Proceedings of the ASME 2022 13th International Pipeline Conference

IPC2022
September 26-30, 2022, Calgary, Alberta, Canada

IPC2022-87108

USING ADVANCED PROBABILITY OF EXCEEDANCE METHODS FOR DEFINING

REMEDIATION OPTIONS

Thomas Bubenik Benjamin Hanna William Harper Matt Ellinger

DNV GL USA, Inc. DNV GL USA, Inc. DNV GL USA, Inc. DNV GL USA, Inc.
Dublin, OH Dublin, OH Dublin, OH Dublin, OH

ABSTRACT
Keywords: Probability of exceedance, in-line inspection,
The Probability of Exceedance (POE) methodology is used
pipeline, corrosion growth rate
by pipeline operators to manage corrosion. It prioritizes
anomalies identified by an in-line inspection (ILI) based on their
likelihood of exceeding one or more failure thresholds. Two types NOMENCLATURE
of thresholds are commonly used: leak and rupture. The 𝜎 Standard deviation of growth rate
calculations represent the likelihood an anomaly will leak or 𝜎 Standard deviation of ILI error
rupture as a function of time. By assessing the likelihood of such 𝜎 Standard deviation of ILI calls in second
a release as a function of time, an operator can design rational inspection
and statistically valid remediation programs. 𝜎 Standard deviation of ILI calls in first
In this study, the likelihood that an anomaly will cause a leak inspection
or rupture was based on the ILI dimensions, the ILI system’s 𝜎 Pooled deviation of ILI calls in both
accuracies, and the corrosion growth rate at the anomaly inspections
location. The ILI dimensions were adjusted by the mean error as CDF Cumulative Density Function
determined using ILI-to-field comparison data. The same ILI-to- CGR Corrosion Growth Rate
field data were used to determine the scatter (standard deviation) d Depth
in reported dimensions for different categories of anomalies. 𝑑 Projected depth at year “k”
The POE (likelihood) of a leak or rupture was taken as a
ILI In-line Inspection
function of the mean and standard deviation of a corrosion
k Number of years
growth rate at the anomaly location. Anomalies were “grown”
MAOP Maximum Allowable Operating Pressure
using mean growth rates that were established by performing ILI
𝑛 Number of ILI calls in first inspection
run-to-run comparisons. By using accurate growth rates and ILI
𝑛 Number of ILI calls in second inspection
performance characteristics, realistic estimates were made of the
PDF Probability Density Function
POE of an anomaly, pipe joint, pipe segment, and the overall
PHMSA Pipelines and Hazardous Materials Safety
pipeline itself.
Administration
This paper discusses improved methods of calculating leak
POE Probability of Exceedance
and rupture POEs using realistic assumptions on how anomaly
POF Pipeline Operators Forum
growth occurs as a function of time. Case studies are used to
wt Wall thickness
demonstrate the effectiveness of the advanced POE method.
x Variable, x
The introduction and select portions of this paper are based,
in part, on “Advanced Probability of Exceedance Method
Overcomes Previous Weaknesses,” by the authors,1 Pipeline
Pigging and Integrity Management Conference, February 2-4,
2022.

