MPWT19-14353

Validating ILI run with and without Data Loss

Ahmad Al Saif
Saudi Aramco
King Fahd Road, Dammam 34223
Dammam, Eastern Province, 34223
Saudi Arabia

ABSTRACT

In-Line Inspection (ILI) technology is considered one of the safest and most efficient and reliable
inspection method to inspect hydrocarbon pipelines. The retrieved data are usually validated and verified
upon successful completion of the inspection. This paper is intended to introduce a new approach to
validate the ILI run based on a statistical analysis comparing the new ILI run with a previous ILI run of
the same pipeline by leveraging a root mean square (RMS) model to quantify the similarity between the
datasets. API-1163 and Canadian Energy Pipeline Association (CEPA) offer consistent criteria as a
validation methodology for a new ILI run. Also, this paper will demonstrate a new scoring criterion for
accepting Magnetic Flux Leakage (MFL) runs with partial data loss as number of MFL runs experience
unexpected data loss, which might affect the minimum reporting threshold of the tool. The approach will
help pipeline operators to identify the criticality of the missed data via a detailed comparison with the
previous MFL run for the same pipeline and detailed analysis of the behavior of the tool during the run.
The scoring criteria is aligned with the Pipeline Operators Forum (POF) requirements for data loss.
Multiple case studies extracted from actual data will be presented throughout the paper.

Keywords: In-Line Inspection, Pipeline Inspection, Pipeline Data Analysis

INTRODUCTION

Using ILI technology to inspect hydrocarbon pipelines in order to revalidate the integrity of pipelines
has been a usual practice for many years. Although there are other inspection methods such as direct
assessment, or hydrostatic testing, ILIs are the most commonly used technology worldwide due to
number of factors, including but not limited to: practicality, cost effectiveness, and accuracy. Pipelines
need to be periodically inspected to detect and evaluate any potential threat that might affect their
integrity. Integrity threats can be divided into three main categories: metal loss due to corrosion, cracks,
and geometry damages. Metal loss corrosion is the general driver of pipelines’ deterioration and leads to
failures. Cracks are usually associated with high pressure fluids and their associated failures are usually
more severe than metal loss corrosion failures. Geometry damages (i.e.: dents, wrinkles, etc.) in the
pipelines are usually caused by excessive external stress on the pipelines. Each one of these threats
require specific ILI technology to be detected, sized, and evaluated. In this paper, only metal loss
corrosion will be discussed. The magnetic flux leakage (MFL) tool is widely used to inspect metal loss
corrosion in pipelines. Upon completion of the MFL run, it has to be validated by the pipeline operator.
API-1163 standard offers three levels of validation for ILI run [1]. In this paper, a new methodology will be
introduced to validate the MFL run based on level 1. However, MFL runs might experience mechanical
damages or electronic malfunctions due to different reasons. Reasons include the upsets in the operating
parameters of the pipeline, partial closure of mainline valves, or unexpected sudden sensor loss.
Accepting such runs require additional engineering analysis that will be addressed in the second section.

Level 1 Validation

API [1] and Canadian Energy Pipeline Association (CEPA) [2] describe different levels of validating
and verifying the ILI run which can be applied to MFL runs.
For API, there are three levels of validations sorted by the risk severity of the pipeline. Level-1 applies
to pipelines with corrosion of low risk level. Therefore, no field verifications are required to validate the
MFL run. However, the pipeline operator will not be able to disqualify the MFL run based on level 1 only.
Thus, level 2 or level 3 is required for the disqualification [1].
For CEPA, the MFL run can also be validated for pipelines that do not have any corrosion features
meet the criteria of immediate repair.
The validation criteria for API and CEPA in the absence need of field verifications are alike. There
are number of approaches that were mentioned to validate the MFL run. One approach is comparing the
ILI run with a previous one for the same pipeline or another similar pipeline or compare this pipeline with
a large-scale data of previous similar MFL runs. In the upcoming section, validating MFL run by
comparing it with previous ILI run(s) will be described in detail.

