EPA 510-K-92-810
                      REVIEW OF EFFECTIVENESS OF STATIC TANK TESTING

                                      Jairus D.  Flora

                                MIDWEST RESEARCH INSTITUTE

                                       April 1,  1988
\

       1.    Introduction

            This  paper 1s a review of  the document "Analysis of Static  Tank Testing
       as  a  Leak Detection Technique  for Used Oil  Tanks at  Retail Outlets"  (Ref.
       No.  1).   The paper deals with  two  Issues.   It first  estimates  the  standard
       deviation  of a stick  reading  and  then uses  that  value  1n calculating  an
       average  error  associated  with a   static  tank   test   under  a   variety  of
       scenarios. The error  in the measured leak rate that results from  the sticking
       error is  computed  for  two tank sizes (500  and 1000 gallons [gal.])  under a
       variety  of filling conditions  and  assuming a variety  of forms of  the static
       tank test.  The  paper then recommends a particular form  of the  static  tank
       test based on  consideration  of the error in the  estimated leak  rate and the
       practicalities  of  the  test.


       2.    Summary of the Paper

            The authors  used  a simulation  to estimate the  standard error  of leak rate
       estimates   (called  the  "Effective  Leak  Rate  Error")  for  a  variety  of  con-
       ditions.  Table 3-4 in that report  (see attachment  for  tables from the report)
       contains  these estimates  together  with a  95%  confidence  interval   for  the
       errors.

            The  "effective   leak  rate errors"  of Table 3-4  were arrived at  by
       considering the volume  error induced by the error  in the  level measurement
       resulting  from manual  sticking of  the tank.   Since the  volume  for a given
       error in  level depends on the  cross sectional  area of  the  tank,  an average
       error was  calculated  for the tank,  assuming  a  cycle of filling  and emptying
       the tank.   A simulation was done resulting  in a  number of  volumes in the tank
       and  associated levels and cross  sectional  areas.   The volume errors corre-
       sponding  to  the  assumed  stick reading  error  (of 0.44  Inches  (1n.)  for the
       difference of  two  readings)  were calculated and  averaged  over the different
       volumes.   Simulations with  larger rates of  filling the  tank  resulted  1n a
       smaller number of  terms 1n the average, but for the most  part, the levels in
       the  faster filling rate simulation  would also  be in the  slower  filling rate
       scenarios, giving the same average error.

            Once a standard error in terms  of volume was  determined for a tank size,
       then  it was scaled to reflect  the length of the test.   For example,  consider
       model tank 1,  scenario  2.a.   If the  tank had an average volume error of 5.56
       gal.  corresponding to  a static test, that  1s divided by the time of the test
       to  convert it  to  an error on the leak rate scale.  Thus,  an average error of
       5 56 gal. is 0.93 gal./h for a  6-h test.   If the static test protocol  suggests
       averaging  a  number  of  static tests,  the  leak  rate  error was  reduced by
                                                                         Recycled/Recyclable
                                                                         Printed on paper that contains
                                                                         at least 50% recycled fiber

-------
give the effective leak rate error (0.93/2.64 = 0.35) gal./h.

                                               ®£
    of the Smum volume error that would occur when the tank is half full.





 fled by deleting 2 scenarios and 2 tolumns (model tanks).
      •HIP tahle nresents an interval stated to be  a 95% confidence interval for








  1f the chi-squared distribution had been used.


                 2?«iu2t fffl-Jflndlng put.,  and  for different residence

-------
                   Table 1.  Maximum Errors for Static Tank Tests
       Capacity
       Diameter
       Length
       Cross-sectional
       area at midpoint
       (square inches)

       Volume error for
       0.44 in. level error
       (gal.)
Tank 1

500 gal.
48 in.
5.5 ft.

3168
6.034
Tank 2

1000 gal.
48 in.
11 ft.

6336
12.0685
Tank 3

2000 gal,
64 in.
12 ft.

9216
17.554
       Test scenario

       Daily 6 h                1.01
       Daily 8 h                0.75
       Daily 12 h               0.50

       Ave of 7 daily tests
           6 h                  0.38
           8 h                  0.29
          12 h                  0.19

       Ave of 30 daily tests
           6 h                  0.18
           8 h                  0.14
          12 h                  0.09

       Ave of 4 weekly tests
           6 h                  0.50
           8 h                  0.38
          12 h                  0.25
          24 h                  0.13
          36 h                  0.08
          48 h                  0.06
        Leak Rate Error (qal./hr)

