Paper No.

07120

CALCULATION OF A CORROSION RATE

USING MONTE CARLO SIMULATION

Dr. John F. Kiefner and Kolin M. Kolovich

Kiefner & Associates, Inc.
585 Scherers Ct.
Worthington, OH 43085

ABSTRACT

This document describes a means of calculating a credible corrosion rate for establishing
integrity reassessment intervals for a pipeline that has sustained corrosion-caused metal loss. The
method uses corrosion pit depths determined at a point in time by either direct measurement or inferred
indirectly from inspection data or a combination of both. The history of the pipeline can be used to
determine logical points in time for the corrosion to have initiated. Probability density functions are
used to represent the time of corrosion initiation and pit depth data, and through a Monte Carlo
simulation the 80 percent upper confidence bound corrosion rate is determined.

Keywords: corrosion rate, Monte Carlo, reassessment interval, pipeline integrity

Corrosion of a pipeline has typically been observed to occur in the form of localized pitting at
areas where the protective coating has become breached or has disbonded from the pipe and where the
levels of cathodic protection have proven insuficient to mitigate corrosion. The amount and
distribution of metal loss may be expected to vary from point to point along the pipeline, and much of
the pipeline may remain unaffected by corrosion. The prudent pipeline operator strives to mitigate
corrosion by maintaining adequate cathodic protection. However, while above-ground electrical
measurements are often quite helpful in determining where corrosion could have taken place or may be
taking place, periodic assessments of the amounts and distributions of metal loss are needed to ascertain
the extent to which the pressure-carrying capacity of the pipeline has been compromised. To cany out
such assessments at appropriate intervals, that is, intervals sufficiently short to prevent leaks or ruptures
Copyright
©2007 by NACE International. Requests for permission to publish this manuscript in any form, in part or in whole must be in writing to NACE
International, Copyright Division, 1440 South creek Drive, Houston, Texas 777084. The material presented and the views expressed in this paper are
solely those of the author(s) and are not necessarily endorsed by the Association. Printed in the U.S.A.

1
from metal loss but not excessively short so as to make operation of the pipeline uneconomical, a
pipeline operator needs to determine a reliable estimate of the rate of corrosion.
The determination of an appropriate reassessment interval requires a mathematical model
relating remaining pressure-canying capacity to the depth of penetration and axial extent of the worst-
case area of metal loss on the pipeline, knowledge of the approximate dimensions of the metal loss, and
the time rate at which the metal loss is increasing. Adequate models for predicting the remainin
pressure-canying capacity are readily available (e.g., ~ s m ~ ~ l - ~ o d B3 i f IG,
i e API 5
d RP 579 ). The
approximate dimensions of metal loss on a pipeline at a given point in time can be assessed with
reasonable accuracy by means of in-line inspection (e.g., a high-resolution magnetic flux leakage tool)
or by direct observation at an adequate number of excavations selected by the techniques described in
NACE Standard RP0502-2002~(External Corrosion Direct Assessment). The third element needed, the
corrosion rate, can be difficult to assess accurately. Yet, it is a critical element because estimating a too-
low rate could result in a pipeline leak or mpture while estimating a too-high rate could result in
unnecessary assessments being camed out.
Corrosion rates can be estimated in a number of ways such as comparing successive in-line
inspections, making actual measurements of corrosion rates, or using the default rate suggested in
NACE Standard RP0502-2002. Each of these methods has drawbacks. Comparing successive in-line
inspections is difficult and time-consuming, and the operator may not have successive inspections that
can be readily compared. Measuring actual rates at particular points on the pipeline may not represent
the rates at other points on the pipeline, and the default rate suggested in NACE Standard RP0502-2002,
16 mils per year (0.41 mmly), likely would result in excessively short re-assessment intervals for a
..
Dineline subiected
" to even moderately effective cathodic protection. To - get around these limitations, it
is common practice to estimate the corrosion rate for calculating re-assessment intervals by considering
the operating history of the pipeline and any knowledge of its condition such as prior in-line inspections
or direct observations of corroded areas on the pipeline. The simplest method is to assume that the
deepest pit started corroding at a particular point in time and that the rate of corrosion was constant.
Dividing the pit depth by the time interval between the present and the start of the corrosion gives the
rate of corrosion. Common assumptions of starting time include, the year of construction (the least
conservative) or some other year after which it is conceivable that cathodic protection has significantly
mitigated the corrosion. Rates determined in this manner will work if the history of corrosion and
cathodic protection on the pipeline is well-understood, but the simple process just described inherently
involves an unknown degree of uncertainty. A pipeline operator can use this simple approach in a more
sophisticated way with the aid of Monte Carlo simulation that results in far less uncertainty.