1 © 2022 by ASME
1. INTRODUCTION The POE method was developed to account for ILI sizing
POE is a method of calculating the likelihood an ILI- errors and future corrosion growth in a straightforward fashion.
identified anomaly reaches a critical size as a function of time. The six overall steps of the POE analyses are given below.
The critical size can be a depth at which a leak occurs or a failure 1. Characterize ILI accuracies in terms of mean errors and
pressure that equals the operating pressure multiplied by a safety standard deviations
factor. The probability of reaching (or exceeding) the critical 2. Determine critical dimensions of all ILI calls
state is calculated for each individual ILI call, after which (anomalies) assuming the anomaly grows in depth and
individual POE values are combined to determine the likelihood length to failure by leak or rupture
one or more anomalies reaches a critical state. 3. Characterize the expected growth rates of each ILI call
POE was originally developed over twenty years ago.2,3 4. Grow each ILI call using the characterized growth rates
Since that time, improvements have been made, but the 5. Calculate the likelihood the anomaly reaches a depth
methodology has been hampered by assumptions that are neither that corresponds to a leak or rupture each year (or at
appropriate nor correct and which produce failure rate alternative intervals as needed)
likelihoods (i.e., POEs) that are not representative. These 6. Combine individual POE values into cumulative POE
limitations primarily exist in the manner in which growth in the values
length direction is considered and rupture POEs are calculated, Each of these steps and recent improvements are described
as described herein. in the following sub-sections. After the POE method and
Typically, POE calculations are based on most-likely ILI improvements are described, a case study is presented to show
errors and corrosion growth rates. POE calculations are meant to how the improved results can be used to optimize a pipeline
provide a best estimate of the likelihood of failure as a function operator’s response program following an ILI.
of time. POE values of individual anomalies can be combined
per joint, per pipeline segment, or for any other length defined 2. POE METHOD
by the pipeline operator to establish a “cumulative” POE. 2.1 Step 1. Characterize ILI Accuracies
Individual and cumulative POEs can be calculated based on The first step in using POE is to assess the degree to which
“boxes” (i.e., individual ILI-reported metal loss anomalies) the inspection met its performance specification. ILI systems
and/or “clusters” (i.e., groups of ILI-reported metal loss often show different accuracies in sizing as a function of
anomalies that interact with one another). anomaly type. In the analyses described here and in traditional
To illustrate, consider the normal distribution curve shown POE analyses, the anomaly categories introduced by the Pipeline
in Figure 1. This curve represents the likelihood that an anomaly Operators Forum (POF) can be used for the characterization:
with a reported depth of 60 percent of the wall thickness (%wt) axial slotting, circumferential slotting, axial grooving,
is actually any value shown on the x-axis. The area under the circumferential grooving, general, pitting, and pinholes.
normal distribution curve is defined as one (100 percent). As an For each anomaly category, errors in sizing are assumed to
example, the shaded area represents the likelihood the actual be normally distributed. The mean error and standard deviation
depth is greater than 70%wt. In this case, there is about a ten between the depth reported by the ILI and the depth measured in
percent likelihood the anomaly is actually 70%wt or deeper (i.e., the field are estimated.
the shaded area represents ten percent of the total area under the In theory, characterizing ILI system accuracies can be
curve). Curves like those shown in Figure 1 can be used to conducted using ILI-to-field measurements after accounting for
determine the likelihood that each reported anomaly exceeds a field measurement errors. In practice, though, it is often difficult
threshold or critical value. to obtain enough data to accurately assess ILI performance for
Normal Distribution all anomaly categories. As a result, it is sometimes necessary to
aggregate data from multiple ILIs to estimate performance as a
function of anomaly type. Alternatively, the ILI system accuracy
can be taken directly from the ILI performance specification,
keeping in mind the performance specifications for some
Relative Likelihood

categories can be inaccurate or unreliable.

Table 1 summarizes depth ILI accuracies used in one POE
analysis. Each POF category has its own mean error and standard
deviation. A negative mean error corresponds to the ILI system
reporting shallower depths than what was measured in the field.
In this case, the ILI tolerances stated by the ILI service provider
were larger than those observed in the measured data. That is,
10 20 30 40 50 60 70 80 90 100 110
Actual Depth (%wt)
the ILI tool performed better than stated by the ILI service
Yield Strength (ksi)
provider.
+/‐10% w/80% Certainty

Figure 1. Normal Distribution of a 60% Deep ILI

Reported Feature Exceeding 70%

2 © 2022 by ASME
 Table 1. Typical Accuracies of ILI Data Based on Field relative rates at which depths and lengths grow. For leaks and
Measurements (in %wt) ruptures, the critical dimensions depend on the length and depth
of the anomaly, both of which are potentially growing. Coupling
these two growths is done by defining length growth as a

General

Pinhole
Groove

Groove
Axial.

Axial.
Circ.

Circ.
Slot

Slot
function of depth growth.