                  2.01
                  1.51
                  1.01
                  0.76
                  0.57
                  0.38
                  0.37
                  0.28
                  0.18
                   1.01
                   0.75
                   0.50
                   0.25
                   0.17
                   0.13
                   2.93
                   2.19
                   1.46
                   1.11
                   0.83
                   0.55
                   0.53
                   0.40
                   0.27
                    1.
                    1.
   .47
   .10
 0.73
 0.37
 0.24
 0.18
OTC73

-------
times  of"  the stick  1n the  product.   Table 5-3  of  the referenced  report
contains estimates of  the standard  deviations  of the  level  reading on  the
stick  for  different  subsets  of  the  data.   (The title  should be  "Standard
Deviation  of ...  H  rather than  "Standard Error	")   The authors  used  the
value  of  0.31  1n.  for all  wood as  the standard  error  1n  a single  stick
reading.   Since  a static tank test requires a stick reading at the  start  and
the end of the test period, the standard error of a single period tank test is
2 times  the standard deviation of a single reading or about  0.44  in.  This
basic  value  of 0.44 1n. has  been Used by the authors throughout the paper as
the standard error 1n height reading for a single static tank test.


3.   Evaluation of the Paper
                                   i.
     In  summary,  the "Effective Leak Rate Errors" presented 1n Table 3-4  are
reasonable estimates for current  practice.   If  special care is  taken  in  the
readings these values can  be  reduced.   Also,  1f  two stick readings were taken
each  time  and averaged each  time) the  tank  1s  stuck  these errors  would be
reduced, by about  3Q%.  This appears a practical  improvement.

     The paper assumes  that volume changes due to  temperature  changes  can be
neglected.   This assumption is questionable.  Assume that the  coefficient of
thermal  expansion for the  product is  about 0.0004 per  Fahrenheit degree.  The
largest  temperature change observed during the  testing on US  EPA's National
Survey of  USTswould correspond to about 0.2 gal./h in the 500- or  1000-gal.
tanks, assuming  that the same heat flow applied  and that the specific heat of
the product  is  the same.   During spring warming we have  observed a  long term
trend  of  product  temperature  in  a half-full  2000-gal.  tank rising  about
0.05 degree  Fahrenheit  per  hour, which  would  correspond  to  an  Increase 1n
volume of  about  0.02 gal./h.   Comparing these volume  changes to the estimated
leak  rate  errors,  the  smaller,  O.;02 gal./h  corresponds  to about 25%  of  the
error  estimated  in the  more precise static tank  tests.   These  examples may be
extreme, but they suggest  that there are conditions under which the effect of
temperature  is not negligible.

     Table  2  is  a  tabulation  of  temperature   changes   that  would  produce
specified  volume changes.  The volume changes included are small, so that they
will  be  comparable  to  a leak rate in  terms  of  gallons per hour.   Thus,  the
temperatures in  Table 2 should be thought of as changes per hour.  Multiplying
these  by the length of  the test  will  give a temperature  change over the test
period that  could affect the  measured leak rate by the amount indicated.


4.    Improvements in Static Tank  Tests
                                   i
     Generally,  it  1s  likely that1 the  temperature effects  will  not be large
relative to  the  sticking error.   However, 1t would be helpful to check this by
making a temperature measurement each  time the tank is  stuck.  Referring to
Table  2,  and   considering   the   recommended  36  hour  test  period,  if  the
temperature   at   the beginning   and  end differs  by  less  than   4  degrees
Fahrenheit,  the  temperature  change per  hour  will  be  0.1 degree Fahrenheit or
less  and the temperature effect  can be  ignored.   However,  temperature changes
Induce a bias or systematic error that  could  cumulate over time.   Expecting

-------
                      Table 2.  Temperature Change Needed to Produce
                                      Volume Changes
                             (Coefficient of Expansion  0.0004)
                                      Product Volume
Volume
Change
0.2
0.1
0.05
0.043
0.021
200
2.50
1.25
0.63
0.54
0.26
300
1.67
0.83
0.42
0.36
0.18
400 500 750
Degrees Fahrenheit
1.25
0.63
0.31
0.27
0.13
1.00
0.50
0.25
0.22
0.11
0.67
0.33
0.17
0.14
0.07
1000
0.50
0.25
0.13
0.11
0.05
1500
0.33
0.17
0.08
0.07
0.04
2000
0.25
0.13
0.06
0.05
0.03
rrrr??

-------
operators "to do  a temperature correction may be impractical,  but  asking that
the temperature be measured and the. results  questioned  1f  a large  temperature
change 1s observed should be possible.

     If the operator would  take  two stick readings and use the  average level
each time the  level is determined,  a substantial reduction in the  error would
be obtained.   This would reduce the standard error of the  difference from the
assumed 0.44 in.  to 0.31 in..  This would reduce the  errors of the leak rates
by about  3Qfc  and would make the procedure better.  (Requiring three readings
and an average each time would reduce  the error from 0.44 in. to O.p4 in. and
the average of four readings each time would reduce it to 0.22 in..)


