the temperature-compensated volume change were estimated after the exponential decay
had become small. The predictions were made for K = 120 cm2 and Tc = 1 h.
Table 8.10. Effects of Tank Elasticity on the Model Predictions*
K
(cm2)
0
15
30
60
90
120
TrfT
(L/h)
1.00
0.84
0.73
0.57
0.47
0.40
Predicted
Volume Rate
(L/h)
-1.20
-1.01
-0.88
-0.69
-0.57
-0.48
Actual
Volume Rate
(L/h)
-1.2
-1.2
-1.2
-1.2
-1.2
-1.2
Predicted:
Actual
1.00
0.84
0.73
0.57
0.47
0.40
Temperature-
Compensated
Volume Rate
(L/h)
0.00
0.19
0.32
0.51
0.63
0.72
Estimates of how K affects the product-volume changes in a 10-cm (4-in.)-diameter fill tube.
The predictions were made for Aeff =81 cm2 and Tc = 1 h.
Tables 8.9 and 8.10 show that, as Aeff increases and K decreases, more accurate estimates
of the actual rate of change of volume can be made. Because only the linear product-level
changes (observed after several time constants, Teff, have elapsed) are being analyzed (e.g., see
8-25
-------
Figure 6.4), the results are independent of the time constant, Tc. A longer time constant would
require a longer test. The results can be interpreted as either a measurement of a leak in the tank
when all other volume changes are negligible or as an estimate of the temperature-compensated
volume rate in a nonleaking tank. In the first case, the actual leak rate should be -1.2 L/h, and in
the second case, the temperature-compensated volume rate should be 0 L/h.
8.3.2.2 Product-Level Changes that Occur During Releveling
Mathematically, the negative of the cumulative product-volume changes predicted by
Eq. (6.32) for a volume rate, C, is nearly identical to the volume changes predicted by Eq. (6.31).
This is true when (1) the cross-sectional area of the fill tube used to estimate product-level
changes with Eq. (6.31) is large enough to approximate a partially filled tank (e.g., Aeff = A >
4560 cm2, which is equivalent to a fill-tube diameter greater than 76 cm), and (2) the releveling
time interval is close to zero (e.g., t' < 5 min). An estimate of the time constant of the
cumulative volume tune series generated by adding the volume changes predicted by Eq. (6.32),
as a function of the periodic releveling interval, t', is presented in Table 8.11. The calculations
were made for K = 120 cm2, Aeff = 125 cm2, Tc = 1 h, and C = 0 L/h. The time constant was
estimated at the point at which 63% of the total product-volume change had occurred. Two
points are noteworthy. First, the rate of change of volume approaches the leak rate after the
initial exponential deformation has occurred. Thus, accurate tests can be conducted if the
waiting period is sufficient for the deformation effects to subside. Second, the time constant is
equal to the time constant of the tank, Tc, when the product is releveled continuously, and it
increases if releveling is done at discrete intervals. The increase in the time constant, Tc, as a
function of the releveling interval is given in Table 8.11.
Table 8.11. Time Constant of the Volume Time Series*
Releveling Interval
(min)
0
5
15
30
60
Time Constant
(h)
1.00
1.04
1.14
1.33
1.79
The time constant was predicted using Eq. (6.32) as a function of periodic releveling interval,
t'. The estimate was made using K = 120 cm2, Aeff= 125 cm2,Tc = 1 h, and C = 0 L/h.
8-26
-------
8.3.2.3 Summary
Unless K is known, it is impossible to accurately estimate the rate of change of volume
from product-level measurements made in the fill tube or standpipe during an overfilled-tank
test. The theoretical and experimental analyses indicate that all volume changes are coupled in a
complicated feedback mechanism.
The Fill-Tube Dynamics Model suggests that the rate of change of volume estimated from
the measured product-level changes in an overfilled tank is in error, even if the measurement is
made after the large changes initially produced by an addition of product to the tank have
stabilized, and even if a measured height-to-volume conversion factor is used. The model
calculations illustrate how Tc, K, and Aeff affect the estimate of product-volume changes in the
fill tube. If K is large or if A,.ff, controlled by the cross-sectional area of the fill tube and other
openings in the tank, is small, large errors can be made in estimating the product-volume
changes produced by a leak or by a temperature-compensated volume rate. The error can be
minimized if K is small, Aeff is large, or the product is continuously releveled within the fill tube.
Increasing the diameter of the container used to measure product-level changes and periodically
releveling the product in the fill tube are two methods that can be implemented as a means
toward conducting accurate tests. In both cases, a waiting period must be included in the test
protocol to allow for the effects of tank deformation to become negligible. A variable level test
can be accurately conducted if K is known. The value of K as well as Tc can also be measured
empirically during a test, but the measurements are difficult and time-consuming to make.
The data collected at the UST Test Apparatus also suggest that K is small I m above grade,
but the data from one set of tanks are insufficient to conclude that conducting a test while
product level is well above grade will eliminate the problem.
8.4 Evaporation and Condensation
The effects of evaporation and condensation in a partially filled tank were not specifically
investigated in this project. The data necessary to quantify the volume changes effected by
evaporation and condensation were collected as part of another program, and the analysis is in
progress. The residual volume changes were compensated for after product temperature changes
had been compensated for. This was accomplished using all three Test Appartaus themiistor
arrays, and was done 6 h or more after any disturbance to the tank. These residual volume
changes were used to estimate the effects of evaporation and condensation. Preliminary analysis
of a number of tests from the evaporation and condensation database suggests for partially filled
tanks that the volume changes are generally small and may be within the experimental
8-27
-------
uncertainty of the temperature compensation estimates (0.04 L/h). No reliable quantitative
estimate of the volume changes associated with evaporation and condensation over a wide range
of tank temperature conditions has been made from the database.
Several observations were made during the UST Test Apparatus experiments. First,
accurate temperature compensation of the product was not possible for approximately 6 h after
any of the tank's fill tubes were opened.
Second, in experiments conducted in Test Apparatus tanks that were filled to within 15 cm
of the top, the effects of evaporation and condensation were small (0.04 L/h) and were highly
correlated with temperature.
Third, the volume changes associated with evaporation and condensation in a
partiallyfilled tank were too complicated to predict using a simple physical model.
8.5 Surface Waves
Mechanical vibrations or other disturbances produce waves travelling along the product
surface in a partially filled tank. These waves, which introduce a modulation of surface height,
are resonant with the length of the tank. Their periodicity is also a function of the tank's
dimension. Typically, waves caused by an external vibration have periods on the order of 2 to
10 s and amplitudes of 0.00127 to 0.0127 cm. A more detailed description of surface waves can
be found in [19, 20, 25, 30].
8.6 Internal Waves
Internal waves usually occur whenever there is a density gradient present, such as the
boundary layer between existing product and newly added product which is at a different
temperature. These waves occur in both partially filled and overfilled tanks. They travel both
longitudinally and latitudinally throughout the tank, below the product surface, with longitudinal
waves being more significant. The internal waves are resonant in the tank, their periodicity
being a function of the tank dimensions and the product density. The passage of the internal
waves modulates the vertical position of the temperature gradient so that a sensor at a fixed
position records a temperature change associated with wave phase rather than with any volume
change.
Internal waves in a 30,000-L tank can have periods of several minutes to an hour or more;
typically, these waves have periods of 5 to 20 min. An internal wave having an amplitude of
8-28
-------
20 cm will produce a temperature change of 0.2°C if the temperature gradient is 1 °C/m. These
conditions can easily arise after a tank has been filled with product whose temperature differs
from that of the in situ product [19, 25, 30].
8.7 Summary of Results from Ambient Noise Experiments
The major conclusions drawn from the ambient noise experiments are as follows.
o Need for temperature compensation and adequate waiting periods. Experiments to
simulate the delivery of product at a temperature different from the tank's surrounding
backfill and from the in situ stored product indicate that the thermal effects decay
exponentially but are large enough, even 24 h after the delivery, to significantly
impair a method's ability to detect small leaks unless the thermally induced volume
changes are compensated for.
When the vertical and horizontal distribution of the temperature was investigated, the
results indicated that the volume-weighted temperature changes measured by a single
array of thermistors at the fill hole of the tank would adequately compensate for the
thermal changes in the product throughout the tank. The results also indicated that
compensation when the product level was at the fill hole was not possible for at least
4 to 6 h after the delivery, but that after the waiting period, the rate of change of
temperature could be measured with sufficient accuracy (with product level at the fill
hole of the tank) to compensate to within 0.04 L/h. During the period immediately
after topping, the effects of horizontal temperature gradients are too large to permit
accurate thermal compensation for 2 to 3 h. The effectiveness of compensation after
the horizontal gradients have stabilized depends on how well the temperature
distribution is measured; compensation improves with increased spatial coverage,
increased test time, and smaller sampling interval.
o Expansion and contraction of trapped vapor. It was found that the expansion and
contraction of trapped vapor could be predicted using the perfect gas law with n = 1.
Atmospheric pressure and product-level changes affect the volume of the trapped
vapor, as do temperature changes. Efforts to develop a simple, practical method of
measuring the amount of trapped vapor during a test were unsuccessful. The model
can do no more than estimate an upper bound of the volume of trapped vapor, because
the effects of trapped vapor can not be separated from the effects of instantaneous
deformation. The tests on the UST Test Apparatus suggested that the effects of
instantaneous deformation were equivalent to those produced by a vapor pocket of
50 L. Accordingly, a vapor pocket estimate of 100 L or more, derived by means of
8-29
-------
the perfect gas law, would indicate that trapped vapor is present in sufficient quantity
to impact the performance of a test method. Even a small vapor pocket can be
important if the product level in the fill tube is changing during the test. Holding the
product level constant in the fill tube eliminates the impact of pressure changes
induced by product-level changes.
o Structural deformation. Structural deformation of the tank leads to significant errors
in precision testing. The product volume changes measured in the fill tube or
standpipe during a variable-level overfilled-tank test cannot be accurately estimated
unless the structural deformation properties of the tank-backfill-soil system are
known. The product-level changes are produced (1) by the instantaneous deformation
of the tank, and (2) by an exponential, time-dependent relaxation of the tank. Even
after the exponential deformation has become small, the product-level changes
resulting from all leak- and non-leak-related volume changes will continue to include
the effects of the deformation. If a test is conducted during the exponential relaxation
period, the measured volume rate may be large; it may introduce a negative bias (flow
out of the tank), and will tend to suggest that the tank is leaking. If a test does not
measure the instantaneous deformation of the tank, or if the test is conducted after the
tank deformation has apparently ended (i.e., after the obvious exponential change),
the measured volume rate will be only a fraction of the true volume rate. Fortunately,
the combined effects of the instantaneous deformation and the trapped vapor can be
measured during a test, and it is possible to wait, before starting a test, until the
exponential product-volume changes become negligibly small. However, even after
the product level has apparently come to equilibrium, the product-volume changes
estimated from the measured product-level changes, using an accurately measured
height-to-volume conversion factor, are always a fraction of the true volume changes;
they depend in a complex way on the time and elasticity constants of the
tank-backfill-soil system, the cross-sectional area of the product surface in the fill
tube or standpipe, the instantaneous defonnation of the tank-backfill-soil system, the
volume of trapped air, and the previous temporal history of product additions to the
tank. Although the product volume can be predicted using a convolution model of the
height changes developed and validated during this program (i.e., the Fill-Tube
Dynamics Model), it is necessary (albeit difficult) to measure the empirical
deformation properties of each tank being tested. These properties vary with product
8-30
-------
level and can vary significantly from tank to tank. Some test results are unaffected by
this effect of structural deformation, while others may be in error by a factor of 2 to
20.
o Constant hydrostatic pressure. In a test conducted under constant hydrostatic
pressure, the product-level changes produced by structural deformation will
eventually become small enough to be negligible. Releveling is one way to achieve a
constant hydrostatic pressure in an overfilled-tank test. Another is to increase the
effective surface area of the fill tube so that this area is much larger than the elasticity
constant. The time constant of the tank will increase if releveling is not done
continuously; a 15-min releveling period will increase the time constant by
approximately 14%. Product must be added carefully during releveling so that the
local temperature field around the sensors will not be disturbed.
o Relationship between compressibility and hydrostatic pressure. The experiments on
the fiberglass and steel tanks at the Test Apparatus indicated that the compressibility
of the tank and ground system decreased with increased head above the top of the
tank, becoming negligible when product was approximately 1.3 m above tank top.
The time and elasticity constants measured in the fill tubes of the steel and fiberglass
tanks were (Tc = 3 h and K = 117 cm2) and (Tc = 2.6 h and K = 75 cm2), respectively.
The fill tubes were 10 cm (4 in.) in diameter.
o Height-to-volume conversion factor. Accurate estimates of performance against small
leaks require that the height-to-volume conversion factor be experimentally measured.
This is essential for all methods that overfill the tank or conduct a test with product
level within the top 10% of the tank. In overfilled-tank test, an error factor of 2 can
occur if the calculated height-to-volume conversion factor, obtained from an accurate
estimate of the area of the product surface and tank geometry, is used instead of the
measured one. This error results from a combination of two factors: a volume change
in any trapped vapor, and instantaneous deformation of the tank when the product
level is changed. Even small amounts of trapped vapor have a significant impact on
the magnitude of the height-to-volume conversion factor. In addition, because the
geometry is not accurately predicted by the tank chart, the calculated
height-to-volume conversion factor can be in error by a factor of 2 during a
partially-filled-tank test in which the tank is filled to within 5 to 10 cm of the top.
Furthermore, the precision of a stick measurement made when the product level in the
8-31
-------
upper 80 to 90% of the tank is generally not sufficient to estimate the conversion
factor with the degree of accuracy needed to detect small leaks, even using a perfect
tank chart.
o Evaporation and Condensation in a Partially Filled Tank. The impact of evaporation
and condensation was not addressed in this program. The Test Apparatus
experiments suggest that the effect of evaporation and condensation will degrade a
method's ability to detect small leaks for up to 6 h after any of the apertures into the
tank are opened. Preliminary analysis of some data obtained after all tank apertures
have been closed for 6 h or more and after the vapor has stabilized suggests that the
volume changes are small (e.g., approximately 0.04 L/h), but their impact on the
reliable detection of leak rates between 0.19 L/h (0.05 gal/L) and 0.76 L/h (0.20 gal/h)
has not yet been quantified.
8-32
-------
9 Evaluation Protocol
Participation
Requirements
I
Test-Method Description
I
Mathematical
Modeling Report
for
Test Method
Test-Method
Model
Development
Field Verification Tests
Performance
Estimates
Test-Method
Model Validation
Evaluation
Report
Evaluation Results
Figure 9.1 Flowchart of the evaluation protocol.
The protocol used to evaluate the performance of the volumetric test methods is
summarized in Figure 9.1. As part of the participation requirements (which were summarized in
Section 1.7), the manufacturer was asked to furnish a description of the test method. Based on
this description, a computer model of the method was developed. In most instances, substantial
interaction with the manufacturer was required in order to specify the method completely. The
major features of the model were then described in a model report that was given to the
manufacturer for review and written concurrence. Once written agreement had been obtained,
the manufacturer was invited to participate in the Field Verification Tests at the UST Test
Apparatus in Edison, New Jersey. The manufacturer was asked to use his method to test the Test
Apparatus tanks under a variety of temperature and leak rate conditions during a three-day
period. In addition, the precision of the manufacturer's product-level and temperature
measurement systems was experimentally estimated. The data collected during the field tests
were used to validate the model of the test method.
9-1
-------
For each method, a performance estimate was made using the computer model of the
method as input to the simulation. (A description of how the performance estimate was made
with the simulation was given in Section 6.) It is important to understand that the performance
estimate was not based on the data collected during the Field Verification Tests. Rather, as
mentioned above, the Field Verification Test data were used to validate the test-method model
that was used as input to the simulation. The performance of each method was presented in a set
of three performance curves.
A 25- to 60-page report describing the results of the evaluation was prepared for each test
method. The first draft was sent to the manufacturer for review and written comment. Each of
the manufacturer's comments was then carefully reviewed for technical relevance by the EPA,
and valid technical comments were incorporated in the report. Each manufacturer was then
invited to participate in a half-day meeting to discuss his evaluation report and the potential for
improvements to his system. The final version of the report was then sent back to the
manufacturer for final review before publication. All of these reports appear as individual
appendices in Volume n of this document.
Each appendix is divided into five sections. Section 1, a brief introduction, gives the name
and manufacturer of the method evaluated. Section 2 describes the most important features of
the manufacturer's instrumentation, the protocol used to conduct a test, the analysis algorithms,
and the detection criterion. Sufficient descriptive material is presented to identify the major
features of the method and to interpret the results of the evaluation; however, no confidential
business information is included. Section 3 describes the results of the Field Verification Tests
conducted by the manufacturer at the UST Test Apparatus in Edison, New Jersey, giving the
results of actual tank tests conducted by the manufacturer's test crew and the conditions under
which the tests were conducted. In addition, the results of a calibration of the manufacturer's
product-level and temperature measurement systems are presented. The Field Verification Test
data were used to validate the test-method model and to determine whether the manufacturer had
actually followed the test protocol. Section 4 describes the validation of the test-method model.
Section 5 presents the performance of the method and makes recommendations that might
improve this performance. The performance was estimated from the results of a large number of
tank tests simulated by using the test-method model and models of the ambient volume
fluctuations that occur in a tank. A large temperature database, collected independently of the
Field Verification Tests, was used as the primary input to the simulation. A list of the pertinent
references is included with each appendix.
9-2
-------
9.1 Test Method Description
With material furnished by the manufacturer at the commencement of the test program,
including brochures, specification sheets, operational manuals, answers to a detailed
questionnaire, and, usually, much discussion by telephone, a special modeling report was
prepared that mathematically described each test method. The report described in detail the
instrumentation, test protocol, data analysis (including temperature compensation) and detection
criterion, as well as the mathematical algorithms developed to represent the method. In almost
all cases, additional contacts with the manufacturer were required, either because important
aspects of the method had not been specified or items in the questionnaire had not been answered
completely. The manufacturer was required to review the report and concur in writing that it
accurately represented his test method. The manufacturer was not permitted to participate in the
Field Verification Tests in Edison, New Jersey, until this review and concurrence process was
complete. Although it was time-consurning, this process (in general) went smoothly.
The model report served two important purposes. First, it clearly specified the important
features of the method that was being evaluated. This was an important point, because many
manufacturers had not specified their test procedures well, and this could have resulted in
differing interpretations of test procedure. Since the performance is greatly affected by changes
in test procedure, it is almost impossible to make a quantitative statement about performance
unless the test procedure is well specified. It was concluded that many manufacturers had not
had a structured procedure before this evaluation, and that, in many cases, they had not
formalized their test procedures until this evaluation. For many manufacturers, specifying the
procedure was difficult, because their operators were allowed to make a variety of judgments as
part of, and during, the test. For example, a test might not be considered valid if the rate of
change of temperature was "too great." While this was a valid approach, the criterion for what
was "too great" was not well specified. It is obvious that performance would be better if the
criterion were 0.01°C/h rather than 0.1°C/h, because the error in temperature compensation
would then be smaller, given that the same degree of compensation was being achieved in all
tests. It is equally clear that using a different criterion for each test in the field would also result
in different degrees of performance. Thus, it was required that each manufacturer quantify these
judgments as part of the test method. For the purposes of this evaluation, no changes to the
method were allowed once the method had been defined in the model report and concurred with
in writing by the manufacturer. It was important to adhere to this procedure, because some
manufacturers were still attempting to make substantial changes to their methods, including
instrumentation and important aspects of the protocol, as they were reviewing the final version of
9-3
-------
the report. This occurred frequently among the manufacturers of the newer methods, who
viewed this evaluation as an opportunity for determining whether their methods were actually
viable.
Second, the model report served as the basis of the computer code developed to model the
method. (A general description of the model was presented in Section 6.5.)
Because some of the mathematical modeling reports include confidential information, they
are considered proprietary, and are not available for public distribution by the EPA. The
manufacturers, however, may distribute copies at their discretion. The modeling reports are
referenced at the back of each appendix. A description of each method, which does not contain
confidential information, is presented in Section 2 of each appendix in Volume n. This
description is more than sufficient to identify the important features of the method and to
interpret the results of a test performed with the method.
9.2 Field Verification Tests
Participation in the Field Verification Tests was a requirement for evaluation. The purpose
of the tests was to
o verify that the method existed and was operational,
o verify that the major aspects of the test protocol were being followed,
o estimate the precision and accuracy of the product-level and temperature
measurement systems through calibration of the instrumentation, and
o validate the model of the test method.
All field verification tests were conducted at the UST Test Apparatus over a three-day
period during which it was planned that up to nine tank tests would be conducted by the
manufacturer's test crew. For each method, three different temperature conditions were
generated by the Test Apparatus engineers to simulate the delivery of product to the tank. In
addition, leak rates up to 0.76 L/h (0.20 gal/h) were generated for some of the tests; however, no
leak rates were induced for the first test after a product delivery. Such deliveries were simulated
by adding product that differed in temperature from the product already in the tank. Except for
temperature and leak rate, no attempt was made to change the test conditions. All valves
connecting the piping with the tank were closed for the tests, so that only the tank was tested.
No vapor pockets were created by the Test Apparatus engineers for any of the tests. There was
no attempt or intention to trick the manufacturer or create conditions that were unusually severe
or different from those normally encountered. The test conditions for individual manufacturers
are given in Section 3 of each manufacturer's appendix (Volume II).
9-4
-------
A new tank temperature condition was generated each day. Approximately 15,000 L
(4,000 gal) of product was added to a half-filled tank containing, approximately, another
15,000 L of product. In the majority of cases, the tank temperature condition was prepared
during the night before testing. Warm gasoline was added to cold; cold gasoline was added to
warm; and gasoline at about the same temperature as that already in the tank was added. The
difference in the temperature between the added gasoline and the ground was between +10°C.
The time series of the tank temperature conditions generated for individual manufacturers are
presented in Section 3 of each manufacturer's appendix.
The performance estimates made for each method were not affected by the conditions
generated during the Field Verification Tests, because a separate temperature database developed
independently of the Field Verification Tests was used to evaluate all of the methods. However,
the range of temperature conditions used in the Field Verification Tests was statistically similar
to that of the database.
Two independent tests of the method, under the same temperature conditions, were
planned for each day of testing. (Each independent test requires that either (1) the test crew
remove their equipment from the tank and then set it up again before performing another test, or
(2) that they start the test after creating a new product condition.) After each independent test
had been completed and the manufacturer had reported a result, a second test was conducted by
simply collecting another set of data over a similar period of time. This second test was not
independent of the first. Time permitting, the manufacturer was then asked to conduct a second
independent test. The same procedure was planned for the second day of testing, using a new
temperature condition; one additional test sequence was planned on the morning of the third day
of testing.
A detailed schedule was developed to conduct these tests. However, it became clear after
the first few manufacturers had started testing that it would be difficult to conduct more than one
independent test per day. Typically, four independent tests were conducted in all, one on the
first day, two on the second, and one on the third. Setting up the test equipment on the first day
generally took longer than anticipated, preventing the conduct of a second test on that day.
9.3 Test-Method Model Validation
The performance estimate made in this evaluation depends on the accuracy of the
mathematical model of each manufacturer's test method. The model was developed by the EPA
from information provided by the manufacturer, and was reviewed for accuracy and
completeness and agreed to in writing by the manufacturer of the method. A systematic
procedure was then used to validate the accuracy of the model.
9-5
-------
With few exceptions, the model is a simple representation of the well-defined steps
established by the manufacturer and followed by him to complete a test. The various parts of the
model, including the time sequence for conducting a test, the data collection and reduction, the
compensation measurements and analysis, the data analysis, and the criteria for a valid test and
those for declaring a leak, were checked by hand and spreadsheet calculations for accuracy. The
main uncertainties in the model are the actual (versus claimed) precision and accuracy of the
instruments, the method of modeling the temperature measurement system when point sensors
are not used, and operational effects. As soon as the field tests were complete, data from those
tests were used to validate the model.
A four-step procedure was used to validate the test-method model used in the performance
simulation. First, the tests were observed on a not-to-interfere basis by one or more of the Test
Apparatus engineers to determine whether the test crew had used the equipment and followed the
test protocol described in the model report. Most protocol deviations were identified during the
post-test analysis after the tests had been completed. Second, the precision and accuracy of the
product-level and temperature measurement systems were estimated experimentally during the
third day of testing. Third, the software was validated using known inputs and outputs to ensure
that it matched the test method described in the model report. This check was independent of the
field tests. Fourth, validation was completed for the model of any method which did not use an
array of point temperature sensors for the manufacturer's temperature compensation.
Not all of the 25 methods were evaluated. Two criteria were used for deciding whether to
make a performance estimate: (1) the ability to obtain at least one valid test with a particular
method, and (2) consistency between test results and expectations for the method. With regard
to the former, it was assumed that a commercially available method should be able to perform
one test in a 72-h period. Thus, any method that failed to perform at least once during the three
days of testing was not evaluated. However, no manufacturer was eliminated from the
evaluation simply because he had not brought a spare set of equipment (as specifically requested
in the invitation letter) to use as a backup in case of instrument failure. The second criterion
applied only to cases in which it was obvious that the manufacturer did not understand what his
method was measuring, that is, when there was an error in the results that was many tunes larger
than would have been expected based on individual errors in the measurement and compensation
systems. Six methods were not evaluated: two methods failed to meet the first criterion, three
methods failed to meet the second, and for one method the Test Apparatus was not properly
configured for the tests. (In this last instance, a valve connecting the above- and below-ground
tanks was not properly closed during many of the manufacturer's tank tests).
9-6
-------
Once the test-method model had been validated, the manufacturer's
temperature-compensated volume-rate test result was compared to that produced by the
simulation in order to assess the operational effects, particularly the disturbance of the product's
temperature field resulting from topping the tank before a test. Because the effects of
temperature changes produced by topping the tank were large enough to produce highly
erroneous results for a method that might otherwise achieve a high level of performance, these
changes were not included in the simulation. It was assumed that as a result of this evaluation
the manufacturer would either eliminate from his protocol the practice of topping the tank, or
wait a minimum of 3 h or more for the resulting thermal effects to become small enough to be
negligible. As a consequence, the performance estimate presented herein was considered invalid
unless the manufacturer had incorporated these changes.
The test protocol specified by the manufacturer assumes that the test will proceed as
prescribed. Unintentional delays in executing the test protocol were a common problem. To
achieve accurate test results requires that the manufacturer wait long enough, after adjusting the
level of the product in the tank, for the temperature inhomogeneities and the deformation effects
to become small. Because of a variety of delays that arose in executing a test, some
manufacturers whose protocol did not include an adequate waiting period accidentally realized
the benefits of waiting. For the tests actually conducted at the Edison facility, delays such as
problems in setting up equipment actually improved the accuracy of the test in some cases, and
would have improved the performance estimate made in this evaluation. The model used to
estimate performance assumes that the manufacturer's protocol is followed exactly as specified,
even though it became clear during the Field Verification Tests that this was not always true.
Consideration was given, for example, to including the effects of delay in the model and
therefore in the performance estimates. This was not done, however, because it would have
violated the protocol given by the manufacturer, and because it was extremely difficult to
determine which delays were typical and which were not.
9.3.1 Adherence to Test Method
During the briefing conducted before any testing began, it was re-emphasized that unless
the method described in the mathematical modeling report was followed, the evaluation would
be invalid. It was stated that no tests would be conducted if the manufacturer had brought a
different system to test, or had made major modifications to the protocol, analysis algorithms,
compensation approach, measurement configuration, or instrumentation.
9-7
-------
The manufacturers' test crews were asked to conduct their tests as they normally did.
These crews, often consisting of senior management and senior technical staff (sometimes
including the inventor of the system) as well as the designated test crew, were professional and
cooperative. There was no intent on the part of the Test Apparatus engineers to police or
interfere with the tests, even though the post-test interpretation of the manufacturers' data would
have been much simpler if the Test Apparatus engineer had been able to make some additional
measurements. The Test Apparatus engineer, whose main role was that of an observer, made
sure that the Test Apparatus sensors were operating properly. He was available to answer
questions for the test crew, but did not actively participate in any of the tests. No information
regarding the nature of the test conditions (i.e., what temperature condition had been created or
whether a leak had been generated for the test) was given to the test crew, so that this evaluation
was equivalent to their testing one of their customers.
Deviations from the prescribed protocol were generally easy to identify during the post-test
analysis. While some test procedures differed from agreed-upon protocols, no obvious
intentional attempts by the manufacturers to make last-minute modifications were observed. In
all instances the deviations that occurred were either unintentional (e.g., operational delays) or
occurred because some part of the protocol did not work and could not have been followed.
Some of the protocol deviations that were discovered during the tests were brought to the
attention of the test crew. They were asked to describe what was being done, and the deviation
was noted in the experimental logs. Operational delays and problems that occurred during the
test were not considered protocol deviations.
The experimental logs, the Test Apparatus's and the manufacturer's raw temperature and
product-level data, and the manufacturer's calculation data sheets and operational notes were all
reviewed during and after the Field Verification Tests in order to identify any variances in the
prescribed test protocol. The most common deviation involved the criteria for determining when
the test was considered completed.
The Test Apparatus was general enough to permit all manufacturers to conduct their
normal tests. No permanent modifications to the Test Apparatus were permitted. The Test
Apparatus sensor systems functioned properly for the large majority of the tests. For a few of the
methods tested, some problems with small but visible leaks in the above-grade thermistor
connections occurred at the beginning of the tests. All Test Apparatus problems were fixed
immediately, or the instrumentation was removed for the duration of the test. Test Apparatus
problems and their impact on the manufacturers' tests are described in the appendices if they
represented more than a simple delay in the test preparations.
9-8
-------
Both the manufacturers' test crews and that of the Test Apparatus demonstrated
professionalism in conducting the Field Verification Tests. The vast majority of the
manufacturers' crews indicated that the tests had been fairly conducted, that the Test Apparatus
crew had been helpful and had not interfered, that the nature of the tests was straightforward and
typical of actual tank tests at retail service stations, and that the Test Apparatus was not as
difficult to test as many of those they had previously performed at retail stations.
The validation procedure is discussed in Section 4 of each appendix in Volume H
9.3.2 Instrument Precision and Accuracy
The manufacturer specified the precision and accuracy of the sensor systems used to
collect the raw data. Typically, these included measurements of product level, product
temperature, and product density. These were used as input to calculate product volume and
volume change due to product-temperature changes, or temperature-compensated volume. The
precision and accuracy of the sensor systems were determined from a calibration check
performed during the Field Verification Tests. The measured precision and accuracy were used
in the test-method model.
Several different methods were used to perform the calibration checks. The standard
method used to calibrate the temperature measurement systems was to place the temperature
measurement sensors in a controlled temperature bath. The temperature measured by the
manufacturer's system was compared directly to the quartz thermometer used as a reference.
Approximately 30 min of measurements were made in at least four different baths between 0 and
30°C; the temperature of the bath was allowed to drift slowly during each 30-min period. The
calibration was performed to estimate the capability of the system to measure the rate of change
of temperature. A least-squares line was fit to a scatter plot of the manufacturer's data and the
reference data. The slope of the line indicates the accuracy of the system to measure absolute
temperature changes, and the standard deviation of the manufacturer's temperature data about
the line is a measure of the precision.
If the sensors were too large to be placed in the temperature bath, an alternate method was
used to perform the calibration check. A direct comparison between the nearest Test Apparatus
thermistor array and the manufacturer's temperature measurement system was made when
horizontal temperature gradients in the tank were small enough to be negligible.
Two methods were used to check the precision and accuracy of the manufacturer's
product-level measurement system. In the first method, where measurements were made in the
tank, bars of different size were used to displace the product level. A bar was inserted in the
9-9
-------
product for 2 min and then removed for 2 min a total of 10 or more times. The precision was
estimated from the standard deviation of the absolute value of the product-level changes after
product-level trends had been removed from the data. The accuracy was determined by the mean
height change. In the second method, which applied to those few instances where the first
method could not be used, the manufacturer was asked to sample the data for several hours at a
rate of 1 sample/min. The precision was then estimated from the white portion of the power
spectrum obtained from the data.
9.3.3 Temperature Compensation
The manufacturer's temperature measurement or temperature compensation system was
modeled using a linear combination of the Test Apparatus thermistors. If the manufacturer's
system used point temperature sensors, the modeling validation was straightforward. If,
however, the temperature measurements were made by a spatially integrating sensor, such as a
thermistor strip or a tube filled with product, the temperature measurement sensor was modeled
as a weighted average of the Test Apparatus thermistors. The model of these spatially
integrating systems was based upon information provided by the manufacturer; if none was
available, the manufacturer was asked to describe what he thought was being measured. The
model was verified by the manufacturer as being representative of his system, and then was
tested experimentally during the field tests. The validity of the models of spatially integrating
temperature systems was not guaranteed; if the temperature measurement model could not be
validated, no evaluation was performed. No attempt was made by the EPA to develop a more
sophisticated physical model of the system, even though it was possible to do this. Four systems
had temperature measurement models that could not be validated.
In some instances it was difficult to validate the spatially integrating temperature
measurement models because of the 1-m horizontal spatial separation between the Test
Apparatus thermistor arrays and the manufacturer's measurement system. Horizontal gradients
in the temperature field were produced by topping product in the fill tube, or by addition of
product to the tank to simulate a delivery. However, this difficulty was overcome in all cases.
9.3.4 Validation Procedure
The data collected during the Field Verification Tests were used to validate the model of
the test method and assess the importance of operations on performance. The largest operational
effects occurred whenever product was added to the tank and whenever delays in executing the
9-10
-------
protocol occurred. In some instances, operational effects actually unproved performance. These
operational effects are not included in the performance model. However, they may have a
significant effect on performance and so are discussed separately in each appendix.
With the exception of the temperature compensation approach, the test-method model was
simply a logical sequence of well-defined steps and algorithms to collect and analyze the data.
The details of this procedure had already been presented in the model report. In what was
essentially a computer-code validation, an effort was made to determine whether the algorithms
produced the same answers as those prescribed by the manufacturer. The computer code was
verified by using known input with known output. As part of the validation, the entire model
code was exercised with complete sets of product-level and temperature data.
9.4 Evaluation Results
The evaluation results in each appendix are discussed in two parts. First, an estimate of
performance is given, and second, recommendations for improving performance are made.
These results are also summarized and discussed in Section 11.
The performance of each method was obtained by combining several different
calculations. First, a quantitative estimate of performance was made for a single tank without
any trapped vapor; it included the instrumentation noise, a wide range of temperature conditions,
and one set of tank deformation characteristics. The performance of each method is presented in
terms of the probability of detection and probability of false alarm for leak rates between
190 ml/h (0.05 gal/h) and 5,000 ml/h (1.3 gal/h). The performance estimate was made for a
2.43-m (8-ft)-diameter, 30,000-L (8,000-gal) steel tank containing unleaded gasoline, and
assumes that the test procedure was followed precisely as specified by the manufacturer.
Second, quantitative estimates of the effects of structural deformation and trapped vapor on
performance (presented in Sections 6, 8, and 10) were made over a wide range of conditions.
The significance of these effects (discussed in Section 5 of each appendix) is that they can
seriously degrade the estimates given by the performance curves. The effects of structural
deformation and trapped vapor were not included in the performance curves, because for a wide
range of petroleum storage facilities, neither the range of tank and backfill properties affecting
the structural deformation of storage tanks nor the distribution of the volume of trapped vapor is
known. For most methods, an arbitrary selection of these conditions could easily result in
anomalously poor performance.
9-11
-------
Third, operational effects were quantified. These effects include topping the tank before a
test, which can significantly impair effective temperature compensation and significantly
increase structural deformation of the tank. These operational effects can seriously degrade the
estimates given by the performance curves.
The multi-step approach described above was taken because a majority of the methods are
similarly influenced by deformation, trapped vapor, and operational effects. These three effects,
which were noted during the actual field tests of each test method, result in large errors that
could invalidate a test, yet they can be corrected by simple modifications to the protocol.
The part of each appendix describing performance improvements is brief. All methods can
be improved to some degree. However, only those suggested modifications are given that can
significantly improve the performance of a method. Many additional requirements which might
produce incremental improvements of up to 10% were not mentioned. This does not imply that
such changes are not warranted.
9-12
-------
10 Performance of Canonical Test Methods
Most of the volumetric test methods are generically similar. They attempt to minimize the
volume changes that occur in a nonleaking tank so that if a leak is present, these volume changes
will be much smaller than those produced by the leak. Minimization of the volume changes
(i.e., noise) is usually accomplished by temperature compensation and by including adequate
waiting periods in the protocol.
Performance calculations are made in this section using a number of canonical methods
which are representative of the 25 methods that have been evaluated. Each canonical method
consists of product-level and temperature measurements that are combined to develop a
histogram of the temperature-compensated volume rates. Unless specifically intended to
demonstrate a deficiency, the canonical methods invoke sound operational, data collection, data
analysis, and temperature-compensation practice. The results of the performance calculations
made for these theoretical test methods are used (1) to help interpret the performance being
achieved by the methods evaluated in this study, (2) to illustrate performance that can be
achieved by such a method, (3) to illustrate the effects of the important sources of noise on
performance so that design and performance improvements can be made, and (4) to estimate the
practical technological limitations of volumetric tank testing.
Three calculations are made in this section. First, an estimate of the degree of temperature
compensation that can be achieved with different measurement schemes is made using the
product temperature database as a function of test duration. The degree of temperature
compensation achieved is dependent on the ability of the temperature measurement system to
estimate the average rate of change of temperature in the tank. There are six basic temperature
measurement schemes employed by the methods evaluated. Second, an estimate of the volume
changes that result from thermal expansion or contraction of trapped vapor is made using the
temperature changes derived from the temperature database and measured by the thermistor
located near the top of the tank. Third, an estimate of the systematic error that results from
structural deformation as a function of the time constant (Tc) and elasticity constant (K) of the
tank-backfill-soil system is made using a method with a vertical array of five thermistors to
compensate for thermally induced volume changes.
10.1 Tesl Mt-lhod Model Description
All calculations were performed using the basic test method described below, and the same
temperature database used to evaluate the 25 commercial methods. Modifications were made to
the method to illustrate specific effects on perfonnance. Unless otherwise stated, all calculations
were performed with this method.
10-1
-------
The analysis was performed for overfilled-tank tests conducted in the 10-cm (4-in.)-
diameter fill tube of a 30,000-L (8,000-gal), 2.43-m (8-ft) diameter tank containing unleaded
gasoline. The precision of the product-level sensor was assumed to be 0.25 mm (0.01 in.),
corresponding to a volume change of 0.002 L (0.0005 gal). This precision is an order of
magnitude better than is necessary for the detection of a leak of 0.19 L/h (0.05 gal/h) with a PD of
0.99 and a P,.A of 0.01. It is further assumed that the height-to-volume conversion factor was
measured experimentally and includes the effects of instantaneous structural deformation and
vapor pockets. Finally, it was assumed that the tank does not deform exponentially in response
to a height change unless the deformation effects are specifically included (i.e., for these tests K
= ()cnr).
