IBP 1304_15

ILI PINHOLE ANOMALY SIZING USING MFL SIGNAL MODELING

Guy Desjardins1, Joel Falk2, Sherif Hassanien 3

Copyright 2015, Brazilian Petroleum, Gas and Biofuels Institute - IBP

This Technical Paper was prepared for presentation at the Rio Pipeline Conference & Exposition 2015, held between September 22-
24, 2015, in Rio de Janeiro. This Technical Paper was selected for presentation by the Technical Committee of the event. The
material as it is presented, does not necessarily represent Brazilian Petroleum, Gas and Biofuels Institute’ opinion or that of its
Members or Representatives. Authors consent to the publication of this Technical Paper in the Rio Pipeline Conference &
Exposition 2015.

Abstract
One of the challenges facing In-line-inspection (ILI) tools is the detection and accurate sizing of small anomalies such
as pinhole corrosion. In order to improve understanding of the interaction of ILI signals with pinhole features,
Desjardins Integrity Ltd. (Desjardins) has obtained matched ILI and in-the-ditch data for a variety of metal-loss defects.
At this stage, the ILI data was restricted to magnetic flux leakage (MFL) data. Comparisons of the depth and length as
reported by the ILI tools and by in-the-ditch (ITD) measurements were made. Those with a discrepancy of more than
10% WT were analyzed in further detail. For each analyzed anomaly, the MFL signal response was calculated using
MFL modelling software. Two responses were calculated: first one with the dimensions as reported by ILI, and a
second one with the dimensions as reported ITD. The expected MFL responses from both the field-reported dimensions
and the ILI-reported dimensions were then compared. The results were analyzed with the goal of finding ways to better
understand the sources of sizing errors and identify areas and methods to improve the sizing performance of pinhole
corrosion. For example, in some cases the discrepancy between ILI and field measurements could be fully explained by
an incorrect ILI determination of the radial position of the anomaly. Preliminary results of this analysis are listed, along
with plans for further work in this area.

1. Introduction
This work is aimed at enhancing in-line inspection management of pinhole corrosion. It is generally true that
the sizing accuracy of MFL ILI tools is less for pinhole corrosion than for larger anomalies, and that in fact they tend to
undersize pinhole depths. For the purpose of this paper, a pinhole is defined as a metal-loss anomaly (MLA) with both
length and width less than 20 mm.
In the forward problem, if the morphology of an MLA is known, it is fairly straightforward to calculate the
MFL field that would be generated by it during an ILI (Edwards and Palmer 1986, Zhang 2009, Lorrain and Corson
1970, Jackson 1962). However, the reverse problem is much more difficult; given an MFL signal detected from an
MLA by an ILI tool, what is the morphology of the MLA? A significant part of the difficulty arises because of the non-
uniqueness of the solution of the reverse problem. It is possible for MLAs with different dimensions to generate very
similar MFL fields. This is the reverse problem that must be solved in processing ILI data. Since similar MFL fields can
correspond to different anomaly morphologies, additional information such as wall thickness and radial position can be
crucial in selecting the correct solution. If some of this information is incorrect, it can result in significant errors in
interpreting the data.

2. Approach
Desjardins obtained ILI and ITD matched data for 162 pinholes from several different pipelines, and
performed a basic statistical analysis on them. Of these, 44 were identified as outliers (anomalies with a discrepancy
greater than 10%), and an attempt was then made to determine a cause for the discrepancy. The anomaly dimensions as
reported by the ILI were used to create a model of the anomaly in software. The MFL field that would be generated by
this anomaly during an ILI was then modelled. The same was done with the dimensions as reported ITD – the anomaly
was modelled using the ITD dimensions, and then the MFL field that would be generated from it was also modelled.
______________________________
1
P.Geoph. – President, Desjardins Integrity Ltd., Calgary, Canada
2
B.Sc. Physics – Senior Pipeline Integrity Specialist, Desjardins Integrity Ltd., Calgary, Canada
3
PhD, PEng. – Manager, Mainlines Integrity Analysis, Integrity Planning, Enbridge Pipeline Inc., Edmonton, Canada
 Rio Pipeline Conference & Exposition 2015
The MFL fields from the ILI and ITD models were then compared to determine if an explanation could be found for the
discrepancy.
The modelled MFL responses from the ILI dimensions and the ITD dimensions were compared to identify
similarities and differences. If the MFL responses are similar, then the error in the ILI reported dimensions are due to
the non-uniqueness of the reverse problem solution. Differences in the signal may assist in identifying when errors in
the reported depth may exist. In addition to the MFL signal, data such as wall thickness and radial position of the MLAs
as reported by ILI and ITD were compared to determine the effect these would have on the sizing accuracy.