5.   Implications of the Error Rates for Method Performance

     If the  "Effective Leak Rate Errors"  are used as  representing the error
associated with  a static tank test, i  the test performance  can be  calculated.
Since the errors  result from  reading errors  on the stick,  they will generally
follow  a  normal  distribution closely.   Assuming the normal  distribution for
the  errors,  there  are two approaches  that  can be taken   in  determining the
performance of the method.  The  first  is to calculate the threshold leak rate
that  would  correspond to  a  IX  false  alarm rate  and  then determine  the
probability of detection of specified  leak rates, particularly 0.2  gal./h. The
second  is to specify the decision threshold as 0.1 gal./h (midway between zero
and the performnce  standard of 0.2  gal./h)  and calculate the probabilities of
false  alarm and  of  detection.   A  third consideration is  that  the procedure
will  be  repeated periodically,  generally monthly,  so that  performance will
Improve over time.

     Let  L denote  the leak rate to,  be  detected,  let C  denote the critical
measured  value for  declaring  the  tank  to be leaking,  and let R denote the
measured  rate.  Then the probability;of  a false alarm 1s given by

     P(FA) »  P(R < C|0),

where  the value  after the vertical bar denotes  the true leak rate (zero if a
false  alarm occurs).   Note that,  following  the usual  convention, leak rates
are represented  as  negative numbers since they correspond to volume reductions
or losses from the tank.  Using the assumption of.the  normal  distribution for
the measuring errors,  and using  s to denote  the  standard error of  the measure-
ment,  we  have

      P(FA)  *  P(  R/s < C/s|0)

            -  D (C/s),

where D  denotes the standard normal  distribution value.   The probability  of
 detecting a leak of size L is the  probability that the measured leak rate,  R,
 is less than  the threshold, C, when the true leak  rate  is  L.   This probability
 1s given by

      P(D) » P(R < C|L)

           = P[(R-L)/s < (C-L)/s|L]

-------
          = D {(C-L)/s}.

     Taking weekly 36-hour static tests, averaged  over  four  weeks each month,
the  estimated  standard  error  is  0.08  gal./h  for  a  500-gal.  tank  and
0.16 gal./h for a 1000-gal. tank.   We  will  use these values  to illustrate the
results.
     First  we  set the  probability  of  a  false  alarm  to  be 0.01.
corresponds to the tabled value of -2.326 from the normal  distribution.
1ng for the threshold, C, gives
                                                                     This
                                                                    Solv-
or
C/s = - 2.326 ,

C = -2.326 s.
     Substituting  the  values  of  the  standard  error,  the  threshold  for
declaring a leak would  be  a  measured  leak rate of -0.186 gal./h or more for a
500-gal. tank  and  -0.372 gal./h or more  for  a 1000-gal.  tank,  with the nega-
tive  sign  denoting that the  rate represents  a  volume  loss (flow  out  of the
tank).   Substituting  these values for C  one  can determine the  probability of
detecting a leak of -0.2 gal./h in a single month as

     P(D|L=-.2) = D {(C +  .2)/s}

                 , D  {0.175}  for the 500-gal. tank
                 *• D  £-1.075} for the 1000-gal.  tank.

     The tabulated values  of the normal  distribution give the  probability of
detection of a leak rate of -0.2 gal./h as 0.57 for the 500-gal, tank and 0.14
for the  1000-gal. tank.

     If we set C = -0.1 gal./h as the threshold for declaring a leak, then the
symmetry of the normal  distribution will make  the probability of the two error
types  (false  alarm   and  missed  detection of a  leak  rate  of -0.2  gal./h)
equal.  Substituting  Into  the formulas gives
     P(FA) = 0.11
     P(D)  = 0.89
for a 500-gal. tank and
     P(FA) = 0.27
     P(D)  = 0.73
for  a 1000-gal. tank.
 6.    Modifications to  Improve Performance

      If the average of two  stick  readings  were used throughout, the standard
 error of level measurement would  be  reduced  from 0.44 in.  to 0.31 in..  This
 would have  the effect of  changing the thresholds,  C,  for  a  probability of
 false alarm of 0.01 to -0.133  gal./h for  a 500-gal. tank and to -0.263 for a
 1000 gal. tank.   The probabilities of detecting  a  leak  of -0.2 gal./h would
 then become 0.88  and  0.29,  respectively.

-------
     If 'the  fixed threshold  of  -0.1 gal./h  were  used with  this  reduced
standard error, the probability of a  false  alarm  for a 500-gal.  tank would be
reduced to  0.04  and  the probability of detecting a  leak  of  -0.2 gal./h would
increase to 0.96.   The corresponding numbers for a  1000-gal. tank would be a
false alarm rate of 0.19 and a probability of detection of 0.81.

