DRAFT REPORT
Submitted by:
Pope-Reid Associates, Inc.
245 E. 6th Street, Suite 813
St. Paul, MN 55101
-------
APPENDIX A
PROBABILITY AND RELEASE VOLUME CALCULATIONS
-------
A.O INTRODUCTION
This appendix consists of two parts: the derivation of formulas used throughout
the failure analysis, and the documentation of the failure probabilities and
leak rates used for the Individual failure events. The general derivations are
presented on pages A-2 through A-ll. The individual failure events are
discussed in the remainder of this Appendix.
A-l
-------
A.I GENERAL DERIVATIONS
The following computations were common to many of the failure events:
Calculation of release rates from underground leaks (fluid bed model);
Calculation of release rates into an air environment (Bernoulli flow);
t Conversion from hydraulic head to pressure;
Computation of fluid velocity in pipes;
Determination of pressure of flowing liquid in pipes; and
Calculation of fill/discharge times.
These calculations are described in the following pages.
A-2
-------
A. 1.1. Underground Releases (for leaks that are impeded by soil or backfill)
EQUATIONS:
We calculated underground leak rates using the Ergun equation for the
pressure drop In beds of mixed particles (Source: Fluldization
Engineering, Dalzo Kun11 and Octave Levensplel, p. /u U969)).
This equation 1s:
where
P « the pressure drop across the bed
Em * the void fraction for the particles
B » viscosity
u0 * velocity of flow
4,5 » sphericity of the particles
.. p « density of fluid
tf
dp average particle size
9c « 1 In metric units
L * the dispersion length
This Is a quadratic equation 1n u0. Thus,
. -B + (82 . 4AC)-5
2A
where A- A-75(1-6.) \ / P
- /1.
\
B . 150(1^)2
Em3 ( s
A-3
-------
The leak rate is simply
dQ/dt » u0A
L for fluid beds is approximately 4 to 5 bed diameters. (Source:
R. H. Perry and C. H. Chi 1 ton, Chemical Engineers' Handbook. 5th
Ed (1973), pp. 5-49). The underground leaks studied in this model,
however. Involve conical dispersion patterns, rather than the cylindri-
cal fluid beds considered by Perry and Chi 1 ton. In addition, L will
also vary-with the dispersion characteristics of the surrounding back-
fill. After extensive analysis and sample calculations, we have chosen
the following dispersion lengths:
Backfill Material
Circular holes
(usual value)
(maximum permissible)
Cracks
Clay Silt Sand
2d , 7.5d 20d
2 cm 7.5 cm 20 cm
Gravel
lOOd "
100 cm
(usual value) 4 w
(maximum permissible) 4 cm
20 w
20 cm
40 w
40 cm
100 w
100 cm
where:
d is the hole diameter, and
w is the crack width
Additional parameters are as follows:
Backfill Material
void fraction
particle size (mm)
sphericity
Clay
0.95
0.002
0.075
Silt
0.75
0.064
0.34
Sand
0.53
0.25
0.65
Gravel
0.50
9.4
0.70
A-4
-------
USER INPUTS:
Pressure
Area of hole
Density of liquid
Viscosity of liquid
Soil characteristics
Type of hole (circular or crack)
Diameter or width of hole
A-5
-------
A. 1.2. Bernoulli Flow (into an air environment)
EQUATIONS:
t This calculation applies to leaks that are unimpeded by soil
resistance.
The loss rate can be calculated by a simple application of
Bernoulli's equation for a sharp-edged orifice.
t dQ/dt « .6A (2gz)-5
where
dQ/dt * leak rate
A » area of hole
g * acceleration due to gravity
z height of liquid
USER INPUTS:
»-
0 Hydraulic head
Area of hole
COMMENTS:
The flow through small holes under relatively low pressure is turbulent,
as can be verified by calculating a typical Reynold's number. The
Reynolds number for loss through a (1/32)" hole for methylene chloride
under 2 feet of head is 7500. This is in the upper end of the tran-
sition zone between turbulent and laminar flow. Use of Poiseville's
Equation for laminar flow gives a considerably higher flow rate, so tur-
bulent flow will be assumed.
A-6
-------
A.1.3. Conversion from hydraulic head to pressure
EQUATIONS:
Some equations In this Appendix use aP (pressure drop). Others call for
z (hydraulic head). These two quantities are related by the following
formula:
AP » P(g/gc)z
where
p « density of fluid
g * acceleration due to gravity
g. » a conversion factor which 1s 1 1n metric units and 32.17 1n
c English units
USER INPUTS:
Hydraulic head or pressure drop
Density of fluid
A-7
-------
A.1.4. Fluid velocity In pipes
EQUATIONS:
For flowing fluid 1n a pipe, the following energy balance equation
applies (Source: M. S. Peters and K. 0. Tiwnerhaus, Plant Design
and Economics for Chemical Engineers. 3rd Ed (1980) pp. 509-515):
P9c 'Sc
where
P! « Inlet pressure
p density
g acceleration of gravity
gc » a conversion factor (1 In metric units, 32.17 1n English unit*
Vj * Inlet velocity
z\ Inlet elevation above arbitrary base
?2 " outlet pressure
\/2 " outlet velocity
Z2 * outlet elevation
hs » hydraulic head added by the pump
hf « frlctlonal losses (measured as loss of hydraulic head)
F Is composed of 3 parts: losses at entrance to pipe, losses along
the pipe, losses at pipe exit.
The following frlctlonal loss equations apply to these losses
(Source: Peters and Tlmmerhaus)
sudden enlargement
Fe - (^"V2} a - 1 (turbulent flow)
A-8
-------
sudden contraction
KCV22
Fc * ?fr-
9
pipes and pipe elbows
F 2fv2(le + I)
gO
1
-05 for
diameter-changes
L
Le
0
friction factor »
5 x 10'3 for
turbulent flow
length of straight
pipe
effective length of
elbows
diameter of pipe
For systems open to the atmosphere Pj * P£ 1 atm
For large Inlet tanks, Vj » 0
Set 21 0
Then
V£^ " 2g (-Z2-hs-hf)
V22 also appears In hf, so
U:2..,,..hS>/l.l .1 . JO-«(L. * L) VI
\2g 2g 4g gD /
gO
A-9
-------
A.1.5. Pressure of flowing liquid in pipes
EQUATIONS:
Pressure In a pipe 1s given by solving the energy balance equation in
Section A.1.4 for P£. Hence:
P2"P1 '' *'2 ..... f
f)
where V * V£ » velocity of flow
The pressure difference between the pipe Interior and the pipe
exterior 1s given by A P " P2*pl« Thus:
P - -/ * 22 * "
At the midpoint of the pipe, the friction losses Include only losses
from the Inlet and half of the effective pipe length. Hence, at the
midpoint of the pipe:
hf . Fc * | F
- -05V2 + 5 x 10-3 y2(Le +
~ -
Thus, substituting this equation 1n the preceding equation gives
AP £.VVi «. ,05V2 + 5 « 10-3 (L. * L)Y» +
+ gh\
.P --
A-10
-------
A.1.6. Fill/discharge time
EQUATIONS:
The length of time necessary for tank filling OP discharge can be calcu-
lated from the fluid velocity in the pipe, the pipe diameter, and the
volume to be transferred:
T « Q/ irr2v
where
Q transfer volume
r « Inside radius of pipe
v * velocity of flow
USER INPUTS:
t Transfer volume
Velocity of flow (from Section A.1.4)
t Pipe diameter (interior)
A-ll
-------
A.2 FAILURE DATA SHEETS
The following data sheets discuss the probabilities and release volumes for
Individual fault tree events.
The events discussed In these data sheets generally represent Individual fault
tree events, but there 1s not always a one-to-one correspondence. When it
facilitates discussion, we have sometimes combined similar events. In addition,
we have occasionally divided our discussion of single fault-tree events in order
to focus more clearly on the differences between various tank designs or operat-
ing conditions. These changes in the fault tree event classifications have been
clearly labeled.
For these reasons, it is not appropriate to read one data sheet without also
becoming familiar with the data sheets for similar events. Important caveats may
apply to a whole series of events, but in order to focus on the differences among
related events, all similarities are not necessarily re-stated. Liberal cross-
references have been supplied, but they are not a substitute for a careful
reading of the related data sheets.
The labels for the events described in these data sheets may be confusing to the
first-time reader. Therefore, the following guide may be useful:
Label Prefix Category of Event Depicted
Txxxx Tank defects (corrosion, rupture, etc.)
Blx Piping, flanges, and gaskets
B2x - B6X Secondary-containment devices
Allllx Spills during discharge
A21x Release routes for overflows (corroded .vent
pipe, corroded flanges, open-topped tank,
etc.)
ANUCAT (I, x) Catastrophic events (flood, fire, etc.)
LIFDEF (I, x) Lifetime defects (Improper Installation,
damage during Installation, etc.)
A-12
-------
xxxxxx Miscellaneous events (including overfill,
operator error, alarm failure, level gauge
failure, etc.) These are listed in alpha-
betical order.
It may also be useful to sunmarize a few Items of commonly-occurring notation:
Notation Explanation.
p Binomial probability.
N(x,y) Normal distribution with mean of x, standard
deviation of y.
pN(xfy) Conditional normal distribution, p is the
binomial probability that failure occurs; -
NXx.y) Is the distribution of times to failure
given that failure does occur.
FNU(x.y) A uniform distribution between x and y.
Maximum y Unless a minimum Is also specified, this
;. means FNU(O.y).
B(a,b,c) Beta distribution with minimum value a,
mode b, and maximum c.
These explanations should make the event sheets more accessible to the general
reader.
A-13
1
-------
HAZARDOUS WASTE TANKS
LABEL: T1121
FAILURE: Localized exterior corrosion
SOURCES: Petroleum Association for Conservation of the Canadian Environment,
"Underground Tank Systems: Review of State of the Art and
Guidelines," PACE Report No. 82-3, Ottawa (1983).
CALCULATIONS:
t The PACE data are presented and analyzed in Appendix C.
0 The results of that analysis are cumulative time-to-failure distribu-
tions for each of 3 categories of Soil Aggressiveness Values (SAV's).
(SAV's are discussed in Appendix C).
Cumulative Probability of Failure
Tank
Low SAV
(0-6)
0
0
6.3
24.0
48.3
69.9
Medium SAV
(7-12)
0
11.1
29.1
54.0
67. 3,,
76.62
High SAV
« 13)
0
26.7
49.9
76.5
79.9
83.3
f'1 "'
9 "
14
19
24i
301
* This age has been chosen as representative of the >. 25 bracket.
2 A probability for this category could not be calculated from the raw
data. In order to complete the table, this number was obtained by
averaging 69.9 and 83.6. It appears to be a reasonable value.
PROBABILITY DISTRIBUTION: Empirical
PROBABILITY: See above table.
DATE OF INITIAL RELEASE:
SAV must first be calculated as shown in Appendix C. This will require
the following inputs:
- Soil resistivity (ohm-cm)
Soil pH
Soil moisture
A-14
-------
- Presence or absence of sulfides
- Differential characteristics
The failure date may now be sampled from the probability table.
A corrosion rate may be obtained by dividing the resulting
failure date by the average tank wall thickness for the tanks
in the PACE survey. This thickness is assumed to be .25
Inches.
The date of failure will be the date at which the sum of accu-
mulated local external corrosion plus general Internal corro-
sion reaches the initial wall thickness of the tank in
question. (General Internal corrosion is calculated under
event T1125). General exterior corrosion does not enter Into
this analysis, for we assume that It 1s already Included in the
corrosion rate for localized exterior corrosion. In addition,
localized Interior corrosion is not Included in this analysis,
for it is unlikely that localized Internal and external corro-
sion will occur at the same point on the tank wall. Thus, it is
unlikely that pits from these two causes will "meet 1n the
middle."
MATERIALS AND CONFIGURATION VARIATIONS:
The pACE-derlved probability distribution is assumed to apply only to
unprotected steel tanks.
Material changes, use of cathodic protection, and use of coatings will
alter corrosion probabilities and hence dates.of failure. These effects
are summarized 1n the following table:
Tank design Effect on probability of failure
Carbon steel (underground, PACE baseline
unprotected)
Carbon steel (underground, Delays onset of corrosion by
coated) N(7,3) years, then corrosion
becomes a certainty. Time-to-
failure distribution 1s given
below.
Carbon steel (above-ground, Corrodes like underground tank
unprotected) with 5X as large a surface area,
located 1n a low-SAV soil.
Carbon steel (above-ground, Delays onset of corrosion by
coated) N(9,3). Then it corrodes like a
low-SAV, underground tank with a
failed coating, and 5X as large a
surface area.
A-15
-------
Carbon steel (underground, PACE baseline after mN(10,5) years
impressed current cathodic
protection)
Carbon steel (underground, Delays onset of corrosion by
Impressed current cathodic max CmN(10,5), N(7,3)]. Then it
protection, coated) corrodes like a similar coated
tank whose coating has failed.
Stainless steel Corrosion rate of 25% of that
applicable to steel.
Fiberglass No corrosion
Concrete Gradually disintegrates due
to chemical attack, but this
effect has been Included with
ruptures.
Sources and Explanations:
Carbon steel (underground, unprotected); PACE
Carbon steel (underground, coated): Corrosion will not begin until the
coating has failed. The coating 1s assumed to fail according to a.nor-
mal distribution, N(7,3). The 7-year mean was obtained from A.H.
Roebuck and 6.H. Brevoort, "Coating Work Costs and Estimating,"
Materials Performance, 22(1): 43-47 (Jan. 1983). Standard deviations
were estimated by doubling the reported variations between average
coating lives 1n differing environments. In addition, the standard
deviation was Increased. 50X to account for variations 1n sub-grade
drainage and the possibility of scratches during Installation. The
coating used was 2-coat coal-tar epoxy with an SP10 (near white blast)
surface preparation. The mean age to failure Is that reported for
coatings In a freshwater environment.
Once the coating has failed, there Is a near certainty of localized
corrosion at the points of failure, for those points present excellent
point anodes. The time-to-failure distribution must therefore be
adjusted to reflect the certainty of a point anode's existence. We do
this by dividing each probability in the PACE baseline distributions by
the corresponding year-30 failure probabilities. The following time-to-
failure distributions therefore apply following coating failures:
Tank Low Medium High
Age SAV(X) SAV(X) SAV(X)
40 00
9 0 14.5 32.0
14 9.0 38.0 59.9
19 34.3 70.9 91.8
24 69.1 87.9 95.9
30 100.0 100.0 100.0
A-16
-------
Carbon steel (abovegrpund. uncoated). We assume .that above-ground tanks
only experience localized corrosion at their saams and their points of
contact with cradles. These areas account for approximately 5% of their
surfaces. We assume that air 1s about as corrosive as a low-SAV soil.
Thus, these tanks corrode like low-SAV, underground tanks of 5% as large
a surface area. The effect of area on corrosion rates is discussed
below under "variations."
Carbon steel (aboveground. coated). The coating lifetime was based
on Roebuck and Brevoort's values for exterior coatings In a moderate
atmosphere. The result Is a coating lifetime at N(9,3). Once the
coating has failed, the tanks will have a 100% chance of developing
point anodes (at the sites of Initial coating failure), and will
corrode like underground, coated tanks of 5X the surface area, in a
low-SAV environment.
Carbon steel (underground. Impressed current cathodic protection).
We chose impressed current as the preferred means of cathodic protection
on the recommendation of the National Association of Corrosion Engineers
(NACE) (Houston, phone conversation). This recommendation was
given because such a system can be repaired without excavation.
Furthermore, a crude .check of Its functioning can be done merely by
observing the ammeter.
The components of such a system are crucial. The wiring Is not subject
to protection, so 1t Is failure-prone. The rectifier Is also subject
to failure, especially 1f much energy 1s dissipated. In addition,
catholic protection can fall if the local electrical environment
changes (e.g. Interference with another protected system across the
street, or if someone buries some unprotected metal nearby).
Impressed-current cathodic protection therefore requires an Inexpensive
check of current distribution every 2-3 months (6 months maximum).
This can be done by a contractor equipped with a hand-held meter, and
takes only 2-3 minutes (plus travel).
With regular maintenance, NACE says such a cathodic protection system
should last for a long time. No failure data appears to be available,
however, and NACE says failures are common among poorly-maintained
systems. We therefore assume that cathodic protection systems fail with
a distribution of (m)N(10,5), where m 1s a random number drawn from (1,
3) Indicating the stochastic quality of the maintenance effort. (1
means no maintenance). Source: BEJ following phone conversations with
NACE.
As long as cathodic protection 1s functioning, corrosion will be negli-
gible.
t
If cathodic protection falls and 1s repaired after a substantial
Interval, It may be too late to restore complete protection. Cathodic
protection does not function well for creviced surfaces (or pits), for
it Is difficult to get an adequate charge density. (Source: NACE/API,
A-17
-------
Corrosion of Oil-and Gas-Well Equipment (1958), p. 71.) Since belated
repair will be insufficient to halt established localized corrosion,
we have not modeled belated repairs of the cathodic protection system.
Ordinary maintenance, however, is already included in our time-to-
failure distriution.
Carbon steel (underground, impressed current cathodic protection.
coated). If cathodic protection and a coating are both used, both the
cathodic protection system and the coating must fail before corrosion
will occur. Once failure occurs, point anodes will exist at the points
of coating failure.
Stainless steel. Stainless steel corrodes four times more slowly than
unprotected carbon steel. Source: Peters & Tlmmerhaus, Plant Design
and Economics for Chemical Engineers. 3rd. Ed., (1980), p. 574.
Concrete. The disintegration of concrete leads to cracking, not to
localized corrosion holes. We will discuss the cracking of concrete
tanks under ruptures, event T1124, below.
VOLUME OF RELEASE:
Tank corrosion represents a growing leak. Eventually the leak rate will
reach a detectable level.
ASSUMPTIONS:
Corrosion holes start small. Assume Initial hole size follows a beta
distribution, with a minimum size of (1/64)", a maximum of (1/4)", and a
mode of (1/32)".
Corrosion holes grow with time. We use the corrosion rate, r, calcu-
lated earlier, as the base rate at which corrosion holes grow in radius.
In addition, once exterior corrosion holes have perforated the tank,
they will also grow because of generalized Interior corrosion (see event
T1125). We therefore add this corrosion rate to r 1n order to.determine
the total rate at which the hole grows In radius.
§ Corrosion holes may occur anywhere on the tank. Depending on such fac-
tors as water table depth, they may be more likely to occur on the bot-
tom of the tank. We have decided to assume that the average corrosion
hole occurs on the tank bottom. This estimate could be refined to
account for fluctuations In fluid levels and the fact that leak rates
are not linearly dependent on hydraulic head. Such refinements greatly
complicate the model, however, without significantly altering the leak
rates. Hence, we use simple "average" hydraulic heads throughout the
model, rather than Integrating loss rates over the cyclical fluctuations
In fluid depths.
r
For storage tanks, we assume that the average fluid depth Is 50% of the
tank height. For treatment tanks we assume that the average fluid
depth 1s SOX of the tank height.
A-18
-------
For underground tanks, the leak rate is calculated according to the
underground leak rate formula. (Section A.1.1). For;.above-ground tanks
or tanks in vaults, the Bernoulli equation (Section A!1.2) applies.
VARIATIONS:
Tank capacity. Localized external corrosion Is an electrochemical pro-
cess which will only occur 1f the tank surface experiences a non-
uniformity which can function as a locale for electrochemical attack.
Such non-uniformities are referred to as "point anodes" and may consist
of scratches in the tank wall, variations In local soil conditions, or
stones or cinders in contact with the tank wall.
Logically the probability that point anodes will be present 1s a func-
tion of tank surface area: the larger the tank, the higher the probabi-
lity that a point anode exists. Similarly, the larger the tank surface
area, the more likely It 1s that there Is an unusually active point
anode somewhere on the tank surface.
The easiest way to account for this factor 1s to adjust .the corrosion
rate r to account for the surface area of the tank. According to a pipe
corrosion model by Rossum (see event 813 for a full discussion), the
corrosion rate for the deepest pit is proportional to the .16 power of
the surface area. Therefore, variations in tank size can be accounted
for by adjusting r according to the following formula:
r r0 (A/A0)-16
where
r » the adjusted corrosion rate
r0 the corrosion rate obtained from the PACE data
A » the surface area of the tank under consideration and
A0'« the average area of the tanks Included 1n the PACE study
These tanks were all service station tanks, many of which were installed
prior to 1970; they probably averaged 5000 gallons in capacity. The
surface area of such tanks 1s approximately 440 square feet.
This area adjustment factor applies to coated tanks and stainless steel
tanks as well as to carbon steel tanks, for in all cases, an Increase in
tank surface area Increases the probability that there Is an unusually
active pit.
Stray currents can accelerate corrosion. Approximately 10X of tanks are
subjected to stray currents due to nearby electrical equipment or
.electrical rail lines. (Source: Warren Rogers, Warren Rogers Associates,
personal communication.) Stray currents approximately double external
corrosion rates. (Source: Warren Rogers, Warren Rogers Associates, per-
sonal communication.) In order to allow greater variability 1n this
effect, we have assumed that stray currents multiply the corrosion rate
A-19
-------
r by a stochastic factor of x. x is distributed according to a beta
distribution with a minimum of 1, a mode of 2, and a"maximum of 4.
If the Teak is detected, the tank will be repaired or replaced, and the
aging process will start over again, using the same parameters as those
used for the original tank.
Cathodic protection will have no effect on hole growth rates; once the
cathodic-protection system has failed, the tank will corrode in the same ma
as an unprotected tank.
Coatings will also have no effect on hole growth rates.
Double-walled tanks; The 2 walls of such tanks are modeled separately.
Exterior corrosion attacks the outside of the outside wall. Interior
corrosion attacks the inside of the inside wall. The interstitial space
is not subject to corrosive attack. The interstitial alarm is modeled
as a leak detector similar to that used in a vault. See event MOALARM,
below.
In-ground tanks or above-ground tanks on-grade. Both the above- and
below-ground sections of such tanks are subject to corrosion. Because
exterior corrosion is more rapid below-ground than above-ground,
however* we have only modeled exterior corrosion of the below-ground
segment. Similarly, since interior corrosion is most important on the
bottom of the tank, we have only modeled interior corrosion of the tank
bottom.
USER INPUTS:
Tank surface area
Corrosion protection methods employed
Tank material
Soil characteristics
Rate of generalized interior corrosion (from event T1125)
Tank wall thickness
Tank location (above-, below-, or in-ground)
Is the tank double-walled?
A-20
-------
HAZARDOUS WASTE TANKS
LABEL: T1123
FAILURE: Localized interior corrosion
SOURCES:
API Tank and Pipe Leak Survey (referenced in SCS Engineers, "Assessment
of the Technical, Environmental, and Safety Aspects of Storage of
Hazardous Waste in Underground Tanks," Draft, 1983).
PACE corrosion study, cited under event T1121.
Best engineering judgment.
ASSUMPTIONS:
Interior corrosion leaks are the same size and grow at the same rate as
exterior corrosion holes.
» *
Volumes are larger, however, for the hole usually occurs at the bottom
of the tank and therefore has more hydraulic head than that assumed for
exterior corrosion holes.
CALCULATIONS:
t Using'-SCS Engineers' analysis of API survey data, we can construct the
following tabulation of underground tank leaks:
Source of Leak t of Failures % of Total Scaled to 77%
- TANKS
localized exterior 988 61.6 77.0*
corrosion
776 + 212
interior corrosion 194 12.1 15.Ot
other 55 3.4 4.3*
loose fitting 9 0.6 0.81
breakage 17 1.1 1.4*
A-21
-------
- PIPES & ANCILLARY
pipe corrosion
loose pipe fitting
flex connector
breakage
other
Explanation;
353
64
38
43
54
TSUI
22.0
4.0
2.4
2.7
3.4
27.5%
5. OX'
3. OX
3.4%
4.3X
- We obtained the last column by scaling the first entry to 77%. We
then multiplied the remaining entries in this column by the same
scaling factor. This scaling factor 1s based on Warren Rogers'
assertion that 77% of all tanks exhibit significant localized
exterior corrosion. This percentage 1s also consistent with the PACE
data, which show that 76.6% of medium-SAV tanks have leaked by year 30.