Pit
In this method, anomalies were assumed to grow to failure
until they reached the critical flaw curve, which is a function of
Measured Mean 2 -2 -1 -5 -3 -4 -1 pipe properties and the local pressure. Whether the anomaly will
Measured Standard leak or rupture is determined by the leak/rupture length at the
10 7 8 8 7 6 6
Deviation local pressure. Anomalies with failure lengths below the
Calculated leak/rupture boundary are assumed to leak, while those with
Tolerance at 80% ±13 ±9 ±10 ±10 ±9 ±8 ±8 lengths greater than the boundary are assumed to rupture.
certainty Using the leak/rupture boundary to delineate leaks and
ILI Performance ruptures differs from the traditional POE approach, which
Specification at None None ±15 ±10 ±10 ±10 None assumes leaks occur at a given depth (e.g., 80% or 100% of the
80% certainty wall thickness) regardless of the anomaly length. Assuming
anomalies leak at a given depth independent of anomaly length
Factors that reduce the ILI system accuracies are not always is incorrect because the depth at which the anomaly will leak
included but should be considered, and care should be taken to depends on its length.
not overestimate the actual impact. For example, many ILI As noted earlier, a relationship between growth in the depth
systems lose accuracy around welds or in regions where the ILI direction and growth in the length direction is necessary to
tool velocity exceeded allowable limits. ILI service providers determine the dimensions at failure. For individual metal loss
often state a tolerance for anomalies near welds that is much anomalies (i.e., anomalies that are not part of a “cluster”), it is
larger (e.g., twice) than the tolerance away from welds, all things reasonable to assume the anomaly’s length growth is twice the
equal. In the authors’ experience, the actual accuracies near depth growth or that the anomaly’s aspect (length/depth) ratio is
welds are often better than given by ILI service providers in their constant. The former is equivalent to assuming a constant growth
(conservative) performance specifications. rate along the periphery of the anomaly. The latter assumes the
Characterizing ILI accuracies is generally done based on cross-sectional shape remains the same. Both options can be
anomaly depths. However, the anomaly length can have an equal evaluated. In traditional POE approaches, only depth growth is
(and sometimes larger) influence on predicted failure by rupture. considered when calculating leak POEs.
Anomaly length accuracies can also be evaluated using field As an example, Figure 2 illustrates how to determine the
data, but care should be taken to account for length sizing errors critical flaw dimensions for an anomaly with a constant aspect
in the field. Length sizing errors from field measurements tend ratio. Here, the reported initial depth of the anomaly is
to be much larger than depth sizing errors due to differences in approximately 25%wt and the reported length is approximately
the way length measurements are made in the field versus the ILI 3 inches. Maintaining a constant aspect ratio, the depth and
system. Because of this, it may not always be appropriate to length increase in proportion until they reach critical dimensions
incorporate field data into length sizing accuracies when corresponding to failure at the MAOP. In the illustrated case, the
conducting a best estimate POE analysis. anomaly fails when the depth is approximately 50%wt and the
length is 6 inches. That is, the anomaly would need to double in
2.2 Step 2. Determine Critical Dimensions size to rupture.
The second step in the POE process is to determine critical
dimensions for each anomaly reported by the ILI system,
assuming growth in depth and/or length. POE methods of
determining critical dimensions vary widely in the literature,
with most relying on an assumption about how quickly the
(predicted) burst pressure decreases with time, rather than how
quickly the defect grows. This approach also ignores the
coupling between growth in the depth and length directions and
its direct impact on predicted burst pressure.
For the method described in this paper, a solution was
developed to predict the dimensions of an anomaly at the time of
failure as a function of growth rates in the depth and length
directions. The initial anomaly dimensions are provided by the
ILI, after accounting for any corrections required in Step 1.
The “final” dimensions (i.e., the dimensions at which the
anomaly is predicted to leak or rupture) are determined from the

3 © 2022 by ASME
 Figure 2. Depth and Length Growth to Critical Figure 3. Cluster Aspect Ratio Multiplier Effects
Dimensions A critical question is at what aspect ratio does corrosion
For clusters, another approach is necessary. Clusters are growth take place. To evaluate corrosion growth aspect ratios,
comprised of multiple metal loss anomalies, each of which the authors reviewed ILI-reported length-to-depth ratios for
essentially grows independently. To capture the change in length individual boxes and for clusters. Figure 4 shows a typical
of a cluster, an assumption is made regarding how quickly the example of how the aspect ratio changes as depth increases.
cluster length changes as the depths of the individual anomalies While more shallow boxes can have very large aspect ratios,
increase. Various approaches were considered before selecting deeper boxes tend to have average aspect ratios that approach a
one for the POE analyses described here. value of two (i.e., the growth in length is twice the growth in
In traditional POE analyses, cluster lengths are not explicitly depth) as the anomalies approach full wall thickness.
considered. Instead, the change in failure pressure is projected to
solely be a function of the change in cluster depth. This approach
ignores differences in how length affects burst pressure and,
hence, critical flaw dimensions.
To study how the aspect ratios of clusters change, a
capability to change the growth aspect ratio was built into the
new POE solver. Here, each cluster’s aspect ratio during growth
was taken as a multiple of the cluster’s initial aspect ratio, as
shown in Figure 3. If the aspect ratio multiplier is taken as one,
the cluster’s aspect ratio remains constant, which can be overly
conservative for clusters. If the aspect ratio multiplier is zero, the
length does not change, which may be unconservative. For
aspect ratio multipliers between zero and one, the growth
trajectory varies.

