You are on page 1of 5

ASSET INTEGRITY INTELLIGENCE

Pipeline Ultrasonic Thickness Data


Analysis Using Gaussian Distribution
Marcos Delgado, Independent Mechanical Engineer

VOLUME 27, ISSUE 2


MARCH | APRIL 2021
Pipeline Ultrasonic Thickness Data Analysis
Using Gaussian Distribution
Marcos Delgado, Independent Mechanical Engineer

Introduction
It is crucial to determine the overall mechanical integrity of our
pipelines. To achieve this, screening examination and inspection
techniques such as guided wave ultrasonics or intelligent pigging
systems have often been utilized. But those techniques can have
some limitations and are often supplemented with visual exam-
inations, as well as conventional spot UT thickness measure-
ments (typically using an A-scan UT instrument). Of course, all of
the data generated during these examinations must be compiled
and properly analyzed. The mechanical integrity evaluation of an
entire field, with a large number of pipelines, can require signif-
icant amounts of resources (i.e., time, economic, expertise) and
can generate large volumes of data to be analyzed. API 570 (fourth
edition, paragraph 7.1.3) allows the use of statistical analysis Figure 1. Normalized Gaussian curves with expected value µ and variance
σ2. (Source: Wikimedia Commons).
methods to establish a representative corrosion rate, remaining
life estimate and/or re-inspection date of piping systems.[1]
This article will discuss a statistical analysis method that was For the application of the Gaussian function, a spreadsheet for
developed to evaluate the integrity of pipelines that can identify, data treatment was developed, superimposing a distribution that
in a first phase inspection, whether the nature of the active dam- we will call “ideal” (blue) and another that we will call actual or
age mechanism(s) are uniform (general) or localized. It can also “real” (red). For the application of the developed method, the fol-
help to identify the most indicative areas, which helps us develop lowing conditions must be met:
a better inspection plan going forward and determine if more
1. Unpredictable corrosion mechanisms are not known in the
advanced (phase 2) inspections are needed. The analysis method
pipeline to be evaluated.
is based on the application of a Gaussian distribution, which
most correctly applies to the type of statistical population that is 2. The schedule (nominal wall thickness) or as-built wall thick-
generated with the application of the non-destructive ultrasonic ness of the pipe must be known.
test (UT) A-scan. The result is a Gaussian bell graph presenta-
3. UT A-scan data must be available for each section (tube) or
tion where the parameterization of the ideal pipe conditions are
accessory in the piping system.
superimposed onto the actual data, thereby clearly showing the
deviation of the actual data from the ideal condition. 4. A large sampling of data from the entire pipeline system, with
a minimum of four thickness measurement locations (TMLs)
Development of the Statistical Method for each condition monitoring location (CML) assigned in at
The proper interpretation of the results obtained through the least two places in the pipe, should be available.
application of the non-destructive evaluation (NDE) UT A-scan
In the event that no design specifications are available regarding
are subjective and reliant on the inspector's experience. Usually
thickness, an experienced inspector can estimate the schedule
low, medium, and high readings are obtained, with respect to the
based on the measurements taken, which requires additional
nominal thickness of the pipe. In view of the data volumes and
time and effort. For these cases, we developed an automated
diversity, the use of descriptive statistics was considered, specif-
search and assignment of the schedule based on the ASME B36.10
ically the applicability of a Gaussian distribution, since through
(Welded and Seamless Wrought Steel Pipe) tables.[2]
this it was possible to represent the behavior of the values of cer-
tain variables. In this case, the thickness variations can be influ- We established two data graphs. The “ideal” (blue) graph is made
enced by random phenomena, such as corrosion that is caused by with data generated randomly in the spreadsheet between 0% and
various damage mechanisms. 12.5%, taking as a reference what API 574 indicates in point 4.1.1,
which allows calculating the mean (a) and the standard deviation
The Gaussian distribution in its simple mathematical expression
(c) from the Gaussian mathematical expression.[3] The actual or
is represented as follows:
“real” (red) graph is the average of the thickness losses calculated
with respect to the nominal thickness of the pipe. To achieve
an adequate visual effect for both graphs, we anchor (b) as zero
... where a is the average, b is the center position of the hood, and
and leave (a) real as a constant. In both graphs, the constant (a) is
c is the standard deviation.

2 Inspectioneering Journal MARCH | APRIL 2021


Figure 2. Spreadsheet used to apply the Gaussian distribution.

calculated based on the real data of the pipeline, which allows us


to visually superimpose the results onto one another.