Due to the possibility that a tank system may experience more than one*
type of failure, the numbers 1n this column do not sum to 100%.
Line 1 Includes 212 tanks: that were not reported as falling by loca-
lized exterior corrosion. We included these tanks because we assumed
that localized exterior corrosion 1s a slower process than the other
four forms of tank failure. Thus, we assume that 77% of the 275 tanks
that failed by these other mechanisms were undergoing undiscovered
localized exterior corrosion at the time they failed. We make no
similar adjustment to account for tanks that might have been
undergoing localized exterior corrosion at the time of piping
failures, because we assume that tank and piping systems are repaired
independently. Thus, the API survey data already account for systems
which suffer both tank and piping failures.
We can read the following results off the above table:
Failure Mechanism
exterior tank corrosion
Interior tank corrosion
seam leaks (the most
likely form of "other")
tank rupture (breakage)
Probability Conditional Distribution
0.77
0.15
0.043
0.014
See event T1121
See below
See event T1124
See event T1124
pipe corrosion
0.28
See event B13
A-22
-------
pipe rupture (breakage) 0.034 See event 811
loose fittings 0.05 See, event B121
PROBABILITY DISTRIBUTION: Conditional normal
PROBABILITY:
,15c where c is a material-dependent coefficient given by the following
table:
Tank material c
steel 1
stainless steel 1
fiberglass 0
concrete 0
Coefficient c merely Indicates that fiberglass does not corrode and that
. Instead of treating the aging of concrete as corrosion, we will treat it
as contributing to rupture under event T1124, below.
TIME-TO-FAILURE:
We. estimate that localized Interior corrosion should occur more quickly
than;-localized exterior corrosion. Since the PACE data show that the
conditional mean date of localized exterior corrosion failure for tanks
in medium-SAV soils Is approximately 16 years, we assume that localized
interior corrosion proceeds twice as fast. Thus, for quarter-inch
steel, we have assigned a localized interior corrosion time-to-failure
distribution of N(8,5). In choosing this distribution, we have used a
relatively large standard deviation In order to cover unknown factors,
Including variations 1n yearly flow rates.
Let x be the failure date sampled from N(8,5). Since the typical ser-
vice station tank is .25" in thickness, the corrosion rate (r) for those
tanks which experience localized Interior corrosion 1s given by:
r « .25
x
In deriving this equation, we have .assumed that localized interior
corrosion was the primary corrosion mechanism for those tanks listed as
falling by Interior corrosion.
» We can now compute the failure date by using the combination of loca-
lized interior corrosion and generalized exterior corrosion to keep
track of remaining wall-thickness. Because generalized exterior corro-
sion 1s usually much slower than localized Interior corrosion it usually
A-23
-------
has only a minor effect on the date of failure. For this reason, we did
not account for it in calculating our corrosion rate r, above. i
Generalized exterior corrosion is only important in those cases when it
occurs unusually quickly.
VOLUME:
Interior corrosion holes begin with the same size distribution as
exterior corrosion holes, and grow 1n radius at the rate r calculated
above.
In addition, corrosion holes will also grow due to the effects of
generalized exterior corrosion. See event T1126, below.
Interior corrosion leaks tend to occur at the bottom of the tank.
For storage tanks, we assume that the tank 1s on average 50* full; for
treatment tanks, we assume that 1t Is 80% full.
For underground tanks, we calculated leak rates according to the
underground leak rate equation (Section A.1.1). For above-ground tanks,
tanks In vaults, or the above-ground sections of In-ground or on-grade
tanks, the Bernoulli equation (Section A.1.2) applies.
VARIATIONS:
Coated, steel. We assume that the coating 1s 3-coat epoxy. Then the
distribution of coating failure times Is N(7,3). Source: Roebuck and
Brevoort (1983). Corrosion will not begin until the coating fails. We
assume that the failure of Interior coatings does not Influence either
the probability of corrosion or the corrosion rate for Interior corro-
sion. Thus tanks with failed Interior coatings experience Interior
corrosion in the same manner as new, uncoated tanks.
Cathodic protection. Cathodlc protection has the same effects and the
same time-to-failure distribution for Interior corrosion as 1t has for
exterior corrosion. We have assumed that when the exterior cathodic
protection system falls, the Interior cathodic protection system also
falls.
Double-walled tanks. See event T1121.
Tank capacity. Unlike localized external corrosion, localized Interior
corrosion 1s not likely to be-correlated with tank capacity. Localized
Internal corrosion 1s most likely to occur beneath the fill tube, and '
the affected area Is likely to be Influenced by factors such as method
.of filling or pumping rate, rather than tank size.
Stainless steel tanks. Stainless steel tanks corrode at 25X of the rate
applicable to carbon steel.
A-24
-------
USER INPUTS:
Rate of generalized exterior corrosion
Tank material
Tank wall thickness
Type of corrosion protection system
A-25
-------
HAZARDOUS WASTE TANKS
LABEL: T1124
FAILURE: Tank Rupture
ASSUMPTIONS:
Tank rupture can occur from Inadequate component strength, settling, or
exterior force.
Tank rupture Includes both large cracks and seam leaks.
SOURCES: See T1123 (Breakage + "other")
CALCULATIONS:
t SCS/API data (see event T1123) Indicates a tank rupture rate of 1.4%.
SCS/API data Indicates a seam leak rate of 4.3%.
SCS/API data Indicates an average tank age of 10.8 years.
Assuming ruptures are approximately uniformly distributed over the 10.8
years of tank lifetime, the average annual probability of ruptures is
5.7X/10.8 yrs - 5.3 x 10'3/yr.
PROBABILITY: 5.3 x 10'3/yr
PROBABILITY DISTRIBUTION: Binomial
VOLUME:
t Ruptures may take 2 forms: large cracks and seam leaks. The data from
event B13 Indicate that 75X of ruptures are seam leaks, while 25% are
large cracks.
Seam leaks are assumed to range 1n length from 0 to 5 ft, and 1n width
from 0 to (1/16)".
4 Large cracks are assumed to range 1n length from 3" to 36" and in width
from 0 to 3".
A-26
-------
VARIATIONS:
Aboveground. in-qroundr or vaulted tanks. Above-grade ruptures occur
because of different types of accidents than occur for below-ground
tanks. Vehicle collisions, collisions with fork-lifts, and inadequate
component strength are major causes of ruptures. We assume that the
probability of rupture is still 5.3 x 10~3. Volume of loss will be
limited by the amount of fluid in the tank.
Fiberglass tanks. Fiberglass tanks are approximately twice as likely to
rupture as steel. The probability of rupture is therefore 1 x 10'2/yr.
Source: Brown Minneapolis Tank Co., personal communication 4/30/85.
Brown Tank Company said that 2% of fiberglass tanks collapse in the
first couple of years. We obtained our scaling factor of 2 by comparing
that percentage to the .5X annual probability of steel tank rupture.
Double-walled tanks. We assume that since the Inner wall of a double-
walled tank 1s subject to fewer stresses than 1s the wall of a single-
walled tank, it 1s only SOX as likely to rupture. The outer wall of
such a tank, though, 1s likely to be just as vulnerable to rupture as
the wall of a single-walled tank. We assume, however, that there is
only a 50% chance-that a rupture of the outer wall will also breach the-
1nner wall. Since ruptures of the Inner wall do not involve large
exterior forces (such as vehicle collisions or settling of the
backfill), we assume that the number of times that Interior ruptures
also breach the outer wall 1s small enough to be Ignored.
When a rupture of a double-walled tank occurs, Infiltrating soil
moisture or leaking hazardous waste will trigger the Interstitial alarm.
There 1s a 10X probability that the alarm will fall to function. See
MOALARM, below.
A-27
-------
HAZARDOUS WASTE TANKS
LABEL: T1125
FAILURE: Thickness loss due to generalized interior corrosion.
ASSUMPTIONS:
This event will cause a loss when remaining wall-thickness reaches zero.
This event will also reduce the time it takes for localized exterior
corrosion holes (if any) to penetrate the tank.
SOURCES: Perry and Chi 1 ton (1973).
CALCULATIONS:
» .
According to Perry and Chi 1 ton, generalized interior corrosion rates of
.002"/yr to .02Vyr are reasonable. We expect that the lower corrosion
rates are the most likely.
VOLUME: See event T1123.
tr
PROBABILITY DISTRIBUTION: Uniform
PROBABILITY:
Cumulative Probability Corrosion Rate
0.00 to 0.65 0.002"/yr
0.65 to 0.90 FNU(.002,.OD
0.90 to 1.00 FNU(.01,.02)
VARIATIONS:
Coatings. See event T1123.
Cathodic protection. See event T1123.
A-28
-------
HAZARDOUS WASTE TANKS
LABEL: T1127
FAILURE: Thickness loss due to generalized exterior corrosion.
ASSUMPTIONS:
This event will reduce the time It takes for interior corrosion holes
(if any) to penetrate the tank, as well as Increasing the rate of
Interior hole growth once a hole occurs.
SOURCES:
t "Prediction of Leaks 1n Unprotected Storage Tanks," Warren Rogers
Associates, Inc., Included in correspondence package from Betsy Tarn,
EPA, to Chris Lough, PRA. . _ .
«
W. H. A1lor, Atmospheric Corrosion (John Wiley and Sons: New York)
1982, p. 33.
CALCULATIONS:
t According to Ailor, generalized corrosion in an air environment occurs
at approximately 1.4 m1ls/yr.
According to Warren Rogers Associates, generalized corrosion 1s unlikely
to cause failure (for a quarter-inch tank) within the 40-year time-
horizon of their data.
t Generalized external corrosion in a soil environment 1s unlikely to
occur more slowly than In air. 1.4 mils/yr 1s therefore a minimum value
for the generalized external corrosion rate.
a 5 mils per year Is a suitable upper bound for the generalized corrosion
rate. This choice correspondes to a 50-year minimum time-to-failure for
quarter-Inch steel tanks in non-extreme environments.
t SAV will have some effect on generalized corrosion. Assuming the effect
Is roughly linear, and noting that according to the Canadian data the
average SAV Is 10, we account for SAV by a multiplying the generalized
exterior corrosion rate by SAV/10. Under no circumstances, however, do
. we use a generalized corrosion rate of less than 1.4 m1ls/yr.
A-29
-------
CORROSION RATE:
where
rgen, ext * tne generalized external corrosion rate.
VARIATIONS:
Direct currents win have the same effect on generalized exterior corro-
sion as they have on localized exterior corrosion.
0 Stainless steel tanks will corrode at one-fourth the rate applicable to
steel tanks. See event T1121.
Above-ground tanks. The atmospheric corrosion rate of 1.4 m1ls/yr
applies (or one-fourth of that for stainless steel tanks).
Cathodlc protection or coatings. Significant generalized exterior
corrosion Mill not begin until corrosion protection fails. Since
generalized exterior corrosion. 1s not Influenced by the presence of
point anodes, we assume that once the coating has failed, generalized
exterior corrosion proceeds in the same manner for coated and untreated
tanks.
A-30
-------
HAZARDOUS WASTE TANKS
LABEL: T1128
FAILURE: Tank falls due to generalized corrosion.
ASSUMPTIONS:
tank corrosion may occur generally, due to a combination of rapid
generalized exterior and generalized Interior corrosion.
CALCULATION:
Tank wall-thickness must be computed even for tanks without localized
corrosion. In a few cases, this will produce corrosion failure.
VOLUME:
We determine the sizes and the growth rates for these holes in the same
manner as we determine the sizes and growth rates for localized corro-
sion holes.
A-31
-------
HAZARDOUS WASTE TANKS
LABEL: Bll
FAILURE: Pipe Rupture
SOURCES: SCS Engineers' analysis of API data (see event T1123).
CALCULATIONS:
According to the API data, 3.4* of service station tanks experienced
pipe rupture. (See the table entry labeled "breakage" 1n event T1123)'.
These tanks had been 1n use an average of 10.8 years.
The annual rupture probability 1s therefore 3.4X/10.8 « 3.15 x 10'3/yr.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 3.2 x 10"3/yr
VOLUME:
We assume minimum and maximum crack sizes of 1" x .1" and 10" x (1/4)",
respectively.
We assume that rupture dimensions are uniformly distributed over the
range of possible values. Therefore:
width - FNU(.1,.25) Inches
length FNUU.10) Inches
Vie determine hydraulic heads by applying the pipe-pressure formula
(Section A.1.5) to the geometry of the system design under con-
sideration. We assume that the rupture occurs at the midpoint of the
pipe.
Loss v<111 only occur during filling or discharging. We calculate
fill/discharge duration according to Section A.1.6.
A-32
-------
VARIATIONS:
Pipe length and system geometry will vary with system design.
Above-ground pipes or pipes in vaults. We assume that the rupture pro-
bability for above-ground pipes is the same as that for underground
pipes. We also assume that the hole size distributions are the same for
both types of pipes.
t Fiberglass pipes. The probability of rupture is doubled. See event
T1124.
Multiple pipes. We evaluate each pipe separately. We assume that the
rupture of one pipe has no influence on the probability that other
pipes will also rupture.
USER INPUTS:
System geometry
Pipe diameter
Pipe length
A-33
-------
HAZARDOUS WASTE TANKS
LABEL: 8121
FAILURE: Welded Flange Leak
SOURCES:
t Nuclear Regulatory Commission, Reactor Safety Study (1975).
t Henley and Kumamota (1981).
. SCS Engineers (1984).
ASSUMPTIONS:
A flange leak can only occur when fluid is in contact with the.flange. -
If the flange is located at the top of the tank, fluid only contacts it
during filling, discharging, or overflow. If the flange is at the bot-
tom of the tank, fluid will always be in contact with it. The location
of the flange is dependent on system design.
CALCULATIONSr
3 x 10'7/hr » welded flange leak rate (Reactor Safety Study).
3 x 10-7/hr - .003/yr
10'4 to 10-} per 10,000 hrs - welded flange leak rateJHenley & Kumamoto),
. lO'4 to 10'1 per 10,000 hrs - 8.8 x 10'5 to 8.8 x 10'2/yr. The
geometric mean of this range of values * .0028/yr.
API service station data gives a loose fitting rate of 5.OX (SCS
Engineers). The average system age is 10.8 yrs. The failure rate is
therefore 5.OX/10.8 - 4.6 x 10'3/yr.
We therefore use a conservative flange leak probability of 5 x 10~3/yr.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 5 x 10'3/yr
A-34
-------
VOLUME:
The flange is at the point of attachment of the pipe to the tank.
The pressure drop due to friction losses therefore includes the entire
length of the pipe.
The elevation change can be calculated from the system geometry.
t We assume a 50% weld failure as upper bound, 0 as lower bound. I.e., we
assume that the length of the breach is distributed according to
FNU(0, 502 of flange circumference). We assume that the width of the
'breach is given by FNU(0,l/8) inches.
0 We assume a 300 to 400 Ib flange. The exterior flange diameter is 6.5".
(Source: Perry and Chi 1 ton, Chemical Engineers' Handbook, 5th Ed.
(1973), pp. 6-66 and 6-67, using a 2"-diameter pipe). The maximum weld
failure is therefore 10" long.
0 For underground flanges, we calculate the leak rate by using these
values in the underground leak rate formula (Section A.1.1). For above--
ground flanges (or flanges in,an air environment such as a vault) we use
the Bernoulli equation (Section A.1.2).
t We calculate the loss per day by multiplying the leak rate by the length
of time that the fluid is in contact with the flange.
* *.
VARIATIONS: "
Flange location* A flange may be at the top of the tank or near the
bottom of the tank. Its location will influence the hydraulic head of
the fluid and will determine the fraction of the time that the flange is
in contact with.fluid. The flange location will be determined by the
choice of system design.
Multiple flanges. We evaluate each flange independently. Thus, we
assume that the failure of one flange has no influence on the probabi-
lity that other flanges will leak.
USER INPUTS:
System design
A-35
-------
HAZARDOUS WASTE TANKS
LABEL: B122
FAILURE: Gasket Falls
ASSUMPTIONS:
We model gasket deterioration as an erosion-like process. We assume
^that under Ideal conditions, 1t will occur relatively slowly, at a rate
of less than 50 mils per year.
SOURCE:
Best engineering judgment based on Nuclear Regulatory Commission,
Reactor Safety Study (1975), and EPA, Case Studies 1-23 (1984) (case
study #4).
PROBABILITY DISTRIBUTION: Empirical
PROBABILITIES (baseline):
Cumulative Probability
0.00 to 0.77
0.77 to 0.83
0.83 to 1.00
Deterioration Rate (m1ls/yr)
FNU(0,12,5)
FNU(12.5,25)
FNU(25,50)
DATE OF FAILURE:
t For a 2" pipe, the standard gasket is a flat disk with an inner radius
of 1" and an outer radius of 2.0625". Since disintegration occurs from
the inside outward, 1.0625" of gasket must be dissolved or eroded before
the fluid can escape. A gasket for a 4" pipe Is slightly thicker
(1.4375"). (Source: Perry and Chllton, 1973).
VOLUME: See event B121 (welded flange leaks).
COMMENTS:
o Under Ideal conditions, we have assumed that a gasket will deteriorate
very slowly. Waste/gasket Interactions will speed up the deterioration
process, however. To model such interactions we use a multiplicative
A-36
-------
factor selected from FNU(1,20). Gasket failure can therefore occur as
early as year 2. This is consistent with the facts of EPA case study
number 4 (Biocraft site).
i-
The Reactor Safety Study quotes a failure rate for containment-quality
gaskets of 3 x 10-°/hr = 2.6%/yr. This may be too low for a conven-
tional hazardous waste tank system, but it serves as an approximate
check for the results of our gasket disintegration model.
A-37
-------
HAZARDOUS WASTE TANKS
LABEL: B13
FAILURE: Pipe Corrosion
SOURCES:
John R. Rossum, "Prediction of Pitting Rates in Ferrous Metals from
" Soil Parameters, Jour. AWWA, 1969, pp. 305-310.
PACE (see event T1121).
SCS Engineer's analysis of API survey data (See event T1123).
PROBABILITY DISTRIBUTION: Cumulative Empirical
CALCULATIONS (underground pipes):
Localized exterior corrosion
Rossum gives the following formula for maximum exterior pit depth for
burijed steel pipes:
p . l.Q6Knr(10-pH)t1nA.16
L P J
where
p the depth of the deepest pit (in mils)
Kn * 170 for soils of good aeration
» 222 for soils of fair aeration
» 355 for soils of poor aeration
pH » the soil pH (must be between 5 and 9)
n « 1/6 for soils of good aeration
1/3 for soils of fair aeration
» 1/2 for soils of poor aeration
the soil resistivity (ohm-cm)
t « the number of years since the pipe was burled
A the surface area of the pipe (in square feet)
A-38
-------
We have added a fourth, aeration category, moderate aeration. We
obtained «n for this category by averaging the values for good aeration
and fair aeration. We obtained the value of n by taking the geometric
mean of the values for good and fair aeration. Thus: '
KM = 196 for soils of moderate aeration
n
n » .236 for soils of moderate aeration
0 Multiple failures. Unlike our tank-corrosion model, which only predicts
the date of first failure, the Rossum pipe-corrosion model can be used
to determine the occurrence of subsequent corrosion holes. The number
*of leaks (L) at time t 1s given by:
n/.16
L - A
|-1.06Knl '« f
L~i~J - L
We make this formula stochastic by converting fractional values of L
Into probabilities. Thus, we have assumed that the first leak occurs
between L » 0 and L » 2. At L * 1, there Is a 50% chance of failure.
Similarly, the second leak occurs between the times when L « 1 and L »
3, the third leak between L » 2 and L » 4, etc. This means, for
example, that a value of L equal to 3.2 means that 2 leaks have occurred
with certainty; there Is a 60% chance that a 3rd leak has occurred, and
if the third leak has occurred, there Is a 10% chance that the 4th leak
has' also begua.
Note that the value of t 1s set to zero whenever detection/repair
occurs.
Caveat: The Rossum model applies only for soil pH between 5 and 9.
Outside that range, other corrosion mechanisms come Into play. These
are not Included in our model.
Generalized Interior corrosion. When fluid 1s 1n contact with the pipe,
we assume that generalized Interior corrosion occurs under the same pro-
bability distribution as applies to tanks. Because these are indepen-
dent events, however, we determine the corrosion rates separately for
each tank and each pipe. When the pipe Is in contact with air, we use a
corrosion rate of 1.4 mils per year (A1lor,. 1982). When we combine
corrosion from the fluid and corrosion from the air, we obtain a genera-
lized Interior corrosion of:
1.4(l-f) » f(r')
where
*
f * the fraction of time the pipe 1s 1n contact with the fluid
r' « the corrosion rate obtained for those times when fluid 1s
present
A-39
-------
Caveat: The Rossum model applies only for soil pH between 5 and 9.
Outside that range, other corrosion mechanisms come into play. These
are not included in our model.
Localized interior corrosion. Since our tank-corrosion model shows that
localized interior corrosion 1s 19* as likely as localized exterior
corrosion, we assume that the same ratio applies to pipes. We cannot
use this Information directly, however, because our localized exterior
pipe-corrosion model only applies 1f soil conditions are known. We
, therefore must use API service station data to determine the probability
that pipes undergo localized Interior corrosion. From this, we can then
obtain the probability of localized Interior pipe corrosion.
>.
According to the API data, 2B% of service station tanks experience pipe
corrosion. Since serv.ice station tanks generally have two pipes, we
convert this value to a per-pipe probability by assuming that corrosion
Is equally likely in either set of piping. Thus, letting "a" be the
probability of localized corrosion for a single pipe, 1t must be the
case that:
.28 * 1 - (l-a)E
or, a « .15. Thus, 1f localized Interior pipe corrosion 1s 19X as com-
mon as localized exterior pipe corrosion, and 1f other forms of corro-
sion failure are uncommon, then approximately 2.4X of the API survey
pipes;,failed by localized interior corrosion, while 12.6% failed by
localized exterior corrosion.
We assume that the time-to-failure distribution for localized interior.
corrosion is the same for pipes and tanks. Thus, the time-to-fallure
distribution for those pipes which exhibit localized Interior corrosion
Is N(8,5).
We adjust for area by multiplying both the probability of localized
corrosion and the corrosion rate by (A/10)*1*. See the discussion of
above-ground pipes, below, for an explanation of this area correction
factor.
Generalized exterior corrosion. Generalized exterior corrosion reduces
the time to pipe failure by localized Interior corrosion in the same way
that It reduces the time to Interior corrosion failure for tanks. We
assume that generalized exterior corrosion rates for pipes follow the
same probability distribution as applies to tanks. See event T1127.
Erosion. Because pipes carry moving fluids, they are subject to erosion
. by suspended solids 1n the waste. Erosion rates are dependent on the
concentration of suspended solids. Since our baseline corrosion values
are derived for pipes carrying a non-erosive fluid (gasoline), we assume
that they do not already account for erosion. We therefore add the
following erosion rates to both our localized Interior and generalized
Interior corrosion rates:
A-40
-------
Fraction of suspended Erosion
solids in the waste rate
_ (ppm) _ (mils/yr)'
<1,000 0
1,000-10,000 FNU(O.S)
>10,000 FNU(5,10)
VOLUME OF RELEASE:
- We assume that localized exterior corrosion holes follow the same size
distribution for pipes as they do for tanks. We calculate their growth
rates 1n the same manner as we calculate growth rates for exterior
corrosion holes In tanks.
Localized Interior corrosion holes also follow the same size distribu-
tion and grow in the same manner as do the corresponding holes in tanks.
If the pipe falls due -to generalized corrosion, the holes will be larger.
than they will be for localized corrosion. We assume that such holes
begin with the dimensions of 'a pipe rupture (see event BID, and double
In length and width -each year until they are detected. In order to
account for the physical limitations Imposed by the dimensions of the
pipe, we do not allow such corrosion holes to become more than 3" wide
or longer than the length of the pipe.
*
We calculate fluid velocity, transfer time, and pressure 1n the same
manner as we used for underground pipe ruptures. For both corrosion
holes and ruptures, we make the simplifying assumption that the leak
occurs at the midpoint of the pipe.