Comparison with the Previous ILI run

In this section, different indicators will be introduced to validate the MFL run. Although API described
this as one method to validate the MFL run, no specific procedure was introduced. On the other hand,
CEPA described in detail the comparison after completing the feature to feature matching of both runs.
However, this paper will focus on performing the comparison without the need of matching features of
both runs and achieving high level of confidence that the MFL run performance was within some
acceptance criteria.
The first indicator is the depth histograms of both MFL runs. Since depth of corrosion is the most
important factor to be accounted for leak failures, the consistency of depth is quite essential. Although
histograms are not expected to be identical due to the tool’s tolerance (usually ± 10%) and the corrosion
growth rate, both histogram shall follow similar trends in terms of the number of corrosion features per
unit depth. As shown in (Figure 1-2), converting the scale to logarithmic provides more details and
increase the resolution of the comparison, especially for high number of corrosion features. It can be
stated that for this specific pipeline, there is a consistency in the reported depth between the new and
the previous ILI runs with no observed growth. Therefore, in terms of depth, the new run passes the
validation test of API-1163 level 1 and CEPA.
 The second indicator is the length histograms of both MFL runs (shown in Figure 3). Corrosion feature
length is quite critical for pipelines that are susceptible to rupture failures. Similar with depth distribution,
length distribution is in logarithmic scale enables for more meaningful comparison (shown in Figure 4).
 If the distribution of length differences for both MFL runs is equal or greater than the distribution of
depth differences, then the pipeline operator and the ILI vendor shall investigate potential causes as the
normal behavior of pipelines’ corrosion is to have a depth corrosion growth faster than length corrosion
growth. It can be stated that for this pipeline, a consistency in the reported lengths between the new and
the previous ILIs with an observed reduced frequency of the large length values between the two ILI runs.
To clarify, the previous ILI run reported long corrosion features more than the recent run. However, this
might be explained by the improvement in the sensor capability for the new run. Therefore, in terms of
length, the new run passes the validation test of API-1163 level 1 and CEPA.
The study of histograms existed for a long time and different approaches were developed to quantify
the similarity between two histograms; one of which is the RMS approach which can be applied to
mathematically confirm the similarity supported by visual similarity.
According to Sergey Bityukov[4] and his colleagues, RMS method can be used to quantitatively
determine the similarity between two histograms. The proposed method to calculate RMS will be applied
to ILI runs histograms with slight changes in the formula to match their configurations.
Consider two ILI runs datasets converted to two histograms. Let k denotes the histogram number, i
the bin value of run 1, j the bin value of run 2, and n frequency value on bins:

𝑙𝑙𝑙𝑙𝑙𝑙(𝑛𝑛�𝑖𝑖1 )− log(𝑛𝑛�𝑖𝑖2 )
𝑆𝑆̂ = (1)
2 2
� 𝑖𝑖1 ) +𝜎𝜎log(𝑛𝑛
�𝜎𝜎log(𝑛𝑛 � 𝑖𝑖2 )

2
Where: 𝜎𝜎log (𝑛𝑛�𝑖𝑖1 ) is the variance of one pin value or the standard deviation squared.

∑𝑀𝑀 ̂ ̅ 2
𝑖𝑖=1(𝑆𝑆𝑖𝑖 −𝑆𝑆 )
𝑅𝑅𝑅𝑅𝑅𝑅 = (2)
𝑀𝑀

Where: 𝑆𝑆̅ is the mean of all 𝑆𝑆�𝚤𝚤 , and M is the number of bins.
The abovementioned methodology can be easily interpreted as follows:
• If RMS = 0, both histograms are identical. However, ILI data sets are not expected to be
identical. Therefore, RMS value of zero is not expected.
• If RMS ~ 1, both histograms are obtained from the same parent distribution, which means in
ILI data sets that both ILI runs have similar distributions and are consistent.
• If RMS >> 1, histograms are obtained from different parent distributions, which means: both
ILI runs datasets are not similar and not consistent.
The above formula was slightly changed from the original RMS formula mentioned in Bityukov’s paper
to match the need of ILIs data comparison. Validation process of this formula was performed for actual
datasets of Saudi Aramco Pipelines.
Since the above criteria is not specified for ILI data sets, a new proposed criterion is proposed to
validate the new ILI run:
• If RMS < 0.5, the two data sets are similar with at least 80% confidence.
• If RMS < 0.8, the two data sets are similar with at least 40% confidence