     If the duration  of the test  -Is  extended,  an improvement in the  size of
the leak that can be detected can be obtained.  Extending the time of the leak
test decreases the leak  rate that  can be  detected in proportion  to the length
of the test.   Extending the time from 36 to 48 hours  reduces the size of the
detectable leak by a factor of 36/48 = 0.75.  Extending the time  from 36 to 56
hours would reduce the size of the detectable leak by 36/56 = 0.643.

     If both  of these modifications  (2  stick readings and  longer tests) are
performed, the performance of the method is improved.  For example, with a 48-
hour  test  in a  500-gallon tank,, the  standard   error  is  reduced  to 0.042
gaWh.   This, 1n turn,  implies  a' threshold of 0.099  gal./h for declaring a
leak with  a 1% false  alarm rate and  means that  a  leak  rate of 0.197 gal./h
could  be  detected with  probability  99%.   Table  3  summarizes  the performance
that would  be expected from this modification for tank sizes of 500,  1000, and
2000 gallons and for tests with a duration of 36,  48, and 56 hours.  Note that
a test of 56 hours could be done by not adding product to the tank from Friday
evening until Monday morning.   (With  a tank used to  hold used oil, this may be
a  reasonable  time  period,  as the  oil is  usually  accumulated  in a drum and
dumped  periodically.)    It  should  be emphasized that  the detectable leaks In
Table  3  are  the  result of  a monthly  average  of  4  weekly  tests  and that
duplicate stick readings are used  at  the beginning and end of each test.

     Table  4  is  a  tabulation of! performance  information  for  a variety of
standard  (steel) tank  sizes.  The  tank lists the nominal  volume of the  tank as
well  as its dimensions  (length and  diameter) in  inches.   Using  the change In
level   that  corresponds  to  the  1%  false  alarm  rate  at  the  threshold
 faDDroximately 0.75 in.,  assuming duplicate stick  readings  at  the beginning
and  end of each  static test), the  average volume change that corresponds to
the  threshold for a single  test  is tabulated.   This average was estimated as
85%  of the maximum volume  that would occur when  the tank was half full.  The
next column in the table is the average volume changes that would be  detected
with 99%  probability for a single test.  The last two  columns are the  1% false
 alara ttSioTds  Sd  the  volumes  detectable  with  99% .probability  for  the
monthly test  that averages 4 weekly tests.

      Table 5  is  a tabulation of performance similar  to  Table 4.  In  Table  5
 the  volume  changes  detectable  with 99%  probably  in  the  monthly  (4  week
 average)  tests  have been converted  to  leak rates  for three test  durations:
 36? 48, and  56  hours.   As can be ;seen,  only the 500  andI 550 gallon tanks can
 achieve detection of 0.2 gal./h leak rates with probability  of 99% and a false
 alarm rate of 1%.

      As a result of somewhat disappointing performance of the static tank test
 for tank sizes  larger than 550 gallons, another modification to  the  test was
 evaluated.   Rather than  require owners and operators to perform  the static
 test  every week and average  the  results monthly, one could require  that the
 test  be  done with the tank at a! specific percent of capacity, for  example,
 95%    This has several  advantages.   First, it  ensures that the test checks
                                        8

-------
                                   TABLE 3

        STANDARD ERRORS, THRESHOLDS, AND DETECTABLE LEAK RATE (gal/h)
                         FOR 4 WEEKLY TESTS AVERAGED3


500 gal. Tank
Standard Error
Threshold at 1%
Leak Detectable at 99%
1,000 gal. Tank
Standard Error
Threshold at 1%
Leak Detectable at 99%
2,000 gal. Tank
Standard Error
Threshold at IX
Leak Detectable at 99%
TEST
36

0.057
0.132
0.264

0.107
0.249
0.498

0.146
0.340
0.679
DURATION
48

0.042
0.099
0.197

0.082
0.191
0.382

0.109
0.254
0.507
(HOURS)
56

0.036
0.085
0.170

0.069
0.160
0.320

0.094
0.219
0.437
a  Assuming duplicate stick readings.
OTC73A

-------
                                                  TABLE 4
                                     TANK SIZES AND THRESHOLD VOLUMES



