\
xvEPA
United States
Environmental Protection
Agency
Office of
Research and Development
Washington, DC 20460
EPA/600/2-91/044a
August 1991
Volumetric Leak
Detection in Large
Underground
Storage Tanks
Volume I
-------
EPA / 600/2-91/044a
Augilst, 1991
VOLUMETRIC LEAK DETECTION IN LARGE
UNDERGROUND STORAGE TANKS
VOLUME I
by
James W. Starr, Richard F. Wise, and Joseph W. Maresca, Jr.
Vista Research, Inc.
Mountain View, California 94042
Contract No. 68-03-3409
Project Officer
Robert W. Hillger
Superfund Technology Demonstration Division
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
$& Printed on Recycled Paper
-------
DISCLAIMER
This material has been funded wholly or in part by the United States Environmental
Protection Agency under Contract 68-03-3409 to CDM Federal Programs Corporation. It has
been subject to the Agency's review and it has been approved for publication as an EPA
document. Mention of trade names or commercial products does not constitute endorsement or
recommendation for use.
-------
FOREWORD
Today's rapidly developing and changing technologies and industrial products 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 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 formulate 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 impacts, 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 engineering basis in support of the policies, programs, and regulations of the EPA
with respect to drinking water, wastewater, 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 presents the results of experiments conducted on 190,000-L (50,000-gal)
underground storage tanks (USTs) to determine how to test large tanks for leaks with volumetric
leak detection systems. The work reported in this document has applications to the UST release
detection technical standards in CFR 280 Subpart D.
E. Timothy Oppelt, Director
Risk Reduction Engineering Laboratory
1U
-------
ABSTRACT
The performance standards established by the EPA underground storage tank (UST)
regulation (40 CFR Parts 280 and 281) for volumetric leak detection systems, which include tank
tightness testing systems and automatic tank gauging systems (ATGS), were based upon
experimental research in tanks having capacities of 30,000 L (8,000 gal) and 38,000 L
(10,000 gal). However, the regulation requires the testing of tanks as large as 190,000 L
(50,000 gal). The performance of volumetric systems in detecting leaks from large tanks is not
well known, and there exist very little data from which an assessment can be made. As a
consequence, there is not enough information to help owners and operators select systems that
will be in compliance with the regulations when it comes to testing large tanks, i.e., those
between 57,000 and 190,000 L (15,000 and 50,000 gal).
This report addresses three important questions about testing the larger underground
storage tanks for leaks. First, can the EPA regulatory standards be met when volumetric
methods are used to test tanks up to 190,000 L (50,000 gal) in capacity? Second, what is the
precision required of the temperature and level sensors and what is the minimum duration of the
data collection period in order for a volumetric system to accurately test larger tanks, particularly
those that are partially filled? Third, what are the important features of a volumetric system that
meets or exceeds the regulatory performance standards?
These questions were addressed in a set of experiments conducted on two partially filled
190,000-L (50,000-gal) underground storage tanks at Griffiss Air Force Base in upstate New
York during late August 1990. The experiments suggested that the time required for the
temperature inhomogeneities within the product and for the structural deformation of the tank
system to become negligible after any large addition or removal of product and after topping is
approximately the same as observed in 30,000-L (8,000-gal tanks); in tests on the 190,000-L
(50,000-gal) tanks, however, the temperature inhomogeneities were greater than in tests on
30,000-L (8,000-gal) tanks. Thus, a system's performance in large tanks depends primarily on
the accuracy of the temperature compensation, which is inversely proportional to the volume of
the product in the tank. Volumetric tank tightness tests use a preset threshold value that, if
exceeded, is the basis for declaring a leak; they employ a waiting period after any addition or
removal of product so that both temperature inhomogeneities and structural deformation will
have a chance to subside. The thermistors used in the Griffiss experiments were calibrated to
better than 0.001°C and spaced at 30-cm (12-in.) intervals. The data from these experiments
IV
-------
suggest that, at the thresholds typical of those used in volumetric tank tightness tests, small leaks
(up to 0.38 L/h (0.1 gal/h)) would be difficult to detect even if the waiting period were sufficient
for temperature inhomogeneities and structural deformation to subside. Vertical gradients in the
rate of change of temperature near the bottom and top of the tank and horizontal gradients
between the centerline of the tank and the wall of the tank were still so large after the waiting
period that the thermistor array used in the Griffiss experiments did not provide sufficient
thermal compensation. The data also suggest, however, that if thresholds typical of monthly
ATGS tests were used, reliable detection of leaks as small as 0.76 L/h (0.2 gal/h) would be
possible.
As a result of the experiments on 190,000-L (50,000-gal) tanks, the important features of a
volumetric leak detection system that would have the performance necessary to meet EPA's
regulatory standards for volumetric tank tightness tests have been identified. These features
include the instrumentation, test protocol, and analysis. The experiments suggest that volumetric
systems now capable of testing 30,000- to 38,000-L (8,000- to 10,000-gal) tanks can be used to
meet the regulatory standard for testing 190,000-L (50,000-gal) tanks if (1) the duration of the
test is increased from 1 or 2 h to 4 h or more to ensure that the vertical gradients are accurately
measured and to reduce the ambient and instrumentation noise, (2) the number of temperature
sensors is increased from 5 to 10 or more so that the accuracy of estimating the average
thermally induced volume change in the layer of product surrounding each sensor increases,
(3) the waiting period after any addition or removal of product is increased from 6 h to 24 h or
longer so that the horizontal and vertical temperature gradients dissipate, (4) the average rate of
change of temperature in any one layer or in the tank as a whole is small enough to allow
accurate temperature compensation, and (5) an accurate experimental estimate of the constants
necessary for converting level and temperature changes to volume is made. The experiments
further suggest that a multiple-test strategy is required to meet the tank tightness regulatory
standard.
The duration of a test depends on the precision of the instrumentation and the amount of
ambient noise present in the measured volume changes. The magnitude and frequency of the
ambient noise observed in the Griffiss experiments suggest that a test should be at least 4 h long.
Given a certain precision of the level and temperature instrumentation, the minimum duration
can be calculated. Calculations based on the Griffiss experiments suggest that, when the level
sensors have a precision of 0.0005 cm (0.00025 in.) and the temperature sensors have a precision
of 0.001°C (0.002°F), the test must be at least 2 h long. When the instrumentation is less precise,
the test must be commensurately longer.
-------
The number of temperature sensors should be sufficient to cover the vertical extent of the
tank, with denser coverage near the bottom and top of the product, where the rate of change of
temperature and the gradients in the rate of change of temperature are greatest. In the Griffiss
experiments it was observed that, during the first 9 h after product addition or removal, the 10
thermistors equally spaced at 30-cm (12-in.) intervals did not provide a sufficiently accurate
estimate of the rate of change of temperature near the bottom of the tank, where the largest
changes in temperature occurred, or near the surface of the product. (Even in the mid-region of
the tank, 30-cm (12-in.) spacing was not sufficient in cases when temperature reversals occurred;
this, however, can be monitored). Thus, it is recommended that the thermistors at the top and
bottom of the tank be spaced at intervals of 15 cm (6 in.) or less. Reducing the space between
sensors reduces the volume of product in the "layer" around each sensor, thus minimizing any
potential measurement errors.
In the Griffiss experiments, a waiting period of 4 to 6 h after the addition or removal of
product was deemed sufficient for the dissipation of horizontal gradients in the rate of change of
temperature that were observed along the long axis of the tank. Those observed between the
centerline of the tank and the wall, however, were large enough during the first 18 h to prevent
the reliable detection of leaks up to 0.38 L/h (0.1 gal/h). It is therefore recommended that, with
large tanks, a waiting period of 24 h be used. A longer waiting period may be required if the rate
of change of temperature is very great. To determine whether the waiting period is sufficient, it
is recommended that a measurement of the temperature changes in the area between the
centerline and the wall be made; if this is not possible, repeated tests should be made until there
is no observable change over time in the measured compensated volume.
Ten potential sources of error in temperature compensation are discussed, any one of
which may be large enough to produce a testing mistake. Since all errors in temperature
compensation are proportional to the average rate of change of temperature during a test, the
most direct approach for improving the accuracy of temperature compensation is to wait until
this rate has decreased substantially before beginning a test. This requires real-time
measurements of temperature.
This report was submitted in fulfillment of Contract No. 68-03-3409 by Vista Research,
Inc., under the sponsorship of the U.S. Environmental Protection Agency. This report covers a
period from 30 November 1989 to 14 September 1990, and work was completed as of
30 September 1990.
VI
-------
TABLE OF CONTENTS
Disclaimer ii
Foreword iii
Abstract iv
List of Figures ix
List of Tables ...'. xii
Acknowledgements xiii
1 Introduction 1
1.1 Objectives 1
1.2 Background 1
2 Conclusions 4
3 Recommendations 10
3.1 Temperature Compensation 10
3.2 Evaluating a Volumetric Leak Detection System 12
4 Measurement Methodology 13
4.1 Temperature Field 14
4.2 Structural Deformation 14
4.3 Waiting Periods after Topping and Delivery 14
4.4 Instrumentation Requirements 15
5 Instrumentation 16
5.1 Configuration of the Tanks and Equipment Used in the Experiments 16
5.2 Temperature and Level Measurement Systems 18
5.3 Data Quality Objective and Calibration 20
6 Experiments 22
6.1 Test Conditions 22
6.2 Weather Conditions 25
6.3 How Data Were Divided for Analysis 25
6.3.1 Coefficient of Thermal Expansion 25
6.3.2 Height-to-Volume Measurements 27
6.3.3 Level Data 29
6.3.4 Surface Waves 30
6.3.5 Temperature Data 32
6.3.6 Temperature Profiles 36
6.3.7 Thermal Volume Time Series 36
6.3.8 Temperature-Compensated Volume Time Series 41
7 Sources of Ambient Noise 43
7.1 Surface Waves 43
7.2 Temperature Inhomogeneities after Product Additions or Removals 45
7.2.1 Temperature Inhomogeneities after a Product Delivery or Product
Transfer 45
7.2.2 Temperature Inhomogeneities after Topping 45
7.3 Horizontal Gradients 47
7.3.1 Horizontal Gradient along the Long Axis of the Tank 47
vii
-------
7.3.2 Horizontal Gradient along the Short Axis of the Tank 48
7.4 Thermistor Spacing 49
7.5 Structural Deformation 51
7.6 Evaporation and Condensation 51
7.7 Internal Waves 54
7.8 Residual Volume Changes after Temperature Compensation (Overnight Tests) 54
7.8.1 Overnight Test Starting on 27 August 1990 57
7.8.2 Overnight Test Starting on 29 August 1990 59
7.8.3 Overnight Test Starting on 30 August 1990 61
7.9 Residual Volume Changes after Temperature Compensation in Topping Tests 62
7.10 Summary 63
8 Test Duration and Instrumentation Precision 65
8.1 Measurement of Small Level Changes 65
8.2 Measurement of Small Temperature Changes 67
9 Temperature Compensation 69
10 Important Features of a Volumetric Leak Detection System for Testing Large Tanks 75
11 References 78
Vlll
-------
LIST OF FIGURES
1 Horizontal plan view of the underground storage tanks at Griffiss Air Force Base
used to conduct the experiments 17
2 Cross-section of the 190,000-L (50,000-gal) tanks used to conduct the
experiments. The experiments were conducted in Tanks 1 and 2, and the
thermistor arrays were located in Manways B and C in Tank 1 and in Manways
A and B in Tank 2. The level sensor was located in Manway B in both tanks 17
3 Configuration of Thermistor Arrays A and B 19
4 Summary of the product level between 27 and 31 August 1990 and of the
experimental analyses performed on the data 22
5 Time series of the air temperature measured at the Griffiss Air Force Base
Weatherstation 26
6 Time series of the barometric pressure measured at the Griffiss Air Force Base
Weatherstation 26
7 Time series of the wind speed measured at the Griffiss Air Force Base Weather
Station 27
8 Time series of the dew point temperature measured at the Griffiss Air Force Base
Weatherstation 27
9 Average height change computed from the height-to-volume calibration data
collected at 0730 on 30 August 1990 28
10 Time series of the volume changes measured with the level sensor beginning
immediately after the initial level change done at (a) 1510 on 27 August, (b)
1441 on 29 August, and (c) 1505 on 30 August, (d) 1441 on 29 August, and (e)
1505 on 30 August. Time series of the volume changes measured with the level
sensor beginning immediately after the addition of a small volume of (a) cold
product to Tank 1 on 29 August and (b) warm product to Tank 2 on 31 August;
time series of the volume changes measured with the level sensor beginning
immediately after the initial level change done at (c) 1510 on 27 August, (d)
1441 on 29 August, and (e) 1505 on 30 August 29, 30
11 Surface-level fluctuations produced by an impulse, (a) over a 34-min period after
the impulse and (b) over a 2-min period between 35 and 37 min after the
impulse 31
12 Time series of the product temperature changes on the vertical portion of
Array A (with thermistor 19 from Array B substituted for 18 on Array A) for the
data collected after the product removal at 1441 on 29 August; (a) shows times
series for Thermistors 20, 22 and 23, (b) for Thermistors 17 and 19, and (c) for
Thermistors 14, 16, and 24 33
13 Time series of the product temperature changes on the horizontal arm extending
from the center of the tank (a) on Array A and (b) on Array B. The arm, located
at a height of approximately 180 cm (71 in.) from the bottom of the tank, is
located at the midpoint between Thermistors 16 and 18 on Array A and
Thermistors 2 and 19 on Array B 34
IX
-------
14 Time series of the (a) air temperature and (b) the vapor thermistors on Array A
for the data collected after the product removal at 1441 on 29 August 35
15 Vertical temperature profile computed 4 h after the initial level change done at
(a) 1510 on 27 August, (b) 1441 on 29 August, and (c) 1505 on 30 August 37
16 Time series of the thermally induced volume changes estimated using Array A
beginning immediately after the initial level change done at (a) 1510 on 27
August, (b) 1441 on 29 August, and (c) 1505 on 30 August; time series of the
thermally induced volume changes estimated from Array A beginning
immediatley after the initial level change done at (d) 1510 on 27 August, (e)
1441 on 29 August, and (f) 1505 on 30 August. Time series of the thermally
induced volume changes estimated from Array A beginning 0.5 to 2 h before the
initial level change done at (a) 1510 on 27 August, (b) 1441 on 29 August, and
(c) 1505 on 30 August; time series of the thermally induced volume changes
estimated from Array A beginning immediately after the initial level change
done at (d) 1510 on 27 August, (e) 1441 on 29'August, and (f) 1505 on 30
August 38,39
17 Time series of the thermally induced volume changes estimated using Array A
beginning immediately after the addition of a small volume of (a) cold product to
Tank 1 on 29 August and (b) warm product to Tank 2 on 31 August 40
18 Temperature compensated volume time series computed following the addition
of a small volume of (a) cold product to Tank 1 on 29 August and (b) warm
product to Tank 2 on 31 August 41
19 Time series of the temperature-compensated volume changes estimated using
Array A beginning immediately after the initial level change done at (a) 1510 on
27 August, (b) 1441 on 29 August, and (c) 1505 on 30 August 42
20 Power spectrum of the time series shown in Figure 11 44
21 Raster display of the temperature compensated volume time series computed
following the addition of (a) cold product to Tank 1 on 29 August and
temperatures measured by upper four thermistors; (b) cold product added to Tank
1 on 31 August and temperatures measured by upper four thermistors; (c) warm
product added to Tank 2 on 31 August and temperatures measured by lower five
thermistors 47
22 Time series of the thermally induced volume changes estimated using Array A
beginning immediately after the initial level change done at (a) 1510 on 27
August, (b) 1441 on 29 August, and (c) 1505 on 30 August 50
23 Time series derived from thermistors located in the vapor and in the uppermost
layers of product: (a) Thermistors 10,11 and 13 on Array A on 29 August, (b)
Thermistors 6,9 and 0 on Array B on August 29, (c) Thermistos 10,11 and 13 on
Array A on 30 August, and (d) Thermistors 6, 9 and 0 on Array B on 30 August.
On 29 August Array A was inserted in a 76-cm (30-in.) manway and Array B in
a 10-cm (4-in.) fill hole; on 30 August Arrays A and B were inserted in two
76-cm (30-in.) - diameter manways 53
24 Time series of the temperature-compensated volume changes estimated using
Array A beginning 3 h after the initial level change done at (a) 1510 on 27
August, (b) 1441 on 29 August, and (c) 1505 on 30 August 56
-------
25 Accurate temperature compensation requires that the average rate of change of
temperature be accurately measured in four regions of the tank 58
26 Scatter plot of the data used to estimate the coefficient of thermal expansion 71
XI
-------
LIST OF TABLES
1 Summary of the Volume of Product in Gallons Surrounding Each Thermistor on
Arrays A and B for Each of the Five Tests 20
2 Specifications of the Measurement Instrumentation 21
3 Summary of the Product Level and Product Volume Measurements 23
4 Depth of the Thermistor Located Closest to the Product Surface 24
5 Summary of the Height-to-Volume Calibration Results 28
6 Summary of the Periods of the Largest Waves Present in the 190,000-L Tank
Filled to 283.8 cm (111.75 in.) 44
7 Summary of the Leak Detection Results for Overnight Tests 55
8 Summary of the Leak Detection Results After Topping 63
9 Precision, S, of the Sensor Estimated at Sm= 0.000174 cm/h (0.000006786 in./h)
for Different Measurement Periods (A level change of 0.000174 cm/h
corresponds to a volume change of 115 ml/h (0.03 gal/h) in a half-filled,
182,000-L (48,000-gal) tank.) 66
10 Precision, S, of the Sensor Estimated for Sm = 0.0005°C/h for Different
Measurement Periods in a Half-filled 182,000-L (48,000-gal) Underground
Storage Tank 68
11 Precision, S, of the Sensor Estimated for Sm = 0.0005°C/h for Different
Measurement Periods in a Full 182,000-L (48,000-gal) Underground Storage
Tank 68
Xll
-------
ACKNOWLEDGMENTS
This research project for evaluating the performance of volumetric testing systems for
large underground storage tanks was prepared for the U.S. Environmental Protection Agency's
(EPA's) Risk Reduction Engineering Laboratory (RREL) on Contract No. 68-03-3409. Robert
W. Hillger was EPA/RREL's Technical Program Monitor on this work. Technical assistance,
experiment coordination, and technical review were provided by Anthony N. Tafuri, Section
Chief, Underground Storage Tank Program, and Robert W. Hillger, both of EPA/RREL. A
special thanks is offered to Mr. Morris Leno of the New York State Department of
Environmental Conservation and his staff for their assistance in locating tanks for these
experiments. The experimental work was done at Griffiss Air Force Base. A special thanks is
also offered to U. S. Air Force Lieutenant Colonel Thomas J. McDonald, Base Civil Engineer,
and Mr. Bruce H. Mero, Chief, Environmental Engineering Branch, for providing access to the
tanks and for their active support of the program. We gratefully acknowledge the coordination
assistance provided by Gina Carpenter, Environmental Protection Specialist. We especially
acknowledge the assistance provided by Technical Seargent Michael Olsen, Liquid Fuels
Maintenance Foreman, and his staff, who provided operational assistance during the conduct of
the experiments. This document was edited by Monique Seibel, who also prepared the technical
illustrations. Pamela Webster prepared the document for publication.