The six temperature sensor configurations given in Table 10.1 were used to illustrate the
degree of temperature compensation that can be achieved as a function of vertical spatial
coverage. For the structural deformation and trapped vapor configurations, an array of five
thermistors, equally spaced and volumetrically weighted, was used to measure the average rate
of temperature change in the tank. The temperature changes were converted to an equivalent
volume assuming that the volume of the product in the tank and the coefficient of thermal
expansion were known perfectly. The precision of the temperature measurement system was
assumed to be 0.001"C, the actual precision of the thermistors used to collect the data at the Test
Apparatus.
The data for each test method were collected and analyzed at a rate of 1 sample/min. The
test duration was varied from 1 to 4 h, and no test was begun until 12 h after a delivery. No other
product-level changes were induced before a test unless such was specifically stated. The
temperature-compensated volume rate was calculated by subtracting the temperature time series
from the product-level time series after each had been converted to an equivalent volume and a
least-squares line had been fitted to the residual volume.
10.2 Thermal FApansion and Contraction of the Product
AH methods attempt to compensate for thermal expansion, either by measuring the rate of
change of temperature directly and converting to the rate of change of volume, as described in
Section 6.4, or by directly measuring the volume changes produced by temperature changes.
The achievable performance depends on how accurately each approach measures the average
rate of change of temperature of the product throughout the entire tank. Temperature
compensation can be accomplished by means of a thermistor array placed at the fill tube of the
tank, providing that (1) there is a waiting period to allow the horizontal gradients to dissipate
after addition of product to the tank, and (2) any addition of product via the fill tube during the
10-2
-------
test does not adversely affect the temperature measurements. The product-level changes in the
tank are estimated assuming that the only product-level changes present are caused by the
thermally induced volume changes as determined from the three thermistor arrays in the tank.
10.2.1 Test Method Description
The six temperature-compensation schemes that were modeled are summarized in Table
10.1. In order that the performance estimates could be used to evaluate the temperature
compensation scheme, no effects of structural defomiation were included here. Schemes Cone,
Thre, Five, and TVMT volumetrically weight the thennistor measurements for thermal
compensation based on the circular geometry of the tank. Method AVGT uses the arithmetic
mean of measurements from one array of submerged thermistors. The results are applicable
equally to methods that overfill the tank without trapping vapor and to those that operate in a
tank filled nearly to capacity. All temperature measurements were made when product level was
near the fill hole, using Array 2 of the Test Apparatus thermistors.
Table 10.1. Temperature Compensation Schemes
Method
Description
None No temperature compensation
One One thennistor located at the center of the tank
Three Three thermistors
Five Five thermistors
AVGT Arithmetic average of all submerged thennistors
VWAT Volumetrically weighted average of all submerged
thermistors
10.2.2 Performance Results
A histogram and a set of performance curves were generated for each canonical method for
test durations of 1, 2, 3, and 4 h. To illustrate performance, the standard deviation of the noise
and the signal-plus-noise histograms estimated from the histogram of the
temperature-compensated volume rates is presented in Table 10.2. The leak rates that can be
detected with a Pn of 0.99 and a PFA of 0.01 under the conditions in Table 10.2 are summarized
in Table 10.3. As an example, the histogram and cumulative distribution function for the
five-thermistor canonical method are shown in Figure 10.1 for a test duration of 1 h. Figure 10.2
illustrates the change in the histogram for One, Five, and VWAT. The results suggest that
performance improves with the number of thennistors and with volumetric weighting.
10-3
-------
Table 10.2. Standard Deviations of Temperature-Compensated
Volume Rate Histogram
Scheme
None
One
Three
Five
AVdT
VWAT
1.0
789
161
112
84
30
34
Standard Deviation
Test Duration
2.0
766
156
106
80
35
26
(nil/h) for
(ID of
3.0
795
157
105
81
35
21
4.0
800
156
105
75
31
19
Table 10.3. Smallest Detectable Leak Rates with a PD = 0.99 and
PIlA = 0.01
S die me
None
One
Three
Five
AVGT
VWAT
1.0
3685
752
523
392
140
159
Detectable
for Test
2.0
3577
729
495
374
163
121
Leak Rate (ml/h)
Duration (h) of
3.0
3713
733
490
378
163
98
4.0
3736
729
490
350
145
89
10-4
-------
020
0 18
0 16
0 14
0 12
0 10
008
OOfi
004
0 02
000
-400
fit
-300
_t^u:
-200 -100
100
200
300
400
Flow Rale - ml/h
0 80
0 60
0 40
0 20
000 "
300
200
B
100
200
300
Flow Rate - ml/h
Figure 10.1. The histogram (A) and cumulative distribution function (B) for the five-thermistor
method with a test duration of 1 h. The dotted line in (B) is a normal model generated with the
mean and standard deviation of the histogram data.
The results of the compensation schemes having three or fewer thermistors should be
interpreted cautiously. The product condition used in the Edison evaluation tends to have some
symmetry that will not always be encountered in the field. The results are somewhat dependent
on the location of the thermistors relative to the initial and final volume of product in the tank
after a del ivery. If the tank was not half-filled at the time of delivery, differences of as little as
15 cm (6 in.) in the location of the temperature sensor and the product level (before the addition
of product) could result in larger errors than those manifest in this study. To illustrate the
magnitude of the error for the one-thermistor case, the location of the thermistor has been moved
to 1/4 and 3/4 of the tank height. The results, designated by the height of the thermistor as a
fraction of tank diameter and summarized in Tables 10.4 and 10.5, suggest the perfonnance that
might be achieved if the product was not delivered to a half-filled tank. The perfonnance is
degraded severely for the 1/4 and 3/4 methods.
10-5
-------
Standard Deviation: 152ml
0.12
c 0.10
O
I °-08
C
I 0.06
a 0.04
.£•
S 0.02
f
*• o.oo
*TN
-600
-400
-200 0
Flow Rate - ml/h
ra
200
400
600
Standard Deviation 84 ml
0.30
0.25
1 0.20
u_
J 0.15
§ 0.10
o
2-
% 0.05
£ 0.00
-600
B
-400
-200
200
400
600
Flow Rate - ml/h
Standard Deviation: 36 ml
0.45
0.40
1 ° 35
1 0.30
1 °'25
3 0.20
a 0.15
i1 0.10
1 0.05
o
i o.oo
-6(
-
-
—
—
—
1
i • — - ' FTT-,^
\/
A
y
-^
V'
X
X
X
V
n
sv/
/•
V
/^
>
^
;x
\,
7 — i
vj
VPvl 1 1
10 -400 -200 0 200 400 6(
Flow Rate - ml/h
D
Figure 10.2. Histograms of the noise compiled from overfilled, constant-head tank tests that
compensate for thermal expansion and contraction of the product using (A) one temperature
sensor located at the midpoint of the tank, (B) five equally spaced temperature sensors that are
weighted voiumetrically, and (C) eleven equally spaced temperature sensors that are weighted
volumetric ally. It is assumed that the tank does not deform (i.e., K = 0 cm2). The standard
deviations of the histograms shown in (A), (B), and (C) are 161 ml/h, 84 ml/h, and 34 ml/h,
respectively. The histograms are compiled from estimates of the temperature-compensated
volume rates for a 1-h test using thermistors with a precision of 0.001 °C. All product-level
changes are measured in a 10-cm (4-in.)-diameter fill tube with a precision of 1 ml/h. The only
product-level or product-volume changes are produced by the thennal expansion or contraction
of the product.
10-6
-------
Table 10.4. Standard Deviations for One-Thermistor Case at Different Heights
Method
1/4
3/4
Height (cm) from
Bottom of Tank
61
182
Standard
1.0
934
948
Deviation (ml/h)
of
2.0
939
952
for Test
3.0
947
958
Duration (h)
4.0
954
962
Table 10.5. Smallest Detectable Leak Rates for One-Thermistor Case at Different
Heights (Pn = 0.99 and P,A = 0.01)
Method
1/4
3/4
1.0
4362
4427
Detectable
for Test
2.0
4385
4446
Leak Rate (ml/h)
Duration (h) of
3.0
4422
4474
4.0
4455
4493
Another calculation was made to estimate the change in performance that would
occur if one thermistor were located 30 cm above the mid-point of the tank. The standard
deviation was reduced from 161 ml/h to 128 ml/h. Moving the thermistor up another
30 cm (3/4 method) dramatically changes the performance.
An estimate of performance was also made for the five thermistor method after
moving the array up by 20 cm (8 in.), or one thermistor location, on Array 2. The
standard deviation increased from 84 ml/h to 121 ml/h as a result of the change. Shifting
the array up and down by different amounts results in similar changes in the standard
deviation. This change in the standard deviation is consistent with the experimental
uncertainties of the temperature measurement.
The results indicate that temperature compensation is essential and that performance will
improve with an increased number of thermistors. The results also suggest that it is possible to
compensate for temperature sufficiently well to reliably detect leaks of 0.19 L/h (0.05 gal/h).
Without compensation, leaks of 4.75 L/h (1.19 gal/h) and larger are detectable with a PD of 0.99
andaP,.Aof().01.
The results also indicate that the increase in test duration did not, in general, improve the
estimate of the mean rate of change of temperature in the tank. This result was initially puzzling,
because it was contrary to previous experience [e.g., 6, 7]. This finding should be interpreted
cautiously, since it does not necessarly indicate that a 1-h test will suffice. The level of
10-7
-------
electronic noise (including electronic drift) in most of the temperature measurement systems
evaluated was higher than the noise level of the temperature measurement system used at the
Test Apparatus, and the additional averaging gained from the longer duration of the test was
needed for high performance. Also, if temperature fluctuations have periods of 20 to 30 min or
longer, a test longer than 1 h is required to avoid aliasing the data. This is particularly true if the
vertical temperature field is undersampled. For the most part, a 2-h test is long enough to avoid
undersumpling the temperature data.
Results can he highly variable for any thermistor array which does not adequately cover
the vertical extent of the tank. Since it is quite conceivable that temperature could be rising in
the upper part of the tank and falling in the lower part, large errors can arise unless the
temperature changes in both areas are monitored. A one-thermistor array will always have the
potential for large errors because the measured temperature change is not representative of what
is going on in the tank as a whole. In general, a three-thermistor array has the vertical coverage
to avoid most of these problems, but the estimates can be poor under some conditions. A
volumetrically weighted five-thermistor system is probably the minimum acceptable
configuration for avoiding spatially induced errors. The magnitude of the compensation error,
however, will continue to decrease with increased spatial coverage of the temperature
measurement system.
10.3 Trapped Vapor
It is difficult to estimate the impact of trapped vapor on an overfilled-tank test that is
conducted under uncontrolled conditions, because the distribution of the volume of trapped
vapor is unknown for the universe of installed tanks. If the volume of trapped vapor were
known, it would be possible to develop a method of avoiding its effects, or perhaps a
compensation scheme for minimizing them. Because of the instantaneous deformation of the
tank, the effects of which are virtually undiscemible from those of trapped vapor, it is not
possible to measure the volume of the vapor directly. Changes in the volume of the trapped
vapor produced by hydrostatic (i.e., product-level) and atmospheric pressure changes can be
minimized by using an experimentally determined height-to-volume conversion factor and
conducting a test when the atmospheric changes are small.
A canonical method was used to estimate the important effects of trapped vapor. A
five-thermistor measurement system was used to compensate for temperature. The precision of
the product-level and temperature measurement systems does not affect the results.
10-8
-------
Tests were simulated to estimate the effect of trapped vapor on test performance.
Saturated vapor pocket volume changes were predicted using Eq- (6.8). The uppermost
thermistor on the Test Apparatus thermistor array, located within 10 cm (4 in.) of the top of the
tank, was used as input into the vapor model. Tests were simulated assuming a linear
atmospheric pressure change of ± 0.5 mb/h. It was found that the vapor pocket volume changes
induced by the combined effects of thermal and pressure changes are very small, less than 40
ml/h, for vapor pockets smaller than 200 L. The vapor pocket volume change caused by
temperature fluctuation can be roughly estimated as 1 ml per liter of trapped vapor per degree
Celsius change in mean vapor temperature. The change in vapor pocket volume due to pressure
changes can be estimated as a 1.5-mJ change in vapor pocket volume for every millibar change
in vapor pocket pressure. Although the changes in vapor pocket volume should, in most cases,
be relatively small during a test, the trapped vapor, if the initial volume is large, can be an
important source of error unless it is included in the height-to-volume conversion factor.
Furthermore, as discussed in Sections 6.4 and 8.2, even small vapor pockets can effect large
changes if the vapor is unsaturated.
10.4 Slrucliirul Dd'orrnalion
Because, for the wide range of tank-backfill-soil conditions among installed tanks, the
probability distributions of the tune constant, Tr, and the tank system elasticity constant, K, are
unknown, the full range of effects of structural deformation was not included in the 25 test
method evaluations. The effects of deformation included in the evaluations were for a single
tank-hackfill-soil condition.
An estimate of the effects of structural defonnation on performance was made for tests on
a single tank and for tests on many tanks, all overfilled. The estimate was made of the rate of
change of the temperature-compensated volume in a nonleaking tank when the only volume
changes are produced by thermal expansion of the product and by structural deformation of the
tank. Two calculations were made to estimate the effects, respectively, of starting a 1-h test
immediately after topping the tank and of waiting 3 time constants (i.e., 3 Tc) before starting the
test. These represent the extreme conditions, that is, cases for which there is no waiting period to
allow for the stabilization of the large volume changes that occur immediately after any
product-level change, and cases for which an adequate waiting period does exist.
10.4.1 Tesl Mel hod Description
The five-thermistor measurement system was used with varying values for Tc and K, and
with A,.,, = A set to 82 cm2. It was assumed that the initial product level in the fill tube was
10-9
-------
15 cm above the top of the tank and that 8.2 L (2.2 gal) of product was added to the fill tube, an
amount sufficient to raise the level 1 m if the instantaneous deformation and trapped vapor
effects are not included.
All calculations were performed assuming that product-level changes are produced by the
thermal expansion and contraction of the product and by the structural deformation of the tank.
10.4.2 Product Volume Estimates in a Nonleaking Tank
The effects of structural deformation upon test performance were estimated for tests that
start immediately after product volume adjustments are made and for tests in which a waiting
period is observed after the tank has been topped.
10.4.2.1 Test Starting Immediately after Topping
Histograms of the temperature-compensated volume rates were generated for increasing K
(30, 60, 120 cm2) for a time constant of the tank of 0.75 h. Figure 10.3 illustrates the effect of
structural deformation on the result when K = 120 cm2. Two observations are noteworthy. First,
the histogram has a large nonzero mean, or bias, which suggests that most tests in a tight tank
would result in a declaration that the tank is leaking; and second, the standard deviation is
significantly larger than would be obtained if deformation were not occurring. The increase in
the standard deviation with K is illustrated in Table 10.6. The standard deviation when K = 0
cm" is approximately 80 ml/h. Clearly, the test results are not predictable, and are dominated by
the large change in product level that occurs immediately after topping the tank. The histogram
of the temperature-compensated volume rate has a larger spread than that produced by thermal
expansion or contraction of the product itself.
Histograms were generated for increasing Tc (0.5, 1, 3 h) with a tank elasticity constant of
120 cm:. The results, shown in Table 10.7, indicate that the mean changed but the standard
deviation remained approxbnately the same. Figure 10.3 illustrates the effect on the mean and
standard deviation as the time constant increases from 0.75 to 3 h. Clearly, the time constant
controls the results.
10-10
-------
0.12
0.10
0.08
006
004
5 0.00
-4000
-3500
-3000
X
x>
x>
,
X
XT{
<>
/K
-2500
Flow Rate - ml/h
-2000
-1500
-1000
0.08000
0.07000
0.06000
0.05000
0 04000
0.03000
0.02000
0.01000
0 00000
-5000
B
-4000
-3000
-2000 -1000
Flow Rate - ml/h
1000
2000
Figure 10.3. Histograms of the noise generated from overfilled, variable-head tank tests
conducted immediately after topping the tank (1-m product-level addition) for (A) Tc = 0.75 h, K
= 120 cm'and (B) Tc = 3 h, K =120 cm2. The standard deviation of the histograms are 502 ml/h
for (A) and 1160 ml/h for (B). The nonzero mean is produced by the exponential
time-dependent structural deformation. The histograms are compiled from estimates of the
temperature-compensated volume rates for a 1-h test with a method using five equally spaced
thermistors having a precision of 0.001 °C. All product-level changes are measured with a
precision of 1 ml/h in a 10-cm (4-in)-diameter fill tube. Other than deformation, the only
product-level or product-volume changes are produced by the thermal expansion or contraction
of the product.
Table !0.6. Standard Deviations for a Nonleaking Tank Immediately
after Topping (Tc = 0.75 h, Aelf = 82.0 cm2)
K (cm2)
30
60
120
1.0
253
373
505
Standard Deviation (ml/h) for
Test Duration (h) of
2.0 3.0
246 249
371 375
505 510
4.0
249
376
513
10-11
-------
Table 10.7. Standard Deviations for a Nonleaking Tank Immediately after Topping
(K = 120 cm2, Aeff = 82.0 cm2)
Tc(h)
0.5
1.0
3.0
Mean (ml) for
Test Duration (h) of
1.0 2.0 3.0 4.0
-4000 -1200 -550 -550
-4500 -1800 -950 -550
-3250 -2000 -1450 -1000
Standard Deviation (ml) for
Test Duration (h) of
1.0 2.0 3.0 4.0
527 506 508 530
536 510 512 546
585 532 535 557
10.4.2.2 Test Starling 3 Time Constants After Topping
Approximately 99% of the deformation will have occurred after 3 time constants.
Therefore, if the total change in volume due to deformation is less that 4 L/h (1 gal/h), the
residual effects wiU be less than 0.04 L/h (0.01 gal/h), sufficiently small to allow a 0.19-L/h
(0.05-gal/h) leak to be detected.
Table 10.8. Standard Deviations for a Nonleaking Tank 3 Time
Constants after Topping (Aet, = 82.0 cm2)
K (cm2)
30
60
90
120
120
120
120
120
Tr(h)
0.75
0.75
0.75
0.75
0.5
1
2
3
Standard Deviation (ml/h)
260
371
453
480
509
512
522
544
The histograms were generated for increasing K (30 to 120 cm2) and for increasing Tc
(0.50 to 3 h). The results, shown in Table 10.8, were for a 2-h test duration and a five-thermistor
temperature compensation scheme (Five). The results of using a 1-, 3-, and 4-h test duration are
almost identical to the results of the 2-h test. The same conditions used to generate the
histograms in Figure 10.3 were used to generate those in Figure 10.4, except that the waiting
period was 3 time constants or more. The mean of the histogram is approximately zero, but the
standard deviation increases with K and is approximately constant. The product-level and
product-volume changes are smaller than those that would occur if defonnation were not present,
and as a consequence, the histogram of the temperature-compensated volume rates has a smaller
10-12
-------
spread than the original temperature volume histogram, but a larger one than the histogram for
the case where deformation is not a problem. As K approaches 0 cm2, the
temperature-compensated volume rate histogram approaches the histogram generated without
deformation (i.e., a standard deviation of approximately 80 ml/h). The effect of the increasing
time constant is an increase in the duration of the test. The measured volume rate is controlled
by K if the waiting period is sufficiently long for the exponential decay to stabilize.
e
a.
014
0.12
0.10
0.08
0.06
0.04
0.02
0.00
•1500
Standard Deviation. 480 ml
V
y
. < X
-1000 -500 0
Flow Rate - ml/h
500
1000
1500
Standard Deviation. 530 ml
0.14
c 012
g
£ 010
§ °08
| 006
Q 004
-S 002
000
-2000
<
KX
B
-1500
-1000
-500 0
Flow Rate • ml/h
500
1000
1500
Figure 10.4. Histograms of the noise generated from overfilled, variable-head tank tests
conducted 3 time constants after topping the tank (1-m product-level addition) for (A) Tc = 0.75
h, K = 120 cnr and (B) Tc = 3 h, K =120 cm2. The standard deviation is 480 ml/h for histogram
(A) and 530 ml/h for histogram (B). The means of the histograms are approximately zero,
suggesting that the exponential time-dependent structural deformation effects have become
small. The histograms are compiled from estimates of the temperature-compensated volume
rates for a 1-h test with a method using five equally spaced thermistors having a precision of
0.001 °C. All product-level changes are measured with a precision of 1 ml/h in a 10-cm
(4-in)-diaineter fill tube. Other than deformation, only product-level or product-volume changes
are produced by the thermal expansion or contraction of the product.
The performance estimates made in the appendices assumed one set of
defonnation characteristics. An estimate of performance was made for a uniform
distribution of K between 0 and 120 cm2 and T(- between 0 and 3 h for a 1-h test with
10-13
-------
a five-thermistor method. The wait time after topping was at least 3 time constants.
One histogram of the temperature-compensated volume rates was generated for K and
for Tc, respectively. The results, shown in Figure 10.5, show that the standard
deviations were 410 ml/h when Tc= 0.75 h and K is uniformly distributed, and
514 ml/h when K = 120 cm2 and Tc is uniformly distributed. Under these conditions,
only leaks greater than 3.8 L/h (1 gal/h) could be detected with a probability of
detection of 0.99 and a probability of false alarm of 0.01.
Standard Deviation: 410 ml
0.12
0.10
008
0.06
0.04
002
000
-1500
-1000
X
KX
KX
500
1000
1500
Flow Rate - ml/h
Standard Deviation 514 ml
500
1000
1500
Flow Rate - ml/h
Figure 10.5. Histograms of the noise in overfilled, variable-head tank tests produced by a
uniform distribution of (A) K (30, 60, 90, and 120 cm2'with Tc= 0.75 h and (B)TC(0.5, 1,2, and
3 h) with K = 120 cm2. It can be assumed that the waiting period after topping is long enough
for the exponential time-dependent structural defonnation effects to become small. The standard
deviations of the histograms are 410 ml/h for (A) and 514 ml/h for (B). The histograms are
compiled from estimates of the temperature-compensated volume rates for a 1-h test with* a
method using five equally spaced thermistors having a precision of 0.001 *C. All product-level
changes are measured with a precision of 1 ml/h in a 10-cm (4-in)-diameter fill tube. Other than
deformation, the only product-level or product-volume changes are produced by the thermal
expansion or contraction of the product.
10-14
-------
10.4.2.3 Summary
Large values of K and Tc degrade the performance of a method which does not maintain a
constant head during the test and may introduce a bias if the waiting period after topping is too
short. In all cases, the spread in the histogram (i.e., standard deviation) of the
temperature-compensated volume rates is greater than that achieved when the tank does not
deform.
The impact of deformation on the noise and signal-plus-noise histograms, and therefore on
performance, was discussed in more detail in Section 5. Because it is impractical to measure K
and Tcunder actual field conditions, the results of tests conducted at a variable hydrostatic
pressure must be considered suspect.
10-15
-------
-------
11 Performance of Commercial Test Mel hods
The performance of 25 commercially available volumetric test methods was evaluated by
the EPA at the UST Test Apparatus. A description of each method and its performance are
presented as a lettered appendix in Volume II. The appendices are listed in alphabetical order by
the name of the method. The results are summarized in this section; for ease of reference, the
description of each method and the recommendations for performance improvements are
condensed into a single appendix in Volume I. These two sections are taken from the Volume n
appendices. As summarized in Section 9.4, the evaluation results are presented and discussed in
three subsections of each lettered appendix. In the first subsection, a quantitative estimate of
performance is presented in tables and curves in terms of the probability of detection and
probability of false alarm. These estimates do not include all of the sources of noise that control
performance. Several sources of noise were not included because either the range of applicable
conditions across the country was unknown and an arbitrary selection of conditions might have
unfairly degrade the performance of some methods (e.g., the range of vapor pocket sizes), or the
inclusion of a noise source had such a degrading effect on performance that it would have been
difficult to interpret the result (e.g., topping the tank). The second subsection discusses the
impact of the sources of noise which were not included fully in the quantitative estimate of the
method, and the third section discusses the impact of operational practice on performance.
The performance estimate was made for a 2.43-m (8-ft)-diameter, 30,000-L (8,000-gal)
tank containing unleaded gasoline, under the assumption that the test followed the
manufacturer's specified procedure precisely. The conditions selected for evaluation, as
estimated from empirical relationships developed in 1111, represent a wide range of temperature
conditions found throughout the United States. Tests conducted under more benign product
temperature conditions and less elastic tank conditions would result in better performance.
The performance can be extrapolated to smaller-capacity tanks, to larger-capacity tanks
with the same nominal diameter, and to other gasoline fuels. In principle, the results should also
be applicable to other fuels, like diesel, providing that: the proper coefficient of thermal
expansion for the fuel, as required for temperature compensation, is used; the evaporation and
condensation characteristics of the fuel are not significantly different than they are for gasoline
fuels; and the difference in density between gasoline and other fuels does not adversely impact
either the test protocol or the deformation characteristics. Field evidence suggests that the
performance of volumetric methods is different for diesel fuels than for gasoline fuels, being
sometimes better and sometimes worse. A statistically significant bias has been identified for
diesel fuels in tests that circulate and mix the product [8J.
ll-l
-------
With one exception, all of the estimates were made using the temperature database,
described in Section 6, that was collected independently of the Field Verification Tests. A
smaller temperature database, in which the product was continuously circulated and mixed
throughout the test, was developed using the Petro Tite circulation system, specifically to
evaluate the Petro Tite test method. This database was generated using the same range of
differences in temperature between the product in the tank and the product added to the tank, but
the standard deviation of the histogram of the thermally induced volume changes for the mixed
conditions was larger than for the nonmixed conditions; this suggests that the test conditions
under which these methods were evaluated were more severe than the nonmixed test conditions.
Whether this is true or not is uncertain, because the circulation does add heat to the product and
increases the rate of change of temperature during the test.
11.1 How to Interpret Performance Results
The quantitative estimate of performance assumes that no trapped vapor is present (for
methods that overfilled the tank), that all tanks have the structural deformation properties of one
tank system, that the temperature field has not been disturbed by topping the tank, that the
deformation effects produced by topping the tank were not included although specific
product-level changes required by the protocol were, and that evaporation and condensation are
negligible. The performance estimate further assumes that all tests are conducted precisely as
prescribed by the manufacturer by a highly trained and competent test crew. The estimate does
not, in general, include the uncertainty in the tank chart or coefficient of thermal expansion. The
error introduced by using the manufacturer's tank chart for tanks at the Test Apparatus, and the
coefficient of thermal expansion for the Test Apparatus tanks, however, are included in the
performance estimates. The degradation in performance that occurs when these assumptions are
violated is discussed in this section and in Sections 6, 8, 10, and 12.
The quantitative estimates of performance did not include the potential adverse effects of
trapped vapor for overfilled-tank test methods, and of evaporation and condensation for
partially-filled-tank test methods. These effects on performance are discussed in the second
subsection of each lettered appendix mid in Section 6.4 of the body of this report. The effects of
trapped vapor were not included in the estimates because the distribution of the volume of
trapped vapor in a tank depends on the tank geometry and the operator's expertise in overfilling
and bleeding the tank; that is to say, the effects are unknown. It is assumed that the effects of
trapped vapor will be minimal with experienced operators, and thus, this effect was not included
in the performance estimate. This assumption is not always valid; the reader should understand
that large testing errors can be made if vapor is trapped in the in the tank system. Because of the
11-2
-------
wide diversity of experience, even among operators of the same method, the performance of
overfilled-tank tests presented in this report may be poorer than predicted. The effects of
evaporation and condensation were not included because no simple model could be developed to
predict their impact on performance; these effects are also operator-dependent. The magnitude
of the effect depends on the change of the partial pressure of the gasoline vapor after the fill tube
has been opened to the ambient environment for equipment setup and test preparations.
The quantitative estimates of performance for each method used one set of tank
defonnation characteristics. The estimates did not include a wide range of conditions because
the nationwide distribution of tank-backfill-sou1 elasticity properties is unknown. This effect is
also discussed in the second subsection of each lettered appendix and in Section 6.4.
The effect of all three sources of noise not included in the performance estimate is to
increase the standard deviation of the noise and to introduce a bias for the defonnation effects.
The effects of trapped vapor and evaporation and condensation may also exhibit a systematic
trend on a seasonal basis.
The perfonnance estimates assume that the height-to-volume conversion factor, Aeff, is
accurately estimated for each method. In general, this is true because most methods estimate Aeff
experimentally. Any method that does not already do so can easily include this simple
measurement in its protocol, and thus no manufacturer was penalized for failing to describe it in
the protocol. It should be noted, however, that the estimates presented in this report could be in
error by as much as 50% or more if this experimental measurement of Aeff is not implemented.
The performance estimates do not include the degradation in perfonnance in
overfilled-tank tests produced by topping the tank. Even when a level change was required, the
spatial inhomogeneities in the temperature field produced by topping were not included. Again,
to realize the performance presented in this report, changes must be made to many of the
methods to account for the temperature and defonnation effects of topping. It should be pointed
out, however, that the deformation effects produced by the volume changes during the test (e.g.,
thermal expansion or contraction of the product) were included in the perfonnance estimates.
The effects of topping will have very little impact on test method perfonnance if there is a
waiting period of at least 2 to 3 h after adding product to the tank and before starting a test.
Topping effects were not overly emphasized because the overall degrading effects of topping the
tank are large enough to invalidate a test, and it is assumed that most manufacturers will change
their protocol to minimize this problem.
11-3
-------
The performance of each method should ultimately depend on how well the volume
changes produced by thermal expansion or contraction of the product are compensated for and
on the magnitude of the evaporation and condensation of the vapor from the product surface and
tank walls. The effects of trapped vapor, structural deformation, and the spatial inhomogeneities
in the temperature field produced by topping and delivery can and should be minimized.
Many of the methods evaluated exhibited a bias (i.e., a noise histogram with a nonzero
mean that is statistically different than zero) in the noise histogram produced, in general, by the
deformation of the tank. The magnitude of the bias for the test conditions used in the evaluation
can be estimated from the data provided in the lettered appendices in Volume II. The bias is
probably not as large as it might be under actual field testing conditions where the range of
deformation conditions will be larger. The bias will change as the test conditions change. The
performance estimates presented in Subsection 11.2 do not include the effects of the bias. In
general, a method cannot be properly evaluated unless the bias can be quantified and removed.
A performance estimate was made for all methods nonetheless. This assumes that the bias has
been removed. This is essential to good performance, and it is assumed that all manufacturers
will identify the reason for the bias and either quantify and remove it, or modify the method to
eliminate its presence. The effects of the bias present in these data can be interpreted using the
PD-vs.-threshold and PFA-vs.-threshold performance curves in the lettered appendices.
It cannot be emphasized enough that the results of this evaluation, particularly those shown
in Tables 11.1, 11.3, and 11.5, should be used cautiously, because the majority of the test
methods evaluated will probably have changes made since this evaluation. The detailed
description of the performance of each method and the suggestions for improving performance
can be found in the lettered appendices in Volume II as well as in the appendix in Volume I.
11.2 Performance Results
The evaluation results are summarized and discussed in four categories: (1) overfilled-tank
test methods at variable head, (2) overfilled-tank test methods at constant head, (3)
underfilled-tank test methods, and (4) those methods for which no estimates of performance
could be made. The results of the quantitative estimates of performance for the first three
categories are summarized in Tables 11.1, 11.3, and 11.5 (in Sections 11.2.1, 11.2.2, and 11.2.3).
The results are presented in both SI and English units; Tables 11. l(a), 11.3(a), and 11.5(a) are in
SI units and Tables 1 l.l(b), 11.3(b), mid 1J .5(b) are in English units. The tables include the
mean and standard deviation of the noise histogram, the number of points in the histogram, the
PD and PFA achieved with the method (using the manufacturer's standard detection criterion or
threshold) against a 0.38-L/h (0.10-gal/h) leak rate, and the smallest leak rate that can be detected
11-4
-------
with (1) a PDof 0.95 and PFAof 0.05, and (2) a Pn of 0.99 and PFAof 0.01. The latter two
estimates are representative of capability, but are not actually being achieved because the
manufacturer's standard detection threshold was changed to make these estimates. To make the
two latter estimates, thresholds that would yield P,.As of 0.05 and of 0.01, respectively, were
derived from the noise histogram. The performance actually being achieved by these methods
may be considerably different, because the actual threshold used by the manufacturer may be
different. Additional factors, without which it is impossible to interpret the quantitative estimate
of performance, are tabulated and presented for each category in Tables 11.2, 11.4, and 11.6. Of
particular note is the length of the waiting time after topping. The six remaining methods could
not be fully evaluated (see Section 11.2.4)
The leak rate measurable by these systems with a threshold established by a PFA of 0.05
ranged from 0.26 to 6.97 L/h (0.07 to 1.84 gal/h), with a probability of detection of 0.95 and
probability of false alarm of 0.05. Five of the methods achieved a performance that was better
than 0.57 L/h (0.15 gal/h), and a total of eight methods had a performance that was better than
0.95 L/h (0.25 gal/h). The leak rate measurable by these systems with a threshold established by
a PFA of 0.01 ranged from 0.47 to 12.95 L/h (0.12 to 3.42 gal/h) when the probability of detection
increased to 0.99 and the probability of false alarm decreased to 0.01. Only one of the methods
achieved a performance better than 0.57 L/h, but five methods achieved a performance between
0.57 L/h and 0.95 L/h.
Over 90% of the methods evaluated claimed a performance of 0.19 L/h (0.05 gal/h), but
none of the methods achieved this performance with a probability of detection of 0.99 and a
probability of false alarm of 0.01. One of the methods (LMS-750 Leak Detection System)
claimed a performance of 3.8 L/h (1.0 gal/h), and one method (TLS-250) claimed a performance
of 0.76 L/h (0.20 gal/h), the latter with a PD of 0.98 and a PPA of 0.005. With this one exception,
none of the methods either specified or attempted to support a PD or PF.A
11.2.1 Overfilled-Tank Test Methods at Variable Head
In twelve test methods, the tank is overfilled to a level within the fill tube or an
above-grade standpipe, and product-level changes are measured and converted to volume using a
height-to-volume conversion factor. The major contributions to the noise are thermal expansion
and contraction of the product, expansion and contraction of the trapped vapor (due to pressure,
temperature, and evaporation and condensation within the vapor space), and structural
deformation. Some of the methods use an experimentally determined height-to-volume
conversion factor and others calculate it from an estimate of the product's surface geometry. All
of the methods compensate for temperature and attempt to use one or more waiting periods to
-------
ensure that the temperature and deformation effects will become small. Trapped vapor may be a
problem because it is difficult to estimate how much trapped vapor, if any, exists in the tank after
it has been overfilled.
The performance of eleven methods is shown in Table 11.1. One method in this category
(AUTAMAT) was not evaluated (see Section 11.2.4).
Table 1 I.I (a). Performance of Overfilled-Tank Test Methods at Variable Head (SI Units)
Test Method Name
AES/Brockman Leak
Detecting System*
Ainlay Tank 'Tegrity
Tester
Computerized VPLT
Tank Leak Testing
System
EZY CHEK
Leak-O-Meter
LiquidManager
Mooney Leak
Detection System
PACE Leak Tester
Portable Small Leak
Detector (PSLD)
S.M.A.R.T.
Tank Auditor
Mean
(L/h)
-0.167
0.076
0.023
0.048
-1.060
0.307
-0.266
0.143
-0.192
-0.033
1.048
Standard
Deviation
(L/h)
0.910
0.470
0.230
0.184
2.072
0.168
0.551
0.810
0.871
0.366
1.107
Number
112
284
99
399
231
79
196
245
135
81
207
Pn and PFA
to Detect
a 0.38 L/h
(0.10 gal/h)
Leak Rale
0.45,0.34
0.50,0.3 1
0.66,0.19
0.86,0.15
0.57,0.49
0.80.0.14
0.47,0.38
0.37,0.32
0.63,0.32
0.58,0.32
0.57,0.43
Smallest
Detectable
Leak Rate
for
PD=0.95
PFA=0.05
(L/h)
6.79
2.97
1.08
0.62
6.96
0.75
3.13
6.97
3.05
2.25
6.27
Smallest
Detectable
Leak Rate
for
PD=0.99
PFA=0.01
(L/h)
12.95
3.93
1.84
0.93
10.80
1.25
4.58
11.12
5.60
3.43
12.52
* Data analysis algorithms for this method had to be modified in order to determine its minimum
detectable leak rate.
The quantitative performance estimates made for variable head, overfilled-tank tests are
affected by three factors: (1) the temperature and deformation effects produced by topping the
tank, (2) the potential for trapping an unknown volume of vapor during a test, and (3) whether or
not the height-to-volume conversion factor is measured experimentally or calculated
theoretically from the tank geometry.
11-6
-------
Table I l.l(b). Performance of Overfilled-Tank Test Methods at Variable Head (English Units)
Test Method Name
AES/Brockman Leak
Detecting System*
Ainlay Tank Tegrity
Tester
Computerized VPLT
Tank Leak Testing
System
EZY CHEK
Leak-O-Meter
LiquidManager
Mooney Leak
Detection System
PACE Leak Tester
Portable Small Leak
Detector (PSLD)
S.M.A.R.T.
Tank Auditor
Mean
(gal/h)
-0.044
0.020
0.006
0.013
-0.280
0.081
-0.070
0.038
-0.051
-0.009
0.277
Stand urd
Deviation
(gal/h)
0.240
0.124
0.061
0.049
0.547
0.044
0.146
0.214
0.230
0.097
0.292
Number
1 12
284
99
399
231
79
196
245
135
81
207
PD and PFA
(o Deled
a 0.38 L/h
(0. 10 gal/h)
Leak Rate
HVW
0.45,0.34
0.50,0.31
0.66,0.19
0.86,0.15
0.57,0.49
0.80,0.14
0.47,0.38
0.37,0.32
0.63,0.32
0.58,0.32
0.57,0.43
Smallest
Detectable
Leak Rate
for
PD=0.95
PFA=0.05
(gal/h)
1.79
0.78
0.28
0.16
1.84
0.20
0.83
1.84
0.81
0.59
1.66
Smallest
Detectable
Leak Rate
for
PD=0.99
PFA=0.01
(gal/h)
3.42
1.04
0.49
0.25
2.85
0.33
1.21
2.94
1.48
0.91
3.31
* Data analysis algorithms for this method had to be modified in order to determine its minimum
detectable leak rate.
The effects of topping were not included in these evaluations, because the degradation that
occurs from topping may be large (i.e., it may degrade the performance presented here by a
factor of 2 or more) but is easy to correct. The degradation will be large if the product added to
the tank is at a significantly different temperature than the product in the tank, or the level
change is a large fraction of the tank diameter. The effects of topping can be avoided by waiting
for the spatial inhomogeneities in the temperature field (e.g.. Figures 8.5 and 8.6) and the
product-level changes due to structural deformation to become negligible. The minimum
waiting tune may be different for each effect. The quantitative performance results presented
here should not be considered valid if the waiting period is too short. Some methods use a
dynamic analysis algorithm to minimize the waiting period for each test. If the algorithm does
not result in a waiting period of 1 to 3 h or more, the method vShould be considered suspect.
