United States
Environmental Protection
Agency
Technology Transfer
EPA/625/9-89/009 Apr. 1989
Volumetric Tank Testing:
An Overview

-------
                                   EPA/625/9-89/009
                                        April 1989
   Volumetric Tank Testing:
           An Overview
Center for Environmental Research Information
    Office of Research and Development
    U.S. Environmental Protection Agency
           Cincinnati, OH 45268

-------
Contents
 1.   Introduction   1

 2.   The EPA Program   3

 3.   Types of Volumetric Tank Tests   5

 4.   Volumetric Tank Testing   7

 5.   Conducting a Volumetric
      Tank Test   13

 6.   Performance   15

 7.   Results of the Edison
      Evaluation   17

 8.   Applying the Results   25

 9,   Checklist for Good
      Performance    27

10.  References   33

11.  Glossary   35

-------
 1.   Introduction
Background

In the United States there are several
million underground storage tanks
containing petroleum  fuels and chemicals, it
is estimated that 10 to 25% of them may
be leaking. This translates to up to one half
million leaking tanks in the U.S. The
contamination of groundwater that results
from such leaks is a serious environmental
threat and one that impacts public health
directly, for in most states at toast 50% of
the potable water supply comes from
underground sources. In 1984 (through the
Hazardous and Solid  Waste Amendments
to the Resource Conservation and
Recovery Act of 1976) the U.S.
Environmental Protection Agency (EPA)
was charged with developing regulations for
the detection of releases from underground
storage tanks. The new regulations [1],
issued in September  1988, state that all
volumetric tank test methods must, within
two years, have the capability of detecting
leaks as small as 0.1  gallons per hour with
a probability of detection1 of 95% and a
probability of false  alarm of 5%.

There are many commercially available
methods for detecting leaks in underground
storage tanks. Those which are the most
widely used in the petroleum industry are in
a category called volumetric tank tests
(also known as "precision," "tank
tightness," or "tank integrity" tests). The
premise of a volumetric tank tost, and
hence its name, is that any change in the
volume of fluid within a tank can be
interpreted as a leak. Detection of these
leaks is difficult because there are  many
physical  mechanisms which produce
volume changes that can be mistaken for
 1 Bold type denotes a term that is defined in the
  glossary.
leaks. Most of the volumetric tank tests on
the market today claim the ability to detect
leaks as small as 0.05 gallons per hour.
(This is the "practice" recommended by
the National Fire Protection Association [2]
for volumetric tests in tanks less than
12,499 gallons in capacity, a category that
includes most of the tanks at retail gasoline
stations, i.e., those addressed in the EPA
regulations.) These volumetric tank tests,
however, do not specify their reliability in
terms of probability of detection and
probability of false alarm against this 0.05-
gallon-per-hour teak rate. A performance
of 0.1 gallons per hour with a probability of
detection of 95% and a probability of false
alarm of 5% represents a realistic goal in
terms of the current technology. A
performance of 0.05 gallons per hour with
the same 95% probability of detection and
5% probability of false alarm would be, at
best, difficult to achieve.
Approach

This report summarizes the technical
findings of an EPA study on volumetric tank
testing.
   • It describes the results of the EPA
     study, which evaluated the viability of
     volumetric tank tests as a means of
     detecting leaks in underground storage
     tanks.

   • It explains the accuracy requirements
     specified in the EPA regulations
     concerning the testing of underground
     storage tanks for leaks.

   • It presents information that will aid the
     user in selecting a volumetric tank test
     method that meets these regulations.

-------
The EPA study found that volumetric tank
tests are a viable means of monitoring
underground storage tanks for leaks. It
determined that the concept of volumetric
tank testing is intrinsically sound, as are
most of the physical devices used in
testing* In general, performance
shortcomings were found to be procedural
in nature. That is, it was not the devices
themselves, but the way in which they were
used, that detracted from performance. The
study concluded that volumetric  testing can
meet regulatory standards and is an
important tool in minimizing the effect of
leaks from underground storage  tanks.
operators. The reader who wants more
technical detail is referred to the EPA
document Evaluation of Volumetric Leak
Detection Methods for Underground Fuel
Storage Tanks [3], a comprehensive report
on the results of the EPA evaluation. It
contains the objectives and the chronology
of the experiments, a thorough explanation
of the engineering principles underlying the
experiments, and a comprehensive
analysis of the results. In addition, the
document includes the evaluation reports
written for each of the 25 test methods. It is
available from the National Technical
Information Service.
This document is intended for state and
local regulators and small tank owners and

-------
  2.   The  EPA  Program
 In 1986 the EPA initiated a program whose
 purpose was to evaluate the performance
 claims made by manufacturers of
 volumetric tank tests. Participation in the
 program was voluntary. Of the 43
 commercially available systems,  the
 manufacturers of 25 elected to participate.

 The objectives of the program were to:

    • provide data to support the
     development of new EPA regulations

    • define the performance oif the current
     technology

    • make recommendations  to improve
     current practice

    • provide information that would help
     users select suitable leak detection
     systems

 The evaluations were conducted  at the
 EPA's Risk Reduction Engineering
 Laboratory (RREL) in Edison,  New Jersey.
 An experimental setup consisting of one
 steel and one fiberglass tank was
 constructed especially for the evaluation.
 This Underground Storage Tank  (UST) Test
 Apparatus, as it is called, permits the
 conduct of full-scale tank tests under
 controlled conditions. Leaks of different
- sizes can be simulated in the  tanks, and
 control can be exercised over the
 temperature of the fluid in the tank and
 other factors that affect the  accuracy of leak
 detection systems.

 Major Findings of the EPA
 Study

 Prior to the EPA study, manufacturers
 commonly claimed  that volumetric test
 methods could detect leaks as small as
0.05 gallons per hour. This claim was not
supported by the Edison experiments. The
experiments did show, however, fihat
volumetric testing is a sound concept and
that available test methods2 work well when
applied properly. The major findings of the
EPA study are:

   • Volumetric test methods are capable
    of meeting regulatory requirements
    (Five of the methods evaluated in the
    EPA study, once they modify their
    criteria for determining whether the
    tank is leaking or not, have the
    capability to meet the EPA
    regulations).

   • To achieve a performance level that
    meets the regulatory requirements,
    most volumetric test methods need
    modification.

   • In most cases, the area  in need of
    modification is the testing procedure,
    not the instrumentation.  Generally, the
    instrumentation is intrinsically sound.

Furthermore, the study identified the
minimum required modifications for each
test method evaluated. With thesis
modifications, most of the test methods
should be able to meet the regulatory
requirements.

Every leak detection method has two
components: equipment and procedure.
The term equipment includes the physical
devices used to measure leaks, as well as
any computer hardware or other
instruments used to make measurements.
The term procedure refers to the way the
2 In this report, the more comprehensive term test
 method is often used instead of the word system.
 "Test method" connotes not only the physical system
 but also the procedures used in operating it.

-------
test is conducted; it includes not only the
role of the test operator but also the data
analysis scheme, detection criterion,
temperature-compensation scheme, and
procedures for calibrating the instruments.
When a method did not function as well as
it could have, the cause was almost
invariably some aspect of procedure. Two
classes of procedural failures were noted:
(1) the instructions were incorrect or were
inappropriate; (2) the instructions were not
followed. It is significant that the limitations
of these volumetric test methods turned out
to be in the realm of procedure rather than
equipment. Procedural changes can
generally be made with less effort than
equipment changes.
The equipment itself has no intrinsic
ranking. The ranking depends on how the
test method is used. Because procedural
changes are so readily implemented, any
ranking implicit in the EPA study is already
outdated the study's value^ lies not in
ranking various methods but in identifying
features common to methods with high
performance.

-------
Volumetric tank tests can be divided into
two categories. In the first, the tank is filled
to capacity, and in the second, the tank is
partially filled. In filling a tank to capacity
the operator does not stop until  the level of
the fluid reaches the fill tube (or a
standpipe located above grade); hence the
term "overfilled" is applied to these tests.
Overfilled-tank tests can be further
categorized according to those conducted
under nearly constant hydrostcitic  pressure
and those conducted under variable
hydrostatic pressure. Hydrostatic pressure
varies with any fluctuations in cjroundwater
level, product level, or atmospheric
pressure that occur during a test.

When a test is conducted in  a partially filled
tank, only that portion of the  tank that
contains fluid is tested for  leaks; the test
cannot assess the integrity of that portion of
the tank located above the product level. A
test conducted in an overfilled tank, on the
other hand, assesses the integrity of the
entire tank.

Constant-Level Tests

In a constant-level test, product is added or
removed in order to maintain a constant
fluid level in the tank's fill tube or
standpipe. To conduct a successful test, it
is necessary, once a tank has been filled
and then again after it has been topped off
prior to testing, to observe a waiting period
long enough to ensure that the tank has
expanded to its maximum capacity. Then, if
the fluid level is kept constant, the tank will
neither expand nor contract during the test,
and measured volume changes will
accurately represent actual volume
changes.