The third indicator is the combination of depth and length histograms for both runs (shown in Figure
5). Although double-category chart is not straightforward to comprehend, the observed findings are
significant. This indicator enables the pipeline operator to determine the criticality of the pipeline based
on the long and deep corrosion features. Hence, the pipeline is more susceptible to both leak and rupture
failure modes. Based on the previous indicators for this pipeline, an important question arises, that is:
where do the long corrosion features fall in the depth distribution? Indicator 3 will help the pipeline
operator to answer that question. For this pipeline, corrosion features with 241 mm length and more fall
in [21-40] and [41-60] depth categories. It is worth mentioning that indicator 3 has two contributions in
the validation process: the first is the fact that the reported deep corrosion features were longer in the
new ILI run compared with the previous run, and the second is the fact that both ILI runs are consistent
for the combined depth and length distributions.
 The classification (internal corrosion, external corrosion, non-corrosion) of the reported features can
be changed from one ILI run to the other. Therefore, classification check is another important indicator to
validate the new ILI run.

(Figure 6) shows a pipeline that is different to the presented one in the previous indicators. In this case,
the internal corrosion features (denoted by YES) in the 1st run have similar distribution to the non-
corrosion features (denoted by N/A) in the 1st run. In such case, the ILI vendor need to be approached to
verify the classification of the reported features or conduct field verification digs to validate the ILI run
results.
ILI run with partial data loss

Next section of this paper will focus on MFL runs that experienced partial data loss. Data loss can be
caused by number of reasons including, but not limited to: sensor malfunction, sensor lift off the pipe wall
due to excessive debris, or sensor damage due to mechanical pipeline damage (i.e.: valve, guide bar,
dent). API-1163 and CEPA have addressed this issue and drawn the general guidance to validate the
run but both standards did not specify solid criteria for rejecting the MFL run. However, the ILI vendor
has to provide a summary of the number of malfunctioned sensors and the new detection and sizing
criteria compared with the original one. On the other hand, Pipeline Operators Forum (POF) added more
rigorous conditions to validate the MFL run in the: “Specifications and requirements for in-line inspection
of pipelines” (2016) [3]. These conditions are:
• Continuous loss of data less or equal to 0.5 % of pipeline length.
• Discontinuous loss of data less or equal to 3% of pipeline length.
• Continuous loss of data from less than 4 adjacent sensors or 25 mm circumference (whichever is
smallest).
To put these conditions in perspective, let’s set an example of 56” pipeline with a total length of 100 km:
• The maximum continuous data loss shall not exceed 500 meters, but it was not specified that the
distance is the travelled distance or the trajectory distance with the circumferential coverage.
• The maximum discontinuous data loss shall not exceed 3 km. Again, it was not specified that the
distance is the travelled or the trajectory.
These two conditions were meant to cover axial data loss. Therefore, the third condition was set to
define the limit of the circumferential data loss.

Guidance to Analysis the Data Loss

When an MFL run experiences data loss, the ILI vendor shall provide the travelled path for each
affected sensor to the pipeline operator. This raw data will be analysed independently and combined with
the previous pipeline corrosion data (i.e.: previous ILI run, previous field findings). The purpose of this
analysis is to confirm the total data loss to be examined throughout the pipeline.
Upon receiving the data loss log file, the following formula can be applied to calculate the coverage of
each sensor:

𝜋𝜋∗(𝐷𝐷−2∗𝑤𝑤𝑤𝑤 )
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 [𝑚𝑚𝑚𝑚] = (3)
𝑛𝑛𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑟𝑟𝑟𝑟

Where: 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 [𝑚𝑚𝑚𝑚] ≡ the total circumferential coverage per sensor