XJUl
-------
SECTION 1
INTRODUCTION
The United States Environmental Protection Agency (EPA) regulation for underground
storage tanks (USTs), published in the Federal Register (40 CFR Parts 280 and 281) on
23 September 1988, specifies the technical standards and a variety of release detection options
for minimizing the environmental impact of tank leakage [1]. With several exceptions, the tanks
covered by the regulation range in size from small (a few hundred gallons in capacity) to very
large, with no clearly defined upper limit. (The regulation covers only shop-assembled tanks;
requirements for large field-erected tanks have not yet been established). The number of large
tanks (defined here as those between 57,000 and 190,000 L (15,000 and 50,000 gal) in capacity)
represents a small but important portion of the total tank population. This number is increasing
because of the preference of tank owners/operators for a smaller number of larger tanks to meet
storage needs. Many large-volume storage facilities have tanks that are nominally 190,000 L
(50,000 gal) in capacity. Unfortunately, there is not enough information to help owners and
operators of large tanks select a testing system that will be in compliance with the regulations.
Furthermore, it is not known whether volumetric leak detection systems can achieve the same
level of performance in large tanks as they do in smaller ones.
1.1 Objectives
The program of experiments conducted at Griffiss Air Force Base was devised to expand
the understanding of large underground storage tank behavior as it impacts the performance of
volumetric leak detection testing. This report addresses three important questions about testing
the larger underground storage tanks for leaks. First, can the EPA regulatory standards be met
when volumetric methods are used to test tanks up to 190,000 L (50,000 gal) in capacity?
Second, what is the precision required of the temperature and level sensors and what is the
minimum duration of the data collection period in order for a volumetric system to accurately
test larger tanks, particularly those that are partially filled? Third, what are the important
features of a volumetric system that meets or exceeds the regulatory performance standards?
1.2 Background
Leak detection systems used as volumetric tank tightness tests on tanks currently covered
by the EPA regulation (i.e., those up to 190,000 L (50,000 gal) in capacity) must be able to
detect leaks as small as 0.38 L/h (0.10 gal/h) with a probability of detection (PD) of at least 0.95
and a probability of false alarm (PFA) of 0.05. Leak detection systems used as monthly tests,
such as automatic tank gauging systems, must be able to detect leaks as small as 0.76 L/h
(0.2 gal/h) with a PD of 0.95 and a PFA of 0.05. Most state regulations, which require that a tank
-------
tightness test be done annually, are more stringent than the federal regulation. These standards
are based on the results of an extensive experimental program conducted by the American
Petroleum Institute (API) on 38,000-L (10,000-gal) tanks at retail stations [2] and by the EPA on
30,000-L (8,000-gal) tanks at EPA's Underground Storage Tank Test Apparatus in Edison, New
Jersey [3-8].
The EPA described the important features of a generic volumetric tank tightness system
that would yield the required level of performance in tests conducted in overfilled and partially
filled tanks [3,7,8]. The important features are the ones that compensate for or minimize errors
in the measurement of volume changes not due to a leak; these errors, which are due to ambient
noise, occur in both leaking and nonleaking tanks. Experiments on a 30,000-L (8,000-gal) tank
showed that an array of five or more equally spaced temperature sensors, each weighted by the
volume of product in the layer surrounding it, was sufficient to compensate for thermally
induced volume changes. This temperature-sensing array is suitable providing that adequate
waiting periods were observed after any addition of product, whether this addition represented a
delivery to the tank or whether it constituted topping of the tank (as is required when testing an
overfilled tank). The addition of product to the tank produced inhomogeneities in the
temperature field that were large enough to prevent an accurate estimate of the mean rate of
change of temperature. As a means of minimizing the effect of these thermal inhomogeneities,
waiting periods of at least 3 h after topping and 4 to 6 h after a delivery were recommended.
Any addition of product also changes the level of the liquid in the tank and, therefore, the
pressure that is exerted on the tank walls. This change in pressure causes the tank to deform;
observing a waiting period before the test allows deformation to subside, thus eliminating any
errors that it might have produced. The waiting period may have to be as long as 12 to 18 h,
depending on the properties of the tank and of the backfill and native soil surrounding it. In tests
conducted on overfilled tanks, there is a third source of potential error: the temperature- and/or
pressure-induced expansion or contraction of any vapor trapped in the tank. If the amount of
trapped vapor is significant, or if the range of temperature and/or pressure changes within the
trapped vapor is great, the error can have a detrimental impact on the test. Unfortunately, it is
extremely difficult to identify the presence of small volumes of trapped vapor, and there is no
satisfactory way to compensate for volume changes due to this phenomenon. A fourth source of
potential error is unique to tests conducted in partially filled tanks: this is the error produced by
evaporation or condensation at the liquid/vapor interface and the vapor/tank-wall interface. It is
believed that the effects of evaporation and condensation are usually small, but available data
show that under some air-temperature and atmospheric-pressure conditions they can be large
enough to produce false alarms and missed detections. At the present time there is no systematic
-------
way to determine when these conditions will adversely affect the test; however, any adverse
effects can be minimized through the use of a multiple-test strategy. The other sources of error,
which are produced by surface and internal waves, can affect the test results in both overfilled
and underfilled tanks. These can be minimized by proper sampling of the level and temperature
data.
For best performance, the instrumentation noise, or system noise, should be less than the
ambient noise. In small tanks, whether partially filled or overfilled, this can be achieved.
However, as the volume or the surface area of the product in the tank increases, this becomes
proportionately more difficult.
Because the EPA regulations (and the recommended "important features" of a volumetric
tank tightness test that are described above) are based on experiments conducted on 30,000-L
(8,000-gal) tanks, it is unclear whether a volumetric test that meets the EPA standard when used
to test smaller tanks can achieve the same level of performance when used to test larger tanks.
At the present time, there are no experimental data that can be used to make such an assessment,
particularly for tanks as large as 190,000 L (50,000 gal). Additional information about the
magnitude of the noise in large tanks would be required before such an assessment could be
made. Based upon previous experimental and analytical work, we expected that four things
would be necessary in order for a volumetric leak detection system to maintain the required level
of performance when used to test large tanks: (1) more temperature sensors, (2) longer waiting
periods after topping or after a delivery of product to the tank, (3) a longer test duration, and
(4) an increase in the precision requirements of the temperature and level measurement systems
-------
SECTION 2
CONCLUSIONS
EPA regulations require that all volumetric leak detection systems meet the same
performance standards regardless of the size of the tank on which such systems are used. The
EPA standards for volumetric tank tightness tests and for automatic tank gauging systems were
developed from research on 30,000- and 38,000-L (8,000- and 10,000-gal) tanks. Performance
evaluations of most of the volumetric systems in use today are based on their application to tanks
30,000 to 38,000 L (8,000 to 10,000 gal) in capacity, the size of tank for which they were
designed. The EPA regulation, however, covers tanks as large as 190,000 L (50,000 gal). Since
the performance of volumetric leak detection systems in large tanks is not known, and the design
features necessary for such systems to meet performance standards when used in large tanks
have not been investigated, it is evident that further research was necessary. Before the study
reported here, there was no published information available on the errors associated with testing
tanks up to 190,000 L (50,000 gal) in size.
In the present study, limited-scope field experiments were conducted during late August
1990 on two partially filled 190,000-L (50,000-gal) underground storage tanks located at Griffiss
Air Force Base in upstate New York. The purpose of the experiments was to determine whether
volumetric systems intended for use on 30,000- to 38,000-L (8,000- to 10,000-gal) tanks could
successfully be used on larger tanks. The Griffiss tanks were 254.3 m (77.5 ft) long and 320 cm
(10.5 ft) in diameter. A level sensor with a precision of 0.0005 cm (0.00025 in.) and an array of
10 submerged thermistors spaced at 30-cm (12-in.) intervals, each thermistor having a precision
of O.OOrC (0.002°F), were used to collect the data. All experiments were conducted in partially
filled tanks. Changes in product level were effected by shunting product from one tank to
another by means of a pump. During the tests, fluctuations in the temperature of the product
produced volume changes of several liters per hour or more.
The following observations were made:
• Temperature fluctuations observed throughout the tank after approximately 38,000 L
(10,000 gal) of product had been added or removed were too great, unless at least 4 h
had elapsed, to permit an accurate leak detection test. (The addition of 38,000 L
(10,000 gal) simulated a delivery of product to the tank.)
-------
Temperature fluctuations observed after the addition of approximately 19 L (5 gal) of
product 25°C warmer or 8°C cooler than the in situ product were too great, unless 2 to
3 h had elapsed, to permit an accurate leak detection test. (The addition of 19 L (5 gal)
simulated the topping of an overfilled tank.)
The difference in the average rate of change of temperature between any two locations
along the long axis of the tank, as measured by any two thermistors at the same height
but distanced horizontally, was less than 0.001°C/h beginning 4 h or more after product
additions or removals. Furthermore, the mean temperature along the centerline of the
tank was the same at each height, i.e., the horizontal gradient was negligible at the
centerline.
A horizontal gradient in the mean temperature was observed in the mid-region of the
tank in the area between the centerline and the wall of the tank. The rate of change of
temperature, as measured by three thermistors spaced at horizontal intervals of 30 cm
(12 in.), was lower in the area of the centerline and higher in the area near the tank
wall, an observation that is consistent with the physical process of heat transfer
between the backfill surrounding the tank and the product inside it. The difference
between the rate of change of temperature measured at the centerline of the tank and at
locations off the centerline was large enough to suggest that a single vertical array is
not sufficient for temperature compensation unless at least 18 h has elapsed.
Additional horizontally oriented thermistors must be incorporated into the arrays to
compensate for these temperature changes, or the test must not be started until the rate
of change of temperature has subsided.
The area near the bottom of the tank exhibited the largest vertical gradient and the
greatest rate of change of temperature. In this region, a spacing of 30 cm (12 in.)
between thermistors was not adequate for correct temperature compensation unless at
least 10 h had elapsed since the last addition or removal of product.
The temperature fluctuations that followed changes in product level in these
experiments were so large that changes due to deformation were masked. As a
consequence, none of the classical exponential level changes associated with tank
deformation was observed in these experiments.
-------
• Analysis of the level data revealed the presence of surface waves (seiches) having
periods of 2 to 10 s and peak-to-peak amplitudes of 1 to 2 L. These waves were
consistent with the fundamental and first harmonic of waves propagating along the
long and short axes of the tank.
• Analysis of the temperature data revealed the presence of internal waves having
periods of 3 to 30 min. These subsurface waves were large enough to produce periodic
surface waves.
• Experimental estimates of the height-to-volume conversion factor were within 5% of
the theoretical estimates made with a tank chart.
• When 10 h had elapsed after a product addition or removal, and thermistors spaced at
30-cm (12-in.) intervals were used for thermal compensation, the residual volume
changes in each of three tests on nonleaking tanks were 0.36, 0.67 and -0.22 L/h
(0.095, 0.177, and -0.58 gal/h) respectively. When 2 h had elapsed after topping and
20 h after a product addition or removal, residual volume changes in each of two tests
were 0.036 and -0.043 L/h (0.0095 and -0.011 gal/h) respectively. These residual
volume changes can be attributed to inadequate temperature compensation. (There was
insufficient coverage near the surface of the product, at the bottom of the tank and in
the area between the centerline and the tank wall, all locations where large temperature
gradients were present.) The effects of evaporation and condensation within the tank
could not be quantified, but their contribution to the residual volume changes appears
to be smaller than the errors in temperature compensation.
The data collected during these experiments, combined with theoretical analysis, were
sufficient for the researchers to address each of the technical objectives of this study. A
summary of the key conclusions of this research project are provided below.
2.1 Volumetric Leak Detection Systems and EPA Performance Standards
Volumetric leak detection systems can meet EPA's performance standards for testing tanks
up to 190,000 L (50,000 gal) in capacity. The experiments on 190,000-L (50,000-gal) tanks
suggest that volumetric leak detection systems can meet the EPA standards for both volumetric
tank testing and automatic tank.gauging. That is, they can detect leaks of 0.38 L/h (0.10 gal/h)
with a PD of 0.95 and a PFA of 0.05, which is the performance required of tank tightness tests.
(By definition they can also meet the requirement for automatic tank gauging systems, which is
the ability to detect a leak of 0.76 L/h (0.20 gal/h) with a PD of 0.95 and a PFA of 0.05.)
However, achieving this goal is very difficult and probably requires a multiple-test strategy.
-------
A 0.76-L/h (0.2-gal/h) leak was detectable with the thermistor array used in these
experiments, but a smaller leak of 0.38 L/h (0.1 gal/h) could not be reliably detected. However,
leaks as small 0.38 L/h (0.1 gal/h) were detectable if the waiting period after any addition or
removal of product from the tank was at least 18 h. After some analysis, this was explained by
the fact that there was an insufficient number of thermistors at the bottom of the tank, near the
surface of the product, and between the centerline and the walls of the tank, areas where either
the rate of change of temperature or the gradient in the rate of change of temperature was greater
than in other parts of the tank. The upper portion of the layer closest to the surface was
influenced by large changes in the temperature of the vapor, so that the implicit assumption that
the rate of change of temperature varied linearly through the layer was violated. The thermistor
closest to the surface was not physically centered in the layer, and the product surrounding it was
not equally distributed above and below it. When the thermistor was too far away from the
surface (e.g., 25 cm (10 in.)), the contribution from the surface was not properly included in the
average. Any error in measuring the average rate of change of temperature was magnified by the
volume of product in the layer. The error in measuring the average rate of change in temperature
was even greater in the layer of product at the bottom of the tank. (In measurements made at the
bottom, however, the magnitude of the error decreased with time; in measurements made near
the surface it did not.) Although the temperature sensor is indeed centered in this bottom layer,
the curvature of the tank is such that the volume of the product above the sensor is significantly
greater than that below it. The average rate of change of temperature and the gradient in the rate
of change of temperature were significantly greater here than in any other layer. While errors of
similar magnitude in measuring the average rate of change of temperature are present in small
tanks, the volume of each layer in a small tank is significantly less than in a large tank,
particularly near the bottom. The largest source of error in measuring the average rate of change
of temperature was due to horizontal gradients in the rate of change of temperature between the
centerline of the tank and the wall of the tank. Better temperature compensation would have
been achieved if additional temperature sensors had been located near the bottom of the tank and
near the product surface, ensuring a better estimate of the rate of change of temperature in the
layer surrounding each thermistor, or if a longer waiting period had been used, thus minimizing
the rate of change of temperature in these layers and throughout the tank as a whole.
2.2 Features of a System That Meets EPA Standards
The important features of a generic volumetric leak detection system that can meet the
EPA performance standards have been identified. The experiments on 190,000-L (50,000-gal)
tanks allowed us to identify the key feature that a volumetric leak detection system used on
-------
larger tanks must possess in order to meet the EPA standard. It must have better temperature
compensation than what is deemed sufficient for a 30,000- to 38,000-L (8,000- to 10,000-gal)
tank, and it must have longer waiting periods and a longer test duration. All the other features
required for accurate detection of leaks in small tanks are applicable equally to the detection of
leaks in large tanks.
Five things are necessary for successful temperature compensation in tanks as large as
190,000-L (50,000-gal). First, a test must not be started until the horizontal gradients in the rate
of change of temperature between the centerline and the tank walls have dissipated. Second, the
number of temperature sensors must be sufficient that the volume of product in the layer around
each sensor is not too great; the smaller the volume in each layer, the less likely it is that a
temperature measurement error, when summed with measurements from the other layers, will
adversely affect the test. Third, the duration of the test must be long enough that (1) the
fluctuations in volume observed 6 h or more after any product additions or removals can be
averaged and (2) the precision of the temperature and level is sufficient to detect a leak with a
specified performance. Fourth, a test should not begin unless the average rate of change of
temperature in the tank as a whole or in any one layer is small enough to allow accurate
temperature compensation. Fifth, an accurate experimental estimate of the constants necessary
for converting level and temperature changes to volume is required: these constants include the
coefficient of thermal expansion, the volume of product in the tank or in each layer, and the
height-to-volume conversion factor.
How long must the waiting period be? In testing tanks up to 190,000 L (50,000 gal) in
capacity, the length of the required waiting period is controlled by the gradient in the rate of
change of temperature between the centerline and the wall of the tank. This waiting period must
be sufficiently long that the temperature fluctuations and tank deformation associated with a
product addition or removal will have time to subside. A waiting period of 24 h or longer may
be required.
The data suggest that the minimum duration of a test should be at least 4 h, long enough
that an average of the ambient volume fluctuations can be made. Whether 4 h is sufficient
depends on the resolution and precision of the temperature and level instrumentation.
The spacing between thermistors in a 190,000-L (50,000-gal) tank may need to be as small
as 15 cm (6 in.) to detect leaks as small as 0.38 L/h (0.1 gal/h), particularly near the bottom and
top of the tank where more dense coverage will result in a more accurate estimate of the rate of
-------
change of temperature. Since errors in temperature compensation increase as the average rate of
change of temperature increases, the most direct way to avoid errors is to wait until the average
rate of change of temperature has diminished before starting a test.
2.3 Test Duration
The precision of the instrumentation used to measure temperature and level changes
establishes the minimum duration of a test. In order for a volumetric leak detection system to
meet the EPA standards, the length of a test must be appropriate for the precision of the system's
instrumentation for temperature and level measurement. Given a certain level of precision, the
optimum duration of a test can be calculated. As part of the experiments, calculations were made
to estimate the minimum duration of a test conducted on a 182,000-L (48,000-gal) tank as a
function of the precision of the temperature and level sensors. It was assumed in the calculations
that the resolution of the sensors was 2 to 3 times smaller than the most extreme level change
that occurred over the duration of the test. The calculations indicated that the test duration must
be at least 2 h in the case of a level sensor having a precision of 0.0005 cm (0.00025 in.) and a
temperature sensor having a precision of 0.001°C (0.002°F). When the instrumentation is less
precise, the test duration must be commensurately longer. For example, if the level sensor had a
precision of 0.0025 cm (0.001 in.), the test would have to be at least 4 h long.
-------
SECTION 3
RECOMMENDATIONS
The recommendations developed from this research project are based on a limited set of
data. The recommendations for controlling the key sources of noise might be further refined if
additional experiments were conducted. We believe, however, that additional data would not
have any substantial impact on the general nature of the recommendations made here, and that
further refinements to these recommendations would not materially change the effort or cost
involved in developing or modifying a method of testing large tanks. Although the experiments
were limited to tests conducted on partially filled tanks, many of the conclusions are applicable
to the use of volumetric systems in overfilled tanks as well. The recommendations that emerged
from this research project fall under three headings.
3.1 Temperature Compensation
The single most important cause of errors in testing large tanks with volumetric leak
detection systems appears to be inaccurate temperature compensation. Two things are necessary
for successful temperature compensation. First, the number of temperature sensors must be
sufficient that the volume of product in the layer around each sensor is not too great. The
smaller the volume in each layer, the less likely it is that a measurement error will adversely
affect the test. This is because any erroneous measurement is averaged with presumably correct
measurements from the other layers. Second, a test must not be started if the average rate of
change of temperature of the product in the tank as a whole, or even in a single layer, is great
enough to prevent the system from detecting a leak of given size.
The following procedure is recommended for compensating for the thermal expansion or
contraction of the product.
• Place the lowest temperature sensor approximately 8 cm (3 in.) from the bottom of the
tank and the uppermost sensor approximately 8 cm (3 in.) below the surface.
• Space the temperature sensors at intervals of 15 to 30 cm (6 to 12 in.) or less along the
vertical axis of the tank; space the sensors at intervals of 15 cm (6 in.) or less in the
bottom 46 cm (18 in.) of the tank and in the 15 to 30 cm (6 to 12 in.) of product located
immediately beneath the surface. (A 30-cm (12-in.) spacing can be used if the rate of
change of temperature between adjacent layers of product throughout the entire tank is
nearly identical.)
• Partition the tank into layers, each of which is centered about a temperature sensor.
Then calculate the volume of product in each layer.