Figure 4. Length-to-Depth Aspect Ratios for Boxes

At one point in time, the deep anomalies were shallow.
When they were shallow, they had a high aspect ratio. As they
grew, their aspect ratio became smaller. In order for the aspect
ratio to become smaller, the length must be growing at a slower
rate than the depth.
Clusters can grow in several ways. Individual boxes within
a cluster may grow toward an aspect ratio of two, which could
slightly change the cluster’s aspect ratio, and clusters could also
grow by interacting with other boxes or clusters. To evaluate
how clusters grow, a similar analysis was conducted on reported
aspect ratios for clusters. Figure 5 shows the results. In this
figure, the aspect ratios are based on effective lengths and depths
rather than maximum lengths and depths.

4 © 2022 by ASME
 data to accurately calculate the error or bias. A much larger data
set for analysis can be obtained by matching metal loss calls in
the first and second inspection. The matched data can then be
used to determine bias in the initial inspection relative to the
subsequent inspection. Changes in mean depths on a joint-by-
joint basis can then be calculated. Such changes could be the
result of bias, growth, or both.
To determine if the initial inspection has a bias relative to
the second inspection, a bias assessment is performed. Typical
components of a bias assessment include (1) performing
anomaly-to-anomaly matching and identifying trends between
matched metal-loss pairs, (2) observing reporting differences
between the subsequent ILI survey data at areas that are unlikely
to contain corrosion growth (e.g., areas containing composite
sleeves, areas in which no growth is observed in the ILI signal
data, etc.), and (3) assessing field-too correlations to establish
Figure 5. Length-to-Depth Aspect Ratios of Clusters actual tool accuracy (as discussed in Step 1).
In the bias assessment, adjustments are made to ensure an
The mean aspect ratio curve in Figure 5 illustrates that the
apples-to-apples comparison between the subsequent ILI survey
aspect ratios of clusters tend to have a higher value – in this case,
data. This allows reported depths to be directly compared, which
the average aspect ratio approaches a value of approximately ten
results in more realistic corrosion growth rates.
as the clusters approach their critical depths. For shallower
clusters and as for individual boxes, the aspect ratios can be Mean and Standard Deviation of the Growth Rate
much higher or lower. Once bias has been considered, the two ILIs can be
Figure 4 and Figure 5 were used to provide guidance on how compared on a joint-by-joint basis to establish spreadsheet
the aspect ratio multiplier in the POE analyses should vary. A derived mean corrosion growth rates along the length of the
sensitivity study was performed using different aspect ratio pipeline segment. Mean growth rates are calculated based on
multipliers to determine which factor gave results that most comparing the average depth of metal-loss boxes between the
closely mimicked the average aspect ratios seen in the figures. A ILI survey data on a given joint.
balance between accuracy and conservatism is used in the new When comparing two ILIs, it is not usual for some of the
POE method. mean growth rates to be negative, implying corrosion in reverse!
Negative corrosion growth rates are replaced with a minimum,
2.3 Step 3. Characterize Expected Future Growth non-zero growth rate for each joint.
Rates Determining standard deviation associated with growth is
The third step in the POE process is to assign mean growth more challenging. The growth standard deviation is taken as a
rates and standard deviations to each pipe joint using the results pooled standard deviation of the older and newer ILI calls on the
from DNV’s CGReal analysis. This is another place where the joint from the two ILI runs. Separately, older and newer standard
POE calculations described here differ from traditional POE deviations are computed on a joint based on the calls reported by
practices. In traditional POE practices, a constant growth rate is each ILI. Then, these two standard deviations are weighed by
assumed and standard deviations (variabilities) are not their associated degrees of freedom to get a pooled standard
considered.
deviation, 𝜎 , as calculated in Equation (1) below:
DNV’s CGReal analysis, which is used to determine local
growth rates and standard deviations, is a comprehensive ILI ∗ ∗
run-to-run comparison that utilizes engineering and statistical 𝜎 (1)
analyses and is supplemented by ILI signal review. Details on
the CGReal process can be found in other references, but a brief where the number of calls on a joint are 𝑛 for the first inspection
overview is provided below.4 and 𝑛 for the second inspection, and the degrees of freedom are
Bias Assessments (𝑛 -1) for the older ILI anomalies on the joint and (𝑛 -1) for the
As a part of the CGReal analysis, the potential for a newer anomalies on that same joint.
difference in reporting or bias by one or both the previous and Finally, the standard deviation associated with corrosion
subsequent ILI tool runs is evaluated. As previously described, growth, 𝜎 , is computed factoring in the sample sizes for
ILI field-tool correlations are used to establish the error in the the number of ILI calls in each year for the joint.
most recent inspection. The mean error is considered the bias in
the most recent inspection. 𝜎 𝜎 (2)
Field-tool correlations could be used to account for bias in
the initial inspection, but as before, it is difficult to obtain enough