We use the underground leak rate equation (Section A. 1.1) to calculate
loss rates. Multiplying this leak rate by the fill or discharge time
(Section A. 1.6) then gives the total volume of loss per filling or
discharging.
ABOVE-GROUND PIPES:
The Rossum Model does not apply to above-ground pipes.
Localized exterior corrosion. We obtain the localized exterior corrosion
rate for above-ground pipes from the following conditional normal
distribution:
fN(16(10/.05A)16. 6.6(10/.05A)-16]
l ' J
where
A-41
-------
A = the surface area of the pipe in square feet. *
Interpretation. The first term of this formula, .1H.05A/10)-16 1|gthe
probability that localized exterior corrosion occurs. The (A/10)'
factor is the same area adjustment factor that we use for tanks, except
that for pipes, the baseline area 1s 10 square feet. This area
corresponds to 20 feet of 2" pipe or 10 feet of 4" pipe. These pipe
dimensions are appropriate for service station fill and discharge pipes,
respectively.
.11 coefficient In this expression 1s the baseline probability of
localized exterior corrosion. We derived this probability from a com-
bination of the PACE tank data and the API pipe-corrosion data.
According to the PACE data, 70X of low-SAV tanks fail by localized
exterior corrosion within 30 years, while 77% of medium-SAV tanks fail
by that mechanism 1n the same period. According to the API data, 12.6%
of service station pipes fall by localized exterior corrosion. If we
assume that this percentage applies to medium-SAV soils and that the
same proportionality factor applies to low- and medium-SAV pipes as
applies to low- and medium-SAV tanks, then (70/77) x 12. 6X of low-SAV
service station pipes fall by localized exterior corrosion. To two
significant figures, this percentage Is lit. If we assume that above-
ground pipes corrode like low-SAV underground pipes (an assumption s~imi--
lar to that which we made for above-ground tanks) then we can apply this
percentage to above-ground pipes.
The second part'of our localized exterior pipe-corrosion formula,
N(16(10/A)-16, 6.6(10/A)-16), gives the conditional date of failure for
pipes that experience localized Interior corrosion. The factor
(10/A)*16 reduces the time to failure according to the Inverse of the
area adjustment factor. Thus, when a corrosion rate 1s calculated from
the sampled time-to-failure, that corrosion rate will be higher than
baseline by a factor of (A/10)-16. .
Our baseline time-to-failure distribution in this expression 1s
N(16,6.6). We derived this distribution in the same manner as we
derived the lit baseline probability of failure. According to the API
survey data, service station pipes show the following distribution of
failure dates:
Year of Percentage of reported
failue pipe failures
0-1 2.1
2-5 6.4
6-10 32.8
11-15 33.1
16-20 18.1
21-25 5.0
26-30 2.3
30* 0.2
A-42
-------
We can approximate this distribution as N(12,5). Assuming that this
distribution applies to medium-SAV pipes, we can obtain"a low-SAV
time-to-failure distribution by comparing the PACE distributions for
medium- and low-SAV tanks. These distributions show conditional mean
dates of failure of 21 and 16 years, respectively. Adjusting both
parameters of the API distribution by 21/16 gives us a low.-SAV time-to-
failure distribution of N(16,6.6).
The final element of our above-ground localized corrosion formula 1s the
.05 factor used to reduce the effective surface area of the pipe. We
use this factor because we assume that like above-ground tanks, above-
ground pipes are only vulnerable to localized exterior corrosion at
their seams and points of support. These regions account for approxima-
tely 5X of their surface areas.
In our pipe-corrosion distribution, we have applied the area adjustment
factor to both the baseline probability and the baseline corrosion rate.
This 1s different from the way that we adjust for area 1n our tank
corrosion model, for in that model, we only applied our area adjustment
factor to corrosion rates. ,
We have modeled pipes and tanks differently because we believe that
three factors control the onset of localized corrosion. One of these 1s
the surface area of the burled metal. The others are the corrosivity of
the soil and the care with which the component 1s Installed. We have
assumed that surface area 1s an Important factor for localized corrosion
event's with low probabilities. In other words, we have assumed that the
probability of pipe corrosion 1s relatively low because a pipe is small
enough that It Is relatively unlikely to experience a point anode. For
tanks, however, the baseline probabilities of localized corrosion are on
the order of 70-85%. We assume therefore, that the principal factors
Influencing the onset of tank corrosion are the corrosivity of the soil
and the care with which the tank 1s Installed. Surface area will be
Important, but Its primary effect will be to determine the number of
point anodes and thus the depth of the deepest pit.
We have generalized the preceding discussion to obtain the following
rule of thumb: whenever the baseline probability of a corrosion event
1s less than SOX, we have assumed that component surface area will
Influence the probability of the onset of corrosion; whenever the base-
line corrosion probability 1s over SOX, we assume that the component 1s
already large enough that surface area has little Influence on probabi-
lity. Since all of our relevant baseline probabilltes are either
greater than 70X or less than 12X, we never had to elaborate this rule
of thumb by developing a model to deal with Intermediate cases. In
addition, since our area adjustment factor requires an area more than
SOOO times larger than baseline to increase a 12X probability to SOX, we
did not need to modify that factor to assure that our area adjustments
do not Increase the probability of failure to a value greater than SOX.
Finally, our model 1s Insensitive to any choice of cut-off probabilities
between 251 and 70X, so it Is unnecessary for us to be precise In our
determination of what value In that range Is the theoretically best
choice.
A-43
-------
Generalized exterior corrosion. Generalized exterior corrosion for
above-ground pipe segments occurs at the atmospheric corrosion rate of
1.4 mils/yr.
Localized interior corrosion and generalized interior corrosion. The
corrosion rates for these types of corrosion are the same for above- and
below-ground pipes.
Combinations of corrosion mechanisms. We take combinations of corrosion
mechanisms into account in the same manner for pipes as we used for
tanks. Thus, to determine the date of interior corrosion failure, we
"compute the remaining wall-thickness for the combination of localized
interior corrosion and generalized exterior corrosion. To determine the
date of exterior corrosion failure, we calculate the remaining wall-
thickness from the combination of localized exterior corrosion and
generalized Interior corrosion. To determine the date of generalized
Interior corrosion, we calculate the remaining wall-thickness from the
combination of the two generalized corrosion mechanisms. We do not com-
bine generalized exterior and localized exterior corrosion (or genera-
lized Interior and localized Interior) because we assume that our
localized exterior corrosion rates already Include both forms of
exterior corrosion.
Pipe thickness. Pipe thickness 1s directly Included as a parameter in
the Rossum localized exterior corrosion model. For our PACE- and
API-derived corrosion rate formulas, we use a baseline pipe thickness of
190 mils, which 1s the average of the thicknesses generally used for
2" and 4" pipe (Peters and Timmerhaus (1980)).
Hole sizes. We use the same hole-size distribution for above- and
below-ground pipes. We calculate hole growth rates in the same manner
for both types of pipes.
VARIATIONS (above- and below-ground):
Coated pipes. When a coating falls, there Is a 100X chance of point
anodes developing at the sites of coating failure. Thus, once the
coating falls, localized exterior corrosion begins with certainty.
Because the Rossum model does not take this factor Into account, we use
the following baseline time-to-failure distributions for low-, medium-,
and high-SAV pipes:
Piping t1me-to-fallure distribution
Soil type following coating failure
Low SAV NU6.6.6)
Medium SAV N(12,5)
High SAV N(9,4)
These distributions are derived from the PACE and API data sets. The
derivation process for high-SAV pipes is the same as that described
earlier for medium- and low-SAV pipes.
A-44
-------
These distributions apply to 190-mil pipes. We model different pipe
thicknesses using the same approach that we used to model variations in
tank wall-thicknesses.
Cathodic protection. We assumed that the entire tank facility uses a
cathodic protection system powered by a single power supply. Thus, when
that system fails and repairs are not undertaken within a reasonable
time, cathodic protection falls for both the tank and the pipes.
Cathodic protection failure Is discussed under events T1121. As long as
it 1s functioning, cathodic protection will prevent both Interior and
"exterior corrosion.
t Loss rates from above-ground pipes or below-ground pipes with secondary
containment. Leak rates in such circumstances will be controlled by the!
Bernoulli equation (see Section A.1.2).
0 Stray currents. If stray currents exist, they will affect pipes and
tanks similarly. See event T1121. Because stray currents are likely to
be equally severe for all elements of a tank system, we only sample the
stray current event once for .the entire facility.
Stainless steel. Stainless steel reduces all corrosion.rates by a fac-
tor of .25.
Fiberglass. Fiberglass pipes do not corrode.
USER INPUTS:
Pipe material
System design
Corrosion protection system
Pipe thickness
Surface area of pipe
Soil characteristics
A-45
-------
HAZARDOUS WASTE TANKS
LABEL: Alllll
FAILURE: Strainer drain left open after maintenance (or during pump-out for
storage or accumulation tanks).
ASSUMPTIONS:
i
We assume that the minimum operator response time is 15 seconds.
Because the operator is likely to be in the vicinity, we assume a maxi-
mum response time of 3 minutes. We therefore assume that the response
time is distributed FNUC.25,3) minutes.
Strainer maintenance occurs once per month.
SOURCES:
JRB Associates (1982)
Nuclear Regulatory Commission, Reactor Safety Study (1975)
'-
*
CALCULATIONS:
1.7 x 10'3 per operation » operator failure rate for operations
embedded in a procedure (JRB).
This is a relatively infrequent procedure (once per month) with no imme-
diate feedback. Therefore, 1 x 10"' (general failure rate for opera-
tions with no status display) is probably a better figure (Reactor
Safety Study).
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 1 x IQ"2/month
VOLUME:
_ We assume that the leak rate is equal to the pumping rate.
VARIATIONS:
t Storage and accumulation tanks. For many of these tanks, the strainer
will be included as part of the pump-out truck. In such cases, error in
A-46
-------
strainer maintenance will be detected and corrected at the first pump-
out following pump maintenance.
t We assume that the pump-out truck visits 40 tanks per month.
t Let n be the number of pump-outs per month for the modeled tank (n may
be less than 1). Then the probability 1s n/40 that the facility is the
first one to be visited after any given strainer maintenance. The
annual probability of a spill due to faulty maintenance of the pump-out
truck's strainer Is therefore given by:
1 x 10'2 x (n/40) per month » 3n x 10'3/yr.
USER INPUTS:
Number of pump-outs per month (n)
A-47
-------
HAZARDOUS WASTE TANKS
LABEL: A11112
FAILURE: Pump drain left open during pumping.
ASSUMPTIONS:
This can only happen following pump maintenance.
^
0 Pump maintenance occurs annually.
The response time In the event of a spill is FNU(.25,3) minutes.
SOURCES:
t Nuclear Regulatory Commission, Reactor Safety Study (1975)
JRB Associates (1982)
Event Alllll (strainer drain left open)
CALCULATIONS:
See event Alllll
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 10'2/yr
VOLUME:
The spill rate Is equal to the pumping rate.
VARIATIONS:
Storage and accumulation tanks. For many of these tanks, the pump-out
pump Is Included as part of the truck. An error In pump maintenance
w11rtherefore be detected and corrected at the first pump-out
following pump maintenance.
We assume that the truck visits 40 sites/month.
A-48
-------
0 Let n be the number of pump-outs per month at the facility being
modeled. Then the probability that the facility being modeled is the
first one to be visited after pump maintenance is n/4Q^ ^-and the
probability that it is the first one visited after faulty pump
maintenance is given by:
(1 x 10-2)(n/40) per year - 2.5n x 10'4/yr
USER INPUTS:
Number of truck visits per month (n)
A-49
-------
HAZARDOUS WASTE TANKS
LABEL: A1112
FAILURE: Hose ruptures (above ground) during pump-out (storage or accumulation
tanks only).
ASSUMPTIONS:
0 -Hose ruptures inside the pump-out pipe will be inconsequential. Leakage
will return to the tank in all but extraordinary circumstances.
SOURCES:
t JRB Associates (1982)
CALCULATIONS:
Hose rupture probability » 1 x 10"4/hr (JRB)
Annual probability of hose rupture « 1 x 10"4/hr x T hr/pump-out
x n pump-outs/week x 52 weeks/yr » 5(nT) x 10"3/yr
tf
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 5(nT) x 10'3/yr
VOLUME:
The maximum leak rate will be the pumping rate for the pump-out pump.
The minimum is approximately 0.
We assume that the leak rate is uniformly distributed between these two
exremes.
t The maximum detection/response time is the entire discharge time (see
Section A.1.6). The minimum detection/response time 1s about 15 seconds.
USER INPUTS:
V
Number of discharges per week (n)
0 Time required for pump-out (from Section A.1.6)
A-50
-------
HAZARDOUS WASTE TANKS
LABEL: A1113, A1114
FAILURE: Pump or strainer ruptures during pump-out (storage or accumulation
tanks only).
SOURCES:
U.S. Coast Guard (1978)
MRS Associates (1982)
CALCULATIONS
1 x 10'8/hr « strainer rupture rate (JRB)
1 x 10'8/hr pump rupture rate (JRB) -
COMMENTS:
These are very low probability events producing spills which will pro-
bably be contained by an above-ground pad. These events are
therefore not included in our model.
A-51
-------
HAZARDOUS WASTE TANKS
LABEL: A1115
FAILURE: Loose flexible hose connection during pump-out (storage or accumula-
tion tanks only).
SOURCES:
9* Nuclear Regulatory Commission, Reactor Safety Study (1975)
CALCULATIONS:
1 x 10"2/demand * general rate of human error (Reactor Safety Study)
0 1 x 10'2/demand « m x 10"2/month where m 1s the number of pump-outs per
month (m will generally be less than 1).
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY:
tlx lt)'2/demand » m x 10"2/month where m 1s the number of pump-outs per
year.
VOLUME:
Assume that the maximum loss rate 1s the hose flow rate. So leak rate
1s FNU(0, hose flow rate).
USER INPUTS:
t Number of pump-outs/year (m)
Tank capacity (Q)
.A-52
-------
HAZARDOUS WASTE TANKS
LABEL: A2113, A2114
FAILURE: Strainer or pump rupture during tank filling.
SOURCES:
t U.S. Coast Guard (1978)
, JRB Associates (1982)
CALCULATIONS:
1 x 10"8/hr » strainer rupture rate
1 x 10*8/hr « pump rupture rate
COMMENTS:
These are very low probability events producing spills which will
probably be contained by an above-ground pad. These events are
therefore not included in the computer model.
A-53
-------
HAZARDOUS WASTE TANKS
LABEL: A2116
EVENT: Pump Corrodes
ASSUMPTIONS:
t Pump corrosion 1s composed of:
*
- generalized Interior corrosion
- localized Interior corrosion
- generalized exterior corrosion
- loca-lized exterior corrosion
- erosion
CALCULATIONS:
Our pump-corrosion model Is very similar to our pipe-corrosion model.
Localized interior corrosion." We model localized Interior corrosion in
the same way for pumps and pipes. Because a pump's complex shape makes
1t more vulnerable to localized Interior corrosion than 1s a pipe, we
use-.the same probability distributions for localized interior corrosion
of pumps and pipes. See event B13, above.
Generalized interior corrosion. Generalized Interior corrosion 1s also
the same for pumps and pipes.
Generalized exterior corrosion. Because pumps are exposed to the atmos-
- phere, we use a generalized exterior corrosion rate of 1.4 mils per
year.
Localized exterior corrosion. Because of the pump's small surface area
and above-ground setting, localized exterior corrosion is highly un-
likely and we have not Included it in our pump-corrosion model.
Erosion. Because of the pump's complex shape, it will be more subject
to erosion than 1s a pipe. We assume the following dependence of ero-
sion rate on fraction of suspended solids:
Fraction of Erosion rate
Suspended Solids (mils per
(ppm) year)
0-10,000 FNU(O.IO)
10,000 FNUU0.20)
A-54
-------
VOLUME:
We use the same hole sizes for pumps as we used for pipes.
COMMENTS:
In some cases, pumps may be located inside the tank. In these cases,
pump corrosion cannot produce a release of fluid.
A-55
-------
HAZARDOUS WASTE TANKS
LABEL: A210
FAILURE: Fluid flows over the top of an open-topped tank during overfill
events.
ASSUMPTIONS:
.Open-topped tanks have a ready overflow route over the top of the tank,
VOLUME:
The leak rate will be equal to the rate at which fluid is pumped into
the system.
The overflow may be detected and remedied immediately, or it may con-
tinue through the entire batch.
We assume that the overflow duration is uniformly distributed between
zero and the entire fill time for the tank. Thus, the overflow volume
is given by FNU(0, volume of entire batch).
A-56
-------
HAZARDOUS WASTE TANKS
LABEL: A211
FAILURE: Fluid flows out the vent of an above-ground tank during overfill
events.
ASSUMPTIONS:
Above-ground tanks have a ready overflow route through their vents.
VOLUME:"
The leak rate will be equal to the rate at which fluid 1s pumped into
the system.
The overflow may be detected and remedied Immediately, or 1t may con-
tinue through the entire batch.
t We assume that the overflow duration is uniformly distributed between
zero and the entire fill time for the tank. Thus, the overflow volume
1s.given by FNU(0, volume of entire batch).
A-57
-------
HAZARDOUS WASTE TANKS
LABEL: A213
EVENT: Pump-out pipe rupture leads to loss during overfill events.
SOURCES: Bll
PROBABILITY DISTRIBUTION:
*
Non-stochastic. This event will occur if the pump-out pipe has ruptured
under event Bll.
PROBABILITY: See event Bll (pipe rupture)
VOLUME:
If there is a pump, we assume that It shuts off automatically. Then, the
pressure at the point of the rupture will be determined by the static
hydraulic head of the backed-up fluid.
We assume that the average rupture occurs at the midpoint of the pump-
out 'pipe. Then the hydraulic head can be determined from the system
1 ayout''.
We assume an overfill detection/response time of FNU(.25,60) minutes
(the operator notices that the fluid is not flowing).
USER INPUTS:
System design
COMMENTS:
t This event is only important for systems for which pump-out is by
flexible hose. For these tanks, this event is the only source of loss
through a ruptured pump-out pipe. For other tanks, losses by this
mechanism will be overshadowed by losses through the same holes during
normal discharge operations.
A-58
-------
HAZARDOUS WASTE TANKS
LABEL: A214
FAILURE: Pump-out pipe corrosion results in leak during overfill events.
SOURCES: B13
PROBABILITY DISTRIBUTION:
t Non-stochastic. This event will occur if the pump-out pipe has corroded
under event B13.
PROBABILITY: See event B13 (pipe corrosion)
VOLUME:
If there is a pump, we assume that it shuts off automatically, then, the
pressure at the corrosion hole will be determined by the static
hydraulic head of the baclced-up fluid.
As in event B13, we assume that the average corrosion hole occurs at the
midpoint of the pump-out pipe. Then the hydraulic head can be deter-
mined 'from the system layout.
We assume an overfill detection/response time of FNU(.25,60) minutes
(the operator notices that the fluid is not flowing).
USER INPUTS:
t System design
COMMENTS:
This event 1s only Important for systems for which pump-out is by
flexible hose. For these tanks, this event 1s the only source of loss
through a corroded pump-out pipe. For other tanks, losses by this
mechanism will be overshadowed by losses through the same holes during
normal discharge operations.
A-59
-------
HAZARDOUS WASTE TANKS
LABEL: A215
FAILURE: Outlet flange or gasket leak produces losses during overfill events.
SOURCES: See B121 or B122
PROBABILITY DISTRIBUTION:
0 Non-stochastic. This event will occur if the flange or gasket is
leaking under events B121 or B122.
PROBABILITY: See event B121 (flange leak) or B122 (gasket leak).
VOLUME: .
If there Is a pump, we assume that 1t shuts off automatically.- Then, the
pressure at the point of"the rupture will be determined by the static
hydraulic head of the backed-up fluid. .This will be determined by the
system design.
t We assume an overfill detection/response time of FNU(.25,60) minutes
(the operator notices that the fluid 1s not flowing).
USER INPUTS:
System design
COMMENTS:
This event Is only Important for systems for which pump-out is by
flexible hose. For these tanks, this event 1s the only source of loss
through a leaking outlet flange or gasket. For other tanks, losses by
this mechanism will be overshadowed by losses through the same hole
during normal discharge operations.
A-60
-------
HAZARDOUS WASTE TANKS
LABEL: A216
FAILURE: Inlet pipe rupture produces leaks during overfill events.
PROBABILITY DISTRIBUTION:
Non-stochastic. This event will only occur if the inlet pipe has-
already ruptured under event Bll.
PROBABILITY: See event Bll (pipe rupture).
VOLUME:
The leak-rate calculations are similar to those used for event A213
(pump-out pipe rupture).
COMMENTS: r
Leakage from this pipe will also produce losses during normal filling.
Cumulative losses from that mechanism will generally be much larger
than losses occurring during overfill events.
A-61
-------
HAZARDOUS WASTE TANKS
LABEL: A217
FAILURE: Fill pipe corrosion produces losses during overflow events.
PROBABILITY DISTRIBUTION:
Non-stochastic. This event will only occur if the fill -pipe is already
, corroded leaking under event B13.
PROBABILITY: See event B13 (pipe corrosion).
VOLUME:
t The leak rate can be obtained from the geometry of event B13,. using
a static hydraulic head. The size of the corrosion hole is determined
under event B13.
COMMENTS:
Leakage from this hole will also produce losses during normal filling.
Cumulative losses from that mechanism will generally be much larger than
losses occurring during overfill events.
A-62
-------
HAZARDOUS WASTE TANKS
LABEL: A218
FAILURE: Inlet flange or gasket leaks during overfill events.
PROBABILITY DISTRIBUTION:
Non-stochastic. This event will only occur if the inlet flange or
,gasket is already leaking under events B121 or B122.
PROBABILITY: See event B121 (flange leak) or B122 (gasket leak).
VOLUME:
t The leakage calculations are identical to those used for event A215
(outlet flange or gasket leak). . . ..
COMMENTS:
Leakage from this flange or gasket Mill also produce losses during nor-
mal .filling. The leak rate will be much higher during overflow,
however, because the pressure will be much higher under the conditions
of static hydraulic head that occur during overflow than it will be when
fluid is in motion during normal filling. During normal filling, the
pressure at this location is very low.
A-63
-------
HAZARDOUS WASTE TANKS
LABEL: A219
FAILURE: Vent pipe rupture produces losses during overfill events.
CALCULATIONS: See event Bll (pipe rupture).
PROBABILITY: See event Bll.
n.
VOLUME:
The volume can be obtained from the system layout, using static
hydraulic heads and sizing the rupture according to the method
used for other pipe ruptures. See event Bll.
A-64
-------
HAZARDOUS WASTE TANKS
LABEL: A220
FAILURE: Vent pipe corrosion produces losses during overfill events.
CALCULATIONS: See event B13 (pipe corrosion).
PROBABILITY: See event B13.
VOLUME:
The volume can be obtained from the system layout, using static
hydraulic heads and sizing the corrosion hole according to the method
used for other pipe ruptures. See event 613.
A-65
-------
HAZARDOUS WASTE TANKS
LABEL: A221
FAILURE: Vent pipe flange leaks during overfill events.
CALCULATIONS: See event B121 (welded flange leaks).
PROBABILITY: See event B121.
VOLUME:
The volume can be obtained from the system layout, using static
hydraulic heads and sizing the flange leak according to the method used
for other flange leaks. See event B13.
A-66
-------
HAZARDOUS WASTE TANKS
LABEL: B21
FAILURE: Asphalt Pad Breached
SOURCES: Conversations with local asphalt contractors.
ASSUMPTIONS:
V
Breakage 1n a 2" pad with 6" class-5 base (a mixture of gravel,
sand, and clay) will become general In an average of 8-12 years,
depending on maintenance. It could occur as early as 3-5 years.
A 3-4* pad with a crushed limestone base should last 15 years
before generalized break-up begins.
Break-up will occur earlier If the pad is not properly maintained.
The entire spill volume 1s lost 1f the pad has broken up.
PROBABILITY DISTRIBUTION: Beta .
PROBABILITY:' ;,
For 2" pad, no maintenance, we use a beta distribution with a
minimum value of 2.5 years, a mode of 8 years, and a maximum of
12 years.