10
-------
• Wait at least 24 h for horizontal gradients in the rate of change of temperature to
dissipate. (These horizontal gradients occur between the centerline and the wall of the
tank.) Alternatively, measure these horizontal gradients directly, and do not attempt to
compensate for temperature until they have dissipated. If the compensated volume rate
exceeds the threshold, continue to test until the measured volume rate ceases to
decrease and remains constant.
• Using real-time measurements, wait for the rate of change of temperature to diminish
sufficiently that the maximum potential error in measuring the average rate of
temperature for each test is small. The acceptable rate of temperature change depends
on the number of thermistors, the precision of each thermistor, and the degree of
compensation that can be achieved with the array of thermistors. A very conservative
approach is to incorporate the following analysis tests.
- Do not begin a test if the rate of change of temperature is great enough in any one
layer to produce a volume change that will exceed the detection threshold. (When
using a threshold of 0.19 L/h (0.05 gal/h) in a tank containing JP-4 fuel, this would
limit the rate of change in temperature to less than 0.008°C in the largest layers of a
190,000-L (50,000-gal) tank divided into ten layers.)
- Do not begin a test if the average rate of change of temperature throughout the tank
is great enough to produce volume changes that exceed the threshold based on an
average level of compensation to be achieved. (When using a threshold of 0.05
gal/h in a tank containing JP-4 fuel, this would limit the rate of change in
temperature to less than 0.019°C throughout a 190,000-L (50,000-gal) tank if on
average the method is able to compensate for 95% of the temperature changes.)
• Use the most precise temperature and level measurement systems available and
calibrate them frequently and properly. It is recommended that temperature sensors
have a precision of 0.001°C and the level sensors have a precision of 0.00025 cm
(0.0001 in.).
• Check that all sensors function properly during a test. If a sensor malfunctions, the test
should be repeated.
• Make sure the test is at least 4 h long so that ambient fluctuations will be properly
averaged and will not affect the test. Longer tests may be required depending on the
resolution and precision of the level and temperature sensors.
• Measure the coefficient of thermal expansion experimentally.
• Determine the height-to-volume conversion factor used to convert level measurements
to volume measurements experimentally.
• Use a multiple-test strategy.
Whether this temperature compensation procedure is sufficiently adequate for a volumetric leak
detection system to meet the EPA's regulatory standard for a tank tightness test (or a monthly
monitoring test) will not be known until an actual performance evaluation [9,10] is conducted on
a system that incorporates some or all of these procedures.
11
-------
3.2 Evaluating a Volumetric Leak Detection System
Volumetric leak detection systems that will be used on large tanks should be
experimentally evaluated according the EPA's standard test procedure for evaluating volumetric
tank tightness tests [9]. This includes the performance of the system in terms of probability of
detection and probability of false alarm. The primary features that should be examined are the
method of temperature compensation, the waiting periods, and the duration of the test. The
results of the present study suggest that, when a volumetric leak detection system is used to test
larger tanks, longer waiting periods, a longer test duration and better temperature compensation
are required if leaks are to be detected with reliability. Unfortunately, none of the existing
facilities specializing in evaluations is equipped with 190,000-L (50,000-gal) tanks. Therefore,
systems must be evaluated at large-volume storage facilities that are operational, and the EPA's
standard test procedure should be modified to accommodate this type of evaluation.
12
-------
SECTION 4
MEASUREMENT METHODOLOGY
The technical objectives of the project were addressed by means of a limited set of
experiments and some theoretical calculations. The purpose of the experiments, which were
conducted on two 190,000-L (50,000-gal) tanks containing a petroleum product, was to
determine whether volumetric leak detection systems that meet the EPA performance standards
when used on 30,000-L (8,000-gal) tanks are valid when used on tanks up to 190,000 L
(50,000 gal) in capacity. These systems were examined in terms of the features they must
possess in order to meet the EPA standards. The experiments sought to determine whether these
features would allow volumetric leak detection systems to accurately test large tanks as well as
small ones, and, if not, to determine what modifications would be necessary in order for them to
do so.
Two such important features are the method of temperature compensation and the length of
the waiting periods that allow structural deformation of the tank and temperature fluctuations in
the product to subside. The experiments were designed to estimate the volume changes
produced by structural deformation and temperature fluctuations (i.e., to estimate the amount of
noise that could normally be expected in a large tank). Several thermistor arrays deployed in the
tank measured thermally induced volume changes in the product and provided the data used
(1) to estimate the magnitude of the horizontal and vertical changes in product temperature and
(2) to compensate for the level changes resulting from the thermal expansion and contraction of
the product.
The effects of temperature and deformation have been characterized in previous work
conducted in tanks of a nominal 30,000-L (8,000-gal) capacity at the EPA's Test Apparatus in
Edison, New Jersey [3-8]. The current work was thus focused on extending the experiments to
tanks significantly larger than those installed at the Test Apparatus. The previous studies were
used as a baseline against which to compare the results of the experiments. While the program
described here is more limited in scope than the original study, it draws extensively from the
instrumentation and experiment designs developed during the course of the previous work.
To find out whether volumetric leak detection systems would be applicable to large tanks,
it was necessary to estimate the following as they relate to large tanks: (1) the vertical and
horizontal characteristics of the product temperature field, (2) the structural deformation
13
-------
characteristics of the tank system, (3) the time required for the temperature inhomogeneities due
to topping and delivery to subside, and (4) the precision required of the temperature and level
instrumentation.
4.1 Temperature Field
Accurate temperature compensation requires that there be a sufficient number of
thermistors along the vertical axis of the tank and that the horizontal differences in temperature
be small. Two arrays of thermistors were used to make an estimate of the vertical and horizontal
differences in temperature before, during, and after the addition of product in each of two
circumstances, a delivery and topping.
4.2 Structural Deformation
Structural deformation of the tank occurs in response to any change in the hydrostatic
pressure on the tank. Previous studies of this phenomenon suggest that the response of a tank to
a change in pressure (or level) is exponential in form. The details of the shape of the exponential
response are largely influenced by the tank material, type of backfill, water-table level, and local
native-soil conditions. Experiments conducted on the EPA's 30,000-L (8,000-gal) tanks in
Edison indicate that, for the backfill/soil conditions prevalent there, a relaxation time constant of
3 h is not uncommon.
The time constant of a tank can be determined in either of two ways. In the first, used
when the tank is overfilled and product level is within the fill tube, a bar of known volume is
inserted into the tank (after any deformation effects due to previous product additions have
subsided), and the time history of the resulting volume changes required to maintain a constant
level is monitored. Then an exponential curve is fit to the cumulative volume changes required
to maintain a constant product level. This curve describes the time constant. The second way,
employed when the tank is partially filled, is to induce one rapid change in the level of product
(and continue as indicated above). The volume changes due to deformation may be difficult to
isolate during the first several hours after a change in level, because the accompanying large
temperature fluctuations cannot be compensated for.
4.3 Waiting Periods after Topping and Delivery
There are two effects of topping and delivery that will affect the performance of a
volumetric leak detection system: (1) temperature inhomogeneities and (2) structural
deformation. Both effects diminish with time and so can be minimized if there is a mandatory
14
-------
waiting period between filling/topping and testing. In these experiments, the sets of temperature
and level data from which the effects of topping and delivery were determined were obtained
immediately before and after product was added to the tank.
4.4 Instrumentation Requirements
In large tanks, the instrumentation noise can be a significant fraction of the total noise.
One way to overcome high instrumentation noise is to increase either the sampling rate or the
duration of the test. The precision required of the instrumentation can be determined from an
analysis that includes the resolution, the system noise, the duration of the test, and the number of
independent samples acquired during a test.
15
-------
SECTION 5
INSTRUMENTATION
The experiments were conducted in two operational 190,000-L (50,000-gal) underground
steel storage tanks containing JP-4 fuel and located at Griffiss Air Force Base in upstate New
York. Each tank was taken out of service for several days to support these experiments. Five
days of experimental data were collected between 27 and 31 August 1990. The experiments
conducted on 28 August had to be repeated because an electrical storm that day caused a power
outage that resulted in loss of data.
5.1 Configuration of the Tanks and Equipment Used in the Experiments
The two tanks used in these experiments are part of a large, hillside storage facility
consisting of five clusters of four tanks. Each of the tanks is cut into the hill, buried under 76 to
91 cm (2.5 to 3 ft) of backfill, and covered by grass. The native soil is sandy, and, because of the
hillside location, groundwater does not reach the area where the tanks are situated. Fuel is
delivered to the tanks by pipeline. A pump house services each cluster of tanks, as shown in
Figure 1, a plan view of one of the four-tank clusters. The labels Tank 1 and Tank 2 in this
figure refer to the order of the experiments. Figure 2 is a cross section of one of the tanks, each
of which is 320.0 cm (10.5 ft) in diameter and 23.62 m (77.5 ft) long and has a nominal capacity
of 190,000 L (50,000 gal). Level measurements made in several of the openings of one tank
suggest that the tank is nearly horizontal, with a difference of only about 1 in. in height between
the two ends. The two tanks were considered identical for the purposes of the experiments.
The pump house, which overlaps the tanks by approximately 5.3 m (17.5 ft), is a
single-story, flat-roofed building approximately 305 cm (10 ft) in height. By means of a pump,
product can be transferred from one tank to another at a rate of up to 1,500 L/min (400 gal/min).
The pump is located in the pump house, 76 cm (2.5 ft) from the end of the tank. Each tank has
an overfill protection device that prevents its being filled above a height of 305 cm (10 ft), or
beyond approximately 98% of its capacity.
In addition to the pump, there are six other openings into the tanks. There are three 76-cm
(30-in.) diameter manways, a 10-cm (4-in.) diameter fill hole, and a 10-cm (4-in.) diameter,
3.7-m (12-ft) high vent located outside the pump house; a 25-cm (10-in.) diameter level control
port is located inside the pump house. The manways provide entry into the tank. They are
connected to the top of the tanks by a flange located 15 cm (6 in.) above the tank top. This
connection is not liquid-tight and, as a consequence, the tanks could not be overfilled in these
16
-------
m
1
T
(t>
ARRAY B
o
~ 0 9 m »N
1
(?) ° O oo
ARRAY A
TANK 2
! ARRAYS
0 t O oo
ARRAY A '
TANK1
23 6 m ^
t
-H
h- !
oo |o° O O
I
I
I
oo ]O° O O
I
I
5.3 m L^
|
PUMP HOUSE
E
01
o
n
I
A
Figure 1. Plan view of the underground storage tank cluster at Griffiss Air Force Base.
PUMP
HOUSE
0.8m-*.
0.9m
E F
Figure 2. Cross-section of the 190,000-L (50,000-gal) tanks used in the experiments. The experiments were
conducted in Tanks 1 and 2, and the thermistor arrays were located in Manways B and C in Tank 1 and in Manways
A and B in Tank 2. The level sensor was located in Manway B in both tanks.
experiments. The first manway (A) is located 76 cm (2.5 ft) from the end of the tank and the
other two manways (B and D) are located 9.1 and 17.5 m (30.0 and 57.5 ft) away. A ladder is
permanently installed in each manway.
Experiments were conducted in Tank 1 between 1030 on 27 August and 0800 on 30
August and in Tank 2 between 1015 on 30 August and 1450 on 31 August 1990. In Tank 1,
Thermistor Arrays A and B were inserted in Manway B and the vent hole at C, respectively, and
17
-------
in Tank 2 in Manways B and A, respectively. The distance between arrays was 5.5 m (18 ft) in
Tank 1 and 9.1 m (30 ft) in Tank 2. The horizontal arms of the arrays, which extend from the
center of the tank to the wall of the tank, were located on opposite sides of the tank in Tank 1 and
on the same side of the tank in Tank 2. The array locations and the configuration of the
horizontal arm on the arrays are shown in Figure 1. In both tanks, the level and pressure sensor
measurements during the tests were made at Manway B. All of the level changes in the tanks
were done by adding or removing product from the tank by means of a pump which was located
8.5 m (28 ft) away from the nearest thermistor array in Tank 1 and 46 ft away from the nearest
thermistor array in Tank 2. A 2,470-ml cylindrical bar used to determine the height-to-volume
conversion factor experimentally was inserted into and removed from the liquid in the tank at
Manway A in Tank 1 and at Manway C in Tank 2; these openings were selected because there
was no temperature or level measurement equipment located there. All stick measurements were
made in these same openings. The small-volume product additions, whose purpose was to
simulate the effects of topping, were done in each tank at the opening where Array B was
located. With the exception of a limited number of experiments conducted to measure surface
fluctuations at 1 sample/s (1 Hz), all data were collected at a sample rate of 1 sample/min
(0.017 Hz).
The experiments adhered to the well-defined operational and security requirements of
Griffiss Air Force Base. The tanks could not be filled above 10 ft, the level at which the overfill
protection devices were set. The main constraint in designing the experiments, however, was
that the technical staff were allowed to be present at the tank site only between 0700 and 1630
each day. Hence, the product transfers, the product additions to simulate topping, and the
height-to-volume measurements, all of which were necessary to set up the conditions for the
tests, had to be made during this period. It was also understood that normal operation of the
tanks would take precedence over the experiments and that product might possibly be dispensed
from the tanks being used in the experiments. As it turned out, the staff at Griffiss were able to
use the other tanks to fulfill their operational needs, and no interruption of the experiments was
experienced.
5.2 Temperature and Level Measurement Systems
The following data were required for these experiments: the change in the temperature of
the product and vapor in the tank and the height and change in level of the product. The product
temperature data were analyzed to estimate the thermally induced volume changes in the tank.
The level data were converted to volume data by means of the experimental estimates of the
height-to-volume conversion factor. The level data were used to estimate the volume of product
18
-------
in the tank during a test. Air temperature was also measured; air temperature data are not central
to the experiments or the analysis, however, and were used only to characterize weather
conditions during the experiments.
Two types of product level measurements were required. The first was a measurement of
the height of the product from the bottom of the tank; a pressure sensor with a precision of
0.5 cm (0.2 in.) or better was used for this. The second was a measurement of the level changes
in the tank; an electromagnetic sensor developed by Vista Research prior to these experiments
was used for this second measurement. The electromagnetic sensor consists of a linear variable
differential transducer (LVDT) and a float. The LVDT is a commercially available sensor that
has a dynamic range of ±0.5 cm (±0.2 in.) over 10 volts. The sensor is read to the nearest
0.0001 volts, which results in a resolution of 0.0005 cm (0.0002 in.). The pressure sensor was
located at the bottom of Thermistor Array A. In addition, each time the level was changed, it
was measured to the nearest 0.3 cm (0.125 in.) with a calibrated stick.
THERMISTOR
THERMISTOR
k
1
1
p I
u
3
J '
1 ! '
J_ ,
30.5 cm
i
t ,
1
T |
1 12.7cm
ARRAY A ARRAY B
> 13 PRODUCT LEVEL 0 '
29 AUQ (COLD TOP) - — — ............ - - Q
1 "** Mir: 6 <
,17 5 <
,~i8~j — »j 51 cm h — 19 •
• 16 27 21 26 12 25 15 j 2 <
t 14 ] 1 <
• THERMISTOR
R t
' 24 t] THERMISTOR NUMBER
I t || j THERMISTOR OUT OF CAUtRATION - (
i. 22 * '
' 20 3J
i
. E
c
I ^
I
.1 _L
i 30.5cm
;'T
iTTcm1
1
Figure 3. Configuration of Thermistor Arrays A and B.
To measure product temperature, two arrays of thermistors were used. Figure 3 shows
Arrays A and B and the channel number of the thermistors on each array. The thermistors,
attached to a stainless steel tube, were spaced at intervals of 30 cm (12 in.) along the vertical axis
19
-------
of the tank. The thermistor closest to the bottom of the array was located approximately 13 cm
(5 in.) from the bottom of the tank. The vertical portion of the array contained a total of
11 thermistors and was used primarily to estimate the average thermally induced volume change
of the product in the tank. Each array, placed in the tank through the fill hole or a manway, was
equipped with a 1.5-m (5-ft) -long pivoting "arm" that could be lowered to a horizontal position
after the array had been positioned. The pivoting arm provided for the measurement of
horizontal thermal gradients between the tank's centerline and its walls. The arm contained three
thermistors located at intervals of approximately 51 cm (20 in.); the thermistor located farthest
from the centerline was within 7.6 cm (3 in.) of the tank wall. Each thermistor was accurate to
within 0.64 cm (0.25 in.). Table 1 shows the nominal height of each thermistor from the bottom
of the tank (except for the top and bottom ones) and shows the volume of product in the 30-cm
(12-in.)-high layer centered about each thermistor.
Table 1. Summary of the Volume of Product Surrounding Each Thermistor on Arrays A and B
Thermistor
(Array A)
13
11
10
17
18
16
14
24
23
22
20
Channel
(Array B)
0
9
6
5
19
2
1
8
7
4
3
Thermistor Height
(cm (in.))
318(125)
113(45)
101 (36)
89 (35)
77 (27)
65 (26)
53(21)
41(16)
29(12)
17(67)
5(2)
Volume of Product
(L (gal))
3884(1096)
3661 (952)
4470(1163)
5532(1438)
5925 (1540)
6075(1580)
6000(1560)
5692(1480)
5109(1328)
4129(1074)
2139 (556)
5.3 Data Quality Objective and Calibration
The data quality objective for the instrumentation used in the experiments is based upon
the EPA performance standard for tank tightness tests [1] and is more fully described in
Section 8 of this report and in the Quality Assurance Project Plan (QAPP) developed for these
experiments [11]. All of the temperature and level measurement systems that were considered
for use in the experiments have the capability to detect a leak of 0.38 ml/h (0.1 gal/h) with a PD
of 0.95 and a PFA of 0.05 in a given measurement period. The range, resolution, precision, and
accuracy of the temperature and level sensors that were actually used are summarized in Table 2.
The minimum duration of the measurement period required for each sensor to meet the data
quality objective is presented in the last column of this table. The values presented in Table 2
are based on the following definitions: (1) accuracy is defined as the difference between the
20
-------
'measured and actual values; (2) precision is defined as one standard deviation of uncertainty in
each measurement; (3) the resolution of a sensor is determined from the A/D converter in each
instrument and is less than or equal to the required precision.
Table 2. Specifications of the Measurement Instrumentation
Sensor
Temperature
Product
Air
Product Level
Absolute Pressure
Electromagnetic
Range
5 to 25°C
5 to 25°C
0 to 370 cm
(0 to 144 in.)
1 cm
(0.39 in.)
Resolution
0.0008"C
0.0008°C
0.0012cm
(0.00047 in.)
0.00025 cm
(0.000097 in.)
Accuracy
0.05'C
0.5'C
5.0cm
(1.95 in.)
n/a
Precision
o.oorc
o.rc
5.0cm
(1.95 in.)
0.00064 cm
(0.00025 in.)
Duration
Ito2h
< 1 min
< 1 min
2h
The sensors were calibrated according to the procedures described in the QAPP [11]. All
sensors used in the analysis (temperature, level, and pressure sensors) were within specification.