Partially-filled-tank tests are  generally
considered constant-level tests;. Because
the surface area of the product is spread
across the width and length of the tank, any
level changes will be quite small,
regardless of the size of the associated
volume change (see Figure 1a). Unless
product is added or removed  during the
test, causing the tank  to deform, a partially
filled tank behaves in the same manner as
an overfilled tank at constant levesl, given
that the appropriate waiting periods are
observed.                     <
Variable-Level Tests

In a variable-level test, the fluid level is
allowed to fluctuate. When such a test is
conducted in an overfilled tank, the surface
area of the product is extremely small — it
is usually limited to the diameter of the fill
tube and other small openings. Any volume
changes will be seen as significant height
changes (see Figure 1b). Unless the
deformation characteristics of the tank
being tested are known (as well as those of
the backfill and surrounding soil) it is not
possible to distinguish between the volume
changes due to a leak and those that
normally occur in a nonleaking tank. These
deformation  characteristics are not known
at the time of the test, and it is impractical
to measure them. There is, consequently, a
high risk of error associated with variable-
level tests.

Since the Edison experiments, many of the
variable-level tests have been converted to
constant-pressure tests. This has been
accomplished by adding either of two
features to the test: (1) releveling the
product to maintain a constant level during
the test or (2) increasing the surface area of
the container in which product level is
measured.

-------
            Partially Filled Tank
               Overfilled Tank
Adding a quart of liquid to this tank would produce
a barely noticeable rise in the level of fluid. Level
changes are distributed over a large surface area,
so that even large volume changes produce only
very small level changes.
Adding a quart of liquid to this tank would cause
the fluid to rise many inches. Here the surface
area is very small. Thus, even a small volume
change can mean a drastic level change, and
consequently, a large change in pressure on the
tank walls.
Figure  1.    The same volume change can produce very different level changes depending on
             whether the tank is partially filled or overfilled.

-------
 4.   Volumetric Tank Testing
Detecting a leak by means of volumetric
testing is akin to the statistical problem of
finding a signal in a background of noise.
The "signal" is the volume change that is
due to a leak, and the "noise" is the sum of
all the volume changes due to factors other
than a leak. A volumetric tank test
measures the change in the volume of fluid
("product") in the tank and attributes this
change, once all other sources of noise
have been accounted for, to a leak. Most
methods measure changes in the level of
the product and convert these to volume
changes using a height-to-volume
conversion factor (h-to-v); this is the
technique used in variable-level tests.
Others measure volume directly; this is the
technique used in constant-level tests. A
leak is defined in terms of flow rate in
gallons per hour and can be positive or
negative; that is, product can flow out of the
tank or groundwater can flow into the tank.
Once the flow rate has been established, a
decision must be made as to whether to
declare the tank in question leaking or
nonleaking. This decision is usually made
by comparing the flow rate to a
predetermined value (called the
"threshold"); if the flow rate exceeds the
threshold, a leak is declared.

Unfortunately, a leak is not the only
physical mechanism responsible for
changes in the level and volume of product.
There are a number of physical
mechanisms that can contribute to either
real or apparent volume changes, whether
the tank is leaking or not. The best-
performing methods can differentiate
between these non-leak-related volume
changes and an actual leak.

It is a common perception that if the
equipment is working properly the  test will
yield the actual leak rate. In reality, there is
always some variation in test results, and it
is likely that even with the same lest
method a different flow rate will be obtained
each time a test is conducted. Even a test
on a nonleaking tank will generally yield a
value different from zero. Variation in test
results stems from three sources:  (1) the
equipment itself, (2) operational practice
(how the test is conducted), and (3)
environmental considerations such as
thermal expansion and contraction of the
product, evaporation and condensation, and
the way these and other environmental
factors interact. For best results, the
instrumentation noise should be less than
the environmental noise. This ensures that
the instrumentation will not limit the
detection of small leaks. Environmental
noise is the more acute problem because it
can be difficult to predict, it can be difficult
to compensate for, and it can be larger than
the leak itself.
Normally Occurring Volume
Changes Can Mask a Leak

There are at least five sources of
environmentally induced product-level or
product-volume changes that are unrelated
to a leak. These non-leak-related product-
level or product-volume changes (called
"ambient noise") can be as large as or
larger than the smallest leaks to be
detected. There are many effective ways to
compensate for ambient noise. The five
sources of ambient noise are by no means
equal in their impact on the accuracy of a
test. The first three are likely to have the
most deleterious effects, because* the error
associated with them is large. The
remainder are responsible for errors that
are small by comparison.

-------
f. Thermal Expansion and
Contraction of the Product
Temperature fluctuations cause expansion
and contraction of the "product  Expansion
and contraction represent changes in
volume that can easily be mistaken for a
leak unless they are taken into
consideration when overall volume changes
are calculated. When product is added to
the tank (for  example, during a delivery or
during topping to fill the tank to its
maximum capacity), the temperature
increases or decreases as the product
seeks thermal equilibrium with the
surrounding  backfill, native soil, and
groundwater. Similarly, newly added
product seeks equilibrium with the product
that is already in the tank, and vice versa.
Thus, temperature fluctuations are at their
peak in the period immediately following a
delivery of product or following topping
(see Rgure 2.) The importance of
compensating for these temperature
changes cannot be overemphasized. The
volume changes produced by expansion
and contraction of the product are real, and
may be as large as 1 gallon per hour,  but
they are not in any way associated with a
leak.

2. Expansion and Contraction
of Vapor Pockets

In addition to their direct influence on the
product, temperature fluctuations can also
cause the expansion or contraction of vapor
pockets that are almost always present
after a tank has been filled to capacity.
Here, though, temperature is not the sole
influence

Vapor pocket size can also be affected by
atmospheric pressure and by pressure
changes resulting from product-level
changes. When the volume of the trapped
vapor changes, there is a resultant change
in the level of the product; this latter
change  may mistakenly be interpreted as a
leak. Despite efforts to "bleed" the tank of
trapped vapor, pockets large enough to
adversely affect the outcome of a test may
still be present. Vapor pockets in  quantities
as small as 10 gallons can influence a test
result in a 10,000-gallon tank. It is virtually
impossible to determine the exact size of
vapor pockets. However, when the vapor
pocket is large, i.e.,  greater than 10 to 20
 A forge amount of product has just been added
 to an underground tank that was already partly
 filled. In this example, the new product is cooler
 than the resident product and sinks.  The
 temperature of the product throughout the tank
 fluctuates greatly.
As the old product cools and the hew warms,
equilibrium is reached. But the temperature as
a whole is cooler, causing the product to contract
and the level to go down. (The inverse is true
when a large amount of warmer product is added,
causing the product to expand and the level to
rise.)
 Figura 2.    How temperature changes can be mistaken for a leafs.

-------
gallons, it Is possible to determine that
trapped vapor is present; The procedure for
doing this is explained in [3]. If vapor
pockets of 10 to 20 gallons or more are .
shown to be present, or if for any reason
(for example, if the tank is tilted) it:ls
suspected that they  are present, the tank
and lines should again be bled; if vapor
pockets are shown to be largely absent,
testing may proceed.

3. Structural  Deformation of
the Tank

Whether it is constructed of steel or
fiberglass, and whether it is embedded in a ,
dense backfill or in a loose one that has
more "give," the tank itself expands and
contracts in response to both level and
temperature changes. This phenomenon is
known as structural deformation. When
the tank expands, the level of the'fluid
inside it goes down; conversely, when it
contracts, the level goes up. The height
changes produced by the change in the
volume of the tank may also be mistaken
for a leak (see Figure 3). There; are two
types of structural deformation; (1) the
instantaneous deformation that appears
immediately after any change in product
level and (2) the time-dependent relaxation
of the tank. Instantaneous deformation is
accounted for when the height-to-volume
conversion factor is measured rather than
calculated. The instantaneous compression
and expansion of trapped vapor is also
accounted for in this way. What is not
accounted for is the time-dependent
relaxation of the tank. In order to do this,
the amount of "give" of the tank," backfill
and surrounding soil must be known. The
length of time it takes for the tank to
expand or "relax" to its maximum capacity
must also be known. Generally, thesd
values are not known during an actual test.
However, the effects of structural
deformation can be minimized by
introducing a waiting period. The waiting
period varies from one tank system  to
another. Efficient testing requires an
analysis algorithm to determine when the
effects of deformation have subsided.  In
the Edison tanks, time-dependent
deformation took 15 hours or more to
subside.

Even when expansion or contraction due to
sudden, large, man-made changers in
product level has reached its maximum,the
tank continues to deform, expanding or ;
contracting in response^ to every  product-   :
level change that occurs during a testxThis
occurs regardless of whether the change
was produced by a leak or by one of the
         Cap
 A         ..• ,                      ..        B                 .           '..---
 An empty underground tank has just been filled    In response to tfje pressure of the product, the
 with product.                                walls of the tank (the sides as well as the ends)
                                            begin to deflect or "deform," and the level of
                                            the product goes down.          \
 Figure 3,    How structural deformation of the tank can be mistaken for a leak.

-------
five environmental noise sources. It can be
a problem during a variable-level test if the
tank/backfill/soil system is highly elastic
and the tank is overfilled. In a case like this.
interpretation of the results is difficult,
because the measured volume changes are
always smaller than the actual volume
changes. The reason for this is that any
increase in product level at the start of a
test causes the tank to expand in response
to the increased pressure; product level
then drops as a result of the expansion of
the tank; when the product level drops, the
pressure is reduced and the tank contracts;
when the tank contracts, the product level
rises again at least part of the way back to
where it was originally. The product level
drops and rises repeatedly. This complex
interaction of forces is dependent on the
tank/backfill/soil characteristics, which in
actual practice are not known. It is best,
therefore, to avoid variable-level tests.