11-7
-------
After a tank has been topped, accurate temperature compensation is difficult to achieve
unless a waiting period is observed; this is because topping produces spatial inhomogeneities in
the product temperature field. A waiting period of at least 2 h, preferably 3 h, should be part of
the test protocol. The errors in temperature compensation that would be produced by topping
were not included in any of the performance estimates. Therefore, unless there is an adequate
waiting period defined in the protocol, the actual performance of a method in the field can be
significantly lower than that which is presented here. The methods subject to the effects of
topping are identified in Table 11.2. The performance estimate made for any method that has a
waiting period of at least 3 h would not be affected. All of the methods in this study, however,
may be adversely affected because none has a 3-h waiting period.
Table 11.2. Factors That Affect the Performance of Overfilled-Tank Tests Conducted at
Variable Head
Test Method
Name
AES/Brockman Leak
Detecting System
Ainlay Tank Tegrity Tester
Computerized VPLT Tank
Leak Testing System
EZY CHEK
Leak-O-Meter
Liquid Manager
Mooney Leak Detection
System
PACE Leak Tester
S.M.A.R.T.
Tank Auditor
Waiting
Period After
Product
Delivery
(h)
4
8
12
6
12
12
12
12
12
8
Waiting
Period
After
Topping
(h)
0
2
0
1
0.75
1
Variable3
0
0
0
H-to-V
Conversion
Factor1
3
4
2
2
3
3
1
3
2
2
Bias2
2
1
1
1
3
2
2
2
1
3
1 The numeral 1 indicates that the height-to-volume conversion factor has been theoretically
estimated from the tank chart or the known geometry of the tank; 2 that it has been obtained
by displacement of fluid using a solid of known volume; 3 that it has been obtained by
displacement through addition or removal of product; and 4 that it has been obtained by
means of a direct volume measurement.
2 The numeral 1 indicates a bias less than or equal to 100 ml/h; 2, greater than 100 but less
than 400 ml/h; and 3, greater than 400 ml/h.
3 Stipulates a variable waiting period after tank has been topped.
11-8
-------
Table 11.2 also shows the minimum waiting periods specified by each test method's
protocol. The performance of methods which test within 4 to 6 h after a product delivery will
also be affected by the spatial inhomogeneities of the temperature field. However, these effects
are included in the performance estimates.
The effects of structual defonnation are an important source of error in overfilled-tank test
methods. These errors may occur whenever the product level in the fill tube or standpipe
changes, specifically after a product delivery or topping of the tank, in tests which require
measurements at two or more levels, or as a result of any product-level changes that occur during
the test itself.
The performance of overfilled-tank test methods that allow the product level to change
during the test is flawed because the induced product-level changes cause the tank to defonn
continuously, making conversion to true volume changes nearly impossible. The product-level
changes measured during a test are either larger or smaller than expected. If the waiting period
after topping the tank is too short (e.g., less than 2 to 3 h), the measured volume rate depends
almost entirely on the product-level change due to the exponential changes caused by
deformation. Such a measurement will tend to suggest that the tank is leaking. The effect of
exponential changes is minimized if the time constant of the defonnation is small or if the
waiting period is long in comparison. If an adequate waiting period is used, the measured rate of
change of product level (and product volume) will always be a fraction of the expected rate.
Because the volume rate estimated for the product-temperature changes is not affected by the
deformation effects in the fill tube, the temperature-compensated volume rate will always be in
error. This flaw is subtle and not always apparent in field measurements, because the
product-level changes will tend to approach a constant value. The error depends on the unknown
structural defonnation characteristics of the tank-backfill-soil system. The flaw can be
minimized either by increasing the surface area of the product so that the hydrostatic pressure
remains essentially constant during the test or by releveling the product in the tank.
The performance estimates were made using only one set of deformation characteristics.
The value of K and Tc used in the analysis of each method is noted in each lettered appendix in
Volume n. The values of K measured in the steel tank of the Test Apparatus indicated that K
was smaller for tests conducted with the product level above grade than for tests conducted with
the product level in the fill tube. As a consequence, the effects of deformation on the
performance estimates made in Table 11.1 are less severe for above-grade tests than for
below-grade tests. The time constant used in the analysis was typically 0.75 h, small enough
(and deliberately selected for that reason) that a 2- to 3-h waiting period, approximately 3 time
constants, would minimize the exponential defonnation effects on performance. The time
11-9
-------
constant of the steel and fiberglass tanks was approximately 3 h, so that a minimum waiting
period of 9 h or more might be required. Estimates of how Tc and K affect performance were
given in Sections 6.4, 8.3 and 10.4.
The performance of any method is degraded when it does not conscientiously attempt to
identify the exponential volume changes produced by structural deformation (i.e., those that
result from changing the level of the product in the fill tube or standpipe) and wait for these to
become small. The deformation effects produced by the product-level change required to top the
tank were not included in these estimates. The 2- to 3-h waiting period required to minimize the
spatial inhomogeneities in the product temperature field may not be long enough to minimize the
effects of deformation. However, the methods in Table 1 1.2 that do not wait at least this long are
affected by product temperature and may also be affected by defonnation. In general, most
methods unintentionally include a "start" waiting period because of the time it takes to top the
tank. Unless it is sufficently long, minimum waiting period is no guarantee that defonnation has
subsided. Several of the methods incorporate an algorithm to identify and wait out the effects of
deformation. The sensitivity of the methods to deformation can be assessed from the waiting
periods observed after product delivery or topping. At the minimum, the waiting period must be
at least as long as the one for product temperature disturbances (those produced by the addition
of product), since the defonnation effects may take longer to subside.
Large testing errors, 50% or more, even for well-bled tanks, can occur if the effects of the
instantaneous tank deformation and trapped vapor are not Included in the height-to-volume
conversion factor, Aeff. A theoretical estimate of the cross-sectional area of the product surface,
A, does not include these effects. Furthennore, accurate estimates of the cross-sectional area of
the product surface area are difficult to make in operational practice because of the unknown
presence of additional fill tubes and piping. Even when these are known to exist, their diameters
may not be known. An experimental estimate of \.lt is therefore required. In addition, methods
which determine Acll by adding product to the tank may introduce spatial inhomogeneities in the
temperature field. Because all methods can easily implement an experimental measurement of
the height-to-volume conversion in their protocol, it was assumed in the perfomiance estimate
that all methods used an experimental estimate of AUI. Table 1 1.2 indicates that only one method
does not use an experimental measurement. In a case such as this, when the height-to-volume
conversion factor is estimated only from the geometry of the tank, the performance estimates
made in this report are likely to be in error by 50% or more. It is assumed that as a result of this
evaluation all methods will implement the technique of experimentally estimating Acff. Actf is not
a valid height-to-volume conversion factor if the product level is allowed to fluctuate in the fill
-------
tube or standpipe during the test; Acff + K is the correct conversion factor. Since K is not usually
known during a test and since K may be large compared to A,.lf, large errors are possible in
variable-level tests. If K is small compared to Acll, then accurate tests can be conducted.
All of the overfilled-tank test methods have the potential for trapping vapor. The effects of
trapped vapor were not included in the performance estimates because no severe problems are
encountered unless (1) the effects of the trapped vapor (as well as the instantaneous defonnation
of the tank) are not included in the height-to-volume conversion factor, (2) the volume of trapped
vapor is large, (3) large changes in atmospheric pressure occur during a test, (4) the temperature
of the vapor is affected by ambient changes, or (5) the trapped vapor is not saturated. It was
observed during the Field Verification Tests that the more experienced test crews did a better job
in removing vapor from the tank before a test was begun. A method of estimating an upper
bound on the volume of the trapped vapor is presented in Section 8.2 of this report.
The magnitude of the bias is also summarized in Table 11.2. To achieve the performance
presented in Table 11.1, the bias must be identified and removed. Removal of a bias is a
complex task. It cannot be correctly done simply by using the magnitude of the bias presented in
this evaluation, because the magnitude of the bias may change with the testing conditions.
Additional developmental work is required by the manufacturer.
In summary, the quantitative estimates of performance in Table 11.1 do not include the
degradation in performance produced by topping; they do assume the benefits of an
experimentally measured height-to-volume conversion factor. The errors produced by topping
and by not measuring the h-to-v conversion factor directly can he corrected immediately, and it
is assumed that they will be, because simple protocol changes are all that is necessary to do so.
If these modifications have not been made, the performance estimate presented in Table 11.1
should be considered invalid. Even if the topping-related exponential defonnation effects are
properly handled, the potential for testing errors in the field is large (in the case of variable-level
tests) because of the problems in making volumetric interpretations of the product-level
measurements in the fill tube and standpipes. Reliable performance requires constant hydrostatic
pressure. Significant improvements in performance would be realized if a waiting period were
incorporated into the protocol after topping, if the test were conducted at a constant hydrostatic
pressure head, and if an experimental estimate of A,,,, were used to convert product-level changes
to product-volume changes.
-------
11.2.2 Overfilled-Tank Test Methods at Constant Ilcud
In three of methods evaluated, the tank is overfilled and the volume of product necessary
to maintain a constant level during the test is measured. The performance of these methods is
presented in Table 11.3.
Table 11.3(a). Performance of Overfilled-Tank Test Methods at Constant Head (SI Units)
Test Method Name
Leak Computer
MCG-110
Petro The
Mean
(L/h)
0.005
0.206
0.002
Standard
Deviation
(L/h)
0.096
0.119
0.209
Number
132
97
25
PD and PFA
to Detect
a 0.38 L/h
(O.lOgal/h)
Leak Rate
-------
Table 11.4. Factors That Affect the Performance of Overfilled-Tank Tests Conducted at
Constant Head
Test Method
Name
Leak Computer
MCG-1100
Petro The
Waiting
Period After
Product
Delivery
(h)
Variable
2
0
Waiting
Period
After
Topping
(h)
Variable
Variable
Variable3
H-to-V
Conversion
Factor1
4
3
3
Bias2
1
2
1
1 The numeral 1 indicates that the height-to-volume conversion factor has been theoretically
estimated from the tank chart or the known geometry of the tank; 2 that it has been obtained
by displacement of fluid using a solid of known volume; 3 that it has been obtained by
displacement through addition or removal of product; and 4 that it has been obtained by
means of a direct volume measurement.
2 The numeral 1 indicates a bias less than or equal to 100 ml/h; 2, greater than 100 but less
than 400 ml/h; and 3, greater than 400 ml/h.
3 Product is circulated (5 to 8 min/1,000 gal at high level and 0 h at low level).
Several of the methods in Table 11.4 do not have a minimum waiting period specified in
the protocol, but instead have an algorithm that analyzes the data, as they are being collected, to
determine if the effects of thermal inhomogeneities and structural deformation have subsided.
These methods are identified by the superscript 3. The thermal effect is also minimized in other
methods that mix the product before or during the test. The success of these approaches can be
assessed from the degree of temperature compensation achieved by each method during the Field
Verification Tests; this is presented in the lettered appendices in Volume n.
11.2.3 Parially-Filled-Tank Test Methods
Ten of the methods that participated in the evaluation program routinely test in partially
filled tanks. All of these tests are conducted at a specified product level which is less than 95%
of the capacity of the tank. While the exact level is not critical, most of these tests require that
product be added to the tank before a test is conducted. The addition of product will affect
performance if the waiting period between the product addition and the start of the test is too
short.
The major contributions to noise are instrumentation, thermal expansion and contraction of
the product, evaporation and condensation from the product surface and tank walls, and
structural deformation. The degrading effects on performance produced by the spatial
inhomogeneities in the product temperature field and the structural deformation resulting from
11-13
-------
product delivery are included in these estimates. Because the test is conducted at a constant
hydrostatic pressure, the measured product-level changes can be converted to product-volume
change using a height-to-volume conversion factor. All of the methods compensate for
temperature and attempt to wait for the deformation effects to become small. None of the
methods listed in Table 11.5 compensates for evaporation and condensation of the product. In
about half of the cases, better performance could be achieved if the system noise were lowered
(i.e., if there were better precision of the product-level and product-temperature sensors).
Table 11.5(a). Performance of Partially-Filled-Tank Test Methods (SI Units)
Test Method Name
Gasoline Tank Monitor
(GTM)*
Gilbarco Tank
Monitor*
Inductive Leak
Detector 3100
Tank Sentry II
TLS-250*
Mean
(L/h)
0.105
0.016
0.055
-0.093
-0.016
Standard
Deviation
(L/h)
0.408
0.075
1.012
0.154
0.142
Number
13
59
45
23
46
PD and PFA
to Detect
a 0.38 L/h
(0.10gal/h)
Leak Rate
(PD^FA)
0.73,0.21
0.96,0.003
0.72,0.33
0.89,0.16
0.15,0.001
Smallest
Detectable
Leak Rate
for
PD=0.95
PFA=0.05
(L/h)
1.35
0.26
4.23
0.58
0.51
Smallest
Detectable
Leak Rate
for
PD=0.99
PFA=0.01
(L/h)
1.91
0.47
9.54
0.89
0.90
* In a precision test mode rather than in its normal operating mode as an automatic tank gauging
system (ATGS)
The performance of five partially-filled-tank test methods is presented in Table 11.5. The
remaining five (DWY Leak Sensor, INSTA-TEST, LMS-750, OTEC Leak Sensor, and
Tank Monitoring Device (TMD-1)), could not be evaluated for the reasons discussed in
Section 11.2.4.
The quantitative estimate of performance made for the methods in Table 11.5 do not
include the effects of evaporation and condensation of product from the product surface and the
tank walls. While these effects are believed to be small, the performance of all the methods
presented here will be degraded somewhat. Preliminary estimates for a limited data set suggest
that evaporation and condensation would affect the detectable leak rate presented in Table 11.5
by less than 0.19 L/h (0.05 gal/h), providing that the evaporation and condensation effects that
occur whenever the tank is opened to the ambient atmosphere have ceased. No quantitative
estimate of this latter effect has been made, but it was seen to be present each time the tanks at
11-14
-------
Table 11.5(b). Performance of Partially-Filled-Tank Test Methods (English Units)
Test Method Name
Gasoline Tank Monitor
(GTM)*
Gilbarco Tank
Monitor*
Inductive Leak
Detector 3 100
Tank Sentry II
TLS-250*
Mean
(gal/hl
0.027
0.004
0.015
-0.024
-0.004
Standard
Deviation
(gal/h)
0.108
0.020
0.267
0.041
0.037
Number
13
59
45
23
46
PD and PFA
to Detect
a 0.38 L/h
(0.10gal/h)
Leak Rate
(PD.PFA)
0.73,0.21
0.96,0.003
0.72,0.33
0.89,0.16
0.15,0.001
Smallest
Detectable
Leak Rate
for
PD=0.95
PFA=0.05
(gal/h)
0.36
0.07
1.12
00.15
0.13
Smallest
Detectable
Leak Rate
for
PD=0.99
PFA=0.01
(gal/h)
0.50
0.12
2.52
0.23
0.23
* In a precision test mode rather than in its normal operating mode as an automatic tank gauging
system (ATGS)
the Test Apparatus were opened to the ambient environment. Typically, a 6-h waiting period is
required for the tank to reestablish a quasi-equilibrium condition after the it has been closed to
the ambient environment. The performance of the methods given in Table 11.5 would probably
be degraded if the waiting period were less than 6 h. The potential degradation in the
performance estimate made here can be estimated from the minimum waiting periods after a
product delivery that are included as part of the manufacturer's test protocol and that are given in
Table 11.6.
Two of the methods evaluated are acoustic systems. Both are subject to a potentially large
seasonal bias in the measurement of product level, because changes in product temperature,
which affect the propagation of the acoustic signal, cannot be completely compensated for. This
is best observed in [21].
The performance estimates for partially-filled-tank test methods do not include the
uncertainty in the coefficient of thermal expansion; they do assume that the tank chart is
sufficiently accurate to estimate the height-to-volume conversion factor accurately, a reasonable
assumption when product level is not in the upper part of the tank.
The high level of performance achieved by three of these methods requires waiting periods
of 18 to 24 h and test durations of 4 to 12 h. This represents a total test duration that is
significantly longer than that of many of the other methods.
11-15
-------
Table 11.6. Factors That Affect the Performance of Partially-Filled-Tank Tests
Test Method
Name
Gasoline Tank Monitor
(GTM)3
Gilbarco Tank Monitor3
Inductive Leak Detector 3 100
Tank Sentry II
TLS-250
Waiting
Period After
Product
Deliven
(h)
24
18
3
24
2
Waiting
Period
After
Topping
(h)
N/A
N/A
N/A
N/A
N/A
l-l-lo-V
Conversion
Factor1
1
1
1
1
1
Bias2
1
1
1
1
1
1 The numeral 1 indicates that the height-to-volume conversion factor has been theoretically
estimated from the tank chart or the known geometry of the tank; 2 that it has been obtained
by displacement of fluid using a solid of known volume; 3 that it has been obtained by
displacement through addition or removal of product; and 4 that it has been obtained by
means of a direct volume measurement.
2 The numeral 1 indicates a bias less than or equal to 100 ml/h; 2, greater than 100 but less
than 400 ml/h; and 3, greater than 400 ml/h.
3 Unknown magnitude of systematic error introduced by acoustic measurement of product.
11.2.4 Methods for Which No Performance Ivslimates Were Made
There were no performance estimates made for 6 of the 25 methods evaluated. Two of the
six methods (INSTA-TEST and TMD-1) did not satisfactorily perform a valid test during the
three days of the Field Verification Tests. Any method that could not complete a valid test
during the three-day period was not evaluated; moreover, no data were obtained to validate the
test method model. The temperature and product-level data obtained for three of the other
methods (AUTAMAT, DWY Leak Sensor, and LMS-750) showed that the measurement
systems were clearly not operating as the manufacturers had indicated; the models of these
methods could not, therefore, be validated. The fourth method (OTEC Leak Sensor) could not
be evaluated because the Tank Apparatus had not been properly configured for some of the tests.
Typically, these were the methods that attempted to compensate for product-temperature changes
without measuring temperature. In the case of these methods, the performance estimates made
using the test method model would have been significantly better than what the field
measurements indicated.
11.2.5 Summary of Test Results
The estimates of performance are summarized in Tables 11.7 and 11.8. Table 11.7
summarizes the methods' actual performance using the manufacturers' detection criteria. Table
I 1-16
-------
11.8 summarizes performance using detection criteria that produce a probability of false alarm of
0.05 and 0.01. The performance results are expressed as the smallest leak rate that can be
detected with probabilities of 0.95 and 0.99, respectively.
Table 11.7. Estimates of Test Method Performance In Terms of PD and PFA for the Detection of
a Leak Rate of 0.38 L/h (0.1 gal/h) Using the Manufacturer's Detection Threshold
0.90
0.65
0.35
0.10
PD
-------
11.3 Instrumentation
For best performance, the instrument-system noise should be less than the ambient noise.
The precision and accuracy requirements of the product-level and temperature measurement
systems for the detection of a given leak rate are more stringent for the partially-filled-tank test
methods than for the overfilled-tank test methods. For methods in which the tank is partially
filled, the precision of the product-level measurements is generally marginal (as compared to the
desired 1.3 (im) if detection of leaks as small as 0.19 L/h (0.05 gal/h) is desired. The
specification for temperature is generally 0.001 °C. If achieved, this is adequate to perform the
temperature compensation necessary to detect 0.19-L/h leaks.
An analysis of the instrumentation system requirements for the Test Apparatus was
presented in Section 7. A similar analysis should be done by each manufacturer to determine the
limitations of the measurement system on performance. With few exceptions, the
instrumentation specifications do not limit the performance of the methods evaluated.
An analysis of the instrumentation of each method was performed to determine the
minimum leak detectable with a PD of 0.99 and a PFA of 0.01. Methods in which the tank is
overfilled generally have more than adequate product-level precision.
The product-level and temperature measurement systems of all the methods evaluated in
this study were calibrated during the Field Verification Tests. As a general rule, the
product-level and temperature measurement systems did not meet their precision claims. The
calibration performed at the Test Apparatus was the first for some of these instruments. Most
methods specify either no regular calibration of the instruments or one that is not sufficient to
verify precision properly. Calibrating the temperature measurement systems is particularly
important, since a precision of O.OOTC is extremely difficult to achieve and maintain under field
conditions.
11.4 Test Operations
All 25 of the test methods that were evaluated participated in the Field Verification Tests at
the UST Test Apparatus in Edison, New Jersey. Each manufacturer was requested to test a tank
with his own equipment and test crew, following his established protocol. The overall testing
experience is summarized below; the specific experiences for each method are described in more
detail in the lettered appendices (Volume II).
1-18
-------
11.4.1 Setup
It generally took the manufacturers 1 to 2 h to set up the test equipment during the Field
Verification Tests. As a rule, most manufacturers believed that the Test Apparatus tanks were
easier to test than those encountered in the field. On several occasions, Test Apparatus failures
(chiefly, minor leaks along thermistor cables) interfered with a few of the setup preparations;
these problems were identified, as appropriate, in the individual manufacturer evaluations. Most
of the test crews took longer than expected, given the schedules indicated in their protocols.
Based on experience during the Field Verification Tests, it is evident that, generally, the test
preparations and conduct of a test, not including the delivery and waiting period, would require
that a service station be closed most of the business day.
11.4.2 Instrumentation
Several of the manufacturers had serious instrumentation problems. This was particularly
true of systems which were in the development and testing phases, or which were one of a kind,
or were not considered "off-the-shelf systems.
11.4.3 Test Protocol
The majority of the methods that were evaluated followed their protocols reasonably well.
The three most common violations of protocol were:
o not following a precise schedule in topping the tank
o not following (or not having) a well defined procedure for stopping a test
o not following (or not having) a well defined procedure for starting a test
11.4.4 Data Tabulation and Calculation
A few calculational and data tabulation errors were experienced during the Field
Verification Tests, but this was not considered a major problem. Software errors were also
identified during the model validation phase in several systems that collected data automatically.
It must be assumed that the best test crews were asked to perform these tests, so that mistakes
would be minimal. Furthermore, in many cases, high-level managers and executives (including
the inventor or head engineer of the team that developed the system) were present for the tests;
presumably, they checked the results. Whether errors in calculation or data tabulation are a
significant problem in the field is unknown. Several testing organizations with quality assurance
programs suggested that the error rate may be as high as 5%; however, they indicated that these
errors could be identifed and corrected.
11-19
-------
11.4.5 Arbitrary Operator Decisions
All of the test methods require one or more operators. In some methods, the role of the
operator is simply to implement a well-structured test protocol. More commonly, the operator is
allowed or even required to alter the protocol based on his own judgment. During the
development of the mathematical model for each method, any operator judgments required to
start or stop a test, or to interpret the results of the test, were quantified by the manufacturer. In
most instances, the manufacturers had a great deal of difficulty either doing this or seeing the
need to do it. Without quantitative judgments, however, the test method could not be evaluated.
The most important judgment concerns what the operator does when a tank is declared
leaking. Usually, another test is conducted by simply continuing the data collection, and the tank
is declared tight if the threshold is not exceeded in the second test. The ramifications of this
approach are discussed in Sections 5 and 12.
11.5 Interpretation of the Evaluation Results
A set of test conditions was developed for this evaluation. The applicability of the results
of the evaluation depends on whether the test conditions used and the performance estimates
made in this evaluation are representative of those generally encountered at underground storage
tank facilities.
The results of the evaluation are controlled by the temperature database and the elasticity
constant of the steel tank at the Test Apparatus. The temperature conditions used in this study
were generated from a climatic analysis (based on 77 cities located throughout the United States)
of the difference in temperature between the product in the tank and that in the truck immediately
prior to delivery. However, the value used for the elasticity constant is the only known value. It
is anticipated that a wide range of tank defonnation conditions will be encountered in the field.
Section 10 describes a small study whose objective was to determine how performance is altered
when different conditions are encountered. This study utilized generic methods representative of
those evaluated in the larger study, and, using these generic methods, developed the performance
norms described in Section 10. Results were compared on a relative basis.
The results obtained at the UST Test Apparatus were compared to field evaluations of
three methods similar to the ones in this study [6-8, 21]. These field evaluations, which used
data obtained at retail and industrial underground storage tank facilities, had already been made,
or were being made at the time of this study. The protocols used by these methods were
different, and the temperature conditions were less encompassing. All of the data were collected
either (a) throughout the year in one state with a moderate climate (i.e., California), or (b)
11-20
-------
throughout the 50 states but during only one season (i.e., late spring and early summer).
Nevertheless, the results of these field evaluations are consistent with the results obtained at the
UST Test Apparatus and presented here.
11.6 Expected Performance after Modifications
Many of the methods that the EPA evaluated in this study are capable of significant
performance improvements. A qualitative estimate was made of the performance that might be
achieved by each method; the results are summarized in Table 11.9. The column labeled "After
Minor Modifications" (e.g., changes relating to the sequencing and timing of data collection)
shows that over 30% of the methods (6 out of 19) should be capable of achieving a high level
performance against a 0.38-L/h (0.10-gaI/h) leak. The last column, labeled "After Protocol and
Equipment Modifications," shows that over 60% of the methods (12 out of 19) should be capable
of achieving an equally high performance against a 0.38-L/h leak.
Table 11.9. Estimate of the Performance of Volumetric Test Methods Evaluated at the UST
Test Apparatus after Two Levels of Modifications; Expressed in Terms of the Smallest Leak
Rate That Can Be Detected with a PD of 0.99 and a PFA of 0.01
Number of Test Methods Able to Detect This Leak Rate
Detectable Leak Rate Evaluation After Minor After Protocol
(L/h) Results Modification and Equipment
(Before Modifications
Modifications)
0.19
-------
A careful review of the instrumentation and protocol of the methods evaluated shows that
those with the best performance are making optimum use of their instrumentation and testing
approaches; their performance comes close to the technological limits. It is equally clear, based
on their instrumentation and testing approaches, that the methods which were not in the upper
third have the potential, as yet unrealized, to do better. Some methods performed more poorly
than expected simply because of an easy-to-correct protocol error. After performance
modifications have been made, many methods are expected to achieve performance levels that
are more or less equal. A number of manufacturers have already begun making the suggested
modifications. Thus, the temptation to rank these methods is premature, and should be avoided.
The effects of topping are not included in the evaluation of the overfilled-tank test
methods. The majority of these methods do not utilize a waiting period that is adequate to avoid
the degradation effects of topping. As a consequence, the perfonnance of these methods may be
significantly poorer than what is presented here. If an adequate waiting period were added to the
methods' protocol, the results presented here would be representative of the methods'
perfonnance.
Before any method is selected, it is advisable that the entire performance evaluation report
be read, including the appendices. The user needs to make an assessment about whether the
method's protocol can be followed and whether it is particularly sensitive to operator influence
or mistakes. The user also needs to determine whether the version of the method being offered
is: (1) the original that was actually evaluated under this study; (2) an improved version that has
also been evaluated under the same protocol used in this study (and whose performance can be
compared on an equal basis with the 25 in the original evaluation); (3) an improved version that
has been evaluated under some other protocol (and whose perfonnance may not be comparable
on an equal basis); or (4) an "improved" version that has not been evaluated (and whose
performance is, therefore, not known).
11-22
-------
12 Performance Improvements
Simple and low-cost modifications will permit many of the methods evaluated in this
program to improve their performance dramatically. The determining factor in whether these
modifications are simple to make, or can be done at low cost, is the method itself. The best
candidates for such modifications are methods (1) in which the test protocol, data analysis, and
detection criterion are implemented manually, (2) for which only a few prototypes exist, or
(3) which are automated systems and can thus easily accommodate the required software
changes. Even minor modifications to a method can be costly when a large number of systems
that must all be upgraded exist in the field. For most methods, some changes are mandatory if a
high level of performance against small leaks is desired. The suggestions for performance
improvements made for each method and given in the appendices in Volume II are also included
in the appendix in Volume I. The current section describes some of the more common problems
in tank testing and suggests some simple ways to correct these problems.
12.1 Instrumentation
A well designed leak detection system should not be limited by its instrumentation. The
technology of product-level and temperature measurement systems is sufficiently developed that
the system noise can be designed to be less than the ambient or operational contributions to the
noise field. In methods of testing in which the tank is overfilled, the instrumentation is generally
more than adequate for the product-level and/or temperature measurements. However, for
methods that test in partially filled tanks, in which the requirements on the product-level sensor
system are 2 to 3 orders of magnitude more stringent, this is not always the case.
12.1.1 System Noise
The performance limit imposed by system noise should be estimated. An estimate of the
noise level of each of the sensor systems is determined by measurements made in the same
frequency band as the leak. Using system noise, one can determine the smallest leak rate that
can be detected with the required probabilities of detection and false alarm. This is done by
adding the variances of the equivalent volumes estimated from the product-level and
product-temperature sensor systems. For example, a system that uses a product-level sensor with
a standard deviation of 0.04 L/h (0.01 gal/h) and a temperature sensor suite with a standard
deviation of 0.06 L/h (0.015 gal/h) would have a combined standard deviation of 0.068 L/h
(0.018 gal/h) and would be able to detect a leak of 0.32 L/h (0.08 gal/h) with a PDof 0.99 and a
12-1
-------
PFAof 0.01 in the case of a test conducted at a constant hydrostatic pressure. Thus, even with no
sources of ambient or operational noise present, this system would not be able to detect leaks
smaller than 0.32 L/h with equivalent reliability. This analysis assumes that the noise is white.
12.1.2 Sensor Geometry
For detection of leaks in underground storage tanks, it is necessary to estimate volume
fluctuations due to thermal expansion or contraction of the product and to subtract them from the
product volume estimate. It is essential to have well calibrated temperature sensors with
sufficient spatial coverage to accurately estimate the mean rate of change of temperature. From a
practical standpoint, it is possible to adequately cover the vertical temperature field by means of
a temperature array containing point sensors. In general, the effects of horizontal variations were
found to be small in if product has not been added less than 6 h before the test. While a 1- to
3-sensor array may obtain a high level of performance under some product-temperature
conditions, the spatial coverage is not generally adequate, and large errors in estimating the
temperature field will sometimes be made. To provide adequate spatial coverage to support a
detectable leak rate of 0.38 L/h (0.10 gal/h) with a PD of 0.99 and a PFAof 0.01, and to ensure
coverage at all product levels, at least five equally spaced thermistors, or methods which achieve
a similar degree of averaging, are recommended.
12.1.3 Calibration
Since calibration is required in order to convert the output of the measurement system to
engineering units, it is essential that all of the instrumentation be calibrated periodically,
including off-the-shelf instrumentation which has already been factory- calibrated. The
sensitivity of performance to thermal compensation makes the calibration of temperature sensors
very important, particularly if a precision requirement of 0.001°C is to be met. While most
instruments have linear calibration curves, the thermistors that many methods have incorporated
into their systems do not. Thus, calibration is more difficult and requires enough data to develop
a calibration curve.
12.2 Protocol
The performance of many of the methods evaluated in this study can be improved by
modifying the test protocol. In general, protocol changes are the simplest type of changes to
make. It should be noted, however, that many of these improvements require an increase in the
total time required to complete a test, and will thus impact the amount of time that a retail station
must remain closed for testing.
12-2
-------
12.2.1 (jroundwater Considerations
Correct interpretation of a test result (i.e., the temperature-compensated flow rate) depends
on the level of the groundwater immediately outside the tank, because the groundwater level
affects the rate of flow through a hole in the tank. Thus, in combination with the product level,
the groundwater level affects the size of the signal to be detected, and, as a consequence, it is
difficult to interpret the test results. When there is any suspicion that the ground water may be
above the level of the tank, it is recommended that an estimate of the groundwater level be made
before conducting any tests. It is also recommended that a formal procedure be included as part
of all test methods to deal with groundwater if it is above the bottom of the tank. This
recommendation is made because of the possibility that a no-flow or a low-flow condition could
occur, in which case a leak might go undetected even by a reliable test method. Thus, when the
groundwater level is high, it is recommended that the test be conducted at two levels (of
product), or that the test be repeated at a later time, after the groundwater level has changed.
When this is not possible, for example, in a tidewater area, test results should be viewed with
caution.
Testing at two different product levels can be beneficial, but since any changes to the
method can affect the performance, this approach needs to be designed, integrated into the test
protocol, and evaluated properly. It is important to consider the effect on performance of the
change in test protocol that results from adding or removing product from the tank for the
purpose of changing the level, and of the new detection criterion required to interpret any
two-level tests. An arbitrary change in the field in response to a high groundwater level will
result in a test whose performance is unspecified.
The magnitude of the noise will also be affected if the groundwater level is actively
changing during the test, and it is recommended that a test not be conducted under such
conditions, which can alter the deformation of the tank and the rate of heat transfer between the
product the backfill. The effects of a changing groundwater level were not fully quantified
during the ambient noise experiments. One experiment that was performed indicated that large
changes in product temperature occurred as the groundwater level in the backfill was lowered.
12.2.2 Height-to-Volume Conversion
The majority of the volumetric test methods measure product level and then convert level
to volume using a height-to-volume conversion factor, A,.ff. Some manufacturers use a
theoretical estimate based on the geometry of the system, while others measure A^ as part of the
test protocol, using cylinder of known volume or a known amount of product to displace the
product level.
12-3
-------
In an overfilled-tank test, it is necessary to estimate Aetf in order to account for the effects
of trapped vapor and the instantaneous structural deformation of the tank. Errors of 50% or more
can result if A,.ff is assumed to be equal to the area, A, of the free product surface. The
calibration experiments necessary to estimate the height-to-volume conversion factor in an
overfilled tank are not time-consuming and are easy to perform. For nearly contant level tests,
the height-to-volume conversion factor is Ac((. For variable-level tests, K needs to be known,
because the height-to-volume conversion factor is Aeft + K.
In a partially-filled-tank test it is commonplace to use the tank chart to estimate A^,,.
Because, for a partially filled tank, Aeff is approximately equal to A, it is reasonable to estimate
Aeff theoretically. Two errors are possible, however, either of which may be large enough to
require experimental estimates. First, the tank chart may not constitute an accurate
representation of the tank, because measurements are usually based on the nominal dimensions
of the tank. Even if the tank chart has been developed from the measured dimensions of the
tank, it is still possible that the theoretical estunate of Aclf near the top of the tank will be in error
because of distortion or tilting of the tank that may have occurred during installation. Distortion
near the top of the tank can lead to differences between the theoretical estimates of A,.ff and the
measured ones. Second, even if the tank chart is perfect, an accurate estimate of the actual depth
of the product in the tank is a prerequisite for the proper use of the tank chart as a means of
estimating A^. Typically, the depth of the product can be measured to within 0.25 cm (0.1 in.)
with an instrument and to within 0.6 cm (0.25 in.) with a stick. In general, if reliable detection of
leak rates as small as 0.19 L/h is desired, this measurement is not accurate enough to estimate
A,.ffin the case of a test that is conducted when product level is in the upper 80 to 90% of the
tank. If the error is sufficiently large, it is recommended that Acff always be estimated
experimentally (by immersing a solid of known volume, measuring the height change, and
calculating Aef, using Eq. (4.1)). As a check on the measurements, the experimental value should
be compared to the theoretical one.
12.2.3 Product-Level Measurements During Overfilled-Tank Tests
Any volume changes that occur during a test will cause the tank to defonn. If the product
is maintained at a constant hydrostatic pressure during the test (this applies to both
partially-filled-tank tests and to overfilled-tank tests in which product is releveled), a waiting
period sufficient to allow the tank to defonn is required.
On the other hand, if the test is conducted with a variable hydrostatic pressure, waiting for
the structural deformation to subside is only a first step. Interpretation of the product-level
changes after the initial exponential deformation has occurred requires a detailed knowledge of
12-4
-------
the time and elasticity constants of the tank-backfill-soil system. Because these variables are not
known a priori, it is advisable to eliminate this method of collecting data. Accurate estimates of
the product level are possible, but only if the elasticity constant, K, approaches zero.
12.2.4 Data Collection
Data collection is discussed in terms of sampling interval and test duration.
12.2.4.1 Sampling Interval
In the majority of the methods, data are undersampled, causing a phenomenon known as
aliasing. This means that the high-frequency product-level fluctuations, which are not in the
same frequency band as the leak signal, are manifested as low-frequency fluctuations in the
frequency band of the signal. To avoid these effects, the data should be sampled at an interval
which is at least 0.5 and preferably 0.33 of the period of the fluctuations in the tank, and then
averaged. The significance of the aliasing depends on the sample interval and the magnitude and
frequency spectrum of the fluctuations. Aliasing should be avoided or shown to be negligible.
Because the highest-frequency fluctuations (those with the shortest period) in a partially
filled tank are produced by seiching of the product surface, the effects of aliasing can be easily
avoided. These seiches or standing surface waves may have a period of 2 to 10 s in a 30,000-L
(8,000 gal) tank, depending on the dimensions of the tank, the level of product in the tank, the
free product surface at that level, and the density of the product and the vapor above it. To avoid
aliasing the surface-wave data, the product level data should be sampled at time intervals of
approximately 1 s.
In a 30,000-L tank, internal waves produce temperature and sometimes product-level
fluctuations with periods between 3 and 60 min. Undersampling the thermal data equates to
incorrect estimation of the rate of change of temperature, and thus can lead to inaccurate
temperature compensation. In the 30,000-L Tank Apparatus tank, temperature fluctuations
produced by internal waves typically had periods between 5 and 20 min. Predicting these
periodic fluctuations is difficult. To avoid aliasing the internal wave data, a sample rate of 1 min
should be sufficient.
To avoid aliasing when both internal and surface waves are present, product level and
product temperature should be sampled at 1-s intervals and averaged to I min for analysis.
12-5
-------
Alaising is a complicated phenomenon, and most methods which collect data manually
will not be able to avoid this error. The error is generally small if the duration of the test is
several hours. Even with alaising, these methods should be able to reliably detectleaks of
0.38 L/h.
12.2.4.2 Test Duration
The duration of the test has a direct effect on performance. Analyses of data from previous
studies have shown that the longer the test duration, the higher the performance achieved
[7, 19, 20]. Longer tests tend to reduce the random fluctuations of the noise if the number and
location of the sensors are adequate.
An optimum test duration can be determined by collecting data over a 4- to 8-h period and
then investigating the effect on performance of varying the test duration. This can easily be done
during the manufacturer's field validation of the test method. For most of the methods evaluated
in this study, a 1-h test is generally too short to obtain accurate results.
Temperature-compensated volume fluctuations with periods between 30 and 60 min have been
observed. A 2-h test is required to properly average out these effects. A 1-h test would suffice if,
for example, the data were adequately sampled, the deformation effects had become negligible,
and the product temperature field had been adequately sampled. The advantage of a longer test
is that uncompensated noise can be averaged out.
12.2.5 Trapped Vapor
One of the sources of error in an overfilled-tank test is trapped vapor. Trapped vapor is a
problem because its presence is difficult to detect. The release of bubbles or a sudden drop in
product level are indicative that trapped vapor is present, but, in general, the volume of this
trapped vapor is not easily ascertained. Even if the volume of the trapped vapor were known, it
would be extremely difficult to compensate for the expansion and contraction of the vapor
without detailed measurements of the temperature and pressure of this vapor, and it would be
impossible to conduct a test if the volume of the vapor became too large. The best solution is to
eliminate trapped vapor from the tank. This is best done during the initial preparations for a test.
