RISK ASSESSMENT OF MODERN PIPELINES

James N. Mihell
Dynamic Risk, Ltd.
1324 17
th
Ave. S.W.
Calgary, Alberta, T2T 5S8,
Canada
Cameron Rout
Dynamic Risk, Ltd.
1324 17
th
Ave. S.W.
Calgary, Alberta, T2T 5S8,
Canada


ABSTRACT
Proponents of new pipeline projects are often asked by
regulators to provide estimates of risk and reliability for their
proposed pipeline. On existing pipelines, the availability of
operating and assessment data is generally considered to be
essential to the task of performing an accurate and defendable
risk or reliability assessment. For proposed or new pipelines,
the absence of these data presents a significant challenge to
those performing the analysis. The reliance on industry incident
data presents problems, since the vast majority of loss-of-
containment incidents relate to older pipelines in which the
design, routing criteria, material properties, material
manufacturing processes, and early operating practices differ
significantly from those that are characteristic of modern
pipelines. As a consequence, much of the available failure
incident data does not accurately reflect the threats or the
magnitudes of the threats that are associated with modern
pipelines. In order to address this problem, 'adjustment factors'
are often applied against incident data to try to account for
threat differences between the source data and the intended
application. The selection of these adjustment factors can often
be quite subjective, however, and open to judgment; therefore,
they can be difficult to justify.
With the rapidly growing practice of regular in-line inspection
(ILI) on transmission pipelines, an extensive repository of ILI
data has been accumulated - much of it relating to modern
pipelines. Through the judicious selection of source data, ILI
data sets can be mined so that an analogue data set can be
created that constitutes a reasonable representation of the
attributes of reliability of a specific new pipeline of interest.
Key reliability properties, such as tool error distribution, feature
incidence rate, feature size distribution, and apparent feature
growth rate distribution can be derived from such analogue
data. By applying these reliability properties in an analysis
along with known pipeline design and material properties and
their associated distributions, and by taking consideration of
planned inspection intervals, a reliability basis can be derived
for estimating pipeline risk and reliability. Estimates of risk and
reliability that are derived in this manner employ
methodologies that are repeatable, defendable, transparent, and
free of subjectivity.
This paper outlines an approach for completing risk and
reliability estimates on new pipelines, and presents the results
of some sample calculations. The reliability estimates
illustrated are based on an approach whereby corrosion feature
size and growth rates are obtained from analogue ILI datasets,
and treated as random variables. In that regard, they constitute
the probability of exceeding a limit state that represents an
approximation of the condition for failure.

INTRODUCTION
One of the challenges of employing a quantitative risk
assessment on a new pipeline is that industry failure statistics
are not directly applicable to modern pipeline designs,
materials, and operating (i.e., assessment) practices. A review
of industry failure statistics indicates that the vast majority of
pipeline failures occur on pipelines that were installed in the
1970s or earlier (1, 2). These pipelines where largely
developed prior to the advent of several risk-critical
technologies, such as:
- High-performance coating systems, such as three-layer
coatings and fusion bonded epoxy coating systems,
- Design-phase identification of interference effects and
development of mitigation plans through diagnostic
testing of cathodic protection systems, and
- Design-phase identification of internal corrosion threat
factors and the design of mitigation plans through
internal corrosion modeling.
Proceedings of the 2012 9th International Pipeline Conference
IPC2012
September 24-28, 2012, Calgary, Alberta, Canada
IPC2012-90072
1 Copyright 2012 by ASME