4. Evaporation  and
Condensation within the Tank

Unless a tank and its fill tube are
completely filled and no air or vapor
pockets are present, it is likely that as
temperatures change fluid will evaporate
from the free product surface or condense
on tank walls. This activity produces
volume fluctuations that may be mistaken
for a leak.

5. Waves

Mechanical vibrations and other
disturbances produce waves; these can be
of two types, surface or internal. (In some
instances, internal waves can produce
surface waves.) Surface waves move along
the exposed area of the product in a
partially filled tank, causing a back-and-
forth motion comparable to a "bathtub"
wave. They may be misinterpreted as
changes in fluid level. Internal waves, which
are found in both filled and partially filled
tanks, usually occur when there are
temperature differences present, such as at
the boundary layer between resident and
newly added  product. The passage of an
internal wave causes this boundary layer to
undulate vertically so that a temperature
sensor at a fixed location records the
temperature changes associated with the
wave rather than those responsible for
volume changes. If the data are
undersampled, the waves may appear to
produce level or temperature changes,
       Sampling Interval
       Sampling Interval
       Wave Period
                                                   Wave Period
An example of  undersampled data. Data  are
collected every 25 minutes, but the length of the
wave period is 30 minutes. The data show only
the alternate or "alias" fluctuations. The mean
rate of change is zero, but the measured rate
is not zero.
 An example of an appropriate sampling interval
 for the length of  the  wave period. Data are
 collected every 1S minutes, so that both the highs
 and lows  are recorded The  average rate  of
 change is zero.
Fjgura 4.    Undersampling the data can produce a problem called aliasing. This problem causes
            errors in estimating the moan rate of change of product level or temperature.
                                         10

-------
even though none exist. This problem is
called aliasing (see Figure 4). However, if
the data are sampled frequently enough
and are averaged, the problem is avoided.
Groundwater Height Can
Affect the Size of a Leak

There is another factor that can interfere
with accurate measurements. Unlike the
five listed above, it does not mimic a leak.
It does, however, have a direct effect on
the size of a leak. This factor is;
groundwater. The water table of the soil
in which a tank is buried can vary in height
depending on factors such as geographic
location, season, and amount of
precipitation. If a tank is leaking, the height
of the water table in relation to the tank has
a direct effect on the flow rate measured
during a test. If the water table is above the
location of a hole or fissure in an
underground tank, the groundwater exerts a
pressure on that hole which counteracts the
pressure exerted on the same hole by the
fluid in the tank. There are three possible
scenarios. Groundwater can restrict the
flow of product out of the tank; it can
prevent flow entirely; or it can cause an
inflow of water into the tank. As shown in
Figure 5, any of these scenarios can alter
the rate of a leak. Whenever the
groundwater is above a hole in the tank, it
may cause even a large leak to go
undetected. Since it is virtually impossible
to determine the location of a hole in an
underground tank, efforts must be
concentrated instead on monitoring the
groundwater level. This is accomplished by
means of a "monitoring well" installed next
to the tank and used to make
measurements of the water table.  It is
important to be aware that when the water
table Is higher than the bottom of  the tank,
any test for leaks will be less sensitive. The
best test results are obtained when the
water table is below the level of the tank.
Flow through the hole is then unrestricted
by groundwater.
                                         11

-------
               Cap


       Product Level
           V
    Water Table
EguatPressure Level
                      Water Table
 No Flow, The pressure exerted by the product
 at the hole is exactly balanced by the pressure
 of the groundwater  at the ho/e.  Because the
 product is less dense than water, there is no flow
 in e/ther direction even though the product level
 is higher than the water table.  The dotted line
 shows the product height reguired to produce an
 equal balance  of  pressure  between the
 groundwater and the product.
       Product Level
                                               Product Level
                   B
                   Flow into tank.  The pressure exerted by  the
                   groundwater is greater than that of the product;
                   therefore, water flows into the tank. Product level
                   would have to be at the "equal pressure" line
                   in order  to achieve an  exact balance with the
                   groundwater.
                                                            Cap
                                                  Product Level
      Water Table
                                                                           Equal Pressure
                                                                               Level
                                                Water Table
Flow out of tank. Since the water table is below
 the tank, there is no counter-pressure against
 the product at the hole; therefore, product flows
 out.
                   Flow out of tank. Here the pressure exerted by
                   the groundwater is less than that of the produce-
                   therefore, product flows out, but at a rate slower
                   than shown in (C).
 figure 5.    Groundwater can affect the rate of flow through a hole in an underground tank.
                                             12

-------
 5.   Conducting  a Volumetric  Tank  Test
Procedures for conducting a volumetric
tank test do not differ greatly from one test
method to another. (The major differences
between test methods are in the
instrumentation used to gather data.) The
general procedure for conducting an
overfilled-tank test is shown in Figure 6; the
procedure for a partially-filled-tank test is
similar to it. The three procedural steps
common to all volumetric test methods are
preparation, testing and analysis.

Preparation

In order to prepare for a test, the tank must
first be filled to the level required for
testing. A waiting period must be observed
to ensure that two major consequences of
filling the tank have subsided: i;he horizontal
differences in the temperature of the
product within the tank and the structural
deformation of  the tank itself. The
instrumentation can then be installed. (This
can be done during the waiting period.) The
instrumentation includes the  physical
devices used to measure the temperature
or volume of the product, for example,
thermistors and height or volume sensors.
Next (in the case of overfilled-tank tests)
the tank is topped. This means making fine
adjustments to the fluid level, ciither by
adding or removing small amounts of
product, in order to bring this fluid level into
conformity with the test method's
requirements for initiating testing. Since any
change in  product level precipitates
temperature differences and  structural
deformation, these fine adjustments
necessitate a second waiting period. The
test operator must then determine values
for the height-to-volume conversion factor
and the coefficient of thermal
expansion. Finally, the operator should
measure the height of the water table. This
is necessary in order to interpret the test
results properly. The preparations are now
complete, and testing can begin.;

Testing

During the test, specialized sensors take
measurements of both the temperature and
the level (or volume) of the fluid in the tank.
In many test methods, temperature and
level data are sampled at the same rate. In
most cases the data are collected manually;
the operator reads an instrument or makes
a measurement and then records it on a
data sheet. Some manufacturers, however,
have automated the data collection process.
The decision to end a test is based on a
criterion predetermined by the
manufacturer. Usually this criterion is
expressed in terms of time; for example,
the test ends 60 minutes after the start of
the data collection.  Sometimes the length of
test is not fixed, the end of the test
depends on some criterion in the data.

Analysis

Making sense of the data is the purpose of
the analysis. Procedures for converting the
level data to  volume data, compensating for
temperature, and  computing a votume rate
are usually well defined by the
manufacturer of the test method. The end
result of the analysis is a calculated
volumetric flow rate that indicates how fast
fluid is escaping from the tank.

The detection criterion (usually a single
threshold value) is applied after the analysis
has been completed and is  used to
determine whether the level changes are
due to  a leak or to normally occurring
volume fluctuations. If the temperature-
compensated volume change exceeds the
detection criterion, then a leak is
                                        13

-------
C Start
X. ^
forEfh
andSt
Preparation
Heigh
Coe

) >»• /V// Tank
I

IVa/f
?cfs o/ Temperature Fluctuations
ructural Deformation to Subside.
\ '

Install Test Equipment
t
Top Tank
\
Wait
\


Determine
t-to -Volume Conversion Factor
fficient of. Thermal Expansion,
and Groundwater Level


I

f
•* <
\
Make Temperature and
Level Measurements

\ <
Analyze Data
\ '

^pply Detection Criter7or£z>—
} F
&




                                           suspected; if not, it is assumed that no leak
                                           is present.
                                           The height-to-volume conversion factor is
                                           employed to convert level changes
                                           measured during the test to volume
                                           changes. For example, if 5 gallons of liquid
                                           were poured into a container, and
                                           measurement with a ruler showed that the
                                           height of the liquid in that container was 15
                                           inches, a height of 3 inches would equal a
                                           volume of 1 gallon, making the h-to-v
                                           conversion factor equal to 3; this is an
                                           example of a direct physical measurement.
                                           An  h-to-v conversion factor that is derived
                                           mathematically from the geometry of the
                                           surface area of the product in the tank  may
                                           be  in error. In an overfilled tank in
                                           particular, the instantaneous change in the
                                           dimensions of the tank due to structural
                                           deformation or the compression or
                                           expansion of vapor may result in height
                                           changes different from those expected for a
                                           given volume change. Thus, calculating the
                                           h-to-v factor is considerably less reliable
                                           than making a physical measurement.

                                           Using the coefficient of thermal expansion
                                           and the total volume of the product in the
                                           tank, compensation  is made for any
                                           temperature changes recorded. The
                                           temperature-induced volume changes are
                                           then subtracted from the measured volume
                                           changes, yielding the temperature-
                                           compensated volume changes. These  are
                                           used to determine the volumetric flow rate.