𝐷𝐷 ≡ Pipeline diameter [mm]
𝑤𝑤𝑤𝑤 ≡ Nominal wall thickness [mm]
𝑛𝑛𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 ≡ Number of sensors per ring
Assuming that each data point reports the center of the sensor, hence, the sensor’s coverage will be:
𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 [𝑚𝑚𝑚𝑚] = 𝑥𝑥 [𝑚𝑚𝑚𝑚] ± 0.5 ∗ 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 [𝑚𝑚𝑚𝑚]
Where: 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 ≡ the range covered by one sensor from a given data point.
𝑥𝑥 ≡ the reported distance of the defected sensor.
 Now, one can specify the range at which the data was lost throughout the pipeline. An additional step
maybe required to combine all adjacent defected sensors. With this data, the total axial and
circumferential data loss can be calculated and represented in a similar format shown in (Figure 7).

The observation for the presented case is: starting from km-2.6, sensors start to fail reporting some
data and the maximum circumferential loss was at km-7.3. The presented figure would enable for more
meaningful interpretation if it was aligned with the previous ILI corrosion data. (Figure 8) aligned the two
data sets and shows previous corrosion features with a metal loss depth of 40-60% were reported and
locations correspond to the maximum data loss. For similar case, the reliability of the new inspection is
severely jeopardized due to the existence of deep corrosion features reported in the previous run.
Therefore, regardless of the overall quantity of data loss, pipeline operators shall take this section for
more detailed analysis to confirm the reliability of this MFL run. For instance, a more detailed analysis
would be to replace the distribution of the data loss in (Figure 8) to the actual path of the defected sensors.
(Figure 9) is a zoomed snapshot of the combined path of the defected sensors and the previous ILI run
results.
 Clearly, there are corrosion features that fall at the path of the defected sensors. A mathematical
algorithm could be developed to find the distribution of the previous corrosion features that falls at the
path of the defected sensors of the recent run. The last representation of the lost data is the distribution
of the data loss along the circumference of the pipeline. (Figure 10) illustrates that the total circumferential
data loss is less than 5% and the least amount of data loss is at the bottom of the pipeline; which indicates
that the MFL tool had a fair amount of rotation throughout the inspection run. Therefore, the rotation of
the tool can reduce the total missed corrosion features as the working sensors will have higher probability
to catch general corrosion in the pipeline. It is worth mentioning that this analysis shall be aligned with
the POF requirement that was previously mentioned.

CONCLUSIONS

API-1163 and CEPA standards offer pipeline operators an option to validate the ILI run for cases that no
field excavations are required as immediate repairs based of the probability of failure and consequence
criteria. The validation process can be achieved by different approaches, one of which is comparing the
new ILI run with the previous one. In the first section of this paper, specific indicators were presented to
validate the new ILI run when compared with previous ILI run of the same pipeline. Indicators included
length, depth, and circumferential corrosion distributions of both ILI runs. These indicators will assist
pipeline operators in the validation process of the ILI run. Also, if a noticeable discrepancy was observed
between the two ILI runs, the operator shall approach the ILI vendor to re-evaluate the data or conduct
field verifications to validate the run. Pipeline operators can add more indicators like the circumferential
distribution of the corrosion along the pipeline.

The second section of this paper introduced a new methodology to qualify an MFL run with partial data
loss. The purpose of the methodology is to identify the criticality of the missed data based on a previous
MFL run and to quantify the distribution of the missed data in both axial and circumferential range of the
pipeline.

The proposed approaches in validating the ILI run with and without data loss introduces systematic data
analyses to confirm that the ILI run can be used as a revalidation for the pipeline
 REFERENCES

1. API-1163 STANDARD (2016), “In-line Inspection Systems Qualification”.

2. Canadian Energy Pipeline Association (January 2016, 1st Edition), “Metal Loss Inline
Inspection Tool Validation Guidance Document”

3. Pipeline Operators Forum (POF) (2016), “Specifications and requirements for in-line inspection of
pipelines”

4. S. Bityukov, N Krasnikov, A Nikitenko, V Smirnova, “A method for statistical comparison of

histograms”, ResearchGate, February 2013,
https://www.researchgate.net/publication/235438674_A_method_for_statistical_comparison_of_histogr
ams (February 2013)