NOMINAL
VOLUME (gal.)
500
550
1,000
1,000 •
2,000
3,000
4,000
5,000
6,000
8,000
10,000
12,000

TANK
DIMENSIONS
DIAM. LENGTH
(IN.) (IN.)
48 66
48 72
48 128
64 73
64 144
64 216
64 288
96 ,. 160
96 192
96 256
96 320
96 384
AVERAGE IX
FALSE ALARM
VOL. CHANGE
THRESHOLD (gal.)
(SINGLE TEST)
8.7
9.5
17.0
12.9
25.4
38.2
50.9
42.4
50.9
67.8
84.8
101.7
AVERAGE VOL.
DETECTED
WITH 99*
PROB. (gal.)
(SINGLE TEST)
17.5
19.1
33.9
25.8
50.9
76.3
101.7
84.8
101.7
135.6
169.6
203.5
AVERAGE 1%
FALSE ALARM
VOL. CHANGE
THRESHOLD (gal.)
(MONTHLY)*1
4.4
4.8
8.5
6.4
12.7
19.1
25.4
21.2
25.4
33.9
42.4
50.9
AVERAGE VOL.
DETECTED
WITH 99X
PROB. (gal.)
(MONTHLY)4
8.7
9.5
17.0
12.9
25.4
38.2
50.9
42.4
50.9
67.8
84.8
101.7
a
   4 test average.
OTC73A

-------
                                   TABLE 5


              TANK SIZE,  THRESHOLDS,  AND  DETECTABLE LEAK RATES
TANK
SIZE
NOHIAL
VOLUME (gal
500
550
1,000
1,000
2,000
3,000
4,000
5,000
6,000
8,000
10,000
12,000
DIMENSIONS
DIAM.
1.) dn.]
48
48
48
64
64
64
64
96
96
96
96
96
LENGTH
i (1n.)
66
72
128
73
144
216
288
160
192
256
320
384
AVERAGE IX
FALSE ALARM
LEAK RATES DETECTABLE .
WITH 99* PROBABILITY (gal./h)b
VOL. CHANGE TEST DURATION (HOURS)
THRESHOLDa(qa1.) 36 48 56
4.4
4.8
8.5
6.4
12.7
19.1
25.4
21.2
25.4
33.9
42.4
50.9
0.243
0.265
0.471
0.358
0.706
1.060
1.413
1.177
1.413
1.884
2.355
2.826
0.182
0.199
0.353
0.269
0.530
0.795
1.060
0.883
1.060
1.413
1.766
2.119
0.156
0.170
. 0.303
0.230
0.454
0.681
0.908
0.757
0.908
1.211
1.514
1.817
a
b
4 Weeks Average (gal.)
Value from last column of Table 4 divided by test duration.
OTC73A
                                      11

-------
nearly all of  the tank for possible holes.  Second, 1t will result in a much
smaller surface  area for the product during the  test,  and, as a result, the
level  change  will  correspond  to a  smaller  volume  change,  increasing the
performance of the test.  Third, it should reduce  the  burden of testing on the
Swne? and operator, since they will now be required to test only each  time the
tank reaches 95% capacity, or, at most once a month.

     Tables  6A-6C  contain  information  about the  static  tank  test  when
performed  at  95*  of  the capacity, of  the tanks.   The  three standard tank
diameters  (for steel tanks) of  48,  64,  and 96  in. are Included.  In addition,
the standard tank  lengths and nominal volumes  are included for tanks  from 500
tra to 12,000 gal.  For each tank diameter, the  distance  from the bottom of the
tank that  corresponds  to 95* of  the  capacity  has been calculated.   Thus,  an
Imlr or operator would be required to allow the  used  oil  tank to  reach this
level and then test.  Following the,test, the oil  could  be pumped out.

     Assuming  that duplicate  stick Readings are used  at the beginning and end
of each test,  the standard error  of the difference in height is 0.31  in.  The
SatS tink  test indicates a  possible  leak if the product  height  drops more
than 3/4 inch  from start to finish (the actual number is 0.31 x 2.326 -  0.721
in   for  a 1*  false  alarm rate,  but  three-fourths of an  inch  is a practical
field  number).  In the first section  of the table, for each tank diameter the
standard error of the  height differences, the height difference threshold,  and
the  difference  detectable with  probability  of  99*  have  been  converted  to
gal.   TMs  conversion  results from  multiplying  the  height times the  cross
sectional  area  of  the  tank at  the  95*  capacity level  and  expressing  the
resulting  volume in  gal.

     Tables  6A-6C also present  the volume  change that can be detected with 99*
probability  in  terms  of the  equivalent leak rate  for  tests  of different
durations.  The  durations  range  from 12  hours to 56 hours, corresponding  to
durations  from overnight to  over a weekend from Friday evening until  Monday
morning.   The results are not restricted to used  oil,  but are valid for  any
product.

     The results  in Tables  6A-6C do not  include any temperature correction.
The convention followed is that  an increase  in volume requires use of water-
finding  paste and  action if water is found.  A  decrease in volume signals a
 leak.   Temperature  changes could cause a-false  alarm  or a missed detection.
 However, incorporating a temperature  correction would substantially complicate
 the static tank test.