MONTE CARLO METHOD

Generically, a Monte Carlo method is any technique that uses statistically sampled values to
approximate the solution to a quantitative problem. This method is particularly useful in problems
where there is a known uncertainty associated with the input parameters. A Monte Carlo simulation can
approximate the solution to such a problem by sampling the uncertain input parameters from a range or
distribution of possible values. The values sampled &om the input distributions are then used to
calculate the desired output or solution to the problem. This process of sampling input values and
calculating output solutions is repeated hundreds or thousands of times resulting in a distribution of
output values. The output distribution can then be used to determine either the most likely outcome, or a
range of outcomes that has a specific level of confidence attached to it. This confidence level, reported
as a percentage, represents how certain one can be that the outcome will fall within the specified range.

2
 INPUT VARIABLE DISTRIBUTIONS

The information needed to establish a corrosion rate via a Monte Carlo simulation includes a
distribution of pit depths and a distribution of times when the corrosion might have initiated. A pit
depth distribution from a prior in-line inspection can be used, but the distribution of tool error should be
included whether that distribution comes from actual comparisons of "called" versus measured pits or
from the vendor's advertised accuracy. The tool error distribution is used as a scale factor to adjust
sampled pit depths to account for the uncertainty in the pit depth measurements. Alternatively, a pit
depth distribution may be created from values measured directly at excavations provided that these are
considered to be representative of corrosion on the pipeline. In either case, a custom-tailored continuous
distribution can be fit to the data using the chi-squared goodness-of-fit statistic as a measure of how well
the distribution represents the data.
A logical corrosion initiation distribution can be defined using historical knowledge of a
pipeline. If no knowledge is available, a uniform distribution representing equal probability of corrosion
initiating at any time during a pipeline's life can be used.

EXAMPLE CALCULATION

A corrosion rate is calculated for a 147-mile segment of a 36-year old 30-inch-OD, 0.375-inch-
wall, X52 pipeline based on in-line inspection data that gave pit depths for 130 called anomalies. Tool
*
tolerance on depth accuracy was assumed to be depth 15% of wall thickness 95% of the time, and a
normal distribution of error was assumed. The assumed distribution of corrosion starting times is
uniform, meaning that the corrosion could have started at any time during the 36-year history of the
pipeline. Note that other distributions could be constructed to better fit the actual corrosion and cathodic
protection history of a particular pipeline. The relevant input distributions are shown in Figures 1
through 3 below.

Figure 1 Distribution of Pit Depths

3
 Tool Tderance Factor

Figure 2 Vendor's Advertised Tool Tolerance

Figure 3 Uniform Dishibution of Corrosion Starting Times

4
 As shown in Figure 1, an extreme value distribution of pit depths was assumed. This distribution
is used in this example for demonstration purposes as it is apparent even to a casual observer that the fit
is not good because of the two-peaked distribution. The chi-squared value for this fit is 13.2. For 11
degrees of freedom (based on subtracting 1 from the 12 bins in the histogram of data), one could say
with 99.5 percent confidence that the distribution of pit depths does not fit the assumed distribution.
The problem with trying to fit these data with one distribution is that the pits are spread over a distance
of 147 miles (237 km). Looking at the actual data by milepost (not done in this example), one would
immediately see that the 147 miles (237 krn) should be broken into three or more segments with more
commonality with respect to pit depth distributions. The outcome of the analysis correspondingly would
be a representative corrosion rate for each segment. In an actual analysis this approach should be taken
whereas in this document the purpose is merely to demonstrate how the process works.
The tool error distribution is shown in Figure 2. In this case the vendor's advertised tolerance
was used. When field verification data are provided, a " u n i ~ ' plot
' comparing field-measured to tool-
called depths is used to develop a mean and standard deviation for particular tool's error margin.
The distribution of corrosion starting times is shown in Figure 3. In this case the uniform
distribution signifies that equal probability is assigned to each of the 36 years of operation.
Distributions of other shapes can be established based on historical data associated with the corrosion
and cathodic protection history of a particular pipeline.
A commercially-available software product is utilized to conduct a Monte Carlo simulation
involving repeated calculations each taking a pit depth at random from the distribution shown in Figure
1. The sampled depth is adjusted based on the tool error probability distribution shown in Figure 2. The
value sampled from the tool error distribution serves to increase or decrease the sampled pit depth based
on the assumed tool accuracy of 15% of wall thickness 95% of the time. The adjusted pit depth is
divided by a "years-of-active-cornsion" value chosen at random from Figure 3. The result is the
distribution of corrosion rates shown in Figure 4.