3. Findings
Analysis of the 162 pinholes showed that:
- Sizing of pinholes by ILI tends to undersize the depth, as shown in Figure 1.
- The Probability of Detection (PoD) decreases with anomaly size. For one of the pipelines, PoD was 54.5%
for pinholes and 82.7% for larger anomalies.
- The Radial Position error rate (internal/external errors) is higher for pinholes than for other metal loss.
Radial position of 8.6% of the pinholes were in error whereas the error rate was only 0.7% for non-
pinholes
- For 24.6% of pinholes, the discrepancy in wall thickness between ILI and ITD measurements were greater
than 0.7 mm. In all cases, the ILI reported a heavier wall thickness than the ITD measurement.

Figure 1 Unity Plot of MFL ILI Pinhole Anomalies

The 44 outlier pinholes were investigated in further detail. In some cases, the ILI and ITD dimensions differed
significantly, but the MFL fields from the ILI and ITD models were very similar. For 64% of the outliers, Desjardins
was able to determine a possible cause for the discrepancy between the ILI and ITD depth measurements:
- Sensor liftoff or sharp anomaly profile (36.4%)
- Wall thickness miscall (13.6%)
- Radial position (internal vs external) miscall (9%)

An example for each of the most common causes is provided below.

3.1. Example 1: Error In Determining Radial Position (Anomaly C71)

The depth of this anomaly was sized as 26% WT by the ILI tool, while the ITD measurement sized it at 41%
WT, a discrepancy of 15% (Table 1). However, the measurements for length and width agreed. It was noted that the ILI
reported the anomaly as being on the internal surface, whereas upon excavation it was actually external. Two anomalies
were modelled in software, one based on the ILI measurements and the other based on the ITD measurements, as shown
in Figure 2. Next, the resulting MFL field that would result from these anomalies was calculated and is shown in Figure

2
 Rio Pipeline Conference & Exposition 2015
3. It is apparent that the field resulting from a 26% internal anomaly is very similar to that from a 41% external one. To
determine the correct sizing of the anomaly, it is necessary to know the correct radial position of the anomaly.

Table 1. Anomaly C71

Model Length (mm) Width (mm) Depth (%WT) WT (mm) Wall Surface
ILI 18 14 26 3.9 Int
ITD 18 14 41 3.9 Ext

ILI 26% ITD 41%

(Interior) (Exterior)

Figure 2. Anomaly C71 Dimensions. Left: ILI, Right: ITD

ITD 41% Ext

ILI 26% Int

Figure 3 Simulated MFL fields. Top: ITD 41%WT External, Bottom: ILI 26%WT Internal

For comparison, the difference between the MFL fields from a 41% internal and a 41% external anomaly is
more pronounced, as shown in Figure 4.

3
 Rio Pipeline Conference & Exposition 2015

41% Ext

41% Int

Figure 4. Simulated MFL Fields. Top: 41%WT External, Bottom: 41%WT Internal

Although the details of the algorithms used by ILI vendors for sizing anomalies are not made public, it appears
that in this case it would be difficult to distinguish between 26% WT internal and 41% external anomalies. If the ILI
analysis software assumes an incorrect radial position, the result could be a significant sizing error.