The temperature sensors, or thermistors, were calibrated both before and after the experiments in
a well-mixed water bath to attain a precision of 0.001°C or better over a range of 0 to 30 °C. The
accuracy of the majority of the thermistors was generally better than 0.02°C. The second
calibration, the one that took place after the experiments, was required because the location of
the sensors in the data acquisition box had been changed in the field. Thermistor 18 on Array A
and Thermistors 1 and 2 on Array B fell outside the required precision for temperatures above
20°C and below 10°C. Since all thermistors brought into the field were within the 0.001°C
precision and 0.05°C specifications, these thermistors presumably functioned properly during the
test, and it is probable that the error (the fact of the three thermistors falling outside the range)
was in the calibration curve developed after the experiments were completed; a third calibration
would probably have corrected this deficiency, but the loss of these thermistors did not impact
the conclusions drawn from the study. These thermistors were removed from the analysis when
the temperature of the product in the tank was above 20°C and below 10°C. The level sensor was
checked with a precision caliper to verify that its response was within the LVDT specification; it
was then calibrated in the 30,000-L (8,000-gal) tanks at the UST Test Apparatus by means of
height-to-volume measurements with several different bar sizes. To calibrate the pressure
sensor, the level of product in a 10-cm (4-in.) -diameter tube was raised from 0 to 3.0 m (0 to
10 ft) in 15-cm (6-in.) increments. The pressure sensor was checked in the field by means of
stick measurements.
21
-------
SECTION 6
EXPERIMENTS
A description of the test conditions, weather conditions, and the temperature and volume
data is provided below.
6.1 Test Conditions
The conditions required to examine each source of noise were produced by adding or
removing product from the tank. The time line presented in Figure 4 summarizes the nominal
product level and the product additions and removals during the measurements. Also shown are
the time of the height-to-volume calibrations and the time at which 19 L (5 gal) of product was
added to the tank to simulate the effects of topping. Table 3 presents an overview of all the
measurements of level and volume made over the five-day period.
300
o
uu
UJ
280 -
260 -
240 -
220 -
200 -
180
TOPPING
HEIGHT-TO-VOLUME MEASUREMENTS
PRODUCT
REMOVAL
COLO PRODUCT
TOP REMOVAL
PRODUCT WARM
ADDITION TOP
0 20
8/27/90
40
8/28/90
60
8/29/90
80
8/30/90
100 120
8/31/90
HOURS ELAPSED/DATE
Figure 4. Summary of the product level measurements between 27 and 31 August 1990 and of the analyses
performed on the data.
22
-------
Table 3. Summary of the Product Level and Product Volume Measurements
Tank Date
1 8-27-90
1 8-27-90
1 8-27-90
1 8-28-90
1 8-29-90
1 8-29-90
2 8-30-90
2 8-30-90
2 8-30-90
2 8-31-90
Time
0800
0845
1510
1126
1400
1455
1008
1430
1513
1519
Level
(cm (in.))
236.5(93.125)
290.83(114.5)
262.25(103.25)
293.37(115.5)
29242(115.125)
253.04 (99.625)
212.73(83.75)
213.36(84.0)
283.84(111.75)
284.16(111.875)
Volume
(L (gal))
150,747 (39,775)
181,583(47,911)
166, 836 (44,020)
182,629(48,187)
182,269(48,092)
161,333 (42,568)
134,280(35,430)
134,735 (35,550)
178,383 (47,067)
178,539 (47,108)
Four experiments were conducted on Tank 1 to estimate the time constant of the structural
deformation and the magnitude of the residual volume changes after temperature compensation.
The best method of conducting this type of experiment is to overfill the tank into a small-
diameter fill hole or standpipe and instantaneously change the level approximately 61 cm (24 in.)
by inserting or removing a bar of known volume. This procedure has the advantage that the
temperature field in the tank is not disturbed, so that accurate estimates of the deformation-
induced volume changes are possible even immediately after the product level has been changed.
The Griffiss tanks, however, could not be overfilled, and a different method had to be used. The
rapid change in level was induced by pumping product in or out with the transfer pumps. In
general, it took 15 to 30 min to change the level by approximately 29 to 70 cm (11.25 to
27.75 in.). The same data used to estimate the time constant of the structural deformation were
used to examine the vertical and horizontal temperature inhomogeneities that are produced by
adding or removing product and to determine what degree of compensation can be achieved for
the thermal expansion and contraction of the product.
Product was added to Tank 1 at 0800 on 27 August and 1100 on 28 August, and was
removed at 1458 on 27 August and 1402 on 29 August. Approximately 19 L (5 gal) of product
that was 5'C warmer than the in situ product was added to Tank 1 at opening C (containing
Array B) at 1543 on 28 August, approximately 7 h after 33 cm (13 in.) of product had been
added. Approximately 19 L (5 gal) of product that was 8"C cooler than the in situ product was
added to Tank 2 at opening A (containing Array B) at 0818 on 29 August, approximately 21 h
after the 33-cm (13-in.) product addition on 28 August. The product was poured into the tank
without the use of a drop tube. The small volumes of product were added to examine the
temperature effects that might be produced by topping the tank; the deformation effects produced
by topping were examined independently.
23
-------
Not all of the data shown in Figure 4 and Table 3 were used in the analysis. The
temperature data collected between 0759 on 28- August and 0744 on 29 August and the level data
collected between 1403 on 28 August and 0747 on 29 August were lost due to the power outage
that occurred during the evening of 28 August. None of the data collected between 1100 on
28 August and 0747 on 29 August were analyzed. Thus, the data collected during the "warm
topping" experiment and the "25-in. product addition" experiment were lost. The warm topping
experiment was repeated in Tank 2 on 31 August. Another set of data that was not analyzed was
that collected between 0845 and 1458 on August 27, the first day of tests, because the product
addition that was completed at 0845 was done before the temperature and level instruments had
been placed in the tank. Collection of temperature and level data was initiated at 1025.
Only one product addition and one topping experiment were done in Tank 2. At 1430 on
30 August, the level of the product in Tank 2 was raised approximately 70 cm (27.75 in.), and at
0820 on 31 August, about 18 h after 70 cm (27.75 in.) of product had been added to the tank,
approximately 19 L (5 gal) of product that was 5°C warmer than the product in the tank was
added.
The five data sets analyzed and discussed in this report are noted at the bottom of Figure 4
and in Table 4. The three tests begun between 1440 and 1505 on 27, 29, and 30 August will be
referred to as overnight tests in this report; the other two tests will be referred to as either the
topping tests or more specifically, as the warm topping or cold topping tests.
Table 4. Depth of the Thermistor Located Closest to the Product Surface
Tank
1
r
i
i
2"
Start Date
8-27-90
8-29-90
8-29-90
8-30-90
8-31-90
Start Time
1510
0819
1441
1505
0820
Nominal Product
Level
(cm (in.))
262.25 (103.25)
292.4(115.125)
253.0 (99.6)
283.8(111.75)
283.8(111.75)
Thermistors
Closest to Surface
Array A/ Array B
10/6
11/9
17/5
10/6
10/6
Nominal
Thermistor
Height
(cm (in.))
256.5 (101)
287.0(113)
226.0 (89)
256.5(101)
256.5 (101)
Nominal
Product Above
Thermistor
(cm (in.))
5.7 (2.25)
5.4(2.125)
26.9(10.6)
27.3 (10.75)
27.3 (10.75)
"Test begun 2 h after topping with colder product
"Test begun after topping with warmer product
The temperature of both the vapor and the liquid in a tank is controlled by the addition or
removal of product and the heating and cooling of the ground around the tank. In general, the
fluctuation in air temperature during a given 24-h period will not have a strong influence on the
temperature inside the tank. However, direct communication with the external, ambient
environment, through the 76-cm (30-in.) -diameter manways, may influence temperatures inside
24
-------
the tank. These manways are large enough that the diurnal temperature changes do affect the
temperature of the vapor in the tank. In addition, the pump house is large enough to block
sunlight and produce shadows, changing the ambient heating and cooling of the ground around
the tanks and the air in the manways. Since the tanks are so long, the temperature changes can
differ from one manway to another.
6.2 Weather Conditions
In addition to the level and thermistor data, the air temperature, atmospheric pressure, dew
point, and wind speed data collected at 30- to 60-min intervals by the Air Force are plotted in
Figures 5 through 8 as a aid to interpreting the results. The vapor pressure in the tank, which
controls the evaporation and condensation and the heating and cooling of the surface layer of the
product, is influenced by these qualities. The air temperature shows a strong diurnal variation,
which was observed in the thermistors in the vapor space on the arrays located in the 76-cm
(30-in.) -diameter manways, but not in the 10-cm (4-in.) -diameter opening. The atmospheric
pressure decreased continuously from 27 August until 2400 on 28 August and then increased
continuously from 1200 on 29 August through the end of the experiments on 31 August. The
wind speed was generally less than 5 m/s during the tests.
6.3 How Data Were Divided for Analysis
As shown in Figure 4 and Table 4, the data from five tests were divided for analysis.
There were three overnight tests that followed the addition or removal of 15,000 to 43,500 L
(4,000 to 11,500 gal) of product from the tank. The intent here was to simulate the effects of a
delivery or product transfer prior to a leak detection test. There were two more tests that
followed the addition of 19 L (5 gal) of warm and cold product, respectively. The purpose of
these two tests was to simulate the effects of topping.
6.3.1 Coefficient of Thermal Expansion
Samples of product were taken from the tank on 27 August 1990, and an estimate of the
coefficient of thermal expansion was made from measurements of the API gravity and from the
API tables. The coefficient of thermal expansion obtained, 0.000104/°C, was used in all
calculations found in this report.
25
-------
' M*
r 3
*«.0. H
3% i
8
o
o
./'
\
\
gs
oa.
2 8
>«
fl-
cl
3 w
It
a.
>*>
-------
I
o
01
01
O.
0
6 -
4 -
2 -
20 40 60 80
TIME - h
100 120
Figure 7. Time series of the wind speed measured at the Griffiss Air Force Base Weather Station. (0 h represents
0000 on 27 August and 120 h represents 2400 on 31 August.)
20 -
o
UJ
HI
cr
O
g 15
O
Q.
LU
° 10 H
/K/v j
20
40
60
TIME - h
80
100
120
Figure 8. Time series of the dew point temperature measured at the Griffiss Air Force Base Weather Stadon.
(0 h represents 0000 on 27 August and 120 h represents 2400 on 31 August.)
6.3.2 Height-to-Volume Measurements
As shown in Figure 4, an experimental estimate of the height-to-volume conversion factor
was made at each level of product (near the end of each run) and compared to the theoretical
27
-------
estimate made from the tank chart. Except for the data collected at 291 cm d 14.5 in.) on
27 August, all of the calibration measurements were done with a 2,470-ml bar. The bar was
carefully inserted into and removed from the tank at 90-s intervals; to ensure that the bar was
completely immersed in or removed from the liquid, and to minimize the large initial waves
produced at the time the bar was inserted or removed, only the data from the last 60 s of each
90-s interval were used in the analysis. Ten or more repetitions were done at each level of
product. The mean level change was calculated from the absolute value of the difference
between the two levels (i.e., the level when the bar was in the liquid and the level when it was
out). An example of the mean level changes induced by the bar is shown in Figure 9. The mean,
standard deviation, and number of repetitions at each level are presented in Table 5. Except for
the height-to-volume conversion factor measured on 31 August, the difference between the
experimental and theoretical estimates was within 5%.
i
0.048 -
0.046 -
0.044 -
0.042 -
TIME - CYCLES
Figure 9. Average level change computed from the heieht-to-volume calibration data collected at 0730 on
30 August 1990.
Table 5. Summary of the Height-to- Volume Calibration Results
Tank Date
1 8-27-90
1 8-28-90
1 8-29-90
1 8-30-90
2 8-30-90
2 8-31-90
Start Time
(h)
1406
0835
0913
0736
1312
1315
Nominal
Level
(cm (in.))
290.8(114.5)
262.3 (103.25)
293.4(115.5)
251.8(99.125)
213.4(84.0)
283.8(111.75)
Measured Standard Tank
HVC Deviation Chart HVC Difference
(L/cm (L/cm (L/cm (%)
(gal/in.)) (gal/in.)) gal/in.))
-
577 (386.69)
401 (268.74)
643 (430.93)
702 (470.47)
423 (283.49)
435 (291.53)
579 (388.04)
417 (279.47)
621 (416.19)
712(477.17)
475(318.34)
-
0.9
4.0
3.4
1.4
12.3
28
-------
6.3.3 Level Data
Figure 10 shows the level data, after conversion to volume, that were obtained from the
two topping tests on 29 and 31 August and from the three overnight tests on 27, 29, and 30
August, initiated after a large amount of product had been added or removed. All three of the
overnight volume times series exhibit large fluctuations associated with the product addition or
removal for a period of 3 to 4 h. The data collected on 27 and 30 August show a minimum of
1 to 2 h of fluctuations before midnight (24 h). The data collected on 29 August show a distinct
change in slope at a point approximately 23.5 h after the start of the experiment. The volume
changes decreased continuously between 18 and 23.5 h and increased continuously between 23.5
and 30 h after the start. This change in slope appears correlated with the change in slope of the
product temperature data on the arm of the thermistor arrays and occurs about the time that the
temperature of the vapor in the upper portion of the tank drops below the temperature of the
vapor near the product surface and the product in the upper portion of the tank.
(a)
9.5
to.s
11.5
12.5
13.5
TIME -h
1S
13 •
12 •
11 •
31 AUQ
(b)
10
12
TIME • h
Figure 10. Time series of the volume changes measured with the level sensor beginning immediately after the
addition of a small volume of (a) cold product to Tank 1 on 29 August and (b) warm product to Tank 2 on 31
August: time series of the volume changes measured with the level sensor beginning immediately after the initial
level change done at (c) 1510 on 27 August, (d) 1441 on 29 August, and (e) 1505 on 30 August.
29
-------
60 -
i 4.H
20 •
15
60
50
40 -
-i
uj 30 •
§ 20 •
10 •
•10
•10 •
•15
•20
15
17
27AUQ
(c)
19
23
TIME • h
27
29AUQ
(d)
14 1« 18 20 22 24 26 21 30 32
TIME -h
30AUQ
19
23 25
TIME • h
27
29
31
33
Figure 10 (concluded). Time series of the volume changes measured with the level sensor beginning immediately
after the addition of a small volume 'of (a) cold product to Tank 1 on 29 August and (b) warm product to Tank 2 on
31 August; time series of the volume changes measured with the level sensor beginning immediately after the initial
level change done at (c) 1510 on 27 August, (d) 1441 on 29 August, and (e) 1505 on 30 August.
6.3.4 Surface Waves
An experiment was conducted to determine the periods of the long waves that might be
present in the tank. A 5,045-ml (1.3-gal) cylindrical bar 8.9 cm (3.5 in.) in diameter and 81.3 cm
30
-------
(32.0 in.) long was inserted and removed with a rapid, gliding motion at the product surface. A
2048-s (34.1-min) time series of level data, which is shown in Figure 11 (a), was collected at a
1-Hz sample rate (1 sample/s) beginning approximately 2 min before the bar was inserted. The
peak-to-peak fluctuation level was approximately 0.75 L (0.2 gal) before the bar was inserted.
After the bar was removed, the peak-to-peak fluctuation level increased to approximately 4.0 L
(1 gal), the equivalent of a height change of 0.0095 cm (0.004 in.) and then decreased
exponentially over the next 30 min to about 0.5 L (0.13 gal), the equivalent of a height change of
0.0012 cm (0.000468 in.).
S
12
10
8 -
i i i i i
(a)
10
20 30
TIME - h
40
9.40
9.20
C 9.00
8.80 -
8.60
(b)
35.00 35.50 36.00 36.50 37.00
TIME • h
Figure 11. Surface-level fluctuations produced by an impulse (a) over a 34-min period after the impulse and
(b) over a 2-min period between 35 and 37 min after the impulse.
31
-------
Periodic fluctuations of 1 to 3 min can be seen in the data shown in Figure 11 (a). Figure
11 (b), which is a 2-min segment of the data recorded between 35 and 37 min after the impulse,
shows periodic fluctuations of approximately 4 and 8 s. The frequency content is too
complicated, however, to allow interpretation of the type of waves that are present and their
periods in the time domain. This is done in Section 7.1, where frequency analysis of the time
series is discussed. Waves with periods less than 10 s are produced by disturbances of the
surface, while the waves with periods of several minutes to tens of minutes are generally a
manifestation of internal waves (subsurface waves generated in regions of strong vertical
temperature gradients). In general, internal waves do not usually affect the surface unless the
temperature gradient is very large, as it was during these experiments. Because of the strong
vertical gradients evident near the bottom and top of the tank, it is possible that internal waves
were present at each location.
As shown in Figure 10, the level data collected during the overnight test beginning on
30 August show a range of wave fluctuation, even though no manmade disturbances of the tank
environment occurred. The peak-to-peak fluctuation level of selected waves from a period in
which the amplitude of the surface fluctuations was small (21 to 23 h after the start of the
experiment) and a period when the amplitude of the surface fluctuations was large (27 to 28 h
after the start) are 0.25 L and 2.0 L, respectively. The large-amplitude waves are comparable to
those produced by a manmade disturbance and are generally associated with the wind blowing
over the vent tube.
6.3.5 Temperature Data
Time series of the temperature data obtained with Arrays A and B can be found in
Appendices A through E, which correspond to each of the five tests shown in Figure 4 and
Table 4. The data for each test are grouped as follows: (1) air temperature, (2) vapor
temperature, (3) product temperature for thermistors on the vertical portion of the array, and
(4) product temperature for thermistors on the horizontal portion of the array. Figures 12
through 15 show the time series of the data from 29 August.
A number of observations can be made directly from the raw temperature data collected on
29 August; these observations are also valid for the data collected on 27 and 30 August.
• The rate of change of temperature is greatest near the bottom of the tank. The rate of
change of temperature measured by the bottom thermistor (No. 20) is significantly
greater than that measured by the thermistor located immediately above it. This
suggests that additional thermistors would be required to accurately estimate the mean
rate of change of temperature in this region of the tank.
32
-------
o
in
UJ
in
tr
a
01
Q
UJ
ff
tt
UJ
Q.
UJ
tr
s
a.
o
co
ui
ui
cr
o
UJ
a
ui
tr
i
i
20.50
20.00
19.50
19.00
18.50
21.80
•**-,
(a)
10
20
TIME - h
30
O 21.60
03
ui
UJ 21.40
Q
21.20
21.00
20.80
i
(b)
10
20
TIME - h
30
21.00
20.60
20.60
H*
I
(c)
10
20
TIME - h
30
Figure 12. Time series of the product temperature changes on the vertical portion of Array A (with Thermistor 19
from Array B substituted for 18 on Array A) from the dara collected after the product removal at 1441 on 29 August:
(a) shows times series for Thermistors 20, 22 and 23, (b) for Thermistors 17 and 19, and (c) for Thermistors 14, 16
and 24.
33
-------
s
W
0 S33HO3C
LU 21.20
i
2 21.00
1-
20.80
hi ' - — """"
fa,
r
— '- 21
- 2S
_ 28AUO
L >
1 1 1 1 1
10 20 30
TIME • h
(a)
21.40
O
W
HI
HI
CC
O
0 21.20
LU
3
>
LU
1 21.00
•-
20 80
^" ""
^.---•-
t '
i; _
. |L • — •
— 15
~ •• 25
•— ~ 18
— — ' a *uo
I I I I I
10 20 30
TIME - h
(b)
Figure 13. Time series of the product temperature changes on the horizontal arm extending from the center of the
tank (a) on Array A and (b) on Array B. The arm, located at a height of approximately 180 cm (71 in.) from the
bottom of the tank, is located at the midpoint between Thermistors 16 and 18 on Array A and Thermistors 2 and 19
on Array B.
• The rate of change of temperature decreases in the lower portion of the tank and
increases in the upper portion of the tank. This occurs between Thermistors 14 and 16,
about 135 and 165 cm (53 and 65 in.) from the bottom of the tank, in all of the
overnight data collected on 27, 29, and 30 August.
• The temperature and the rate of change of temperature measured on the vertical portion
of Arrays A and B are nearly the same, suggesting that there are no horizontal
gradients in temperature along the long axis of the tank and that the temperature
measurement required for temperature compensation can be made at any convenient
location in the tank. This was true even when Array B was located within 0.8 m of the
end of the tank.