A method, described in Section 8, was developed to estimate the volume of the trapped
vapor. Unfortunately, because the volume changes caused by the instantaneous deformation of
the tank are inextricably linked to those caused by trapped vapor, only an upper bound estimate
of the amount of trapped vapor can be made. If Aeff is approximately equal to A, however, it can
be assumed that the volume of trapped vapor is small.
12-6
-------
It is recommended that this technique be used in all overfilled-tank tests in order to make
an estimate of the volume of trapped vapor. The procedure, which can be completed within
minutes, will not necessarily produce a good estimate of how much vapor is present, but it will
indicate whether trapped vapor is largely absent. If the results of the procedure show that the
volume of trapped vapor is large (e.g., 100 L), it should be assumed that trapped vapor is present,
and appropriate action should be taken to remove it before a test is conducted. In the Test
Apparatus tests, the volume of trapped vapor could not have exceeded 60 L, and was estimated
to be about 10 L (even after extensive efforts to reduce it).
Regardless of the level of experience of the test crew, it is likely that some vapor will be
trapped in the tank. As a consequence, the height-to-volume conversion factor must be measured
experimentally, because small amounts of trapped vapor affect the magnitude of this factor.
During the course of these experiments at the Test Appartus, no detailed study was made
of the product-level changes due to partial pressure changes in the tank before equilibrium has
been reached. (This phenomenon was observed in the preliminary experiments leading up to the
Test Apparatus experiments [25]). It is recommended that sufficient time be allowed after
overfilling the tank for the trapped vapor to approach equilibrium with the temperature and
pressure environment. In small containers (e.g., 1 L), this tends to occur over a period of
minutes; thus, the time allowed for the structural deformation effects to become small (typically
hours) should be more than sufficient to ensure that equilibrium has been reached.
12.2.6 Thermal Disturbance of the Vapor in a Partially Filled Tank
Any disturbance of the tank's thermal environment tends to degrade performance.
Experiments in partially filled tanks at the Test Apparatus suggested that, for at least 6 h after
opening the fill hole and preparing for a test, the temperature-compensated volume fluctuations
were too large to permit a high level of performance against small leaks. The residual volume
fluctuations evident in these experiments were probably caused by the escape of vapors when the
fill tube was opened, and by the evaporation of product into the vapor space after the test was
begun. For this reason, it is recommended that the test protocol specify that a waiting period be
observed after test preparations have been completed.
12.2.7 Product-Level Changes
Any product-level changes will affect the thermal environment of the product and induce
structural deformation of the tank.
12-7
-------
12.2.7.1 Product Delivery
The term "product delivery," as used in this report, refers to the addition of very large
amounts of product to a tank (usually the night before the test) to reach the approximate level
needed for the test. Large horizontal temperature gradients develop during and after delivery.
The data suggest that a minimum waiting period of up to 6 h is required before accurate thermal
compensation in the tank can be achieved. The effects of structural deformation take at least 6 h
to become small enough to be negligible.
Accurate tests of the tank cannot be reliably performed immediately following a delivery.
Theoretically, when a method calls for circulating product in the tank, a test can be started as
soon as the product is uniformly mixed. In reality, this may take several hours in a 30,000-L
tank. Any method for which the manufacturer claims that a test can be started immediately
following a delivery should be considered suspect.
12.2.7.2 Topping the Tank
Topping the tank, probably the largest source of error in overfilled-tank tests, is usually
done the day of the test to bring the product to the exact level needed for the test. Like a
delivery, it impairs accurate temperature compensation and induces structural deformation.
The effects of topping the tank, described in Section 8, are to disturb the local temperature
field and to cause horizontal temperature gradients to develop in the tank. Thus, accurate
temperature compensation is not possible for 2 to 3 h after topping, that is, until the disturbances
have subsided.
Another more subtle effect was identified in methods that overfill the tank. The
temperature of the product in the fill tube or above-ground standpipe is affected by the ambient
changes in air temperature. If the product sinks, it can disturb both the horizontal and vertical
distribution of temperature in the tank and the local temperature field around the sensors. The
effect is similar to but not as severe as the effect induced by topping.
Unfortunately, the product-level changes (and the changes in hydrostatic pressure) that
occur during topping are usually as large if not larger than those that occur during a delivery of
product. While most protocols include a long waiting period after a delivery, in many cases no
provisions are made for a waiting period after topping the tank. As a consequence, the effects of
structural deformation may be large enough to seriously degrade the performance of a test. The
length of the waiting period depends on the time constant and the magnitude of the deformation.
Waiting 3 time constants will reduce the fluctuation level by 99%; this may or may not,
depending on the magnitude of the deformation, be enough. It is hard to suggest a universally
12-8
-------
applicable waiting period because the deformation characteristics of individual tank-backfill-soil
systems are unique. The available data at the Test Apparatus suggest that the waiting period
should be a minimum of 6 h and can be as long as 24 h.
12.2.7.3 Releveling
In order to properly interpret volume changes, in the fill tube or standpipe of an overfilled
tank in terms of volume changes, the product must be releveled during the test. If the releveling
is not done carefully, the temperature field in the tank as a whole, as well as locally around the
temperature sensors, can be disturbed when product is added to the tank. These effects can be
minimized by releveling the product so that only very small amounts of product are added to the
tank at any time and a thermal adjustment can take place. It is desirable to add product that is at
approximately the same temperature as the product in the tank and to add it at a location that
does not affect the local temperature field around the sensors. This might be accomplished by
adding product through a tube that extends to the bottom of the tank.
12.2.8 Coefficient of Thermal Expansion
The accuracy of a temperature-compensation scheme depends in part on the accuracy of
the coefficient of thermal expansion used to convert temperature to volume. Most tests either
assume a coefficient for a given type of product (e.g., gasoline or diesel fuels), or measure the
specific gravity of the product before or after a test and calculate the coefficient using the API
tables. The tables are based on an average value of the coefficient generated from many
different products. The uncertainty for the latter method can be estimated from the uncertainty
of the coefficient in the tables for the measured specific gravity. In general, this uncertainty is
approximately 5% providing the uncertainty of the measured specific gravity is small. The
uncertainty for the former (the assumed value) also includes the change in the coefficient with
specific gravity, which could result in an uncertainty of as much as 10% or more. While this
error is very important for the detection of small leaks in large-capacity tanks, it was less
significant than many of the other errors commonly encountered in volumetric testing.
12.3 Data Analysis
The need for adequate temporal sampling and averaging has already been noted. After
data collection, the volume rate of change should be inferred from the temperature-compensated
product-level time series (or mass time series) by a technique, such as least-squares, which
estimates the linear trend in the time series.
12-9
-------
12.4 Detection Criterion
The performance of a leak detection system is determined by the detection criterion,
usually a threshold volume rate. Once the system performance has been determined for a
particular criterion, this criterion should be used exclusively. If another criterion is substituted,
the performance of the method is changed. Applying the given detection criterion leads to a
known performance for the method, presumably established with an acceptable error rate in
terms of the probability of missed detections and false alarms. If the error rate is unacceptable,
the detection criterion should be changed, and the performance of the method should be
determined again in light of the change(s).
The threshold is determined by the PFA. To satisfy the PD= 0.95 and PFA= 0.05
requirements, a detection threshold of one half the desired detectable leak rate should be
established; i.e., if the desired detectable leak rate equals 0.38 L/h (0.1 gal/h), the threshold for
declaring a leak should be set to 0.19 L/h (0.05 gal/h). If a method claims a high performance
against a leak rate, the threshold used to detect leaks should always be smaller than this leak rate
by a factor of 2 or more. A high performance claim also implies that the precision of the leak
detection system in the tank environment should be no larger than approximately 0.2 times the
desired detectable leak rate.
12.5 Operator Influence
Because arbitrary changes to the protocol will alter performance, any method in which the
operator is required to make judgments during the test is not recommended. This is particularly
true if the operator adjusts the product level during the test or changes the sampling interval or
the duration of the test by making arbitrary judgments about when to start and stop based on the
specified detection criterion. Once the tank has been prepared for testing and the equipment has
been properly set up, the test should be conducted exactly as specified. Changes should not be
made during the test.
Special attention should be paid whenever any product-level adjustments are made. Since
the magnitude of the volume changes due to deformation depend on the time history of the
product level before the adjustments were made and on the details of how the product level is
adjusted, it is recommended that all product-level adjustments be accomplished using a set,
repeatable procedure.
A common field practice, not observed in the Field Verification Tests because of the
requirement to follow the test protocol exactly, is to conduct more than one test consecutively if
the measured volume rate is near, but exceeds, the threshold, or to arbitrarily declare the tank
12-10
-------
tight even though the threshold for flow into the tank is exceeded. Both pactices will lower
probability of detection and may lead to a missed detection. The impact of the former depends
on how the additional tests are interpreted with regard to the first and on whether the second test
is independent or correlated with the first. If a multiple-testing procedure is desired, it should be
incorporated into the test protocol and should be evaluated. In general, the number of tests
should be fixed.
12.6 Multiple-Test Strategy
Some test methods employ a multiple-test strategy. Sometimes the multiple-test strategy is
incorporated in the data collection and analysis, and sometimes it is incorporated into the
detection criterion. As commonly practiced, the strategy is arbitrary and therefore has
deleterious consequences on system perfonnance. Also, the tests are not independent, and the
expected improvement in PD, PFA, or both is not achieved because of the correlation between tests
(e.g., systematic errors that remain essentially the same between tests). With a properly designed
multiple-test strategy, however, improved leak detection perfonnance can be achieved. As
measured by leak rate, the system performance will improve for a constant PD and PFA.
To illustrate, let us propose to conduct five statistically independent tests of a tank. Let the
tank be declared leaking if the detection threshold was exceeded in at least three of the five tests
and declared tight if the threshold was exceeded in two or fewer tests. For this m-out-of-n
strategy (where m = 3 and n = 5), the detectable leak rate is almost halved vis-a-vis the detectable
leak rate of a single test, while the detection and false alarm probabilities remain at
approximately 0.99 and 0.01, respectively. Such a result can be achieved only if the five tests
are independent.
The multiple-test strategies utilized by the commercial methods can be improved. In
general, the strategies are not properly designed, and therefore, the PD and PFAcannot be
interpreted. Most methods use an m-out-of-n approach that reduces the PFA; the PD, however, is
also reduced, sometimes in a substantial way.
Examples of strategies being used appear below.
o The minimum volume rate out of two or more tests is used to determine whether
the tank is leaking.
o After the volume-rate threshold has been exceeded, an arbitrary number of tests is
conducted, until such time as the measured volume rate fails to exceed the
threshold.
12-11
-------
These testing strategies are typically applied at the end of a test; they may be an inherent
part of the test, or they may be used only because the test indicates the possibility of a leak.
Most approaches are based on the assumption that the product-level changes in a nonleaking tank
will become smaller with time. If this assumption is true, selecting the volume rate from the
smallest or last test to determine whether the tank is leaking will lead to more accurate test
results. If the noise has already decayed to a value small enough to be negligible, as required by
most test designs, this approach will result in a substantial reduction in the PFA but will also result
in a PD below that which would be obtained with a single test. If the number of tests that may be
performed is large and/or not fixed, the PD will approach 0.
It is difficult to interpret the impact of the multiple-test strategy on performance unless the
tests can be shown to be independent (or the correlation between them is known) and unless the
number of tests is fixed. The degree of performance achieved will depend strongly on the
independence assumption. Most multiple-test strategies are implemented at the end of the first
test. The second test is not apt to be independent of the first because of systematic errrors that
are present during the first test and that are still present in the second test. The second test will
not show a performance improvement if the first and second test both contain a vapor pocket
large enough to impact performance. Neither will performance improve if uncompensated noise,
such as evaporation and condensation, structural defonriation, or thermal expansion and
contraction of the product, is still present during the second test. It is easy to show statistically
the dramatic improvement gained by using a multiple-test strategy; however, it is difficult to
achieve this performance unless that test strategy is carefully designed and evaluated.
Achieving independence is difficult. One method of improving the chances of an
independent test is to completely remove the equipment from the tank and perform another test.
This may not always be sufficient if the same source of error persists for each test. For example,
when vapor is trapped in a tilted tank during an overfilled-tank test, a nearly identical amount of
vapor may be trapped during each succeeding test. For methods that overfill the tank, improving
the chances of an independent test includes dropping the product level below the top of the tank
before conducting another test. Chances improve if the second test is conducted on another day.
Finally, if the first test is performed in an overfilled tank, the chances for independence improve
if the second test is conducted when product level is near, but below, the top of the tank. Such
an approach is impractical for many methods because of insufficient precision in the
product-level instrumentation. In the latter approach there is the additional problem that the leak
may not be located below the product level but in that section of tank or piping above the tank
top. Finally, the impact on performance of any changes of product level prior to the conduct of
additional tests must be carefully evaluated.
12-12
-------
References
1. Niaki, S., and J.A. Broscious. "Underground Tank Leak Detection Methods: A
State-of-the-Art Review." EPA/600/2-86/001. IT Corporation, Pittsburg, Pennsylvania.
Prepared for the Hazardous Waste Engineering Research Laboratory, U.S. EPA,
Washington, D.C. (January 1986).
2. Westat, Inc., Midwest Research Institute, Batrelle Columbus, Inc., and Washington
Consulting Group. "Underground Motor Fuel Storage Tanks: A National Survey."
Prepared for the Office of Pesticides and Toxic Substances, U.S. EPA, Washington, D.C.
(May 1986).
3. U.S. Environmental Protection Agency. "Underground Storage Tanks; Proposed Rules."
Federal Register, Vol. 52, No. 74 (17 April 1987).
4. U.S. Environmental Protection Agency. "Underground Storage Tanks; Proposed Rules."
Federal Register, Vol. 53, No. 185 (23 September 1988).
5. National Fire Protection Association. "Underground Leakage of Flammable and
Combustible Liquids." NFPA Pamphlet 329, National Fire Protection Association,
Quincy, Massachusetts (1987).
6. Wilcox, H.K., J.D. Flora, C.L. Haile, M.J. Gabriel, and J.W. Maresca, Jr. "Development
of a Tank Test Method for a National Survey of Underground Storage Tanks."
Final Report, Midwest Research Institute Project 8501-A(25), Midwest Research
Institute, Kansas City, Missouri. Prepared for the Office of Pesticides and Toxic
Substances, U.S. EPA, Washington, D.C. (June 1986).
7. Maresca, J.W., Jr. "Analysis of the Pilot Study Tank Test Data." Final Report,
Vista Research Project 2012, Vista Research, Inc., Palo Alto, California. Prepared for the
Office of Pesticides and Toxic Substances, U.S. EPA, Washington, D.C. under
subcontract to Midwest Research Institute, Kansas City, Missouri (July 1985).
8. Maresca, J.W., Jr., C.P. Wilson, and N.L. Chang, Jr. "Detection Performance and
Detection Criteria Analysis of the Tank Test Data Collected on the U.S. Environmental
Protection Agency National Survey of Underground Storage Tanks." Final Report,
Vista Research Project 2013, Vista Research, Inc., Palo Alto, California. Prepared for the
Office of Pesticides and Toxic Substances, U.S. EPA, Washington, D.C. under
subcontract to Midwest Research Institute, Kansas City, Missouri (September 1985).
9. Wilson, C.P., J.W. Maresca, Jr., H. Guthart, J.A. Broscious, S. Niaki, and D.E. Splitstone.
"A Program Plan to Evaluate Underground Storage Tank Test Methods." Program Plan,
13-1
-------
Vista Research Project 2011, Vista Research, Inc., Palo Alto, California. Prepared for the
Hazardous Waste Engineering Research Laboratory, U. S. Environmental Protection
Agency under subcontract to IT Corporation, Pittsburg, Pennsylvania (March 1985).
10. Starr, J.W., J.A. Broscious, S. Niaki, J.S. Farlow, and R. Field. "An Approach to
Evaluating Leak Detection Methods in Underground Storage Tanks. Proceedings of the
1986 Hazardous Material Spills Conference, U.S. Environmental Protection Agency,
St. Louis, Missouri (1986).
11. Starr, J.W., and J.W. Maresca, Jr. "Protocol for Evaluating Volumetric Leak Detection
Methods for Underground Storage Tanks." Technical Report, Contract No. 68-03-3244,
Enviresponse, Inc., Livingston, New Jersey, and Vista Research, Inc., Palo Alto,
California. Prepared for the Hazardous Waste Engineering Research Laboratory, U. S.
Environmental Protection Agency (June 1986).
12. Maresca, J.W., Jr., and M. Seibel. "Volumetric Tank Testing." Final Report, Vista
Research Project No. 2028, Mountain View, California. Prepared for the Center of
Environmental Research Information, U. S. Environmental Protection Agency under
subcontract to JACA Corporation, Fort Washington, Pennsylvania (in press).
13. Maresca, J.W., Jr., J.W. Starr, R.D. Roach, and J.S. Farlow. "Evaluation of the Accuracy
of Volumetric Leak Detection Methods for Underground Storage Tanks Containing
Gasoline." To be published in the Proceedings of the 1989 Oil Spill Conference
sponsored by the U. S. Coast Guard, American Petroleum Institute, and the U. S.
Environmental Protection Agency, San Antonio, Texas (in press).
14. Metric Practice Guide, ASTM E. American Society of Testing Materials, 1916 Race
Street, Philadelphia, Pennsylvania 19103.
15. U.S. National Bureau of Standards. "Volume Correction Factors." In Manual of
Petroleum Measurement Standards. American Petroleum Institute, Washington, D.C.
(August 1980).
16. Urick, R.J. Principles of'Underwater Sound. 3rd ed. New York: Me Graw Hill (1983).
17. Skolnik, M. I. Introduction to Radar Systems. 2nd ed. New York: McGraw-Hill (1980).
18. Burdick, W.S. Underwater Acoustic System Analvsis. Englewood Cliffs: Prentice Hall
(1984).
19. Maresca, J.W., Jr., P.C. Evans, R.A. Padden, and R.E. Wanner. "Measurement of Small
Leaks in Underground Gasoline Storage Tanks Using Laser Interferometry." Final
Report, American Petroleum Institute, SRI Project 7637, SRI International, Menlo Park,
California (September 1981).
13-2
-------
20. Maresca, J.W., Jr. "A Method of Determining the Accuracy of Underground Gasoline
Storage Tank Leak Detection Devices." Proceedings of the Underground Tank Testing
Symposium, Toronto, Ontario, May 25-26, 1982, Petroleum Association for Conservation
of the Canadian Environment (May 1982).
21. Maresca, J.W., Jr., N.L. Chang, Jr. and P.J. Gleckler. "A Leak Detection Performance
Evaluation of Automatic Tank Gauging Systems and Product Line Leak Detectors at
Retail Stations." Final Report, American Petroleum Institute, Vista Research Project
2022, Vista Research, Inc., Mountain View, California (January 1988).
22. Weast, R.C.,ed. CRC Handbook of Chemistry and Physics. 57th ed. Cleveland:
CRC Press, Inc. (1976-1977).
23. Roach, R.D., J.W. Starr, C.P. Wilson, D. Naar, J.W. Maresca, Jr., and J.S. Farlow.
"Discovery of a New Source of Error in Tightness Tests on an Overfilled Tank."
Proceedings of the Fourteenth Annual Resarch Symposium, Hazardous Waste
Engineering Laboratory, Office of Research and Development, U.S. Environmental
Protection Agency, Cincinnati, Ohio (in press).
24. Vista Research, Inc. "Quality Assurance Project Plan: Evaluate Leak Detection Methods
for Petroleum USTs." Document Control No. VRI-87-01, Rev. 0
(EI-03-70-0307000, Rev. 2), Vista Research, Inc., Palo Alto, California. Prepared for the
Hazardous Waste Engineering Research Laboratory, U. S. Environmental Protection
Agency under subcontract to CDM Federal Programs Corporation, Virginia
(3 April 1987).
25. Maresca, J.W., Jr., C.P. Wilson, N.L. Chang, Jr., and H. Guthart. "Preliminary
Experiments on the Ambient Noise Sources in Underground Tank Testing."
Technical Report, Vista Research Project 1006, Vista Research, Inc., Palo Alto,
California. Prepared for the Hazardous Waste Engineering Research Laboratory, U. S.
Environmental Protection Agency under subcontract to Enviresponse, Inc., Edison, New
Jersey (May 1986).
26. Maresca, J.W., Jr., N.L. Chang, Jr., and M. Seibel. "Operating Manual for LIDAS."
Technical Paper, Vista Research Project 1006, Vista Research, Inc., Palo Alto, California.
Prepared for the Hazardous Waste Engineering Research Laboratory, U. S.
Environmental Protection Agency under subcontract to Enviresponse, Inc., Edison, New
Jersey (June 1986).
27. Roach, R.D., J.W. Maresca, Jr., and M. Seibel. "Operating Manual for TPDAS."
Technical Paper, Vista Research Project 1006, Vista Research, Inc., Palo Alto, California.
Prepared for the Hazardous Waste Engineering Research Laboratory, U. S.
Environmental Protection Agency under subcontract to Enviresponse, Inc., Edison, New
Jersey (June 1986).
13-3
-------
28. Maresca, J. W. Jr., R. D. Roach, J. W. Starr, and J. S. Farlow. "U.S. EPA Evaluation of
Volumetric UST Leak Detection Methods." Proceedings of the Thirteenth Annual
Research Symposium, Hazardous Waste Engineering Laboratory, Office of Research and
Development, U.S. Environmental Protection Agency, Cincinnati, Ohio (September
1987).
29. Maresca, J.W. Jr., and J.W. Starr. "Evaluation of Volumetric Leak Detection Methods
for Petroleum Underground Stoage Tanks." Internal EPA Report, EPA Contract No.
68-03-3255. Vista Research Project 1019, Vista Research, Inc., Palo Alto, California.
Prepared for the Hazardous Waste Engineering Research Laboratory, U. S.
Environmental Protection Agency under subcontract to Enviresponse, Inc., Edison, New
Jersey (January 1987).
30. Maresca, J.W., Jr., and P.C. Evans. "Measurement of Leaks in Underground Gasoline
Storage Tanks Using Laser Interferometry." Technical Report 1, American Petroleum
Institute, SRI Project 7637, SRI International, Menlo Park, California (June 1979).
13-4
-------
Appendix
Descriptions of the 25 Test Methods Evaluated in the EPA Study
and Suggestions for Improving the Performance of Each
(Source: Sections 2 and 5.2 of Each Appendix in Volume II)
A-l
-------
-------
A. 1.6 Mathematical Model
While the description provided above is sufficient to identify the key features of the
AES/Brockman Leak Detecting System as it was evaluated and to interpret the results of the
performance evaluation, it should be noted that a mathematical model of the system was
developed based on the more detailed description of the test method provided in [A-l].
During the validation of the model, it was found that not all of the key features of the
AES/Brockman system were in accord with the definitions in [A-l]. A discussion of the
discrepancies is presented in Section A.4.
A.2 Improving System Performance
The AES/Brockman System as defined in [A-l] can realize performance improvements.
The recommended changes include additional instrumentation, adjustments to the protocol, and a
new leak-rate calculation algorithm.
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual performance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. The 0-h waiting period
after topping is too short to minimize the impact of the (1) spatial inhomogeneities and (2)
structural deformation produced by the addition (or removal) of product to reach the level
required to begin the test. A waiting period of at least 2 to 3 h is required to minimize the spatial
inhomogeneities in temperature. This period is long enough to minimize the effects of a 7°C
difference in temperature between the product in the tank and the product used to top the tank. A
longer period may be required to minimize the deformation-induced product-level changes. This
can present operational problems, because the time required to minimze the effects of
deformation in some tanks may be 1 to 5 times (or more) longer than the time required to wait
for the temperature effects to stabilize. The time constants of the Test Apparatus tanks, for
example, are 2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or longer;
the waiting period would be longer if the deformation effects were large and more than three
time contants were required for them to become small compared to the smallest leak to be
detected. Since the temperature and deformation effects associated with topping are independent
of each other, the protocol and/or analysis algorithm use to minimize this effect must address
each disturbance. A single waiting period based on a maximum disturbance would adequately
address the temperature instability. However, this may not suffice for the deformation effect.
A.2.1 Instrumentation
The largest deficiency in the instrumentation is inadequate temperature compensation. A
comparison of temperature volumes from the volumetrically weighted array with the
A-5
-------
manufacturer's estimate (see Figure B.3.4) shows that only in one case was there agreement
between the two; i.e., the AES/Brockman estimate of temperature effects can contain large
errors. This is caused by undersampling of the vertical temperature field, since the manufacturer
measures the temperature at the base of the tank only. The manufacturer has observed that near
the base of the tank the temperature changes at a greater rate than in the rest of the tank and,
because of this, has incorporated a factor that serves to reduce the measured temperature volume.
Although the present evaluation has not rigorously attempted to determine whether this
temperature volume estimate is correct, a better estimate of the temperature effects might be
obtained if the manufacturer included an array of thermistors. Since the AES/Brockman system
is computer-based, incorporating the measurements from additional thermistors into the data
analysis could be done with relative ease.
A.2.2 Protocol
The following protocol changes are recommended:
o Wait 6 to 12 h or longer after product delivery
o Relevel the product at 15-min intervals or less and record the volume changes
o Wait at least 3 to 6 h or longer after topping off the tank, and maintain constant
product level during the waiting period
The protocol employed by the AES/Brockman system involves large product-level
adjustments each hour during a 3-h test. These adjustments are detrimental in two ways: (1)
they cause measurable structural deformation effects, and (2) the added product disturbs the
temperature field. It is recommended that the manufacturer consider raising the product once
and performing the entire test at that level. Because the AES/Brockman system is
computer-based, consecutive 1-h tests could be run to determine when structural deformation
effects have subsided. Furthermore, if the manufacturer were to relevel product during the tests,
underestimating the volume change due to the coupling of volumetric noise sources with
structural deformation could be avoided. Both the coupling effects and the constant-level
approach are discussed in detail in the body of the report.
A.2.3 Analysis
It is recommended that the manufacturer reexamine, as discussed above, the attempt to
estimate the total temperature effect and that he remove TCC from the algorithm. This would
require the incorporation of additional thermistors, since it is not evident that one thermistor
located at the base of the tank is adequate to represent temperature.
A-6
-------
The algorithm employed to evaluate the effects of structural deformation is not accurate
and can cause gross errors in the measured leak rate results. The manufacturer would obtain
better performance if he were to establish a method of determining, as a prerequisite for initiating
a test, when structural deformation has become negligible. Performance would improve if the
manufacturer removed the structural deformation estimate from the algorithm and calculated a
thermally compensated leak rate. The criterion for determining whether a tank is leaking causes
missed detections of leaking tanks, but at the same time, it greatly reduces the probability of false
alarm. The implementation of this criterion therefore reduces the performance of the system. It
is recommended that the criterion be removed from the analysis that determines whether a tank is
leaking.
A.2.4 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-7
-------
-------
B.I Ainlay Tank 'Tegrity Test Method Description
The Ainlay Tank Tegrity Tester is a volumetric tank tightness test which is conducted in
an overfilled tank. The product level is allowed to vary depending upon environmental changes
occurring in the tank, and no effort is made to maintain a constant level during the test, for which
a threshold of 0.189 L/h (0.05 gal/h) is used. A tank is declared leaking if the themially
compensated flow rate is greater than this threshold. If the flow rate is less than the threshold,
the tank is declared nonleaking.
B.I.I Instrumentation
The Ainlay Tank 'Tegrity Tester uses two measuring devices to estimate a
temperature-compensated flow rate: product level as determined by means of a slope tube
indicator (manometer), and product temperature as determined by three temperature probes. The
testing period of 1 h applies only to situations in which the capacity of the tank being tested is
less than 37,800 L (10,000 gal). For tanks with a greater capacity, the increase in test duration is
15 min for every additional 9,450 L (2,500 gal). Test duration and sampling intervals are
constant. The test is carried out in a storage tank that has been overfilled to a level within the fill
tube.
Volumetric changes are determined by monitoring pressure changes in a tube inserted
below product level in the tank's fill tube. Product height changes in the fill tube alter the
amount of pressure required to force propane gas bubbles through the inserted tube, and these
pressure changes are reflected as fluctuations in the height level of the water column in the slope
tube indicator. Refer to the mathematical modeling report [B-l] for a schematic diagram of the
Ainlay Tank Tegrity Tester.
Horizontal movement in the slope tube indicator corresponds to the changes in product
level within the fill tube. Although a change in product level in the fill tube is barely noticeable,
it causes exaggerated movement in the slope tube indicator, where it can be easily detected and
measured.
The slope tube indicator consists of approximately 150 cm of clear plastic tubing coiled
into a helix approximately 8 cm in diameter and 5 cm high. A 5-cm change in product level
therefore produces a liquid movement of approximately 150 cm in the slope tube indicator.
Since volume change is shown directly on the slope tube indicator, product level is not
used to determine volume change, and the device contains no scale. At the conclusion of the
test, product is added to (or removed from) the tank by means of a graduated cylinder in order to
return the slope tube indicator level to its starting point. The amount added (or removed)
represents the gross volume change during the test period.
A-9
-------
Temperature is measured with three thermistors (probes) placed at the vertical centroids of
three volume slabs so as to have a weighting factor of 25% at the top, 50% in the middle, and
25% at the bottom. The input signal voltage from a thermistor is compared to the thermistor
response data supplied by the thermistor manufacturer, and is encoded in a data logger.
Calibration of the temperature sensors takes place before the first use of the instrument and is
recommended at intervals of not more than one year. Precision and resolution of the sensors are
those reported by the manufacturer. They were experimentally conftnned during the Field
Verification Tests at Edison, New Jersey.
The volume change is determined by adding to or removing from the fill tube an amount of
product sufficient to return the slope tube level to its starting point. This is done at the end of the
test. The product is added by means of a graduated cylinder whose precision and resolution, as
claimed by the manufacturer, are presented in Table B.I. This cylinder, with a capacity of 0.45
L, has numbered divisions every 0.038 L and five subdivisions between each numbered mark.
The amount of product added or removed represents the gross change in volume during the test
period.
Table B.I. Precision and Resolution of the Product Level and Temperature Measurement
Systems
Product Level Measurement
Temperature Measurement
Precision
0.0038 L
o.oorc
Resolution
0.0076 L
o.oorc
Three 6,000-Q thermistors are used to measure temperature changes. The nominal
temperature-resistance curve supplied by the manufacturer is used to convert from resistance to
temperature. The resolution varies with temperature because the number of ohms per degree C is
not constant. The resolution is less than 0.001°C for temperatures greater than 13.6°C. The
temperature required for the calculations is represented by the difference between the
temperatures at the beginning and end of the test. The thermistors have an operating range of-17
to 50°C.
B.I.2 Equipment Setup and Test Preparations
The manufacturer indicated that approximately 1 h would be required to set up the test
equipment and prepare the tank for testing. The manufacturer requires that the tank be filled to a
level within the fill tube the night before the test, but states that the test crew need not be present
during the delivery of the product. Unless the crew is on site and personally supervises the
overfilling of the tank, this requirement is unlikely to be met in actual tests. However, the crew
A-IO
-------
must monitor product level to ensure that the product does not drop out of the fill tube and into
the tank because of structural deformation and thermal contraction of the product, or overflow
out of the top of the tank because of thermal expansion. If, in the absence of monitoring by the
crew, such events do ensue, 38 to 380 L (10 to 100 gal) of product may need to be added to the
tank immediately before the start of the test just to raise the product to the prescribed level within
the fill tube.
B.I.3 Test Protocol
The test protocol requires at least an 8-h wait between the overfilling of the tank and the
start of a test. If final topping off occurs immediately prior to the start of a test, an additional 2-h
waiting period is required.
The test duration is 1 h in length, with temperature values being recorded at the beginning
and end of the test period. For each of the three thermistors, data are recorded at the start and
end of the test. In order to gauge temperature fluctuations, the operator also observes the
temperature readings every 5 min on the field recorder during the test period. Based upon an
undefined procedure, the operator utilizes this information to decide whether to continue the test.
The basic parameters of the test protocol are summarized in Table B.2.
Table B.2. Important Parameters of the Ainlay Tank Tegrity Tester
Sample Period 60 min
Test Duration 60 min
Number of Samples per Sensor per Test 2
Test Start After Filling Tank 8 h
Test Start After Topping Off Tank 2 h
Test Type Overfilled Tank
If, due to the height of the water table, the tank is sitting in water, compensation is made
for the external pressure exerted by the water by fitting an extension to the fill tube and raising
the level of product in the fill tube. This procedure is intended to neutralize the pressure effects
of tank immersion in groundwater, and in effect restores dry tank conditions.
No compensation for evaporation is made, since experiments conducted by Ainlay showed
an evaporation rate of less than 0.004 L/h in a fill tube at 23°C.
A-n
-------
B.I.4 Data Analysis
The Ainlay Tank 'Tegrity Tester calculates a temperature-compensated volume rate for the
single-point product-level and temperature measurements obtained at the beginning and end of
the specified test period. Product-level measurements are converted to volume by adding
sufficient liquid to the fill pipe to bring the slope tube manometer reading back to its starting
point. The amount of liquid added is taken to be the volume change which occurred during the
test.
Thermal expansion and contraction of the product is calculated from the volumetrically
weighted thermistors placed in the tank. The volume change, ATV, is estimated using
ATV = CeVAT (B.2.1)
where the change in temperature, AT, is the volumetrically weighted value determined from the
three thermistors, the total volume, V, is the tank volume as determined from the tank chart
issued by the tank's manufacturer, and Cc is the coefficient of thermal expansion of the product.
The value of Ce is determined immediately prior to a test by taking a sample of product and
measuring its gravity with a hydrometer. The temperature of this sample is also measured, and
the API gravity at 15.5°C calculated. The resulting value is then used, in conjunction with the
API Tables, to determine the value of the coefficient.
Although the standard test duration is 60 min, the operator may terminate a test at any time
if an irregular temperature profile is observed on the recorder. The criteria for establishing a
definition of irregular temperatures are not available.
B.I.5 Detection Criterion
A leak is declared if the temperature-compensated volume rate exceeds 0.189 L/h
(0.05 gal/h).
B.I.6 Mathematical Model
While the description provided above is sufficient to identify the key features of the Ainlay
Tank Tegrity Tester as it was evaluated and to interpret the results of the performance
evaluation, it should be noted that the mathematical model of the system was developed based on
the more detailed description of the test method provided in [B-l], After it had been validated,
this model was used to estimate the performance of the method.
A-12
-------
B.2 Improving System Performance
The Ainlay Tank 'Tegrity Tester as defined in [B-l] can realize substantial performance
improvements without hardware modification. The system could be further improved by
additional instrumentation.
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual perfonnance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. The 2-h waiting period
after topping may be too short to minimize the impact of the (1) spatial inhomogeneities and (2)
structural deformation produced by the addition (or removal) of product to reach the level
required to begin the test. A waiting period of at least 2 to 3 h is required to minimize the spatial
inhomogeneities in temperature. This period is long enough to minimize the effects of a 7°C
difference in temperature between the product in the tank and the product used to top the tank.
Thus, for many tests this effect is minimized by the 2-h waiting period required by the method.
A longer period may be required to minimize the deformation-induced product-level changes.
This can present operational problems, because the time required to minimze the effects of
deformation in some tanks may be 1 to 5 times (or more) longer than the time required to wait
for the temperature effects to stabilize. The time constants of the Test Apparatus tanks, for
example, are 2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or
longer;The time constants of the Test Apparatus tanks, for example, are 2 to 3 h, which means
that the minimum waiting period would be 6 to 9 h or longer; the waiting period would be longer
if the deformation effects were large and more than three time contants were required for them to
become small compared to the smallest leak to be detected. Since the temperature and
deformation effects associated with topping are independent of each other, the protocol and/or
analysis algorithm use to minimize this effect must address each disturbance. A single waiting
period based on a maximum disturbance would adequately address the temperature instability.
However, this may not suffice for the deformation effect.
B.2.1 Protocol
The following protocol changes are recommended:
o Wait 6 to 12 h or longer after product delivery
o Sample the temperature data every 3 min
o Relevel the slope-tube manometer at 15-min intervals or less and record the
volume changes
o Wait 3 to 6 h or longer after topping off the tank, and maintain constant product
level during the waiting period
A-13
-------
Releveling the product represents the most significant performance improvement that can
be achieved by the Ainlay Tank Tegrity Tester. Releveling removes the deleterious effect of
dynamic structural deformation described in Section 6 of the report. Eliminating the dynamic
structural deformation would improve the performance of the system by approximately 100%.
B.2.2 Analysis
It is recommended that the temperature-compensated volume rate be computed from the
data taken at the recommended higher sample rates using least-squares techniques. If four
volume measurements were used to estimate the volume-rate change and 20 temperature
measurements to estimate the thermal volume change, the performance could improve by as
much as 100%.
B.2.3 Instrumentation
Performance improvement could be achieved by adding more thermistors to improve
spatial sampling of the vertical temperature field. Based upon studies of canonical methods
presented in Section 10 of the report, it is estimated that a 30% performance improvement could
be achieved by increasing the number of thermistors from three to five.
B.2.4 Detection Criterion
The probability of false alann is determined by the threshold. The manufacturer should
determine what probability of false alann is desired and adjust the threshold accordingly.
A-14
-------
C.I The AUTAMAT Test Method Description
The Automatic Tank Monitor and Tester (AUTAMAT) is a volumetric leak detection
system
C.I.I Instrumentation
The AUTAMAT system is designed so that the product level measured in a reference tube
can be used as the thermal compensation for product-level changes measured in the tank. Both
the product level in the reference tube and that in the tank are measured using absolute pressure
sensors located near the base of the tank and offset vertically from one another by approximately
40 cm; a third absolute pressure sensor is used to measure and compensate for ambient pressure
changes. The precision and resolution of the AUTAMAT system as given by the manufacturer
are presented in Table C.I.
The reference tube, constructed of copper, has a diameter slightly smaller than the fill rube
and is slightly shorter than the diameter of the tank. Another tube, smaller in diameter and
approximately 1 m in length, is attached to the top of the copper reference tube. Inside the
reference tube are four pieces of "deadwood" (material whose purpose is to displace volume)
symmetrically located about the center of the tank in order to approximate the changing chord
length across the tank's diameter.
A pressure sensor at the base of the tube monitors product level within the reference tube;
another near the base of the tank monitors product level within the tank. A data set consisting of
10 samples gathered at a rate of 3 samples per second is collected every 2 min. The data are
averaged and then stored for subsequent analysis. A graph of the product level in the fill tube
versus that in the reference tube is displayed and updated during the test. This graph is used to
determine whether the reference tube has come to equilibrium with the tank and whether
structural deformation effects have ended.