                                           The manufacturers of some test methods
                                           use all the data available when  they
                                           perform the analysis to calculate flow rate.
                                           Others do not. One analysis approach is to
                                           subtract end-of-test data from start-of-test
                                           data and divide by the time that has
                                           elapsed between the two. Another is to add
                                           all the cumulative volume changes and
                                           then divide this sum by the duration  of the
                                           test. A third approach is to fit a line to the
                                           data. Based on the result of the analysis,
                                           the tank is declared either leaking or
                                           nonleaking.
Figure 6.    General    procedure     for
            conducting an overfilled-tank
            test.
                                         14

-------
 6.   Performance
Performance is the accuracy and reliability
of a test method in ascertaining whether a
tank is leaking or not. The key to how well a
test method performs is its ability to
discriminate between the "signal," or
volume changes produced by a leak, and
the "noise," or other volume changes that
normally occur in both leaking and
nonleaking tanks. Noise can mask or mimic
the signal  and thus be confused with the
leak. Better performance is expected when
the detectable leak is large; i.e., large leaks
can be detected with greater reliability than
small ones.                •

How Performance Is Defined

Performance is defined by the test
method's probability of detection (PD) and
probability of false alarm (PFA) for each leak
rate that the method claims to be able to
detect. The probability of detection refers to
the test's chances of correctly identifying a
leaking tank compared to its chances of
failing to detect a leak that is actually
present. The probability of false alarm
refers to a test's chances of reporting the
presence of a leak when in facl: none exists.
There are four possible outcomes of a leak
detection test: a correctly identified leak, a
correctly identified tight tank, a false
alarm, and a missed detection. These are
summarized in Table 1.

Two of the outcomes in Table 1 constitute
accurate test results — the measured
conditions reflect the actual conditions. A
leaking tank is correctly identified, or the
integrity of a nonleaking tank is confirmed.
The other two represent erroneous test
results — the measured conditions do not
reflect actual  conditions. In a false alarm, a
test result mistakenly indicates a leak in a
tank that is tight. In a missed detection, a
test result indicates that a tank is tight when
in reality it is  leaking. The likelihood of a
missed detection hinges on the probability
of detection. If the PD is 75%, the likelihood.
of a missed detection is 25%.

Both the PD and the PFA are dependent on
the criterion for declaring a leak, that is, on
the threshold value set by the
manufacturer. If the calculated flow rate
exceeds the threshold, it is assumed that a
leak is present. Once this threshold value
has been selected, the PFA is established; it
does not change, even if the leak rate to be
detected is changed. The PD, however,
does change. The PD increases as the leak
to be detected increases. Stated simply,
there is a better chance of finding large
leaks than small leaks. The threshold can
be changed in order to balance the PD and
PFA in such a way that there is also an
acceptable balance between economic and
environmental risks. If the threshold is  low
(i.e., if very small leaks are to be detected),
the probability of detection is high, but so is
the probability of false alarm. On the other
hand, if the threshold is high, there exists
less chance of false alarm but also a
greater probability of missed detections
(because the PD is lower). Any adjustment
made to the  threshold for the purpose of
improving the PD carries with it an
increased  risk of false alarm. Conversely,
any adjustment made to the threshold for
the purpose  of lowering the PFA
automatically implies an increased risk of
missed detections.

The most commonly used threshold is 0.05
gallons per hour. A threshold of 0.05
gallons per hour might yield high
performance (e.g., a PD greater than 99%)
against a leak rate of 1 gallon per hour, but
low performance (e.g., a PD less than  10%)
against a leak rate of 0.025 gallons per
hour. The  0.05-gallon-per-hour threshold is
often confused with the leak rate to be
detected. If  the threshold is equs\ to the
                                         15

-------
Table 1. Possible Outcomes of a Leak Detection Test
                                                  Actual Conditions
 Conditions as Measured by a Test
Leak
No Leak
Leak
No Leak
Correct Declaration (Leak)
Incorrect Declaration (Missed
Detection)
Incorrect Declaration (False
Alarm)
Correct Declaration (Tight)
leak rate to be detected, the' Pp is only
50% against a leak of  that size. The EPA
requires test methods to have a minimum
detectable leak rate of 0.1 gallons per hour.
In order for a test method to meet this
requirement, its threshold must be less
than 0.1 gallons per hour,

Balancing the PD and PFA

Choosing the right balance between the PD
and PFA is a very difficult task. Missed
detections result in the release of product
into the ground. False alarms lead to the
expense of additional testing and/or the
repair or replacement of tanks that are not
leaking. The clean-up costs resulting from
a missed detection must be weighed
against the cost of unnecessary testing and
repairs resulting from a false alarm.  The
EPA requires that tests be capable of
detecting a leak with a probability of
detection of 95% and a probability of false
alarm of 5%. These, though, are only
   minimum standards, and the tank
   owner/operator may want better accuracy
   as protection'against the possibility of a
   testing mistake.

   Interpreting Manufacturers'
   Claims of  Performance

   It Is anticipated that state and local
   regulatory agencies, as well as owners and
   operators of underground storage tank
   facilities, will hear claims of high
   performance from manufacturers of teak
   detection systems. Because of the highly
   technical nature of volumetric testing, it can
   be difficult to interpret these claims of
   performance. It would be easy if the EPA
   could present  a simple ranking of test
   methods according to their performance
   during the Edison evaluations. Such a
   ranking, however, would be misleading,
   because the Edison experiments showed
   that simple procedural modifications could
   significantly alter the original ranking.
                       Interpreting Claims of Performance

 1.  Is the test method applicable to the specific conditions found at the storage facility
    where it will be used?

 2.  What is the performance of the test method? (Performance should be stated in terms
    of probability of detection and probability of false alarm.)

 3.  Does the performance satisfy the regulatory requirements?

 4.  Does the balance between PD and  PFA adequately minimize the costs associated with
    testing and testing mistakes?

 5.  Has the test method been evaluated? Under what conditions was it evaluated? Does
    the evaluation cover the conditions under which it will actually be used? How much
    data forms the basis of the evaluation?

 6.  What are the main features of the test method? Is the protocol well defined enough
    that it can be repeated identically for each test? Is the test method comparable to the
    one described in "A Reliable Test Method" (in Chapter 9 of this report)?
                                        16

-------
 7.   Results of  the  Edison  Evaluation
Evaluation Approach

The experiments conducted at the EPA
facility in Edison represent the first attempt
at a systematic evaluation of volumetric test
methods. The program was designed to
accommodate a large number of methods,
an approach made possible  by the fact that
there is a great similarity among volumetric
test methods. Based on information
provided by the manufacturer, a
mathematical description was generated for
every test method in the program, and a
report describing each method was
completed. The manufacturer reviewed the
report and concurred that the description
was an accurate representation of his test
method. It was the method described in the
report that was evaluated at Edison. (The
salient features of each test method  can
readily be found in the technical
appendices  to [3], from which proprietary
information has been excised.)

A large computer database of the volume
changes that normally occur in nonleaking
tanks was developed at the Test Apparatus.
These normally occurring changes can
mimic or mask a leak. In particular, the
database included a wide range of
temperature conditions for product, for
example, conditions representative .of
seasonal temperature changes nationwide.

This database was entered into a computer
along with the mathematical description of
each test method. The result obtained was
a simulated  performance of the test  method
under many conditions. Moreover, the use
of computer simulation made it; possible to
replicate this broad range of conditions
identically for each method.  The validity of
the simulated performance was confirmed
during three days of field testing (at  Edison)
for each test method. During that time the
manufacturers, using their own crews and
equipment, conducted tests whose
performance was compared with the
simulated performance of their test
methods. (In addition, the manufacturers'
equipment was calibrated, and the
precision and accuracy of the instruments
were used as input to the simulation.) A
primary goal of the field testing was to
verify that the test method was described
accurately in the report. It is important to
note that the evaluation results presented
here are based on a computer simulation.
The field test data were used to validate the
test-method models that were used  as input
to the simulation. Each performance
estimate was made on the basis of many
tank tests simulated with the test-method
model.


Quantitative Performance
Estimates

The names of the 25 commercially
available volumetric test methods that were
evaluated by the EPA at the UST Test
Apparatus are presented, along with their
manufacturers, in Table 2. The performance
of 19 of the 25 test methods is summarized
in Tables 3 and 4. Proper interpretation of
ttie quantitative performance estimates
given in Table 3 requires the use of Table
4. Table 5 shows the methods for which no
quantitative performance estimates could
be made.

In Table 3, test methods are arranged by
alphabetical order in three categories: (1)
partially-filled-tank test methods, <2)
variable-level, overfilled-tank test methods,
and (3) nearly constant-level, overfilled-tank
test methods. The reason for this
arrangement is that performance 5s largely
controlled by characteristics particular to
each of these categories. For example,
vapor may be trapped during an overfilled-
                                       17

-------
Table 2.  Participants Completing the EPA Volumetric Test Method Evaluation Program
        Test Method Name
    Test Method Manufacturer
Telephone Number
 AESlBrockman Leak Detecting
  System
 Ainlay Tank 'Tegrity Tester
 Automatic Tank Monitor and
  Tester'(AUTAMAT)
 Computerized VPLT Tank Leak
  Testing System
 DWY Leak Sensor
 Gasoline Tank Monitor (GTM)
 Gilbarco Tank Monitor
 Inductive Leak Detector 3 100
 1NSTA-TEST
 Leak Computer
 Leak-O-Meter
 LtguidManager
 LMS-7SO
 MCG-1100
 Mooney Leak Detection System
 OTEC Leak Sensor
 PACE Leak Tester
 Petro Tite
 Portable Small Leak Detector
  (PSLD)
 SMA.R.T.