34
-------
E3 22
o
LU
Q
£ 18
14
10
J I
(a)
10
20
TIME - h
30
28
O 26
en
uj
01
24
I
20
18
(b)
10
20
TIME - h
30
Figure 14. Time series of Array A thermistors measuring (a) air temperature and (b) vapor temperature obtained
from data collected after product removal at 1441 on 29 August.
• The temperature and the rate of change of temperature measured on the horizontal arms
extending from the midpoints of Arrays B (Thermistors 15, 25, 12) and A (Thermistors
27, 21, and 26) show that there is a horizontal gradient in the region between the tank
wall and the centerline, and that it may be large enough to significantly affect the
accuracy of temperature compensation if only the vertical portion of the array is used.
The error could not be quantified given that there was only one horizontal arm on each
array. Several arms at different heights would have been necessary.
• The rate of change of temperature in the vapor space of the tank, as measured by
Arrays A and B in the 76-cm (30-in.) -diameter manways, reflected the rate of change
of the ambient air temperature; this was as expected. On all three nights, 27, 29, and
30 August, the temperature of the vapor near the top of the tank (as measured by
Thermistor 13 on Array A and Thermistor 0 on Array B) dropped below the
temperature of the vapor closer to the product surface (as measured by Thermistor 11
on Array A and Thermistor 9 on Array B) and the upper layers of the product in the
tank (as measured by Thermistors 17 or 10 on Array A and Thermistors 5 or 6 on
Array B). Array B, located in the 10-cm (4-in.) -diameter opening at location C, which
is 5.5 m (18 ft) away from Manway B containing Array A, did not show this same
35
-------
behavior during tests initiated on 27 and 2Q August. This suggests that the
communication between the air and the tank is localized to the manway. This makes it
difficult to interpret the influence of the vapor space on evaporation and condensation.
6.3.6 Temperature Profiles
To illustrate the gradient, profiles of the temperature field during the three overnight tests
were generated from the data from Array A. (A 4-h wait after any addition or removal of
product during the tests allowed the strong temperature fluctuations associated with such level
changes to subside.) These profiles are shown in Figure 15. All three profiles are similar and are
consistent with summer ground conditions. There is a very strong gradient in the bottom 50 cm
(20 in.) of the tank and another near the top of the tank. The strength of the gradient near the
bottom of the tank suggests that additional thermistors are necessary if the rate of change of
temperature is to be accurately measured. During the summer the thermistors must be more
densely spaced than they would have to be during the winter, when the profile would be more
uniform from the top to the bottom of the tank.
The strong gradients support the propagation of internal waves. The volume and thermally
induced volume data collected after 19 L (5 gal) of warm product was added to the tank at 0820
on 31 August show a strong 6-min periodicity in the data, an observation consistent with the
presence of internal waves.
6.3.7 Thermal Volume Time Series
The data source for all estimates of the thermally induced volume changes in the tank was
Array A. These time series were developed from Eq. (1) and are shown in Figure 16 for the
three overnight tests on 27, 29, and 30 August and in Figure 17 for the two topping tests on 29
and 31 August. Data collection in all five cases started immediately after the level change had
been completed. Since, as evidenced in the calibration, the ability of Thermistor 18 on Array A
to measure temperatures above 20°C was suspect. Thermistor 19 on Array B was used in its
place. Because of the horizontal gradients in the rate of change of temperature along the long
axis of the tank, it was assumed that the temperature measured by Thermistor 19 would be in
good agreement with that measured by Thermistor 18 after the strong initial temperature
fluctuations had subsided.
36
-------
§
t-
X
X
300
200 -
(a)
19.1 19.5 19.9 20.3 20.7 21.1
TEMPERTURE - DEGREES C
300
200 •
29AUQ
(b)
19 19.4 19.8 20.2 20.8 21 21.4 21.8
TEMPERATURE - DEGREES C
300
200 •
100 -
30AUQ
(C)
19.5
19.9 20.3 20.7
TEMPERATURE • DEGREES C
21.1
Figure 15. Vertical temperature profile computed 4 h after the initial level change at (a) 1510 on 27 August,
(b) 1441 on 29 August, and (c) 1505 on 30 August.
37
-------
20
27AUG
(a)
14 16 18 20 22 24 26 28 30 32
14 18 11 20 22 24 26 28 30 32
_j -20 •
S -40 •
40 •
•80 •
1
-
4
30AUO
#v~
16 18 20 22 24 28 28 30 3
TIME-h
Figure 16. Time series of the thermally induced volume changes estimated from Array A beginning OS to 2 h
before the initial level change done at (a) 1510 on 27 August, (b) 1441 on 29 August, and (c) 1505 on 30 August;
time series of the thermally induced volume changes estimated from Array A beginning immediately after the initial
level change done at (d) 1510 on 27 August, (e) 1441 on 29 August, and (f) 1505 on 30 August.
38
-------
20
S
I
27AUG
(d)
15
23
TIME • h
31
14 16 18 20 22 24 26 2S 30 32
•74
•78
•78
•80
(f)
15
1«
23
TIME • h
27
31
Figure 16 (concluded). Time series of the thermally induced volume changes estimated from Array A beginning
OJ to 2 ft before the initial level change done at (a) 1510 on 27 August, (b) 1441 on 29 August, and (c) 1505 on
30 August: time series of the thermally induced volume changes estimated from Array A beginning immediately
after the initial level change done at (d) 1510 on 27 August, (e) 1441 on 29 August, and (f) 1505 on 30 August.
39
-------
I
0 -
-0.2 -
-0.4 -
-0.6 -
-0.8 -
-1 -
-1.2 -
-1.4 -
-1.6 -
-1.8 -
29AUQ
(a)
13
|
(b)
-0.9 -
-1.2 -
Figure 17. Time series of the thermally induced volume changes estimated from Array A beginning immediately
after the addition of a small volume of (a) cold product to Tank 1 on 29 August and (b) warm product to Tank 2 on
31 August.
The temperature-volume time series developed after the product removals on 27 and 29
August show fluctuations of about 7.5 and 12 L (1 and 3 gal) during the 16-h test. The large
volume fluctuations observed during the first 5 h of the time series are associated with vertical
and horizontal mixing after product had been removed. The presence of large-amplitude,
periodic internal waves is easily seen during the first 6 h in the 27 August time series; the period
of these waves is approximately 7 min. The rate of change of volume is relatively large and is
decreasing slowly over time.
The time series of the thermally induced volume changes estimated for the test begun after
the product addition on 30 August shows an increase of about 4.5 L (1.2 gal) over the first 12 h
before it levels outs abruptly at 2700. The increased level of temperature fluctuations is also
observed during the first 6 h after the product addition. The abrupt shift in the thermally induced
volume changes seems to be correlated with the abrupt shift in the temperature of the vapor
during the same period.
40
-------
6.3.8 Temperature-Compensated Volume Time Series
The temperature-compensated volume time series were compiled from the data from
Array A. Figure 18 shows the time series for the topping tests initiated on 29 and 31 August, in
which warm or cold product was added. Figure 19 shows the time series for the three overnight
tests initiated on 27, 29 and 30 August, each of which followed a large product addition or
removal. Several observations can be made about the temperature-compensated-volume time
series shown in Figure 19.
• The effects of the strong temperature fluctuations that are associated with the addition
or removal of product from the tank are present during the first 3 to 4 h of each time
series.
• The distinct changes of volume observed in the volume time series were not removed
by temperature compensation with the vertical portion of Array A.
• The fluctuations in volume have periods of 2 to 4 h.
• The data collected from Tank 1 on 27 and 29 August show that the rate of change of
volume is still increasing long after the temperature effects due to product addition or
removal have subsided; the data collected from Tank 2 on 30 August show that it is
decreasing.
(a)
9 It
TIME-h
(b)
10 S 115
TIME-It
Figure 18. Temperature-compensated volume time series computed after the addition of a small volume of (a) cold
product to Tank 1 on 29 August and (b) warm product to Tank 2 on 31 August.
41
-------
1* 16 18 20 22 24 26 28 30 32
15 17 19 21 23 25 27 29 31 33
Figure 19. Time series of the temperature-compensated volume changes estimated with Array A beginning
immediately after the initial level change at (a) 1510 on 27 August, (b) 1441 on 29 August, and (c) 1505 on
30 August.
42
-------
SECTION 7
SOURCES OF AMBIENT NOISE
An analysis of the data gathered in the experiments is presented below.
7.1 Surface Waves
Any disturbance of the product surface will generate a seich in an underground tank. A
seich is a long wave whose fundamental period and higher harmonics are controlled by the
boundaries of the basin. Waves will propagate both along the long axis of the tank and across its
short axis. In a closed tank, wind blowing across the vent tube will create pressure fluctuations
in the tank, which will generate waves. Any manmade disturbance, such as those that occur
when equipment is inserted into or removed from the tank, will also create waves. Long waves
can also be present in the product at places where the temperature (i.e., density) is high. At
times, these internal waves can have enough energy to effect surface waves.
The fundamental frequency of the waves (and their higher harmonics) is predictable [13].
The period, T, of long waves in a rectangular basin in which the width is very much smaller than
the length can be estimated by
T = (2L)/(gh)", (1)
where n = 1, 2, 3,..., L is the length of the tank (23.6 m (77.5 ft)), h is the height of the product
in the tank (284 cm (111.75 in.)), and g is the acceleration due to gravity. In the Griffiss
experiments, the fundamental frequency (n = 1) estimated with Eq. (1) is 8.95 s. It is also
possible to predict the frequency of the surface waves produced by internal waves, but this is
beyond the scope of this project.
The spectra indicate the presence of highly periodic waves in the 0.1- to 0.5-Hz region
(2 to 10 s), typical of surface waves, and in the 0.001- to 0.1-Hz region (10 to 1000 s), typical of
internal waves. The spectra in Figure 20 were smoothed in the frequency domain with a
10-point running average and were used to estimate the period of the waves with frequencies
between 0.001 and 0.1 Hz. The period of the waves in the 0.1- to 0.5-Hz region was estimated
with the spectrum after smoothing with a 25-point running average. Table 6 summarizes the
periods of the main waves present in the tank. The spectra suggest that the fundamental (n = 1)
and the first harmonic (n = 2) of waves propagating along both the long and short axes of the
tank are present. The uncertainty in estimating the spectral peaks is several tenths of a second.
In all cases the first harmonic (n = 2) is the larger wave. The fundamental and first harmonics
for each pair of waves is also noted in Table 6. The experimental estimate of n for the first
harmonic, which is estimated by dividing periods of the fundamental and the first harmonic, is
43
-------
also shown. The predominant period of the surface waves, which is between 8 1 and 8.4 s. is in
reasonable agreement with the 9.0-s prediction, for the fundamental period of waves moving back
and forth along the long axis of the tank. Higher harmonics of the fundamental can be obtained
by dividing the fundamental by 2, 3, 4, etc.
10
o
LLJ
Q.
2
1
1
2
2
-
Meas.
n
-
.
-
1.98
1.94
..
-
1.93
1.84
-
Generation Mechanism
Internal Wave
Internal Wave
Internal Wave
Internal Wave
Internal Wave
Surface Movement
Surface Movement
Surface Movement
Surface Movement
Surface Movement
Because the amplitude of the waves can be large, equivalent to several liters of product, it
is important during a leak detection test to sample the data at a high enough frequency to avoid
aliasing, and to average the data over a long enough time to make an accurate estimate of the
trend. To minimize the effects of aliasing, the data should be sampled at 1 s or better and
averaged to 1 min. This will avoid aliasing the surface undulations produced by internal waves.
44
-------
7.2 Temperature Inhomogeneities after Product Additions or Removals
Any addition and removal of product will alter the temperature field in an underground
tank. The horizontal and vertical mixing of the product creates large temperature fluctuations
that last for many hours. The horizontal gradients that develop make it difficult to accurately
compensate for the thermally induced volume changes, because one array is not sufficient to
measure the average temperature in the tank. The strong vertical mixing of product also prevents
accurate estimates of the mean temperature changes in the tank, because the fluctuations are too
large to permit an accurate estimate of the rate of change of temperature. These large
temperature fluctuations can be observed in the raw temperature time series, the thermally
induced volume time series, and the temperature-compensated time series. The temperature-
compensated time series is also affected by the evaporation and condensation that occurs during
the period of strong gradients because of the nonsaturated condition that is created as vapor is
pushed out of the tank or air is drawn into the tank. Evaporation and condensation can be
significant until a saturated condition is re-established in the tank; although no measurements of
the evaporation and condensation were made in these experiments, it is known that a saturated
condition will generally not be reached until fluctuations in product temperature have subsided.
7.2.1 Temperature Inhomogeneities after a Product Delivery or Product Transfer
The temperature fluctuations that occur immediately after product has been added to or
removed from the tank can be clearly seen in the thermally induced volume changes shown in
Figure 16; these are from the overnight tests conducted on 27, 29 and 30 August. The
temperature fluctuations can also be observed in the raw temperature time series presented in
Appendices A through C. The temperature fluctuations are particularly large for the first 3 to
4 h. During this period, any attempt to compensate for the thermally induced volume changes
would result in large errors. This can be verified by inspection of the temperature-compensated
volume time series presented for these three tests in Figure 18.
The level and duration of the temperature fluctuations observed in these tests are nearly
identical to those observed in tests conducted on the 30,000-L (8,000-gal) tanks at the Test
Apparatus and presented in [3, 4]. The data suggest that it takes at least 4 h for the temperature
fluctuations to subside.
7.2.2 Temperature Inhomogeneities after Topping
Topping is a common practice with all volumetric leak detection systems that require that
the tank be overfilled for a test. The temperature effects due to topping were investigated by
adding 19 L (5 gal) of product that was approximately 25°C warmer or 8°C cooler than the
product in a partially filled tank. The warm topping experiment was conducted at the end of the
45
-------
overnight test that began on 29 August; at the time of the topping nearly 21 h had elapsed since
the beginning of the test. The cold topping experiment was conducted at the end of the overnight
test that began on 30 August; in this case nearly 17 h had elapsed since the beginning of the test.
Figure 21 shows raster displays of the temperatures measured on Array A. In these displays, the
mean from the time series of each thermistor was removed and an offset of 0.02°C greater than
that of the thermistor below it was added. In Figure 21 (a) and (b) the temperature fluctuations
take approximately 2 h to subside after the addition of the cool product, and in Figure 21 (c) and
(d) they take slightly longer than 3 h to subside. The temperature fluctuations can also be
observed in the thermally induced volume time series shown in Figure 17 and the raw
temperature time series presented in Appendices D and E. It is important to note that when the
cold and warm topping tests were done, the tanks had not been disturbed for a long period of
time. During this time, the mean rate of the thermally induced volume change had decayed to
less than 0.38 L/h (0.1 gal/h), and it is safe to assume that the volume changes due to residual
deformation had become negligible. As a consequence, it was expected that the temperature-
compensated volume rate would approach zero.
Figures 17 and 19 present time series of the thermally induced volume changes and the
temperature-compensated volume changes for both topping experiments. When a waiting period
of 2 h was used, the temperature-compensated volume rate for the 29 August cold topping test
was 0.04 L (0.01 gal/h), and for the 31 August warm topping test, -0.04 L/h (-0.01 gal/h). When
there was no waiting period, the temperature-compensated volume rates were -0.53 L/h
(-0.14 gal/h) and -0.41 L/h (-0.11 gal/h), respectively. Part of the error in estimating the
temperature-compensated volume rate during the first 2 h may be due to large volume changes
that are effected by evaporation and condensation, which occur when the saturated vapor in the
tank is disturbed (when the tank is opened to the ambient environment so that product can be
added).
Since the data used in this analysis came from a partially filled tank, the volume of product
added to the tank was not large enough to produce a level change that would induce significant
deformation. In an overfilled tank, however, the addition of even a small volume of product can
produce a large level change, and the effects of deformation must therefore be accounted for
during a leak detection test. As in tests on the 30,000-L (8,000-gal) tanks at the Test Apparatus,
temperature fluctuations of equal magnitude were observed in both arrays, which were separated
by a distance of 6.1 m (20 ft).
46
-------
(a)
024
022
02
0 18
0 IS
014
012
0 1
00»
006
(b)
TIME-h
Figure 21. Raster display of the temperature-compensated volume time series under each of two conditions on
different days: (a) cold product added to Tank 1 on 29 August and temperatures measured by upper four
thermistors; (b) cold product added to Tank 1 on 31 August and temperatures measured by upper four thermistors;
(c) warm product added to Tank 2 on 29 August and temperatures measured by lower five thermistors; and (d) warm
product added to Tank 2 on 31 August and temperatures measured by lower five thermistors.
7.3 Horizontal Gradients
Accurate temperature compensation requires an accurate estimate of the average rate of
change of temperature of the product in the tank. A single array of temperature sensors can be
used provided that the horizontal gradients in the rate of change of product temperature
throughout the tank are negligible.
7.3.1 Horizontal Gradient along the Long Axis of the Tank
To examine the horizontal gradients, the temperatures measured by each thermistor on
Array A were differenced with those measured by the thermistors on Array B that were located
at the same height. The horizontal separation between Arrays A and B was either 5.5 or 8.2 m
(18 or 27 ft), depending on the test. Individual time series of the temperature differences during
the overnight tests conducted on 27, 29, and 30 August are presented in Appendices A through
C. The largest differences in temperature were generally less than ±0.001°C/h and showed no
obvious bias in the vertical. Some differences showed an increase and some showed a decrease,
47
-------
suggesting that the errors are randomly distributed in the vertical. Because Thermistors 18, 1,
and 2 were out of calibration, no estimate of the thermally induced volume changes could be
calculated. However, the sum of all temperature differences along the vertical axis would
probably be less than 0.001°C/h, which corresponds to an error of less than 0.19 L/h (0.05 gal/h)
in a 190,000-L (50,000-gal) tank filled with product.
Thermistor 19 on Array B was used in place of Thermistor 18 on Array A for all estimates
of the thermally induced volume. This substitution results in an error of less than 0.023 L/h
(0.006 gal/h), as can be seen if the differences between Thermistors 17 and 5 (the two located
directly above Thermistors 18 and 19) are measured.
7.3.2 Horizontal Gradient along the Short Axis of the Tank
The horizontal gradient in temperature and the horizontal gradient in the rate of change of
temperature between the centerline of the tank and the wall of the tank were examined by
comparing the temperature time series on the horizontal arm of each thermistor array. In all
three overnight tests there is a difference of several thousandths of a degree per hour between
Thermistor 27 on Array A and Thermistors 21 and 26 on Array A's horizontal arm (the two
thermistors closest to the wall) and between Thermistor 15 on Array B and Thermistors 25 and
12 on Array B's horizontal arm (again the two thermistors closest to the wall). Large differences
might be expected between the thermistors near the wall and those near the centerline. Because
of the large volume of product, however, it is significant that there is a difference in the rate of
change of temperature measured by the thermistors located in the middle of each arm
(Thermistors 21 on Array A and 25 on Array B) and those closer to the center of the tank
(Thermistors 27 on A and 15 on B). Figure 13 illustrates the amount of this difference during the
overnight test initiated on 29 August; similar time series displays for the other two overnight
tests (27 and 30 August) are presented in Appendices A and C. The thermistor closest to the
wall shows that the mean temperature of the product and presumably that of the ground outside
the tank is warmer than that of the product near the center of the tank. This explains why the
temperature of the product is increasing in this region of the tank. Modeling calculations suggest
that the differences in the rate of change of temperature will decrease over time, even though a
difference in the mean temperature persists. This can be observed in the thermistor data taken
during the topping experiments on 29 and 31 August. Except during the first 2 h after topping,
the differences in the rate of change of temperature between thermistors on the arm are small.