Application of the Statistical Method


The statistical method allows one to quickly and efficiently iden-
tify pipes with thinning that requires a second phase of more spe-
cialized NDE. Figure 2 shows an example of a spreadsheet used
to apply the Gaussian distribution.
Next, results of the application of the Gaussian-based statistical
method are presented, which illustrate whether or not pipelines
need attention regarding their localized or generalized in-service
thinning. Graphically, the deviation of the results, if any, from the
UT A-scan thickness is observed in the “ideal” and “real” Gaussian
distribution superposition.
The first case, represented in Figure 3, corresponds to a pipeline
that we will refer to as Pipeline 1, which had outside and inside
diameters of six and four inches, where 3,852 readings were
obtained at 963 CMLs. As shown at the top of Figure 3, the red
Gaussian graph (real) is superimposed on that of the blue (ideal).
Therefore, according to the method, the system is affected by Figure 3. Gaussian Graph vs. Scatter Plot of Percentage of Mate-
generalized thinning and requires attention. What the Gaussian rial Loss for Pipeline 1.
graph shows us can be confirmed on the bottom in Figure 3,
where a large number of points are observed that represent per-
centages of thickness loss ranging from 12.5% to 35%.
Figure 4 corresponds to a pipeline that we will refer to as Pipeline
2, which had inside and outside diameters of three and four
inches, where 3,152 readings were obtained at 788 CMLs. As can
be seen at the top of Figure 4, the Gaussian graph in red (real)
does not superimpose on that of blue (ideal), therefore, accord-
ing to the method, the pipeline does not require attention due
to generalized thinning. Although there are four isolated points
outside the ideal (blue) readings (visible in the scatter plot) with
a maximum of 17% loss of thickness that, in this case, corre-
sponds to a 90° elbow, there are no conditions that require our
immediate attention.
The third case, which we will identify as Pipeline 3 is represented
in Figure 5. Pipeline 3 is two inches in diameter, and 872 readings
were obtained at 218 CMLs. As shown in Figure 5, the results are
similar to those of Pipeline 2. According to the method, Pipeline
3 does not require attention because there is no evidence of dam-
age mechanisms impacting the pipeline in a generalized way. In a
localized way, three isolated points are observed that are outside Figure 4. Gaussian Graph vs. Scatter Plot of Percentage of Mate-
rial Loss for Pipeline 2.

MARCH | APRIL 2021 Inspectioneering Journal 3


the ideal (blue) readings with a maximum of 17% loss of thickness,
corresponding to a straight section and a 45° elbow. However, as
confirmed by the scatter plot, there are no thickness losses that
require attention for additional testing.
Finally, Figure 6 shows the results of Pipeline 4, which has an
inside and outside diameter of two and three inches, where 1644
readings were obtained at 411 CMLs. As can be seen, the Gaussian
graph of red (real) color overlaps only a little of the blue (ideal)
color. Therefore, according to the method, the results tell us that
it behaves very similarly to an ideal pipe with thickness losses
between 0% and 12.5%. However, the scatter plot shows there were
13 isolated points outside of the ideal readings with a 20% loss of
thickness. In cases such as this one, it would be recommended to
expand the sample and apply the study again to establish more
reliably whether or not there is generalized thickness loss that
requires attention.

Conclusion
During the mechanical integrity evaluation of pipelines, a first
stage inspection plan generally includes the examination of the
Figure 5. Gaussian Graph vs. Scatter Plot of Percentage of Mate- pipeline using UT A-scan which can generate large volumes of
rial Loss for Pipeline 3. data requiring significant amounts of technical and financial
resources to analyze. But when properly evaluated using the sta-
tistical analysis method as outlined in this article, the process can
be much more efficient. The statistical case presented here, based
on the superposition of the graphs of the Gaussian distribution
of “ideal” and “real” data, enables the inspector to make better
decisions regarding the degree to which a pipe system is affected
by generalized or localized thinning. The graphic endorsement
that results from a population of UT A-scan data can support
decisions for justifying additional or subsequent and specific
non-destructive examination(s). It should be noted that pipelines
are often susceptible to additional damage mechanisms (not only
thinning), so a full mechanical integrity analysis or assessment
is dependent on identifying those additional damage mechanism
susceptibilities and detecting associated damage and damage
rates. This statistical analysis method presented herein, although
contributing to the efficiency of the integrity assessment, is not
intended to replace the judgment of a knowledgeable and expe-
rienced inspector. It only establishes parameters for better and
more efficient decision-making. ■

For more information on this subject or the author, please email


us at inquiries@inspectioneering.com.
Figure 6. Gaussian Graph vs. Scatter Plot of Percentage of Mate-
rial Loss for Pipeline 4. REFERENCES

1. API 570, “Piping Inspection,” Fourth Edition, February 2016, Appendix March 2,
2018, Errata April 1, 2018. The American Petroleum Institute

2. ASME B36.10, “Welded and Seamless Wrought Steel Pipe,” 2018. The American
Society of Mechanical Engineers.

3. API Recommended Practice 574, “Inspection Practices for Piping System


Components,” Fourth Edition, November 2016. The American Petroleum
Institute.

4 Inspectioneering Journal MARCH | APRIL 2021


CONTRIBUTING AUTHOR

Marcos Delgado
Marcos Miguel Delgado Lucero graduated from Universidad de Oriente in Venezuela, and is a
Mechanical Engineer, API 570 Piping Inspector, and ASNT Level II VT, PT, and UT Technician.
He has experience in static asset integrity, such as vessels, atmospheric storage tanks,
and pipelines. Mr. Delgado has knowledge in Visual Basic applied to Microsoft Excel and the
Python programming language.

MARCH | APRIL 2021 Inspectioneering Journal 5

You might also like