 Tank Auditor
 Tank Monitoring Device (TMD-1)
 Tank Sentry II
 TLS-250 Tank Level Sensing
  System
Associated Environmental Systems      (805) 393-2212

Solltest, Inc.                          (312) 869-5500
Exxon Research and Engineering        (201) 765-3786
Co.
NDE Technology, Inc.                  (213) 212-5244

DWY Corp.                           (715) 735-9520
Homer Creative Products, Inc.      -  .(517) 684-7180
Tidel Systems                        (214) 416-8222
Gilbarco, Inc.                         (919)292-3011
Sarasota Automations, Inc.             (813) 366-8770
EASI, Inc.                            (219)239-7003
Tank Audit, Inc.                       (619) 693-8277
Fluid Components, Inc.                (SI9) 744-6950
Colt Industries                        (813)882-0663
Pneumercator Co., Inc.                (516) 293-8450
L&JEngineering, Inc.                 (312)396-2600
The Mooney Equipment Co., Inc.         (504) 282-6959
OTEC, Inc.                           (715) 735-9520
PACE (Petroleum Association for        (416) 298-1144
 Conservation of the Canadian
 Environment)
Heath Consultants, Inc.                (617)344-1400
TankTech, Inc.                        (303) 757-7876

Michael & Associates of Columbia,      (803) 786-4192
 Inc.
Leak Detection Systems, Inc.        .    (617)740-1717
Pandel Instruments, Inc.       ~       (214)660-1106
Core Laboratories, Inc.                (512)289-2673
Veeder-Root Co.                      (203)527-7201
tank test but not during a parttally-filled-
tank test. Evaporation and condensation
can be an important source of error in a
partialiy-fiHed-tank test, but they are not a
significant source of error in an overfilled-
tank test. In overtilted-tank tests the tank
must be topped prior to the test, an action
that carries with it the risk of degrading  the
            performance, while in a partially-filled-tank
            test topping, is not. necessary.
            The first column in Table 3 lists the name
            of each method. The second and third
            columns show the mean and the standard
            deviation of the simulated temperature-
            compensated tank test data that were used
            to estimate  performance. The fourth
                                           18

-------
Table 3.  estimates of the Performance of 19 Volumetric Test Methods

        (1)              (2)          (3)-         (4)          (5)          (6)
(7)






Test Method Name





Mean
(gal/h)




Standard
Deviation
(gal/h)
P0 and PFA
to Detect a
" 0.38-Uh
(O.W-gal/h)
Leak Rate
Number of (PD, PFA)
Tests
Smallest
Detectable
leak Rate
with
P0 = 0.95
PFA = 0.05
(gal/h)
" Smallest
Detectable
" Leak Rate
with
PD =0.99
PFA = 0.01
.(gallh)
                              Partially-Filled-Tank Test Methods
Gasoline Tank'
Monitor (GTM)"
Gilbarco Tank
Monitor"
Inductive Leak
Detector 31 00
Tank Sentry II
TLS-250"
- 0.027

0.004

O.OJ5

-0,024
•0.004
O.J08

0.020

0.267

0.041
0.037
13

59

45

23
46
0.73, 0.21

0.96, 0.003

0.72,0.33.

6.89,0.16
0.15,0.001
0.36

0.07

1.12

• 0.15
0.13
0.50

0.12

2.52

0.23
0.23
                       Overfilled-Tank Test Methods/Nearly Constant Level
Leak Computer
MCG-110
Petro Tite
0.001
0.054
0.000
0.025
0.03 1
0.055
132
97
25
0.97, 0.04
0.97, 0.09
0.79, 0.21
0.03
0.10
0.2 1
0.17
0.23
0.29
                          Overfilled-Tank Test Methods/Variable Level
 AES/Brockman        -0.044      0.240        112     0.45,0.34      1.79        3.42
  Leak Detecting
  System**
 Ainlay Tank'Tegrity     0.020      0.124        284     0.50,0.31      0.78        1.04
  Tester
 Computerized VPLT     0.006      0.061         99     0.66,0.19      0.28        0.49
  Tank Leak Testing                                   ••     *           -             ;     :
  Sysfem
 EZYCHEK              0.013      0.049        399     0.86,0.15      0.16        0.25
 Leak-O-Meter         -0.280     '0.547        231     0.57,0.49      1.84        2.85
 LJguidManager         0.081      0.044         79     0,80, 0.14      0.20        0.33
 MooneyLeak         -O.070      0.146        196     0.47,0.38      0.83        1.21
  Detection System
 PACE Leak Tester      0.038      0.214     .'  245     0.37,0.32      1.84        2.94
 Portable Small        -O.051      0.230        135     0.63, 0.32      0.81        1.48
  Leak Detector      .        •     •                                              .       •
  (PSLD)
 S.M.A.R.T.   ,      ...   -0.009      0.097         81     0.58,0.32   .. ... 0.59    „   0.9T ,
 Tank Auditor           0.277      0.292	207     0.57, 0.43	if.66	3.3T

* These test methods were employed in a special precision test mode rather than in their normal
  operating mode as automatic tank gauging systems (ATGS)
"Data analysis algorithms for this method had to be modified in order to determine performance.
                                             19

-------
Table 4. factors That Affect Performance
            Test Method Name
Waiting Period after
Product Delivery (h)
Waiting Period after
    Topping (h)
                             Partially-Filled'Tank Test Methods
Gasoline Tank Monitor"
Gilbarco Tank Monitor*
Inductive Leak Detector 3 WO
Tank Sentry II
TLS-250 '
24
T8
3
24
2
HI A
NIA
N/A
NIA
N/A
                     Overfilled-Tank Test Methods/Nearly Constant Level
 Leak Computer
 MCG-1100
 Petro Tite
     Variable
        2
        0
     Variable
     Variable
     Variable
                        Qverfilled-Tank Test Methods/Variable Head
Alnlay Tank 'Tegrify Tester
AESIBrockman Leak Detecting System
Computerized VPLT Tank Leak Testing
System
EZYCHEK
Leak-O-Meter
LiquidManager
Mooney Leak Detection System
PACE Leak Tester
S.M.A.R.T.
Tank Auditor
8
4
12

6
12
12
12
12
12
8
2
0
0

1
0.75
1
Variable
0
0
0
'Unknown magnitude erf the systematic error introduced by the acoustic measurement of the product
Table 5. Ust of Test Methods Not Evaluated

              Test Method Name
     Reason for Not Evaluating Test Method
 AUTAMAT

 DWY Leak Sensor

 INSTA-TEST

 LMS-750

 OTEC Leak Sensor

 TMD-1
 Operational principles could not be verified
 Operational principles could not be verified
 Did not successfully conduct a tank test
 Operational principles could not be verified
 Improper configuration of the Test Apparatus
 Did not successfully conduct a tank test
column gives the number of simulated test
runs used to make the estimate. Using
these simulated tank test data,
performance curves were generated for
each method. The fifth column presents the
performance that the method should be
able to achieve in actual tank tests. This
performance is expressed in terms of the
probability of detection and probability of
false alarm for leak rates as small as 0,1
  gallon per hour, using the manufacturer's
  detection threshold as defined at the time
  of the evaluation. (The majority of the
  manufacturers used a detection threshold
  of 0.05 gallons per hour to determine
  whether the tank was leaking or not.) The
  sixth and seventh columns present the
  potential performance of the method in
  terms of the minimum detectable leak rate
  that might be achieved if two different
                                          20

-------
probabilities of detection and false alarm
(95 and 5% and 99 and 1%, respectively)
were required. The manufacturer's
detection threshold was not employed in
these performance estimates; instead,
thresholds were selected which yield
probabilities of false alarm of 5% and 1%,
respectively.

Basis for  Performance
Estimates

The quantitative performance estimates
shown in Table 3 should not be used
unless the reader first understands what
sources of noise were included in the
estimates, what sources were not included,
and what assumptions were used in the
analysis. If these are not understood, the
test method that is chosen may yield
significantly poorer performance than would
be expected based on the numbers shown
in Table 3.

Three general remarks apply to nearly all
of the methods evaluated. First, the
majority of the methods exhibit a bias; i.e.,
the mean of the data is statistically different
from zero. Because this means that  the
measured volume is systematically
different from the actual value, a bias can
adversely affect the conclusions drawn
from test results (i.e., declarations of false
alarm, missed detection, tank tightness,
etc., may be in error). The magnitude of the
bias found in most methods is  evidenced In
the mean values shown in Table 3. In
general, the performance of methods with a
bias cannot be accurately estimated,
because unless the physical mechanisms
producing the bias are known and can be
quantified (so that the bias can be
removed), performance can change from
test to test. The bias was removed for  the
estimates presented in Table 3, i.e., the
estimates were made with the assumption
that the bias was zero. If the bias is  large,
i.e., if it represents a large percentage of
the detectable leak, the test method should
be considered suspect.