The extent of the differences between the centerline and the first thermistor on the arm
(Thermistor 27 on Array A and Thermistor 15 on Array B) cannot be accurately estimated
because there is no thermistor located at the center of the tank at the same height as the arm. It
48
-------
was assumed that the rate of change of temperature at the centerline could be estimated by
averaging the temperatures measured by the two thermistors bracketing the arm. Unfortunately,
the arm was located in the region of the tank where the vertical gradient is large and the accuracy
of the average is not known. The thermistor on the vertical portion of the array that is directly
above the arm showed an increase in temperature over time and the thermistor below the arm
showed a decrease.
7.4 Thermistor Spacing
The thermally induced volume changes estimated from thermistors spaced on 61-cm
(24-in.) centers were calculated and compared to the estimates made from thermistors spaced on
30-cm (12-in.) centers. Array A was used in the calculations. The first calculation began with
the thermistor closest to the bottom of the tank (Thermistor 20). The total number of thermistors
in each estimate depended on the level of the product in the tank. Estimates made from the
overnight test initiated on 29 August were based on 4 and 8 thermistors while those made from
the tests initiated on 29 and 30 August were based on 5 and 9 thermistors. Figure 22 shows the
thermally induced volume changes for the estimates made from the thermistors having a 61-cm
(24-in.) spacing. A comparison of Figures 16 and 22 shows that there are large differences in the
volume changes regardless of the spacing of the thermistors. The thermally induced volume
time series made with 5 thermistors, from the test initiated on 27 August, suggests that the
volume changes level off after 5 h; however, the time series made with 9 thermistors continues to
show large volume changes. Slightly better agreement might occur if the bottom thermistor of
the array having 61-cm (24-in.) spacing between thermistors were located 30 cm (12 in.) from
the bottom of the tank rather than 13 cm (5 in.). The analysis suggests that 5 thermistors do not
provide adequate thermal compensation in a 190,000-L (50,000-gal) tank.
49
-------
*u -
O 30 -
CO
IU
g
o
Q 20 -
g
3
K
1 -
3
H
1
1 27 AUG
mt
( ^^^^..j
i 19 23 27 31
TIME • h
(a)
35 -
30H
29 AUG
(b)
14 16 18 20 22 24 26 28 30 32
(C)
31
Figure 22. Time series of the thermally induced volume changes estimated by thermistors spaced at 24-in. intervals
on Array A beginning immediately .after the initial level change at (a) 1510 on 27 August, (b) 1441 on 29 August,
and (c) 1505 on 30 August.
50
-------
7.5 Structural Deformation
If deformation is an important source of error in detecting leaks from a particular tank, the
temperature-compensated volume times series will exhibit an exponential change in volume
immediately after the level of the product in the tank has been changed [2, 3,6, 13]. This
behavior was not observed in the temperature-compensated volume time series of the three tests
initiated on 27, 29, and 30 August, in which level changes of 29, 33, and 70 cm (11.25, 12.94,
and 27.75 in.) were produced. These time series are shown in Figure 18. The strong temperature
inhomogeneities produced immediately after product has been added or removed make it
impossible to fit an exponential curve to the data to quantitatively estimate the time constant of
the deformation (the time constant is the time required for 63% of the total volume change to
occur). Also, as discussed in more detail in Section 7.8, the temperature-compensated volume
estimates made for the first 10 h after the product addition or removal are controlled by the
exponential error in compensating for the large thermally induced volume change in the layer
closest to the bottom of the tank. Clearly, the dominating noise source centers around the
thermally induced volume changes. A qualitative inspection of the data suggests that the time
constant cannot be more than several hours, which is about the same as that observed in the
30,000-L (8,000-gal) tanks at EPA's UST Test Apparatus. These data suggest that after a
waiting period of 6 to 10 h, structural deformation will no longer affect estimates of the
temperature-compensated volume rate. It is important to note that the deformation observed in
the large tanks at Griffiss Air Force Base does not differ significantly from that observed in tanks
of smaller capacity. This is not unexpected; while a 190,000-L (50,000-gal) tank might be larger
by a factor of 8 than a 30,000-L (8,000-gal) tank, important aspects of the tank's geometry,
particularly its diameter, which controls end and wall deflection, are within 25% regardless of
the size of the tank. The native soil surrounding the tanks in the Griffiss experiments is similar
to the soil found around the tanks at the UST Test Apparatus. With only one set of tanks, it is
not possible to generalize too much about the deformation. However, there is no indication that
deformation is significantly greater in large tanks than it is in smaller tanks.
7.6 Evaporation and Condensation
The volume changes due to evaporation and condensation at the vapor/product and the
wall/product interfaces are extremely difficult to quantify. At the present time, there is no simple
measurement or combination of measurements that can be used reliably to identify or quantify
the volume changes due to evaporation and condensation in an underground storage tank. This
process is extremely complicated because it is not just the evaporation and condensation from the
product surface that must be predicted, but the losses and gains from the tank walls. In general,
contributions from the walls can be much larger than those from the surface. If the vapoj is
51
-------
saturated, evaporation would be expected to increase as the temperature of the vapor relative to
the temperature of the product increased or the pressure within the vapor space increased. If the
vapor is not saturated, simple temperature and pressure measurements may not be sufficient to
even indicate that volume changes due to evaporation and condensation are present.
The presence of evaporation or condensation can only be identified if the error in
compensating for the thermal expansion and contraction of the product is small. Its presence is
best identified when there is a distinct shift—one that cannot be explained by thermal
fluctuations—in the measured temperature-compensated volume changes, or when there is a
strong correlation between temperature and pressure measurements in the vapor space of the
tank. The magnitude of these volume changes is best determined (1) after the volume time series
has been accurately temperature-compensated, (2) after the effects of deformation and the
temperature fluctuations associated with product addition or removal have subsided, and (3) after
the vapors lost or gained when product is added or removed have reached a saturated condition.
It is assumed that there are no other mechanisms, except for evaporation and condensation, that
can generate large volume changes.
All of the data were extensively analyzed to determine if there were any simple and
consistent correlations between atmospheric pressure, air temperature, vapor temperature, wind
speed, and evaporation or condensation that were large enough to impact the accuracy of a leak
detection test. The vapor temperatures measured in the 76-cm (30-in.) -diameter manways
suggested that condensation was a dominant process and was responsible for the increase in
volume still evident even 24 h after the start of each of the tests. Inspection of the vapor
temperatures measured during each of the overnight tests, especially the 30 August test, shows a
large decrease in the temperature within the vapor space, with the temperature of the vapor
becoming colder than that of the product near the surface after 24 h. The data also indicate that
the temperature in the vapor space becomes colder near the top of the tank than near the product
surface, creating an unstable condition. The time at which the vapor temperatures became
unstable or dropped below the product temperature is correlated with the time at which the
volume changes dramatically increased. Similar temperatures were observed at both arrays
during the 30 August test, when the arrays were located in 76-cm (30-in.) -diameter manways,
but not during the 27 and 29 August tests, when one of the arrays was located in a 10-cm (4-in.)
-diameter fill hole. Figure 23 illustrates these differences with data from the same thermistors on
29 and 31 August. The lowest thermistor in the vapor space was located within 5 cm (2 in.) of
the surface, and the highest thermistor immersed in the liquid was located 25 cm (10 in.) below
the product surface. The data collected on 29 August, from the thermistor array positioned in the
52
-------
smaller opening, suggest that the temperatures measured in the manways are localized in those
manways. The array located in the manway detected an unstable temperature condition and a
drop in the temperature of the vapor below that of the product, an indication that condensation
was occurring. This did not happen when the array was placed in the smaller opening. While
condensation was probably occurring at each of the three manways, it is not clear that
condensation was also occurring at other locations in the tank. As a consequence, the vapor
temperatures measured at the smaller opening were probably more representative of the vapor
temperature within the tank as a whole. This is also suggested by the lack of correlation
between the vapor temperature fluctuations and the volume fluctuations.
til
HI
cr
UJ
HI
o
111
-------
7.7 Internal Waves
A frequency-domain analysis of the temperature data collected by the thermistors near the
top and bottom of the tank suggests the presence of large-amplitude internal waves with periods
ranging from 3 to approximately 30 min. Power spectra of the average of three thermistors
during the warm topping test done on 31 August showed clearly defined spectral peaks at 3, 7,
18, and 25 min, or longer. Waves shorter than 2 min could not be observed because they were
sampled at 1-min intervals; that they were present, however, is suggested by the level data.
Slightly different wave periods were observed in the region near the top of the tank than in the
region near the bottom of the tank. However, the 3-min waves were present in both, and the
long-period waves were also present. These waves can introduce large errors if the data are not
sampled fast enough to avoid aliasing and if the duration of a test is not long enough to average
to two periods. The data suggest that a sample interval of 1 min and a test duration of several
hours will suffice.
7.8 Residual Volume Changes after Temperature Compensation (Overnight Tests)
Accurate temperature compensation requires an accurate estimate of the average rate of
change of temperature of the product in the tank. A single array temperature sensors can be used
provided that the horizontal gradients in the rate of change of product temperature throughout the
tank are negligible and that the vertical spacing of thermistors is dense enough to permit an
accurate estimate of the average rate of change in each layer. Each layer must be thin enough
that the change in temperature is linear within the layer and any errors in measuring the rate of
change of temperature within the layer are small in comparison to the rate of change of volume.
The largest errors occur when the layers are too thick or in those layers where the rate of change
of temperature is largest or changes sense, usually the layers nearest the bottom and top of the
tank. The larger the tank, the more significant these errors become, unless the volume of the
product in each layer can be minimized through the use of additional thermistors. Errors that can
affect the accuracy of temperature compensation are discussed in Section 9.
It is normally assumed that a leak detection test will result in an accurate estimate of the
leak rate (1) after the volume changes due to deformation, product temperature fluctuations, and
evaporation and condensation produced by product addition or removal have subsided, and
(2) after the temperature contribution to the measured volume changes has been compensated
for. For a partially filled tank, this assumption is valid provided that the effects of evaporation
and condensation at the vapor/product interface and at the wall/product interface are small during
a test. For an overfilled tank, the assumption is valid if the volume of trapped vapor is
negligible. Accurate temperature compensation requires that the horizontal gradients in the rate
54
-------
of change of temperature be small enough that a single array of thermistors at any location along
the centerline of the tank can be used for compensation. For an accurate estimate of the rate of
change of temperature, the volume of product around each thermistor must be reduced, and to do
this, an adequate number of thermistors is required.
In all three overnight tests, indications are that both tanks are tight, which means that the
measured volume rate should be 0.0 L/h (0.0 gal/h). As observed in Figure 18, during the first
4 to 6 h after a product addition or removal, the temperature fluctuations are too large to permit
an accurate estimate of leak rate or to allow us to determine if the effects of deformation are
important. All three overnight tests began in late afternoon, between 1500 and 1600. If the three
tests are examined in terms of a time line beginning at 0 h (the start of a test), it can be seen that
the rate of change of the temperature-compensated volume exhibits a distinct shift at 19 h
(approximately 4 h after the product addition or removal that occurred at 15 h) and again at 24 h
and continuing on to 30 h. Figure 24 shows the same temperature-compensated volume data
used to generate Figure 18, but enlarges an area of detail (the time between 18 h and 31 h). The
data in Figure 24 were analyzed to determine why two distinct compensated volume rates were
observed. The validity of the above-mentioned assumptions was investigated. The best estimate
of leak rate would be provided by the data gathered during the second period, when the waiting
period was longest; these estimates are presented in Table 7.
The temperature-compensated volume changes from the three overnight tests are discussed
separately.
Table 7. Summary of the Leak Detection Results for Overnight Tests
Tank
Start Date
Start Time
(h)
Nominal Level
(cm (in.))
TCVR
(L/h (gal/h))
1
1
2
8-27-90
8-29-90
8-30-90
24-29
24-29
25.5 - 29.5
262.3 (103.25)
252.1(99.25)
283.8(111.75)
0.36 (0.094)
0.66(0.172)
-0.22 (-0.057)
55
-------
18 20 22 24 28 28 30 32
28
22 -
20
120
29AUQ
(b)
22
26
TIME-h
18 20 22 24 29 28 30 32
TIME • h
Figure 24. Time series of the temperature-compensated volume changes estimated from Array A beginning
approximately 3 h after the initial level change done at (a) 1510 on 27 August, (b) 1441 on 29 August, and (c) 1505
on 30 August.
56
-------
7.8.1 Overnight Test Starting on 27 August 1990
The period between 18 and 23 h on the time line suggests that the volume rate is
approaching zero, as typically happens when deformation due to a rapid drop in product level is
subsiding. However, at about 23.5 h, the temperature-compensated volume rate begins to
increase and changes linearly at a rate of 0.31 L/h (0.08 gal/h) over a 6.5-h period between 24.5
and 31.0 h on the time line. This increase cannot be explained by deformation. It suggests that
another mechanism is controlling the volume rate during both periods. As seen in Figure 16 (a),
the thermally induced volume time series estimated from the vertical portion of Array A has a
nearly linear change of approximately -0.4 L/h (-0.1 gal/h) between 18 and 31 h, much too large
to account for the nearly 0-L/h (0-gal/h) volume changes observed in the volume time series
shown in Figure 10 (a). In addition, the thermally induced volume time series does not exhibit
the small-amplitude fluctuations observed in the volume time series. Inadequate temperature
compensation or condensation of product could explain the observed 0.36 L/h (0.09 gal/h)
temperature-compensated volume rate.
Since the tank was not leaking, the measured leak rate should have been 0 L/h. The
0.36-L/h (0.09-gal/h) error in measurement could have been reduced if the thermal volume
changes had not decreased at such a fast rate. The shaded portions in Figure 25 indicate those
regions of the tank where accurate temperature compensation is critical. In these regions, large
horizontal or vertical gradients in the rate of change of temperature or an insufficient number of
temperature sensors can produce errors large enough to affect the accuracy of temperature
compensation. An error in estimating the temperature changes in one or more of these regions is
the most likely cause of the 0.36-L/h error in the measurement of the leak rate.
In the overnight test initiated on 27 August, the largest errors in temperature compensation
were likely to occur in the region near the bottom of the tank and near the walls of the tank. The
error in the layer closest to the top of the tank was minimized because Thermistor 10 was located
close to the surface (within 5 cm (2 in.) of it); thus, the rate of change of temperature in the upper
10 cm (4 in.) of the tank, which is due to heat transfer at the vapor/liquid interface, was
accounted for. (In interpreting the overnight tests done on 29 and 30 August, it is important to
note that the rate of change of temperature measured by Thermistor 10 is much greater than that
measured by the thermistor located immediately below it.) As suggested by the time series of
temperature for Thermistor 20 in Appendix A, the bottom region of the tank can be a significant
source of error because the rate of change of temperature here is very large, as is the gradient in
the rate of change. One thermistor is not enough to estimate these gradients, given their size.
Inaccurate temperature measurement in the bottom layer explains almost all the error in the
57
-------
VAPOR SPACE
LAYERS CENTERED
ABOUT THERMISTORS
PRODUCT
AREAS OF
NONLINEAR
TEMPERATURE
CHANGE
Figure 25. Accurate temperature compensation requires that the average rate of change of temperature be measured
in four regions of the tank.
compensated volume rate during the first period, between 19 and 23.5 h on the time line. The
discrepancy in the rate of change of temperature measured by Thermistors 20 and 22, the two
closest to the bottom, suggests that the gradient in the bottom layer was large. Between 25 and
27 h on the time line, however, the discrepancy was negligible, and it can be assumed that by this
time the thermally induced volume changes in the bottom layer were being accurately estimated.
The error in the temperature-compensated volume rate during the second period might be
explained by the horizontal gradient in the rate of change of temperature observed between the
center of the tank and the wall, where an increase in the rate of change of temperature occurred
between 24 and 31 h.
The time series displays of the temperature measured by the horizontal arms of Arrays A
and B, shown in Appendix A, suggest that there are differences in the rate of change of
temperature between the centerline of the tank and the tank walls. Beginning at about 23.0 h, the
rate of change of temperature increased as the distance from the centerline to the wall increased;
prior to this time, the rate of change of temperature was fairly uniform. The rate of change of
temperature measured by Thermistor 21 was 0.0039°C/h, 0.0019°C/h greater than that measured
by Thermistor 27. This would account for 0.201 L/h (0.052 gal/h) of the error in the
58
-------
temperature-compensated volume rate if this temperature change included 76.000 L (20.000 gal)
of product. The rate of change measured by Thermistor 26 (the one closest to the wall) is about
twice that measured by Thermistor 21 (the next thermistor in). The estimated contribution would
be larger if it included the larger rate of change of temperature measured near the wall. This
alone can explain the majority of the error. With only one horizontal arm, no quantitative
estimate could be made. Additional arms at different levels along the vertical extent of the
temperature array would have been required. Fortunately, a waiting period can resolve the
problem; as noted in the discussion of the cold topping experiment, these horizontal gradients, as
well as the vertical gradients near the bottom of the tank, had disappeared 21 h after the addition
of product.
It is also possible that condensation, particularly from the wall, could have contributed to
the residual volume changes, but there is no simple way to' quantify the contribution or even to
confirm this notion. Observations of temperature from Array A, located in the 76-cm (30-in.)
manway at opening B, suggest that condensation occurred at 23.5 h on the time line, when the
temperature measured by Thermistor 13 (in the vapor space) dropped below that measured by
Thermistor 11 (located 12 in. below it), and also at 25.5 h when it dropped below that measured
by the thermistor in the upper 5 cm (2 in.) of the product. However, this was not observed at
Array B, located in the 10-cm (4-in.) -diameter opening at C. Here, the temperature measured by
the immersed thermistors closest to the surface and the one 30 cm (12 in.) above the surface
increased. This suggests that the temperature changes observed at the manway are not
representative of the overall vapor temperature throughout the tank. The rate of change of
temperature measured by the uppermost product thermistor was identical at both Array A and
Array B, suggesting that the temperature change was governed mainly by the temperature
differential between the product and the vapor. It is reasonable to conclude, therefore, that
measurements of vapor temperature made in the 76-cm (30-in.) -diameter manways are
representative of local conditions only and cannot be used as an indicator of vapor temperature in
the tank as a whole.
7.8.2 Overnight Test Starting on 29 August 1990
The overnight test initiated on 29 August was similar to the one initiated on 27 August
except that the product level was located 5 cm (2 in.) above Thermistor 10. Thus, the thermistor
closest to the surface, Thermistor 17, was located 25 cm (10 in.) below the surface. Based on the
observed rate of change in the 27 August data, it is likely that the temperature contributions near
the surface, which are large, are not included in the estimates of the thermally induced volume
changes used for compensation. In the 29 August test, the thermal contraction estimated in
59
-------
Figure 16 (b) was greater than it should have been if the error had been due entirely to
temperature compensation. In Figure 10 (b), measurements by the level sensor show that the
volume decreased at a rate of -0.35 L/h (-0.091 gal/h) between 18 and 23.5 h on the time line
and, after an abrupt change, increased at a rate of about 0.16 L/h (0.042 gal/h). The compensated
volume rate also shows this distinct shift. The compensated volume rate for the period between
18 and 23.5 h is 0.38 L/h (0.099 gal/h); it increases abruptly to 0.70 L/h (0.18 gal/h) after 23.5 h.
While the first period could reflect volume changes due to deformation that has not yet subsided,
the increase in the compensated volume rate during the second period cannot be explained by the
same mechanism. As with the 27 August overnight test, the increased rate again suggests that
another mechanism is controlling the volume rate during both periods.