For a 2" pad, with maintenance, (or a 3-4" pad without maintenance)
we use a beta distribution with parameters (4, 12, 15).
0 For a 3-4" pad with maintenance, we use a beta distribution with
parameters (5, 15, 18).
VOLUME: Entire volume of the spill.
A-67
-------
HAZARDOUS WASTE TANKS
LABEL: B31
FAILURE: Concrete Pad Breached
SOURCES:
Telephone conversations with concrete contractors and officials 1n the
Minnesota Department of Transportation.
L
PROBABILITY DISTRIBUTION: Normal
PROBABILITY: N(30,5)
VOLUME: Entire volume of spill.
A-68
-------
HAZARDOUS WASTE TANKS
LABEL: 833
FAILURE: Breach of Concrete Berm
ASSUMPTIONS:
t Concrete berms will age similarly to concrete pads (see event B31).
SOURCES: 'Event B31
PROBABILITY DISTRIBUTION: Normal
Probability: N(30,5)
VOLUME: Total spill volume.
A-69
-------
HAZARDOUS WSTE TANKS
LABEL: B41
FAILURE: Concrete Vault Fails
SOURCES:
0 Best engineering judgment after converstatlons with concrete contrac-
tors. .
s.
PROBABILITY DISTRIBUTION: Normal
PROBABILITY: N(35,10)
VOLUME: Total volume of spill.
A-70
-------
HAZARDOUS WASTE TANKS
LABEL: 851
FAILURE: Synthetic Liner Fails
SOURCES:
0 EPA, Liner Location Report (1984-).
PROBABILITY DISTRIBUTION: Normal
PROBABILITY: N(35,10)
VOLUME: Total volume of spill.
A-71
-------
HAZARDOUS WASTE TANKS
LABEL: ANUCAT(I.l)
FAILURE: Vehicle Crash
ASSUMPTIONS:
This event 1s Included 1n tank or pipe rupture (events T1124 and 811),
A-72
-------
HAZARDOUS WASTE TANKS
LABEL: ANUCAT (1,2) (aboveground or In-ground tanks only)
FAILURE: Vandalism of tank system resulting in total system loss.
ASSUMPTIONS:
We assume that the probability of catastrophic release due to vandalism
is of the same order of magnitude as the probability of catastrophic
release due to a vehicle crash.
s.
SOURCES:
SCS Engineers (1983), Figure 4-18, p. 4-29.
JRB Associates (1982), Exhibit 3-5.
CALCULATIONS:
< 1 x 10-10/hr - failure rate for vehicle crash (JRB).
0 10-10/hr - (1 x 10'10/hr)(24 hr/day)(365 day/yr) - 10-6/yr.
«
PROBABILITY: " 1 x 10'6/yr
PROBABILITY DISTRIBUTION: Binomial
VOLUME: Entire tank contents.
A-73
-------
HAZARDOUS WASTE TANKS
LABEL: ANUCAT (1,3)
FAILURE: Tornado/hurricane (I.e. high wind storm event) resulting in total
system loss (above-ground or in-ground tanks only).
ASSUMPTIONS:
t ^The facility is in coastal area subject to periodic hurricanes or in a
tornado-prone area.
0 The annual probability of a great hurricane (winds exceeding 125 mph)
for 50-mile segments along the U. S. coastline ranges from 1% to 7X
(Petak and Atkisson).
Assume 20X damage for hurricanes with wind speeds of 125 mph (Petak and
Atkisson).
0 Assume an average of 2.5 tornado strikes per 10,000 square miles for the
continental U.S. (U.S. Weather Bureau).
0 Approximately 35X of all tornadoes have a Fujita classification of
F2 or above (135 - 290 mph) (Petak and Atkisson).
*
0 Assume approximately 50% damage for structures affected by tornadoes
with a Fujita classification of F2 or above (Petak and Atkisson).
0 Assume a facility has an area of approximately 10 acres.
0 Assume that the average tornado strike is 2 mi x 300 yards * 200 acres.
(U.S. Weather Bureau.)
SOURCES:
0 Petak, William J. and Arthur A. Atkisson, Natural Hazard Risk Assessment
and Public Policy; Anticipating the Unexpected. Sprlnqer-Verlag, New
York (1982).
0 United States Weather Bureau, Minneapolis Office, Personal
Communication.
CALCULATIONS:
0 7% annual probability of hurricanes x 20% chance of damage » 1.4X chance
of a damage due to a hurricane.
A-74
-------
(2.5/10,000) tornado strikes per square mile x (200 acres/strike) x
(Imi2/640 acres) x 10 acres/facility x (.35 x .5) probability of damage
=» 1.5 x 10'4/yr per facility.
PROBABILITY:
t 0.014/yr for hurricanes
1.5 x 10'Vy'r for tornadoes
PROBABILITY DISTRIBUTION: Binomial
^
VOLUME: Entire tank contents
USER INPUTS:
Is facility 1n a tornado zone?
Is facility 1n a hurricane zone?
COMMENTS: This event only applies for above-ground or In-ground facilities.
A-75
-------
HAZARDOUS WASTE TANKS
LABEL: ANUCAT (1,4)
FAILURE: Earthquake causes total system loss.
ASSUMPTIONS:
t The facility Is located in a seismically active area. Recurrence
intervals for damaging earthquakes in Los Angeles and San Francisco
areas are between 100 and 125 years.
\
Assume .5-15% damage to commercial structures (built in California
after 1933) in response to an earthquake with intensity of 7 or above.
SOURCES:
California Institute of Technology (personal communication).
CALCULATIONS: .
(1/125) earthquakes per year x 10* average probability of damage
» 8 x 10'* damaging earthquakes per year.
\ '
PROBABILITY:'' .0008/yr
PROBABILITY DISTRIBUTIONS: Binomial
VOLUME: Entire tank contents
USER INPUTS:
Is facility 1n a seismically active area?
A-76
-------
HAZARDOUS WASTE TANKS
LABEL: ANUCAT (1,5)
FAILURE: Flood causes total system loss (above- or In-ground tanks only).
ASSUMPTIONS:
The facility 1s located in a flood prone area.
0 The facility 1s designed to withstand up to a 100-year flood.
SOURCES:
Thomas Dunne and Luna Leopold, Water In Environmental Planning.
W. H. Freeman and Company, San Francisco (1978).
t William J. Petak and Arthur A, Atklsson, Natural Hazard Risk Assessment
and Public Policy: Anticipating the Unexpected, Springer-Verlag, New
York (1982).
CALCULATIONS:
According to Petak and Atklsson there Is a 50% chance that a flood will
result in damage to an above-ground tank.
There 1s a 1% chance per year of a 100-year flood.
t IX x 50* » .5%
PROBABILITY: .005/yr
PROBABILITY DISTRIBUTION: Binomial
VOLUME: Entire contents of tank.
USER INPUT:
Is facility located in a flood-prone area?
COMMENTS: This event applies only for above- or In-ground facilities.
A-77
-------
HAZARDOUS WASTE TANKS
LABEL: ANUCAT (1,6)
FAILURE: Ignition source available to Ignite waste in tank system (all
systems).
ASSUMPTIONS:
The tank 1s properly grounded.
The operator 1s reasonably cautious In handling the waste.
PROBABILITY: 1 x 10*6/yr
PROBABILITY DISTRIBUTION: Binomial
VOLUME: Entire tank contents
COMMENTS:
This event applies 1f the waste Itself Is the source of the fire. If the
tank 1s ruptured by a nearby fire or explosion, ANUCAT (1,7) applies.
This event applies only if the waste 1s flammable.
A-78
-------
HAZARDOUS WASTE TANKS
LABEL: ANUCAT (1,7)
FAILURE: Nearby fire or explosion causes complete system loss.
ASSUMPTIONS:
A nearby fire or explosion is 1/3 as likely to damage an underground
.tank as it is to damage an above-ground tank.
SOURCES:
JRB Associates (1982), Exhibit 3-5
CALCULATIONS:
t 3 x 10'7/hr - probability of nearby fire (JRBj
(3 x 10-7/hr)(24 hr/day)(365 days/yr) » 2.6 x 10'3/yr
PROBABILITY?
0 3 x 10'3/yr (above-ground or in-ground) tank
1 x lO'Vyr (below-ground tank)
PROBABILITY DISTRIBUTION: Binomial
VOLUME: Entire tank contents
A-79
-------
HAZARDOUS WASTE TANKS
LABEL: LIFDEF (1,1) and LIFDEF (1,2)
FAILURE: Vibrational/tortlonal stress causes rupture due to Inadequate
support or due to a construction defect.
ASSUMPTIONS:
The only part of the system subject to vibration is the pump.
i
We assume that these losses are included in pipe rupture (event BID,
PROBABILITY: Zero. This loss mechanism is Included- in event A1113.
A-80
-------
HAZARDOUS WASTE TANKS
LABEL: LIFDEF (1,3)
FAILURE: Inspection falls to detect installation damage or fabrication errors.
ASSUMPTIONS:
'Assume four levels of inspection/testing:
1) none
2) lowvisual inspection
3) mediumvisual inspection and weld testing
4) highvisual inspection, weld testing, and tightness testing
SOURCES:
0 Best engineering judgment based on human error probabilities listed in
Nuclear Regulatory Commission, Reactor Safety Study (1975)
PROBABILITY (by inspection level):
None -1.00
Low -"0.50
Medium - 0.25
High - 0.05
PROBABILITY DISTRIBUTION: Binomial
USER INPUTS:
Level of inspection or testing.
A-81
-------
HAZARDOUS WASTE TANKS
LABEL: LIFDEF (1,4)
FAILURE: Off-spec materials used In construction.
ASSUMPTIONS:
t Use of poor-grade materials would accelerate the onset of
.various leaks and ruptures. Since our probability distributions
for these events are based on empirical data (the API/SCS
survey) we assume that these probability distributions already
account for off-spec materials.
A-82
-------
HAZARDOUS WASTE TANKS
LABEL: LIFDEF (I,5)A
FAILURE: Tank damaged during installation.
SOURCES:
Best engineering judgment
*
PROBABILITY: 2 x 10'2
PROBABILITY DISTRIBUTION: Binomial
VOLUME:
t We assume that the damage is similar to a seam leak.
t The volume will therefore be identical to volume loss from seam leaks
(see event T1124). Leakage will begin in year 1.
VARIATIONS:'. ,
tf
Concrete and Stainless Steel Tanks. Since steel and stainless steel are
approximately the same strength, we assume that stainless steel is just
as vulnerable to Installation damage as is steel. Based on conver-
sations with concrete contractors, we assume that concrete also has a 2%
chance of cracking due to improper installation.
t Fiberglass tanks. Fiberglass tanks are twice as likely to rupture as
are steel tanks (see event T1124). We therefore assume that they are
also twice as vulnerable to installation damage.
COMMENTS:
Due to a transcription error, we used a value of .03 for the installa-
tion damage probability for steel, sta-inless steel, and concrete. This
error did not substantially alter our results. It will be corrected in
subsequent versions of the model.
A-83
-------
HAZARDOUS WASTE TANKS
LABEL: LIFDEF (1,5)8
FAILURE: Underground piping damaged during Installation.
SOURCES: Best engineering judgment
PROBABILITY: 1 x 10'2
PROBABILITY DISTRIBUTION: Binomial
VOLUME:
We assume that the damage is similar to that from a pipe rupture.
t The leak rate will therefore be Identical to that from-event Bll. Loss-
will begin in year 1.
VARIATIONS:
0 Stainless steel pipes. Since steel and stainless steel are approxima-
tely the same strength, we assume that they are equally vulnerable to
Installation damage.
Fiberglass pipes. Fiberglass 1s twice as likely to rupture as 1s steel
T]
Installation-damage probability of 2 x 10'
(see event T1124). We therefore assume that fiberglass pipes have an
A-84
-------
HAZARDOUS WASTE TANKS
LABEL: LIFOEF (I,5)C
FAILURE: Above-ground piping damaged during installation.
SOURCES: Best engineering judgment
PROBABILITY: 1 x 10'2
**
PROBABILITY DISTRIBUTION: Binomial
VOLUME:
The leak rate is the same as that from an above-ground pipe rupture
(see event BID.
Leakage will begin in year 1.
VARIATIONS:
t Stainless steel pipes. Since steel and stainless steel are approxima-
tely the same strength, we assume that they are equally vulnerable to
installation damage.
Fiberglass pipes. Fiberglass is twice as likely to rupture as is steel
(see event T1124). We therefore assuem that fiberglass pipes have an
installation-damage probability of 2 x 10'2.
A-85
-------
HAZARDOUS WASTE TANKS
LABEL: LIFDEF (1,5)0
FAILURE: Welded flange damaged during installation.
SOURCES:
^Best engineering judgment
PROBABILITY: 2 x 10'2
PROBABILITY DISTRIBUTION: Binomial
VOLUME:
We assume that the leak rate is the same as that for a welded flange
leak (event B121).
t Leakage will begin in year 1.
A-86
-------
HAZARDOUS WASTE TANKS
LABEL: LIFDEF (I,5)E
FAILURE: Gasket damaged during installation (or improperly installed).
SOURCES:
0 Best engineering judgment
PROBABILITY: 1.5 x 10'2
PROBABILITY DISTRIBUTION: Binomial
VOLUME:
We assume that the leak rate is the same as that for a gasket failure
(event B122).
Leakage begins In year 1.
A-87
-------
HAZARDOUS WASTE TANKS
LABEL: LIFDEF (1,6) and (1,7)
FAILURE: Stresses due to settling.
ASSUMPTIONS: This event is already included in tank and piping ruptures.
A-88
-------
HAZARDOUS WASTE TANKS
LABEL: FLVCNl
FAILURE: Automatic level controller fails.
ASSUMPTIONS:
t Assume that the failure of the controller, the controller settings, and
the Impulse lines may each cause controller malfunction
SOURCES:
Anyakora, Engel and Lees, Table V, p. 400
CALCULATIONS:
0.29/yr « 7.9 x 10'Vdy. » controller failure rate (Anyakora, Erigel and
Lees).
0.14/yr « 3.8 x 10~4/dy » controller settings failure rate (Anyakora,
Engel and Lees).
0.77/yr » 2.1 x 10'3/dy « impulse lines failure rate (Anyakora, Engel
and Le^s).
t 7.9 x lO'4 + 3.8 x lO'4 + 2.1 x 10'3 - 3.3 x 10'3/dy
t 3.3 x 10-3/dy » 9.4 x 10-2/mo
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 9.4 x 10'2/mo
A-89
-------
HAZARDOUS WASTE TANKS
LABEL: FLVCN2
FAILURE: Emergency shut-off level controller fails to function.
ASSUMPTIONS:
We assume that the emergency shut-off controller Is inspected monthly.
A
SOURCES: See event FLVCN1
CALCULATIONS:
According to event FLVCN1, the probability of automatic level controller
failure is 9.4 x 10"2/mo.
If the level controller is inspected monthly, then the average failure
lasts half a month. The probability that the controller is in a
failed state at any given time is therefore .5 x 9.4 x 10'S which is
approximately 5 x 10'2.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: .OS/Demand
A-90
-------
HAZARDOUS WASTE TANKS
LABEL: LEVIN2
FAILURE: Emergency shut-off level sensor falls to function.
ASSUMPTIONS:
0 We assume that the published failure rates include failures in the
meter, the sensor, and the impulse lines.
i
We assume that the emergency shut-off level sensor is inspected monthly.
SOURCES:
Anyakora, Engel, and Lees (1971), Table V, p. 400.
CALCULATIONS:
t .2/yr « failure rate for a capacitance-type level transducer (Anyakora,
Engel and Lees).
.2/yr - 5.5 x 10'4/day » 1.6 x 10'2/mo.
» . »
If the level controller is Inspected monthly, then the average failure
lasts half a month. The probability that the controller is in a
failed state at any given time is therefore .5 x 1.6 x 10'2, which is
8 x lO'3.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 8 x 10'3 .
COMMENTS:
Due to a transcription error, this event's probability was set to the
probability calculated for event FLVCU2. The result was a conservative
error but the number will be changed to the value calculated above in
future versions of the model.
A-91
-------
HAZARDOUS WASTE TANKS
LABEL: MOALARM
FAILURE: High liquid level alarm system failure.
ASSUMPTIONS:
We assume that the high level alarm Is tested once per year. Thus, the
^average alarm failure will persist for 6 months.
SOURCES:
Lawley (1974), p. 54, note 6.
CALCULATIONS:
t 0.2/yr » frequency of dangerous high level alarm failures (LawTey).
6 months » duration of average undetected failure.
0.2 (6/12) « 0.1 » fractional dead time for high level alarm.
i «
PROBABILITY: ".I/demand
PROBABILITY DISTRIBUTION: Binomial
COMMENTS:
The same failure probability applies to leak detectors in vaulted tanks
or interstitial alarms in double-walled tanks or pipes.
We also model Interstitial alarms according to this probability distri-
bution. Many of these alarms, however, have status lights which can be
checked at any desired frequency. Thus, if status is checked conscien-
tiously, the per-demand failure probability may be considerably lower.
Our value of 102 is therefore a conservative estimate.
A-92
-------
HAZARDOUS WASTE TANKS
LABEL: MOFILL
FAILURE: Tank 1s to be filled nearly to capacity.
ASSUMPTIONS:
Treatment tanks are always filled to their operating capacities.
^
0 We assume that pump-out schedules for storage or accumulation tanks
generally allow sufficient margin for error that the tank is not filled
completely to capacity unless something Interferes with the normal pump-
out schedule or there is an unexpected upsurge In the generation of
waste. We conservatively assume that this happens once per year.
CALCULATIONS:
1 per year » .0027 per day « 1 - X1-.0027)30 .079 per month.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: '
t 1.00 for treatment tanks.
t .079/mo. for storage or accumulation tanks.
A-93
-------
HAZARDOUS WASTE TANKS
LABEL: MOLEVIN
FAILURE: Level indicator malfunction results in attempted overfill.
ASSUMPTIONS:
We assume that the published failure rates include failures in the
meter, the sensor, and the impulse lines.
*
We assume that a fault is detected and repaired after one faulty
transfer.
SOURCES:
Anyakora, Engel, and Lees (1971), Table V, p. 400.
CALCULATIONS:
.22/yr » failure rate for a capacitance-type level transducer (Anyakara,
Engel, and Lees).
.22 x .5 » ,1/yr « rate of overfill events due to level transducer
failure.
.2/yr - 5.5 x 10'4/day « l-(l-5:5 x 10-4)30/mo.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: .16/mo.
VOLUME:
Assume that the overflow consists of between 0 and 100X of a batch.
Thus Q - FNU(0, volume of one batch).
USER INPUTS:
0 Volume transferred per batch
COMMENT:
Due to a round-off error our model uses a probability of .15/mo. The
difference between this and the correct value is not significant.
A-94
-------
HAZARDOUS WASTE TANKS
LABEL: MOPMCE
FAILURE: Outlet pump fails to start on demand (extreme environment)
ASSUMPTIONS:
t 75% of pump failures are failures to start. The remaining 25% are
^failures to run under event MOPMOE
SOURCES:
t Nuclear Regulatory Commission, Reactor Safety Study (1975),
Table III 2-3.
Southwest Research Institute (1982).
Henley and Kumamoto (1981), Figure 6.7, p. 278.
CALCULATIONS:
0 1 x, 10*4 to 1 x 10*3 per operating hour » probability of pump failure
in. extreme environment (Reactor Safety Study)
The geometric mean of this range of values « 3 x 10~4 per operating
hour.
(3 x 10-4/hr)(.75) « 2.25 x 10'4 per operating hour » probability that
pump fails to run.
We convert this per-hour probability into a per-demand probability. We
do this by noting that the pump must have started properly when the pre-
vious batch drained. Otherwise, the operator would have noticed the
failure and repaired it. Thus, if the pump fails, it does so during the
fill time for the present batch. The probability of this is given by:
(2 x 10-4 x fill time)
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: (2 x 10'4 x fill time per batch) per demand^
A-95
-------
USER INPUTS:
The per-batch fill time
COMMENTS:
For continuous systems, the fill time 1s the entire operating day.
Due to a round-off error our model uses a probability of 3 x 10-* x
fill time per batch. The difference between this and the correct value
Is not significant.
A-96
-------
HAZARDOUS WASTE TANKS
LABEL: MOPMCN
FAILURE: Outlet pump fails to start on demand (normal environment)
ASSUMPTIONS:
t 75X of pump failures are failure to start. The remaining 25% are
failures to run under event MOPMON.
SOURCES:
0 Nuclear Regulatory Commission, Reactor Safety Study (1975),
Table III 2-3.
Southwest Research Institute (1982).
Henley and Kumamato (1981), Figure 6.7, p. 278.
CALCULATIONS:
1 x 10'* to 1 x 10~5 per operating hour « probability of pump failure
in normal environment (Reactor Safety Study).
The geometric mean of this range of values » 3 x 10"^ per operating
hour.
t (3 x 10'6/hr)(.75) « 2.25 x 10"6 per operating hour « probability that
pump fails to run.
We convert this per-hour probability into a per-demand probability.
We do this by noting that the pump must have started properly when the
previous batch drained. Otherwise, the operator would have noticed the
failure and repaired it. Thus, if the pump fails, it does so during
the fill time for the present batch. The probability of this is given
by:
(2 x 10'6 x fill time)
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: (2 x 10'6 x fill time per batch) per demand.
A-97
-------
USER INPUTS:
The per-batch fill time
COMMENTS:
For continuous systems, the fill time Is the entire operating day.
Due to a round-off error our model uses a probability of 3 x 10-6 x f
time p.er batch. The difference between this and the correct value is
-not significant.
A-98
-------
HAZARDOUS WASTE TANKS
LABEL: MOPMOE
FAILURE: Pump fails 1n "on" position, preventing emergency shut-off (extreme
environment).
ASSUMPTIONS:
" Assume that failure of the solenoid or the controller can cause a pump to
fail in the open position. We assume that these subcomponent failures
make up approximately 2556 of all pump malfunctions.
SOURCES:
Nuclear Regulatory Commission, Reactor Safety Study (1975).
Southwest Research Institute.(1982).
Henley and Kumamoto (1981), Figure 6.7, p. 278-.
CALCULATIONS:
1 x 1
extreme environment (Reactor Safety Study).
The g
hour.
1 x 10" to 1 x 10 per operating hour « probability of pump failure in
The geometric mean of this range of values 3 x 10 per operating
(3 x 10"4/hr) (.25) - 7.5 x 10"5 per operating hour » probability that
the pump fails in the "on" position.
We need to convert this per-hour probability into a per-demand probabi-
lity. We do this by noting that the pump must have shut off properly
after the previous batch finished filling. Otherwise, the operator
would have noticed the failure and repaired it. Thus, if the pump fails
It does so during the fill time for the present batch. The probability
of this 1s given by:
(7.5 x 10"5 x fill time)
PROBABILITY DISTRIBUTION: Binomial
V
PROBABILITY: (7.5 x 10"5 x fill time per batch) per demand
A-99
-------
USER INPUTS:
t The per-batch fill time
COMMENTS:
For continuous systems, the fill time is the entire operating day.
A-100
-------
HAZARDOUS WASTE TANKS
LABEL: MOPMON
FAILURE: Pump fails 1n "on" position, preventing emergency shut-off (normal
environment).
ASSUMPTIONS:
Assume that failure of the solenoid or the controller can cause a pump
to fail in the open position. We assume that these subcomponent
failures make up approximately 25X of all pump malfunctions.
SOURCES:
Nuclear Regulatory Commission, Reactor Safety Study (1975).
t Southwest Research Institute (1982).
Henley and Komamoto (1981), Figure 6.7, p. 278.
CALCULATIONS:
1 x. 10;"7 to 1 x 10"^ per operating hour » probability of pump failure
in normal environment (Reactor Safety Study).
The geometric mean of this range of values » 3 x 10~6 per operating
hour.
t (3 x W6/hr)(.25) - 7.5 x 10'7 per operating hour « probability that
the pump fails In the "on" position.