5 © 2022 by ASME
Special considerations, not discussed in this paper, are needed to usually accounted for in POE analyses. As discussed earlier,
address situations where the number of calls on one or both ILIs individual boxes are placed in a POF anomaly category to
is small. determine the mean and standard deviation associated with the
ILI error.
ILI Signal Reviews Assigning a POF category for clusters is more difficult than
At this point in the CGReal analysis, spreadsheet derived for individual boxes because each cluster consists of two or more
corrosion growth rates and standard deviations have been boxes, which may or may not represent a single POF category.
calculated on a per joint basis along the length of the pipeline Because each cluster’s depth typically equals the maximum
segment. Verification is needed that corrosion is taking place at depth of the cluster’s boxes, the ILI accuracy for the POF
locations where the mean corrosion growth rate is greater than category associated with the deepest box in each cluster can be
zero. To verify the growth, signal reviews by a skilled ILI analyst used in the POE calculations. Alternatively, the accuracy can be
are used to confirm, modify, or reject the mean growth rates. based on the overall cluster footprint. The accuracy is assumed
Joints with the highest corrosion growth rates, as well as to be constant as the anomaly grows – i.e., the analyses currently
additional joints based on the authors’ experience1, are included do not account for the potential for a cluster to change from one
in ILI signal review. Growth is confirmed when the shape POF category to another.
(footprint), length, and/or circumferential extent of the anomaly The cumulative standard deviation associated with anomaly
changes or additional anomalies appear. Growth rates are growth is also considered, and so, the uncertainty in the projected
modified when anomaly depths in the specific area(s) of anomaly depth and length increases with time. The total standard
observed growth are not the same as the mean anomaly depths deviation (i.e., the standard deviation due to ILI error and
within the entire joint. Finally, growth is rejected when no corrosion growth) as a function of time is needed to determine
change is observed between the subsequent ILI signal data. the likelihood the anomaly’s depth or predicted failure pressure
A minimum corrosion growth rate can be assigned to joints reaches a critical value. Accounting for the effects of uncertainty
that do not show evidence of growth, adding a degree of in future corrosion growth rates is not usually part of traditional
conservatism and accounting for growth too small to be seen in POE analyses.
the signal reviews or statistical analyses. Confidence in the In the approach documented here, the combined standard
growth rates depends in part on the number of ILI features on a deviation is a function of the ILI error (i.e., the standard deviation
joint in each inspection. For confidence levels between 90% and determined in Step 1) plus a term representing the standard
99%, the minimum detectable growth rates are between 0.5 and deviation of the growth rate applied from the CGReal analyses
1.0 mils per year (mpy). (determined in Step 3):
Step 4. Grow ILI Calls 𝜎 𝜎 𝑘 𝜎 (4)
The fourth step in the POE process is to calculate the depth
of each anomaly as a function of time, once the mean growth rate where 𝜎 is the total standard deviation applied in any one
and standard deviation are assigned. Here, the approach in this year, 𝜎 is the standard deviation associated with the ILI error,
paper follows traditional POE practices. The mean (expected) 𝜎 is the standard deviation associated with defect growth,
depth is simply the adjusted depth (from Step 1) times the and 𝑘 remains the number of years since the beginning of the
number of years from the most recent ILI: simulation. Equation (4) is based on an assumption that the
uncertainty in growth rate is independent of the uncertainty in
𝑑 𝑑 𝑘 𝐶𝐺𝑅 (3) ILI accuracy. This is a reasonable assumption because the
uncertainty associated with growth from the CGReal analysis is
where 𝑑𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑 is the expected depth in the year being analyzed,
a measure of the distribution of growth rates within a joint. It
𝑑 is the initial ILI depth incorporating the mean error from
uses reported calls from successive ILIs.
Step 1, 𝑘 is the number of years since the beginning of the Once the mean growth has been applied, and knowing the
simulation, and 𝐶𝐺𝑅 is the rate defined by the signal analyst standard deviation each year, the individual POE for each
or the mean corrosion growth rate when signal reviews are not anomaly is calculated using the equations listed below, where
performed. In addition to depth growth, the cumulative standard 𝑃𝐷𝐹 𝑥 is the probability density function for anomaly number
deviation of the growth rate also increases as the anomaly grows. “i”, 𝐶𝐷𝐹 𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 𝑑𝑒𝑝𝑡ℎ is the cumulative distribution
Step 5. Calculate Likelihood Each ILI Call Reaches a function, and 𝑃𝑂𝐸 is the probability of exceedance for anomaly
Critical Value number “i”.
The fifth step in the POE process calculates the likelihood . ∙
each anomaly’s depth or predicted burst pressure reaches a 𝑃𝐷𝐹 𝑥 𝑒 (5)
∙√
critical value. The POE calculations account for uncertainties in
the reported (ILI) dimensions, which is common in POE
analyses, and the established corrosion growth rates, which is not
1
This includes joints that contain a large change in anomaly counts, joints a given confidence level, joints that have new or disappearing metal loss
that exhibit a statistically significant difference in average or maximum depths at anomalies, joints that have relatively deep reported metal loss anomalies, etc.