3.2. Example 2: Possible Sensor Liftoff (Anomaly B137)

Although this feature does not meet the criteria of being a pinhole outlier, it is a good example of the possible
scenario of sensor liftoff. The ILI undersized Anomaly B137 by 6% in depth, as well as undersizing both length and
width. The anomalies as measured by ILI and ITD are listed in Table 2 and modelled in Figure 5. The MFL fields
resulting from these anomalies are plotted in Figure 6.

Table 2. Anomaly B137

Model Length (mm) Width (mm) Depth (%WT) WT (mm) Wall Surface
ILI 23 18 17 5.16 Ext
ITD 27 30 23 4.91 Ext

ILI 17% ITD 23%

Figure 5 Anomaly B137 Dimensions. Left: ILI, Right: ITD

4
 Rio Pipeline Conference & Exposition 2015

ITD 23%

ILI 17%
assuming
no sensor
liftoff

Figure 6 Simulated MFL fields. Top: ITD 23%WT External, Bottom: ILI 17%WT External

It was noted that this anomaly was located just 18 cm downstream from a girth weld. The possibility was
considered that the MFL sensor head might not have fully returned to the pipe wall surface after passing the weld, and
various liftoff distances were modelled. It was found that a 23% WT anomaly with sensor liftoff of just 1.8mm resulted
in an MFL signal very similar to that from a 17% anomaly assuming no sensor liftoff (Figure 7). In other words, in this
case a liftoff of less than 2mm can appear to reduce the sizing depth by 6%.

23% with
1.8mm liftoff

Figure 7 Simulated MFL Field of 23% Anomaly with 1.8 mm Sensor Liftoff

3.3. Example 3: Incorrect Wall Thickness (Anomaly C115)

As shown in Table 3 and Figure 8, anomaly C115 was undersized by 11.4% although the length and width
were sized correctly. However, the ILI determined the wall thickness to be 3.96 mm, while the actual thickness was 2.8
mm. Desjardins investigated what effect this might have on sizing, and Figure 9 shows some modelled MFL fields.
It was found that the MFL signal from a 23.6% WT anomaly in 3.96 mm pipe was almost exactly the same as
that from a 33% WT anomaly in 2.8mm pipe. In other words, if the correct wall thickness had been used, the ILI results
would have been consistent with a 33% defect instead of a 23.6% one. By comparison, if the anomaly actually were
23.6% WT in 2.8 mm pipe, the MFL field would be expected to look significantly different, as shown in Figure 10.

Table 3. Anomaly C115

Model Length (mm) Width (mm) Depth (%WT) WT (mm) Wall Surface
ILI 18 14 23.6 3.96 Int
ITD 18 14 35 2.8 Int

5
 Rio Pipeline Conference & Exposition 2015

ILI 23.6% ITD 35%

Figure 8 Anomaly C115 Dimensions. Left: ILI, Right: ITD

33% WT Anomaly in
2.8 mm wt pipe

23.6% WT Anomaly
in 3.96 mm wt pipe

Figure 9 Anomaly C115 Simulated MFL Fields

23.6% WT Anomaly
in 2.8 mm wt pipe

Figure 10 Anomaly C115

3.4. Example 4: Non-Ellipsoidal Shape

Anomaly D3 was determined by ILI to have a depth of 23.1% WT, almost 12% less than the ITD measurement
of 35% WT. The ILI also significantly undersized the width, although the length was correct. A comparison of the
anomaly morphology as reported by ILI and ITD appears in Table 4 and Figure 11, and the simulated MFL signals from
these models appears in Figure 12. There is a very large difference between these signals, and Desjardins proposes two
possible explanations. One is that there has been sensor liftoff, as in the case of example 2, although this feature is not
located close to an upstream weld. On the other hand, sensor liftoff could also be caused by debris in the pipeline
(Burger et al 1981).
6
 Rio Pipeline Conference & Exposition 2015
Table 4. Anomaly D3

Model Length (mm) Width (mm) Depth (%WT) WT (mm) Wall Surface
ILI 12 4 23.1 6.35 Ext
ITD 12 11 35 6.35 n/a

ILI 23.1% ITD 35%

Figure 11 Anomaly D3 left: ILI, right: ITD

Another possibility is that an incorrect assumption was made about the shape of the anomaly. Metal-loss
anomalies are typically assumed to have a semi-ellipsoidal cross-sectional shape, as defined by:

2 2
+ + =1 (1)

where , , and are the anomaly length, width and depth respectively.