- Design-phase identification of geotechnical threats,
and mitigation through threat avoidance
Because of these factors, the use of historical incident data
is not a sound foundation for estimating failure frequency in
modern pipelines.
The use of statistical analysis and reliability methods,
supported by high-resolution in-line inspection data has become
common practice in making failure likelihood estimates for
existing pipelines in which such data exists (3, 4). As these
methods intrinsically address variables such as wall loss feature
incidence rate, growth rate and size distributions, they can be
employed to make estimates of reliability and failure likelihood
for both external corrosion and internal corrosion. By carefully
selecting ILI data sets to address conditions that influence
factors that affect these variables, an analogue data set can be
obtained that can be superimposed on the design conditions
representing a new pipeline. In this manner, estimates of
reliability and failure likelihood can be derived for the new
pipeline for any future timeframe, given exposure to similar
corrosive environments and the implementation of similar
protective measures.
NOMENCLATURE
ILI In-Line Inspection
Reliability Methods A statistical approach used to
estimate the probability of a system
or structure completing its expected
function during an interval of time.
PDF Probability density function: The
density function of a continuous
random variable
CDF Cumulative distribution function:
The function that describes the
probability that a random variable
will be found at a value less than or
equal to a set value.
MIC Microbially Induced Corrosion
Monte Carlo A class of computational algorithms
that rely on repeated random
sampling to compute results.
SELECTION OF ANALOGUE ILI DATA
The reliability approach described in this paper employs
the superposition of an analogue ILI dataset upon the
contemplated new pipeline design. In this respect, the
reliability analysis of a new pipeline using analogue ILI data
essentially models how the new pipeline materials and design
responds to an anticipated degradation process and establishes
how quickly the reliability degrades to a point where failure
likelihood becomes significant. For this reason, it is essential
that the analogue ILI dataset is representative of the
degradation process, including defect incidence rate, defect size
distribution, and defect growth rate distribution.
Several factors must be considered in selecting an
appropriate ILI dataset that can be held as being representative
of corrosion performance anticipated for a new pipeline. First
and foremost among these factors is the quality of the ILI
dataset itself. Tool quality factors such as detection reliability
and sizing accuracy on both length and depth should be
reviewed to ensure that they represent the current state of the
art in tool performance. Ultimately, as will be discussed later,
tool error plays a critical role in making reliability predictions,
and the use of data that is associated with a high tool error
distribution may give rise to an excessively conservative result.
ILI data should be reviewed to identify and remove wall
loss data that is not attributed to active corrosion. Otherwise,
the presence of wall loss that is associated with benign features,
such as manufacturing imperfections might result in a gross
over-estimation of corrosion feature density. Furthermore, non-
zero wall loss rates will be inferred from the presence of such
benign features, and those wall loss rates will be assigned to
what are in fact static features. The misinterpretation of benign
wall loss features in this fashion will result in an overly-
conservative estimate of pipeline reliability.
When evaluating internal corrosion susceptibility, one of
the simplest methods to perform screening is to view
orientation charts for internal wall loss features. Where water
drop-out and accumulation is an essential aspect of the internal
corrosion mechanism that is associated with the product and
flow characteristics being considered, wall loss that is
associated with internal corrosion should be expected at the
bottom of the pipe, as is illustrated in Figure 1. This is
especially true where concentrations of internal wall loss can be
seen to coincide with steeper pipeline inclination angles or
receipt points.


Figure 1. Internal Wall Loss Characteristic of Internal
Corrosion

On the other hand, a random distribution of internal wall
loss features around the circumference of the pipe, with no
apparent trends relative to inclination angle or receipt points
might be more representative of benign manufacturing
imperfections, as is represented in Figure 2.

2 Copyright 2012 by ASME


Figure 2. Random Internal Wall Loss Characteristic of
Manufacturing Imperfections

One effective method that can be used to screen for active
wall loss is to use data derived from pit-matching of separate
in-line inspections. Figures 3 through 6 are derived from ILI
data that is representative of the same section of pipeline.
Figure 3 shows the distribution of apparent external wall loss
rates obtained by pit matching ILI data taken four years apart.
When plotted as a probability density function, Figure 4
illustrates that the external wall loss is, in this example,
normally distributed, with a mean of 0.00 mm/yr, and a
standard deviation of 0.01 mm/yr. The inclusion of negative
growth rate values over half the distribution is indicative of the
fact that the apparent feature growth rate is insignificant
relative to pit matching error. Figure 5 shows the distribution
of apparent external wall loss rates for the same section of
pipeline that is inferred from only one set of ILI data. In this
case, estimates of growth rates were derived by assuming linear
growth to the date of in-line inspection. This is generally
considered to be a conservative assumption, since it ignores the
tendency for growth rates to attenuate over time due to
polarization caused by the build-up of corrosion products.
When plotted as a probability density function, Figure 6
illustrates that the external wall loss is, for this example, best
represented as a Weibull distribution, having a mean of 0.03
mm/yr and a standard deviation of 0.02 mm/yr. It is apparent
from a comparison of Figures 3/4 with Figures 5/6 that in this
case, the vast majority of external wall loss features are
dormant (quite possibly benign manufacturing wall loss
features), and if only un-matched data were used as the basis of
the analysis, this would result in highly conservative reliability
results for the pipeline.