      Calculations show that the cross sectional  area of  a  *** at.  95*  of
 capacity  is about  59* of the maximum cross sectional area of the tank   As a
 result, requiring that  the static tank test be performed when  a tank  is filled
 to 95* of capacity  reduces the volume  error by only  a  factor  of 0.59 (i.e. to
 about 60* of  the maximum volume  error  when the tank  is half furl).   As can be
 seen in Table 6,  a  leak rate of  0.2  gal./h is detectable with 99*  probability
 onlj for tanks of about 550 gal. or less.  Thus,  the  performance of the  single
 static test at 95* capacity  is  about  the same as the  average  performance of
 the  average  of  four  weekly  tests  of the same  duration  (see  Table 5).    It
 should be noted that the weekly tests use an average  error that is  about  85*
 of the maximum  error when the  tank is half full.   The test at 95% capacity  has
 a fixed error size.               ;
                                       12

-------
 TANK
DIAMETER
                                      TABLE 6A

                   PERFORMANCE OF MONTHLY STATIC TANK TEST AT 95X
                  OF CAPACITY FOR VARIOUS TANKS AND TEST DURATIONS
                         TANK LENGTH  Mn.)
                        66	~~^I12
                        NOMINAL VOLUME (qal.)
                                        500
                                 550
1,000
48
(43.4)a
Standard Error         2.515    2.743     4.877
1* Threshold (gal)     5.849    6.381    11.344
99*DrtSlonSal)   11-698   12.762    22.688
           .eak
     Detectable with 99% Probability fgal/h);
12-hour test
24-hour test
36-hour test
48-hour test
56-hour test
0.975
0.487
0.325
0.244
0.209
1.063
0.532
0.354
0.266
0.228
1.891
0.945
0.630
0.473
0.405
• • '•
    value 1n parenthesis 1s the distance from tank bottom for 95% capacity,
    OTC73A
                                           13

-------
                                      TABLE 68

                   PERFORMANCE OF MONTHLY STATIC TANK TEST AT95*
                  OF CAPACITY FOR VARIOUS TANKS AND TEST DURATIONS
 TANK
DIAMETER (1n.)
                              TANK  LENGTH  (in.)	
                        73	144     216       288
                              NOMINAL VOLUME  (gaT-1
                                       1,000    2,000    3,000     4,000
64
(57.8)a
Standard Error         3.709    7.316    10.973
1% Threshold (gal)     8.626   17.016    25.524
99% Detection (gal)   17.252   34.032    51.048
14.631
34.032
68.064
           Leak  Rates Detectable with 99% Probability (gal/h):
                 12-hour test
                 24-hour test
                 36-hour test
                 48-hour test
                 56-hour test
1.438
0.719
0.479
0.359
0.308
2.836
1.418
0.945
0.709
0.608
4.254
2.127
1.418
1.063
0.912
5.672
2-836
1.891
1.418
1.215
 a  Value in parenthesis is the distance from tank bottom for 95% capacity.
    OTC73A
                                          14

-------
                                      TABLE 6C

                   PERFORMANCE OF MONTHLY STATIC TANK TEST AT 95X
                  OF CAPACITY FOR VARIOUS TANKS AND TEST DURATIONS
 TANK
DIAMETER (1n.)
                                                   TANK LENGTH (1n.)
  160      192       256
        NOMINAL VOLUME (gal.)
                     320
                  384
96
(86.7)a
                Standard Error
                1% Threshold (gal)
                99% Detection (gal)
                                       5,000    6,000     8,000   10,000  12,000
12.19
28.36
56.72
14.63
34.03
68.06
19.51
45.38
90.75
 24.39
 56.72
113.44
          Leak Rates Detectable with 99% Probability (gal/h):
 29.26
 68.06
136.13
12-hour test
24-hour test
36-hour test
48-hour test
56-hour test
4.727
2.363
1.576
1.182
1.013
5.672
2.836
1.891
1.418
1.215
7.563
3.781
2.521
1.891
1.621
9.453
4.727
3.151
2.363
2.026
11.344
5.672
3.781
2.836
2.431
a  Value  1n parenthesis  is the distance from tank bottom for 95% capacity.
   OTC73A
                                         15

-------
     One'can  obtain  somewhat  better1 performance by averaging  the  four weekly
tests.   However,  the single test at 95* capacity may  be preferred by some in
that  It requires  less  testing  for  about the  same preformance.   It  may be
easier to comply with than the averaging of four weekly tests.