Second, experimental estimates of the
precision and accuracy of each method's
instrumentation were derived from data
obtained from a calibration of the level and
temperature measurement systems; this
calibration was done as part of the
evaluation. The precision and accuracy
derived from the calibration data were used
to make the performance estimate. It is
assumed that the instruments used in
actual practice are accurately and routinely
calibrated, and that the precision and
accuracy of these instruments are equal to
or better than the precision and accuracy
used in the performance estimate. It was
observed during the evaluation field tests
that the calibration procedures that many of
the manufacturers used or claimed to use
were generally inadequate, and that in
many cases the instruments had never
been calibrated at all.

Third, deviations from the protocol alter
performance; during these evaluations,
performance was seen to improve as well
as to suffer as a result of changes in
protocol. In order to make the performance
estimates, therefore, it was assumed  that
the test protocol as given by the
manufacturer is always followed precisely.
This implicitly assumes that only the  best
test crews are used to execute a test, that
is, the type of crew that participated in the
evaluation program.

Paiiially-Fllled'Tank Tests

The effects of evaporation and
condensation were  not included in the
estimates made for methods that tost in
partially filled tanks. In general, these
effects are small, but in some
circumstances they can be large enough to
produce testing errors. (Neither were  these
effects included in the estimates made for
methods that test in overfilled tanks;
because of the small product-surface area
in an overfilled tank, evaporation and
condensation are negligible.)

Overfilled-Tank Tests/Nearly
Constant  Level

The effects of a product delivery are
included in all the performance estimates
for methods that overfill the tank and
maintain a nearly constant level of product.
However, the  effects of topping the tank
during  an overfilled-tank test were not
included in the performance estimates, nor
were the effects of any produet-ieval
                                        21

-------
changes that are required before starting a
test. Furthermore,  these estimates include
neither the degrading effects of spatial
inhomogeneittes in product temperature
nor the large change in tank volume due to
structural deformation. Waiting periods,
which allow the temperature fluctuations
and the structural deformation produced by
topping to subside, are usually
incorporated into test protocols, but in
general they were found to be too short. It
should be noted that improper topping
protocol (i.e., waiting periods that are too
short) can seriously degrade performance.
The short waiting times after topping shown
in Table 4 suggest the magnitude of the
problem. Thus, in actual practice,
performance could be significantly  poorer
than that shown in Table 3.

In all overfilled-tank tests the potential
exists for trapping vapor in the top of the
tank or in its associated piping. The effects
of trapped vapor were not included in the
performance estimates in Table 3 for two
reasons: most manufacturers  claim to be
able to remove vapor before a test begins,
and the distribution of the volume of
trapped vapor is unknown. That vapor will
be trapped, however, is almost inevitable,
and the performance  estimates  shown in
Table 3 will be reduced if this vapor is not
removed before a test.

Overfilled-Tank Tests/Variable
Level

The same effects that have the  potential to
degrade the performance of constant-level
tests also impact the  performance of
variable-level tests. In addition, when the
product level within the fill tube or
standpipe is allowed to fluctuate during a
test, it is  nearly impossible to convert
product-level measurements to volume.
Test methods that allow the product level
to fluctuate should be considered suspect.

Methods Which Could  Not Be
Evaluated

There were no quantitative performance
estimates made for 6 of the 25 methods in
the evaluation program; these methods are
listed in Table 5. The reason was one of
three.

   1. In two cases, the manufacturer's test
     crew could not perform a satisfactory
     tank test during the 72-hour period
     allotted to them during the field tests
     at the UST Test Apparatus.

   2. In three cases, the data obtained
     during the field tests clearly indicated
     that the method was not behaving as
     the manufacturer had said it would.

   3. In the final case, the Test Apparatus
     was not properly configured, thus
     preventing a fair field test of the
     method.

Summary of Test  Results

Estimates of the potential performance of
each test method, detailed in Table 3, were
summarized in order to show the total
number of methods meeting the detection
standard of 0.1 gallons per hour with
various probabilities of detection and false
alarm. Table 6 shows the number of test
methods attaining  a certain PD and PFA.
with each test method using its own
detection threshold. For example, three of
the test methods have a probability of
detection greater than 90% with an
accompanying probability of false alarm of
10% or less.

Table 7 summarizes potential performance
in terms of the leak rate detectable with  twc
different sets of PD and PFA (95 and 5%
and 99 and  1%, respectively). Five test
methods were able to detect leaks between
0.05 and  0.15 gallons per hour with the PD
of 95% and PFA of 5% required by the EPf
tank tightness regulations. One test methoc
was able to detect a leak of the same size
with the PD of 99% and the PFA of 1 %. A
total of eight methods evaluated could
detect leaks of 0.25 gallons per hour or
less with the PD and PFA specified by the
regulations.

Table 8 gives another summary, an
estimate of the performance that could be
achieved with these methods after
improvements have been made; these
estimates are based on the experimental
                                        22

-------
raWe 6. Summary of Performance Estimates (Performance Is expressed in terms of PD and
        PFA tor the detection of a leak of 0.1 gallon per hour, each manufacturer using his
        own detection threshold.)
                                       PFA (%)
                            Number of Methods Having
                                 This P0 and PFA
           90-100

            65-90

            35-75

            10-20
         0-10

         10-25

         25-50

          0-1
           3

           6

           9

           1
Table 7. Potential Performance in Terms of Leak Rate for Two Different Sets of PD ana PFA

                                    Number of Methods Able to Detect This Leak Rate
  Detectable Leak Rate (gal/h)
with PD = 0.95, PfA = 0.05
with PD = 0.99, PM -• 0.01
          0.05-0.15

          0.15-0.25

          0.25-0.35

          0.35-0.55

          0.55-0.75

          0.75-0.95

            0.95
                                       1

                                       5

                                       2

                                       2

                                       0

                                       2

                                       7
Table 8. Estimate of the Performance after Two Levels of Modifications, Expressed in Terms
        of the Smallest Leak Rate That can be Detected with P0 = 0.39 and PFA =  0.01
          Detectable Leak Rate (gal/h)
              Number of Test Methods Able to Detect This
                  Leak Rate after Minor Modifications
                           (Protocol Only)
                  0.05-0.15

                  0.15-0.25
                                 6

                                13
and theoretical work done during the
program. Table 7 shows that, without
modifications, most systems were not able
to detect leaks smaller than 0.20 (±0.05)
gallons per hour. In Table 8, however, it is
evident that with minor modifications, i.e.,
primarily protocol changes, all the systems
should be able to do at least as well as this.
With more elaborate modifications, the
             majority of systems should be able to
             detect leaks as small as 0.10 (± 0.05)
             gallons per hour. Thus, for many methods,
             a significant increase in performance can
             be achieved by means of protocol changes
             alone. The actual performance
             improvement would depend, however, on
             the specific changes made by the
             manufacturer.
                                          23

-------
Page Intentionally Blank

-------
 8.   Applying  the Results
The temptation to use only those methods
that were ranked highest in this evaluation
should be avoided for two reasons. First,
Table 8 shows that with modification all
methods should be able to perform
accurate tests. Since more than a year has
elapsed since the evaluations were
performed, many methods have had
changes made that should improve their
performance. Second, the quantitative
estimates presented in Table 3 are not
sufficient, when used alone, to assess the
performance of a method, toy system that
meets the EPA regulations and that has
been satisfactorily evaluated should also be
considered.
Five main conclusions are drawn from the
Edison evaluations. First, at the time the
EPA evaluations were done, performance
was significantly less than claimed by most
test-method manufacturers. Second,
volumetric tank testing is complex, but test
methods can achieve high performance if
they follow the principles listed below.
Third, minor modifications should enable
most test methods to follow thesa
principles and thus meet the EPA
requirements. Fourth, evaluation results
should be presented in terms of probability
of detection and probability of false alarm
because this gives a quantitative estimate
of performance. Finally, reliable lank testing
takes time; appropriate waiting periods
should always be observed.


The Edison experiments demonstrated that
volumetric testing is sound in principle and
that most test methods evaluated under the
EPA program can achieve a high level of
performance with only minor modifications.
It is procedure that matters, and procedure
can be changed. Changes based on the
Edison results have already been made to
many test methods.


The ranking of test methods that evolved
from the Edison experiments, therefore, can
be considered valid only for that particular
evaluation. A similar evaluation today might
yield an entirely different performance
ranking. -


If the performance ranking cannot be used,
what was gained from the Edison
experiments? The value of the experiments
lies in the fact that all the test methods
were evaluated under the same sel of
conditions, allowing the features common to
reliable, high-performance tests to be
identified.