Figure 3. Apparent External Wall Loss Rate Based on
Matched ILI Data



Figure 4. Apparent External Wall Loss Rate
Distribution Based on Matched ILI Data



Figure 5. Distribution of Apparent External Wall Loss
Rate Based on Single Set of ILI Data


3 Copyright 2012 by ASME


Figure 6. Apparent External Wall Loss Rate
Distribution Based on Single Set of ILI Data


Beyond the quality of ILI data, several other factors must
be considered in selecting an appropriate ILI dataset. For the
evaluation of external corrosion threats, factors such as coating
type, coating specification, operating environment, and
cathodic protection standards and history should be evaluated
to confirm that those conditions are similar between those
represented by the analogue ILI dataset and those that are
proposed for the new pipeline.
For reliability estimates of the threat of internal corrosion,
an evaluation should be made of the corrosion susceptibility of
the pipeline from which analogue ILI data is being considered,
relative to the corrosion susceptibility that would be
characteristic of the new pipeline. A variety of internal
corrosion evaluation techniques exist, and while it is not the
intent of this paper to delve into these evaluation techniques in
detail, some mention is made here of factors that require
consideration.
Internal corrosion evaluation techniques are largely based
on product stream characteristics and flow rates. For liquid
products, the important parameters that should be included in a
comparison of Corrosivity are water content, erosion and
erosion/corrosion, flow velocity, temperature, susceptibility to
under-deposit corrosion (solid deposition, MIC potential, and
water chemistry), and mitigation measures (use of inhibition,
biocides, or pigging). In order to ensure that the corrosion
mechanism and corrosivity that is represented by the analogue
ILI dataset is representative of that which would be expected in
the new pipeline, an evaluation of all of these parameters must
be conducted.
An evaluation of water content should include
consideration of bottom sediment and water (BS&W), as well
as characterization of the flow regime as either turbulent (in
which water will be entrained) or lamellar (in which there may
be opportunities for water stratification and accumulation).
Consideration of solids should include an evaluation of the
potential for settling of the solids, given consideration of flow
velocity and flow properties.
Microbially Induced Corrosion (MIC) can greatly enhance
corrosion rates, especially in conjunction with solid deposition,
and the temperature range 15
o
C - 70
o
C is known to be
associated with increased MIC activity, leading to localized
attack.
Besides MIC activity, chloride concentration also plays an
important role in the overall corrosion and under-deposit
mechanisms. In respect of the overall corrosion rate, chlorides
increase the conductivity of the corrosive solution due to a
change in ionic strength, leading to an increase in the corrosion
rate. In the under-deposit corrosion mechanism, chlorides can
promote the removal of protective scales (if there are any),
leading to localized attack. In addition, strong changes in
temperature might form hydrochloric acid from the reaction of
the chlorides with water vapour. However, pipeline operators
have not experienced this phenomenon with temperature
changes below 150 C.
Sulphur may be co-produced with sour gas and sour oil.
As a result of pressure and temperature changes (in most cases
a reduction) in the flowing conditions or in shut-in conditions,
sulphur may precipitate from polysulphides and cause plugging
and / or corrosion problems. From the corrosion viewpoint,
sulphur can increase corrosion in many ways, including
compromising protective iron sulphide layers, enhancing
cathodic reactions, and impairing inhibitor performance.
Erosion-corrosion could be a factor that enhances wall loss
rates if the flow velocity is higher than approximately 3 m/s,
and BS&W exceeds 0.5%.
Changes in temperature can lead to a rapid increase in the
corrosion rate if the corrosion mechanism is controlled by
diffusion. However, in a pipeline environment, temperature
also affects other important driving parameters, such as scale
deposition, chemical reactions rates, microbiological activity,
and presence of undissociated organic acid (short chain e.g.
acetic acid) .

ANALYSIS APPROACH
The above section described screening considerations for
selecting an appropriate analogue ILI dataset that is
representative of the corrosion mechanism and severity that is
expected on the new pipeline. This screening and selection of
appropriate analogue ILI data must be completed separately for
internal corrosion and external corrosion. Once this has been
completed, the analogue ILI dataset can be incorporated into
the procedure described in this section in order to establish
estimates of pipeline reliability as a function of year of
operation.
A Monte Carlo approach has been developed to assimilate
distributions derived from size and growth rate distributions
derived from the analogue ILI dataset, and to apply those
distributions against a failure limit state for volumetric wall
loss. In the example described here, the failure limit state
function is the modified ASME B31.G criterion, which, for the
purposes of the analysis, is rearranged to determine depth at the
limiting condition:
4 Copyright 2012 by ASME


( )
( )
(
(
(
(
(

|
|
|
|
|
.
|

\
|
|
|
.
|

\
|

=
o
o
o o
M
t
t . d
op
op
f
85 . 0
, 8 0 MIN
(1)
Where,
d
f
= depth at the limiting condition;
t = Wall thickness;

op
= Operating Stress;
o = Flow stress;