As with the overnight test done on 27 August, almost the entire error in the compensated
volume rate during the first period, between 18 and 23.5 h, can be explained by an inaccurate
estimate of the rate of change of temperature in the bottom layer. The data collected by the
thermistors on the horizontal arm suggest that a portion of the error might also be explained by
the increase in temperature as one moves from the centerline of the tank to the wall. By 25 or
26 h on the time line, however, the difference between the bottom two thermistors was
negligible, and it can be assumed that the thermally induced volume changes near the bottom of
the tank were being accurately estimated after 27 h.
The error in the temperature-compensated volume rate during the second period, between
23.5 and 29 h, might be explained by the horizontal gradient in the rate of change of temperature
observed between the centerline and the wall of the tank, where the rate of change of temperature
at Thermistors 26 and 21 during the second period was nearly twice the rate of change at
Thermistor 27. The temperature change measured at Thermistors 26 and 21 was 0.004°C/h,
while at Thermistor 27 it was 0.0022°C/h. Similar changes were observed in the arm on Array
B. As in the 27 August test, this might account for 0.20 L/h (0.05 gal/h) of the 0.70-L/h
(0.18-gal/h) error. Another portion of the error can probably be accounted for by the rate of
increase in temperature near the surface of the product, which is not accurately represented by
Thermistor 17, located 25 cm (10 in.) below the surface. Assuming the same rate of change of
temperature in the upper 11.5 cm (4.5 in.) of the product (7258 L (1915 gal)) that was observed
in the 27 August overnight test, this rate of change of 0.02°C/h would account for 0.15 L/h
(0.039 gal/h) of the error. An analysis shows that the error in estimating the volume in the region
between Thermistors 14 and 16, where the rate of change of temperature changes sense, is
negligible.
60
-------
The thermistors in the vapor space exhibited the same behavior as in the 27 August test. It
is possible that there was condensation, particularly on the walls in the vicinity of the manways,
but its total contribution cannot be estimated.
7.8.3 Overnight Test Starting on 30 August 1990
The overnight test that began on 30 August differed from the other two in that product was
added to the tank and the thermally induced volume changes increased over time. Although this
test was conducted in Tank 2 (whereas the 27 and 29 August tests had been conducted in Tank
1), there is no reason to believe that there were any substantial differences due to the change in
tanks. As with the 27 and 29 August tests, the compensated volume rate showed a distinct
change at about 24 h on the time line. The compensated volume change during the first period
showed a large exponential increase until 24 h on the time line. At 24 h, the compensated rate
and the fluctuation level changed abruptly. The compensated volume rate between 24 and 31 h
was 0.57 L/h (0.15 gal/h). As in the 29 August test, the thermistor closest to the surface, in this
case Thermistor 10, was located 25 cm (10 in.) below the surface.
As can be seen in Figure 24, the data show that fluctuations in the temperature-
compensated volume rate increased at about 24 h on the time line. The reason for this remains
unclear. The most likely explanation, that it was a response to weather changes, fails to provide
a satisfactory answer. Usually, increased wave activity is associated with high winds. In this
case, however, the wind was highest when the fluctuations were lowest, and vice versa. The
thermistor monitoring the vapor space indicated a rapid decrease in temperature in the area above
the product surface until 24 h on the time line, and then a leveling off until 31 h. As would be
expected, the submerged thermistors showed cooler temperatures near the surface than farther
down. (Thermistor 13, in the vapor space, gave a lower reading than the thermistors in the upper
layers of product, but Thermistor 11, the submerged thermistor nearest the product surface (5 cm
(2 in.) below it), gave a reading closer to that of Thermistor 13.) If the temperature measured by
this one thermistor was not localized but was representative of conditions in the vapor space
throughout the tank, it would indicate a high degree of condensation, which could be the cause of
the increased fluctuation level. Although measurements of temperature made in the vapor space
in the manways usually represent a highly localized phenomenon, the condensation explanation
remains a possibility because the temperature in the vapor space was cooler than it was 25 cm
(10 in.) below the surface. A final possibility is that the machinery in the pump house was
turned on at about 24 h on the time line and that vibrations from this equipment induced a
mechanical disturbance in the tank.
61
-------
The exponential increase in the compensated volume rate during the first period, between
18 and 24.5 h, can be attributed to an error in estimating the mean rate of temperature in the
bottom two layers. Thermistor 20 shows that the rate of change was exponential and was very
high. By 27 h, the rate of change of temperature measured by the bottom two thermistors was
approximately the same, suggesting that this source of error had diminished.
The compensated volume suggests that the product in the tank was expanding at a greater
rate than estimated by the vertical array of thermistors. This could also be explained by the
existence of horizontal gradients and the lack of a thermistor in the region immediately below the
surface (during the 30 August test, product level was such that the uppermost submerged
thermistor was 10 in. below the surface). As can be seen in the time series of the thermistors on
the horizontal arm located on Array A in Appendix C, a horizontal gradient in the rate of change
of temperature is present. However, there is no gradient in the rate of change of temperature
measured by the thermistors on the arm attached to Array B. This is because Array B is located
within 0.9 m (3 ft) of the end of the tank, so that all of the thermistors on the horizontal arm are
being affected equally by the heat transfer from the end wall of the tank, as well as from the side
walls. The temperature change measured by Thermistors 26 and 21 on Array A is 0.0073°C/h,
over twice that measured by Thermistor 27 (0.0033°C/h). Assuming that this increase in
temperature affects 75,800 L (20,000 gal) of product, this could account for 0.315 L (0.082 gal)
of the 0.57-L/h (0.15-gal/h) change. As with the 29 August test, the uppermost submerged
thermistor (Thermistor 10) is located 25 cm (10 in.) below the surface. It is difficult to
determine the extent of the possible error in estimating the true rate of temperature change in this
upper layer. However, it is likely that in the upper 11 cm (4.5 in.) of product, temperature would
be increasing at a higher rate than in the area measured by Thermistor 10, a fact that would
reduce the observed error in the compensated volume.
7.9 Residual Volume Changes after Temperature Compensation in Topping Tests
In both the cold and warm topping tests, the tank was filled to over 93% of capacity. As
shown in Table 8, the temperature-compensated volume rates calculated after a 2-h waiting
period were less than 0.05 L/h (0.013 gal/h). The duration of these tests was approximately 2.5
to 3 h, which may be a little shorter than would be desired for estimating the leak rate. The tests
were done 21 and 17 h after product had been added to the tank. Inspection of the temperature
time series shows that the horizontal and vertical gradients had dissipated, because the rate of
change of temperature between adjacent thermistors near the top and bottom of the vertical array
and along the horizontal array had diminished. Thus, the estimate of the thermally induced
volume changes should be accurate. The compensated volume rate during the first 2 h after
62
-------
topping showed a decrease. While there was an increase in temperature fluctuations due to
topping, it does not appear to change the overall linear trend in the thermal volume over the
duration of the test. The error in the compensated volume is probably due to evaporation
resulting from the manway being opened and the interior of the tank being exposed to the
atmosphere. When the tank is opened, both unsarurated air entering the tank and saturated vapor
leaving the tank will produce evaporation until such time as the vapor is saturated. In both
morning tests, the air was at a much lower temperature than the vapor in the tank. Thus, the
vapor, which was warmer than the air, escaped immediately from the tank and was replaced by
the cooler air.
Table 8. Summary of the Topping Tests
Tank
1*
2**
Start Date
8-29-90
8-31-90
Start Time
(h)
10.3 - 13.2
11.5- 13.0
Nominal Level
(cm (in.))
253.0 (99.625)
283.8(111.75)
TCVR
(ml/h (gal/h))
36 (0.009)
-43 (-0.011)
*Test begun 2 h after topping with product colder than product in tank
**Test begun after topping with product warmer than product in tank
7.10 Summary
The experiments yielded some important results that impact the performance of volumetric
leak detection systems; these are summarized below.
• The temperature inhomogeneities produced by adding 19 L (5 gal) of product either
8°C cooler or 25°C warmer than the product in a nearly full tank persisted for 2 to 4 h.
The temperature fluctuations were observed throughout the tank. It is not
recommended that a leak detection test be initiated during this period.
• In a nearly full tank the temperature inhomogeneities produced by adding or removing
15,000 to 45,500 L (4,000 to 11,000 gal) of product were extremely violent and
persisted for approximately 4 to 6 h. During this period, the temperature fluctuations
were large enough to mask the deformation effects. The temperature fluctuations were
observed throughout the tank. It is not recommended that a leak detection test be
initiated during this period.
• After the temperature fluctuations due to product addition or removal had subsided, the
differences in temperature along the long axis of the tank were generally less than
63
-------
0.00rC/h. However, when only a single vertical nrray was used, the difference in
temperature between the centerline of the tank and the walls of the tank remained large
enough to affect the accuracy of compensation for at least 18 h.
• It was difficult to determine the duration of the deformation effects due to the errors in
temperature compensation, but deformation effects did not appear to be large.
Qualitative inspection of the data suggests that the time constant of the deformation
could not have been larger than 1 to 2 h.
• The waiting periods required for the temperature inhomogeneities and deformation due
to product addition or removal to subside in tests on 190,000-L (50,000-gal) tanks are
the same in large tanks as in small tanks.
• The impact of evaporation and condensation on these test results is also difficult to
assess, but their contribution to the error in the compensated volume rate appeared to
be smaller than the error due to thermal expansion and contraction.
• The accuracy of temperature compensation increased as the number of temperature
sensors increased. A minimum of 10 thermistors is recommended if the EPA
requirements are to be met. Because of the large volume of product in the tank, one
poorly calibrated or damaged thermistor or a small error in the coefficient of thermal
expansion will significantly degrade the ability to compensate for the thermal
expansion and contraction of the product.
• Both the fundamental and the first harmonic of surface and internal waves propagating
along the long and short axes of the tank were observed. Peak-to-peak amplitudes of
0.5 to over 2 L (0.13 to over 0.52 gal) were observed.
• Internal waves with periods between 2 and 25 min were observed in the product.
Internal waves were large enough after a manmade disturbance to affect the surface.
• Surface waves with periods of 2 to 10s were observed propagating along the long and
short axes of the tank. The spectra suggest that the first harmonic contained the most
energy. Internal waves produced surface waves with periods of 20 to 190 s.
• Low-frequency fluctuations in the compensated-volume-rate data had periods of 0.5 to
2 h, which suggests that the minimum duration of a leak detection test be 4 h.
64
-------
SECTION 8
TEST DURATION AND INSTRUMENTATION PRECISION
To meet the EPA performance standard for a tank tightness test, the instrumentation used
to measure temperature and level in large tanks should not inhibit the detection of a leak of
380 ml/h (0.1 gal/h) and must allow for a PD of 0.95 and a PFA of 0.05. This requires that the
instrumentation noise be sufficiently small that a one-standard-deviation uncertainty in the
measurement of the rate of change of volume is less than 115 ml/h (0.03 gal/h). This 115-ml/h
estimate assumes that the instrumentation noise is normally distributed with a zero mean and that
the signal adds linearly with the noise. For instrumentation noise, the assumption of normality is
generally justified. This estimate also assumes that the resolution of the sampled data is smaller
than the standard deviation of the noise. Thus, the instrumentation or system noise estimated in
terms of volume can be characterized by its standard deviation; the precision of the instrument is
defined by the standard deviation of the instrument noise. If the resolution is greater than the
inherent precision of the instrument, it is more difficult to characterize the performance of the
instrument. To satisfy this data quality objective, the height and temperature sensors must have
resolution and precision adequate to sense changes of 115 ml/h (0.03 gal/h).
The sensors, thus, must be able to measure changes to within a one-standard-deviation
uncertainty of 115 ml/h (0.03 gal/h). To estimate the resolution and precision of these sensors, it
is necessary to specify the duration of the test. For tanks less than 38,000 L (10,000 gal) in
capacity (for which the EPA performance standard was developed), the duration of a test is
usually 1 to 2 h. This is the minimum amount of time required to make a reasonable estimate of
the rate of change of volume in the tank from level and temperature measurements. Larger
tanks, which are usually only partially filled during testing, may require tests longer than 1 to
2 h, because the sensors required to measure level and temperature changes approach the
technological limits of non-laboratory and affordable equipment. At present, most systems that
are used to conduct a test on a partially filled tank are automatic tank gauging systems (ATGSs).
The duration of tests conducted with ATGSs is typically 4 to 8 h. The precision of the level
sensors is typically between 0.00025 and 0.0025 cm (0.00010 and 0.001 in.). In the Griffiss
experiments, the duration of the measurements was based on the resolution and precision of the
sensors being used [14].
8.1 Measurement of Small Level Changes
Tests conducted in half-filled tanks require the highest degree of precision because the
height-to-volume ratio is lowest when the surface area of the product is greatest, as it is at the
half-way point in the tank. A volume-change uncertainty of 115 ml/h (0.03 gal/h) in a half-filled
65
-------
182,000 L (48,000 gal), 3.5-m (11.5-ft) underground tank will result in an uncertainty of
0.000174 cm/h (0.000069 in./h) in the corresponding product-level changes. The sensor
precision required to measure a level change of 0.000174 cm/h can be estimated from
Sm2 = nS2/(nXt2-(Xt)2) (2)
where Sm = standard deviation of the slope of the least-squares line in centimeters per hour, S =
standard deviation or precision of the sensor in centimeters, n = number of independent points
(i.e., degrees of freedom), and t = time in hours.
Eq. (2) describes the one-standard-deviation error in the slope of a least-squares line fit to a
number of independent points taken over a period of time in terms of the standard deviation of
the ordinate (i.e., sensor precision). Eq. (2) can be used to estimate the minimum duration of the
measurement required to obtain the desired Sm, given a sensor with a precision of S. Estimates
made with Eq. (2) are valid providing that (1) the standard deviation of the sensor is greater than
the resolution, and (2) each sample is independent. The standard deviation represents an
estimate of the system noise or precision of the sensor. Eq. (2) can be used to estimate the
minimum duration of a measurement made with a sensor whose precision is known, or it can be
used to estimate the precision of a sensor given that the duration of the measurement period is
specified. In general, the number of independent samples, and therefore the number of degrees
of freedom, will be significantly less than the number of points acquired by the sensor, because
ambient level and temperature data are highly correlated for periods less than 5 to 15 min.
Table 9 presents the level sensor precision required to obtain a specified Sm with
measurement periods of 1, 2, 4, and 8 h. The way to use Table 9 is to match the precision, or
standard deviation, S, of the sensor, found in the last column of the table, to the corresponding
test duration shown in the first column; all corresponding elements in this table yield an Smof
0.000174 cm/h (0.000069 in./h).
Table 9. Precision, S, of the Sensor Estimated at Sm = 0.000174 cm/h (0.00006786 in./h) for Different Measurement
Periods (A level change of 0.000174 cm/h corresponds to a volume change of 115 ml/n (0.03 gal/h) in a half-filled,
182,000-L (48,000-gal) tank.)
Duration of
Measurement
(h)
1
2
4
8
Number of
Independent
Points (n)
13
25
49
97
Standard Deviation
of Rate of Change
of Sensor (SJ
(cm/h (inJh))
0.000174 (0.00006786)
0.000174 (0.00006786)
0.000174 (0.00006786)
0.000174 (0.00006786)
Standard Deviation
of Sensor (S)
(cm (in.))
0.000196 (0.00007644)
0.000523 (0.00019897)
0.001435 (0.00055965)
0.004000 (0.00156000)
66
-------
For these calculations, it was assumed that the data were sampled once every 5 min. Thus,
it was assumed that there are only 12 degrees of freedom (i.e., 12 independent points) each hour.
If a level sensor with a precision of 0.0025 cm (0.001 in.) is used, the one-standard-deviation
uncertainty in the measured level changes, Sm, is 0.00226 cm/h (0.00089 in./h) for a 1-h test with
12 degrees of freedom; this is equivalent to a volume change of 1,494 ml/h (0.395 gal/h) in a
half-filled 182,000-L (48,000-gal) tank. Previous measurements of level and temperature in an
underground storage tank suggested that the number of degrees of freedom might be as low as 3
or 4 each hour. As the number of degrees of freedom decreases, the duration of the measurement
must increase or more stringent requirements must be placed on the precision of the sensor.
If the resolution of the level sensor is greater than the inherent precision of the
measurement system, and the level changes are less than the resolution of the sensor, then the
smallest level change that can be measured with a two-point estimate is a resolution cell divided
by the measurement period. If the level changes are larger than a resolution cell, however, level
changes can be estimated to better than a resolution cell by fitting a least-squares line to the data.
The accuracy of estimating the rate of change depends on the number of resolution cells
exceeded and the duration of the measurement. A better estimate can be made if we measure the
time at which the level is located at intervals of one-half a resolution cell and fit a least-squares
line to the data. The number of degrees of freedom is equal to the number of resolution cells
minus 1. A more detailed discussion of how to estimate the performance of a system limited by
resolution is given in [11].
8.2 Measurement of Small Temperature Changes
We can estimate, for both half-filled and completely filled tanks, the precision requirement
of the product temperature measurement system assuming an uncertainty of 115 ml/h
(0.03 gal/h) in the leak rate, a value of 0.00125/°C for the coefficient of thermal expansion, and a
182,000-L (48,000-gal), 3.5-m (11.5-ft) tank. A 115-ml/h volume change corresponds to a
standard deviation of 0.0010 and a 0.00055°C/h rate of change of temperature in a half-filled
tank and a full tank, respectively. Eq. (2) was used to estimate the thermistor precision required
to obtain the specified standard deviation of the rate of change of temperature, Sm, in both a
half-filled tank and a completely filled tank as a function of measurement period. This
calculation also assumes that the number of independent degrees of freedom was 12 per hour.
The results are presented in Tables 10 and 11. The minimum measurement period required to
obtain a precision of 0.001°C is less than 1 h for the half-filled tank and approximately 1.5 h for
the completely filled tank. If the test duration were 2 h or longer, the precision of the thermistors
would not have to be as great (the precision could be higher than 0.001°C).
67
-------
Table 10. Precision, S, of the Sensor Estimated at Sm = 0.0005°C/h for Different Measurement Periods in a
Half-filled 182,000-L (48,000-gal) Underground Storage Tank
Duration of
Measurement
(h)
1
2
4
8
Number of
Independent
Points (n)
13
25
49
97
Standard Deviation
of Rate of Change
of Sensor (S J
rc/h)
0.0010
0.0010
0.0010
0.0010
Standard Deviation
of Sensor (S)
rc)
0.0011
0.0030
0.0084
0.0233
Table 11. Precision, S, of the Sensor Estimated at Sm = 0.0005°C/h for Different Measurement Periods in a Full
182,000-L (48,000-gal) Underground Storage Tank
Duration of Number of Standard Deviation Standard Deviation
Measurement Independent of Rate of Change of Sensor (S)
Points (n) of Sensor (SJ
(h) (°C/h) (°C)
1 13 0.0005 0.0006
2 25 0.0005 0.0015
4 49 0.0005 0.0042
8 97 0.0005 0.0116
68
-------
SECTION 9
TEMPERATURE COMPENSATION
The recommended practice for compensating for the thermal expansion and contraction of
product in a tank during a leak detection test is to estimate the average thermally induced volume
change with an array of sensors that measure the change in temperature at many levels in the
tank. The thermally induced volume change, Av, is usually estimated from the following
equation:
Av = Ce Z [(AVi/V) (Av, AT,)] (3)
The product in the tank is divided into i layers, and the thermally induced volume changes, Av,,
produced by the temperature change, ATt, in each layer, are summed. The temperature sensors
are uniformly spaced from the top to the bottom of the tank, and each layer is centered on a
temperature sensor; thus, each layer has the same vertical dimension. Normally, only one value
for the coefficient of thermal expansion, Ce, is used in the calculation. The tank chart is used to
estimate the volume of product in each layer, vi; and in the tank as a whole, V,. The coefficient
of thermal expansion is estimated from a table by means of API gravity measurements made
with product samples taken from the tank. This method of compensation makes the following
assumptions:
• Assumption 1. The number of temperature sensors (i.e., the number of layers) is
sufficient to estimate the average rate of change of temperature throughout the vertical
extent of the tank. Therefore, it is assumed that the rate of change of temperature
measured at the center of each layer accurately reflects the rate of change of
temperature throughout the layer, even at depths where the rate of change of
temperature changes sense, where the gradients are largest (near the bottom or top of
the tank), or where the sensor may not be centered in the layer (near the top of the
tank).