     One could  improve  the performance by increasing  the level of product in
the tank when the test is to be done.  Requiring the  test at 99* of capacity
would reduce  the  size of  the  detectable leak rates  by a factor of 0.6 for the
numbers  in Table 6.  This would require product levels of at least 46.6 in. in
a 48-in. diameter tank, 61.9  1n. 1n.a 64-1n. diameter tank, and 9,5.8 in. in a
96-ln   diameter  tank.    From a  practical   standpoint,  requiring  any higher
levels  would  almost  amount to requiring that the product be brought into the
fill nine for testing.  This  would probably require monitoring to ensure that
tempeJatuWchanges  did not  influence  the  test and to  ensure that expansion
did not cause an  overflow.          '

      In summary,  either the weekly static test averaged  over four weeks with a
duration of  48i hours or longer or a single test of 56 hours duration when the
oroduct levels is at  least  95* of capacity will  give  a  detection level of
about 0.2  gll./h in  small (550 gal.[ or less) tanks.   Operators could be given
the  ootion  of which  they prefer to use.  Static tank  tests in larger tanks do
not  reach  the 0.2gal.%  leak  rate detectable  with 99* probability for tests
of reliable duration.   Requiring that  the test be done at  95* capacity and
averaged over 4  weeks  would  extend the  size  of tanks  up  to 1,000 gal.,.and
almost UP to  2,000  gal.  if a 56-hour test  is used.   Requiring the test  level
to be  99*  of capacity would result in  about  a 40*  improvement (reduce the
detectable leak by a factor of 0.6);,  but  may not be practical.


 7.   Suggested Operating  Protocol

      The following is suggested at a practical  operating procedure for using  a
 static tank test as  a leak detection method for used oil tanks.

  1.  Take a stick reading on a tank and write down  the depth of the product.

  2.  Wipe off the stick, take a secpnd stick reading,  and write down the depth
      of the product.

  3.  Add these two depths and divide by two to find an average depth.

  4   Convert this average depth of  product to gallons  using the tank chart.
      Record  the  volume and the time of the stick readings.

  5.  After the set  period (e.g., 48 hours) has  elapsed, stick the tank again
      and record  the depth of the product.

  6.  Wipe off the stick,  stick  the ;tank a second time,  and record the depth of
      the product.

  7.  Add these two  depths and  divide by two to obtain an average reading.

   8.   Convert this  depth to  volume  by  using  the  tank chart and  record the
       volume  and time  at  the  end of the static  test period.
                                        16

-------
 9.  If the depth at the end of  the  period 1s more then 3/4 inch greater than
     that at the  start  period,  wipe off the  stick,  apply water-finding  paste
     to the lower end, and re-stick the tank.

10.  If the water-finding paste  Indicates  that water is  present,  the  increase
     in  level  could  be cause  by water  coming  into the  tank.   Unless  an
     alternative source for the  water is  identified, this  should  be  regarded
     as evidence that the tanks  may  have a hole  in it,  allowing water to come
     in.

11   If the  depth at the end of the period  is  more than 3/4  inch less than
     that at the  start  of the period, this  1s evidence  that product  has been
     lost from the tank, and the tank may be leaking.

12   If the depth at the beginning and end of the static tank test period does
     not change by more than 3/4 in. 1n either direction,  subtract the volume
     at the end of the period from that recorded at the start of the period.

13.  At the  end of each four-week period,  add the four differences in volumes
     corresponding to the weekly static tests.  Divide this sum by four.

14.  For  a 500-gal.  tank  (nominal size)  and  a  48-hour test,   if  the average
     volume  change  indicates  a  loss of more than 4.75  gal. per weekly  test,
     the tanks fails the static tank test.

     The  next  to  last column in  Table  4 provides volume thresholds for other
tank sizes.   Table 3 gives the  leak rate detectable with  99%  probability for
some tank  sizes and  test durations.   These should be used  1n  determining the
appropriate  duration of the static tank test.

     The  above protocol  may  be  modified to  improve the performance of the
static tank test  by requiring   the  level  1n the tank  to  be  at 95%  of tank
capacity.   The modification required ensures that the depth of the product In
the tank  is  high  enough before beginning the  static  test.

     If the  diameter of the tank is  48  in., the product  level must be at least
43.3 1n.


     If the diameter of the tank is  64  in.,  the product  level in the tank must
be at  lest 57.75  in.

     If the diameter of the tank 1s  96  in.,  the product  level must be at least
86.65  in.

     After ensuring  that the required level of product  has been met, refer to
Table  6 to  determine an appropriate test duration  for  the tank size and the
required detectable  leak  rate.

     When the minimum product  level has been determined at the test duration
 set, proceed as before 1n steps 1 through 11 above.