0 We convert this per-hour probability Into a per-demand probability in
the same way that we do for pumps in an extreme environment (event
MOPMOE). Thus, the probability 1s given by:
(7.5 x 10-7 x fill time)
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: (7.5 x 10*7 x fill time per batch) per demand
A-101
-------
USER INPUTS:
0 The per-batch fill time
COMMENTS:
0 For continuous system, the fill time is the entire operating day.
A-102--
-------
HAZARDOUS WASTE TANKS
LABEL: MOVLCE
FAILURE: Outlet valve fails in the closed position, preventing emergency shut-
off (extreme environment).
ASSUMPTIONS:
0 We assume that 50% of valve failures occur in the closed position.
SOURCES:
Anyakora, Engel, and Lees (1971), Table V.
Southwest Research Institute (1982), Table 2, p. 32.
Henley and Kumamoto (1981), Figure 6.7, p. 278.
Nuclear Regulatory Commission, Reactor Safety Study, (1975), Table III-
2-3.
CALCULATIONS':
0 0.60/yr * expected number of control valve failures in a normal
environment. (Anyakora, Engel and Lees (1971).
0 Valve failures are 10 times more common in extreme than normal environ-
ments. (Source, best engineering judgment based on Henley and Kumamoto
(1981) and Reactor Safety Study). Therefore, 6.0/yr » the expected
number of control valve failures in an extreme environment.
0 If SOX of these failures occur in the closed position, then 3.0/yr = the
expected number of control valves failing in the closed position.
0 3.0/yr « 3.4 x 10'4/hr
0 Since the valve must have been functional at the time the previous batch
drained, this event can only occur If failure occurs during the time
when the tank 1s being filled. Thus, the per demand failure probability
1s given by:
(3.4 x lO'4) x (fill time per batch)
PROBABILITY DISTRIBUTION: Binomial (per demand)
A-103
-------
PROBABILITY: (3.4 x 10"4) x (fill time per batch) per demand
USER INPUTS:
Fill time per batch
Number of operating hours per day
COMMENTS:
For continuous systems, the fill time is the entire operating day.
A-104
-------
HAZARDOUS WASTE TANKS
LABEL: MOVLCN
FAILURE: Outlet valve fails in the closed position, preventing emergency shut-
off (normal environment).
ASSUMPTIONS:
Failures are only 10X as likely in normal as extreme environments
(see sources cited under event MOVLCE).
SOURCES: See event MOVLCE
CALCULATIONS:
The per demand probability of failure in an extreme environment Is
3.4 x 10'4 x (fill time per batch).. See event MOVLCE.
PROBABILITY DISTRIBUTION: Binomial (per demand)
PROBABILITY: "(3.4 x 10"4) x (fill time per batch)
USER INPUTS:
_ Fill time per batch
Number of operating hours per day
COMMENTS:
t For continuous systems, the fill time is the entire operating day.
A-lOi
-------
HAZARDOUS WASTE TANKS
LABEL: MOVLOE
FAILURE: Inlet valve falls 1n the open position, preventing emergency shut-off
(extreme environment).
ASSUMPTIONS:
*
t Assume that SOX of valve failures occur in the open position.
SOURCES:
Anyakora, Engel, and Lees (1971), Table V.
Southwest Research Institute (1982), Table 2, p. 32.
Henley and Kumamoto (1981), Figure 6.7, p. 278.
Nuclear Regulatory Commission. Reactor Safety Study (1975) Table III-
2-3.
CALCULATIONS:''
0.60/yr « expected number of control valve failures in a normal
environment (Anyakora, Engel and Lees (1971)).
Valve failures are 10 times more common In extreme than normal environ-
ments. (Source, best engineering judgment based on Henley and Kumamoto
(1981) and Reactor Safety Study). Therefore, 6.0/yr « the expected
number of control valve failures in an extreme environment.
If SOX of these failures occur 1n the open position, then 3.0/yr » the
expected number of control valves failing in the open position.
3.0/yr - 3.4 x 10'4/hr.
Since the valve must have been functional at the time the batch began to
fill, this event can only occur if failure occurs during the time when
the tank is being filled. Thus, the per demand failure probability 1s
(3.4 x 10'4) x (fill time per batch).
PROBABILITY DISTRIBUTION: Binomial (per demand)
PROBABILITY: (3.4 x 10"4) x (fill time per batch)
A-106
-------
USER INPUTS:
0 Fill time per batch
Number of operating hours per day
COMMENTS:
t ,For continuous systems, the fill time 1s the entire operating day.
This event can also cause an attempted overfill. We only model this
form of attempted overfill for gravity-fed systems, however, because
for pump-fed systems, the failure can be remedied by shutting off the
pump. In theory, the pump could also fail 1n the "on" position, causing
an attempted overfill even for pump-fed systems, but such simultaneous
failure 1s extremely unlikely, and 1s overshadowed by the other types
of failure (e.g. operator error) which are more likely to cause attempt-
ed overflows.
A-107
-------
HAZARDOUS WASTE TANKS
LABEL: MOVLON
FAILURE: Inlet valve falls 1n the open position, preventing emergency shut-off
(normal environment).
ASSUMPTIONS:
Failures are only 10X as likely 1n normal as extreme environments
(see sources cited under event MOVLOE).
SOURCES: See event MOVLOE
CALCULATIONS:
The per demand probability of failure 1n an extreme environment 1s
(3.4 x 10'4) x (fill time per batch). See event MOVLOE.
PROBABILITY DISTRIBUTION: Binomial (per demand)
PROBABILITY:'' (3.4 x 10"5) x (fill time per batch)
USER INPUTS:
Fill time per batch
Number of operating hours per day
COMMENTS:
0 For continuous systems, the fill time 1s the entire operating day.
This event can also cause an attempted overfill. We only model this
form of attempted overfill for gravity-fed systems, however, because
for pump-fed systems, the failure can be remedied by shutting off the
pump. In theory, the pump could also fall 1n the "on" position,
causing an attempted overfill even for pump-fed systems, but such
. simultaneous failure Is extremely unlikely, and 1s overshadowed by the
. other types of failure (e.g. operator error) which are more likely to
cause attempted overflows.
A-108
-------
HAZARDOUS WASTE TANKS
LABEL: OFTROM
FAILURE: Operator falls to respond to high level alarm.
ASSUMPTIONS:
"Failure may be due to failure to hear alarm, failure to take corrective
action, or inability to take corrective action.
SOURCES:
Lawley (1974), p. 54, note 7.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 3 x 10~2/demand
A-109
-------
HAZARDOUS WASTE TANKS
LABEL: OFTRCOM
FAILURE: Operator erroneously responds to high level alarm.
ASSUMPTIONS:
* At the time when the alarm first sounds, the operator feels no sense
of panic. He responds 1n a routine manner.
SOURCES:
Nuclear Regulatory Commission, Reactor Safety Study (1975).
CALCULATIONS:
3 x 10'3/demand « probability of human error of commission (selecting
wrong switch, etc.), Reactor Safety Study. Table III 6-1.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 3 x 10'3/demand
COMMENTS:
If the operator panics, the probability of error will be much higher.
The Reactor Safety Study gives an error rate of 20-30X for trained
personnel under high stress levels where dangerous activities are
occurring rapidly.
A-110
-------
HAZARDOUS WASTE TANKS
LABEL: OPCOMM
FAILURE: Operator error in batch start-up leads to attempted overfill.
ASSUMPTIONS:
Operate action is required to Initiate the transfer of fluid at the
start of each batch. Mistakes may result in an attempt to transfer
*too much fluid.
Operator action is also necessary whenever a continuous process is
started up. We assume that this occurs once per operating day.
SOURCES:
Nuclear Regulatory Commission (1975)
CALCULATIONS:
3 x 10"3/demand « estimated rate of human errors of commission (e.g.
selecting a wrong switch). Source: Nuclear Regulatory Commission.
Let n-be the number of batches per day. (Let n « 1 for continuous
systems).
t Then the probability that the operator makes no errors is 1-3 x 10~3
per batch, or
(l-.003)30n
per month. The probability of 1 or more errors is
l-(l-.003)30n
per month.
If n » 1 this value is .086/mo.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY:
1 - .99730n/mo for batch systems
A-lll
-------
.086/mo for continuous systems
USER INPUTS:
0 Number of batches per day (n)
A-112
-------
HAZARDOUS WASTE TANKS
LABEL: OPVLOE
FAILURE: Inlet valve falls to close, causing overflow (extreme/environment).
ASSUMPTIONS:
This event only applies for automatic valves. Manual valves are very
unlikely to fall since they have no automated components.
SOURCES: See event MOVLOE.
CALCULATIONS:
. According to event MOVLOE, the probability of such an event Is
3.4 x 10'4/hr.
t The probability of failure during any given batch 1s therefore
(Tb).(3.4 x 10-4) where Tb 1s the fill time per batch.
0 The probability of failure during any given month 1s
l-[l-Tb<3.4 x 10-4)]nbm
where nb 1s the number of batches per day, and m is the number of
operating days per month. For continuous systems, nb 1s 1 and Tb Is the
entire operating day.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 1- fl - Tb(3.4 x 10'4)] nbm
USER INPUTS:
0* Fill-time per batch (Tb)
0 Number of batches per day (nb)
0 Number of operating day per month (m)
A-113
-------
HAZARDOUS WASTE TANKS
LABEL: OPVLON
FAILURE: Inlet valve falls to close, causing overflow (normal environment)
ASSUMPTIONS:
Yhls event only applies for automatic valves. Manual valves are very
unlikely to fall since they have no automated components.
SOURCES: See event MOVLON
CALCULATIONS:
According to event MOVLON, the probability of such an event 1s 3.4 x
10-5/hr.
0 The probability of failure during any given failure 1s therefore
(Tb>(3.4 x 10-5) where *Tb Is the fill time per batch.
0 The probability of failure during any given month 1s
1- [l-Tb(3.4 x 10-4)]nbm
where nb 1s the number of batches per day, and m 1s the number of
operating days per month. For continuous systems, nb 1s 1 and Tb 1s the
entire operating day.
PROBABILITY DISTRIBUTION: Binomial
PROBABILITY: 1- [l-Tb(3.4 x 10'5)] nbm
USER INPUTS:
0 'Fill-time per batch (Tb)
0 Number of batches per day (nb)
0 Number of operating days per month (m)
A-114
-------
HAZARDOUS WASTE TANKS
LABEL: PADINSF, VLTINSF, CRBINSF
FAILURE: Visual Inspection falls to detect secondary containment failure
(pad, yault, curb)
\
ASSUMPTIONS:
We assume that this visual Inspection 1s a passive "walk-around."
We assume that the visual Inspection 1s Infrequent enough that 1t does
not become monotonously routine to the operator.
SOURCES:
t Best engineering judgment based on Nuclear Regulatory Commission,
Reactor Safety Study (1975), and Lawley (1974).
PROBABILITY'. DISTRIBUTION: Binomial
f,.
PROBABILITY: .1
COMMENTS:
t For simplicity, we assume that all secondary-containment failures occur
at the beginning of the month and that they are repaired Immediately
after detection. Thus, the secondary-containment system 1s 1n a failed
state for a minimum of one complete month. A* similar result would be
obtained by assuming that cracks occur 1n the middle of the Inspection
cycle, and that repair takes two weeks.
We assume that all secondary-containment Inspections are statistically
Independent events. Thus a failure to detect a breach 1n a vault does
not Influence the probability that the Inspector will also fall to
detect-a breach in a pad or curb. Similarly, a failure to detect a
fault 1n one month does not change the probability that 1t will be
detected during the next Inspection-.
A-115
-------
APPENDIX B
Statistical Analysis of PACE
Tank Corrosion Data
-------
INTRODUCTION
A group of Canadian oil companies, working through the Petroleum Association for
Conservation of the Canadian Environment (PACE) have compiled data on 300 under-
ground gasoline storage tanks.* It is unclear what sampling techniques were
used to select the 300 tanks, but it appears that the intent was to obtain a
"snapshot" of the contemporary situation. It is not clear whether the data dis-
tinguishes interior and exterior corrosion, but since the intent was to deter-
mine the effect of soil variations on tank leakage, it must be assumed that the -
survey focused on exterior corrosion. The Canadian survey therefore represents
raw data distinct from the API tank leak survey? and independent from either
Warren Rogers'3 preliminary or revised statistical model.
In raw form, these data are pr.esentedMn Figures 1 and 2.* They consist of
scatter diagrams of tank age and "Soil Aggressiveness Values" (SAV) for 108
leaking tanks and 192 non-leaking tanks. Each dot on the scatter diagrams
represents one or more tanks, with overlapping points tallied by the small
numerals adjacent to the relevant dots. Numerical listings of all 300 points
are presented in Tables 1, 2, 3, and 4.
SAV is calculated according to the formula depicted in Figure 3. It is designed
to Incorporate soil resistivity, pH, moisture content, and sulfldes, as well as
the effect of variations in resistivity and pH over the tank installation site.
The resultant numerical index 1s designed to present a cardinal ranking of soil
^This data 1s discussed in PACE, "Underground Tank Systems: Review of State of
the Art and Guidelines," PACE report No. 82-3. Ottawa (1983).
2American Petroleum Institute, Tank and Piping Leak Survey, 1977 to 1980.
^Warren Rogers Associates, Inc., "Prediction of Leaks 1n Unprotected Steel
Storage.Tanks," Included in correspondence package from Betsy Tarn, EPA to Chris
Lough, PRA.
4These data were provided by Esso Petroleum, Canada. Esso Canada participated in
the PACE study.
B-l
-------
00
O£
s
O H-
M 00
5^ O
O
LU
40
35
30
25
~ 20
15
10
MliUKt 1
LEAKING TANK CHART
AGE VS. SOIL AGGRESSIVENESS
-
,
,
2
t
'
t
3
, ,
,3
2
i
L
k i
I
,
,2
»-j_
I i
,4,
I J
^
5,
2
<
,3
2
h
"
,2,
*3
'
>g
... J
>,
21
,2
»
1
__
M
2
»,_
2
'
k
4
(2
>, ,
f
»*
'
5
,2
'
2
4
'
j
1
1
,
1
(
1'-
!
i
*
i
i
i
"
I i
I
1
I
,
1
1
j
1 * '
1
i
1
i
i
1
!
I
i
i
1
i
i
i
i
i
t
i
t
'
. .j._
t
;
' 1
t
10 B-2 15 .
SOIL AGGRESSIVENESS VALUE
20
25
-------
FIGURE 2
TASK FORCE WON-LEAKING TANK CHART
35
30
CO
c:
<
LJ
»
sx
U.
o
UJ
o
25
20
15
10
1
1 .
I
3
4
' 2<
2
3 2 '
. .2
2,
?-1
*
"*
4
4
2
<
2-
5
*
<
2
-id
1 . > '
" j I
' 4
3
, *3
* r 2
*
* 9 2
t
<
»
r .
* . .2-
»2 <
>2 *2
> -t
8
3 »
2
6:
3
i
3
* * 2
3
3»3 3
- .4I.
^
i ^^
0
,
4
*
; 4
> 4^
2
3
i
.
'
' ; .
. *
.
5 10 15 . 20
SOIL AGGRESSIVENESS VALUE
B-3
25
November 15
-------
TABLE 1
Boil Aggressiveness Values (SAV) and ages for leaking tanks.
Leaking tanks
Leaking tanks
BAV
Ag«
Age»SAV
8AV
Age
Age*BAV
s
3
1O
9
13
4
4'
6
6
6
16
10
14
14
3
6
9
10
1O
7
7
10
10
10
6
6 '
6
9
a
14
9
12
10
9
9
9
9
11
18
15
10
9
9
17
11
11
11
9
12
23
7
19
6
2O
20
19
19
19
6
11
B
a
39
2O
19
14
14
20
20
13
19
19
29
29
29
3O
19
11
18
14
17
19
19
19
19
16
10
12
18
20
20
II
17
17
17
21
60
69
70
79
78
80
BO
9O
90
9O
96
110
112
112
117
120
139
140
140
140
140
190
130
19O
190
190
190
ISO
192
194
162
168
17O
171
171
171
171
176
180
1BO
ISO
ISO
18O
187
187
187
187
189
Leaking tanks
BAV Age Age*SAV
9
10
10
16
16
12
12
11
9
10
1O
10
10
10
14
11
11
10
12
13
13
9
9
14
14
16
16
12
12
10
a
14
14
14
14
IS
IS
IS
13
16
13
13
12
10
21
IS
IS
12
21
19
19
12
12
16
16
18
22
20
20
20
20
20
IS
20
20
22
19
IB
IB
26
26
17
17
15
IS
20
2O
24
30
IB
18
IB
18
17
17
17
20
17
21
21
23
29
14
2O
20
25
189
19O
190
192
192
192
192
198
198
200
2OO
2OO
200
200
21O
220
220
220
228
234
234
234
234
238
238
240
240
240
24O
240
24O
232
232
252
252
25S
255
255
260
272
273
273
276
29O
294
3OO
30O
.3OO
10
16
12
16
9
16
16
16
16
16
16
14
30
2O
2B
22
40
29
29
29
29
29
26
30
3OO
32O
336
352
360
400
4OO
4OO
4OO
400
416
420
-------
TABLE 1
Soil Aggressiveness Values (SAV) and ages for leaking tanks.
Leaking tank*
Leaking tank*
BAV
CO
i
ui
Agi
Age»SAV
BAV
s
3
to
5
13
4
4
6
6
6
16
to
14
14
3
6
9
10
10
7
7
10
10
10
6
6
6
S
a
14
9
12
10
9
9
9
9
11
18
IS
10
9
9
17
11
11
11
9
12
23
7
IS
6
2O
2O
IS
IS
IS
6
11
a
a
39
20
IS
14
14
20
2O
IS
IS
IS
23
23
23
30
19
11
18
14
17
19
19
19
19
16
10
12
18
2O
2O
11
17
17
17
21
60
69
70
73
78
80
BO
9O
9O
90
96
110
112
112
117
120
133
140
140
140
140
130
130
150
ISO
ISO
ISO
ISO
132
IS4
162
168
I7O
171
171
171
171
176
ISO
IBO
IBO
18O
ISO
187
187
187
187
189
Age
Age*BAV
Leaking tanks
BAV Age Age»6AV
9
10
to
16
16
12
12
11
9
10
10
10
to
10
14
11
11
10
12
13
13
9
9
14
14
16
16
12
12
10
a
14
14
14
14
13
IS
IS
13
16
13
13
12
10
21
IS
IS
12
21
19
19
12
12
16
16
18
22
20
20
20
20
20
IS
20
2O
22
19
. 18
IB
26
26
17
17
IS
IS
20
2O
. 24
3O
18
IB
IB
18
17
17
17
2O
17
21
21
23
29
14
20
20
25
189
190
19O
192
192
192
192
198
198
20O
200
2OO
200
200
210
22O
220
22O
228
234
234
234
234
238
238
240
240
240
240
240
240
232
232
232
2S2
235
233
233
26O
272
273
273
276
29O
294
300
300
3OO
10
16
12
16
9
16
16
16
16
16
16
14
30
20
28
22
40
23
23
23
23
23
26
30
3OO
32O
336
332
360
400
4OO
4OO
4OO
400
416
420
-------
Soil Aggressivmnmmm Value*
-------
Boil Aggre«*ivenese Values and ages for leaking tank..
TABLE 3
Leaking tank*
Leaking tank*
Leaking tank*
BAV
Age
Age«8AV
SAV
Agi
Age*SAV
BAV
13
16
1O
14
14
18
to
14
17
9
15
16
16
1O
1O
12
21
9
6
6
6
9
1O
1O
10
14
16
16
11
12
12
1O
11
11
11
14
14
19
19
19
16
9
1O
11
13
13
14
14
6
6
7
a
8
10
11
11
11
12
12
12
12
14
14
14
14
19
19
19
19
19
19
19
19
19
19
19
16
16
16
17
17
17
17
17
17
17
17
17
17
18
18
IB
IB
18
18
18
78
96
70
112
112
14
14
a
9
9
18O 9
11O 9
194 10
187 1O
6O 12
180 4
192 4
192 6
14O 7
140 7
168 9
294 9
79 1O
9O 1O
90 10
9O 1O
139
190
19O
190
21O
240
24O
176
192
192
170
187
187
187
238
10
11
11
12
12
13
13
19
16
9
9
13
13
9
1O
238 16
239 3
299 12
299 10
272 6
162 6
180 6
198 12
234 16
234 16
232 16
232 16
18
IB
19
19
19
19
19
19
19
19
20
2O
20
2O
20
20
20
20
20
20
2O
20
20
2O
20
2O
20
20
2O
20
21
21
21
21
22
22
22
23
23
24
23
23
23
23'
23
25
25
25
252
252
152
171
171
171
171
19O
19O
228
80
BO
120
140
140
180
ISO
2OO
2OO
2OO
2OO
2OO
220
22O
240
24O
26O
3OO
3OO
32O
189
189
273
273
198
22O
332
69
276
24O
13O
ISO
ISO
3OO
4OO
400
4OO
4OO
Age«SAV
16
9
9
16
12
TO
3
8
10
14
3
9
23
26
26
26
28
29
3O
3O
30
3O
39
40
400
234
234
416
336
290
ISO
240
3OO
420
117
360
-------
Boil Aggresciveness Value*
-------
Boil Aggr«.«lvene.« Value. (SAV) and age< for non-leaking tanks.
8AV
Ag.
I
to
9
9
2
2
9
9
7
7
13
2
9
9
7
7
10
11
9
9
3
*
10
12
14
16
16
9
9
9
13
13
13
13
14
14
i4
9
9
7
7
It
II
11
16
2
2
3
3
3
3
3
3
3
7
7
7
7
8
8
a
a
a
a
a
a
a
a
.9
9
9
9
9
9
9
9
SAV*Age
IO
18
6
6
19
19
21
21
39
a
20
20
28
28
4O
44
29
29
18
24
24
24
72
28
28
70
84
98
112
112
4O
40
4O
104
104
104
IO4
112
112
112
49
49
63
63
99
99
99
144
SAV
Age SAVftAge
SAV
Agi
6AV«Age
SAV
2
2
10
to
10
3
9
12
12
12
14
14
17
3
4
9
9
12
13
13
13
16
16
16
2
2
2
4
4
6
9
6
6
6
IO
12
2
2
6
6
6
IO
12
12
12
»?
16
16
10
10
IO
10
IO
12
12
12
12
12
12
12
12
12
12
12
13
13
13
13
13
13
14
14
14
14
14
14
19
19
19
19
19
19
15
15
IS
IS
15
IS
20
20
100
1OO
too
33
99
132
132
132
154
194
187
36
48
108
108
144
156
196
156
192
192
192
26
26
26
92
92
78
7O
84
84
84
140
168
3O
30
90
90
90
150
ISO
ISO
ISO
ISO
24O
24O
3
12
5
. 9
12 '.
12
12
12
19
16
6
8
8
a
9
9
9
9
IO
10
10
14
14
14
2
2
2
2
9
9
9
9
9
9
2
2
2
9
9
7
10
IO
10
IO :.
IO
10
IO
10
16
16
17
17
17
17
17
17
17
17
18
IB
18
18
18
18
18
IB
18
IB
18
18
18
18
19
19
19
19
19
19
19
19
19
19
20
20
20
2O
20
20
20
20
20
20
20
20
2O
2O
48
192
85
85
2O4
2O4
204
2O4
255
272
108
144
144
144
162
162
162
162
180
ISO
ISO
292
252
292
38
38
38
38
99
171
171
171
171
171
4O
40
4O
1OO
1OO
140
2OO
20O
2OO
20O
2OO
2OO
20O
2OO
Age SAV*Age
IO
IO
11
13
13
13
13
13
13
17
17
17
17
9
9
13
13
3
IO
IO
IO
12
16
12
12
12
12
12
12
12
12
9
9
9
9
9
16
16
16
16
6
12
2
2
2
7
8
B
2O
2O
20
2O
20
20
20
2O
20
20
20
2O
20
22
22
22
22
24
24
24
24
24
24
25
25
25
25
25
25
25
25
26
26
26
26
26
26
26
26
26
28
28
30
30
30
30
3O
30
2OO
200
220
26O
26O
26O
26O
26O
26O
340
34O
340
340
198
198
286
286
72
240
240
240
288
384
3OO
3OO
3OO
3OO
3OO
3OO
30O
3OO
234
234
234
234
234
416
416
416
416
168
336
6O
6O
60
210
24O
240
-------
TABLE 4
Boil Aggre»*ivene«« Value* (SAV) and age* for non-leaking tanks.