• Assumption 2. The rate of change of temperature measured at each height in the tank
is the same across the entire horizontal axis of the tank, i.e., only one vertical array is
needed for compensation.
• Assumption 3. The coefficient of thermal expansion is the same in each layer, i.e., the
coefficient of thermal expansion does not vary with depth.
• Assumption 4. The method used to estimate the coefficient of thermal expansion is
sufficiently accurate that the required levels of compensation can be achieved.
69
-------
• Assumption 5. The tank chart used to estimate the volume of product in each layer
and the total volume in the tank is sufficiently accurate that the required levels of
compensation can be achieved.
• Assumption 6. The temperature sensors have sufficient precision to measure the
temperature changes occurring in each layer.
Eq. (3) indicates that any errors in the value of Ce or AT, will vary with tank size. Errors in
temperature compensation that were negligible in small tanks may be significant in larger tanks
(those between 57,000 and 190,000 L (15,000 and 50,000 gal) in capacity). The key to accurate
temperature compensation is to divide the tank into enough layers that the uncertainty in the
thermally induced rate of change of volume estimated in each layer is small. This is particularly
important in large tanks. For example, if five temperature sensors (the recommended number in
a 30,000-L (8,000-gal) tank) are spaced at equal intervals in a 3.2-m (10.5-ft) diameter,
190,000-L (50,000-gal) tank, the layer of product surrounding each sensor is 64 cm (25 in.) deep
and contains six times more product than the equivalent layer in the smaller tank. Since there is
only one thermistor in each layer, the precision of this thermistor is very important. The greatest
errors occur in the layers where the gradient in the rate of change of temperature is largest or
changes sense, or where the temperature change is also great. The error in the estimate of the
average rate of change of temperature in the layer is thus a function of the layer's thickness.
Even a 30-cm (12-in.) vertical spacing between thermistors may be too much if the rate of
change of temperature is very high.
The experiments suggest that the rate of change of temperature and the horizontal and
vertical gradients in the rate of change of temperature occurring in the 190,000-L (50,000-gal)
tank are about the same as they are in a 30,000-L (8,000-gal) tank. However, even though the
rates are comparable, the volume of product in each layer is so much greater in a large tank that
errors in temperature compensation can translate to errors in volume measurement that are 5 to
10 times greater than what they would be in a small tank. In a small tank, the effect of these
errors on the test is negligible, whereas in a large tank it is significant.
The tables used to estimate the coefficient of thermal expansion were generated from
measurements (made with a hydrometer) of the specific gravity of a large number of products.
The coefficient is based not on a specific product but on many types of products having similar
properties (e.g., different kinds of gasoline fuels). The tables have an uncertainty of 3.6%;
therefore, the method used to estimate the coefficient of thermal expansion is accurate to 3.6%.
The uncertainty is best illustrated in Figure 26, a plot of the raw data (i.e., the coefficient of
thermal expansion at 15.56°C and the density of the product) used to generate the tables [15].
70
-------
The tables are based on the least-squares line fit to these data. The one standard deviation about
the ordinate is 0.000034/°C, which is 2.7% of the mean coefficient, and the one standard
deviation about the abscissa is 10.2 kg/m3, which is 2.4% of the mean density. The coefficient of
thermal expansion at 15.56°C changes with density, which is estimated from specific gravity
measurements. Any error in the tables stems from (1) incorrect measurements of specific
gravity, (2) fluctuations of the data about each specific gravity, or (3) miscalculation of the
coefficient of thermal expansion at temperatures other than 15.56°C. The fluctuation about each
specific gravity is a function of the many different products used in the analysis. If the specific
gravity is accurately measured, then the error is mainly a function of the scatter about the line.
To reduce this error, a new set of curves would have to be generated for specific fuels, or the
coefficient would have to be determined directly from measurements of specific gravity more
accurate than those made with a hydrometer. Assuming that there is a 3.6% error in the
coefficient, the error associated with a 0.01°C/h change in the temperature of JP-4 fuel as
measured by a thermistor in a 30-cm (12-in.) layer located at the center of a 190,000-L
(50,000-gal) tank would be 0.009 L/h (0.0023 gal/h). This translates into a 0.072-L (0.019-gal/h)
error if the tank is completely filled. In the tank chart used to estimate volume, there can be an
uncertainty of as much as 5%. In practice, the error in the coefficient and/or the volume of the
product used for compensation cannot be reduced without significant effort or cost.
!
0.0015 -
0.0014 -
0.0013 -
0.0012 -
0.0011 -
650
670
690
710
730
750
770
DENSITY - kg/m
Figure 26. Scatter plot of the data used to estimate the coefficient of tbennal expansion at 15.56°C. (Source: [15])
In larger tanks, the increased volume of product in each layer places greater importance on
the instrumentation used to measure temperature. Even one malfunctioning sensor in the vertical
array, or one that is out of calibration, would be sufficient to produce errors of 0.19 L/h
(0.05 gal/h), which is the threshold normally used to declare a leak in a tank tightness test. Even
if the rate of change of temperature is measured to within 0.001°C/h, a malfunctioning sensor
71
-------
would result in an error of 0.023 L/h (0.006 gal/h) in a 30-cm (12-in.) layer of product located
near the center of a 190,000-L (50,000-gal) tank containing JP-4 fuel. If the error is random, the
total error for 10 thermistors would be 0.003°C/h, which would result in" an error of 0.061 L/h
(0.016 gal/h).
Whether the error in estimating the thermally induced volume change involves one layer,
several layers, or the entire tank, compensation becomes a problem only if this error is large.
Errors in temperature compensation can be minimized by increasing the number of thermistors
and by waiting long enough before starting a test that the potential error is less than a given
value, e.g., the threshold used to declare a leak. Since the rate of change of temperature and the
gradients in temperature decrease over time, the error will also decrease over time. Enhanced
software, real-time temperature measurements, and additional sensors are required. More
accurate temperature compensation is achieved if the rate of change of temperature during a test
is low and if there are enough temperature sensors that the volume of product in each layer is
small. The number of temperature sensors now used for temperature compensation in 30,000-L
(8,000-gal) tanks will not suffice in tests on 190,000-L (50,000-gal) tanks.
Inaccurate temperature compensation can be due to any combination of the following
errors.
(1) Error Due to Too High a Rate of Change of Temperature. If the rate of change of
temperature is too high, accurate temperature compensation cannot be achieved. When the rate
of change is acceptably low, compensation can remove 90 to 95% of all the thermally induced
volume changes in any given layer or in the tank as a whole. Under optimal circumstances, it
can remove up to 99%.
(2) Error in the Coefficient of Thermal Expansion. Since any error in Ce is magnified
as the volume of product in the tank increases, the only way to minimize this error is to conduct a
test only when the rate of change of temperature is less than a given value. It is possible to
estimate the potential error thanks to three factors: the uncertainty in Ce is known, the volume of
the product in the tank is known (it is derived from the tank chart, assuming that this chart is
accurate), and the volumetrically weighted average of temperature is measured during a test.
The one-standard-deviation uncertainty in Ce is 3.6%.
(3) Error in the Volume of Product in the Tank. An error in estimating the volume of
product in the tank that stems from an inaccurate tank chart has the same effect as an error in Ce,
because it magnifies any errors in estimating the rate of change of temperature. The error in a
tank chart can be as much as 5% of the total volume of product in the tank.
72
-------
(4) Error Due to Horizontal Temperature Gradients along the Long Axis of the Tank.
Any horizontal differences in the rate of change of temperature will limit the accuracy of
temperature compensation. Because only one array is used for compensation, it is not possible to
estimate this error during a test. Available data suggest that horizontal differences in
temperature along the centerline of a tank are less than 0.001°C/h.
(5) Error Due to Horizontal Temperature Gradients between the Center of the Tank
and the Tank Wall. In the Griffiss experiments, horizontal differences in temperature near the
tank wall were greater by a factor of 2 than those measured near the centerline of the tank. A
waiting period that is long enough for the horizontal gradients to become negligible is the only
way to minimize this error; at least 24 h is the recommended length of time. A shorter waiting
period may be used if real-time measurements are made with one or more horizontal temperature
arrays, so that the point at which horizontal gradients have sufficiently dissipated can be
identified.
(6) Error Due to Significant Vertical Gradients. It is assumed that the rate of change of
temperature varies linearly with depth. If the rate of change of temperature fluctuates rapidly in
the vertical, and particularly if the sense changes, one or more layers can be subject to large
errors. The potential error can be estimated by comparing the measured rate of change of
volume within a layer to the interpolated rate of change of volume in the layers bracketing it.
The error can be corrected by increasing the number of layers (i.e., thermistors) or by adjusting
the position of either the array or the thermistors on the array.
(7) Error Due to Temperature Sensor Characteristics. The accuracy of temperature
compensation is also limited by the precision of the temperature sensors. If the rate of change of
temperature in each layer surrounding a temperature sensor is uncorrelated with the layers
bracketing it, then the total error due to adding the contributions from each layer is equal to the
average precision of the temperature sensors on the array; if, on the other hand, there is a
correlation, the total error is greater than the average precision. Providing that the temperature
changes that occur iri each layer are greater than the resolution of the sensor, the total error can
be reduced by increasing the duration of the test. Frequent and accurate calibration of the
sensors is essential. The required precision of the sensors can be estimated from the procedure
described in Section 8.
(8) Error Due to Sensor Malfunction. If one sensor malfunctions during a test in a small
tank, the rate of change of temperature in that layer is sometimes estimated by linear
73
-------
interpolation from the sensors in the bracketing layers. This is done when the number of sensors
exceeds the minimum number needed to achieve a given performance. Tests in a 190,000-L
(50,000-gal) tank should be redone if a sensor malfunctions.
(9) Error Due to Level Sensor Characteristics. The accuracy of the temperature-
compensated volume also depends on the precision of the level sensor. The uncertainty in
measuring the level of product over the duration of a test can be estimated with the approach
described in Section 8. Since the height-to-volume conversion factor will change as a function
of the amount of product in the tank, the effective uncertainty of the level sensor will change
accordingly. The uncertainty of the level sensor is estimated from the height-to-volume
conversion factor at the product level at which a test is to be conducted.
(10) Error in the Height-to-Volume Conversion Factor. An experimental estimate of
the height-to-volume conversion factor is required for each leak detection test. The uncertainty
in this estimate can be determined from the standard deviation of the level changes used to
estimate the height-to-volume conversion factor. The total error is obtained by multiplying the
standard deviation of the height-to-volume measurements by the rate of change of level
measured during a test.
The magnitude of these errors can range from several thousandths to tenths of a gallon per
hour. A 0.00 rC/h error would result in a 0.197 L/h (0.052 gal/h) error in a 190,000-L
(50,000-gal) tank containing JP-4 fuel.
The size of the error stemming from thermal compensation can be reduced if tests are
conducted only when the rate of change of temperature is low. Since the rate of change of
temperature generally decreases with time after a delivery, it is possible to wait until it has
dropped below some given value before beginning a test. Many volumetric leak detection
systems use this approach in testing small tanks. The other approach is to reduce the total
volume of product in each layer by increasing the number of temperature sensors.
Temperature compensation is not the only error that can influence a leak detection test.
Volume changes due to such factors as residual deformation and evaporation/condensation may
also occur. However, the data collected during the Griffiss experiments suggest that the 24-h
waiting period that allows the horizontal gradients to dissipate is also sufficient to allow the
effects of deformation effects subside.
74
-------
SECTION 10
IMPORTANT FEATURES OF A VOLUMETRIC LEAK DETECTION
SYSTEM FOR TESTING LARGE TANKS
The important features of a volumetric leak detection system that is capable of meeting the
EPA performance standards for testing large tanks (up to 190,000-L (50,000-gal) in capacity) are
described below. They are nearly identical to the recommended features of systems designed to
test smaller tanks (up to 30,000-L (8,000-gal)), except that (1) the minimum number of
thermistors required for temperature compensation has been increased from 5 for a 30,000-L
(8,000-gal) tank to 10 or more for a 190,000-L (50,000-gal) tank, (2) the minimum duration of a
test has been increased from 1 to 2 h to 4 h or more, (3) the minimum waiting period required for
the horizontal and vertical gradients to dissipate has been increased from 6 to at least 24 h,
(4) the average rate of change of temperature in any one layer or in the tank as a whole is small
enough to allow accurate temperature compensation, and (5) accurate experimental estimates of
the constants necessary for converting level and temperature changes to volume are made. The
24-h waiting period also allows adequate time for structural deformation of the tank to subside (it
may take 6 h in a 30,000-L (8,000-gal) tank) and for the violent temperature fluctuations that
occur immediately after any addition or removal of product from the tank to flatten out (these
last approximately 4 to 6 h). The data suggest that, if a test is initiated within 10 h of a product
addition or removal, more than 10 thermistors are required for adequate temperature
compensation, and the thermistors must be more densely spaced near the surface of the product
and the bottom of the tank; however, if there is a waiting period of at least 24 h, 10 thermistors
are sufficient. While the instrumentation requirements are no different for larger tanks than for
smaller tanks, they are more difficult to achieve. A longer test duration can be used to offset a
lesser sensor precision providing that the level and temperature changes exceed at least one
resolution cell.
Horizontal gradients in the rate of change of temperature between the tank's centerline and
its walls appear to be the controlling source of error in temperature compensation. A waiting
period of at least 24 h is recommended so that this gradient has time to subside. Unless the
temperature is sampled with a horizontal array of thermistors similar to the one used in these
experiments, however, it will not be possible to assess whether even a 24-h wait is long enough.
This is true even if the horizontal gradients are small. The only alternative to direct
measurement of the horizontal gradient is to conduct additional tests. These are recommended as
a means of determining whether the volume rate is decreasing over time in cases when the
threshold is exceeded during the first test. A measured volume rate that is decreasing indicates
75
-------
that the waiting period is inadequate. Testing should not be concluded until the measured
volume rate has approached a constant value. A properly designed multiple-test strategy will, in
general, reduce the number of false alarms and missed detections for all sources of noise.
The important features of a volumetric leak detection system suitable for testing 190,000-L
(50,000-gal) tanks are:
(1) The level and temperature instrumentation has a precision that permits measurements
of volume changes 3.3 times smaller than the leak to be detected (i.e., the
instrumentation must be capable of measuring leaks as small as 0.11 L/h (0.03 gal/h)).
• The precision of level sensor depends on how large the surface area of the product
is. (If the tank is overfilled, the surface area is very small; if the tank is underfilled,
the surface area is very large.)
• The precision of temperature sensor depends on the volume and type of product in
the tank during a test.
(2) The test is conducted at a nearly constant pressure.
(3) The height-to-volume conversion factor, the coefficient of thermal expansion, and any
other numerical constants are measured experimentally.
(4) Temperature sensors are spaced at intervals of 15 to 30 cm (6 to 12 in.), with denser
spacing if vertical gradients are large; or, an equivalent method of temperature
compensation is used.
(5) There is a waiting period after delivery which is sufficient to allow
• temperature inhomogeneities to become negligible (4 to 6 h)
• deformation to subside (0 to 24 h)
• horizontal gradients in temperature to dissipate (24 h or longer)
(6) There is a waiting period after topping which is sufficient to allow
• vertical temperature inhomogeneities to become negligible (generally 2 to 4 h)
• deformation to subside (0 to 24 h)
(7) The test duration is 4 h or longer.
(8) The sample rate is sufficient to achieve instrument precision and avoid aliasing (data
are sampled at 1 Hz and averaged to 1 min).
76
-------
(9) An upper limit is placed on the rate of change of temperature that can occur during a
test.
(10) A multiple-test strategy is used to minimize the number of false alarms and missed
detections.
77
-------
SECTION 11
REFERENCES
1. U. S. Environmental Protection Agency. "Underground Storage Tanks; Technical
Requirements and State Program Approval; Final Rules." Federal Register, 40 CFR Parts
280 and 281, Vol. 53, No. 185 (23 September 1988).
2. J. W. Maresca, 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).
3. U. S. Environmental Protection Agency. "Evaluation of Volumetric Leak Detection
Methods for Underground Fuel Storage Tanks." EPA Contract No. 68-03-3409, Risk
Reduction Engineering Laboratory, U.S. Environmental Protection Agency, Cincinnati,
Ohio (December 1988).
4. J. W. Maresca, Jr., J. W. Starr, R. D. Roach, D. Naar, R. Smedfjeld, J. S. Farlow, and
R. W. HiHger. "Evaluation of Volumetric Leak Detection Methods Used in Underground
Storage Tanks." /. Hazardous Materials, Vol. 26 (1991).
5. J. W. Maresca, 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, EPA/600/9-87/015, Hazardous Waste Engineering Research
Laboratory, Office of Research and Development, U.S. Environmental Protection Agency,
Cincinnati, Ohio (July 1987).
6. R. D. Roach, 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 Research Symposium, EPA/600/9-88/021, Risk
Reduction Engineering Laboratory, Office of Research and Development, U.S.
Environmental Protection Agency, Cincinnati, Ohio (July 1988).
7. J. W. Maresca, Jr., J. W. Starr, R. D. Roach, J. S. Farlow and R. W. Hillger. "Summary of
the Results of EPA's Evaluation of Volumetric Leak Detection Methods." Proceedings of
the Fifteenth Annual Research Symposium, EPA/600/9-90/006, Risk Reduction
Engineering Laboratory, Office of Research and Development, U.S. Environmental
Protection Agency, Cincinnati, Ohio (February 1990).
8. J. W. Maresca, 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." 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 (February 1989).
9. U. S. Environmental Protection Agency. "Standard Test Procedures for Evaluating Leak
Detection Methods: Volumetric Tank Tightness Testing Methods." EPA Contract
No. 68-01-7383, Office of Underground Storage Tanks, Washington D. C. (March 1990).
10. U. S. Environmental Protection Agency. "Standard Test Procedures for Evaluating Leak
Detection Methods: Automatic Tank Gauging Systems." EPA Contract No. 68-01-7383,
Office of Underground Storage Tanks, Washington D. C. (March 1990).
11. J. W. Starr and J. W. Maresca, Jr. "Quality Assurance Project Plan: Evaluation of Leak
Detection Methods for Large Underground Storage Tanks." Vista Research Project 1026,
Vista Research, Inc., Mountain View, California (January 1990).
78
-------
12. J. W. Maresca, 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. J. W. Maresca, 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).
14. J. W. Maresca, Jr., J. W. Starr, R. F. Wise, R. W. Hillger, and A. N. Tafuri. "Evaluation of
Internal Leak Detection Technology for Large Underground Storage Tanks." Proceedings
of the Sixteenth Annual Research Symposium, EPA/600/9-91/002, Risk Reduction
Engineering Laboratory, U.S. Environmental Protection Agency, Cincinnati, Ohio (August
1990).
15. U. S. National Bureau of Standards. "Volume Correction Factors." In Manual of
Petroleum Measurement Standards. American Petroleum Institute, Washington, D.C.
(August 1980).
79
------- |