2
2 2
003375 . 0 6257 . 0 1
|
.
|

\
|
+ =
Dt
L
Dt
L
M

) 50 (for Dt L s

Dt
L
M
2
032 . 0 3 . 3 + =

) 50 (for Dt L >

L = Defect length (which is described as a
probability distribution characteristic of the dataset,
incorporating tool error on length measurement)

In the Monte Carlo analysis, the variables of pipe diameter,
wall thickness, yield strength, and operating pressure that are
specific to the new pipe design are considered dynamic
segments. Each dynamic segment of the new pipeline for
which the risk analysis is being evaluated requires a separate
reliability analysis. Corrosion feature incidence rates, and the
distribution parameters for corrosion feature length and depth
are determined from the analogue ILI data, as are corrosion
feature growth rates. When using ILI data for the purposes of
establishing these parameters, it is important to recognize that
the quantities derived represent values at a particular point in
time (i.e., the date of last inspection). Furthermore, these
quantities are subject to tool measurement error. Corrosion
feature size is therefore considered characteristic of the size
after some period of time. For example, when applied to a new
pipeline, the depth distribution must be adjusted downwards
(accounting for some assumed corrosion growth rate) when the
modeled pipeline age is smaller than that from which the
analogue ILI data was obtained. Similarly, the depth
distribution must be adjusted upwards when the modeled
pipeline age is larger than that from which the analogue ILI
data was obtained. This is illustrated in Figure 7, which shows
how the flaw distribution flattens and translates with time, t.
Specifically, as can be seen in this Figure, as time increases, the
mean of the flaw depth distribution increases, and that the
standard deviation of the flaw depth distribution also increases.


Figure 7. Illustration of How Flaw Depth Distribution
Changes With Time

In the absence of any other information pertaining to how
growth rate varies with time, a linear growth rate assumption
can generally be considered a reasonable, yet conservative
approximation, since it ignores the polarizing effects of the
accumulation of corrosion product.
The high-performance coating systems that are
characteristic of modern pipelines, such as fusion bonded epoxy
are not susceptible to time-dependent coating degradation to the
extent that older vintage coating systems are. Therefore, it is
often realistic to assume that any coating damage that is
inferred from the presence of a corrosion feature was created at
the time of installation, and that the arial extent of coating
damage, and hence the potential for increases in wall loss area
(i.e., length and width) does not change appreciably with time.
Similarly, the introduction of new corrosion features may be
limited by the absence of time-dependent coating breakdown.
In the Monte Carlo simulation, corrosion feature size, as a
function of time, is sampled stochastically, based on the
probability density functions for those parameters derived from
the analogue ILI dataset. A further stochastic adjustment is
made to account for the tool error associated with the ILI tool
from which the analogue data was derived. In performing this
stochastic adjustment for tool error, it is preferred to use
correlations derived from <Tool-Predicted> to <In-Ditch
Measurement> data pairs. When using this approach, it is
important to minimize variability of in-ditch measurements
through standardization of in-ditch measurement practices, and
the implementation of quality control measures. If direct
measurement of tool measurement error isnt available, the ILI
vendors published tool error can be used. For instance, tool
error on depth measurement is often characterized at 10% w.t.,
80% of the time. In statistical terms, this corresponds to a
normal error distribution having a mean of 0, and a standard
deviation of 7.8% of the wall thickness.
Assuming a linear growth model, the stochastically-
sampled flaw depth estimate is adjusted to account for the
difference between the age of the analogue pipeline at the time
that the ILI data was acquired, and the modeled age of the new
pipeline:

5 Copyright 2012 by ASME


ILI
A
o
A
o
T
T d
d

=
(2)

Where,
0
A
d
= Stochastically sampled flaw depth at the specific
time assumed in the analysis;

0
d = Stochastically sampled flaw depth, derived from the
analogue ILI dataset (incorporating stochastic
adjustment for analogue ILI tool error)
T
A
= Year of operation for the new pipeline that is being
assumed in the analysis
T
ILI
= Year of operation that the for the analogue pipeline
when the ILI assessment was completed.