BAV
Agi
I
»-
o
3
9
2
2
9
9
7
7
13
2
9
9
7
7
1O
11
9
9
3
4
4
4
12
4
4
1O
12
14
16
16
9
9
9
13
13
13
13
14
14
14
9
9
7
7
11
11
11
16
2
2
3
3
3
3
3
3
3
4
4
4
4
4
4
4
9
9
6
6
6
6
6
7
7
7
7
7
7
7
B
0
8
a
a
a
a
a
a
a
BAV«Age
10
18
6
6
19
19
21
21
39
a
20
20
28
28
40
44
29
29
18
24
24
24
72
28
28
70
84
98-
112
112
40
40
4O
104
104
104
104
112
112
112
49
49
63
63
99
99
99
144
SAV
Agi
SAVttAga
BAV
Agi
SAV«Age
SAV
Age
SAV*Age
2
2
1O
10
10
3
9
12
12
12
14
14
17
3
4
9
9
12
13
13
13
16
16
16
2
2
2
4
4
6
9
6
6
6
10
12
2
2
6
6
6
1O
12
12
12
12
16
16
to
10
10
10
10
11
II
11
11
11
11
11
11
12
12
12
12
12
12
12
12
12
12
12
13
13
13
13
13
13
14
14
14
14
14
14
19
19
19
19
19
13
IS
19
13
13
IS
IS
20
20
IOO
100
100
33
99
132
132
132
134
134
187
36
48
108
108
144
136
136
136
192
192
192
26
26
26
92
92
78
7O
84
84
84
140
168
30
3O
9O
90
90
ISO
ISO
ISO
180
ISO
240
240
3
12
3
12
12
12
12
. IS
16
6 '
a
8
8
9
9
9
9
IO
10
IO
14
14
14
2
2 '
2
2
3
9
9
9
9
9
2
2
2
9
3
7
IO
10
10
10 .
10
10
10
10
16
16
17
17
17
17
17
17
17
17
18
18
18
18
IB
IB
18
18
18
18
18
18
IB
IB
19
19
19
19
19
19
19
19
19
19
2O
20
20
20
20
2O
2O
2O
20
20
20
20
20
20
48
192
as
as
2O4
204
204
204
253
272
1O8
144
144
144
162
162
162
162
ISO
ISO
ISO
232
232
232
38
38
38
38
93
171
171
171
171
171
40
40
40
100
IOO
14O
200
20O
2OO.
2OO
2OO
2OO
2OO
2OO
IO
10
11
13
13
13
13
13
13
17
17
17
17
9
9
13
13
3
10
10
10
12
16
12
12
12
12
12
12
12
12
9
9
9
9
9
16
16
16
16
6
12
2
2
2
7
a
8
20
20
20
2O
20
20
20
2O
20
20
20
20
20
22
22
22
22
24
24
24
24
24
24
23
25
23
23
23
23
25
23
26
26
26
26
26
26
26
26
26
28
28
30
3O
30
30
30
3O
200
2OO
2 2O
26O
26O
26O
26O
26O
260
340
340
340
340
198
198
286
286
72
240
240
24O
288
384
300
3OO
3OO
30O
3OO
3OO
300
300
234
234
234
234
234
416
416
416
416
168
336
6O
6O
60
210
24O
24O
-------
FIGURE 3: COMPUTATION OF SAV
I BASIC CHARACTERISTICS
o Soil Resistivity
o Soil pH
o Soil Moisture
. <300
300 - 1,000
1,000 - 2,000
2,000 - 5,000
5,000 - 10,000
10,000 25,000
>25,000
3 -
5 -
6.5 -
7.5 -
Saturated
Damp
Dry
<3
5
6.5
7.5
9
>9
POINTS
12
10
8 .
6
3
1
0
8
6
4
2
1
0
3
2
0
II DIFFERENTIAL CHARACTERISTICS
o Resistivity
(ratio of extremes)
Soil pH
(Difference 1n
pH Value)
1.5
0
5
3
3
3
3
1.5
3
2
1
0
2
1
0
III SULFIDES
Positive
Negative
4
0
B-ll
-------
corrosivities; that is, a soil with an SAV of 2x is expected to be (on average)
twice as corrosive as a soil with an SAV of x.
On first consideration, the Canadian data reveals one important fact: there is
a lot of scatter. But it 1s also clear that non-leaking tanks are clustered
somehwat closer to the lower left-hand corner of the diagram than are leaking
tanks. Thus, age, SAV, or the combination of the two does appear to have some
predictive effect on the probability of leakage.
This tentative conclusion can be verified by simple statistical analyses. Using
chi-squared techniques, age, SAV and SAV x age can be tested for statistically
significant effects on the probability of leakage. In addition, SAV can be
tested to determine if 1t has any effect Independent from the effect of age.
The following section will describe each of these analyses.
STATISTICAL TESTS
1. SAV x Age
PACE asserts,that the product of SAV and tank age 1s an appropriate predictor
for tank leakage. This assertion makes Intuitive sense, for 1t would appear to
allow for the continuing effects of various types of soils. It 1s also born out
by the data, as 1s Indicated by the following contingency table:
B-12
-------
TABLE 5
SAV x Age
0-49
50 ~ 99
100 - 149
ISO - 199
200 249
250 - 299
300 - 399
> 400
Number
of
Leaking Tanks
108
Number
of Non-
Leakinq Tanks
0 / 16.92
11 / 13.68
10 / 13.32
36 / 24.12
22 / 18.00
14 / 10. 08
8 / 7.92
7 / 3.96
47 / 30.08
27 / 24.32
27 / 23.68
31 / 42.88
28 / 32.00
14 / 17.92
14 / 14.08
4 / 7.04
192
47
38
37
67
50
28
22
11
300
Each cell of this table contains two numbers. The first represents the observed
observed number of tanks 1n each category; these numbers were obtained from
Tables 1 and 2. The second number 1n each cell represents the expected number
of tanks falling Into that category 1f SAV x Age had no effect on the probabi-
lity of leakage. These numbers are computed by multiplying the row total by the
column total and dividing by the grand total (-300).
One of the assumptions underlying the ch1-squared test Is that the sampled size
1s large enough to allow a large-sample approximation. Often, this assumption
1s expressed as a requirement that there be at least 5 observations In each
cell, but more rigorously, the assumption may be stated as a requirement that
B-13
-------
the expected values be greater than 5 in at least 30% of the 'sample cells.5
This assumption is clearly met.
The test statistic (T) is simply the summation of (x^j - E7-j)2/Ej, over all
cells, where x^j is the number of observations in cell,j and EJJ is the corres-
ponding expected number of observations. For Table 3, T « 45.11, With (r-1)
(c-1) 7 degrees of freedom (r and c are the numbers of rows and columns, res-
pectively). This 1s highly significant, Indicating that there 1s far less than
a .IX chance that the difference between the SAV x Age distributions of leaking'
and non-leaking tanks 1s random. SAV x Age 1s therefore a statistically signi-
ficant factor in the differentiation of leaking and non-leaking tanks. In par-
ticular, 1t appears from Table 5 that SAV x Age has a trichotomous effect. For
very low values of SAV x Age, there were no observed leaking tanks (the raw data
Indicates that for all leaking tanks SAV x Age > 59). For Intermediate values
(50 < SAV x Age < 150), approximately 28X were leaking, and for high values (SAV
x Age > 149), approximately 49X were leaking.
2. Tank Age
The conventional wisdom 1s that age 1s a very poor predictor of .tank leakage.
This, however, 1s an overstatement, as Is evident from the following contingency
table:
5W. J. Conover, Practical NonparametHc Statistics (John Wiley & Sons: New
York), 1971, p. 15?:
B-14
-------
Tank
Age
(yrs)
0-4
5 - 9
10 - 14
15 - 19
20 - 24
25-29
> 29 '
TABLE 6
Number
of
Leaking Tanks
108
Number of
Non-Leaking
Tanks
0 / 5.76
5 / 13.32
12 / 17.28
41 / 31.32
30 / 24.12
14 / 11.88
6 / 4.32
16 / 10.24
32 7.23.68
36 / 30.72
46 / 55.68
37 / 42.88
19 / 21.12
6 / 7.68
192
16
37
48
87
67
33
12
300
For this table, T * 28.17, with 6 degrees of freedom. This 1s significant at
something 1n excess of the 99.9X level. Age therefore 1^ a statistically signi-
ficant determinant of the probability of leakage. It 1s clear from Table 1 that
while some tanks are leaking at ages 5 to 14, they represent a fairly small
fraction (17X) of the entire sample. After age 15, however, the percentage of
leakers Increases to 46%.
Another Interesting observation also emerges from Table 6i 1f a new contingency
table 1s constructed only for tanks of age 15 or higher, there 1s no statisti-
cally significant effect of age upon the probability of leakage:
B-15
-------
Tank
Age
15 - 19
\
20 - 24
25 29
> 29
TABLE 7
Number of
Leaking Tanks
Number of
Non-Leaking Tanks
41 / 39.78
30 / 30.68
14 / 15.09
6 / 5.49
46 / 47.22
37 / 36.36
19 / 17.91
6 / 6.51
91
108
87"
67
33
12
199
For this table, T .» .33, with-3 degrees of freedom. This 1s not significant,
even at the 75* level. Thus, age seems to have an effect only for tanks younger
than 15 years; above that age, the probability of leakage Is apparently con-
stant.
3. SAV
A contingency table can also be set up to test the effect of SAV upon the proba-
bility of leakage:
B-16
-------
SAV
0-4
5-9
10 - 14
> 14
TABLE 8
Number of
Leaking Tanks
Number of
Non-Leaking Tanks
4 / 13.32
28 / 31.68
53 / 47.52
23 / 15.48
33 / 23.6?
60 / 56.32
79 / 84.48
20 / 27.52
37
108
192
132
43
300
For this table, T * 17.55 with 3 degrees of freedom. This Is significant at the
99.9X level. .
4. Interaction of SAV and Age
Unfortunately, Age and SAV are not Independent variables, as 1s shown by Table
9:
TABLE 9
Age of
Tank
Low
(0-6)
Soil Aggressiveness Value
Medium High
(7-12) O12)
0-14 years
15 - 20 years
w
> 20 years
37 / 24.91
26 / 33.05
11 / 16.03
32 / 47.47
72 / 62.98
37 / 30.55
32 / 28.62
36 / 37.97
17 / 18; 42
74
141
85
101
134
65
300
B-17
-------
The T statistic for this table is 17.26 with 4 degrees of freedom. This indi-
cates that there is less than a .5% chance that SAV and Age are uncorrelated.
An examination of Table 9 shows that the strongest correlation appears to occur
for low and medium SAV's. High-SAV tanks are fairly randomly distributed across
all three age groups.
This correlation would be easy to explain if older tanks were more likely common
to be found In low-SAV soils; In that case, the relationship between Age and SAV
would simp-ly be due to a survival factor (a disproportionate number of older
tanks in aggressive soils would already have been replaced long before the sur-
vey was taken). Such, however, 1s not the case. Instead, younger tanks are
more likely to be found in low-SAV soils. This cannot be a survival effect.
Instead, It probably represents a shift in Installation practices in favor of
the less corrosive soils.
The correlation between tank age and SAV makes it difficult to determine which
Is the dominant variable. It Is possible, for example, that the observed
effects of SAV and SAV x Age are actually the effects of Age, transmitted
through the linkages among these variables.
tf
This hypothesis can be tested by constructing contingency tables examining the
effects of SAV upon tank leakage for each of the tank age-groups. This will
reveal whether SAV has any effect Independent from Age.
B-18
-------
TABLE 10
Low SAV
Medium SAV
High SAV
Young Tanks
(T » 12.34, 99.5X significance)
Number of
Leaking Tanks
Number of
Non-Leaking Tanks
1 / 6.23
5 / 5.39
11 / 5.39
36 / 30.77
27 / 26.61
21 / 26.61
17
84
37
32
32
101
Low SAV
Medium SAV
High SAV
Medium-age Tanks
(T - 4.67, 90X significance)
Number of .
Leaking Tanks
Number of
Non-Leaking Tanks
7 / 11.84
35 / 32.78
19 / 16.39
19 / 14.16
37 / 39.22
17 / 19.61
61
73
26
72
36
134
B-19
-------
Low SAV
Medium SAV
High SAV
Old Tanks
(T * 2.44, Insignificant)
Number of
Leaking Tanks
30
Number of
Non-Leaking Tanks
6 / 5.08
14 / 17.08
10 / 7.85
5 / 5.92
23 / 19.92
7 / 9.15
35
11
37
17
65
These tables Indicate that SAV and Age have some Independent effect, but. only
for the younger tanks. Combining the results of Tables 10 and 7, it appears
that neither Age nor SAV have much effect for tanks older than 20 years.
B-20
-------
DATA ANALYSIS
Proving that there are statistically significant linkages between leakage rates
and SAV, Age, and SAV x Age does not conclude the analysis, however, for it is
also necessary to determine the correct interpretation of these linkages. The
ultimate goal of such an interpretation is to deduce a cumulative probability
distribution for tank failure over a range of SAV x Age categories.
Unfortunately, this 1s not a straightforward task.
i
One approach to this problem is to assume that the data actually represent the
desired cumulative distributions. This assumption would be correct if leaking
tanks were never repaired or replaced, for in that case, the number of leaking
tanks in any SAV x Age bracket would include both new and pre-existing leaks.
Under this simple assumption, the Canadian data yields the following distribu-
tion (obtained from the numbers in Table 5):
TABLE 11
Cumulative
Leak Probability
SAV x Age . (X)
0-49 0
50-99 28.9,
100-149 27.0
150-199 53.7
200-249 44.0
250-299 50.0
300-399 36.4
> 400 63.6
Since a cumulative probability distribution Is by definition non-decreasing, the
fluctuation In 300-399 category must be assumed to be anomolous. It can be
reduced by combining the two highest brackets: *
B-21
-------
TABLE 12
Cumulative
Leak Probability
SAV x Age X
0-49 0
50-99 28.9
100-149 . 27.0
150-199 53.7
200-249 .44.0
250-299 50.0
> 300 45.0
This distribution reveals that tank failure occurs in two spurts: one at SAV x
Age between 50 and 99, and the other at SAV x Age between 150 and 199. Other
fluctuations in failure rates are statistically insignificant, as-can be
verified by constructing the appropriate contingency tables.-
This distribution has the advantage that It conforms to the expected sigmoidal
pattern, with most of the failures occurring during the middle brackets and with
some tanks which are effectively immortal even in highly aggressive soils, but
the numbers in Table 12 do not seem appropriate. Leakage should not occur at
such tightly defined intervals; i.e. there are too few leakers In the 100-149
category. Even more Importantly, there are too many Immortals. It is very
unlikely that half of the tanks would still survive after 40 years In a soil of
SAV « 10 (or 20 years with SAV » 20). Yet that is what this distribution seems
to indicate.
The problem with the distribution in Table 12 is simple: some leaking tanks
will have been replaced relatively soon after they began to leak. Thus, in the
upper brackets the survey self-selects for norr-leaking tanks (since they are
more likely to still be in use), and the cumulative percentages are too low.
One way.to cure this problem would be to convert the survey data Into a cumula-
tive distribution by computing the number of missing tanks. This could be done
by determining the relative numbers of tanks burled in each year and assigning
these to SAV brackets according to the distributions in Table 10.6 To the
*>Th1s apportionment assumes that the missing tanks in each age group followed the
B-22
-------
extent that Table 11 underrepresents certain SAV x Age groups, it can then be
assumed that the missing tanks are those which have previously leaked, and the
cumulative probabilities can be adjusted accordingly. In order to normalize
these calculations, 1t must be assumed that one SAV x Age-bracket Is fully
represented. Presumably, this would be bracket 0-49.
Unfortunately, this analysis Is fraught with difficulties. Not only are the
calculations complex (and somewhat recursive), but the necessary tank burial
data 1s not available. Instead, the best alternative are three data sources on
service station construction, and even these are difficult to obtain for Canada
(U.S. data are presented in Table 13). Furthermore, once the data are
assembled, It appears that the Canadian survey underrepresents younger tanks.
This Indicates either serious problems in the application of U.S. service sta-
tion data to Canadian tank burials, or .it Indicates that the Canadian survey was
not random, but Instead favored older tanks. In either event, tank attrition
cannot be computed, and another approach .must be used to obtain a more reaso-
nable cumulative distribution.
A simpler approach to the problem of self-selection of older non-leaking tanks
may be found by varying the assumptions underlying Table 12. Instead of
assuming that leaks are never detected before the survey, it can be assumed
instead that all leaks are detected and the tanks replaced before enough time
has passed for the tank to move Into the next SAV x Age category. This assump-
tion requires the detection period to be Inversely proportional to SAV, but that
requirement would be sensible If monitoring 1s better for tanks known to be in
more aggressive soils.
Under this assumption, the failure rates 1n Table 12 become elements of a proba-
bility density function and the corresponding cumulative distribution may be
calculated.
same SAV distributions as their surviving kin. Such an assumption 1s probably
not accurate, but it 1s better than nothing.
B-23
-------
TABLE 131
ro
Date
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972 |
Hunter of
Service
Stations2
179,647
»
180,347
180,697
181,040
181.390
181.747*
188.100
194,600
200,100
206.755
207,800
208,800
209,700
210,600
211,473*
212.600
213.550
214.500 ,
216,059*
219,100
222.200
222.000
220,000
fc 226, 459*
Change In API Reports3
Nunter of of New
Service Service
Stations Stations
350
350
350
350
343
350
357
6353
6500
5500
6655
1045
. 1000
900
900
873
1127
950
950
1559
3041 * 3740
3100
(200) 2508
(2000) 2068
6459 1689
Ratio of
Rehabilitations
API Building* to new
Deacti vat Ions3 Permits Constructions5
9.021
9.826
10,615
5,391
7,801
8.050
6,080
6,150
6,500
6,275
6,606
4554 6000-7000
6,200
3586
3630
34*
24.30
0.55
0.63
0.00
0.17
6.70
5.96
4.46
5.84
5.60
3.24
1.14
1.00
Number of
Rehabilitations6
8664
3473
4115
(109)
1146
7005
5207
5023
5550
5325
5047
3500
3100
-------
TABLE 13 (Continued)
Date
Number of
Service
Stations
Change in
Number of
Service
Stations
API Reports
of New
Service
Stations
API
Deactivations
Building
Permits
Ratio of
Rehabilitations
to new
Constructions
Number of
Rehabilitations
ro
in
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
215.880
196.130
189.480
186.340
176.400*
172,300
164,790
158,540
151,250
147.000
(579)
(19.750)
(6.650)
(3.140)
(9.940)
(4.100)
(7.510)
(6.250)
(7.290)
(4.250)
1172
206
206
319
'284
353
286
169
297
9342
6041
4127
5676
5683
5138
3724
3380
4273
9.0
3.8
3.03
6.16
8.34
9.68
1854
783
967
1749
2385
1636
* Source: National Petroleum News, annual Fact Book of petroleum statistics.
2 Obtained by National Petroleum News (NPN) from U.S. government data and NPN
estimates. Numbers marked by a star (*) are census totals.
3 Obtained by NPN from American Petroleum Institute totals for a somewhat varying number of responding companies.
^ Obtained by NPN from the Department of Labor, Bureau of Labor Statistics.
5 From 1974 through 1980, these are obtained by NPN from an API survey. For 1954 through 1969, these numbers are
obtained from columns 3 and 8. . .
6 For 1954 through 1969, these numbers are the difference between the number of building permits Issued and the change
in the total number of service stations. For 1974 through 1980, these numbers are obtained from columns 4 and 7.
-------
TABLE 14
Probability of Cumulative Failure
SAV x Age New Failure (X) Probability (X)
0-49 0 0
50-99 28.9 .28.9
100-149 27.0 46.7
150-199 53.7 75.3
200-249 44.0 86.2
250-299 50.0 93.1
300-399> 36.4 95.6
> 400" 63.6 98.4
Unfortunately, these cumulative probabilities seem to be too high. Anecdotal
data would seem to Indicate that a larger percentage of tanks have extremely
long lifetimes, even 1n very aggressive soils. Furthermore, 1t 1s likely that
some leaks remain undetected over very, long periods of time. For both of these
reasons, the calculated failure rates are probably too high, though they pro-
bably can safely be used as upper bounds on the actual probabilities.?
«
tf
A third assumption may be used 1n an effort to split the difference between the
two polar cases discussed above: 1t may be assumed that SOX of the leaks
detected 1n each bracket are new, while the remaining leaks are ones that have
been continuing since a previous bracket. Under this assumption,.the first two
brackets are unaltered,** but for the other brackets, the previously leaking
7There 1s another, more theoretical problem with Table 14: the cumulative proba-
bility distribution 1s highly dependent on the width of the SAV x Age brackets
used 1n Its computation. Decreasing their width Increases the number of cate-
gories (without significantly changing the second column of the table), thereby
causing the cumulative probability to converge, upon 1.0 at a considerably more
rapid rate. Broadening the categories has the reverse effect.
In non-mathematical terms, this problem Is related to the detection-period
problem discussed earlier. If leaks are detected rapidly, then the data Indica-
tes a high rate of new leak formation. If leaks are detected slowly, then the
cumulative probability approaches the no-detection assumption depicted 1n Table
11. Since these problems relate to the proper Interpretation of real-world
data, they are not the same as the scenarios ultimately to be studied 1n the
computer model. Instead, the goal Is to determine just how conservative the
Canadian oil companies1 detection/repair policies actually were prior to 1977.
The use of 50-po1nt brackets seem to be a reasonable assumption for the tech-
nology then 1n use.
8The first bracket Is unaltered because there are no leaks. The second bracket
B-26
-------
tanks must be removed from both the leaking tank category and the bracket
totals. With these modifications, the data becomes:
1s unaltered because to alter 1t would be contrary to the observation that there
are no failures 1n the first bracket.
B-27
-------
TABLE 15
SAV x Age
0-49 ,
50-99
100-149
150-199
200-249
250-299
300-399
> 400
Number of
Leaking Tanks
0
11
5
18
11
7
4
3.5
Number of
Previously Non-
Leaking Tanks
47
38-
- 32
49
39
21
18
7.5
Percent
of New
Leaks
0
28.9
15.6
36.
28,
33,
22.2
50.0
,7
,2
,3
Cumulative
Leak
Probability
(X)
0
28.9
40.0
62.0
72.7
81.8
86.0
93.0
These cumulative probabilities from these three assumptions can be combined in
a single table:
TABLE 16
SAV x Age
0-49
50-99
100-149
150-199
200-249
250-299
300-399
> 400
Lower
Bound
(Assumption 1)
0
28.9
27.0
53.7
44.0
50.0
45.0
45.0
Upper
Bound
(Assumption 2)
0
28,
46,
75,
86,
93,
95.6
98.4
Assumption
3
0
28.9
40.0
62.0
72.7
81.8
86.0
93.0
Average of
Assumption
1-3
0
28,
37,
63.
67.6
75.0
75.5
78.8
The last column 1s the one which will be used fn the Monte Carlo model, though
It 1s subject to revisions as better data become available.
CAVEATS
The cumulative probability distribution presented 1n Table 16 must be used with
caution, for unfortunately, the Canadian data set does not represent a random
B-28
-------
survey of existing tanks. Instead, the data were collected in 3 ways:9
o Over a 6-month period In 1977, all PACE member companies were requested
to report leak Incidents. Soil samples were taken at the leaking tanks'
sites.
o During this same time period, PACE member companies were requested to
report tank decommissioning*. Decommissioned tanks were then tested for
leaks, and soil samples were taken.
o Other tanks on the same site as a leaking or decommissioned tank were
also tested.
Thus, the survey is biased both toward leaking tanks and toward older tanks.