Qw& Rahl (mils perpar)

r~
,5977
I
112411
2.u% ,
Figure 4 Distribution of Corrosion Growth Rates

5
 Because the answer is a distribution, not a specific value, one must decide how to select a single
credible value from the distribution. A precedent for a credible value is the default rate given in
Appendix D of NACE RP0502-2002 (ECDA). Quoting from the Standard, the default rate of 16 mils
per year (0.41 mrnly) "... represents the upper 80% confidence level of the maximum pitting rates for
long-term (up to 17-year duration) underground corrosion tests of bare steel coupons without CP in a
variety of soils including native and nonnative backfill." Therefore the representative credible rate is
selected as the upper 80% confidence level of the distribution derived from the Monte Carlo simulation,
i.e., from Figure 4. For this example, that rate is 11 mils per year (0.28 mmly).
It is noted that an improved assumption about corrosion starting times can be used if more
information about the history of the pipeline is available. For example, no cathodic protection was
installed for the first year of operation. On this basis it is assumed that the likely starting time is at a
maximum for the year of installation. For the next 17 years the cathodic protection system was used but
not rigorously monitored and upgraded. On this basis it is assumed that the likelihood of corrosion
starting anytime in this interval is 50% of the maximum. After that time the operator provided rigorous
monitoring of pipe-to-soil potential levels and added cathodic protection where required to maintain a
target level consistent with good industry practice. On this basis it is assumed that the likelihood of
corrosion starting anytime is 20% of the maximum for the most recent 18 years. An overriding
requirement is that the area under the curve in the probability density function is unity indicating a
certainty that corrosion did start at some time during the life of the pipeline.
The representative rate drops to 6.7 mils per year (0.17 mrn/y) under these assumptions. The
custom distribution of corrosion starting times is shown in Figure 5 and the resulting corrosion rate
distribution is shown in Figure 6. The results of this analysis show that if the history of the pipeline is
documented in a manner that helps establish more definitive starting times, a more favorable answer is
obtained than when one assumes the uniform distribution shown in Figure 2.

2.60 35.68
Figure 5 Custom Distribution of Corrosion Starting Times

6
 -Rwmi*ww)
n
,5289
I
6.6313
a7.4 ,
Figure 6 Distribution of Corrosion Rates Using the Custom Starting Times Distribution

CONCLUSION

The method described in this document for determining a corrosion rate for a buried pipeline can
be considered a viable alternative to other methods such as consecutive in-line inspection comparisons
or calculating a corrosion rate based on the deepest pit and a single assumed corrosion starting time. In-
line inspection comparisons require multiple inspections and can be difficult and time consuming.
Determining a corrosion rate based on the deepest pit and an assumed starting time considers only one
possible solution. Using Monte Carlo simulation, distributions of pit depths and corrosion starting times
calculate a distribution of corrosion rates. A corrosion rate for determining reassessment intervals can
then be established as the upper bound confidence level of 80%. This method can also take into account
uncertainty in pit depth measurements by representing this uncertainty by a probability density function.
The two examples presented herein utilize actual in-line inspection pit depth data and a general
tolerance on the accuracy of those measurements. In both cases, a corrosion rate less than the default
rate given in Appendix D of NACE RP0502-2002 is determined. For the case where more information
on the history of the pipeline is used to create a more favorable but realistic custom distribution of
possible corrosion starting times, the determined corrosion rate is lower.

REFERENCES

(1) Kieher, J. F., Vieth, P. H., and R o m a n , I. "ContinuedValidationof RSTRENG",PRC

International, Catalog No. L5 1749 (1996).
(2) "Fitness-for-Service",American Petroleum Institute Recommended Practice 579,First Edition
(January 2000).
(3) Standard Recommended Practice, Pipeline External Corrosion Direct Assessment Methodology,
NACE Standard RP0502-2002