High-performance test methods pay careful
attention to:


   • instrument calibration and
     maintenance

   « waiting periods after product delivery
     or adjustments

   • vapor pocket removal {in tests on
     overfilled tanks)

   • adequate spatial coverage  by the
     sensors used to measure temperature

   • data acquisition, processing and
     analysis

   » maintaining a nearly constant product
     level during the test

   • identical execution of each test
     (minimal operator influence)
                                        25

-------
Page Intentionally Blank

-------
 9.   Checklist for  Good  Performance
                            A, Reliable Test Method

 Experimental studies at the UST Test Apparatus suggest that a  test method having the
 characteristics described below should, with proper execution, meet or exceed the EPA
 regulatory requirements for testing tanks of approximately 10,000 gallons in capacity.
 Whether such a system  does or does not meet the regulatory standards depends on the
 implementation of these features.                                           V

    • There are 5 or more temperature sensors (or the equivalent).

    » The sensors can measure volume changes of 0.025 gallons per hour. Temperature
     sensors with a precision of 0.001 °C are adequate.

    • There is a waiting period of at least 6 hours after delivery of product to minrmize
     temperature instabilities.

    « There is a waiting period of at least 3 hours after topping the tank to minimize
     temperature instabilities.

    « Checks are made to identify the presence of structural deformation and to wait for it
     to subside.

    • There is a single threshold value used as a detection criterion.

    « To avoid aliasing, data are sampled at intervals of 1 second in the case of a tank that
     is partially filled or at intervals of 1 to 5 minutes in the case of an overfilled tank.

    « Test length is between 1 and 2 hours. The test is longer If the precision of the
     instruments is less than that given above.

    « The test is conducted  at a nearly constant hydrostatic pressure. For overfilled-tank
     tests this may require  that the product be releveled at regular intervals during the
     test, or that the cross-sectional area of the measurement container be enlarged.

 A reliable test method need not he identical to the system described above, nor contain
 the same features. In order to meet the regulatory requirements, a system need  only be
 capable of detecting a leak of 0.1  gallons  per hour with a PD of 95% and a PFA of 5%.
A test method with good performance is       question should be whether the test method
one that meets or exceeds the EPA           has been evaluated systematically. If it is
regulations. When a regulator is considering    one of those evaluated in the EPA program,
a test method for approval in his/her          the results of the evaluation are available to
jurisdiction, or when an owner/operator is      the public. If the manufacturer has made
considering purchasing one, the fi^st          changes to an evaluated test method,
                                       27

-------
improved performance can only be
determined through reevaluation. If the test
method is not among those evaluated by
the EPA, the regulator should ask to see
other evaluations that may exist.

A checklist for good performance is given
below. It represents a list of features found
in successful test methods.
1. Instrumentation

A well-designed testing system is not
limited by its instrumentation.

   •  Instrumentation noise (fluctuations in
     level and temperature produced by
     the system itself) should be 3 to 5
     times less  than the minimum
     detectable leak rate {the fluctuations
     due to a leak).

   •  The temperature sensors must provide
     adequate spatial coverage of the tank,
     so that the data they record are
     representative of product conditions
     throughout the tank. Generally, one
     sensor is not sufficient, because the
     temperature at the top of the tank may
     be increasing while the temperature at
     the bottom is decreasing. The Edison
     experiments showed that one vertical
     array of five thermistors, or its
     equivalent, provides good spatial
     coverage. At times, however,
     temperature measurement systems
     with less coverage are adequate. For
     example, circulating (or  mixing) the
     product in  the tank can eliminate the
     problem of uneven temperature
     distribution. Thus, methods that
     circulate the product in the tank can, if
     the product is sufficiently mixed to
     eliminate this problem, obtain
     satisfactory results with  a single probe.
     Methods that use averaging probes
     also eliminate this problem.

   « Calibration ensures that
     measurements made by the sensors
     are accurate. All instrumentation
     should be  calibrated periodically. A
     reliable test method makes provisions
     for field calibration checks before each
     test (as part of the protocol) as well as
     for regularly scheduled calibration
     checks.

2. Protocol

The predetermined steps that are followed
in conducting a test are known as the
protocol. With only slight variations from
one test method to another, the protocol
generally follows the description in Figure
6. The seven items below, whether they
occur during preparation, testing, or
analysis, are in the realm of protocol.

   •  Groundwater level can affect the size
     of a leak. If there is any possibility that
     the groundwater is above the  level of
     the tank, an estimate of the height of
     the water table should be made. There
     should be a formal procedure for
     dealing with high groundwater levels.
     For example, a test can be postponed
     until the water table drops below the
     tank. Alternatives can be  used only if
     they are part of the test design and if
     their impact on performance has been
     evaluated.,

     A test should not be conducted while
     the water table is fluctuating. If it is
     impossible to meet this condition, as it
     is sometimes for a tank located in a
     tidewater area, the test for leaks will
     be less sensitive.
   *  The height-to-volume conversion
     factor should be measured directly
     rather than calculated from the tank's
     geometry. The former procedure is
     simple to do and is less subject to
     error than the latter.
   •  Many test methods have temperature
     compensation schemes whose
     accuracy is directly influenced by the
     coefficient of thermal expansion. The
     coefficient is usually determined in
     one of two ways: it is calculated from a
     specific gravity measurement of the
     product in the tank, or an average
     value is calculated for a given type of
     product. The uncertainty  of the
     coefficient is typically 5 to 10%.
   »  Tests should be conducted under
     nearly constant hydrostatic pressure.
     If a test is conducted under variable
                                         28

-------
 hydrostatic pressure, the product level
 is always fluctuating, making it
 impossible to convert product level to
 product volume. It is advisable,
 therefore, to eliminate from
 consideration any test conducted
 under variable hydrostatic pressure.

 A variable-pressure test can be made
 into a constant-pressure test by
 releveling  the product. Another way to
 turn a variable-pressure test into a
 constant-pressure test is by increasing
 the cross-sectional area of the
 measuring container. With  either
 approach, volume changes in the
 product can be determined from the
 height- to-volume conversion factor, as
 they are in any other constant-
 pressure test.

 To minimize the thermal disturbances
 and deformation associated with
 releveling, newly introduced product
 should be at approximately the same
 temperature as the product in the tank;
 it should be added at a location as
 remote as possible from the
 temperature sensors (i.e., through a fill
 tube other than the one where the
 sensors have been inserted or through
 an extension tube that reaches to the
 bottom of the tank, away from the
 sensors); and it should  be added in
 very small increments.
> Adequate  waiting periods must be
 observed after any change in product
 level, whether such a change
 represents the initial product delivery
 or a subsequent adjustment (topping)
 prior to starting the test. A  change in
 product level can disturb the
 distribution of temperature in the tank;
 it also produces structural deformation
 of the tank itself. These two effects of
 a product-level change will eventually
 dissipate: the horizontal temperature
 differences will become small, and the
 tank will cease expanding. However,
 unless the waiting period is long
 enough to allow for this dissipation,
 the volume changes produced by the
 disturbances will obscure the leak,
 rendering  the test results invalid. It is
 recommended that a waiting period of
4 to 6 hours be observed after the
initial filling of the tank, and that at
least 3 hours be allowed to elapse
after the occurrence of any other
change. These waiting periods will
ensure that thermal equilibrium has
returned. Methods that adequately
circulate the product in the tank
theoretically minimize the temperature
effects produced by topping. However,
the time required to mix product
adequately is comparable in length to
the waiting periods that are used when
product is not circulated.

It is much more difficult to know how
much time must be allowed for
structural deformation to subside. It
could be less than for thermal
equilibrium, but it could be much
more, as much as  12 to 24 hours. The
only way, then, to know how long to
wait for structural deformation to
subside is to watch the data. The data
will indicate the presence of structural
deformation and will show when it has
subsided. It is crucial that a test
method incorporate this data
monitoring in its overall design.

The last consideration in that area of
the protocol that deals with waiting
periods is thermal disturbance of the
vapor in a partially filled tank, a
phenomenon that occurs when the fill
hole, is opened in preparation for a
test.Thermal disturbances in the
vapor space cause residual
fluctuations in the product. Limited
experimental data suggest that it takes
at least 6 hours for these to dissipate.

The various waiting periods can be
observed simultaneously providing
that the total period is long enough  to
compensate for all of the above
phenomena.

Data collection is examined in terms
of sampling interval and test duration.
As a rule, the more data the better.
Undersampling the data can lead to
aliasing. To avoid aliasing, level data
should be sampled frequently enough
to measure the fluctuations at arid
beneath the product surface.
                                      29

-------
    Studies have shown that, when the
    instrumentation characteristics are
    marginal, longer tests may be required
    in order to achieve acceptable
    performance. This is especially true of
    many of the temperature
    measurement systems. Higher
    performance is achieved with a longer
    test because averaging of the data
    tends to reduce the uncertainty. The
    widely used 1-hour test duration would
    appear to be acceptable if the
    instrumentation has a high degree of
    precision and accuracy (for example, if
    it has no electronic drift) and if the
    spatial coverage is adequate for
    temperature compensation.
    Temperature measurement systems
    with a precision of 0.001 °C are difficult
    to make and to keep calibrated. As a
    consequence, a test duration of 2
    hours or more  (after the appropriate
    waiting periods have been observed)
    is generally a good choice when high
    performance is desired. The test
    duration to use, however, is the one
    that gives the desired performance.

  • Dealing with vapor pockets is an
    important part  of the protocol. The
    best solution is to eliminate trapped
    vapor from the tank. Although the
    amount of trapped vapor cannot be
    accurately estimated with the current
    technology, its presence (or absence)
    can be determined. A reliable test
    method will check for the presence of
    trapped vapor  and will call for bleeding
    the tank and lines until a subsequent
    check shows that trapped vapor is
    largely absent.