6 © 2022 by ASME
 𝐶𝐷𝐹 𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 𝑑𝑒𝑝𝑡ℎ 𝑃𝐷𝐹 𝑥 𝑑𝑥 (6) increasing as a function of the growth rate assigned to the joint.
In this case, the growth rate is less than in Figure 6, as seen in
the flatter slopes of the POE curves for each anomaly. Each curve
in Figure 7 has a similar slope, because all anomalies on the joint
𝑃𝑂𝐸 1 𝐶𝐷𝐹 𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 𝑑𝑒𝑝𝑡ℎ (7)
are assumed to be growing at the same rate.
This approach to calculating the individual POE [i.e., Equations
Leak POE
(5), (6), and (7)] is the same as used in traditional POE analyses. 1.E+00
1.E‐01

Step 6. Determine Cumulative POE Values 1.E‐02

1.E‐03
The sixth and final step in the POE analysis is to assemble 1.E‐04

the cumulative leak and rupture POE values for each pipe joint 1.E‐05
1.E‐06
and pipeline segment (or other length of interest) using the

Likelihood
1.E‐07
following equation: 1.E‐08
1.E‐09
1.E‐10
𝑃𝑂𝐸 1 ∏ 1 𝑃𝑂𝐸 (8) 1.E‐11
1.E‐12
1.E‐13
1.E‐14
Capital pi (∏ in Equation (8) represents the product of all 1.E‐15
(1-POEi) terms. Similar to the above, this approach is the same 1.E‐16
2020 2022 2024 2026 2028 2030 2032
as used in traditional POE analyses. Year

3. RESULTS AND DISCUSSION Figure 7. Example Leak POE Results

Examples for a Single Pipe Joint
As an example, Figure 6 is a plot of the leak POE for a single
anomaly. The curve begins with a leak POE value around 1E-15 The curves shown in Figure 7 represent different initial
and steadily increases and approaches one (100 percent). depths and, possibly, different POF categories. As discussed
Because no growth has taken place in year 0, the initial POE above, the start of each curve (the far-left point on each curve) is
value is a function of only the reported depth of the anomaly and a function of the reported depth and the ILI accuracy for the POF
the ILI accuracy associated with the anomaly’s POF category. category of interest. As expected, the deeper anomalies have
The slope of the curve depends on the mean and standard higher starting leak POEs than shallower anomalies. Two of the
deviation of the growth rate, both of which increase the projected anomalies do not reach the minimum POE value of 1E-16 until
depth, length, and the combined standard deviation. Recall that several years into the simulation.
the standard deviation of the growth rate is not usually included Figure 8 shows another set of leak POE curves. This time,
in traditional POE analyses but is used here. each curve represents the cumulative POE of a joint (different
curves represent different joints). Here, the curves have different
Leak POE slopes, representing different growth rates along the line
1.E+00
segment. The dominant joint (i.e., the joint with the highest leak
1.E‐01
1.E‐02 POE in any given year) can change as anomalies on different
1.E‐03 joints grow to become threatening.
1.E‐04
1.E‐05
1.E‐06
Likelihood

1.E‐07
1.E‐08
1.E‐09
1.E‐10
1.E‐11
1.E‐12
1.E‐13
1.E‐14
1.E‐15
1.E‐16
2020 2022 2024 2026 2028 2030 2032
Year

Figure 6. Example Leak POE Results for a Single

Anomaly
Figure 7 is a similar plot with individual curves for five
anomalies on a single joint. Here, the curves are somewhat
similar to that seen in Figure 6, starting at a low POE value
(based on the initial depth and ILI accuracy for the anomaly) and

7 © 2022 by ASME
 Leak POE the most highly ranked locations through signal reviews. If the
1.E+00 signals show no signs of growth, the corrosion growth rate used
1.E‐01
1.E‐02 in the POE analysis can be reduced for the joint in question and
1.E‐03 the analysis can be repeated.
1.E‐04
1.E‐05 Acceptance Limits
1.E‐06
An important part of using POE results is setting allowable
Likelihood