In many cases this is a valid model, as shown in Figure 13 for a different anomaly (A285). A cross-section
micro-photograph was available for this feature. Using the ITD dimensions, a semi-ellipsoidal model was produced and
scaled to the same size. It is apparent that the shape of the sides of the anomaly and the model match reasonably well.

ITD 35%

ILI 23.1%

Figure 12 Simulated MFL fields. Top: ITD, Bottom: ILI

7
 Rio Pipeline Conference & Exposition 2015

Figure 13 Cross-section view of anomaly A285 (grey), with semi-ellipsoidal model (color)

However, this is not always the case. Anomalies can have the same length, width and depth, but the shape of
the cross sections can be different, as shown in Figure 14. It is evident that the will be significant differences in the
volume of metal loss, with a resulting difference in the MFL signal (Bluck et al 2010). To model these, Desjardins
defined a “sharpness” parameter, n, such the shape of the anomaly is defined by equation 2 below:

2 2 (2)
+ + =1

Note that in the case where n=2, the equation reduces to an ellipsoid. Examples of anomalies with varying
values of appear in Figure 14.

Figure 14 Variations in cross-section shapes for constant dimensions

Figure 15 Variations in sharpness of an anomaly

Desjardins then investigated the effect that various values of n would have on the MFL field, and it was found
that an anomaly with a “sharpness” value of 0.55 was the best fit (Figure 16). The MFL field expected to be generated
by this anomaly is shown in Figure 17, and this matches very closely to what the ILI tool is expected to have recorded.
In other words, an anomaly of the dimensions recorded ITD but with a sharpness value of 0.55, would produce a very
similar MFL signal as what was detected by the ILI tool.

8
 Rio Pipeline Conference & Exposition 2015

Figure 16 Anomaly 35% WT n=0.55

Figure 17 Simulated MFL field, 35% WT, n=0.55

4. Conclusions
It was found that the accuracy of MFL ILI tools when recording pinholes was significantly reduced than when
recording larger anomalies, as detailed in Section 3. PoD, sizing accuracy, and accuracy of radial position are all
decreased.
For 28 of the 44 outlier pinholes, Desjardins was able to determine a possible explanation for the discrepancy.
The most likely causes were sensor lift-off, sharp anomaly profile, incorrect radial position determination, and incorrect
wall thickness determination. If increased efforts were made to address those issues, sizing accuracy of pinholes could
be increased. Desjardins plans to continue research on this area.

5. References

Pipeline Operator’s Forum, "Specifications and Requirements for Intelligent Pig Inspection of Pipelines," 2009.

C. Edwards and S. B. Palmer, "The magnetic leakage field of surface-breaking cracks," J. Phys. D: Appl. Phys. 19, pp.
657-673, 1986

H. O. Zhang, Y. Z. Yang, G. L. Wang and H. Zou, "Numerical Simulations of Electromagnetic Flux Leakage in
Application of Internal Defects Prediction of Metal Parts.," Piers Online Vol 5, No 2., pp. 105-108, 2009.

P. Lorrain and D. Corson, Electromagnetic Fields and Waves, San Francisco: W. H. Freeman and Company, 1970.
J. D. Jackson, Classical Electrodynamics, John Wiley & Sons Inc, 1962.

M. Bluck, E. Quick, A. Bain, J. Sutherland and J. Pearce, "Validation of Latest generation MFL ILI Technology Leads
to Improved Detection and Sizing specification for Pinholes, Pitting, Axial Groving, and Axial Slotting," in Pipeline
Pigging and Integrity Management, Houston, TX, 2010.

E. D. Burger, T. K. Perkins and H. Striegler, "Studies of Wax Deposition in the Trans Alaska Pipeline," JOURNAL OF
PETROLEUM TECHNOLOGY, June 1981.