(The latter bias occurs because older tanks are more likely to be
decommissioned). The age bias Is relatively unimportant. The bias toward
leaking tanks, however, means that the resulting data present a worst-case
portrait of the existing tank situation. This bias may not be overly severe,
however, for the fact that only 36X of the sample tanks were leaking Indicates
that the other two sampling methods may have predominated. Furthermore, this
bias may be offset by the fact that the second and third sampling techniques
tend to self-select for non-leaking tanks. Nevertheless, an unknown net bias
probably results, and the data must be viewed as only an approximation of the
results of a truly random survey.
AN ALTERNATIVE APPROACH
Instead of complex calculations based on Age x SAV, it may be more appropriate,
given the data biases discussed above, to attempt a simpler model. With this
in mind, the data can be grouped Into high, middle, and low SAV soils, and
failure rate versus age may be calculated for each soil group. The results are
presented 1n Table 17.
^PACE, "Underground Tank Systems," supra ru-4?-p;-43, and personal communication
with J.R. Clendenlng, Esso Petroleum. Canada, June 1985.
B-29
-------
TABLE 17
SAV Medium SAV High SAV
6> (7-12) (> 13)
Tank Age Leakers Non Leakers Leakers Non Leakers
0-4
5-9
10-14
15-19
20-24
> 25
* 0
0
1
4
4
_5
14
8
13
15
14
6
_.* . ~
60
0
1
4
22
19
' _8
54
7
8
12
25
18
17
87
0
4
7
15
7 .
_7
40
1
11
9
7
13
" JL:-
45
B-30
-------
Table 17 may be Interpreted under the same 3 assumptions that were used in
Tables 11, 12, and 14-16. The resulting probability distributions are presented
in Tables 18-20.
The same approach can also be used employing only two SAV categories. If soils
are classified as benign when SAV 1s 9 or less and aggressive when SAV 1s 10 or
greater, then the data can be summarized in Table 21.
Table 21 can be used to calculate cumulative probability distributions as in
Tables 18-20. The results, using the same three assumptions, are presented in
Tables 22-23.
The 2-part and 3-part SAV distinctions have certain similarities. In both
cases, there are clear differences between soils of different aggresslvttles.
These differences can most readily be appreciated by presenting the results in a
single table, as is done in Table 24.
These distributions can be plotted graphically, as can the probability of tank
failure versus.-SAV x Age (from Table 16). This 1s done 1n Figures 4-9. For
Interpretive purposes, these graphs have converted the cumulative distributions
reported in the tables Into the underlying probability densities. Thus, these
histograms represent the probability that the tank failure will originate in
each of the designated Intervals.
These graphs Indicate that SAV x Age. 1s probably not the best measure of tank
deterioration. The reason for this conclusion 1s the bimodal nature of the SAV
x Age probability density. While such bimodal1ty might possibly be an accurate
reflection of the real world, 1t Is more likely that the bimodal distribution
results from Improperly aggregating unlike distributions. This latter explana-
tion appears particularly appropriate In the present situation. As Figures 4-6
Indicate, the probability distributions are differently-shaped for low-SAV and
Mgh-SAV soils. Low-SAV soils produce a relatively steady failure rate for all
years after year 9, while Mgher-SAV soils produce much higher failure rates in
the lower years, but declining failure rates 1n later years.W Combining these
10The reason for the low failure rates after year 19 1s simply that by that year,
a large fraction of h1gh-SAV tanks have already failed.
B-31
-------
TABLE 18. LOW SAV SOILS (SAV < 6)
Cumulative Probability of Leakage (X)
Lower Bound Upper Bound Assunption 31
Tank (Detection within (No detection (75* of survey
Age 5 years) prior to survey) leaks are new) Average
0-4 0 0 0 0
5-90 0 0 0
10-14 " 6.3 6.3 6.3 6.3
15-19 22.2 27.1 22.8 24.0
20-24 40.0 56.3 48.5 48.3
> 25 55.6 80.6 73.6 69.9
1 Assumption 3 in this table has been adjusted from that used In Tables 15 and
16 in order to be consistent with the lower leak rates for low-SAV tanks.
TABLE 19. MEDIAN SAV SOILS (7-12)
Cumulative Probability of Leakage (%)
Lower Bound Upper Bound Assumption 3*
Tank (Detection within (No detection (SOX of survey
Age 5 years) prior to survey) leaks are new Average
0-40 0 00
5-9 11.1 11.1 11.1 11.1
10-14 25.0 33.3 28.91 29.1
15-19 46.8 64.5 50.6 54.0
20-24 51.4 82.8 67.7 67.3,
> 25 32.0 88.3 73.8 - 2
1 This calculation uses a 75X rate of new leak development 1n order to be
consistent with the lower-bound estimates in the previous column.
2 No number Is calculated for this range, for the anomalous decline'In pro-
bability for the "lower bound" would produce an equally anomalous fluctuation
in the average.
B-32
-------
TABLE 20. HIGH SAV SOILS (SAV > 13)
Cumulative Probability of Leakage (X)
Lower Bound Upper Bound Assumption 3
Tank (Detection within (No detection (SOX of survey
Age 5 years) prior to survey leaks are new Average
0-40 0 0 0
5-9 26.7 26.7 26.7 26.7
10-14 ' 43.8 58.8 47.2 49.9
15-19 68.2 86.9 74.5 76.5
20-24 35.0 91.5 79.9 79.91
> 25 63.6 96.9 89.3 83.3
1 Obtained by Interpolation between the values for ages 15-19 and > 25. An
average of Assumptions 1, 2, and 3 is dominated by the anomalous~~value for
Assumption 1. .
TABLE 21
Benign Soils (SAV < 9) Aggressive Soils (SAV > 10)
Number of Number of Number of Number of Uon-
Age Leaking Tanks Non-leaking Tanks Leaking Tanks Leaking Tanks
0-40 13 03
5-9 0 15 5 17
10-14 .1 18 11 18
15-19 11 26 30 20
20-24 11 9 19 28
> 25 9 12 11 13
8-33
-------
TABLE 22. BENIGN SOILS (SAV < 9)
Cumulative Probability of Leakage (X)
Upper Bound Assumption Average
Lower Bound (Assuming all 3 (67X of of
(Assuming no leaks detected observed Assumptions
Tank detection prior and repaired within leaks are 1, 2, and
Age to survey 5-year age bracket) new) 3
0-40 0 00
5-90 0 00
10-14 5.3 5.3 5.3, 5.3
15-19 29.7 33.4 32.4* 31.8
20-24 55.0 70.0 62.8 62.6
> 25 42.9 82.9 75.2 67.0
1 Calculated under the assumption that only one of the observed leakages was
pre-existing, in order to be consistent with the preceding bracket's low-leak
rate.
TABLE 23. AGGRESSIVE SOILS (SAV > 10)
Cumulative Probability of Leakage (X)
Upper Bound Assumption
Lower Bound (Assuming all 3 (50% of
(Assuming no leaks repaired observed
Tank detection prior within 5-year leaks are
Age to survey age bracket) new) Average
0-40 0 00
5-9 22.7 22.7 22.7, 22.7
10-14 37.9 52.0 48.I* 45.0
15-19 60.0 80.8 70.3 70.4
20-24 40.4 88.6 77.8 72.62
> 25 45.8 93.8 84.4 74.7
* Calculated under the assumption that SOX of the observed leaks are new, in
order to be consistent with the previous bracket's observed low leak rate.
2 Calculated by Interpolation between the preceding and following brackets, in
order to prevent the decline In the value for assumption 1 from causing
anomalous results.
B-34
-------
TABLE 24.
Cumulative Probability of Leakage (X)
Tank
Age1
4
9
14
19
24
> 25
Low SAV
(0 to 6)
0
0
, 6.3
24.0
48.3
69.9
Medium SAV
(7 to 12)
0
11.1
29.1
54.0
67.3
-
High SAV
(> 13)
0
26.7
49.9
76.5
79.9
83.3
Benign Soil
(0-9)
0
0
5.3
31.8
62.6
67.0
Aggressive
Soil
(> 10)
0
22.7
45.0 '
70.4
72.6
74.7
Tank ages have been changed from age brackets to the age corresponding to the
top of each bracket.
B-35
-------
25 T
20
FAILURE
PROBABLITY
FIGURE 4
LOV SAV SOLS (0-6)
0-4 3-9 10-14 15-19 20-24
-------
FAILURE
PROBABILITY
FIGURES
MEDUMSAV SOLS (7-12)
0-4 5-9 10-14 15-19 20-24 >24
TANK ACE
B-37
-------
' 20
FAILURE
PROBABILITY
FIGURES
HIGH SAV SOLS (13-21)
10-14 13-19
TANKAGE
B-38
-------
FAILURE
PROBABILITY
10
FIGURE 7
BENIGN $OLS (0-9)
0-4 5-9 10-14 15-19 20-24 >24
B-39
-------
FAILURE
PR08ABLITY
FIGURE 8
AGGRESSIVE SOLS(> 10)
5-9 10-14 15-19 20-24 >24
B-40
-------
FIGURE 9
FAIURE PROBABLITY USING SAV X ACE
FAILURE
PROBABLITY
30
20
13
10
0-49 50-99 100- ISO- 200- 250- 300- XOO
149 199 249 299 399
B-41
-------
two dissimilar age distributions could easily produce the bimodal distribution
depicted in Figure 9, even after the data have been converted from age cate-
gories to SAV x Age categories. It therefore appears that the data has greater
usefulness if SAV and Age are both used as separated variables than it
does if they are combined Into the single variable of SAV x Age.
This analysis, however, does not Indie-ate whether two or three SAV categories
are preferable. This decision can be facilitated, though, by a re-examination
of the raw data in Figures 1 and 2. These scatter diagrams indicate that a
substantial percentage of the tanks, particularly leaking ones, are to be found
between SAV»9 and SAV»11. There 1s no theoretical reason for dividing
aggressive soils from benign soils at an SAV of either 9, 10, or 11, yet because
of the clustering of the data, this arbitrary division can significantly .alter
the probability density functions when only two SAV categories are used. Thus,
the natural clustering of the data favors the use of three SAV categories, and
therefore three such categories will be used in the computer model.
COMPUTER MODELING
In order to carry out the Monte Carlo simulation on a year-by-year basis, it is
necessary to calculate failure probabilities for each year between 1 and 20.
This can most conveniently be done by straight-line Interpolation between the
age brackets used in Table 24. The results are presented in Tables 25 and 26.
(Table 25 presents cumulative probabilities, while Table 26 presents probabili-
ty densities). Once SAV 1s determined, these tables can then be used to deter-
mine annual probabilities of failure.
SAV can be determined 1n one of two ways: 1t can either be postulated as an
exogenous parameter, or 1t can be determined stochastically. The deterministic
approach.1s the simplest, and 1s to be preferred for the Initial simulations,
but the stochastic approach may be useful for modeling more complex'scenarios.
The PACE data can be used to obtain a distribution of SAV's for the 300 tanks
covered by the survey. This distribution 1s presented in Table 27.
While there Is no guarantee that this distribution 1s representative of U.S.
soils, 1t 1s probably a reasonable approximation, and it can be used to calcu-
B-42
-------
TABLE 25. CUMULATIVE PROBABILITIES OF FAILURE IN LOW-,
MEDIUM-, AND HIGH-SAV SOILS
Cumulative Failure Probability (%)
Tank
Age
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
IS
19
20
Low
SAV
(0-6)
0
0
0
0
0
0
0
0
0
1.26
2.52
3.78
5.04
6.30
9.84
13.38
16.92
20.46
24.00
28.86
Medium
SAV
(7-12)
0
0
0
0
2.22
4.44
6.66
8.88
11.10
14.70
18.30
21.90
25.50
29.10
34.08
39.06
44.04
49.02
54.00
56.66
High
SAV
(> 13)
0
0
0
0
5.34
10.68
16.02
21.36
26.70
31.34
35.98
40.62
45.26
49.90
55.22
60.54
65.86
71.18-
76.50
77.18
B-43
-------
TABLE 26. PROBABILITY DENSITIES FOR FAILURE IN LOW-,
MEDIUM-, AND HIGH-SAV SOILS
Probability Density (X)
Tank
Age
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15.
16
17
IS
19
20
Low SAV
(0-6)
0
0
0
0
0
0
0
0
0
1.26
1.26
1.26
1.26
1.26
3.54
3.54
3.54
3.54
3.54
4.86
Medium SAV
(7-12)
0
0
0
0
2.22
2.22
2.22
2.22
2.22
3.60
3.60
3.60
3.60
3.60
4.98
4.98
4.98
4.98
4.98
2.66
High SAV
(> 13)
0
0
0
0
5.34
5.34
5.34
5.34
5.34
4.64
4.64
4.64
4.64
4.64
5.32
5.32
5.32
5.32
5.32
0.68
B-44
-------
TABLE 27. DISTRIBUTION OF SOIL AGGRESSIVENESS VALUES
AV Number of Tanks
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
0
. 20
7
10
21
16
10
7
34
43
12
35
22
20
7
28
6
1
0
0
1
Probability
Density
0
6.7
. 2.3
3.3
7.0
5.3
3.3
2.3
11,3
. 14.3
4.0
11.7
7.3
6.7
2.3
9.3
2.0
.3
0
0
.3
Cumulative
Probability «i
0
6.7
9.0
12.3
19.3
24.7
28.0
30.3
41.7
56.0 .
60.0
71.7
79.0
85.7
88.0
97.3
99.3
99.7
99.7
99.7
100.0
B-45
-------
TABLE 28. CONDITIONAL SAV DISTRIBUTION UNDER A REQUIREMENT
THAT SAV NOT EXCEED 10
Probability Density Cumulative Probability
SAV (X) (XJ
1 0 0
2 11.9 11.9
3 4.2 16.1
4 5.9 22.0
5 12.5 34.5
6 9.5 44.0
7 5.9 . 50.0
8 4.2 ' 54.1
9 20.2 74.4
10 25.6 100.0
B-46
-------
late the probability that any given tank falls into each of the three SAV cate-
gories used in Tables 25 and 26. In addition, the SAV distribution can be used
to obtain the conditional SAV-distribution under various regulatory scenarios.
Consider, for example, a regulation requiring that SAV not exceed 10. If this
regulation has no effect on the distribution of acceptable SAV's, then the con-
ditional SAV distribution can be obtained simply by dividing the numbers in
Table 27 by 56X (the unconditional probability that SAV < 10). The results are
presented in Table 28.
Similar computations could be undertaken for any other SAV cut-off. These
results could then be used to determine the probability that the tank falls in
each of the three SAV-categories used-to predict the probability of failure.
More complex regulatory scenarios could also be modeled under this approach.
For example, a proposed regulation might make the use of cathodic protection or
secondary containment dependent on the aggressiveness of the soil in question.
This could be modeled by first sampling a value for SAV and then using that
value to determine other system parameters. Such a scenario is considerably
more complicated than the scenarios that have been modeled to date, but if it is
desired that such composite scenarios be studied, they are well within the capa-
bilities of the model.
' B-47
-------
APPENDIX C
TANK FAILURE CASE STUDIES
-------
TANK FAILURE CASE STUDY #1
SITE: Transformer manufacturing plant of Federal Pioneer ltd
LOCATION; Regina, Saskatchewan, Canada
RELEASE MECHANISM; Pipe rupture
DATA SOURCE
*x
"A Case Study of a Spill of Industrial Chemicals: Polychlorinated
Biphenyls and Chlorinated Benzenes," National Research Council Canada,
NRCC No. 17586, 1980.
DESCRIPTION OF RELEASE
In mid-1976 an underground pipe carrying PCB's from a 31,000 liter tank
ruptured.
CAUSE OF RELEASE; Underground pipe ruptured
RELEASE MATERIALS
How Detected:
Material Types: PCB's (70%)t chlorobenzenes (30X)
0 Quantities Released: 6800 - 21,000 liters
RELEASE DURATION
RELEASE ENVIRONMENT
Land, Water, Air, Unknown: land
0 Description of contamination: underground
COMMENTS
0 Additional Information on this site can be found In:
Roberts, Russell J., John A. Cherry, Franklin W. Schwartz, "A Case
Study of a Chemical Spill: Polychlorinated Blphenyl, (PCB's)1.
History, Distribution, and Surface TransTocation," In Water
Resources Research, Vol. 18, No. 3, pp. 525-534, June 1982.
C-l
-------
Roberts, Russell J., John A. Cherry, Franklin W. Schwartz "A Case
Study of a Chemical Spill: Polychlorinated Biphenyls (PCB's)--2
Hydrological Conditions and Containment Migration," in Water
Resource Research. Vol. 18, No. 3, pp. 535-545, June 195?:
C-2
-------
TANK FAILURE CASE STUDY #2
SITE; Unknown
LOCATION; Unknown
RELEASE MECHANISM; Catastrohlc release
DATA SOURCE
Dartnell Jr., R.C., T.A. Ventrone, "Explosion of a Para-Nltro-Meta-
Cresol Unit," Chemical Engineering Progress, Vol. 67, No. 6, pp. 58-61."
DESCRIPTION OF RELEASE
t A temperature Indicator on the feed-tank Indicated a temperature of 154°
C for the entire holding period up to the time of explosion. This was ~
also the temperature of the product leaving the process step Immediately
upstream. Prior to the explosion, the pressure on the feed tank
Increased from 40 to 10056. No product was being fed.
CAUSE OF RELEASE; Explosion
RELEASE MATERIALS
How Detected: pressure sensor
Material Types: para-nltro-meta-cresol (PNMC)
Quantities Released: 1500 gallons
RELEASE DURATION
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
Equipment Material: stainless steel
C-3
-------
TANK FAILURE CASE STUDY #3
SITE; Unknown
LOCATION; Unknown
RELEASE MECHANISM; Unknown, probably tank corrosion
v
DATA SOURCE
0 Eagen Jr., H.B., et al. "Removal of Hazardous Fluid from the Groundwater'.
1n a Congested AreaA Case History," Control of Hazardous Material
Spills, Proceedings of 1976 National Conference on Controls of Hazardous
Material Spills.
DESCRIPTION OF RELEASE .
Hydrocarbon migrated through the top of a shallow water table. It
seeped at the land surface 1n low lying areas discharging 200 gallons
per day Into a perennial stream. Domestic wells were abandoned and
product seeped Into sewer lines.
CAUSE OF RELEASE; Unknown
RELEASE MATERIALS
t How Detected:
0 Material Types: Hydrocarbon product (80% gasoline)
0 Quantities Released: 500,000 gallons
RELEASE DURATION "long period of time"
RELEASE ENVIRONMENT
0 Land, Water, A1r, Unknown: water
0 Description of contamination: underground Into groundwater and from
groundwater Into surface water
COMMENTS
0 This paper deals mostly with recovery operations and doesn't describe
the failure event very well.
C-4
-------
TANK FAILURE CASE STUDY #5
SITE; Bulk Terminals, tank storage farm
LOCATION Calumet Harbor Area, Chicago, Illinois
RELEASE MECHANISM; Pipe rupture
DATA SOURCE
Hampson, T.R. "Chemical Leak at a Bulk Terminals Tank Farm," Control of
Hazardous Material Spills. Proceedings of 1976 National Conference on
Control of Hazardous Material Spills
DESCRIPTION OF RELEASE
Silicon tetrachloride leaked from a pipe rupture, forming.an add cloud
with the moist air. A rain storm worsened the situation, causing such
dense fumes, that electrical.lines and transformers corroded and failed:
CAUSE OF RELEASE
A block''valve on an Inlet line and a pressure relief valve were inadver-
tently closed. Pressure in the line began to build up. At about 12:30
p.m. on April 26, 1974, a flexible coupling on the Inlet line burst
under the pressure. The entire piping system shifted and a second line
also cracked.
RELEASE MATERIALS
How Detected: fumes
Material Types: silicon tetrachloride
t Quantities Released: plume contained 40 ppm of HC1; 284,000 gallons
were leaked; Initially the acid cloud was about .25 miles wide, 1000
to 1500 feet high, and 1 mile in length, but due to the storm, it grew
to 9 miles In length.
RELEASE DURATION
It was 2.5 to 3 days before leak was sealed. However, 7 days passed
before there was no threat of additional releases.
C-5
-------
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage, steel tank with dry air or nitrogen.padding equip-
ment, capacity of 1,500,000 gallons. The tank contained 750,000
gallons of fluid.
Tank/Treatment Components, Ancillary Equipment: extra -automatic
pressure vents, special valves, and closed transfer pumps.
Equipment That Failed: flexible coupling, piping system, tank
Dikes/Berms: present, they contained the liquid spill.
RELEASE ENVIRONMENT
"V
Land, Water, Air, Unknown: air and land
Description of contamination: silicon tetrachlorlde poured out of tank
Into diked area and reacted vigorously with water In the air and rain-
fall to form HC1 vapor. The enormous acid cloud spread over the far
south side of the city.
COMMENTS
Additional information on this site 1s located under:
- Hoyle, W.C. and Melvln, 6.L. "A Toxic Substance Leak in Retrospect:
Prevention and Response." Control of Hazardous Material Spills,
Proceedings of 1976 National Conference on Control of Hazardous
Material Spills.
C-6
-------
TANK FAILURE CASE STUDY #6
SITE; Service station
LOCATION; Cresskill, New Jersey
RELEASE MECHANISM; Tank corrosion
DATE SOURCE
Kramer, William H. "Ground-water Pollution from Gasoline," GWMR.
Spring 1982, pp. 18-22.
CAUSE OF RELEASE; Leaks in four 4000-gallon steel tanks due to corrosion.
RELEASE MATERIALS
How Detected: routine inventory check
Material Types: gasoline
Quantities Released: 1200 gallons
RELEASE DURATION two or three days
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: gasoline storage for service station
Equipment That Failed: tank
Equipment Material: steel
Equipment Age: 17 years
RELEASE ENVIRONMENT
Land, Water, Air, Unknown: land
Description of contamination: Underground
COMMENTS
.
t The article 1s very detailed; It Indicates how the gasoline was reco-
vered and the cost of recovery.
C-7
-------
TANK FAILURE CASE STUDY #7
SITE: Unknown
LOCATION; Unknown
RELEASE MECHANISM; Interior tank corrosion
n.
DATA SOURCE
Bosich, Joseph F. Corrosion Prevention for Practicing Engineers. Barnes
and Noble, Inc. New York, 1970, p. 186.
DESCRIPTION OF RELEASE
CAUSE OF RELEASE
A workman accidentally dropped a 1" diameter hexagon-shaped nut to the
bottom of the tank, causing localized Interior corrosion.
RELEASE MATERIALS
How Detected: visual detection
Material Types: concentrated sulfurlc acid
Quantities Released: Unknown
RELEASE DURATION
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
Equipment Material: steel
-------
TANK FAILURE CASE STUDY #8
SITE: Unknown
LOCATION; Sussex, Wisconsin
RELEASE MECHANISM; Unknown, probably tank corrosion
^
DATA SOURCE .
Llndoff, David E., Keros Cartwrlght, Groundwater Contaminat1 on;
Problems and Remedial Actions, Environmental Geology Notes, Illinois
State geological Survey, May 1977, No. 81, Case history 88, p. 50.
DESCRIPTION OF RELEASE
CAUSE OF RELEASE
RELEASE MATERIALS
How Detected: complaints that water from some wells tasted and smelled
like petroleum
Material Types: petroleum products
Quantities Released:
RELEASE DURATION
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
RELEASE ENVIRONMENT
Description of contamination: surflclal material, surface water, and
probably groundwater
C-9
-------
TANK FAILURE CASE STUDY #9
SITE: Unknown
LOCATION; Spring Mills, Pennsylvania
RELEASE MECHANISM; Unknown, probably corrosion or rupture
DATA SOURCE
Llndoff, David E., Keros Cartwrlght, Ground-Water Contamination;
Problems and Remedial Actions, Environmental Geology Notes, Illinois
State Geological Survey, May 1977, No. 81, Case History 24, p. 30.
DESCRIPTION OF RELEASE
CAUSE OF RELEASE: Storage tank leak
RELEASE MATERIALS
How. Detected: explosion
Material Types: gasoline
Quantities Released: 200-250 gallons
RELEASE DURATION; 2 weeks
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
RELEASE ENVIRONMENT
Land, Water, Air, Unknown: water
Description of contamination: groundwater
C-10
-------
TANK FAILURE CASE STUDY #10
SITE; Unknown
LOCATION; Southeastern Pennsylvania
RELEASE MECHANISM: Unknown, probably tank corrosion
i
DATA SOURCE
Llndoff, David E., Keros Cartwrlght, Ground-Water Contamination;
Probems and Remedial Actions, Environmental Geology Notes, Illinois
State Geology Survey, May 1977, No. 81, p. 35, Case History 19.