Table C.I. Precision and Resolution of AUTAMAT Measurements as
Specified by the Manufacturer
Precision Resolution
Product Level 0.05 cm 0.01 cm
C.1.2 Equipment Setup and Test Preparations
The setup of the AUTAMAT system and preparation for a test at the Edison facility
typically took 1.5 h. This assumes that the product was at a satisfactory level prior to
installation.
A-15
-------
C.I.3 Test Protocol
Test protocol requires a wait of 12 h between filling the tank and starting the test. If more
product must be added prior to the test, an additional waiting period of 1 h is necessary so as to
allow structural deformation effects to subside. Additionally, a waiting period of 30 min is
required, after the instrumentation is hooked up and started, for the temperature sensor to
stabilize; the temperature sensor is used to compensate the pressure sensor electronics, and the
30-min waiting period may overlap with the other waiting periods. The 30-min waiting period
may be lengthened if deemed necessary by the operator.
A complete data set consists of 61 data points collected at a rate of one every 2 min. A
data point, which consists of 10 samples gathered at a rate of 3 samples per second, is collected
every 2 min during the 2-h period.
The important features of the AUTAMAT system are summarized in Table C.2.
Table C.2. Important Aspects of the AUTAMAT Test Protocol
Sampling Period 2 min (average of 10 samples collected at 3 Hz)
Test Duration 2 h
Waiting Period after Product Addition 12 h
Wait after Sensor Installation 30 min
Test Type/Product Level Overfilled Tank
C.I.4 Data Analysis
The concept underlying the AUTAMAT system is that thermal changes in the tank can be
represented using a reference tube located in the tank. In order to best approximate the relative
effects of temperature changes at different levels in the tank, the volume of the reference tube is
varied as a function of depth so as to represent the varying chord length of the tank as a function
of depth. With this approximation, the height changes in the reference tube are equated to
thermal fluctuations in the tank, and these thermal fluctuations can be compensated.
C.I.5 Detection Criterion
A tank is declared leaking if the leak rate as determined by the AUTAMAT system is
greater than a threshold of+/- 0.189 L/h (0.05 gal/h). If the leak rate is less than the threshold,
the tank is declared nonleaking.
A-16
-------
C.I.6 Miilhcmalical Model
While the description provided above is sufficient to identify the key features of the
AUTAMAT system as it was evaluated and to interpret the results of the performance
evaluation, it should be noted that a mathematical model of the system was developed based on
the more detailed description of the test method provided in [C-l].
C.2 Improving System Performance
The AUTAMAT system can realize improvements by changing the data analysis. In
addition, the system employed in the field can be improved by correcting the errors that affect
temperature compensation.
C.2.1 Protocol
The protocol as defined in [C-l] does not give the operator a systematic method of
determining whether structural deformation is occurring. The AUTAMAT method, which is
computer based, could include specific criteria for making such a determination. It is
recommended that a specific criterion be employed in order to standardize implementation of the
method in the field. The criteria should include sufficient time for the probe to stabilize, for the
temperature field to stabilize, and for structural deformation effects to subside.
C.2.2 Instrumentation
To determine why the measured temperature was not representative of the tank, it would
be necessary to fully evaluate and analyze the instrumentation. It is recommended that the
manufacturer determine exactly what the temperature probe is measuring. An empirical study
could also help determine (1) what waiting period is necessary to obtain equilibrium and (2) the
thermal conditions that cause the probe to incorrectly estimate temperature. In order to
empirically determine what the sensor is actually measuring, the manufacturer might attach sets
of thermistors inside and outside the probe. Comparison of these sets of thermistors would
present an independent validation of the assumption that the probe is representative of the overall
tank temperature. Furthermore, by experimenting with different product temperatures and
ambient temperatures, the manufacturer could obtain useful data that would aid in determining
the time necessary to achieve equilibrium, and in understanding the lag that might be present
between temperature changes outside and inside the tube. Permanent instrumentation as
described above would allow the manufacturer to determine whether the sensor had reached
equilibrium with the product.
A-17
-------
C.2.3 Analysis
A major improvement which can be made in the data analysis concerns the conversion of
height changes to volume changes. As described in [C-l], the method assumes that the height
measurements can be converted to volume by multiplying by the measured equivalent
cross-sectional area of all risers from the tank. When conversion is done in this manner, an error
results that is caused by the coupling of the volumetric noise sources with the deformation of the
tank. As a consequence, the volumetric estimation is grossly underestimated.
There are two approaches to correcting this error: (1) it can be minimized by testing with a
large cross-sectional area at product level; or (2) it can be virtually eliminated by releveling the
product and tracking the volume added and subtracted. A discussion of this error, and of
corrective action to be taken, is presented in the body of this report.
A-18
-------
D.I The VPLT Test Method Description
The VPLT system is an HP 9807 computer-based leak detection system which calculates a
thermally compensated leak rate. The system also compensates for vapor pockets, structural
deformation, ground water effects, pressure changes, and evaporation. Data on product
temperature, air temperature, product level, and ambient pressure are collected during a test. If
the VPLT system detects a leak in the full system test, further testing is done to determine the
location of the leak. The VPLT system employs an extensive set of test procedures which
address numerous conditions that may be encountered in the field. For this program, only one
specific protocol has been evaluated. The system is capable, according to the manufacturer, of
detecting leaks as small as 189 ml/h (0.05 gal/h).
D.I.I Instrumentation
Two independent methods of measuring product level are used: an electromechanical
sensor which floats directly on the product surface, and a gauge pressure sensor, referenced to
ambient pressure, located within 10 cm (4 in.) of the product level. Although data are collected
from both sensors, the data used for the calculations are primarily those from the
electromechanical sensor. The pressure sensor is used as an independent measurement in order
to avoid false results due to equipment failure. In the case where the product level is elevated
above grade a third sensor, whose purpose, again, is only to determine whether there are any
equipment failures, is used to determine product-level change (the product is observed in the fill
tube extender).
Both product-level sensors are calibrated prior to a test to determine the height-to-volume
scale factor, HtoV. This is performed by inserting a bar of known volume into the liquid. The
pressure transducers are calibrated every six months for pressure and temperature sensitivity
versus sensor output. Another pressure sensor measures changes in the ambient (atmospheric)
pressure.
For a 30,000-L (8,000-gal) tank, product temperature is measured using an array of five
thermistors equally spaced at 48.6-cm (19.1-in.) intervals, with the first thennistors located
24.3 cm (9.5 in.) from the base of the tank. The resolution and precision of the sensors, as
reported by the manufacturer, are summarized in Table D. 1.
A-19
-------
Table D.I. Precision and Resolution of VPLT System Sensors
Electromechanical Level Measurement
Pressure Measurement
Temperature Measurement
Precision
2.54 urn
4.0 um
0.00006°C
Resolution
2.54 urn
4.9 urn
0.00006°C
D.I.2 Equipment Setup and Test Preparations
Tank preparation and instrumentation deployment typically require about 1 h to complete,
and are relatively straightforward. Instrumentation is generally installed at the fill hole, and the
tank is overfilled nearly to grade prior to initiating preliminary calibration checks and
establishing the stability of the tank. Considerable efforts are expended to ensure that all
accessible trapped vapor is vented. In addition, a sample of product is extracted from the tank to
enable API hydrometer measurements and expansion coefficient determinations to be made.
D.I.3 Test Protocol
The significant features of the VPLT test method are summarized in Table D.2.
Table D.2. Important Aspects of the VPLT Test Protocol
Sampling Period of Temperature Sensor 3 min (30,000-L tank)
Sampling Period of Product Level Sensor 3 min (30,000-L tank)
Test Duration 2 h (nominal)
Number of Samples per Sensor per Test 41 (nominal)
Waiting Period after Addition 12 h
Test Type/Product Level Overfilled Tank
Testing with the VPLT system can be done using either of two approaches. The first
approach, not evaluated in this report, is to mix product in the tank for approximately 1.5 h
before performing the test. If the temporal gradient of any of the thermistors is greater than
0.056°C/h, the mixing must continue for an additional 30 min. The existence of this approach
was not indicated or described to the EPA until two days before the manufacturer was to arrive
for the Field Verification Tests, and, due to scheduling constraints inherent in the overall test
program, has not been evaluated. The second approach is to fill the tank prior to testing and not
mix the product. It is claimed that with this approach, a test can normally be completed within 3
to 5 h of filling the tank.
A-20
-------
The test protocol th?t was evaluated requires that the tank be filled a minimum of 12 h
prior to the test. A test pipe adapter is attached to the fill tube(s) so that product level can be
brought 5 to 7 cm above grade. If the groundwater level is too high, product is raised above
grade to counteract any masking effects which may occur due to the additional pressure of the
groundwater. The manufacturer then attempts to remove any trapped vapor pockets. A product
sample is taken and a hydrometer measurement made to determine the specific gravity of the
product being tested. The specific gravity, measured to three significant figures, and the
temperature of the product are entered into the computer where they are converted to the
coefficient of thermal expansion of the product. Nominally, the test lasts 2 h, with the second
hour of data being used to determine the reported leak rate. This test period, however, may be
extended if either the geometry band (for unknown volume in the tank and piping) or structural
deformation effects indicate that a longer test is necessary. Details of the test protocol are
included in [D-l].
The presence of structural deformation is determined by whether the leak rate over a
60-min period (calculated every 3 min) is within a range of +/- 0.01 gal/h; unless it is within this
range, the tank is not considered stabilized. The VPLT system determines the presence of vapor
pockets by detecting oscillatory movement in pressure level and leakage rate plots. If the
oscillatory behavior has a peak value exceeding 0.189 L/h in the leak rate values, it is concluded
that an excessively large vapor pocket is present, and attempts are made to remove it. If this
occurs, the product level is lowered until the vapor pocket is located and released.
D.I.4 Data Analysis
The VPLT system incorporates an averaging technique into its leak rate calculations in
order to reduce random sensor noise. No more than five data points are incorporated into the
average. The algorithm differentiates between the first-hour and post-first-hour testing. The
second hour is defined so that in the averaging technique five samples each are used for both the
new and the old values.
The VPLT system uses five methods of determining a leak rate. The leak rate reported by
the test, however, is determined by means of the averaging technique combined with the
temperature stratification compensation, using data from the primary (electromechanical) sensor.
The other four methods are used as a comparison. The leak rate calculated with the secondary
sensor (pressure sensor) is used as a check on the primary sensor. If the values do not fall within
a range of +/- 0.01 gal/h, both sensors are checked.
A-21
-------
Using a 1-h average of the data collected over the last hour of the test, the system
computes a temperature-compensated volume rate from the product pressure and temperature
measurements. Thermal compensation is performed by volumetrically weighting the readings
from the five thermistors:
ATV= I CeV,AT, (D.2.1)
where ATV is the change in volume, Cc (the coefficient of thermal expansion of the product) is
determined from hydrometer readings, V is the volume of fluid, and AT is the change in
temperature of that fluid volume.
Using the temperature and time-averaging technique, the leak rate is calculated. Over any
time interval in the test (not to exceed 1 h), a temperature-and-pressure-compensated volume
change is calculated. The compensated volume change over the time interval represents the
calculated leak rate for that sample. Because the pressure is defined as the differential pressure
in the model, ambient pressure effects are automatically compensated for. The VPLT system
actually incorporates two absolute pressure sensors, measuring atmospheric and in-tank pressure,
and compensates for atmospheric changes numerically. The temperature volume and pressure
values are obtained by using the averaging technique outlined above. Although a leak rate is
calculated at every sample period, the leak rate reported by the test is the one obtained during the
last 1-h period.
If the leak rates calculated over a 1 -h period do not fall within an error band of
+/- 0.0378 L/h (0.01 gal/h), the test must continue because the tank has not stabilized
sufficiently. The test does not end until a 1-h period of leak-rate data falls within this range. If
this occurs before 2 h of testing have been completed, the test continues to the 2-h mark. If all
criteria are met, the test is considered successful.
D.I.5 Detection Criterion
A leak is declared if the compensated volume rate for the last hour of testing is greater than
a threshold of 0.189 L/h (0.05 gal/h). This flow rate is based on the volume rate recommended
by NFPA. If the flow rate is less than the threshold, the tank is declared nonleaking. An
inflowing compensated volume rate exceeding the threshold is considered to indicate an
inconclusive test.
D.I.6 Mathematical Model
While the description provided in Section D.2 is sufficient to identify the key features of
the VPLT system as it was evaluated and to interpret the results of the performance evaluation, it
A-22
-------
should be noted that the mathematical model of the system was based on the more detailed
description of the method provided in [D-l]. After it had been validated, the model was used to
estimate the performance of the method.
D.2 Improving System Performance
The Computerized VPLT Tank Leak Testing System system could realize performance
improvements without making any hardware modifications.
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual performance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. A waiting period after
topping is required to minimize the impact of the (1) spatial inhomogeneities and (2) structural
deformation produced by the addition (or removal) of product to reach the level required to begin
the test. The effective waiting period for this method is determined by an analysis algorithm;
the effectiveness of this algorithm was not evaluated.
A waiting period of at least 2 to 3 h is required to minimize the spatial inhomogeneities in
temperature. This period is long enough to minimize the effects of a 7°C difference in
temperature between the product in the tank and the product used to top the tank. A longer
period may be required to minimize the deformation-induced product-level changes. This can
present operational problems, because the time required to minimze the effects of defonnation in
some tanks may be 1 to 5 times (or more) longer than the time required to wait for the
temperature effects to stabilize. The time constants of the Test Apparatus tanks, for example, are
2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or longer; the waiting
period would be longer if the deformation effects were large and more than three time contants
were required for them to become small compared to the smallest leak to be detected. Since the
temperature and deformation effects associated with topping are independent of each other, the
protocol and/or analysis algorithm use to minimize this effect must address each disturbance. A
single waiting period based on a maximum disturbance would adequately address the
temperature instability. However, this may not suffice for the deformation effect.
D.2.1 Protocol
The manufacturer of the Computerized VPLT Tank Leak Testing System could improved
its performance by introducing a waiting period of 3 to 6 h or longer after topping off the tank,
thus reducing errors due to structural defonnation and to the destruction of a stable temperature
field. Instead of allowing the product level to drift, the manufacturer could periodically adjust
product level. This would remove the error present in interpreting product-level changes.
A-23
-------
D.2.2 Detection Criterion
The convergence criterion should be revised. The probability of false alarm is determined
by the threshold. The manufacturer should determine what probability of false alarm is desired
and adjust the threshold accordingly.
A-24
-------
E.3 The DWY Leak Sensor Test Method Description
The DWY Leak Sensor is a volumetric leak detection system capable, according to the
manufacturer, of detecting leaks as small as +/- 0.189 L/h (0.05 gal/h) that occur in the walls or
product distribution lines of an underground storage tank. For the performance results described
in this report, it was employed as a precision (volumetric) leak detection test operating in an
underfilled tank.
E.I.I Instrumentation
The DWY Leak Sensor uses two measuring devices, a float and a detector rod, both of
which measure product level, to estimate a temperature-compensated flow rate.
The position of the float, which is measured over a period of time, is related to the volume
changes by means of the phenomenon that supports a buoyant object in a liquid. The float
assembly is dependent on the size and type of the tank. A float shell is selected from a group of
shells of varying length suitable for a variety of tank sizes. The float length and weight are
chosen based on the diameter of the tank, the height of product in the tank, and the density of the
test product. These measured values are determined prior to a test.
The detector rod assembly consists of a photoelectric cell and a float attachment hinge.
The float movement forces an ink solution into or out of the photocell, and the resulting change
in light transmittance in the photocell causes a voltage drop across the cell. This voltage change,
which is a function of the product-level change, is measured by a voltmeter that has been
calibrated prior to the test.
Volumetric changes are determined by monitoring product height changes with the float.
The float senses these liquid-level changes in the tank, and the detector rod measures the relative
position of the float with respect to the rod. The data are recorded on a strip chart.
The DWY Leak Sensor compensates for thermal effects by taking advantage of the natural
temperature compensation inherent in the specifically designed float. This phenomenon is
established by the relationship that exists between the temperature and three other factors,
namely, liquid density, liquid volume, and liquid buoyancy forces (which are a function of
density). A change in temperature (thus a change in volume) can therefore be exactly offset by
density (buoyancy) changes providing that the floating object (i.e., float) is at a point in the tank
where the volumetric and density changes caused by the temperature variations have an equal
effect.
A-2.S
-------
The manufacturer's claimed precision and resolution of these measurements are
summarized in Table E.I, and were experimentally verified at the UST Test Apparatus in
Edison, New Jersey during the Field Verification Tests.
Table E. I. Manufacturer's Precision and Resolution Claim for the DWY System Measurements
Precision Resolution
Product Level 0.0005 cm 0.0005 cm
E.I.2 Equipment Setup and Tesl Preparations
The manufacturer indicated that it would take approximately 1 h to install the equipment
and prepare for the start of the test. This assumes that the product level is satisfactory for the
conduct of a test.
E.I.3 Test Protocol
The important features of the DWY test protocol are summarized in Table E.2. Test
protocol requires that the tank be filled to 50% to 82% of capacity. There is a waiting period of
at least 6 h between the filling of the tank and the start of a test (i.e., calibrations), so as to allow
the temperature of the product in the tank to stabilize. During the 1-h period immediately
preceding data collection a 1/3-hp pump is used to spray product into the vapor space in order to
saturate it. The 1-h period is the approximate time required to complete this procedure in a
30,000-L (8,000-gal) tank. Instrumentation warm-up and calibration also take place during this
time. Calibration and testing continue at 1-h intervals until two consecutive readings give the
same calculated (leak rate) result, or until a 4-h maximum test period has been completed.
Table E.2. Important Aspects of the DWY Test Protocol
Sampling Period 1 h
Test Duration 4 h (maximum)
Waiting Period after Addition 6 h
Test Type/Product Level Underfilled Tank/50% to 82% of capacity
E.I.4 Data Analysis
A thermally compensated volume change is computed from the change in the height of a
weighted float immersed at the null point of the tank where thermal changes are exactly offset by
A-26
-------
buoyancy forces. Data are obtained by sampling once per hour over the duration of the test. The
voltage change over the 1-h period is converted to volume by multiplying by a conversion factor
determined from the calibrations made before the test.
A tank chart, supplied by the tank manufacturer, is used to convert product-level
measurements to volume. Height-to-volume conversion is accomplished by using linear
interpolation between pairs of data points tabulated at 2.54-cm (1-in.) intervals of tank depth.
All tests are considered successful providing that data are collected for the entire duration
of the test period, and that the appropriate initial waiting period is observed. A test is considered
successful when two consecutive 1 -h tests give the same calculated result, or when the 4-h
maximum test period has been completed. The reported leak rate is that computed over the last
1-h testing period.
E.I.5 Detection Criterion
A leak is declared if the thermally compensated volume rate derived for the test period
exceeds a threshold of+/-0.189 L/h (0.05 gal/h). If the flow rate is less than the threshold, the
tank is declared nonleaking.
E.I.6 Mathematical Model
While the description provided in Section E.2 is sufficient to identify the key features of
the DWY system as it was evaluated, and to interpret the results of the performance evaluation, it
should be noted that a mathematical model of the system was developed based on the more
detailed description of the test method provided in [E-l], After it had been validated, this model
was used to estimate the performance of the method.
E.2 Improving System Performance
A systematic investigation of the DWY system should be made. There are two sources of
error that should be investigated: the implementation of the float system for temperature
compensation and the effect of spraying product into the vapor space before the test.
Theoretically, the float system should behave as predicted, but whether it actually does is
unknown. The Test Apparatus experiments suggests that any disturbance to the product or vapor
immediately before a test can result in large testing errors; such disturbances should be avoided.
A-27
-------
-------
F.I The EZY CHEK Test Method Description
The EZY CHEK leak detection system estimates a temperature-compensated volume rate
from measurements of product level and product temperature. This is an overfilled-tank test.
Although test length is considered constant, the sampling interval can vary. The manufacturer of
the EZY CHEK leak detection system claims the ability to detect leak rates of 0.076 L/h
(0.02 gal/h).
F.I.I Instrumentation
Volumetric changes are determined indirectly with a pressure sensor, calibrated before and
after each tank test, that measures the pressure required to force air bubbles into the product. A
simple calibration performed for each test relates the observed value (lines on a chart recorder) to
the apparent change in volume using a cylinder of known volume (0.19 L (0.05 gal)). Because
the measurement is made in a standpipe (fill tube extender) located above ground, the tank must
be overfilled to a level within the fill tube. It should be noted that, in normal situations, product
is added until it is approximately 50.8 cm (20 in.) above grade in the standpipe.
Temperature is measured using a temperature averaging probe consisting of platinum
sensing wire encased in a coiled spring of plastic tubing. The coil is wound more tightly toward
the center of the probe, thus weighting the temperature of the measured product more heavily
toward the center of the tank. A quantitative description of the higher density of coils toward the
center is presented in Appendix B of the technical memorandum on the mathematical model
[F-l].
Both temperature and volumetric data are sampled at 15-min intervals. The temperature
data, obtained from a digital multimeter, are digitized with an integration time of 30 to 45 s,
while the volumetric data, obtained from a chart recorder, are manually digitized. The chart
recorder is circular and has 29 divisions per inch. Because the recorder and pressure gauge have
a combined range of 2.5 cm (1 in.) of water, the bubbler height must be adjusted and the chart
pen repositioned if the height change is greater than 2.5 cm. A 2.5-cm change equals
approximately a 463-ml change in volume in a 15-cm (6-in.) diameter standpipe.
According to the EZY CHEK representative, a 22-line change is nominally observed for a
0.19-L calibration rod for a tank with a 5-cm (2-in.) diameter vent and a 10-cm (4-in.) diameter
fill tube. Taking into account that the chart recorder is read to a resolution of 1/2 line, a
resolution of 0.004 L would be obtained.
A-29
-------
The precision of the sensor is 0.013 cm (0.005 in.) for water. For gasoline, the precision is
0.018 cm (0.007 in.) due to the lower density of that fluid. Precision and resolution calculations
for product-level measurement are presented in Table F. 1. The model of the method assumed a
5-cm diameter vent and 15-cm diameter standpipe.
Table F.I. Precision and Resolution of the EZY CHEK Leak Detection
System as Specified by the Manufacturer
Level Measurement
Temperature Measurement
Precision
0.004 L
0.001'C
Resolution
0.004 L
0.0006°C
F.1.2 Equipment Setup and Test Preparation
For a 30,000-L (8,000 gal) tank there is a 6-h minimum wait required between the filling
of the tank and the commencement of the test. After this wait, the tank is topped off, the
standpipe is filled with product, and testing begins immediately. A test is not complete unless
2 h of "good" data are obtained [F-l].
F.I.3 Test Protocol
The important features of the EZY CHEK test method are summarized in Table F.2. The
test is conducted 6 h after a product addition and lasts for a period of 2 h. The temperature data
are collected manually at a 15-min interval. The fluid-level data are recorded on an analog chart
recorder.
Table F.2. Important Aspects of the EZY CHEK Test Protocol
Sampling Period 15 min
Test Duration 2 h
Waiting Period after Addition 6 h
Test Type/Product Level Overfilled Tank/Above Grade
F.I.4 Data Analysis
A temperature-compensated volume rate is computed from the product-level and
product-temperature measurements using an average of the data obtained during the last hour of
the test.
The thermal expansion and contraction of the product, ATV, is estimated using
A-30
-------
ATV = CCVAT (F.2.1)
where the rate of change of temperature, AT, is estimated from the thermal probe, the total
volume, V, of the product in the tank is given as the capacity of the tank, and a value of the
coefficient of thermal expansion, Ce, is calculated by measuring both the specific gravity and the
temperature of the product and using an ASTM petroleum measurement table.
A test is considered successful if the eight values for the measured change in volume are
within +/- 0.024 L of each other and the eight values for the measured temperature change per
15-min period are all within +/- 0.003°C of each other.
F.I.5 Detection Criteria
The tank is declared leaking if both of the following criteria are met: (a) the test is
successful (valid), and (b) the leak rate is greater than 0.19 L/h and represents an outflow. An
inflow of 0.076 L/h or greater is considered an inconclusive test.
F.I.6 Mathematical Model
While the description provided above is sufficient to identify the key features of EZY
CHEK as it was evaluated and to interpret the results of the perfonnance evaluation, it should be
noted that a mathematical model of the system was developed based on the more detailed
description of the test method provided in [F-IJ. After it had been validated, this model was
used to estimate the performance of the method.
F.2 Improving System Performance
The EZY CHEK system defined in [F-l ] can realize perfonnance improvements. One
involves a change in protocol and the other a change to the analysis algorithm.
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual perfonnance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. The 1-h waiting period
after topping is too short to minimize the impact of the (1) spatial inhomogeneities and (2)
structural defonnation produced by the addition (or removal) of product to reach the level
required to begin the test. A waiting period of at least 2 to 3 h is required to minimize the spatial
inhomogeneities in temperature. This period is long enough to minimize the effects of a 7°C
difference in temperature between the product in the tank and the product used to top the tank. A
longer period may be required to minimize the defonnation-induced product-level changes. This
can present operational problems, because the time required to minimze the effects of
deformation in some tanks may be 1 to 5 times (or more) longer than the time required to wait
A-31
-------
for the temperature effects to stabilize. The time constants of the Test Apparatus tanks, for
example, are 2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or longer;
the waiting period would be longer if the deformation effects were large and more than three
time contants were required for them to become small compared to the smallest leak to be
detected. Since the temperature and deformation effects associated with topping are independent
of each other, the protocol and/or analysis algorithm use to minimize this effect must address
each disturbance. A single waiting period based on a maximum disturbance would adequately
address the temperature instability. However, this may not suffice for the deformation effect.
F.2.1 Protocol
The following protocol changes are recommended:
o Wait 6 to 12 h or longer after product delivery
o Relevel the product at 15-min intervals or less and record the volume changes
o Wait 3 to 6 h or longer after topping off the tank, and maintain constant product
level during the waiting period
o Sample the temperature data every 3 min
The performance of the EZY CHEK system is significantly affected by (1) the lack of an
adequate method for avoiding systematic errors due to structural deformation and (2) a time
period inadequate for the temperature field to stabilize after product delivery. After product
delivery a period of 6 to 12 h or longer is necessary to allow the temperature field to stabilize.
The manufacturer should stipulate this in his protocol. Furthermore, he must ascertain whether
the measured flow rate is being significantly affected by the systematic error resulting from
structural deformation. The problem here lies in having, a priori, an adequate estimate of the
time and tank elasticity constants of the tank in order to determine the length of time necessary to
allow tank deformation effects to become negligible. A general discussion of methods to
determine whether structural deformation has subsided is presented in the body of the report. It
is recommended that the manufacturer increase the waiting period after topping off the tank in
order to allow the temperature field to stabilize. Finally, the manufacturer could remove the
possibility of misinterpreting height changes by releveling during a test and keeping track of the
amount of product added and removed for the volume measurements.
Releveling the product represents the most significant performance improvement that can
be achieved by the EZY CHEK system. Releveling removes the deleterious effect of dynamic
structural deformation described in Section 6 of the report. Eliminating the dynamic structural
deformation would improve the performance of the system by approximately 50%.
A-32
-------
F.2.2 Analysis
The averaging scheme used by the manufacturer results in a
first-value-subtracted-from-last-value algorithm. The EZY CHEK analysis therefore makes only
partial use of the available data, utilizing only the first and last measurements of the 1-h analysis
period, and failing to utilize those collected at each 15-min interval. Intennediate measurements
presently have no other value than to allow the operator to spot any change in the rate of
temperature or volume. By performing a least-squares fit of the data collected over a 2-h period,
the manufacturer could reduce the instrumentation limitation of his system to a standard error of
18 ml/h, thus reducing the instrumentation-limited detectable leak rate to 84 ml/h with a
PD= 0.99 and a PFA= 0.01. By increasing to the recommended sample rate and using a
least-squares linear analysis, the manufacturer could further decrease the instrumentation
limitation to a standard error of less than 6 ml/h and improve the detectable leak rate to 30 ml/h
with a PD = 0.99 and PFA = 0.01. This procedure could improve performance by as much as
125%.
F.2.3 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-33
-------
-------
G.I The Gasoline Tank Monitor (GTM) Test Method Description
The Gasoline Tank Monitor (GTM) is an automatic tank gauging system capable,
according to the manufacturer, of detecting leaks as small as 3.785 L (1.0 gal) when operated in a
precision test (shutdown) mode. This is the minimum volume loss the GTM system can detect
over the duration of a test (i.e., this is the threshold value for which the system is programmed);
thus, for the 10-h test specified by test protocol, the minimum detectable flow rate would be
0.379 L/h (0.10 gal/h). The monitor, consisting of a tank sensor and a remote control console, is
mounted permanently in the tank, and can perform a leak detection test when dispensing
operations have ceased. For the tests described in this appendix, the GTM was employed as a
volumetric test method using the protocol outlined in [G-l] and summarized below. The
protocol of the GTM system employed as a volumetric test is different from that used when the
system is in an automatic tank gauging mode.
(i.l.l Instrumentation
The GTM system uses two measuring devices to estimate a temperature-compensated flow
rate: product level as determined by means of an acoustic device (ultrasonic transducer), and
product temperature as determined by a temperature probe (bimetallic coil).
Volumetric changes are determined by monitoring fluctuations in product level (height)
with an ultrasonic transducer that has been excited by a 4-(is voltage pulse to produce a 250-kHz
incident signal, and recording them on an 8031 microprocessor. The resonant characteristic of
the transducer creates an envelope-shaped waveform that propagates through a 5.08-cm (2-in.)
diameter fiberglass tube containing a fixed reference target, a bimetallic coil having a position
that is a function of temperature, and a third target consisting of the liquid/air surface boundary.
Each of these targets produces a reflected wave that is received by the transducer. The
microprocessor records height changes digitally.
The transducer output, shaped by a threshold detection circuit to produce a pulse, is
detected by a precision rectifier/low-pass filter circuit, the leading edge of the output pulse
identifying that of the reflected signal envelope. A programmable timer with a 4-MHz reference
clock that begins counting when the transducer is energized is used to measure the round-trip
travel time of the acoustic wave. The round-trip travel time for each echo is stored by the
microprocessor. Echoes from the fixed reference target and the liquid surface boundary are used
to calculate the product level, and the echoes from the reference target and the bimetallic coil
reflector are used to calculate the distance between the transducer and the coil reflector. As the
product level or temperature changes, so does the travel time of the acoustic signal. This
distance, in turn, is used to calculate the temperature of the product in the vicinity of the coil.
A-35
-------
Temperature is determined by using the ultrasonic transducer to monitor the position of a
bimetallic coil. The angular movement of the coil is a function of its temperature. The nominal
temperature-displacement curve supplied by the manufacturer is used to convert from angular
displacement to temperature. The bimetallic coil has an operating range of 4 to 38°C (40 to
100°F). Its resolution, which varies with temperature because the number of angular degrees per
degree Celsius is not constant, is less than 0.06°C for temperatures less than 38°C. The
bimetallic coil is located approximately 4.5 cm from the bottom of the tank.
Product temperature data, like height data, are sampled continuously between the start and
end of the test, and are used to determine the change in temperature, which will in turn be used in
the thermal compensation calculations.
The claimed precision of these measurements is summarized in Table G. 1, and was
experimentally verified at the UST Test Apparatus during the field verification tests.
Table (1.1. Manufacturer's Precision and Resolution Claim for the GTM System Measurements
Precision Resolution
Product Level 0.0254 cm 0.0254 cm
Temperature 0.06°C 0.06°C
(J.I.2 Equipment Setup and Test Preparations
The GTM system sensor can be installed permanently in the tank via a 7.6-cm (3-in.) or
10.2-cm (4-in.) riser located at the top of the tank. Installation of the sensor on the Test
Apparatus, along with the associated temporary wiring to the console, required approximately 1
h. Assuming the proper liquid level had been established in the tank, testing could then have
commenced immediately.
(1.1.3 Test Protocol
Employing a tank test in its volumetric mode requires that the tank be filled to between
50% and 95% of capacity, with a minimum product height of 45.7 cm (18 in.). Because the
tank's temperature field may be significantly disturbed by the addition of product to attain this
level, a waiting period of 24 h after filling must be observed. After this wait, a test may be
initiated. The total test duration is specified as 24 h. The significant aspects of the test protocol
are summarized in Table G.2.
A-36
-------
Table (J.2. Important Aspects of the GTM Test Protocol
Sampling Period 1 s
Test Duration 24 h
Waiting Period after Addition 24 h
Test Type/Product Level Underfilled Tank/50% - 95% of capacity
(J.I.4 Data Analysis
Volume and temperature are calculated from an average of 32 measurement samples
obtained for each second of the test. The volume rate and temperature-volume rate are then
calculated, and a thermally compensated volume rate is continuously developed during each
second of the test. The total leak rate for the test period is calculated from the sum of the
temperature-compensated volumes divided by the total test time.
A tank chart, supplied by the tank manufacturer, is used to convert product-level
measurements to volume. Height-to-volume conversion is accomplished by using linear
interpolation based on a 32-point table.
Thermal expansion or contraction of the product, ATV, is estimated from
ATV = CCVAT (G.2.1)
where the temperature change, AT, is estimated from the output of the ultrasonic detector; the
total volume, V, of the product is estimated from a product level measurement; and a value of the
coefficient of thermal expansion, Ce, for gasoline is determined, by the operator, from the
American Petroleum Institute (API) type-6B algorithm which is incorporated in the GTM
console. In order to use this algorithm, a measurement of the API gravity of the product must be
made and the value centered into the console.
All tests are considered successful providing that data are collected for the entire duration
of the test period, and that the appropriate initial waiting period is observed.
(J.1.5 Detection Criterion
A leak is declared if the thermally compensated volume rate exceeds a threshold of 0.189
L/h (0.05 gal/h). If the flow rate is less than the threshold, the tank is declared nonleaking.
(J.I.6 Mathematical Model
While the description provided above is sufficient to identify the key features of the GTM
as it was evaluated and to interpret the results of the performance evaluation, it should be noted
A-37
-------
that a mathematical model of the system was developed based on the more detailed description
of the test method provided in [G-l]. After it had been validated, this model was used to
estimate the performance of the method.
G.2 Improving System Performance
The GTM system can realize performance improvements by making changes to the
instrumentation, protocol, analysis, and detection criterion.
(i.2.1 Instrumentation
Probably the largest sources of errors are due to inadequate measurement of the vertical
temperature field. This source of error is two-fold:
o The distance between the reference fiducial and the surface produces a
systematic error in the estimation of the product height. This error has not been
quantified for this evaluation, but it will control performance. It can result in a
large systematic error.
o The performance of the Gasoline Tank Monitor is also limited by its
point-source temperature system. By increasing the number of temperature
probes between the tank bottom and product surface the temperature field could
be better estimated.
The change in performance as a function of the number of thermistors (temperature
probes) on a vertical array is discussed in the body of the final report. Substantial improvement
occurs when additional thermistors are used, allowing better sampling of the temperature field.
C.2.2 Analysis
The Gasoline Tank Monitor system could improve performance further by performing a
least-squares regression on the data. This could significantly alleviate the problem associated
with aliasing internal waves, and makes maximum use of the infonnation obtained from the data
sampling. Refer to the main body of the report for a discussion of performance as a function of
sampling and test duration.
(J.2.3 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-38
-------
H.I The Gilbarco Tank Monitor Test Method Description
The Gilbarco Tank Monitor is an automatic tank gauging system capable, according to the
manufacturer, of detecting leaks as small as 0.757 L/h (0.20 gal/h) when operated in a precision
test (shutdown) mode. The monitor, consisting of a tank sensor and a remote control console, is
mounted permanently in the tank, and can perform a leak detection test when dispensing
operations have ceased. For the tests described in this appendix, it was employed as a volumetric
test method using the protocol outlined in [H-l] and summarized below. The protocol of the
Gilbarco Tank Monitor employed as a volumetric test is different from that used when it is in an
automatic tank gauging mode.
H.I.I Instrumentation
The Gilbarco Tank Monitor system utilizes three types of sonic measurements to
determine the temperature-compensated volume rate: fuel surface ("1"), calibration reflectors
(fiducials) ("2"), and water ("3"). The measurements are taken using the sequence 1,2, 1,3, 1,2,
1, 3..., etc., with a time differential of approximately 70 ms between each measurement.
Volumetric changes are determined by monitoring fluctuations in product level (height)
with an ultrasonic transducer that has been excited by a voltage pulse nominally of 465 ns to
produce a 1.1-MHz incident signal, and digitally recording them on a microprocessor. The
resonant characteristic of the transducer creates an envelope-shaped waveform that propagates
through the product, which contains several fixed reference targets (fiducials). These fiducials
are attached to a vertical aluminum rod at equally spaced intervals of 30.5 cm (12 in.), beginning
30.5 cm from the tank bottom. A second medium consisting of the liquid/air surface boundary
serves as the variable target. Each of these targets (i.e., the fixed targets and the liquid/air
surface) produces a reflected wave that is received by the transducer.
The transducer output, shaped by a threshold detection circuit to produce a pulse, is
detected by a precision rectifier/low-pass filter circuit. The leading edge of the output pulse
identifies the presence of the reflected signal envelope. A programmable timer with a
19.66-MHz reference clock that begins counting when the transducer is energized is used to
measure the round-trip travel time of the acoustic wave. The round-trip travel time for each echo
is stored by the microprocessor. Echoes from the fixed reference fiducials and the liquid surface
boundary are used to calculate the product level, while temperature is calculated using the echoes
from the bottom and top fiducials. It should be noted that the word "top" refers to topmost
submerged fiducial.
A-39
-------
The product height is measured by means of an ultrasonic transducer located 15.2 cm
(6 in.) from the tank bottom. The round-trip travel time of the reflected acoustic signal is used to
monitor the liquid surface. The travel time of the acoustic signal changes in proportion to the
product-level changes. The operating range of the product level-measurement device is from 19
to 320 crn (7.5 to 126 in.).
Temperature is also measured by using the ultrasonic transducer, which is used to monitor
the acoustic velocity changes. (Acoustic velocity is determined by measuring the round-trip time
for sound to travel from the the transducer to the highest reflector that is submerged at least 2.54
cm below the product surface.) The nominal temperature curve supplied by the manufacturer is
used to convert the velocity of the acoustic signal to temperature. The resolution is less than
0.0006°C for temperatures greater than 38°C, varying with temperature because the velocity of
sound per degree Celsius is not constant. The nominal operating range of temperatures is -32 to
38°C(Oto 100°F).
Product temperature data, like height data, are sampled continuously between the start and
end of the test, and are used to determine the change in temperature, which in turn is used in the
thermal compensation calculations.
The manufacturer's claimed precision and resolution for these measurements, summarized
in Table H.I, were experimentally verified at the UST Test Apparatus in Edison, New Jersey
during the field tests of the Gilbarco system.
Table H.I. Manufacturer's Precision and Resolution Claim for the Gilbarco Tank Monitor
System Measurements
Precision Resolution
Product Level 0.000254 cm 0.000178 cm
Temperature 0.001°C 0.0006°C
H.I.2 Equipment Setup and Test Preparations
Under actual field conditions, the sensor is installed pennanently in the tank via a 10-cm
(4-in.) riser located at the top of the tank. Installation of the sensor in the Test Apparatus tank,
along with the associated temporary wiring to the console, required approximately 2.5 h.