1.E‐07
1.E‐08 leak and rupture POEs. To provide guidance, data on reported
1.E‐09
1.E‐10
incident rates for U.S. gas and liquid transmission pipelines were
1.E‐11 reviewed between year 2001 and year 20205. In this period, there
1.E‐12 were, on average, 625 incidents per year (see Figure 10) on the
1.E‐13
1.E‐14 U.S. transmission pipeline network, which is comprised of about
1.E‐15 550,000 miles. Dividing the annual incident count by the
1.E‐16
2020 2022 2024 2026 2028 2030 2032
pipeline mileage gives about 1.1E-3 incidents per mile-year.
Year A pipeline operator might select target maximum allowable
Figure 8. Example Leak POE Results for 12 Joints leak and rupture POE values several orders of magnitude lower
than the average number of incidents per mile-year. An
Families of joint POE curves like those shown in Figure 8 acceptance limit of 1E-4 incidents per pipeline segment per year
can be prepared and used to rank pipe joints in terms of their (not per mile) seems reasonable. Acceptance limits for pipe
impact on the likelihood of exceeding either the leak or rupture joints and for individual anomalies should be less. A maximum
critical limits in any given year. Depending on the year chosen allowable POE of 1E-6 incidents per anomaly-year has been
(e.g., three or five years after the most recent inspection), the used, as well as other limits.
ranking order could be different. This occurs when one or more
POE curves cross each other.
Finally, joint POEs can be combined to create cumulative
POEs for the entire line segment or other desired length of pipe,
just as individual anomaly POEs were combined to create joint
POEs. Figure 9 shows a typical result. In this case, the
cumulative leak and rupture POEs are much higher than those
shown earlier in Figures 6 through 8. This is the result of
statistically adding the POEs using Equation (8) from numerous
joints on a long pipeline.
Cumulative Leak and Rupture POE Figure 10. PHMSA Pipeline Incident Counts (2001-
1.E+00 2020)
1.E‐02

1.E‐04 4. EXAMPLE CASE STUDY

1.E‐06
The authors have analyzed many pipeline segments for
multiple operators using this POE methodology. Figure 11
Likelihood

1.E‐08
illustrates the leak and rupture POEs for the top joints on one of
1.E‐10 the segments previously analyzed. The starting leak POE (in year
1.E‐12 2021) is approximately 5.1E-4, which is higher than the
1.E‐14
acceptance limit established by the operator.
To determine where to conduct additional analyses or
1.E‐16
2020 2022 2024 2026 2028 2030 2032 excavations, the authors identified the pipe joints that largely
Year controlled the leak and rupture POEs. Figure 11 shows leak POE
Pipeline Cumulative Leak POE Pipeline Cumulative Rupture POE curves for 25 of the highest ranked joints on the segment, based
on leak POEs five years after the initial inspection (i.e., in 2026).
Figure 9. Example Cumulative Leak and Rupture POE
The dashed curve in Figure 11 represents the cumulative leak
Results for a Pipeline Segment POE for the segment and is identical to the leak POE curve seen
in Figure 9. In Figure 11, there are three joints with leak POEs
Use of Results above 1E-6 in 2026. Remediating these joints could substantially
The POE method described above allows an operator to reduce the segment’s leak POE, as discussed below.
identify and rank “critical” anomalies and/or joints within the
pipeline system. These critical joints can then be addressed either
through excavation and remediation or by further analysis. A
valuable step is to verify that growth is actually taking place at

8 © 2022 by ASME
 Leak POE POEs; (4) use the results to rank pipe joints for remediation; and
1.E+00 (5) establish maximum allowable leak and rupture POE values.
1.E‐01
1.E‐02 Table 2 summarizes the differences between traditional POE
1.E‐03 methods and the method described in this paper.
1.E‐04
1.E‐05 Top 3 Joints
1.E‐06 Table 2. Differences between Traditional and Advanced POE
Methods
Likelihood