DESCRIPTION OF RELEASE
CAUSE OF RELEASE; Leak 1n burled 10,000-gallon -tank
RELEASE MATERIALS
How Detected: appeared in a stream
t Material Types: fuel oil
Quantities Released: 60,000 gallons
RELEASE DURATION
TANK DESIGN AND OPERATING CHARACTERISTICS
t Tank Use: storage
RELEASE ENVIRONMENT
Land, Water, A1r, Unknown: water
Description of contamination: surface and groundwater
c-n
-------
TANK FAILURE CASE STUDY #11
SITE; Unknown
LOCATION; Unknown
RELEASE MECHANISM; Unknown, probably tank corrosion or rupture
"t.
DATA SOURCE
Undorff, David E., CartwMght, Keros; Ground-Water Contamination;
Problems and Remedial Actions, Illinois State Geological Survey
Environmental Geology Notes, May 1977, No. 81, case history 5, p. 32.
DESCRIPTION OF RELEASE
A leak was discovered 1n a gasoline storage tank at a service stast1on.~~
Further Investigation Indicated that several thousand gallons of gaso-
line had been lost over a period of three weeks
CAUSE OF RELEASE; Leak 1n storage tank
RELEASE MATERIALS
How Detected: fumes in nearby houses
Material Types: gasolines
Quantities Released: several thousand gallons
RELEASE DURATION 3 weeks
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
RELEASE ENVIRONMENT
Land, Water, Air, Unknown: groundwater, explosive concentrations of fumes
In four houses
t Description of contamination:
C-12
-------
TANK FAILURE CASE STUDY #12
SITE; Essex Industrial Chemicals, Inc. chemical processing plant
LOCATION; Baltimore, Maryland
RELEASE MECHANISM; Tank rupture
DATA SOURCE
"News In Brief: Hazardous Materials," Hazardous Materials
Intelligence Report, 30 December, 1983, pp. 3-4.
DESCRIPTION OF RELEASE; An outdoor storage tank burst
CAUSE OF RELEASE
RELEASE MATERIALS
t How Detected: storage tank burst
0 Material Types: sulfuric acid
Quantises Released: 485,000 gallons
RELEASE DURATION
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
RELEASE ENVIRONMENT
Land, Water, Air, Unknown:
Description of contamination: approximately 388,000 gallons traveled
over the frozen ground at the plant and spilled Into the Cabin Branch
waterway, which leads Into Curtis Creek and eventually Into the
Chesapeake Bay.
C-13
-------
TANK FAILURE CASE STUDY #13
SITE; Allied Chemical Corporation
LOCATION; Louisiana
RELEASE MECHANISM; Tank rupture
DATA SOURCE
Shields, Edward, Dessert, W.J., "Learning a Lesson from a Sulfuric Acid
Tank Failure," Pollution Engineering, December 1981, pp. 39-40.
DESCRIPTION OF RELEASE; Into the ground
CAUSE OF RELEASE "
An Inlet nozzle for the addition of add to the tank was located too
close to the tank wall.- The cast Iron Inlet pipe broke due to corrosion
and the high velocity of the Incoming add stripped the protective
coating on the sides of the tank. Eventually the tank ruptured.
»
RELEASE MATERIALS
How Detected: rupture In tank
Material Types: 93X sulfurlc add
Quantities Released: 2500 tons
RELEASE DURATION; 625 minutes (10.4 hours)
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
Tank/Treatment Components, Ancillary Equipment: outlet fittings, inlet
nozzle manhole
Equipment That Failed: vertical weld
Equipment Material: A-283 grade C steel plate
Equipment Age: Unknown
Corrosion Protection: ferrous sulfate film
C-14
-------
RELEASE ENVIRONMENT
Land, Water, Air, Unknown:
Description of contamination: effluent discharge pumps to Mississippi
River were turned off and the rain water drainage pipe from the tank
farm Impoundment area was sealed.
COMMENTS
A 3000-ton sulfrlc acid tank In southern Canada also failed because
the inlet nozzle was too close ta the tank wall.
C-15
-------
TANK FAILURE CASE STUDY #14 .
SITE; Unknown
LOCATION; Unknown
RELEASE MECHANISM; Pipe rupture
A
DATA SOURCE
§ Vervalln, Charles H., "Learn from HPI plant fires," Hydrocarbon
Processing
DESCRIPTION OF RELEASE
About 1000 gallons of the hydrocarbon mixture flowed through a 3/8" pipe
opening 1n the pump housing from which the pipe plug had fallen. Most
was absorbed by the ground, however some flowed about 50 feet fr-om the
tank through a shallow ditch.
CAUSE OF RELEASE
A pipe plug had fallen or had been blown. The plug was non-metal He.
RELEASE MATERIALS
How Detected: fire
a Material Types: hydrocarbon mixture of cyclohexane and n-heptane
0 Quantities Released: 1000 gallons
RELEASE DURATION
TANK DESIGN AND OPERATING CHARACTERISTICS -
a Tank Use: storage
RELEASE ENVIRONMENT
Land, Water, A1r, Unknown: land
Description of contamination: The leaking fluid was absorbed by the
ground. Some of 1t flowed about 50 feet from the tank to a shallow
ditch.
C-16
-------
TANK FAILURE CASE STUDY #15
SITE; Tank farm of General American Transportation Co.
LOCATION; San Pedro, California
RELEASE MECHANISM; Pipe rupture
a
DATA SOURCE
§ Vervalin, Charles H., "Learn from HPI Plant Fires," Hydrocarbon
Processing, December 1972, pp. 49-50.
DESCRIPTION OF RELEASE
A tank truck collided with a pipe, and fire engulfed the tank and truck.
The fire spread to nearby trucks.
CAUSE OF RELEASE
A tank, truck apparently sheared a pipe leading to a 30,000 gallon tank.
RELEASE MATERIALS
How Detected: fire
0 Material Types: vinyl acetate
Quantities Released: 30,000 gallons
RELEASE DURATION
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
Equipment That Failed: pipe
Dikes/Berms: diked area around tank farm
RELEASE ENVIRONMENT
V
t Land, Water, Air, Unknown: land
Description of contamination:
C-17
-------
TANK FAILURE CASE STUDY #16
SITE; 011 processing and reclamation facility, Bridgeport Rental and Oil
Services
LOCATION; Southern New Jersey
RELEASE MECHANISM; Tank corrosion
i
DATA SOURCE
Whittaker, Kenneth T., Goltz, Robert, "Cost Effective Management of an
Abandoned Hazardous Waste Site by a Staged Clean-up Approach,"
Management of Uncontrolled Hazardous Waste Sites, 1982, pp. 262.
DESCRIPTION OF RELEASE
CAUSE OF RELEASE; Tank corrosion
RELEASE MATERIALS
How Detected:
Material Types: uncharacterized hydrocarbons, various benzene and
phenolic polyaromatic hydrocarbons (phenanthene, napthalene) lagoon
surface high concentration of solvents and PCBs
0 Quantities Released: 5 of the 88 on-site 300,000-gallon tanks were empty
RELEASE DURATION
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
RELEASE ENVIRONMENT
t Land, Water, Air, Unknown: water (surface and groundwater)
Description of contamination:
C-18
-------
TANK FAILURE CASE STUDY #17
SITE: An electronic components manufacturing plant
LOCATION; A suburban area adjacent to a major city. The site 1s surrounded by
residential neighborhoods and small farms.
RELEASE MECHANISM; Unknown, presumably tank corrosion
DATA SOURCE
e Assessment of the Technical, Environmental and Safety Aspects of Storage
of Hazardous Waste In Underground Tanks, Vol. I, SCS Engineers. Reston,
Virginia, August 1983, pp. 3-36 - 3-46.
DESCRIPTION OF RELEASE.
Lack of Inventory and/or environmental monitoring, tank Inspection, or
tank testing programs at this site allowed a waste solvent storage tank
. leak to go undetected for approximately H years. The leak material
contaminated soil and ground water. As a result of the duration and
size of the leak and the hydrogeology of the site, transport of the con-
tamination Into-three aquifers and over an area of about 1/3 square mile
occurred.
CAUSE OF RELEASE
RELEASE MATERIALS
t How Detected: a mass balance analysis on the solvents entering and
exiting the plant disclosed a leaking tank
Material Types: solvents - Acetone, l-l-D1chloroethylene, Freon 113,
Isopropyl alcohol, 1.1.1-Trlchloroethane, and Xylene
Quantities Released: 58,000 gallons
RELEASE DURATION 1* years
TANK DESIGN AND OPERATING CHARACTERISTICS
Tank Use: storage
RELEASE ENVIRONMENT
Land, Water, Air, Unknown: land
t Description of contamination:
C-19
-------
TANK FAILURE CASE STUDY #18
SITE; A manufacturing plant that produces electronic computing equipment,
semi-conductors and related devices.
LOCATION; A suburban area adjacent to a major city and Is surrounded by resi-
dential neighborhoods, small farms, a hospital, and a golf course.
RELEASE.MECHANISM; Unknown, but Includes a corroded pipe
DATA SOURCE
t Assessment of the Technical, Environmental and Safety Aspects of Storage
of Hazardous Waste 1n Underground Tanks, Vol. I, SCS Engineers. Res ton,
Virginia, August 1983, pp. 3-47 - 3-64.
DESCRIPTION OF RELEASE
CAUSE OF RELEASE
Some of the probable causes are Improper disposal of the chemicals,
past operational problems, and a corroded drainline, but most of the
causes are unknown.
RELEASE MATERIALS
How Detected: unknown
Material Types: solvent (acetone; ethyl amyl ketone; Freon 113; Isopro-
pyl alcohol; 1,1,1-tHchloroethane; 1,1,1-trlchloroethylene; or Xylene)
t Quantities Released:
RELEASE DURATION
TANK DESI6N AND OPERATING CHARACTERISTICS
Tank Use: storage, treatment, sumps
Equipment Material: concrete, fiberglass, carbon steel, stainless steel
RELEASE ENVIRONMENT
Land, Water, Air, Unknown: land
Description of contamination:
C-20
-------
COMMENTS
a Lack of inventory and/or environmental monitoring, tank inspection or
tank testing programs at this site allowed many leaks to go undetected
for as long as 11 years before detection. The source of pollution has
been determined for only one of the three areas found to have soil and
groundwater contamination. Transport of the released chemicals into
three aquifers for over a mile away from the site resulted from the
duration and size of the leaks and the hydrogeology of the area.
This appears to be the same site as the one discussed in the preceding.
, case study, but this incident apparently involves a separate set of
failures.
C-21
-------
TANK FAILURE CASE STUDY #19
SITE: Agricultural chemical manufacturing plant
LOCATION; Northern California
DATE SOURCE
-^United States Environmental Protection Agency, Case Studies 1-23:
Remedial Response at Hazardous Waste Sites. EPA-540/2-84-0026, March
1984 (Case Study #2).
RELEASE MECHANISM
0 Pipe corrosion or rupture, accidental pills, and tank corrosion or
rupture.
DESCRIPTION OF RELEASE
The tanks in question were part of the treatment system for rainfall
run-off and rinsewater from the plant's chemical handling areas. 15,000
gaMons leaked from an underground "skimmer tank." There were also a
number of" small-scale chemical spills, and leakage from 2 joints in a
300' chemical drain used to connect parts of the system.
CAUSES OF RELEASE; Unknown, but probably include corrosion or rupture of both
the tank and the drain.
RELEASE MATERIALS
0 How Detected: "Foul taste" it) nearby drinking wells.
Material Types: Toluene and various herbicides.
a Quantities Released: In excess of 15,000 gallons.
RELEASE DURATION
Unknown. The system was constructed In 1971; the leak was discovered in
1979.
RELEASE ENVIRONMENT
Land, Water, Air, Unknown: land
Description of contamination: There was contamination of shallow
groundwater. Nearby drinking wells had a "foul taste," but no detec-
table chemicals.
C-22
-------
.TANK FAILURE CASE STUDY #20
SITE: Blocraft Laboratories
LOCATION; Waldwick, NJ
RELEASE MECHANISM; Gasket leak
DATA SOURCE;
United States Environmental Protection Agency, Case Studies 1-23:
Remedial Response at Hazardous Waste Sites. EPA-540/2-84-0026, March
1984 (Case Study 14).
DESCRIPTION OF RELEASE
A gasket in a fill pipe for an underground storage tank disintegrated
due to incompatibility with the waste.
CAUSE OF RELEASE
t Incompatibility between gasket and waste
RELEASE MATERIALS
How Detected: Groundwater testing
Material Types: Methylene chloride, N-butyl alcohol, dimethyl aniline,
acetone, and a variety of trace organics.
t Quantities released: Uncertainpossibly as much as 360,000 Ibs.
RELEASE DURATION: Probably 3 years
TANK DESIGN AND OPERATING CHARACTERISTICS
0 Tank use: storage
t Equipment that failed: gasket
0 Tank design: steel tank with no secondary containment and apparently no
'corrosion protection
Age of system: new (constructed In 1972, failure detected 1n 1975)
C-23
-------
RELEASE ENVIRONMENT;
Land, Water, Air, Unknown: land
Extent of contamination: Groundwater was contaminated. The release was
also the probable cause of a fish kill in a nearby stream in 1973.
C-24
-------
TANK FAILURE CASE STUDY #21
SITE; General Electric transformer manufacturing and repair facility
LOCATION; Oakland, CA
RELEASE MECHANISM; Accidental spills, overflows, and tank rupture
r
DATA SOURCE;
United States Environmental Protection Agency, Case Studies 1-23:
Remedial Response at Hazardous Waste Sites. EPA-540/,2-84-0026, March
1984 (Case Study f9).
DESCRIPTION OF RELEASE
Over the operating history of the plant, a number of spills had occurred
(a) at an above-ground tank farm used for a petroleum-based thinner and
oil; (b) near two above-ground 5,000-gallon tanks used for Pyranol
(contains PCS); (c) 1o the area where rail tank cars were unloaded by
pumping; (d) possibly due to minor leakage from oil-warming operations
Inside the building; and (e) from a mobile filtering unit that would
occasionally "blow" from too much pressure.
Additional contamination came from trench burial of liquid PCB's and
contaminated solids such as d1alectrie paper, and from continued
discharges from a lab sink following the collapse of a septic tank (date
of collapse Is unknown. Discharges continued until the mid 1960's).
CAUSES OF RELEASE
Above-ground spills
0 Overflows
0 Tank rupture
RELEASE MATERIALS
0 How detected:
0 Material types: Hydrocarbon products and PCB-contaminated oils
0 Quantities released: 20,000 gallons
*
RELEASE DURATION; Miscellaneous spills between 1927 and 1975.
C-25
-------
RELEASE ENVIRONMENT
Land, Water, Air, Unknown: land ^ :
0 Description of contamination: surface spills and underground leaks,
There was widespread contamination of on-site soils and groundwater,
C-26
-------
TANK FAILURE CASE STUDY #22
SITE; Houston Chemical Company
LOCATION; Houston, MO.
RELEASE MECHANISM; Tank rupture and pipe design error
DATE SOURCE
United States Environmental Protection Agency» Case Studies 1-23;
Remedial Response at Hazardous Waste Sites, EPA-540/2-84-0026, March
1984 (Case Study fi3>.
DESCRIPTION OF RELEASE
The tank was a 21,000-galIon,.steel, horizontal, above-ground storage
tank. There was no containment system. It collapsed for several
reasons:
The saddle support blocks were not sufficient either in spacing or
number.
- there were weaknesses due to corrosion and previous abuse.
» The saddle support blocks were not engineered to fit the curvature
of the tank.
A drain pipe and valve were Installed on the underside of tank.
When the tank collapsed, the drain control valve and piping
sheared off. This was a design error.
An overflow pit contained 5X of the spill. The remainder bypassed the
pit due to the absence of suitable dikes. The oil flowed Into a road-
side ditch, under a culvert and Into a catch basin where 1t remained
temporarily. It then infiltrated the ground, reappearing 125' downgra-
dlent. Eventually it flowed into a farm pond where the water level was
low enough that It remained.
. The Initial report was not received by EPA until 4 days after the spill.
That report falsely stated that the spill had been contained by a dike
and that clean-up was under way. .The volume of the spill was initially
reported as 10,000 gal.
C-27
-------
RELEASE MATERIALS
How Detected:
0 Material Types: 5X solution of PCP in diesel oil
Quantities released: 15,000 gal.
RELEASE DURATION: Release occurred on June 14, 1979.
TANK DESIGN AW OPERATING CHARACTERISTICS
Tank Use: Storage
Tank Design: 21,000-gallon, steel, horizontal, above-ground, on
cradles.
Equipment that failed: cradles
Secondary containment: present, but Inadequate.
RELEASE ENVIRONMENT
t Land, Water, A1r, Unknown: land and water
Description of Contamination: There was a total fish kill 1n the farm
pond and a threat of overflow Into a navlgatable river known as a
valuable wildlife habitat. Soil along the path of flow was con-
taminated.
C-28
-------
TANK FAILURE CASE STUDY #23
NAME; Howe, Inc.
LOCATION: Brooklyn Center, MN
RELEASE MECHANISM; Hazardous materials carried off-site by smoke from fire and
run-off water from fire-fighting efforts.
*
DATE SOURCE
United States Environmental Protection Agency, Case Studies 1-23;
Remedial Response at Hazardous Haste Sites. EPA-540/2-84-0026, March
1984 (Case Study fI4j.
DESCRIPTION OF RELEASE
In January, 1979., a fire occurred at a warehouse site containing 100
different pesticides totaling 80 tons of active Ingredients. Water
used to fight the fire flowed off-site, carrying with 1t dissolved
pesticides and herbicides. Several additional dangers were Involved:
.'-;.A1r pollution from combustion of organic solvents. Pigeons
flying through the plume died Immediately. Eleven fire fighters
became 111.
- Fallout from the plume.
- Contaminated building debris.
- Run-off from contaminated soils.
CAUSE OF RELEASE; Faulty acetylene torch
RELEASE MATERIALS
0 How detected: Immediate visual detection
Material types: Pesticide- and herbicide-contaminated water; fumes.
e Quantities released: 500,000 gallons of contaminated water; unknown
amount of. fumes.
RELEASE DURATION; several hours
C-29
-------
RELEASE ENVIRONMENTS
0 Land, Water, Air, Unknown: land and air
Description of contamination: A1r and land surface. Because of cold
temperatures, the contaminated water froze on the ground surface.
C-30
-------
TANK FAILURE CASE STUDY 124
NAME; N. W. Mauthe, Inc.
LOCATION: Appleton, Wisconsin
RELEASE MECHANISM; Accidental spills; cracks 1n concrete floor.
*
DATA SOURCE
United States Environmental Protection Agency, Case Studies 1-23;
Remedial Response at Hazardous Waste Sites, EPA-540/2-84-0026, March
1984 (Case Study 116).
DESCRIPTION OF RELEASE
A blower vent for a chrome-plating tank discharged chromium-laden m1st~"
to the outside;
Drippings from chromatlng tanks were channeled to a sanitary sewer by a
trough 1n the floor. Cracks 1n the trough and the concrete flooring led
to -seepage Into underlying soil.
»
CAUSE OF RELEASE
Poor design and aging of concrete floor
RELEASE MATERIALS
How Detected: In March, 1982, yellow puddles were observed on adjacent
property.
Material Types: Chromium-contaminated water.
Quantities released:
RELEASE DURATION
The shop operated from 1966 through 1976. The releases were probably
ongoing through much or all of that time.
C-31
-------
RELEASE ENVIRONMENT
Land, Water, Air, Unknown: land
Description of contamination: Continuing spills contaminated soils at
the site. The contaminants migrated off-site, where they were disco-
vered as "yellow puddles." There was a threat to nearby residences and
schools and a threat of run-off to storm sewers leading to the Fox
River.
C-32
-------
TANK FAILURE CASE STUDY #25
NAME; Quanta Resources
LOCATION; Queens, NY
RELEASE MECHANISM; Vandalism
^
DATA SOURCE
United States Environmental Protection Agency, Case Studyes 1-23;
Remedial Response at Hazardous Waste Sites. EPA - 540/2-84-0026,
March 1984 (Case Study fl9).
DESCRIPTION OF INCIDENT
The facility was a processing facility containing about 500,000 gal. of
miscellaneous wastes, Including PCB-contam1nated oils, cyanides, heavy
metals, and low flash-point chlorinated solvents. Bankruptcy of the
owner left the facility without security against arson or vandalism.
*t
CAUSE OF RELEASE
Noneno release actually occurred. However, the unguarded state of
the facility posed a substantial risk of arson or vandalism.
C-33
-------
TANK FAILURE CASE STUDY #26
NAME; White's Septic Tank Services
LOCATION; OuPage County, IL
RELEASE MECHANISM; Operator error
DATA SOURCE
Landfilling of Special and Hazardous Waste 1n Illinois; A Report to
the Illinois General Assembly. Illinois Legislative Investigating
Commission, 19//.
DESCRIPTION OF INCIDENT
White operated a land treatment facility on his 30-acre farm. He was *
permitted to accept only domestic septage. He could spread It only In
good weather and not near the river. Never- theless, he accepted.com-
mercial wastes of unknown nature, mixing them with the septage In his
trucks. He professed Ignorance of his permit requirements and was
"cordial.11 His record-keeping was virtually non- existent. According
to the report, he was "totally lacking in the skill, equipment,
knowledge, and desire necessary [for] toxic waste disposal...--blythely
spread Cingj Industrial wastes over farm land."
COMMENTS
White had also been 1n the same business at different times at other
Illinois sites.
This case study does not Involve hazardous waste tanks. It Is never.
theless relevant, because tank operators can be just as untrained as
landfarm operators.
C-34
-------
TANK FAILURE CASE STUDY 127
NAME; White's Septic Tank Services
LOCATION; DuPage County, IL
RELEASE MECHANISM; Inspection error
DATA SOORCE
Landfill1ng of Special and Hazardous Waste In Illinois; A Report to
the Illinois General Assembly. Illinois Legislative Investigating
Commission, 19/7.
DESCRIPTION OF INCIDENT (see preceding case study)
An Illinois Department of Public Health Inspector made a routine visit
to White's landfarm on 7/6/78, accompanied by an Illinois Legislative
Investigation Commission observer. The visit Involved an applica-
tion for license renewal, the IDPH Inspector:
- Was uncertain of what the law required;
Was''unaware of the danger of leachate contaminating river;
- Took no water samples;
- Placed a checkmark next to "Inspection of servicing equipment"
without ever going near the one truck that made a dump while he was
on the site; and .
Did not seem particularly knowledgeable about the operating require-
ments.
COMMENTS
This case study does not Involve hazardous waste tanks. It Is never-
theless relevant, because similar Inspection errors could occur at any
type of hazardous waste facility.
C-35
-------
TANK FAILURE CASE STUDY #28
NAME; Oestructol Carol awn
LOCATION: Kernersvllle, NC
DATA SOURCE; EPA/SCS Remedial Action Cases, Slte'D
^
RELEASE MECHANISM; Catastrophic release (vandalism)
DESCRIPTION OF RELEASE
On June 3, 1977, vandals opened the valves on six storage tanks at a
commercial hazardous waste Incinerator. There were no locks on the
valves. There was no secondary containment.
RELEASE MATERIALS:
t How Detected: Immediate visual observation
Material Types:
Quantities Released: 30,000 gal.
tf
RELEASE ENVIRONMENT;
Land, Water, Air, Unknown: land and water
t Description of contamination: There was a 90-99* fish kill in a nearby
50-acre reservoir. 200 local residents were temporarily evacuated.
Temporary water rationing and Industrial layoffs resulted, as the muni-
cipality sought an alternative water source. So far, there has been
permanent loss of the reservoir as a drinking water source (the state
refused to approve the reservoir as a drinking water source as long as
the threat of future contamination remained).
C-36
------- |