Assuming the proper liquid level had been established in the tank, testing could then have
commenced immediately.
A-40
-------
H.I.3 Test Protocol
Employing a tank test in its volumetric mode requires that the tank be filled to between
80% and 95% of capacity. Because the tank's temperature field may be significantly disturbed
by the addition of product to attain this level, a waiting period of 18 h after filling is specified.
After this wait, a test may be initiated. The total test duration is specified as 5 h. The significant
aspects of the test protocol are summarized in Table H.2.
Table H.2. Important Aspects of the Gilbarco Tank Monitor Test Protocol
Sampling Period 7 to 8 s
Test Duration 5 h
Waiting Period after Addition 18 h
Test Type/Product Level Underfilled Tank/80% to 95% of capacity
H.I. 4 Data Analysis
Volume and temperature are calculated from 50 measurement samples obtained for each
hour of the test. The volume rate and temperature volume rate are then calculated, and a
thermally compensated volume rate is developed for each hour of the test. The total leak rate for
the test period is calculated from the sum of the hourly temperature-compensated volumes
divided by the total test time.
A tank chart, supplied by the tank manufacturer, is used to convert product-level
measurements to volume. Height-to-volume (HtoV) conversion is accomplished by using linear
interpolation between pairs of data points tabulated at 2.54-cm intervals of tank depth.
Thermal expansion or contraction of the product, ATV, is estimated from
= CVAT H.2.1
where the temperature change, AT, is estimated from the output of the ultrasonic detector; the
total volume, V, of the product is estimated from a product-level measurement; and a value of
the coefficient of thermal expansion, Ce, for gasoline is furnished by the petroleum company.
All tests are considered successful providing that data are collected for the entire duration
of the test period, and that the appropriate initial waiting period is observed.
H.1.5 Detection Criterion
A leak is declared if the thermally compensated volume rate for the test period exceeds a
threshold of +/- 0. 189 L/h (0.05 gal/h). If the flow rate is less than the threshold, the tank is
declared nonleaking.
A-41
-------
H.I.6 Mathematical Model
While the description provided above is sufficient to identify the key features of the
Gilbarco Tank Monitor as it was evaluated and to interpret the results of the performance
evaluation, it should be noted that a mathematical model of the system was developed based on
the more detailed description of the test method provided in [H-l]. After it had been validated,
this model was used to estimate the performance of the method.
H.2 Improving System Performance
The Gilbarco Tank Monitor system can realize performance improvements. These involve
changes to the analysis and detection criterion, and are discussed below.
11.2.1 Instrumentation
The main source of error is the measurement of product level. By judicial selection of the
depth of the product relative to the reference fiducial, this error has been minimized. This effect
has not been thoroughly investigated and the magnitude of this error is not known.
H.2.2 Analysis
The Gilbarco Tank Monitor accurately measures a vertical average of the temperature
field. A weighted average would improve performance. A discussion of the performance
improvements achieved by using a volumetrically weighted average of temperature is given in
the main body of the report.
The Gilbarco Tank Monitor system could improve performance by performing a least
squares regression on the data. This makes maximum use of the information obtained from data
sampling.
H.2.3 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-42
-------
I.I The Inductive Leak Detector 3100 Test Method Description
The Inductive Leak Detector 3100 is a system that provides inventory reconciliation
(automatic tank gauging) and is capable, according to the manufacturer, of detecting leaks as
small as +/- 0.189 L/h (0.05 gal/h) when operated in a precision test (shutdown) mode. The
monitor, consisting of two tank sensors (product height and temperature) and a remote control
console, is mounted permanently in the tank, and can perform a leak detection test when
dispensing operations have ceased. For the tests described in this appendix, it was employed as a
volumetric test method using the protocol outlined in [1-1] and summarized below. The protocol
of the Inductive Leak Detector 3100 employed as a volumetric (precision leak) test is different
from that used when the system is in an automatic tank gauging mode.
1.1.1 Instrumentation
The Inductive Leak Detector 3100 system when in leak check mode utilizes two measuring
devices to estimate a temperature-compensated flow rate. Product level is determined by means
of a digital rule and float, and product temperature is determined by a temperature probe.
Temperature is measured by means of a single resistance temperature detector (RTD) placed
87.6 cm (34.5 in.) from the bottom of a 244-cm (96-in.) diameter tank.
A Minco resistance temperature detector (RTD) is used to measure temperature changes.
The RTD has an operating range of -40 to 100°C (-40 to 212°F). The nominal
temperature-resistance curve supplied by the manufacturer is used to convert from ohms to
temperature. The resolution is less than 0.17°C for temperatures less than 0°C, varying with
temperature because the number of ohms per degree Celsius is not constant.
Volumetric changes are determined by monitoring fluctuations in the product level
(height). The product-level sensor consists of a float device riding the product surface, and is
attached to a digital rule that senses the product-level fluctuations due to the motion of the
surface waves. These height changes are digitally recorded on a microprocessor.
Product-temperature data, like height data, are sampled continuously for 1 min at the start
and end of the test, and are used to determine the change in temperature, which will in turn be
used in the thermal compensation calculations.
The claimed precision and resolution of these measurements are summarized in Table I.I,
and were experimentally verified at the UST Test Apparatus in Edison, New Jersey during the
Field Verification Tests.
A-43
-------
Table I.I. Manufacturer's Precision and Resolution Claim for the Inductive
Leak Detector 3100 System Measurements
Product Level
Temperature
Precision
0.254 cm
0.17°C
Resolution
0.127cm
0.17°C
1.1.2 Equipment Setup and Test Preparations
The sensor is installed permanently in the tank via a 10-cm (4-in.) riser located at the top
of the tank. Installation of the sensor on the tank, along with the associated temporary wiring to
the console, required approximately 1 h. Assuming the proper liquid level had been established
in the tank, testing could then have commenced immediately.
1.1.3 Test Protocol
Testing a tank in the volumetric test mode requires that the tank be filled to approximately
93% of the tank diameter (97% of tank capacity for a tank 243.84 cm in diameter). Because the
tank's temperature field may be significantly disturbed by the addition of product to attain this
level, a waiting period of 3 h after filling must be observed. After this wait, a test may be
initiated. The total test duration is specified as 10 h. The significant aspects of the test protocol
are summarized in Table 1.2.
Table 1.2. Important Aspects of the Inductive Leak Detector 3100 Test Protocol
Sampling Period 10 s (first and last minute only)
Test Duration 10 h
Waiting Period after Addition 3 h
Test Type/Product Level Underfilled Tank/93% of tank diameter
1.1.4 Data Analysis
Volume and temperature are calculated from 14 measurement samples (seven each from
the first and last minutes) obtained from each sensor. The volume rate and temperature volume
rate are then calculated, and a thermally compensated volume rate developed for each test.
A tank chart, supplied by the tank manufacturer, is used to convert product-level
measurements to volume. Height-to-volume conversion is accomplished by linear interpolation
between pairs of data points tabulated at 2.54-cm (1-in.) intervals of tank depth.
Thermal expansion or contraction of the product, ATV, is estimated from
A-44
-------
= CCVAT (1.2.1)
where the temperature change, AT, is estimated from the output of the temperature probe, the
total volume, V, of the product is estimated from a product-level measurement, and a value of
the coefficient of thermal expansion, Ce, for gasoline is furnished by the petroleum company.
All tests are considered successful providing that data are collected for the entire duration
of the test period, and that the appropriate initial waiting period is observed.
1.1.5 Detection Criterion
A leak is declared if the thermally compensated volume rate for the test period, At, derived
from Eq. (1.2.2), exceeds a threshold1 of +/- 0.189 L/h. If the flow rate is less than the threshold,
the tank is declared nonleaking.
AV-ATV
Flow rate = (1.2.2)
At
1.1.6 Mathematical Model
While the description provided in Section 1.2 is sufficient to identify the key features of the
Inductive Leak Detector 3100 as it was evaluated and to interpret the results of the performance
evaluation, it should be noted that a mathematical model of the system was developed based on
the more detailed description of the test method provided in [1-1 ]. After it had been validated,
this model was used to estimate the performance of the method.
1.2 Improving System Performance
The Inductive Leak Detector 3100 system can realize performance improvements if
changes are made to the instrumentation, test protocol, analysis, and detection criterion. Since
the system is limited by the system noise changes to the instrumentation are highly
recommended.
1.2.1 Instrumentation
Probably the largest source of error is due to inadequate temperature compensation, a
result of improper determination of the vertical temperature field. The change in perfonnance as
a function of the number of thermistors (temperature probes) on a vertical array is discussed in
1 Currently the Inductive Leak Detector 3100 system does not set a threshold to declare a tank
leaking or nonleaking, but reports a leak rate. The threshold value of 0.189 L/h conforms to the
current NFPA standard and is used as the detection threshold for the purposes of this report.
A-45
-------
the body of the final report. Substantial improvement occurs when additional thermistors are
used, allowing better sampling of the temperature field. In addition, improving the precision of
the temperature sensors will substantially improve performance.
Based upon studies of canonical methods presented in Section 10 of the report, it is
estimated that a 50% performance improvement (assuming thermistors with better precision)
could be achieved by increasing the number of thermistors to five.
1.2.2 Protocol
The following protocol changes are recommended:
o Wait 6 to 12 h or longer after product delivery
o Sample the temperature data every 3 min
Increasing the minimum waiting period after a product delivery could further improve the
Inductive Leak Detector 3100 system. The thermal effects of a delivery are minimized only
because the current protocol stipulates a long test duration. The test results are influenced also
by the effects of unknown temperature changes due to horizontal gradients, as well as by
structural deformation effects, as discussed in the main body of the report.
1.2.3 Analysis
The performance of the Inductive Leak Detector 3100 system could be improved by
increasing the sample rate over the duration of the test rather than taking the first and last minute
of data and performing a least-squares regression. This could significantly alleviate the problem
associated with aliasing internal waves and maximize the information obtained from the data
sampling. Refer to the body of the report for a discussion of performance as a function of
sampling and test duration.
1.2.4 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly. This is
also discussed in more detail in the body of the report.
A-46
-------
J.I The INSTA-TEST System Test Method Description
The INSTA-TEST system is a volumetric leak detection system which, for this evaluation,
was employed strictly as a test of the tightness of underfilled tanks.
J.I.I Instrumentation
The INSTA-TEST leak detection system measures the differential change in pressure
obtained by comparing open and closed tubes placed in the tank. The tubes are made of copper
and are 1.90 cm in diameter and approximately 3.3 m long. They are inserted into the tank
through the fill tube before a test. The top of each copper tube is open, as is the base, which is
within a few inches of the center line of the tank. After installation is completed, the base of one
tube is closed, preventing any product from entering. The system remains in this condition for a
variable time period, at the end of which the tops of both tubes are closed and connected via a
differential pressure transducer that permits comparison of their respective pressures. The base
of the closed tube is then opened, and the pressure change resulting from the level changes in the
open and closed tubes is read from a digital meter. This change in pressure is converted to
volume using a scale factor determined during a previous calibration. Since thermal
compensation is performed physically by the integration of temperature in the closed tube, no
calculations are performed.
Test length is determined as a function of tank size and product level. The precision and
resolution of the INSTA-TEST system, as given by the manufacturer, are presented in Table J. 1.
Table J.I. Precision and Resolution of INSTA-TEST System as Specified by the Manufacturer
Product Level
Precision
34.2 Jim
Resolution
10.2 urn
J.I.2 Equipment Setup and Test Preparations
The manufacturer did not specify the amount of tune usually required for installation of the
equipment and test preparations.
J.I.3 Test Protocol
The important features of the INSTA-TEST test method are summarized in Table J.2. Test
protocol requires that product be added to the tank at least 1 h prior to testing. For a precision
test, the tank must be approximately 95% full, hi order to avoid transient effects, the actual
differential pressure measurement is made approximately 2 min after the tube is opened. The
sensor's electronics have an integration period of 10 s.
A-47
-------
Table J.2. Important Aspects of the INSTA-TEST System Test Protocol
Test Duration Variable
Waiting Period after Addition 1 h
Test Type/Product Level Underfilled Tank/95% of tank capacity
J.I.4 Data Analysis
A thermally compensated volume change is computed from the change in pressure
resulting from the level changes in the open and closed tubes. The pressure change is converted
to volume using a previously determined calibration factor. Thermal compensation is performed
physically by the integration of temperature in the closed tube and, as a result, no direct
temperature measurements or calculations are performed.
J.I.5 Detection Criterion
A tank is declared leaking if the flow rate as determined by the test is not equal to 0.0 ml/h.
J.I.6 Mathematical Model
While the description provided above is sufficient to identify the key features of the
INSTA-TEST system as it was evaluated and to interpret the results of the performance
evaluation, it should be noted that a mathematical model of the system was developed based on
the more detailed description of the test method provided in [J-l].
J.2 Improving System Performance
No recommendations for improving system performance were made as part of this
evaluation.
A-48
-------
K.I The Leak Computer Test Method Description
The Leak Computer system, designed to function in an overfilled tank, measures changes
in product volume and product temperature in order to determine a volumetric flow rate.
Changes in product volume are determined by measuring the amount of product that is added or
removed to keep the product level in the fill tube constant. The amount of product added or
removed is determined by the change in weight (mass) of a supply of product maintained outside
the tank (approximately 13.2 L (3.5 gal)), while a control loop using two optical detectors
determines the time at which this addition or removal takes place.
K.I.I Instrumentation
An array of three thermistors, volumetrically weighted by a factor of 15%, 70%, and 15%,
respectively, is used to determine a weighted average of the product temperature.
The volume of the tank is determined from the tank manufacturer's strapping table, which
specifies that the coefficient of thermal expansion used in the thermal compensation calculations
for gasoline is 0.0012402/°C (0.000689/T), When a test is performed, the measured volume
changes are compensated for thermal expansion prior to calculating a leak rate. The precision
and resolution of the temperature and volume measurement systems as given by the
manufacturer are presented in Table K.I.
Table K.I. Precision and Resolution of the Leak Computer System as Specified by the
Manufacturer
Volume Measurement
Temperature Measurement
Precision
10.0 ml
0.00 17'C
Resolution
3.3ml
0.0005 °C
K.I.2 Test Protocol
Test protocol does not require a waiting period between the filling of the tank and the start
of a test, but the duration of a test must be at least 70 mm. Volume and temperature data are
digitally recorded at a rate of 120 samples per cycle. The cycle period, approximately 50-70 s in
length, remains constant for the duration of a test. The test is finished either when the standard
deviation of the leak rates is less than 0.189 L/li (0.05 gal/h) or when the following criterion is
satisfied 10 times consecutively: the standard deviation of 30 leak rates is less than half of the
last leak rate determined. Table K.2 lists the important aspects of the test protocol.
A-49
-------
Table K.2. Important Aspects of the Leak Computer Test Protocol
Sampling Period between 50 and 70 s (120 samples averaged
per 50 to 70-s cycle)
Test Duration 70 min (minimum)
Wait Period 0 h
Test Type/Product Level Overfilled Tank/Constant (Below Grade)
K.I.3 Data Analysis
Thennal expansion and contraction of the product is calculated from the volumetrically
weighted thermistors placed in the tank. The volume change is estimated using
ATV = CCVAT (K.2.1)
where the change in temperature, AT, is the volumetrically weighted value determined from the
three thermistors, the total volume, V, is the tank volume as determined from the tank chart
issued by the tank's manufacturer, and Ce is the coefficient of thermal expansion of the product.
Thermal volume changes calculated from temperature measurements are subtracted from
the volume changes determined by measuring the amount of product added or removed.
K.I.4 Detection Criterion
The manufacturer considers the test result to represent the actual leak rate with a 95%
confidence interval less than or equal to the leak rate. The tank is declared leaking if the leak
rate exceeds 0.189 L/h.
K.I.5 Mathematical Model
While the description presented in Section K.2 is sufficient to identify the key features of
the Leak Computer as it was evaluated and to interpret the results of the performance evaluation,
it should be noted that a mathematical model of the system was developed based on the more
detailed description provided in [K-l ]. After it had been validated, the model was used to
estimate the performance of the method.
K.2 Improving System Performance
The manufacturer of the Leak Computer system can improve performance by making
changes to the protocol, the detection criterion, and the instrumentation.
A-50
-------
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual performance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. A waiting period after
topping is required to minimize the impact of the (1) spatial inhomogeneities and (2) structural
deformation produced by the addition (or removal) of product to reach the level required to begin
the test. The effective waiting period for this method is determined by an analysis algorithm;
the effectiveness of this algorithm was not evaluated.
A waiting period of at least 2 to 3 h is required to minimize the spatial inhomogeneities in
temperature. This period is long enough to minimize the effects of a 7°C difference in
temperature between the product in the tank and the product used to top the tank. A longer
period may be required to minimize the deformation-induced product-level changes. This can
present operational problems, because the time required to minimze the effects of deformation in
some tanks may be 1 to 5 times (or more) longer than the tune required to wait for the
temperature effects to stabilize. The time constants of the Test Apparatus tanks, for example, are
2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or longer; the waiting
period would be longer if the deformation effects were large and more than three time contants
were required for them to become small compared to the smallest leak to be detected.
Since the temperature and deformation effects associated with topping are independent of
each other, the protocol and/or analysis algorithm use to minimize this effect must address each
disturbance. A single waiting period based on a maximum disturbance would adequately address
the temperature instability. However, this may not suffice for the deformation effect.
K.2.1 Protocol
It is recommended that the manufacturer reassess the ability of the waiting period
algorithm to differentiate between the effects of structural deformation and a leak.
K.2.2 Instrumentation
Performance improvement could be achieved by adding more thermistors to improve
spatial sampling of the vertical temperature field. Based upon studies of canonical methods
presented in Section 10 of the report, it is estimated that a 30% performance improvement could
be achieved by increasing the number of thermistors from three to five.
K.2.3 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-51
-------
-------
L.I The Leak-O-Meter Test Method Description
Leak-O-Meter is a precision tank testing device capable, according to the manufacturer, of
detecting leaks as small as 189 ml/h (0.05 gal/h). The apparatus, attached to the fill tube of a
tank, functions when dispensing operations have been suspended. For the tests described in this
report, the system was employed using the protocol described in [L-l] and summarized below.
L. 1.1 Instrumentation
The Leak-O-Meter system determines a thermally compensated flow rate by measuring
volumetric fluctuations and temperature. The precision and resolution of the Leak-O-Meter
measurement system, as given by the manufacturer, are presented in Table L. 1. Volumetric
fluctuations are measured directly with a flow transducer having a range of 60 L/h and located in
an extension of the fill tube. The transducer is calibrated by injection of a known amount of
liquid before and after each test, and has a reported response time of 5 s. The tank is overfilled
to a level approximately 1.22 m above grade. The sensor used to measure temperature consists
of a single thermistor from which readings are sampled digitally at 25-s intervals; these are
displayed on a chart recorder.
Table L.I. Precision and Resolution of the Leak-O-Meter System as Specified
by the Manufacturer
Volume Measurement
Temperature Measurement
Precision
0.00061 L/h
o.orc
Resolution
0.00378 L/h
o.oorc
L. 1.2 Equipment Setup and Test Preparations
The manufacturer indicated it would take approximately 45 min to install the
Leak-O-Meter equipment and calibrate the device. This assumes that the product level is
satisfactory for the conduct of a test.
L.I.3 Test Protocol
The important features of the Leak-O-Meter test method are summarized in Table L.2.
Incorporated within the test are a 12-h wait after product addition and a maximum test duration
of 129 min consisting of three 43-min tests. The test may be shorter than the nominal 129 min if
the data collected over the first 86 min agree within 10%. There is also a waiting period after
topping the tank which corresponds to the time it takes to install and set up the Leak-O-Meter
system, nominally 45 min.
A-53
-------
Table L.2. Important Aspects of the Leak-O-Meter Test Protocol
Sampling Period 25 s
Test Duration 129 min (maximum)
Waiting Period after Filling the Tank 12 h
Waiting Period after Topping Off 45 min (approximate)
Test Type/Product Level Overfilled Tank
L.I.4 Data Analysis
A temperature-compensated volume rate is computed from the product's flow rate and
temperature measurements. For each test, data are collected in three sets. Each set, following
identical procedures for data collection and analysis, is comprised of three individual tests during
which temperature data are monitored continually. A detailed flowchart of the data analysis
algorithm is given in [L-l]. The rate of temperature change is estimated from the average
temperature change measured over a 15-min period, during which only temperature data are
recorded. The data are then used to estimate the thermal rate of change for the entire set. The
rate of thermal volume change is estimated by the coefficient of thermal expansion (Ce) and the
volume of the tank (V,ank). Ce is determined by measuring both the specific gravity and the
temperature of the product, using ASTM tables D1250-80; tank volume is determined from the
manufacturer's tank chart. If the estimate of the thennal rate of change (TLR) differs by more
than 25% from the value determined during the test, the TLR value is updated. Flow-rate data
are collected for 6 min after the end of the 15-min period, and the average value is used as the
volumetric flow rate. After a 5-min wait, the same sequence is repeated a second and then a
third time. The three measured flow rates, subsequently compensated for thermal effects, are
then averaged to yield the leak rate for that set.
For each set, the thermistor is deployed at a different location in the tank. During Set A,
the thermistor is positioned at two-thirds of the tank height as measured from the bottom; during
Set B it is positioned at one-half the height of the tank; and during Set C at one-third. If the
thermally compensated leak rates determined during Set A and Set B do not agree within 10%,
then Set C is performed. If the leak rates determined for two of the three sets are not within 10%
of one another, the sequence must be repeated until the percentage criterion is satisfied. The leak
rate reported at that point is an average of the two closest leak rates. The thennal compensation
is estimated using
= CVAT (L.2.1)
A-54
-------
where the rate of change of temperature, AT, is estimated from the thermistor data; the total
volume of the tank, V, is estimated from a tank chart; and a value of the coefficient of thermal
expansion, Cc, for gasoline is estimated from the API tables.
L.I.5 Detection Criterion
A leak is declared if the measured temperature-compensated volume rate exceeds +/- 189
ml/h.
L.I.6 Mathematical Model
While the description provided in Section L.2 is sufficient to identify the key features of
the Leak-O-Meter as it was evaluated and to interpret the results of the performance evaluation,
it should be noted that a mathematical model of the system was developed based on the more
detailed description of the test method provided in [L-l]. After it had been validated, this model
was used to estimate the performance of the method.
L.2 Improving System Performance
The Leak-O-Meter system as defined in [L-1 ] can realize performance improvements with
changes to the system protocol and analysis procedure. By also making minor hardware
modifications, the system performance could be markedly unproved.
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual perfomiance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. The 0.75-h waiting period
after topping is too short to minimize the impact of the (1) spatial inhomogeneities and (2)
structural deformation produced by the addition (or removal) of product to reach the level
required to begin the test. A waiting period of at least 2 to 3 h is required to minimize the spatial
inhomogeneities in temperature. This period is long enough to minimize the effects of a 7°C
difference in temperature between the product in the tank and the product used to top the tank. A
longer period may be required to minimize the deformation-induced product-level changes. This
can present operational problems, because the time required to minimze the effects of
deformation in some tanks may be 1 to 5 times (or more) longer than the time required to wait
for the temperature effects to stabilize. The time constants of the Test Apparatus tanks, for
example, are 2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or longer;
the waiting period would be longer if the deformation effects were large and more than three
time contants were required for them to become small compared to the smallest leak to be
detected. Since the temperature and deformation effects associated with topping are independent
A-55
-------
of each other, the protocol and/or analysis algorithm use to minimize this effect must address
each disturbance. A single waiting period based on a maxunum disturbance would adequately
address the temperature instability. However, this may not suffice for the deformation effect.
L.2.1 Protocol
The following protocol changes are recommended:
o Sample the temperature data every 3 min
o Relevel the product at 15-min intervals or less and record the volume changes
o Wait 3 to 6 h or longer after topping off the tank, and maintain constant product
level during the waiting period
The performance of the Leak-O-Meter system as modeled is significantly affected by
deformation effects. Accurate measurement of volume changes in the fill tube requires
releveling of the product during a test and an adequate time for the deformation effects to
become negligible.
Releveling the product represents the most significant perfonnance improvement that can
be achieved by the Leak-O-Meter system. Releveling removes the deleterious effect of dynamic
structural deformation described in Section 6 of the report. Eliminating the dynamic structural
deformation would improve the performance of the system by approximately 100%.
The Leak-O-Meter system samples at only three locations in the tank and provides the
option of accepting data from only two of the three locations. In order to thermally compensate
to the degree claimed by the manufacturer, the tank must be sampled at more (at least five)
vertical locations in the tank for shorter periods of tune.
L.2.2 Analysis
The present method of estimating leak rates depends exclusively upon operator
intervention and interpretation of the collected data. The Leak-O-Meter system could benefit
greatly from the introduction of digital data collection and processing equipment. The flow
meter presently used in the Leak-O-Meter system appears to display a noisy signal. This
situation could also be improved if a digital-signal processing scheme were implemented.
It is recommended that the temperature-compensated volume rate be computed from the
data taken at the recommended higher sample rates using least-squares techniques. If four
volume measurements were used to estimate the volume-rate change and 20 temperature
measurements to estimate the thermal volume change, the perfonnance could improve by as
much as 100%.
A-56
-------
L.2.3 Instrumentation
Performance improvement could be achieved by adding more thermistors to improve
spatial sampling of the vertical temperature field. Based upon studies of canonical methods
presented in Section 10 of the report, it is estimated that a 30% perfonnance improvement could
be achieved by increasing the number of thermistors to five.
L.2.4 Detection Criterion
The probability of false alarm is detemiined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-57
-------
-------
M.l The LiquidManager Test Method Description
LiquidManager is a precision leak detection device capable, according to the manufacturer,
of detecting leaks as small as 38 ml/h (0.01 gal/h) in the absence of noise sources other than
thermal noise. This capability, when factored for other uncompensated noise sources, will allow
for the detection of leaks of 189 ml/h (0.05 gal/h). The unit is installed in an unsealed fill tube
from which any drop tube that might have been present has been removed.
M.I.I Instrumentation
The LiquidManager leak detection system measures the change in product level and
compensates for thermal fluctuations. The precision and resolution of the LiquidManager
measurement system as given by the manufacturer are presented in Table M. 1. The
LiquidManager measures fluid level by measuring the natural period of oscillation resulting from
the combined buoyancy and gravitational forces exerted on a probe ("displacer") that is
submerged to its midpoint in the product. The period is inversely related to the square root of the
total vertical force on the probe. The probe location is adjusted for fluid-level changes so as to
keep the period of oscillation constant. Product level is sampled at a frequency of 1 Hz.
The temperature measurements are made by raising and lowering a thermistor through the
product throughout the test. The initial and final temperature values, which are the only values
used in the leak rate calculations, are obtained by holding the thermistor stationary for 90 s at the
center of each of 17 equally spaced segments as it gradually ascends from the base of the tank to
the top. (Its initial position is actually 25.4 cm (10 in.) from the base of the tank, and its final
position is 25.4 cm from the top.) Temperature data are read at a sample rate of 8 Hz. The
temperature probe is said by the manufacturer to have a response time constant of 1 s.
Table M.l. Precision and Resolution of the LiquidManager System as Specified by the
Manufacturer
Product Level
Temperature
Precision
0.25 mm
o.oo rc
Resolution
0.15mm
0.0005 °C
M.1.2 Equipment Setup and Tesl Preparation
The installation of the LiquidManager precision tank testing device at the Edison UST
facility took up to 1.5 h. This assumes the product was at a satisfactory level for testing.
A-59
-------
M.I.3 Test Protocol
The important features of the LiquidManager system are summarized in Table M.2. The
LiquidManager method requires that the tank he filled to a level within the fill rube at least 12 h
prior to the start of the test. At the end of this 12-h period, product is added until a pressure of
just under 0.35 kg/cm2 (5 psi) is attained at the base of the tank, and at the end of another 1-h
waiting period the product level is lowered to obtain 0.28 kg/cm2 (4 psi) at the base of the tank;
both pressure levels are determined according to a set of relations developed by the
manufacturer. No additional product-level changes are required, and after a 15-min period has
elapsed, the test may start. Test duration for a 243.8-cm (96-in.) diameter tank is approximately
92 min; this comprises 16 min for temperature data collection prior to the 1 h of product-level
measurements, and 16 min at the end of these measurements.
Table M.2. Important Aspects of the LiquidManager System Test Protocol
Sampling Period 0.125 s
Test Duration 92 min
Waiting Period After Addition 12 h
Test Type/Product Level Overfilled Tank/Above Grade
Prior to the start of a test, LiquidManager checks the thermal stability of the tank. If the
standard deviation of the temperature measured over the 90-s period at each segment is greater
than 0.01°C, the test is delayed an additional 15 min.
M.1.4 Data Analysis
A temperature-compensated volume rate is computed from the product-level and
product-temperature measurements. The level measurements used for the calculations are an
average of the first and last 40 samples collected over the 1-h test period. As the product-level
sensor has a sampling rate of 40 Hz, the 40 sample segments of data represent the first and last
40 s of the 1-h test period. The temperature data, which are collected prior to and after the 1 h of
product-level sensor data, represent temperatures at 17 equally spaced points beginning 25.4 cm
from the tank bottom and ending 25.4 cm from tank top. The temperature sensor stabilizes at
each of the 17 locations for 90 s and collects data at an 8-Hz sample rate. An average of the last
30 samples, representing 3.75 s of data, of each 90-s period is used in the calculations. The two
sets of temperature data points are then volumetrically weighted and differenced to give an
estimation of the average overall temperature change of the product in the tank, AT. The thermal
expansion or contraction of the product, ATV is estimated using
A-60
-------
ATV = CCVAT
where V is the total volume of product in the tank and Ce is the coefficient of thermal expansion
for gasoline furnished by the manufacturer.
A height-to-volume (HtoV) calibration is conducted with a known volume at the
0.28-kg/cm2 level prior to and after each test. By using the known exposed surface area and the
HtoV factor obtained during calibration, an estimation of vapor pocket size is made. The
product level sensor data in conjunction with the HtoV conversion factor are used to estimate the
gross volume change.
M.1.5 Detection Criterion
A leak is declared if the measured temperature-compensated volume rate exceeds
+/- 189ml/h(0.05gal/h).
M.1.6 Mathematical Model
While the description provided in Section M.2 is sufficient to describe the key features of
the LiquidManager system as it was evaluated and to interpret the results of the perfonnance
evaluation, it should be noted that a mathematical model of the system was developed based on
the more detailed description of the test method provided in [M-l]. After it had been validated,
this model was used to estimate the perfonnance of the method.
M.2 Improving System Performance
The LiquidManager system can realize perfonnance improvements by changing the
protocol of the test.
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual perfonnance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. The 1-h waiting period
after topping is too short to minimize the impact of the (1) spatial inhomogeneities and (2)
structural deformation produced by the addition (or removal) of product to reach the level
required to begin the test. A waiting period of at least 2 to 3 h is required to minimize the spatial
inhomogeneities in temperature. This period is long enough to minimize the effects of a 7"C
difference in temperature between the product in the tank and the product used to top the tank. A
longer period may be required to minimize the defonnation-induced product-level changes. This
can present operational problems, because the time required to minimze the effects of
deformation in some tanks may be 1 to 5 times (or more) longer than the time required to wait
for the temperature effects to stabilize. The time constants of the Test Apparatus tanks, for
A-61
-------
example, are 2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or longer;
the waiting period would be longer if the deformation effects were large and more than three
time contants were required for them to become small compared to the smallest leak to be
detected. Since the temperature and deformation effects associated with topping are independent
of each other, the protocol and/or analysis algorithm use to minimize this effect must address
each disturbance. A single waiting period based on a maximum disturbance would adequately
address the temperature instability. However, this may not suffice for the deformation effect.
The performance of the LiquidManager system can be affected to varying degrees by
volume changes due to structural deformation. Because structural deformation effects vary
between tanks, an adequate quantitative estimate is difficult to ascertain. It is therefore
recommended that the LiquidManager utilize a method for maintaining a constant product level
during the test period while keeping other protocol requirements the same.
ML2.1 Protocol
The following protocol changes are recommended:
o Relevel the product at 15-min intervals or longer and record the volume changes
o Wait 3 to 6 h or longer after topping off the tank, and maintain constant product
level during the waiting period
Releveling the product represents the most significant perfonnance improvement that can
be achieved by the LiquidManager system. Releveling removes the deleterious effect of
dynamic structural deformation described in Section 6 of the report. Eliminating the dynamic
structural deformation would improve the performance of the system by approximately 80%.
M.2.2 Analysis
It is recommended that the temperature-compensated volume rate be computed from the
data taken at the higher sample rates using least-squares techniques. If four volume
measurements were used to estimate the volume-rate change and at least 20 temperature
measurements were used to estimate the thermal volume change, the performance could improve
by as much as 100%.
M.2.3 Detection Criterion
The probability of false alann is determined by the threshold. The manufacturer should
determine what probability of false alann is desired and adjust the threshold accordingly.
A-62
-------
N.I LMS-750 Test Method Description
The LMS-750 is a leak detection system capable, according to the manufacturer, of
detecting leaks as small as approximately 3.785 L/h (1 gal/h). The LMS-750 system measures
product level and temperature in order to determine a temperature-compensated leak rate. The
test length is constant, as is the sampling period.
N. 1.1 Instrumentation
The LMS-750 system uses two measuring devices to estimate a temperature-compensated
flow rate: product level as determined by means of the Pneumercator Direct Lift Model 2-412
level sensor, and product temperature as determined by two thermistors located at 33% and 66%
of the tank height.
Volumetric changes are determined by monitoring fluctuations in product level using a
float device. A data set containing 48 readings is collected every 26 s and recorded. The 48
readings are averaged, and the average recorded as the representative value of each data set.
Product temperature is recorded using two thermistors located at 33% and 66% of the tank
height (0.80 m and 1.6 m, respectively, from the bottom of the tank). As with product-level data,
temperature sensors are sampled 48 times every 26 s, and an average is calculated to represent
that data set.
The precision and resolution of the LMS-750 measurement system, as specified by the
manufacturer, are given in Table N.I.
Table N.I. Manufacturer's Precision and Resolution Claim for the LMS-750 Measurements
Level Measurement System
Temperature Measurement System
Precision
0.06 cm
0.2°C
Resolution
0.06 cm
0.2°C
N.I.2 Test Protocol
The important features of the LMS-750 test method are summarized in Table N.2. The test
protocol defined by the manufacturer in IN-11 and followed in this evaluation requires that the
product level be brought to approximately 90% of tank capacity a minimum of 2 h prior to
testing. The test, 6 h in length, is begun as soon as the equipment is installed.
A-63
-------
Table N.2. Important Aspects of the LMS-750 Test Protocol
Sampling Period 26 s
Test Duration 6 h
Waiting Period after Addition 2 h
Test Type/Product Level Underfilled Tank/90% capacity
N.I.3 Data Analysis
Product level and temperature are sampled 48 times every 26 s, and an average is
calculated to represent that data set. The average of each product-level data set is converted to a
volume by multiplying the height value by a scale factor (HtoV), which is determined from a
tank chart. This volume is compensated for thermal effects using the temperature data collected
during the test.
Thermal expansion or contraction of the product, ATV, is estimated from
ATV = CeVAT (N.2.1)
where Ce is the coefficient of thermal expansion, V is the volume of product in the tank, and AT
is the change in temperature.
Six temperature-compensated values from six data sets are used to perform a least-squares
linear fit. A coefficient of determination (the square of the correlation coefficient) is calculated
in order to determine whether the data collected for that group of six sets are acceptable or not.
These calculations are performed every hour, producing six leak rate values. An average
temperature-compensated value of the six data sets represents a 2.6-min group, if the data are
acceptable. Otherwise, the value for that group is replaced with the value of the most recent
group. The difference between the temperature-compensated volume at the beginning and end of
each hour results in the leak rate for that hour. The average of the six values at the end of the test
is reported as the final leak rate.
N.I.4 Detection Criterion
A tank is declared leaking if the flow rate as determined by the test is greater than a
threshold of+/- 3.785 L/h (1 gal/h). If the flow rate is less than the threshold and the coefficient
of determination is greater than 0.85, the tank is determined to be leaking. At these low flow
rates, no rigid specification of the leak rate is made but, rather, a recommendation is made to
perform a precision volumetric test to confirm the indication. If the leak rate is less than
3.785 L/h but is not zero, and if the coefficient of determination is less than 0.85, the test is
considered invalid.
A-64
-------
N.I.5 Mathematical Model
While the description provided above is sufficient to identify the key features of the
LMS-750 as it was evaluated, it should be noted that a mathematical model of the system was
developed based on the more detailed description of the test method provided in [N-l].
N.2 Improving System Performance
The LMS-750 system can realize significant performance improvements if instrumentation
capabilities are enhanced.
Changes to the test protocol and the data analysis algorithm could also result in an
improved ability to detect small leaks. These improvements, which may require substantial
modifications, should be approached from the standpoint that they will have to comply with
regulatory standards which may be promulgated in the future.
N.2.1 Instrumentation
The precision of both the temperature and height sensors can be significantly improved.
Based upon the calibration analyses perfonned during the Field Verification Tests, a volume rate
no smaller than 8.7 L/h could be detected with a PD of 0.99 and a PFA of 0.01. Improving the
height precision (or the volume rate attributable to height precision) by a factor of 2 could reduce
the minimum detectable leak rate by roughly half, bringing the expected performance into
relatively close agreement with the current claim of 3.875 L/h. This reduction in detectable leak
rate may also be achieved by raising the level at which the test is conducted so that the free
surface area (and thus the height-to-volume coefficient) is reduced by half. This practice would
not be inconsistent with conducting a test with the tank at full capacity, provided that a
reasonable degree of confidence in the height-to-volume coefficient can be maintained.
N.2.2 Test Protocol
Errors due to structural deformation and inadequate temperature compensation can be
minimized by increasing the waiting time between filling the tank and initiating a test.
Obviously, the optimum waiting time for structural deformation will be site-specific. However,
analyses conducted at the Test Apparatus suggest that thermal compensation will be difficult to
achieve sooner than 12 h after product addition.
The 2-h waiting period after topping may be too short to minimize the impact of the (1)
spatial inhomogeneities and (2) structural defonnation produced by the addition (or removal) of
product to reach the level required to begin the test. A waiting period of at least 2 to 3 h is
required to minimize the spatial inhomogeneities in temperature. This period is long enough to
minimize the effects of a 7°C difference in temperature between the product in the tank and the
A-65
-------
product used to top the tank. The 2-h waiting period is long enough to minimize the
temperature-induced effects for many testing situations. A longer period may be required to
minimize the deformation-induced product-level changes. This can present operational
problems, because the time required to minimze the effects of deformation in some tanks may be
1 to 5 times (or more) longer than the time required to wait for the temperature effects to
stabilize. The time constants of the Test Apparatus tanks, for example, are 2 to 3 h, which means
that the minimum waiting period would be 6 to 9 h or longer; the waiting period would be longer
if the deformation effects were large and more than three tune contants were required for them to
become small compared to the smallest leak to he detected. Since the temperature and
deformation effects associated with topping are independent of each other, the protocol and/or
analysis algorithm use to minimize this effect must address each disturbance. A single waiting
period based on a maximum disturbance would adequately address the temperature instability.