     With this modification (testing at 95%  of capacity), there is no need to


                                       17

-------
<*.
         loss of product and a positive sign a gain of product).
         8.   Effect of Monthly Tests
         ind                                       is? J
         Sonth to determine whether there was  evidence of a leak.







          leak eacl! ,  55h for n months is P raised to the n-th power.

               Tf the error in failing  to detect a leak results from random errors in




          the leak woSld be 0^ The nSmber of monthly tests required  is given by

               n = log(0.01)/log(P),




          Teak for each of 7 consecutive months.
           perpetually.
                                                 18

-------
Reference "

Shareef, G.  S.  and Dicker-man,  J.  C., "Analysis of  Static Tank Testing  as a
Leak  Detection  Technique  for Used Oil  Tanks at  Retail Outlets."   American
Petroleum Institute, February 1987.
                                       19

-------

-------
Attachment
                              MiS'-VEST RESEARCH KG
0
1 ,
                                 TABLE 3-1.  MODEL TANK PARAMETERS

Model
Tank No.
1
2
3
4
Nominal
Tank Capacity

-------

-------
CVI
*.•>
o:
-i:
IU
CM
in
o:
                                                  u.> m •«•»
      stroi Of0*0 uoi*i
      SU'O) Old IWI
      OM-OI OU'I ««•«
                         «trn
                         IMI1
                         «*»•*
                          tiro
*WOI ««'0
KI-0) l«"0
OK'O) 09»*0

Mf *0» W'O
1»'«» »»*•
           uoro - xsroi oc»*o
           1IWI - IM'OI OW«
           inn • TWC*oi owo
IOMTO - 010*01 OWO
twro - »!»•« HO*O
iwro - iwoi wro
nwo - otroi o»r*o
roi owo
                                           I0f»*0 - MfOI OWO
1M»*0 - Stt'OI

«§«rs -
            IM*0  IS»>-0 - 90TOI

            iU'9  ««W9 - WE'08
            WO  1CTO-8 - 0»KO»
UWI - «I*OI OTW-0
ms-r- own ozn
ISWI - »WII OMM
                                                 - iwo> owo
                                           IH»*1 - »IMI OIC'I
                                                 - KS'» OHM
     I9t9*o - rUro> otro
     1000*1 • 0«'0» OWO
     I0»*l - IH'OI OWO
IOWO - 110*01 OWO
i«ro-«wo> 000*0
uiro - iwoi OEro
IOK-O - uroi ««-o
»««•« - wro> *«-o
uwo - wrw ot»-t
            tWT9 - KC'OI 0»»'0

            1010*0 - IWm 099*0
            C1WI - OWOI 000-0 I9K'I
            io»ro - «o-oi owo iiM-o
              ro - cwoi otro «o«ro
                • - twoi oiro
     wwo - «rot
            nc-oi owo
                                    uu*o -
                             toifo
                        9WO IllfO
            iw-o - 9tro» off-o uwo
                                                            ((O'OI OWO
                                                            «io*oi oero
                                                            »woi oiro
            oii'oi oiro
            itroi oiro
           -101*01 ost*o
            iwi *o -
            lowo-
            uori -
                                                0(9*0
      IMJ'O - 9fi*OI OM*«
      iftsro - itcot owo
      ww.j. PW.9j «.«•«
                                                                                                                 *«-«  "»
                                                                                                                 J1-0  *«
                                             *»

                                             ••-

                                        Jf-M  f

                                        J»-«  •>
                                        JWI-f  'I
                                                                                        J1-0  *«

                                                                                        JM-f  '•
                                                                                                        «O>N •mi
                                                                                                                       *»
                                                                         ••»• ot -HiMH-i'nm *u««  *«
                                                                                                                          •s
                                                                                        Jt-9 *•
                                                                 urn Mrii j»iiM*n  *»-E

-------

-------
Attachment                  MIDWEST RESEARCH KG                               P.  4/4
                TABLE S-.3.   STANDARD ERROR  OF KAKVAL GAUGING RESULTS
                     '         FOR DIFFERENT SUBDIVISIONS OF. THE DATA
                                                       Standard Error
                            Daea S«c      -                (Inehea)
                       All dtea                              0*^0
                       Fib*rsl«ai>                            0.25
                       Wood                                  0.31
                       7ib«rgla*;t» no p*»t«                   0.20
                       FibarfUso, p«ae«                     0.27
                       Woodi &e p*0e«                        0.36
                       Wood, P*»:«                           0.2S

-------

-------
                •a
                 V


                 ?
           ««

           i
      o

 ,-   <:

I'
                r
0

3
 6   I
 «n
 £
 <»   \5
 f*   C^
           I?
-   N

 6    6
 S
 i
 4
 5
            u


            I
            o
                 V)
                 
-------

-------