1.E‐07
1.E‐08
1.E‐09
Traditional Advanced
1.E‐10
1.E‐11 Area POE POE Comments
1.E‐12 Based on Based on ILI
1.E‐13 overall ILI accuracy by Proximity to
1.E‐14 ILI errors
accuracy or POF category welds can have
1.E‐15 and standard
ILI plus proximity a significant
1.E‐16
deviations
2020 2022 2024 2026 2028 2030 2032 Performance to welds and effect
Year
Specification velocity
Figure 11. Leak POE Curves for Joints in a Pipeline Critical flaw
Segment curve for leak
Fixed depth
and rupture;
for leak;
Figure 12 shows the results of remediating the top three leak and
critical flaw Large impact on
joints based on their leak POEs in 2026. Remediating a joint rupture based
curve for leak POE;
removes it from the POE calculations and reduces the cumulative Critical flaw on leak/rupture
rupture; all Moderate
POE (black dashed curve). In this case, by remediating these dimensions length at the
anomalies impact on
local MAOP;
three joints, the operator was able to reduce the expected have both a rupture POE
all anomalies
likelihood of experiencing a leak from 5.1E-4 before remediation leak and
have either a
to 3.5E-6, a decrease of over two orders of magnitude. rupture POE
leak or rupture
Leak POE
POE
1.E+00 Based on
1.E‐01 New Cumulative  CGReal
Deterministic
1.E‐02 Segment POE analysis, which
1.E‐03 and constant Increasing
produces
1.E‐04 Corrosion with time; impact on POEs
means and
1.E‐05
growth rates typically the as time
1.E‐06 standard
same for all increases
deviations on a
Likelihood

1.E‐07
1.E‐08 anomalies
1.E‐09
joint by joint
1.E‐10 basis
1.E‐11 Provides better
1.E‐12
1.E‐13
insight and
1.E‐14 accuracy as to
1.E‐15 actual corrosion
1.E‐16 growth. May
2020 2022 2024 2026 2028 2030 2032
reduce
Year Signal review
conservatism
Figure 12. Leak POEs after Remediating Three Joints performed on
Not included associated with
joints with
ILI signal in a spreadsheet
Similar analyses were performed on the cumulative rupture review traditional
critical POE
based and/or
POE for the segment. In this case, there were more pipe joints values to
analysis POE assumed growth
that significantly affected the calculated rupture POE. confirm (or
anlaysis. rates. May also
Nonetheless, a similar process was used to identify joints for negate)
identify areas of
urgency.
remediation. accelerated
corrosion that
5. SUMMARY AND CONCLUSIONS have been
The POE method described in this paper includes a number missed by
spreadsheets
of refinements not used in traditional POE evaluations. Most
alone.
important are refinements to (1) address ILI inaccuracies as a Growth in Implied
function of POF category, proximity to welds, and sensitivity to Explicitly More accurate
length when Pburst is
tool velocity; (2) establish corrosion growth rates based on a included POEs
direction used
comprehensive ILI run-to-run comparison; (3) explicitly
consider depth and length growth when evaluating rupture

9 © 2022 by ASME
 Traditional Advanced lic%20Website%2F_portal%2FSC%20Incident%20
Area POE POE Comments Trend&Page=All%20Reported.
Calculated
Anomaly using Same as in
POE cumulative Traditional
calculations distribution approach
functions
Statistically
Cumulative Same as in
sums
POE Traditional
individual
calculations approach
POE values
Based on
Based on
anomaly and
anomaly
joint POE More focused
Corrosion POE values
values as well remediation
Management and overall
as overall programs
(segment)
(segment)
POEs
POEs

A case study demonstrated the utility of the POE analysis

methodology. Based on the calculated POE values, the leak POE
dropped by over two orders of magnitude based on remediation
of three pipe joints. Similar improvements can be expected when
considering rupture POEs.

6. REFERENCES
1. Thomas A. Bubenik, et al, “Advanced Probability of
Exceedance Method Overcomes Previous
Weaknesses,” Pipeline Pigging and Integrity
Management Conference, February 2-4, 2022.
2. Rafael G. Mora and Curtis Parker, “Probability of
Exceedance (POE) Methodology for Developing
Integrity Programs Based on Pipeline Operator-
Specific Technical and Economic Factors,” 4th
International Pipeline Conference, Paper Number
IPC2002-27224, 2002.
3. Dennis C. Johnston and Carolyn E. Kolovich, “Using
a Probability Approach to Rank In-Line Inspection
Anomalies for Excavation and for Setting
Reinspection Intervals,” 51st Annual Pipeline
Conference & Cybernetics Symposium (API), April
18-20, 2000.
4. “A Modified Criterion for Evaluating the Remaining
Strength of Corroded Pipe," Battelle Memorial
Institute, 1989.
5. 1 E.S. Graf, M.A. Ellinger, S.G. Hickey, W. V.
Harper, P.J. Moreno, and A. Nenciu “A New
Approach for Establishing Corrosion Growth Rates
Based on Subsequent In-Line Inspections”, Pipeline
Pigging and Integrity Management Conference,
Proceedings of the 2021 Pipeline Pigging and
Integrity Management Conference, 2021.
6. https://portal.phmsa.dot.gov/PDMPublicReport/?url
=https://portal.phmsa.dot.gov/analytics/saw.dll?Por
talpages&PortalPath=%2Fshared%2FPDM%20Pub

10 © 2022 by ASME