However, this may not suffice for the deformation effect.
IS.2.3 Analysis
The effect of the analysis algorithm employed by the LMS-750 as defined in [N-l] is
difficult to ascertain without performing the simulation, but a few generalizations can
nevertheless be made. By performing replacement of uncorrelated data as defined in [N-l], the
system may be removing data and possibly degrading performance by effectively reducing the
system resolution. It is recommended that the LMS-750 system not replace data, but rather
employ either an averaging or least-squares type of analysis over all data collected. The
manufacturer should also be aware that the algorithm for declaring whether the tank is leaking is
biased in that it declares that an inflow is not a leak.
IS.2.4 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-66
-------
O.I The MCG-1100 Test Method Description
The MCG-1100 is a volumetric leak detection system capable, according to the
manufacturer, of detecting leaks as small as 151 ml/h (0.04 gal/h). The MCG-1100 method
determines a thermally compensated leak rate in an overfilled tank. Product level is above grade
and is maintained constant during a test.
O.I.I Instrumentation
The MCG-1100 system utilizes a microprocessor-based data acquisition system to digitally
monitor product level and temperature. The precision and resolution of the MCG-1100
measurement system, as given by the manufacturer, are presented in Table O.I. The
product-level measurement system, situated outside the tank, consists of three interconnected
units: the level-control chamber, the fill chamber, and the by-pass chamber. The level-control
chamber is attached directly to the tank by means of a fill tube extension so that product may be
filled to a level above grade. Both the fill chamber and the by-pass chamber are connected by
means of tubing to the level-control chamber and electronically to the MCG-1100 electronics
box.
The level-control chamber is set up so that product is continually released into the by-pass
chamber at a predetermined rate. If the product level decreases, the float in the level-control
chamber opens a passage allowing product to enter from the fill chamber. A needle valve is
manually set to a predetermined minimum flow rate, allowing product to flow from the
level-control chamber to the by-pass chamber. The minimum flow rate is determined from
temperature measurements taken prior to the test.
Product level is measured both in the fill chamber and the by-pass chamber by a sensing
board, located on the outside the chambers. This board monitors the movement of a magnetic
float in response to level changes the chambers. Temperature is measured using five thermistors
located at the volumetric centers of five equal volumes.
Table (). 1. Precision and Resolution of MCG-1100 Measurements
Product Level
Temperature
Precision
3.17 mm
0.0028'C
Resolution
3.17 mm
0.00056'C
O.I.2 Equipment Setup and Test Preparations
The test crew typically took 1.5 h to set up the equipment and prepare for the first stage of
the test protocol. This assumes that the product level is satisfactory for the conduct of a test.
A-67
-------
O.I.3 Test Protocol
The important features of the MCG-1100 test protocol are summarized in Table O.2. Test
protocol requires that the tank be filled up to the base of the fill tube a minimum of 2 h prior to
testing. Upon the arrival of the test personnel at the site, the product level is brought above grade
in two stages. In the first stage, the tank is filled to 38 cm (15 in.) above grade. Product level is
then observed to assure that the tank has stabilized. The tank is considered stabilized if the level
does not drop below 10 cm (4 in.) above grade in the ensuing 30-min period. If it does, the level
is again raised to 38 cm above grade and monitored. This step is repeated until the level remains
above 10 cm for a period of 15 min. In the second stage, product is brought to a level 122 cm
(48 in.) above grade and is monitored for a minimum of 1 h. If the level drops below 107 cm
above grade, product is added to bring the level back to 122 cm. To complete the second stage,
the change in volume for each of three consecutive 15-min periods must be less than 0.3785 L
(0.1 gal). Otherwise, additional waiting periods in increments of 15 min ensue until such time as
the change-in-volume criterion has been satisfied. Once the second stage is completed, normally
within 1 to 3 h from the start of the first stage, the level is brought to 46 cm (18 in.) above grade.
A 15-min wait ensues, and the test may then commence.
Table O.2. Important Aspects of the MCG-1100 System Test Protocol
Sampling Period 1 h
Test Duration 1 h
Waiting Period after Addition 2 h
Waiting Period after Topping Off Variable
Test Type/Product Level Overfilled Tank/Constant (Above Grade)
O.I.4 Data Analysis
A thermally compensated volume rate is computed from the product-level and
product-temperature measurements sampled at the beginning and end of the 1-h test. Each
sample comprises five measurements averaged over a 2-s period. The measured level changes
are converted into volume changes by multiplying by the cross-sectional area of each chamber.
The thermal expansion and contraction of the product, ATV, is estimated using
ATV = CCVAT (0.2.1)
where the rate of change of temperature, AT, is estimated from the volumetrically weighted
average of five thermistors; the total volume, V, of the product in the tank is determined using a
A-68
-------
tank chart; and a value of the coefficient of thermal expansion, Ce, for the product is determined
using the specific gravity measured with a hydrometer and adjusted to 15.6°C
(60°F) using API Table VI-B. The leak rate, LR, is calculated from
LR = AVF-AVB-ATV (O.2.2)
where AVF is the rate of change of volume in the fill chamber, AVB is the rate of change of
volume in the bypass chamber, and ATV is the thermal expansion and contraction of the product
as calculated in Eq. (O.2.1).
The flow rate from the level-control chamber to the by-pass chamber is detennined using
thermal data obtained during the first 15 min of the 30-min period prior to the test. The flow rate
is manually set to a rate 20% greater than the expected thermal change as estimated using
Eq. (O.2.1) for the 15-min period. The flow rate setting is not to be less than 75.7 ml/h.
The test is successful if the set flow rate to the by-pass chamber is not less than the
measured thermal change. Unless this criterion is satisfied, the test is not valid and must be
performed again.
O.I.5 Detection Criterion
A leak is declared if the measured temperature-compensated volume rate for the test period
exceeds +/- 151 ml/h (0.04 gal/h).
O.I.6 Mathematical Model
While the description provided in Section O.2 is sufficient to identify the key features of
the MCG-1100 as it was evaluated and to interpret the results of the performance evaluation, it
should be noted that a mathematical mode of the system was developed based on the more
detailed description of the test method provided in [O-1 ]. After it had been validated, this model
was used to estimate the performance of the method.
O.2 Improving System Performance
The two major problems that exist in the MCG-1100 system, as it was evaluated, can be
remedied through protocol adjustments and require no changes to the instrumentation. An
improvement in performance can also be achieved by making alterations to the analysis.
O.2.I Protocol
The following protocol changes are recommended:
o Wait 6 to 12 h or longer after product delivery
o Sample the temperature data every 3 min
A-69
-------
o Relevel the product at 15-min intervals or less and record the volume changes
o Wait 3 to 6 h or longer after topping off the tank and maintain constant product
level during the waiting period
The 2-h time period between product delivery and the start of a test is not long enough for
the tank to become thermally quiet, thus precluding adequate thermal compensation.
Experiments at the Test Apparatus have shown that a period of up to 12 h is required before a
system can adequately perform thermal compensation. It is evident, therefore, that significant
improvement could be obtained by lengthening the waiting period before a test.
The largest error present in the MCG-1100 system is the systematic error caused by
structural deformation. Raising the product level and then lowering it so that deformation effects
cancel out may work for some tanks, but because there is no systematic approach to verifying
that the cancellation has occurred, the system is very susceptible to defonnation effects arising
from the large level changes. If the protocol were adjusted so that product level was raised once
and then kept constant, it would be necessary only to detennine when defonnation effects from
the initial product-level change had subsided. A detailed analysis of the data over a sufficient
period of time would be required to ascertain the relaxation constant of the tank so that an
appropriate waiting period could be defined. Available data suggest that a period of 4 h would
be required for stability to be achieved in the Test Apparatus tank. Distortion of the tank's
temperature field, caused by topping off the tank, could also be diminished by an appropriate
waiting period.
O.2.2 Analysis
It is recommended that the temperature-compensated volume rate be computed from the
data taken at the recommended higher sample rates using least-squares techniques. If four
volume measurements were used to estimate the volume-rate change and 20 temperature
measurements to estimate the thennal volume change, the performance could improve by as
much as 100%.
O.2.3 Defection Criterion
The probability of false alarm is detennined by the threshold. The manufacturer should
detennine what probability of false alann is desired mid adjust the threshold accordingly.
A-70
-------
P.I The Mooney Test Method Description
The Mooney Leak Detection System estimates a temperature-compensated volume rate
from measurements of product level and product temperature in an overfilled storage tank. Test
length (1 h) and data sample rate are constant.
P.I.I Instrumentation
Volumetric changes are determined by measuring product height changes with a
mechanic's ruler. A known volume of product (0.47 L (0.125 gal)) is added to the tank before
the test in order to empirically determine the scale factor relating change of product level to
change of volume. Height measurements are then made at 15-min intervals by inserting the ruler
into the product and recording the results.
Temperature is measured by five thermistors placed at the volumetric vertical centers of
five equal volumes. The thermistors are checked in a bath to determine that the resistance values
are in the right range, but they are not calibrated. The ohmmeter is checked with a 10,000-Q
plug.
The precision and resolution of the Mooney measurement system as given by the
manufacturer are presented in Table P.I.
Table P.I. Precision and Resolution of the Mooney System as Specified by the Manufacturer
Height Measurement
Temperature Measurement
Precision
0.05 cm
0.0014°C
Resolution
0.05 cm
1 ft at 720 ft/°C
Both volumetric and temperature data are sampled at 15-min intervals. The product height
is measured with the mechanic's ruler twice at each 15-min mark, with the two samples taken 1
min apart; after each sample has been read by two test personnel, the four readings are averaged.
The temperature data are read as the resistance from a digital ohmmeter.
P. 1.2 Test Protocol
Test protocol requires a 12-h wait between the filling of the tank and the commencement
of the test. It may be necessary to add product to the tank in order to raise the liquid level to the
top of the fill tube. In such cases, an additional waiting period of 5 min per 2.54 cm (1 in.) of
added head is required after topping off the tank. This waiting period is not to exceed 30 min
A-71
-------
unless the change in product level is greater than 0.318 cm (0.125 in.) during the last 10 min of
the waiting period, in which case an additional waiting period of 15 min is required. Salient
aspects of the test protocol are presented in Table P.2.
Table P.2. Important Aspects of the Mooney System Test Protocol
Sample Period 15 min
Test Length 1 h
Number of Samples per Test 5
Test Start after Filling Tank 12 h
Test Start after Topping Off Tank Variable (5 to 54 min)
Test Type Overfilled Tank
P. 1.3 Data Analysis
Data are obtained by sampling every 15 min for 1 h, thereby collecting five data points per
sensor per test. Only product level and temperature data are collected. Product level is sampled
twice at the 15-min mark. Temperature and product-level data are differenced to obtain the
change in quantity of fluid per 15-min sampling interval.
Compensation for evaporation effects is perfonned by measuring the evaporation of
product from an evaporation cup placed in the fill tube. The amount of change in product level
within the evaporation cup is used to compensate for evaporation effects.
The coefficient of thermal expansion (C,) is obtained from the station owner and is verified
by determining Ce from a specific gravity measurement and the manufacturer's tables; the
volume of the tank is determined from a tank chart. The change in volume due to thermal effects
is calculated and is then subtracted from the measured volume change for each period, as shown
in the equation
Thermal volume change = ATV = ATCeV (P.2.1)
where AT is the change in temperature, Ce is the coefficient of thermal expansion, and V is the
volume of product in tank at beginning of test. The average compensated change in volume per
15-min period is shown by
A Vcomp = A Vn_un:d -A TV (P .2.2)
v AVmmp(il
nvMf= I —--^ (P.2.3)
t=1.4 4
A-72
-------
P.1.4 Detection Criterion
If the leak rate as determined by the test is within +/- 0.076 L/h (0.02 gal/h) of the 0.19-L/h
(0.05-gal/h) leak rate criterion, it is possible that the tank is leaking, and another test is required
for confirmation. If the average of the two tests is greater than 0.19 L/h, the tank is considered
leaking.
P. 1.5 Mathematical Model
While the description provided in Section P.2 is sufficient to identify the key features of
the Mooney System as it was evaluated and to interpret the results of the performance evaluation,
it should be noted that a mathematical model of the system was developed based on the more
detailed description of the test method provided in [P-l J. After it had been validated, this model
was used to estimate the performance of the method.
P.2 Improving System Peformance
The Mooney Leak Detection System as defined in [P-1 ] can realize substantial
performance improvements without hardware modification.
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual performance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. The 5-min to 54-min
waiting period after topping is too short to minimize the impact of the (1) spatial
inhomogeneities and (2) structural defonnation produced by the addition (or removal) of product
to reach the level required to begin the test. A waiting period of at least 2 to 3 h is required to
minimize the spatial inhomogeneities in temperature. This period is long enough to minimize
the effects of a 7°C difference in temperature between the product in the tank and the product
used to top the tank. A longer period may be required to minimize the deformation-induced
product-level changes. This can present operational problems, because the time required to
minimze the effects of defonnation in some tanks may be 1 to 5 times (or more) longer than the
time required to wait for the temperature effects to stabilize. The time constants of the Test
Apparatus tanks, for example, are 2 to 3 h, which means that the minimum waiting period would
be 6 to 9 h or longer; the waiting period would be longer it the defonnation effects were large
and more than three time contants were required for them to become small compared to the
smallest leak to be detected. Since the temperature and defonnation effects associated with
topping are independent of each other, the protocol and/or analysis algorithm use to minimize
this effect must address each disturbance. A single waiting period based on a maximum
disturbance would adequately address the temperature instability. However, this may not suffice
for the deformation effect.
A-73
-------
P.2.1 Protocol
The following protocol changes are recommended:
o Sample the temperature data at least every 3 min
o Relevel the product at 15-min intervals or less and record the volume changes
o Wait 3 to 6 h or longer after topping off the tank, and maintain constant product
level during the waiting period
Releveling the product represents the most significant perfonnance improvement that can
be achieved by the Mooney Leak Detection System. Releveling removes the deleterious effect
of dynamic structural deformation described in Section 6 of the report. Eliminating the dynamic
structural deformation would improve the performance of the system by approximately 80%.
P.2.2 Analysis
It is recommended that the temperature-compensated volume rate be computed from the
data taken at the recommended higher sample rates using least-squares techniques. If four
volume measurements were used to estimate the volume-rate change and 20 temperature
measurements to estimate the themnal volume change, the performance could improve by as
much as 100%.
P.2.3 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-74
-------
Q.I The OTEC Test Method Description
The OTEC Leak Sensor is a volumetric leak detection system capable, according to the
manufacturer, of detecting leaks as small as +/-(). 189 L/h (0.05 gal/h) that occur in the walls or
product distribution lines of an underground storage tank. For the performance results described
in this report, it was employed as a precision (volumetric) leak detection test operating in an
underfilled tank.
Q.I.I Instrumentation
The OTEC Leak Sensor uses two measuring devices, a float and a detector rod, both of
which measure product level, to estimate a temperature-compensated flow rate.
Calculations show that the buoyant force acting on a partially submerged object will be
independent of changes in liquid density if the ratio of displaced volume to displaced surface is
equal to the ratio of total liquid volume to total liquid surface. A float that is tapered at the upper
end can be submerged so as to match the volume-to-surface ratio in a horizontal cylindrical tank
filled to any level. Changes in buoyant force on such a float will therefore reflect changes in
tank inventory, since the buoyant force is independent of density.
The position of the conical float, which is measured over a period of time, is related to the
volume changes by means of the phenomenon that supports a negatively buoyant member in a
liquid. The float assembly is dependent on the size and type of the tank. A conical float shell is
selected from a group of floats of varying length suitable for a variety of tank sizes. The conical
float's length and weight are chosen based on the diameter of the tank, the height of product in
the tank, and the density of the test product. These measured values are determined prior to the
test.
The detector-rod assembly consists of a photoelectric cell and the conical float attachment.
The movement of the conical float forces an ink solution into or out of the photocell, and the
resulting change in light transmittance in the photocell causes a voltage drop across the cell.
This voltage change, which is a function of the product-level change, is measured by a voltmeter
that has been calibrated prior to the test.
Volumetric changes are determined by monitoring product height changes with the conical
float. The conical float senses these liquid-level changes in the tank, and the detector rod
measures the relative position of the float with respect to the photocell. The data are recorded on
a strip chart.
A-75
-------
The manufacturer's claimed precision and resolution of these measurements are
summarized in Table Q.I, and were experimentally verified at the UST Test Apparatus in
Edison, New Jersey during the Field Verification Tests.
Table Q.I. Manufacturer's Precision and Resolution Claim for the OTEC Sensor
Measurements
Precision Resolution
Product Level 0.0001 cm 0.0005 cm
Q.I.2 Equipment Setup and Test Preparations
The manufacturer indicated that it would take approximately 1 h to install the equipment
and prepare for the start of the test. This assumes that the product level is satisfactory for the
conduct of a test.
Q.I.3 Test Protocol
The important features of the OTEC test protocol are summarized in Table Q.2. Test
protocol requires that the tank be filled to 50% to 100% of capacity. There is a waiting period of
at least 4 h between the filling of the tank and the start of a test (i.e., calibrations), so as to allow
the temperature of the product in the tank to stabilize. If the tank is not filled to capacity, a
1/3-hp pump is used to spray product into the vapor space in order to saturate it during the 1-h
period immediately preceding data collection. The 1-h period is the approximate time required
to complete this procedure in a 30,280-L (8,000-gal) tank. Instrumentation warm-up and
calibration also take place during this time. Calibration and testing continue at 1-h intervals until
two consecutive readings give the same calculated (leak rate) result, or until a 3-h maximum test
period has been completed.
Table Q.2. Important Aspects of the OTEC Test Protocol
Sampling Period 1 h
Test Duration 3 h (maximum)
Waiting Period after Addition 4 h
Test Type/Product Level Underfilled Tank/50% to 100% of capacity
Q.I.4 Data Analysis
A thermally compensated volume change is computed from the change in the height of a
weighted float immersed at the null point of the tank where thermal changes are exactly offset by
A-76
-------
buoyancy forces. Data are obtained by sampling once per hour over the duration of the test. The
voltage change over the 1-h period is converted to volume by multiplying by a conversion factor
determined from the calibrations made before the test.
A tank chart, supplied by the tank manufacturer, is used to convert product-level
measurements to volume. Height-to-volume conversion is accomplished by using linear
interpolation between pairs of data points tabulated at 2.54-cm (1-in.) intervals of tank depth.
All tests are considered successful providing that data are collected for the entire duration
of the test period, and that the appropriate initial waiting period is observed. A test is considered
successful when two consecutive 1-h tests give the same calculated result, or when the 3-h
maximum test period has been completed. The reported leak rate is that computed over the last
1-h testing period.
Q.I.5 Detection Criterion
A leak is declared if the thermally compensated volume rate derived for the test period
exceeds a threshold of+/-0.189 L/h (0.05 gal/h). If the flow rate is less than the threshold, the
tank is declared nonleaking.
Q.1.6 Mathematical Model
While the description provided in Section Q.2 is sufficient to identify the key features of
the OTEC system as it was evaluated, and to interpret the results of the performance evaluation,
it should be noted that a mathematical model of the system was developed based on the more
detailed description of the test method provided in [Q-1J. After it had been validated, this model
was used to estimate the performance of the method.
Q.2 Improving System Performance
The OTEC system was not evaluated because the above-ground tanks used to heat and
cool product was left open during the Field Verification Tests. Nevertheless, a systematic
investigation of the OTEC system should be made. There are two sources of error that should be
investigated: the implementation of the float system for temperature-compensation and the effect
of spraying product into the vapor space before the test. Theoretically, the float system will be
subjected to errors whenever the rate of change of temperature is not vertically uniform.
Experimental data taken during the ambient noise experiments at the Test Apparatus suggest that
any disturbance to the vapor space during a partially filled test can degrade performance unless
these efects are allowed to subside. The effect of spraying product into the vapor space was not
evaluated.
A-77
-------
-------
R.I The PACE Test Method Description
The PACE Leak Tester is a volumetric test capable, according to the developer, of
detecting leaks as small as 0.237 L/h (0.063 gal/h). The PACE apparatus is temporarily installed
at the fill hole of a tank. The test is normally conducted only when dispensing operations have
been suspended. For the tests described in this report, the protocol in [R-1J was employed.
K.I.I Instrumentation
The PACE Leak Tester system uses two measuring devices to estimate a
temperature-compensated flow rate: product level as determined by means of a dip-tube
manometer, and product temperature as determined by three temperature probes.
Volumetric changes are determined by monitoring product-level changes in a dip tube.
This measuring device consists of two units: a dip tube mounted in the tank, and a measuring
rule (dipstick) located inside the dip tube. The dip tube, measuring approximately 150 cm in
length and 2.54 cm in diameter, is welded into a cap that has been threaded to fit the fill tube in
such a way that the dip tube extends approximately 5 cm above the top of the fill tube. The dip
tube extends below the product level in the tank, and product-level changes are measured on the
dipstick.
Once the tank has been filled to within 30 to 45 cm of the top of the fill tube, all vents and
openings to the tank are sealed except for the fill hole fitted with the dip tube. This creates a
vapor space between the product level and the cap. The dip tube now provides the only access to
the storage tank during testing, and sealing the system in this manner causes the dip tube to
behave like a manometer. Any product-level changes within the tank produce an amplification
of the height change in the dip tube, a phenomenon due to the pressure difference between the
atmosphere and the vapor space in the fill tube.
To determine height changes, product level is monitored by means of the dipstick. This is
accomplished by notching the dipstick and measuring the distances between notches at the
conclusion of the test.
Temperature is measured with three thermocouples (probes) placed at the vertical centers
of three equal volumetric slabs at the dip tube location.
A height-to-volume (HtoV) calibration, performed just before the start of a test, is done by
removing a known volume of product from the fill tube (0.237 L) and measuring the
corresponding height change.
The stated precision and resolution of the measurement systems used in the PACE Leak
Tester are summarized in Table R.I.
A-79
-------
Table R.I. Precision and Resolution of PACE Measurements
Product Level Measurement
Temperature Measurement
Precision
0.3 1 8 cm
().<)06°C
Resolution
0.159cm
0.006'C
R.1.2 Equipment Setup and Test Preparations
Equipment deployment in the tank, because of the nature of the test, requires extensive
purging and bleeding of vapor from the tank risers. Normal equipment installation requires
approximately 1 h, after which calibration checks and testing may be initiated. This assumes that
a product level satisfactory to testing has been established.
R.I.3 Test Protocol
The important features of the PACE Leak Tester are summarized in Table R.2. The test is
carried out in a storage tank that has been overfilled into the fill tube. It consists of three
half-hour segments (experiments), resulting in a total test duration of 1.5 h.
An experimental leak rate (flow rate) is obtained for each experiment (i.e., each segment).
At the conclusion of the test (i.e., 1.5 h from the start), the three experimental flow rates are
statistically analyzed to arrive at a flow rate for the test. The test duration, experiments, and
sampling intervals are constant.
Test protocol requires a wait of at least 12 h between the filling of the tank and the start of
a test. Prior to testing, the tank is filled to a level within 35 to 40 cm of the top of the fill tube.
Table R.2. Important Aspects of the PACE Leak Tester
Sampling Period 30 min
Test Duration 90 min
Waiting Period After Addition 12 h
Test Type/Product Level Overfilled Tank/Below grade
R.I.4 Data Analysis
A temperature-compensated volume rate is computed from the measured height and
temperature data. The measured height changes are converted to gross volume changes by the
calibration tests which are performed prior to each 0.5-h experiment. For each experiment, the
thermal volume change calculated from the measured temperature change is computed. The
thermal expansion and contraction of the product, ATV, is estimated from
A-80
-------
where the rate of change of temperature, AT, is estimated from the three thermistor arrays; the
volume, V, of the product in the tank is estimated from the developer's tank chart; and the value
of the coefficient of thermal expansion of the product, Ce, is 0.00108/°C (0.00060/T). This
volume is then subtracted from the gross volume change to obtain a temperature-compensated
volume rate. The procedure is repeated for each of the three experiments. The results of the
three experiments are then analyzed statistically to determine whether the test is successful and,
if so, the mean and standard deviation of the leak rate.
R.I.5 Detection Criterion
A leak is declared if the mean flow rate (i.e., the temperature-compensated volume rate) is
greater than 0.237 L/h. If the flow rate is less than this value, the tank is declared nonleaking.
R.I.6 Mathematical Model
While the description provided in Section R.2 is sufficient to identify the key features of
the PACE Leak Tester as it was evaluated and to interpret the results of the performance
evaluation, it should be noted that a mathematical model of the system was developed based on
the more detailed description of the test method in \R-l]. After it had been validated, the model
was used to estimate the performance of the method.
R.2 Improving System Performance
The PACE Leak Tester system defined in [R-l j can realize performance improvements.
These improvements are discussed in terms of changes to the instrumentation, protocol and data
analysis.to not appear to validate these predictions.
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual perfonnance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. The 0-h waiting period
after topping is too short to minimize the impact of the (1) spatial inhomogeneities and (2)
structural deformation produced by the addition (or removal) of product to reach the level
required to begin the test. A waiting period of at least 2 to 3 h is required to minimize the spatial
inhomogeneities in temperature. This period is long enough to minimize the effects of a 7°C
difference in temperature between the product in the tank and the product used to top the tank. A
longer period may be required to minimize the deformation-induced product-level changes. This
can present operational problems, because the time required to minimze the effects of
defonnation in some tanks may be 1 to 5 tunes (or more) longer than the time required to wait
for the temperature effects to stabilize. The time constants of the Test Apparatus tanks, for
A-81
-------
example, are 2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or longer;
the waiting period would be longer if the deformation effects were large and more than three
time contants were required for them to become small compared to the smallest leak to be
detected. Since the temperature and deformation effects associated with topping are independent
of each other, the protocol and/or analysis algorithm use to minimize this effect must address
each disturbance. A single waiting period based on a maximum disturbance would adequately
address the temperature instability. However, this may not suffice for the deformation effect.
R.2.1 Instrumentation
The largest source of error is the volumetric noise created by excessive structural
deformation. This is a direct result of the small diameter of the fill tube. While decreasing the
tube diameter tends to increase the gain of the height measurement, it also increases the
magnitude of changes in the hydrostatic head, which increases the structural deformation and
volumetric noise. At some point, the volumetric noise increases so much that the gain in the
height measurement does nothing to increase the performance of the system.
If the developer were to use a larger tube diameter, such as 7.6 cm, volumetric noise would
be greatly reduced, while an acceptably small level of instrumentation noise would still be
maintained. Using the parameters from Section R.5.3.1 with a 7.6-cm diameter tube, the
standard error in the volume measurement is 7.29 ml/h. This gives a minimum detectable leak
rate (based on instrumentation) of 163 ml/h, 2 ml/h greater than with the 2.54-cm diameter tube.
Although, due to the same interaction, error will still be present, the developer can eliminate or
minimize it by keeping product level constant. This approach is discussed in the body of this
report.
Performance improvement could be achieved by adding more thermistors to improve
spatial sampling of the vertical temperature field. Based upon studies of canonical methods
presented in Section 10 of the report, it is estimated that a 30% perfonnance improvement could
be achieved by increasing the number of thermistors from three to five.
R.2.2 Protocol
The following protocol changes are recommended:
o Sample the temperature data every 3 min
o Relevel the product at 15-min intervals or less and record the volume changes
o Wait 3 to 6 h or longer after topping off the tank, and maintain constant product
level during the waiting period
A-82
-------
Releveling the product represents a significant performance improvement that can be
achieved by the PACE Leak Tester. Releveling removes the deleterious effect of dynamic
structural deformation described in Section 6 of the report. Eliminating the dynamic structural
deformation would improve the performance of the system by approximately 80%.
The PACE Leak Tester system could be further improved by increasing the sample rate for
temperature measurements. This would ensure that internal waves in the tank were not being
aliased into the measurement. Sampling temperature every 3 min is recommended in the case of
7-min internal waves.
In addition to more frequent temperature sampling, there could be an extra waiting period
of 4 h after the height calibrations are performed. Topping off the tank prior to testing may
cause the local temperature field (below the fill tube) to become unrepresentative of the rest of
the tank. This requires an additional wait of 4 h after topping off the tank.
R.2.3 Analysis
It is recommended that the temperature-compensated volume rate be computed from the
data taken at the recommended higher sample rates using least-squares techniques. If four
volume measurements were used to estimate the volume-rate change and 20 temperature
measurements to estimate the thermal volume change, the performance could improve by as
much as 100%.
R.2.4 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-83
-------
-------
S.I The Petro Tite Test Method Description
The Petro Tite leak detection system uses two measuring devices to estimate a
temperature-compensated flow rate: product level as determined by means of a graduated
cylinder, and product temperature as determined by a temperature probe. For a 30,000-L
(8,000-gal) storage tank, the testing period consists of four phases with a total duration of
400 min. Test duration and sampling intervals are constant. The test is carried out at two levels
in a storage tank that has been overfilled into the fill tube extender so that the product level is
approximately 70 cm above grade in the first level and 30 cm above grade in the second. These
levels are dependent on the height of the water table and relative height of the fill tube extender
(standpipe) above grade. If the water table is high enough that it exerts pressure on the tank,
enough product is added to bring the level in the fill tube to the equivalent of dry soil conditions.
S.I.I Instrumentation, Equipment Setup and Test Preparations
The precision and resolution of the Petro Tite measurement system as given by the
manufacturer are presented in Table S.I.
Table S.I. Precision and Resolution of the Petro Tite System as Specified by Manufacturer
Precision Resolution
Level Measurement 0.00095 L 0.0189 L
Temperature Measurement 0.000031 'C 0.000031 °C
There are four phases in a Petro Tite Test: (1) create circulation in the product to produce a
uniform temperature throughout the tank; (2) sample the product to obtain a coefficient of
thermal expansion; (3) perform the high-level test; and (4) perform the low-level test.
13.1.1 Phase 1 - Circulation
During the first phase of the Petro Tite test, a circulating (jet) pump draws product by
means of a suction tube from an area at least 15 cm (6 in.) below the top of the tank. The
product is discharged, under pressure of approximately 1.76 kg/cm2 (25 psi), through a discharge
hose into sections of tubing which have been coupled together to form an outlet nozzle at the
bottom of the tank. This nozzle is adjusted so as to be above any water at the bottom of the tank
and below any drop tube that may be present. The nozzle is directed at a 45° angle upwards from
the center line of the longitudinal axis of the tank. This suction and nozzle system creates a
vortex-like swirling motion in the tank in an attempt to produce a uniform (homogeneous)
temperature throughout its contents. At the same time, the operator is making visual
observations for above-grade and gross system leakage. The duration of this phase is based on
A-85
-------
the viscosity and volume of product in the tank, i.e., 5 to 8 min per 3,800 L (approximately 5 min
for light fuels such as gasoline and 8 min for heavy fuels such as diesel) or until predictable
thermal readings are achieved. After the completion of Phase 1, circulation continues throughout
the duration of the test.
13.1.2 Phase 2 - API Sample
The second phase requires the operator to obtain a sample of the product, which is tested
for the purpose of determining its coefficient of expansion. During this phase the jet pump is
shut down for approximately 30 s to prevent air from being sucked in by the pump while the
sample is being removed through a bleed valve.
13.1.3 Phase 3 - High-Level Test
The third phase requires the operator to take periodic thermal and product-level
measurements based on the diameter of the tank, as shown in Table S.2. During this phase the
operator is making visual observations for changes or fluctuations in product volume due to
structural deformation, thermal stabilization due to circulation (Phase I) and/or vapor pockets.
The sampling interval is every 15 min, so for a large tank requiring 12 measurement readings,
Phase 3 could last as long as 3 h.
Table S.2. Number of Readings Required for High-Level Test
Tank Diameter (m)
1.22
1.63
1.83
2.13
2.44
2.59
2.74
3.05
3.25
Number ol
4
4-6
6
6-8
8
Readings
8-10
10
10-
12
2
13.1.4 Phase 4 - Low-Level Test
At the start of the low-level test, which is Petro Tite's precision test, the product level in
the fill tube extender is dropped to approximately 30 cm above grade, with product volume and
thermal changes being recorded every 15 min. Volumetric changes are determined by
A-86
-------
monitoring the product height changes in the fill tube extender. This is done by means of a
3.785-L (1-gal) graduated cylinder that measures the amount of product added or drained to
maintain a constant product level.
S.I.2 Test Protocol
Test protocol requires that the tank be filled to a level above grade so as to achieve the
same testing conditions as for dry soil. The procedure for achieving the correct height is based
on the external pressure exerted on the tank by the surrounding water table; nominal testing
pressure is 0.28 kg/cm2 (4 psi). There is no waiting period between the filling of the tank and the
commencement of a test. Important aspects of the protocol are shown in Table S.3.
Table S.3. Important Aspects of the Petro Tite Test Protocol
Sampling Period
Duration of Equipment Setup
Duration of Phase 1
Duration of Phase 2
Duration of Phase 3
Duration of Phase 4
Total Test Duration
15 min
15 min
Variable within 5 to 8 min per
3,800 L
Variable
60 to 180 min
60 min
400 min
Number of Samples per Sensor per Experiment 2
Test Start after Filling Tank 0 min
Waiting Time after Product Addition 0 min
Test Type/Product Level* Overfilled Tank/Constant
'Product level is kept constant during any one phase of testing
S.I.3 Data Analysis
A thermally compensated volumetric flow rate is computed from the product-level and
product-temperature measurements obtained by sampling once every 15 min. One data point is
collected for each sensor at the start of the test and one at the end. Temperature and volume are
differenced to obtain the change in volume and change in temperature per sampling interval,
with the result representing the total change in temperature and volume through the duration of
the experiment.
A-87
-------
The coefficient of thermal expansion is determined by the operator from API gravity
tables, and the tank volume from strapping charts. Change in volume is calculated according to
the equation
Thermal volume change = ATV = CCVAT (S.2.1)
where Ce is the coefficient of thermal expansion of the test liquid, V is the volume of product in
the tank at the start of the test, and AT is the change in temperature.
The flow rate reported as the test result is the sum of four consecutive experimental flow
rates, each of which represents the thermally compensated volume change divided by the
duration of the experiment as shown in the equation
AV-ATV
LR = -—--- (S.2.2)
where LR is the flow rate, AV is the change in product volume, and At is the duration of the
experiment (15 min).
S.I.4 Detection Criterion
A tank is declared leaking if the flow rate as determined by the test is greater than a
threshold of-0.189 L/h (0.05 gal/h). If the flow rate is greater than the threshold and is also
positive, the test is considered invalid, since the nature of the test design precludes the detection
of inflowing (positive) leaks.
S.I.5 Mathematical Model
While the description provided in Section S.2 is sufficient to identify the key features of
the Petro Tite system as it was evaluated and to interpret the results of the performance
evaluation, it should be noted that a mathematical model of the system was developed based on
the more detailed description of the test method provided in [S-l |. After it had been validated,
this model was used to estimate the performance of the system.
S.2 Improving System Performance
The Petro Tite system could realize performance improvements if alterations were made to
the test protocol and data analysis. No major design modification is required.
S.2.1 Protocol
The manufacturer could improve the performance of the Petro Tite system by lengthening
the high- and low-level tests. Structural deformation of the tank can cause testing errors. The
A-88
-------
duration of the high-level tests is too short, unless the tank elasticity constant of the tank is small,
as it was for the Test Apparatus tanks. This is also true of the low-level tests. If the tank
elasticity constant were larger, as may be the case with other tanks, deformation could affect the
test results. The effects of structural deformation had only minimal impact on this evaluation.
S.2.2 Data Analysis
Using least-squares techniques to estimate the leak rate would maximize the information
obtained from the 15-min sample interval.
S.2.3 Detection Criterion
The probability of false alann is determined by the threshold. The manufacturer should
determine what probability of false alann is desired and adjust the threshold accordingly.
A-89
-------
T.I The PSLD Test Method Description
The PSLD system measures product level, product temperature and ambient pressure,
which are combined to determine a volumetric leak rate. The PSLD system is a self-contained
unit which is deployed by a technician in a customer's tank for data collection and is retrieved
the following day for analysis. The technician at the site conducts no analysis of test data.
Rather, the raw data are transmitted to a central processing facility where a staff of technical
analysts examine the results in detail. As such, no on-site operator intervention is required.
When the PSLD system is used to monitor product level, product may be added or
removed during the data collection period, but the leak rate is detennined during a "quiet"
period. When the PSLD system is used as a precision leak test, however, product additions and
deletions are not allowed during the test period. It is the latter mode which this evaluation
addresses. The manufacturer describes this as an overfilled or nearly filled tank test. Although
test length is variable this evaluation is for a 4-h test. The manufacturer of the PSLD system did
not claim a detectable leak rate.
T.I.I Instrumentation
The product-level measurement system consists of a 4.445-cm (1.750-in.) diameter float
located inside a 4.458-cm (1.755-in.) diameter fiberglass tube. Conversion from height to
volume is performed using the combined cross-sectional area of all open surfaces when the tank
is overfilled, and the area at product level as obtained from a numerical integration of the tank
dimensions when it is underfilled.
The temperature measurement system, nominally used in a 30,000-L (8,000-gal) tank,
incorporates 16 RTD temperature sensors spaced at intervals of 20.32-cm (8-in.) on a 3.05-m
(10-ft) rod. Temperature sensors are thus found both in the product and in the vapor space up to
the top of the fill tube.
An absolute pressure sensor is located near the top of the fill tube in order to measure
ambient pressure changes.
Table T.I lists the instrumentation precision and resolution of the PSLD as given by the
manufacturer.
A-91
-------
Table T.I. Instrumentation Precision and Resolution as Given by the
Manufacturer
Level Measurement
Temperature Measurement
Precision
0.0013cm
().()() 14°C
Resolution
0.000254 cm
0.00277°C
T.I.2 Test Protocol
Test protocol specified for this evaluation requires that product be added immediately prior
to the test to a level approximately 30.5 cm (12 in.) above the base of the fill pipe (as measured
by a stick). The temperature of the product deposited is measured to within 0.5°C. Normally, a
sample of product is subjected to a chromatographic analysis in order to evaluate the constituent
properties of the fluid. The results are then used to determine the coefficient of thennal
expansion, the bulk modulus, the coefficient of heat transfer, and the specific heat of the fluid
being tested. The analysis to determine the leak rate is perfomied later, and a report is
subsequently sent to the customer. For the Field Verification Tests, however, the sample
analysis was performed prior to the tests so that the final leak rate could be calculated at the site.