For the purposes of the Monte Carlo simulation, all pipe
parameters that are contained in the limit state function shown
in Equation (1) (i.e., pipe wall thickness, operating stress level,
and flow stress) correspond to the new pipeline for which
failure probability values are being sought.
Limit state exceedance is predicted when the stochastically
sampled flaw size has a depth component that exceeds the
limiting condition (derived from Equation (1)). When the
Monte Carlo simulation is performed through multiple
iterations, the probability of limit state exceedance for the given
year of analysis is defined as the proportion of those iterations
that return a result that exceeds the limit state. This probability
is defined as the conditional probability, given the presence of a
corrosion feature, P
f,F
. The overall probability for a given
dynamic segment of the new pipeline in the year of operation
being considered in the analysis is defined as:

F e DS
ILI
N
ILI DS e
P L
D
D
P
, ,
=
(3)


DS e
P
,

= Probability per dynamic segment
ILI

= Corrosion feature density per unit length of pipeline

derived from the analogue ILI dataset
N
D = Diameter of the new pipeline
ILI
D
= Diameter of the pipeline from which the analogue
ILI data was derived
DS
L = Length of the dynamic segment in the new pipeline
F e
P
,
= Conditional probability, given the presence of a
corrosion feature


Note that in the above expression, the ratio: D
N
/D
ILI
is
employed in recognition of the assumption that the frequency
of occurrence of coating defects (and hence corrosion features)
will be proportional to the pipe surface area (with pipe surface
area being proportional to diameter).
By performing a separate analysis for each year of
operation, and for each dynamic segment, a profile can be
generated that represents the probability of exceeding the
reliability limit for each year of operation of the new pipeline,
an example of which is provided in Figure 8.


Figure 8. Frequency Profile Example

In order to support a risk analysis, the output from the
above analysis must be relevant to the consequence analysis.
Therefore, the results should specify more than frequency of
occurrence; instead, the frequencies of occurrence should be
tied to an outcome, with outcome being related to magnitude of
a potential release, and hence hole-size. The proportion of
ruptures can be derived by first calculating the critical through-
wall flaw size as a function of material properties and operating
parameters of the new pipeline. There are a variety of through-
wall fracture criteria relationships; an example of which is the
NG-18 flaw equation (5):

(

=

=
o
o t
t
o
2
sec ln
8 12
2
2
h T
c
v
c
M c
A
E C
K
(4)

The above relationship can be used to determine the
maximum size defect that will leak rather than rupture. At high
toughness values, it represents a flow-stress or plastic
instability criterion (typical of the failure mode of most
corrosion features), whereas at lower toughness values, it may
represent a conservative representation of the leak/rupture
boundary for corrosion features.
As is illustrated in Figure 9, the cumulative distribution
function for flaw length, derived from the analogue ILI dataset
is compared against the critical through-wall flaw length for the
new pipeline. Using this approach, the proportion of features
that have the potential to penetrate through-wall at a length
greater than the critical through-wall flaw length can
6 Copyright 2012 by ASME

conservatively be said to have the potential to fail in rupture
mode, while the remainder of the flaws will fail as leaks.





Figure 9. Determination of Fraction of Leaks and
Ruptures from Corrosion Feature Length CDF and
Critical Through-wall Flaw Size

The breakdown of leak sizes can be obtained from the
distribution of flaw areas (length x width) for those flaws that
are predicted to fail by leak mode. A reasonable, yet
conservative representation of the outcome associated with a
leak can be represented by the 50
th
percentile of flaw size area,
as is depicted in Figure 10.

Figure 10. Determining Hole Area From Distribution of
Flaw Area for Leak-Designated Corrosion Features
CONCLUSIONS
An approach based on probability of exceedance
methodologies has been advanced for estimating failure
likelihood to support risk assessments of new pipelines. This
approach addresses the problems that have historically been
associated with the use of failure incident data. These failure
incident data are unduly influenced by older vintage pipelines
for which the material performance and design features are not
a good representation of modern pipelines. The approach
described is transparent, defendable and repeatable, and for
these reasons is a sound basis for making realistic estimates for
the purposes of supporting risk assessments in modern
pipelines.
REFERENCES
1. Keifner, J.F., Trench, C.J., Oil Pipeline
Characteristics and Risk Factors: Illustrations from the
Decade of Construction API Publication, December,
2001.
2. U.S. Department of Transportation Pipeline &
Hazardous Materials Safety Administration Pipeline
Safety Stakeholder Communications, Significant
Incident Data, 1986-2001.
3. Chan, P.D., and Webster, D., Probabilistic
Assessment of ILI Metal Loss Features, ASME, 2010
International Pipeline Conference (IPC), IPC2010-
31298, September, 2010.
4. Mora R.G., Parker C., Vieth P., Delanty B.,
Probability of Exceedance (POE) Methodology for
Developing Integrity Programs Based on Pipeline
Operator-Specific Technical and Economic Factors,
ASME, 2002 Internal Pipeline Conference (IPC),
IPC2002-27224, October 2002.
5. Eiber, R.J., and Leis, B.N., Fracture Control
Technology for Natural Gas Pipelines Circa 2001,
PRCI Report No. PR-003-00108, July, 2001.
7 Copyright 2012 by ASME