3. Data Analysis and Detection
Criterion

Many test methods lack a defined data-
analysis protocol. In others, an attempt is
made to implement complicated analysis
algorithms whose effect on the test results
is not known. The first step in the analysis
is to compute the volume changes from the
product-level changes (unless volume is
measured directly)  and from the product-
temperature changes. The second is to
calculate the temperature-compensated
volumetric flow rate. Thirdly, the volumetric
flow rate, once it has been estimated, must
be compared to a detection criterion, which
has been predetermined as part of the test
design. In order for a test method to
perform well against small leaks, the
threshold value used as a detection
criterion must be smaller by a factor of 2 or
more than the smallest detectable leak.

4. Operator Influence

Since performance depends on a set,
repeatable procedure,  the most reliable test
methods are those least subject to operator
influence. Changes to the test protocol will
alter the performance of the method. Thus,
any test method which requires or allows
the operator to make subjective decisions
during the test should be avoided. This
does not imply that a trained operator
should not be present. The presence of a
trained operator is essential in preparing
the tank prior to testing, especially in the
case of overfilled tanks. He/she, however,
should not interfere with the test once data
collection has begun.

5. Multiple-Testing Strategy

The essence of a multiple-testing strategy
is that the declaration of a leak depends not
on one test but on the  results of two or
more tests. A multiple-testing strategy can
be useful if it is well defined and has been
evaluated as an integral part of the test
method. Multiple-testing strategies can
increase the PQ and/or lower the PFA if
implemented properly.

Unfortunately, the evaluation of multiple-
testing strategies is complicated because
the tests in a given series are not
necessarily independent. Tests are not
independent when one test is correlated
with the previous one.  When this occurs,
systematic errors remain the same from
one test to another. It is possible to
theoretically calculate the performance of a
multiple-testing strategy if two or more
independent tests are combined.  If the
tests are not independent, experimental
estimates are required. As a general rule,
all multiple-testing strategies should be
evaluated experimentally.
                                        30

-------
=urthermore, many of the multiple-testing
strategies currently in use are arbitrary. For
jxample, if the results show that the
calculated flow rate exceeds the threshold
jy only a small amount, an operator may
Jecide to do a second test even though it
s not a part of the protocol; if the threshold
s not exceeded in the second test, the tank
s declared tight. This procedure lowers the
Drobability of false alarm (i.e., the incidence
Df a "tight" tank being declared leaking) but
at the same time it lowers the probability
that a leaking tank will be detected.

A multiple-testing strategy should be
considered suspect if the number of tests
is not fixed. This means that the PD and
PFA cannot be interpreted. Sometimes one
test after another is conducted until such
time as the leak rate  is smaller than the
threshold value. This approach will
eventually result in a probability of
detection of zero. (A  leak will never be
detected.)

While multiple-testing strategies are
desirable, performance claims based on
multiple-testing strategies should be
viewed cautiously.
                                         31

-------
Page Intentionally Blank

-------
10.   References
  U.S. Environmental Protection Agency.
  Underground Storage Tanks;Part 280-
  TechnicaJ Standards and Corrective
  Action Requirements for Owners and
  Operators of Underground Storage
  Tanks. Federal Register, Vol. 53, No.
  185 (September 23,1988).

  National Fire Protection Association.
  Underground Leakage of Flammable
  and Combustible Liquids. NFPA
  Pamphlet 329, National Fire Protection
  Association, Quincy, Massachusetts
  (December 30, 1986).
3.  U.S. Environmental Protection Agency.
   Evaluation of Volumetric Leak Detection
   Methods for Underground Fuel Storage
   Tanks, Vol. I (EPA/600/2-88/068a) and
   Vol. II (EPA/609/2-88/068b). Risk
   Reduction Engineering Laboratory,
   Edison, New Jersey (December 1988),

       This two-volume report 
-------
Page Intentionally Blank

-------
 11.   Glossary
Accuracy » The difference between a
measured value and the corresponding
actual value; "bias."

Aliasing • An error that occurs when data
are sampled too infrequently to reflect all
the fluctuations in the data.

Bias « See "accuracy."

Coefficient of thermal expansion * A
measure of how much a liquid expands  :
given a certain temperature change,

Database * A collection of many pieces of
information, stored in a computer and
arranged for ease and speed of retrieval.

Deformation * See "Structural
deformation."

Detectable leak rate * The smallest leak,
expressed in terms of gallons per hour or
liters per hour, that a test can reliably
discern with a certain probability  of
detection and a certain probability of false
alarm.

Detection criterion * A predetermined
rule to ascertain whether the tank is leaking
or not. Most volumetric tests use a
"threshold" value as the detection criterion.

False alarm * The  situation that occurs
when a test indicates a leak in a tank that
is, in reality, "tight" (not leaking). See
"missed detection."

Flow rate • The rate, expressed  in gallons
per hour or liters per hour, at which fluid is
escaping from a hole or fissure in a tank.
Groundwater * Water that is found
underground between saturated soil and
rock and that supplies wells and springs.

Height-to-volume conversion factor * A
value used to convert measurements made
in inches or centimeters to a volume
measurement. For example, a value of 3
inches per gallon for the h-to-v might mean
that, in a specific container, the surface of
the liquid rises 3 inches for every gallon of
liquid added.

Mean • A  number that represents the
average value of a group of data.

Missed detection * The situation that
occurs when a test indicates that a tank is
"tight" when in fact it is leaking. See "false
alarm."

Noise • Product-level or product-volume
changes occurring during  a test that are not
related to a leak but that may be mistaken
for one.

Performance • A test's accuracy and
reliability in detecting leaks; expressed in
terms of leak rate in gallons or liters per
hour and defined by a probability of
detection and a probability of false alarm.

Performance Curve •  A graphic
representation of performance data.

Precision • The repeatability of a
measurement.

Probability of detection (PD) • The
likelihood, expressed as a percentage, that
a test method will correctly identify a
leaking tank.
                                        35

-------
Probability of false alarm (PFA) • The
likelihood, expressed as a percentage, that
a test method will incorrectly identify a
"tight" tank as a leaking tank.

Probability of missed detection • The
likelihood, expressed as a percentage, that
a test method will incorrectly identify a
leaking tank as a "tight" tank.

Product • The contents of a storage tank
(in the present context, a petroleum fuel).

Product level • The height of the product
as measured in inches or centimeters from
the bottom of the tank.

Protocol • A series of detailed steps for
conducting a tank test.

Signal • The volume or product-level
change produced by a leak.

Standard Deviation • A number that
represents the amount of variation within a
group of data.

Structural deformation • The distortion
observed in the walls of a tank after liquid
has been added or removed.

Temperature-compensated • Having
been offset or counterbalanced for the
effects of temperature.

Threshold • A predetermined value
against which measurements made during
a test are compared and which serves as
the basis for declaring the presence of a
leak. If, for example, measurements
indicate that fluid is flowing out of a tank at
a rate of 0.2 gallons per hour, and the
protocol of a test states that any
measurement greater than a threshold of
0.1 gallons per hour constitutes, a leak, the
tank is declared leaking.

Tight • In reference to a tank, not leaking.

Topping • The addition of a small amount
of product 1o a nearly filled or overfilled
tank to raise the fluid level in the fill tube or
an above-ground standpipe  to the level
required for  starting a test. Topping often
results in a large height change and
thermal instability of the product in the
tank.

Uncertainty • A measure of the spread in
the data about its mean value.

Variable • A quantity or quality that is
allowed to change within the context of an
experiment.

Volumetric tank test • One of several
types of tests to ascertain the physical
integrity of a storage tank. In a volumetric
test, the volume of fluid in the tank is
measured directly or calculated from
product-level changes; a change in volume
indicates a leak.

Water table • The height of the
groundwater in relation to the surface of th«
earth.
                                         36

-------
This document was written by Joseph W.
Maresca, Jr., and Monique Seibel of Vista
Research, Inc., Mountain View, California,
for the Center for Environmental Research
Information, U.S. Environmental Protection
Agency, Cincinnati, Ohio 45268. Contents
of the document are based on the EPA
report entitled Evaluation of Volumetric
Leak Detection Methods for Underground
Fuel Storage Tanks, Volumes I and II,
available from the National Technical
Information Service (see References).
Technical assistance was provided by
Thomas Young of the EPA Office of
Underground Storage Tanks in Washington,
D.C., and by John S. Farlow of the EPA
Risk Reduction Engineering  Laboratory in
Edison, New Jersey.
Technical reviewers were James H. Pirn,
Suffolk County (New York) Department of
Health Services; Ken Goldstein, New
Jersey Department of Environmental
Protection; Dr. Robert Koerner, Drexef
University; and Dr. H, Kendall Wilcox of
Geraghty and Miller, Inc.

This report has been reviewed in
accordance with U.S. Environmental
Protection Agency policy and approved for
publication. Mention of trade names or
commercial products does not constitute
endorsement or recommendation for use.
Other alternatives may exist or may be
developed.
                                               . GOVERNMENT WUNTING OFflCE: M»2 - MS4BS/4U3Z
                                        37

-------