An instrumentation check is conducted immediately preceding the test, which then
proceeds in the following sequence. At a point 34 min after the start of the test, 550.6 ml (1 pint)
of product are removed. At the end of the next 30-min period another 550.6 ml are removed; this
is repeated twice more at intervals of 30 min. At the end of the fourth 30-min mark, 189 L (50
gal) are removed to bring the product to a level approximately 5.1 to 10.2 cm (2 to 4 in.) below
the fill tube. Product level is then measured by means of a dipstick, and data are collected for a
period of 1 h before the test is stopped.
The important features of the PSLD test method are summarized in Table T.2.
Table T.2. Important Aspects of the PSLD Test Protocol
Sampling Period 1 min
Test Duration 4 h
Waiting Period After Addition 0 h
Test Type/Product Level Overfilled/Nearly filled Tank
A-92
-------
T.I.3 Data Analysis
Rather than supplying the algorithms for obtaining the leak rate, the manufacturer supplied
a "black box" set of code which was incorporated into the simulation for the evaluation of the
PSLD method.
The PSLD system collects data at a rate of 10 Hz. The data are then filtered and wild
points are removed. In the mathematical model of the system, however, data are sampled at a
rate of 1/60 Hz, and the filtering scheme is not included.
T.I.4 Detection Criteria
A tank is declared leaking if the flow rate as determined by the test is greater than a
threshold of+/- 0.189 L/h (0.05 gal/h). If the flow rate is less than the threshold, the tank is
declared nonleaking.
T.I.5 Mathematical Model
The description provided above is sufficient to identify the key features of the PSLD
system as it was defined by the manufacturer. The extensive number of product-level changes
specified by the manufacturer were not included in the evaluation. The simulation was
performed as an underfilled tank test because (1) the manufacturer did not follow his specified
protocol during the Field Verification Tests, and (2) the code supplied by the manufacturer
would have seen all of these product-level changes as possible leak rates.
T.2 Improving System Performance
Given the length of the PSLD test and the number of thermistors, the precision of the
instrumentation in the PSLD system should have performed much more accurately than was
observed. Therefore, it is suggested that there may be a problem in the data analysis. Because
no information was supplied by the manufacturer concerning the algorithm employed to
determine the leak rate, it is not possible to make specific suggestions for improving
performance. However, general suggestions are made based on data obtained at the Test
Apparatus and on the results of a canonical method study described in Section 10 of the report.
Since it appears that the PSLD system can be used as an overfilled-tank test, it is suggested
that the manufacturer examine the discussion in the body of this report that deals with the error
due to interaction of the volumetric noise sources in the fill tube. This causes the conversion
from height to volume to be in error, even if a calibration is performed prior to the test. This
evaluation did not examine the performance of the PSLD system as an overfilled-tank test.
A-93
-------
T.2.1 Protocol
The following protocol changes are recommended:
o Wait 6 to 12 h or longer after product delivery
o Relevel the product at 15-min intervals or less and record the volume changes
o Wait 3 to 6 h or longer after topping off the tank, and maintain constant product
level during the waiting period
Releveling the product represents the most significant performance improvement that can
be achieved by the PSLD system. Releveling removes the deleterious effect of dynamic
structural deformation described in Section 6 of the report. Eliminating the dynamic structural
deformation would improve the performance of the system by approximately 100%.
T.2.2 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-94
-------
U.I The S.M.A.R.T. Test Method Description
The S.M.A.R.T. system determines a volumetric leak rate by measuring product-level
change in an overfilled tank. The manufacturer assumes that temperature effects are
automatically compensated for by the fact that pressure is measured from the base of the tank.
The test length is variable within a maximum of 3 h.
U.I.I Instrumentation
A differential pressure sensor located 7.62 cm (3 in.) from the base of the tank is used to
measure changes in product level, accomplished by measuring the change in pressure needed to
force air into the product. The change in pressure is converted to a volume change using a scale
factor, HtoV, calibrated prior to the test. Since the manufacturer states that the measured
pressure change is physically compensated for temperature effects, no temperature data are
collected. The S.M.A.R.T. system contains another pressure sensor, located at the base of the
tank, which is used to determine the ingress of water when the system is used as a tank monitor,
but which is not incorporated in the precision leak detection calculations. Precision and
resolution of the sensor as reported by the manufacturer are presented in Table U.I.
Table U.I. Precision and Resolution of the Product-Level
Measurement System as Claimed by the Manufacturer
Precision Resolution
Product Level
0.019cm 0.0254cm
U.I.2 Equipment Setup and Test Preparation
Test protocol requires that the tank be filled to a level within the fill tube 12 h prior to the
test. If, after the 12 h have elapsed, product level has dropped below the fill tube, more product
can be added to the tank without an additional waiting period before the start of the test.
U.I.3 Test Protocol
The important features of the S.M.A.R.T. test method are summarized in Table U.2. The
product level data are collected at non-periodic intervals during each hour, but the only data
employed in the leak rate calculations are those collected at each hourly interval.
A-95
-------
Table U.2. Important Aspects of the S.M.A.R.T. System Test Protocol
Sampling Period 1 h
Test Duration 1 h
Test Start After Filling Tank 12 h
Test Type/Product Level Overfilled Tank/Below Grade
U.I.4 Data Analysis
The test lasts 1 to 3 h, with the difference between the first and last values over a 1-h
period used to determine volume change for that period. If the absolute leak rate calculated
during the first hour is greater than 0.189 L/h (0.05 gal/h), the test will continue for another hour.
If the absolute leak rate calculated during the second hour still exceeds 0.189 L/h, the test will
continue one more hour. In this case the leak rate determined during the third, or final, hour is
the one reported as the test result. If the leak rate determined during either the first or second
hour is less than 0.189 L/h, the leak rate determined during that hour is the one reported.
U.I.5 Detection Criteria
The tank is declared leaking if the absolute value of the calculated leak rate is greater than
0.189 L/h.
U.I.6 Mathematical Model
While the description provided above is sufficient to identify the key features of the
S.M.A.R.T. system and to interpret the results of the performance evaluation, it should be noted
that a mathematical model of the system was developed based on the more detailed description
of the test method provided in [U-l]. After it had been validated, this model was used to
estimate the performance of the method.
U.2 Improving System Performance
The S.M.A.R.T. system as defined in [U-l ] can realize performance improvements. If the
present approach is to be followed, a major change to protocol or instrumentation will be
required, as discussed below. Another alternative would be to abandon the attempt to do
automatic thermal compensation and to make temperature measurements instead. For a
discussion of system performance for a range of methods measuring height in the fill tube and of
the role of an array of thermistors, refer to the canonical study discussed in the body of the
report.
A-96
-------
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual performance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. The waiting period after
topping is too short to minimize the impact of the (1) spatial inhomogeneities and (2) structural
deformation produced by the addition (or removal) of product to reach the level required to begin
the test. A waiting period of at least 2 to 3 h is required to minimize the spatial inhomogeneities
in temperature. This period is long enough to minimize the effects of a 7°C difference in
temperature between the product in the tank and the product used to top the tank. A longer
period may be required to minimize the deformation-induced product-level changes. This can
present operational problems, because the time required to minimze the effects of deformation in
some tanks may be 1 to 5 tunes (or more) longer than the tune required to wait for the
temperature effects to stabilize. The time constants of the Test Apparatus tanks, for example, are
2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or longer; the waiting
period would be longer if the deformation effects were large and more than three tune contants
were required for them to become small compared to the smallest leak to be detected. Since the
temperature and deformation effects associated with topping are independent of each other, the
protocol and/or analysis algorithm use to minimize this effect must address each disturbance. A
single waiting period based on a maximum disturbance would adequately address the
temperature instability. However, this may not suffice for the deformation effect.
U.2.1 Protocol
The performance of the S.M.A.R.T. system can be improved by establishing a method for
determining the length of the waiting period after topping off the tank. An adequate waiting
period will ensure that the effects of structural deformation are negligible. While no algorithm
for determining the length of the waiting period is presented here, a general discussion of
structural deformation and how to determine its magnitude is presented in the body of the report.
I).2.2 Temperature Compensation
The claim that temperature compensation occurs automatically is incorrect when pressure
measurements are made with the product at a level in the fill tube. The error is best illustrated by
example. The following analysis is performed assuming that the temperature, density, and rate
of temperature change throughout the tank are uniform. When measurements are made in a fill
tube with a diameter of 10.16 cm, a temperature change of 0.0 PC results in a level change of
approximately 4.5 cm (362.4 ml). Because of the large volume of gasoline in a full tank, and the
relatively small cross-sectional area of the fill tube, temperature changes result in large
hydrostatic head changes compared to the small change in product density. In this example, with
A-97
-------
the initial product level 3 m above the pressure sensor, the measured gauge pressure changed by
a factor of 0.015, and the density changed by a factor of 0.0000125. This demonstrates that the
S.M.A.R.T. pressure measurements are not automatically compensated for temperature effects.
Automatic temperature compensation is better approximated when the product level is below the
fill tube. For example, at a level of 236 cm in the steel tank of the Test Apparatus, a 0.01 °C
change in temperature results in a product-level change of approximately 0.004 cm. Assuming
the change could be measured precisely, the curvature of the tank walls would induce a 150-ml
apparent product voidage error. At mid-tank, a 0.01°C change in temperature results in a
product-level change of 0.0003 cm and an error of 133 ml. These estimates do not include any
coupling effects due to structural deformation. This simple analysis indicates that thermal
compensation cannot be performed by measuring pressure at the base of the tank when the
product is in the fill tube. The induced error in automatic compensation can be reduced, but not
eliminated, by testing in a partially filled tank.
If the'automatic thermal compensation approach is to be retained, product level must be
below the fill tube when test measurements are made. Given this situation, the S.M.A.R.T.
system would be limited by its instrumentation resolution for a 1- to 3-h test. To improve
performance the test could be lengthened or the instrumentation improved. If the test were
lengthened, data could be averaged, thereby reducing the need for better instrumentation. The
longer the test, the less stringent the resolution requirement becomes. Although the assumption
that thermal compensation occurs automatically when pressure measurements are made is an
erroneous one, the error is minimized when product is below the fill tube during testing.
U.2.3 Analysis
Because of the assumption that thermal compensation is automatic, system performance
would not be improved by making changes in the analysis. If the protocol were changed such
that the product level was below the fill tube during testing, a recommendation could be made
that a more sophisticated algorithm (such as a least-squares fit) be employed to determine the
flow rate.
U.2.4 Detection Criterion
The probability of false alann is detennined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-98
-------
V.I The Tank Auditor Test Method Description
The Tank Auditor Leak Detection System is a precision volumetric leak detection system
capable, according to the manufacturer, of reliably detecting leaks as small as 189 ml/h
(0.05 gal/h). The test equipment is temporarily installed in a tank in order to perform the
measurements. The test is done in an overfilled tank, and the resulting test data, obtained during
a period in which dispensing operations have been suspended, are used to determine a
temperature-compensated volume change in the tank.
V.I.I Instrumentation
The Tank Auditor system gives computerized temperature and height measurements. The
precision and resolution of these measurement systems, as reported by the manufacturer, are
summarized in Table V.I. Product level is measured by monitoring the movement of a small
buoyant displacer suspended in the liquid. The movement of this displacer is converted, via an
RF transducer, to an electrical signal which can then be recorded. The magnitude of the
transducer output is directly related to the motion of the displacer. Product level is converted to
volume by inserting a bar of known volume into the product and recording the
displacer/transducer response.
Temperature is measured by a single quartz thermometer probe located at the geometric
vertical center of the tank during testing. The probe is suspended in the tank via a thin, weighted
cable, and is not rigidly located spatially in the tank. The probe, consisting of a precisely
manufactured quartz crystal, has a frequency of oscillation that is a precise function of
temperature. Changes in product temperature result in changes in probe oscillation frequency,
which are then converted and recorded by appropriate electronics.
Table V.I. Resolution of the Tank Auditor Leak Detection System
Level Measurement System
Temperature Measurement System
Precision
0.0133cm
o.oo rc
Resolution
0.0018cm
o.ooo rc
V.1.2 Equipment Setup and Test Preparations
On the first day of testing, preparation of the tank and setup of the equipment took
approximately 2 h. Progressively less time was required on subsequent days, as the test crew
became more familiar with installation requirements for the test tank. On the second day, these
preparations took 1 h, and on the third, 30 min. Preparation time included tank level adjustments
required prior to initiating a test.
A-99
-------
V.I.3 Test Protocol
The test consists of two phases. During the first phase, product-level and temperature
measurements are made over a period of I h. During this period, total system integrity is tested,
and the presence of leaks, tilting of the tank, vapor pockets, and tank deformation ascertained.
After this time, the product level is reduced to approximately 2.5 cm above the tank top, and
another 30 min of data are recorded. The low-level testing confirms the presence of vapor
pockets, tilting, and leaks. The important features of the Tank Auditor Leak Detection System
are summarized in Table V.2.
Table V.2. Important Aspects of the Tank Auditor Leak
Detection System
Sampling Period 2 s
Test Duration 1.5 h
Test Duration of Phase II 0.5 h
Waiting Period after Addition 8 h
Test Type/Product Level Overfilled Tank
V.I. 4 Data Analysis
A temperature-compensated volume rate is computed from the product-level and
temperature measurements made during each phase of testing. For a complete test, a
least-squares linear regression is performed on both the temperature and level measurements in
order to assess the rate of change of the variables. Height changes measured during a test are
converted to volume by means of periodic calibrations performed using displacement bars of
known volume.
The change in volume due to thermal fluctuations is calculated by the equation
= CVAT V.2.
where ATV is the change in volume due to temperature change; the coefficient of thermal
expansion of gasoline, Ce, is determined by API hydrometer measurements made on a sample of
product removed from the tank; the volume of product in the tank, V, is determined by the tank
manufacturer's chart; and the change in temperature, AT, is that measured by the temperature
probe in the tank.
A-JOO
-------
The compensated volume rate is calculated by subtracting the calculated thermal volume
change from the gross volume change as indicated hy the corresponding product height
measurement, after converting that height to volume via the appropriate height-to-volume
calibration coefficient.
V.I.5 Detection Criterion
A leak is declared if the calculated temperature-compensated volume rate exceeds a
threshold of-189 ml/h (-0.05 gal/h). If the compensated volume rate is less than the threshold,
the tank is declared nonleaking.
V.I.6 Mathematical Model
While the description provided above is sufficient to identify the key features of the Tank
Auditor Leak Detection system as it was evaluated and to interpret the results of the performance
evaluation, it should be noted that a mathematical model of the system was developed based on
the more detailed description of the test method provided in | V-1J. After it had been validated,
this model was used to estimate the performance of the method.
V.2 Improving System Performance
The Tank Auditor system could realize significant performance improvements without any
hardware modifications.
The effects of topping are not included in the quantitative performance estimates made for
this method, and as a consequence, the actual performance that would be achieved during actual
testing would be less than the estimates presented in this evaluation. The 0-h waiting period
after topping is too short to minimize the impact of the (I) spatial inhomogeneities and (2)
structural deformation produced by the addition (or removal) of product to reach the level
required to begin the test. A waiting period of at least 2 to 3 h is required to minimize the spatial
inhomogeneities in temperature. This period is long enough to minimize the effects of a 7°C
difference in temperature between the product in the tank and the product used to top the tank. A
longer period may be required to minimize the deformation-induced product-level changes. This
can present operational problems, because the time required to minimze the effects of
defonnation in some tanks may be 1 to 5 times (or more) longer than the time required to wait
for the temperature effects to stabilize. The time constants of the Test Apparatus tanks, for
example, are 2 to 3 h, which means that the minimum waiting period would be 6 to 9 h or longer;
the waiting period would be longer if the deformation effects were large and more than three
time contants were required for them to become small compared to the smallest leak to be
detected. Since the temperature and deformation effects associated with topping are independent
A-IOI
-------
of each other, the protocol and/or analysis algorithm use to minimize this effect must address
each disturbance. A single waiting period based on a maximum disturbance would adequately
address the temperature instability. However, this may not suffice for the deformation effect.
V.2.1 Protocol
The recommended protocol changes incorporate procedures that were followed during the
Field Verification Tests. A standardized test protocol should be developed that eliminates
operationally induced variation in test performance.
The following protocol changes are recommended:
o wait 6 to 12 h or longer after product delivery
o relevel at 15-min intervals or longer and record the volume changes
o wait 3 to 6 h or longer after topping off the tank, and maintain constant product
level during the waiting period
The system could be vastly improved by introducing a waiting period after the tank has
been topped off with product. The volume measurement calibrations would be more accurate if
performed at the end of this waiting period. The two-phas; test protocol was not conducted
during the field tests. If the two-level test is reinstituted, a waiting period should be introduced
between the test phases to allow the tank to deform after the product level has been lowered.
Periodic product releveling would minimize the error associated with the coupling of
temperature volume and structural deformation. Because the volume measurement system has
limited dynamic range, a virtual releveling was performed between tests during the field testing.
Instead of allowing the product level to drift, the manufacturer could periodically readjust
product level. This would remove the error present in interpreting product-level changes.
V.2.2 Analysis
A longer test period would improve the estimation of temperature-volume changes. The
spatial decorrelation distance is as short as 20 cm when temperature trends less than 15 min in
duration are being measured.
V.2.3 Instrumentation
Performance improvement could be achieved by adding more thermistors to improve
spatial sampling of the vertical temperature field. Based upon studies of canonical methods
presented in Section 10 of the report, it is estimated that a 50% performance improvement could
be achieved by increasing the number of thermistors from one to five.
A-102
-------
V.2.4 Performance Estimate Based on the Field Verification Tests
Assuming that the the standard deviation of the field tests reflects the perfonnance of a
modified Tank Auditor without bi-level testing, and that the criteria for test success could be
better defined, the performance of the system as field-tested would be much better than that of
the system defined by the manufacturer in the mathematical modeling report [V-l]. The
minimum leak rate detectable by the modified system is estimated to be 2.4 L/h (0.63 gal/h),
with a PD of 0.99 and a PFA of 0.01.
V.2.5 Detection Criterion
The probability of false alann is determined by the threshold. The manufacturer should
determine what probability of false alann is desired and adjust the threshold accordingly.
A-103
-------
-------
W.I The TMD-1 Test Method Description
The TMD-1 is a volumetric leak detection system capable, according to the manufacturer,
of detecting leaks as small as 0.189 L/h (0.05 gal/h) which can occur in an underground storage
tank.
VV.1.1 Instrumentation
The TMD-1 system is composed of an In-Tank Emulator (ITE), which is used to determine
the loss or gain of fluid in the tank. The function of the TMD-I is to measure the level of fluid in
the ITE, which is done by means of an HP 5528A laser interferometer. The TMD-1 makes no
direct temperature measurements except to predict evaporation effects; this is done at the
beginning of a test. The TMD-1 method employs a constant sampling rate with a variable test
length. The test must be at least twice the length of the longest period observed in the data by
the operator. For the purpose of the mathematical model, it is assumed that this period is 10 min.
Therefore, the test must last 20 min.
The ITE, inserted in the tank through the fill tube, consists of a closed tube containing a
fluid with a low coefficient of thermal expansion (0.00004/°C) and a density of 0.84 gm/cc. A
flexible pac, or bladder attached to the bottom of the ITE rube expands and contracts in such a
way as to mimic the pressure (height) changes in the surrounding product. Due, however, to its
low coefficient of thermal expansion, the fluid in the ITE expands and contracts only minutely in
response to thermal fluctuations. Thus, the system is effectively a U-tube device sensing
pressure at the base of the tank. The sensor is designed to measure changes of product mass,
since level changes due to thermal expansion and contraction of product in the tank are expected
to be compensated automatically by the associated change in product density (i.e., pressure
change). The laser interferometer system measures the fluid level inside the ITE tube. The sum
of evaporation effects subtracted from the product-level change measured in the ITE predicts the
volume change or leak. The claimed precision and resolution of the system are given in Table
W.I.
Table W.I. Precision and Resolution of TMD-1 System as Specified by the
Manufacturer
Level Measurement
Precision
12.7 urn
Resolution
63.5 |lm
Data are collected at a rate of 20 Hz and smoothed using an 11-point smoothing window.
This is done by means of a simple moving average, using the statistical software package
developed by Statistical Graphics Corporation (SGC). After collection and smoothing of the
A-105
-------
data, these SGC routines fit a third-order polynomial, an exponential, and a linear equation to the
data by least-squares methods. The mean percentage error (MPE) and the mean absolute
percentage error (MAPE) from the first two forms (exponential and third order) are used to
determine whether the data are good; errors must be within +/- 0.5% and 1.00%, respectively, in
order for the data to be considered acceptable. If, within the 20-min test length, the data are
determined not to be good, no conclusions can be drawn and the test must be repeated.
W.I.2 Equipment Setup and Test Preparations
Equipment setup normally requires approximately 2 h. After this time, testing may begin,
assuming that the proper level has been established in the tank.
W.I.3 Test Protocol
Test protocol requires that the tank be filled to 95% of capacity at least 12 h prior to
testing. The liquid level in the tank is determined manually by means of a dipstick. This liquid
level is used to determine a trigger value in length units. If the absolute slope of the
product-level change is greater than this value, the tank may be leaking. If the slope is within a
range of+/- 20%, the average of the tests is reported; otherwise the leak rate of the second test is
reported. It is important to note that the data are not acceptable unless the MAPE and MPE
values for each leak rate are within the prescribed limits.
The important features of the TMD-1 test are given in Table W.2.
Table W.2. Important Aspects of the TMD-1 Test Protocol
Sampling Period 0.05 s
Test Duration Variable (1 -2 h)
Waiting Period After Addition 12 h
Test Type/Product Level Underfilled/95% filled tank
W.I.4 Data Analysis
Prior to conducting a set of tests, the expected evaporation rate (ER) is calculated. ER is
then subtracted from the rate of volume change measured to determine the final leak rate.
As noted previously, three different curves are fit to the data collected. The first two, a
third-order polynomial and an exponential, determine whether the data are good. The third
curve, a linear equation, determines the rate of height or product-volume change. If the data are
determined to be good at the end of a 20-min test, the linear equation is fit and a leak rate is
determined. If the data are not good, the test is repeated until acceptable data are obtained. Once
A-106
-------
the data are considered acceptable, a leak rate is calculated and compared to the threshold value
or trigger rate. As described in [W-l |, if the value is within +1-20% of the trigger rate, the test is
performed again, this time at a test duration of 1.5 times the original test (30 min). Again the
leak rate is calculated. If both leak rates are within the range of +/- 20%, the average of the two
is reported; otherwise the second leak rate is reported.
W.I.5 Detection Criterion
A tank is declared leaking if the flow rate as determined by the test is greater than
0.189 L/h (0.05 gal/h). If the leak rate is less than the threshold, the tank is declared nonleaking.
W.I.6 Mathematical Model
While the description provided above is sufficient to identify the key features of the
TMD-1 as it was evaluated and to interpret the results of the perfonnance evaluation, it should be
noted that the mathematical model of the system was based on the more detailed description of
the test method described in [W-1J. After it had been validated, this model was used to estimate
the performance of the method.
W.2 Improving System Performance
No recommendations for improving system perfonnance were made as part of this
evaluation.
A-107
-------
-------
X.I The Tank Sentry II Test Method Description
The Tank Sentry n is a precision volumetric leak detection system capable, according to
the manufacturer, of detecting leaks as small as 189 ml/h (0.05 gal/h). The sensor consists of a
level measuring device and a temperature measurement system. The test is conducted in a nearly
full tank in which dispensing operations have been suspended.
X.I.I Instrumentation
The Tank Sentry El leak detection system uses two measuring devices to estimate a
temperature-compensated flow rate: product level as determined by means of a dip-tube
manometer and product temperature as determined by a temperature probe.
Volumetric changes are determined by monitoring pressure changes with a dip-tube
pressure sensor (manometer). This measuring device consists of two units: a dip tube mounted
in the tank and a reference cell mounted on a work station close by. The dip tube is inserted
below the product level in the tank and is matched with a similar external device connected to a
micrometer mounted in the reference cell. Product-level changes in the tank alter the pressure
required to force gas bubbles through the dip tube. To monitor these changes, the pressure
exerted on the dip tube is compared, by means of a sensitive differential pressure (DP) transducer
at a zero differential pressure, to the pressure excited on the reference-cell probe. The
reference-cell duplicates the pressure changes of the dip tube in the tank. When the
reference-cell probe is at the same depth as the dip tube, the DP transducer meter reads zero (null
point). The depth of the reference-cell probe is measured with a micrometer.
Product level is monitored by the DP transducer to determine height changes during the
test, and a hard copy is generated on a strip chart for backup arid verification purposes. The
depth changes measured between the start and end of the test (i.e., between the initial and final
readings on the micrometer attached to the reference cell) represent the total height change
during the test period.
Temperature is measured with a thermocouple (probe) placed at the vertical volumetric
center of the product in the tank. The resolution and precision of the instrumentation are
summarized in Table X.I.
Table X.I. Precision and Resolution of the Tank Sentry II System as Specified by the
Manufacturer
Product Level
Temperature
Precision
0.0025 cm
0.06°C
Resolution
0.00025 cm
0.06°C
A-109
-------
X.1.2 Equipment Setup and Tesl Preparations
Equipment setup in the tank prior to a test is relatively straightforward, and does not
require extensive tank preparation. In general, test equipment can he deployed in an hour or two,
barring unforeseen instrumentation difficulties.
X.I.3 Test Protocol
The important features of the Tank Sentry II Leak detection system are summarized in
Table X.2.
Table X.2. Important Aspects of the Tank Sentry II Test Protocol
Sampling Period 12 h
Test Duration 12 h
Waiting Period after Addition 24 h
Test Type Filled Tank
Product Level Below Top of Tank
Test protocol requires a wait of at least 24 h between the filling of the tank and the start of
a test. The tank is filled to a level just below the fill tube (within a few centimeters of the top of
the tank), thus keeping the surface area to a minimum.
Product temperature data are sampled only at the start and end of the test in order to
determine the change in temperature, which is used in the thermal compensation calculations.
The product level (height) data are also sampled in this manner.
X.I.4 Data Analysis
A temperature-compensated volume rate is computed from the product-level and
product-temperature measurements made at the beginning and end of the test period.
A tank chart supplied by the tank manufacturer is used to convert product-level
measurement to volume. Conversion is accomplished by interpolating between points on the
chart.
X.I.5 Detection Criterion
A tank is declared leaking if the leak rate as determined by the test is greater than a
threshold of+/- 0.189 L/h. If the leak rate is less than the threshold, the tank is declared
nonleaking.
A-110
-------
X.I.6 Mathematical Model
While the description provided in Section X.2 is sufficient to identify the key features of
the Tank Sentry II leak detection system as it was evaluated and to interpret the results of the
performance evaluation, it should be noted that a mathematical model of the system was
developed based on the more detailed description of the test method provided in [X-1J. After it
had been validated, this model was used to estimate the performance of the system.
X.2 Improving System Performance
The performance of the Tank Sentry II system could be improved by making minor
changes. Refer to the body of the report for further discussion of improving system
performance.
X.2.1 Instrumentation
Changing to at least five relatively inexpensive thermistors similar to those used on the
Test Apparatus would significantly improve system performance, both in tenns of spatial
resolution in the tank and improved sensor resolution and precision.
X.2.2 Protocol
Due to large errors in the tank chart when the product level is within a few centimeters of
the top of the tank, height-to-volume calibrations should be done prior to conducting a test to
better determine any volume change.
X.2.3 Analysis
By increasing the sample rate and performing a least-squares regression on the data, the
Tank Sentry II system could significantly alleviate the temperature compensation problem
associated with aliasing internal waves.
X.2.4 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
A-11
-------
-------
Y.I The TLS-250 Tank Level Sensing System Test Method Description
The TLS-250 Tank Level Sensing System is an automatic tank gauging system capable,
according to the manufacturer, of detecting leaks as small as 757 ml/h (0.20 gal/h) with a
probability of detection of 0.98 and a probability of false alarm of 0.005. It is usually
permanently installed in the tank, and it can perform a leak detection test when the dispensing
operations are closed. For the tests described in this report it was employed as a volumetric test
using the protocol described in [Y-ll and summarized below. The protocol of the TLS-250
system employed as a volumetric test is different than the one used when the TLS-250 is
permanently installed in the tank.
Y.I.I Instrumentation
The TLS-250 system gives computerized measurements of product level and product
temperature. The precision and resolution of the TLS-250 measurement system as given by the
manufacturer are presented in Table Y.I. Product level is measured using an eight-section
cylindrical capacitance sensor that is tubular in shape. Its hollow center permits a temperature
probe to be suspended within the sensor and also provides an exit for the sensor's wiring. The
capacitance of the vapor space and of the product are measured in situ prior to testing. Given
these values and the distance between the capacitor plates, the dielectric constant for both media
can be estimated. The dielectric constants are then used to calculate height changes in the tank.
Temperature is measured using one thermistor located inside a sealed glass tube
approximately 2.43 m in length, with the thermistor positioned about l.lm from the bottom.
The glass tube is sheathed in an open-ended aluminum tube of the same length. The
manufacturer describes the effective temperature weighting function as a half sine wave with the
peak at the 40% volume level. This function is an approximation determined empirically, and
has not been rigorously tested by the manufacturer.
Table Y.I. Resolution and Precision of the TLS-250 System as Specified by the Manufacturer
Rcsol ul ion Precision
Level Measurement 0.0025 mm 0.0025 mm
Temperature Measurement 0.006'C ().0()6*C
Y.I.2 Equipment Setup and Test Preparations
The manufacturer indicated that it would take approximately 1 h to install the equipment
and prepare for the start of the test. This assumes that the product level is satisfactory for the
conduct of a test.
A-113
-------
Y.I.3 Test Protocol
The important features of the TLS-250 test protocol are summarized in Table Y.2. The test
consists of two phases, the first being a 2-h minimum waiting period between the delivery of
product to the tank and the start of data collection, and the second being the 8-h duration of the
data collection. The data are collected at a sampling interval of 0.67 s and averaged to obtain a
1-h sample. Although the minimum duration of a test is 5 h, the maximum is not fixed by the
manufacturer; it is usually determined by the number of hours for which normal dispensing
operations cease each day.
Table Y.2. Important Aspects of the TLS-250 System Test Protocol
Sampling Period 1 h (average of 0.67-s samples)
Test Duration 8 h
Waiting Period after Product Addition 2 h
Test Type/Product Level Underfilled Tank/80% of capacity
Y.I.4 Data Analysis
A temperature-compensated volume rate is computed from the product-level and
product-temperature measurements using a I -h average of the data obtained during the first and
last hour of the test. A tank chart furnished by the manufacturer is used to convert the
product-level measurements to volume. Height-to-volume conversion is accomplished using
curvilinear interpolation between 20 pre-loaded tank-specific calibration points. The data that
are entered represent 5% height increments across the diameter of the tank. These points can be
automatically generated from a single volume and diameter input for a steel tank (assuming a
right regular cylinder), or from the calibration points for a fiberglass tank with rounded ends, or
by entering all 20 points as obtained from a tank chart or on-site strapping tests.
The thermal expansion and contraction of the product, ATV, is estimated using
ATV = CCVAT (Y.2.1)
where the rate of change of temperature, AT is estimated from the one thermistor probe; the total
volume, V, of the product in the tank is estimated from a product-level measurement; and a value
of the coefficient of thermal expansion, Ce, for gasoline is furnished by the manufacturer.
All tests are considered successful providing that product-level and temperature data are
collected for the entire duration of the test.
A-M4
-------
Y.I.5 Detection Criterion
A leak is declared if the measured temperature-compensated volume rate exceeds
+/- 568 ml/h (0.15 gal/h). Based upon this criterion, the manufacturer claims that a leak of
757 ml/h (0.20 gal/h) can be detected with a probability of detection of 0.98 and a probability of
false alarm of 0.005.
Y.I.6 Mathematical Model
While the description provided above is sufficient to identify the key features of the
TLS-250 as it was evaluated and to interpret the results of the performance evaluation, it should
be noted that a mathematical model of the system was developed based on the more detailed
description of the test method provided in [Y-l |. After it had been validated, this model was
used to estimate the performance of the method.
Y.2 Improving System Performance
The performance of the TLS-250 system can be improved by means of changes to the
instrumentation and the protocol.
Y.2.1 Instrumentation
Probably the largest source of error is inadequate temperature compensation, a result of
undersampling the vertical temperature field. The change in performance as a function of the
number of thermistors on a vertical array is discussed in body of the final report. Substantial
improvement occurs when additional thermistors are used, allowing better sampling of the
temperature field. In the Field Verification Tests, the TLS-250 temperature measurement system
did not perform as the manufacturer indicated. The empirical measurements made during the
Field Verification Tests suggest that the system behaved more like a single point sensor rather
than the sinusoidal probe indicated by the manufacturer. It is recommended that the
characteristics of the probe be quantified.
Y.2.2 Protocol
The TLS-250 system could be further unproved by increasing the minimum waiting period
after a product delivery. At present the thermal effects of a delivery are minimized only because
the current protocol stipulates a long test duration. The test results are also influenced by the
effects of the unknown temperature changes due to horizontal gradients, as well as by structural
deformation effects. A longer waiting period after product delivery would ensure that effects of
thermal changes and of structural deformation have subsided.
A-115
-------
Y.2.3 Detection Criterion
The probability of false alarm is determined by the threshold. The manufacturer should
determine what probability of false alarm is desired and adjust the threshold accordingly.
Improved probability of detection can be attained by decreasing the threshold (i.e., increasing the
probability of false alarm).
A-11 ft
-------
References
(Appendix)
A-l. Naar, Daniel. "Mathematical Model for the AES/Brockman Leak Detection System."
Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research. Inc., Mountain View, California (23
May 1987).
B-l. Chang, N.L., Jr. "Mathematical Model for the Ainlay Tank Tegrity Tester™." Technical
Memorandum, Vista Research Project 2020, U.S. EPA Contract No. 68-03-3409, Work
Assignment 01, Vista Research, Inc.. Palo Alto, California (30 January 1987).
C-l. Naar, Daniel. "Mathematical Model for the Automatic Tank Monitor and Tester
(AUTAMAT)." Technical Memorandum, Vista Research Project 1008. U.S. EPA
Contract No. 68-03-3409, Work Assignment 01, Vista Research, Inc., Mountain View,
California (2 July 1987).
D-l. Naar, Daniel. "Mathematical Model for the Computerized VPLT Tank Leak Testing
System." Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Mountain View, California
(17 June 1987).
E-l. Chang, N.L., Jr. "Mathematical Model for the DWY Leak Sensor." Technical
Memorandum, Vista Research Project 1008, U.S. EPA Contract No. 68-03-3409, Work
Assignment 01, Vista Research, Inc., Mountain View, California (18 June 1987).
F-l. Naar, Daniel. "Mathematical Model for the EZY CHEK Leak Detection System."
Technical Memorandum, Vista Research Project 2020, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Palo Alto, California (3 February
1987).
G-l. Chang, N.L., Jr. "Mathematical Model for the Gasoline Tank Monitor (GTM)."
Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Mountain View, California
(22 June 1987).
H-l. Chang, N.L., Jr. "Mathematical Model for the Gilharco Tank Monitor." Technical
Memorandum, Vista Research Project 1008, U.S. EPA Contract No. 68-03-3409, Work
Assignment 01, Vista Research, Inc., Mountain View, California (28 May 1987).
1-1. Chang, N.L., Jr. "Mathematical Model for the Inductive Leak Detector 3100." Technical
Memorandum, Vista Research Project 1008, U.S. EPA Contract No. 68-03-3409, Work
Assignment 01, Vista Research, Inc., Mountain View, California (27 July 1987).
A-l 17
-------
J-l. Naar, Daniel. "Mathematical Model for the INSTA-TEST Leak Detection System."
Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Mountain View, California
(28 May 1987).
K-l. Naar, Daniel. "Mathematical Model for Leak Computer." Technical Memorandum,
Vista Research Project 2020, U.S. EPA Contract No. 68-03-3409, Work Assignment 01,
Vista Research, Inc., Palo Alto, California (16 March 1987).
L-l. Cervantes, Joseph. "Mathematical Model for the Leak-O-Meter Leak Detection System."
Technical Memorandum, Vista Research Project 2020, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Mountain View, California
(18 May 1987).
M-l. Naar, Daniel. "Mathematical Model for the LiquidManager Leak Detection System."
Technical Memorandum, Vista Research Project 2020, U.S. EPA Contract
No. 68-03-3409, Work Assignment 01, Vista Research. Inc., Palo Alto, California
(18 February 1987).
N-l. Naar, Daniel. "Mathematical Model for the Pneumercator LMS-750 System." Technical
Memorandum, Vista Research Project 1008, U.S. EPA Contract No. 68-03-3409, Work
Assignment 01, Vista Research, Inc., Mountain View, California (20 July 1987).
O-l. Naar, Daniel. "Mathematical Model for the MCG-1100 Leak Detection System."
Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research Inc., Mountain View, California (1
June 1987).
P-l. Naar, Daniel. "Mathematical Model for the Mooney Leak Detection System." Technical
Memorandum, Vista Research Project 2020, U.S. EPA Contract No.68-03-3409, Work
Assignment 01, Vista Research, Inc., Palo Alto, California (19 February 1987).
Q-l. Chang, N.L., Jr. "Mathematical Model for the OTEC Leak Sensor." Technical
Memorandum, Vista Research Project 1008, U.S. EPA Contract No. 68-03-3409, Work
Assignment 01, Vista Research, Inc., Mountain View, California (22 June 1987).
R-l. Chang, N.L., Jr. "Mathematical Model for the PACE Leak Tester." Technical
Memorandum, Vista Research Project 1008, U.S. EPA Contract No. 68-03-3409, Work
Assignment 01, Vista Research, Inc., Palo Alto, California (17 February 1987).
S-l. Chang, N.L., Jr. "Mathematical Model for the Petro Tite Leak Detection System."
Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Palo Alto, California (24 April
1987).
A-II8
-------
T-l. Naar, Daniel. "Mathematical Model for the Portable Small Leak Detection (PSLD)
System." Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Mountain View, California
(29 June 1987).
U-l. Naar, Daniel. "Mathematical Model for the S.M.A.R.T. Leak Detection System."
Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Mountain View, California
(19 June 1987).
V-l. Chang, N.L., Jr. "Mathematical Model for the Tank Auditor Leak Detection System."
Technical Memorandum, Vista Research Project 2020, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Mountain View, California
(4 May 1987).
W-l. Naar, Daniel. "Mathematical Model for the Pandel Tank Monitoring Device (TMD-1)."
Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract
No. 68-03-3409, Work Assignment 01, Vista Research, Inc., Mountain View, California
(30 June 1987).
X-l. Chang, N.L., Jr. "Mathematical Model for the Tank Sentry II Leak Detection System."
Technical Memorandum, Vista Research Project 2020, U.S. EPA Contract No.
68-03-3409, Work Assignment 01, Vista Research, Inc., Palo Alto, California
(10 February 1987).
Y-l. Naar, Daniel. "Mathematical Model for the TLS-250 Tank Level Sensing System."
Technical Memorandum, Vista Research Project 1008, U.S. EPA Contract
No. 68-0303409, Work Assignment 01, Vista Research, Inc., Mountain View, California
(7 May 1987).
A-119
------- |