Calculation of a laser-scan-like

3D defect profile
from conventional MFL data

by Johannes Palmer, Andrey Danilov

ROSEN Technology & Research Center, Germany

15thPipeline Technology
Conference

Estrel Congress Center, Berlin, Germany

30 March - 2 April 2020

Organized by
EITEP Institute Marie-Jahn-Str. 20
D-30177 Hannover
Germany
 Pipeline Technology Conference 2020, Berlin

Abstract
Calculation of a laser-scan-like 3D defect profile from conventional MFL data –
For more-accurate defect assessment and prediction of safe operating pressure
Reliable pipeline safe pressure prediction is increasingly required. Complex shaped
volumetric metal loss is a challenge in this regard. Methodological In-line Inspection (ILI)
restrictions and principal reporting simplification reduce the precision of ILI results and
output tables.
Today, phenomenological Magnetic Flux Leakage (MFL) data parameterization is
rearranging complex metal loss structures and often results in reduced accuracy.
Nevertheless, the established maximum depth boxes do not necessarily prevent cases
where MFL data interpretation is insufficiently conservative.
The presented ground-breaking MFL data evaluation technique directly calculates the
accurate 3D metal loss geometry as an alternative to the currently practiced boxing. The
resolution of the innovative result opens a new dimension for MFL. Auxiliary components
of the typical MFL indirect interpretation method lose importance with this new system.
Experience-based practices using human expertise, artificial intelligence or elaborate
sizing models play a role still, but the conclusive accuracy of the calculated metal loss
profile becomes significantly independent from the variability of these experience-based
practices.
This technology enables new services for difficult-to-access areas, areas with complex
corrosion, or instances in which a laser map cannot be obtained, resulting in a more
accurate, reliable and detailed corrosion growth assessment. This paper will detail
concrete case examples from blind tests with high-resolution laser maps and will compare
the new approach with results from the traditional MFL evaluation. Performance and
reliability analyses help rationalize this innovative technique and demonstrate the
contribution to more reliable management of metal loss threats.
As part of describing this new evaluation technique, theoretical and practical advantages
will be discussed of combining axial and transverse MFL measurements for detection and
sizing of metal loss anomalies.

1 Introduction
In-line Inspection (ILI) result requirements continuously increase. In the case of complex
metal loss, more accurate safe pressure predictions need not only more accurate
maximum depth values. More detailed, more accurate and more relevant metal loss
profiles are required for more precise and economically adequate predictions
[Kariyawasam et al. 2019]. Established Magnetic Flux Leakage (MFL) methodological
characteristics hardly follow.
The relevance of reliable and accurate safe pressure prediction increases, but MFL
parameterization and result simplification introduces conservatism, which increases the
number of digs too early or superfluous. Nonetheless, even this conservatism still does
not completely prevent principally erroneous data interpretation caused by the ambiguity
associated with the MFL method. Specifically complex corrosion is responsible for difficult
________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 2 of 11
 Pipeline Technology Conference 2020, Berlin

data interpretation and potential threats. This situation motivates investigating the principal
possibility to replace the established MFL parameterization and simplification with
calculating the 3D metal loss geometry with the highest resolution.

2 MFL Attributes: Indirectness and Ambiguity

The MFL signals do not directly show the metal loss. A direct analytical calculation is also
not possible (see below). The example (Fig.1a) shows, how different MFL field
disturbances in the axial and circumferential direction are generated from the same seven
ft. long complex metal loss. The established signal data interpretation of the observed
signals will prefer most probable solutions. Hence, apparently dominating data structures
will also dominate the metal loss interpretation.
Figure 1a
ILI measured MFL signals; left: axial magnetizer tool, right: circumferential magnetizer tool

When comparing the ILI data with the corresponding laser scan (Fig.1b), it appears
typically more probable, that the axial continuity of the axial structures tends to volume
loss overestimation in the circumferential and underestimation in the axial field. Obviously,
both data sets in combination contain an enormous amount of detail and information. Two
elementary questions arise: How precise are the ILI measurements? How sound is
methodological MFL understanding?

Figure 1b
External corrosion of slotting/grooving dimension under folded, disbonded coating (46% maximum depth)

The mathematical forward calculation from the laser scan (Fig.1b) is a complex task, but
basically requires just one principal calculation step (Fig.1c). The excellent agreement
between this calculation (Fig.1c) and the ILI measurements (Fig. 1a) confirms both: the
understanding of MFL principles as well as the high quality of the ILI measurement.

________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 3 of 11
 Pipeline Technology Conference 2020, Berlin

Figure 1c
Synthetically calculated magnetic field left: axially magnetized, right: circumferentially magnetized

The “direct” task from 3D scan to magnetic field is straightforward attainable, the “inverse”,
from measurement to metal loss, is not. In academic jargon it is therefore referred to as
an “incorrectly formulated task” [Militzer, Weber, 1984: 1.5.3].
Two different source geometries may cause the same magnetic field disturbance.
Therefore, the measured field is ambiguous. In all field directions [Militzer, Weber, 1984:
5.2.1.6.1] and at different measurement heights [Militzer, Weber, 1984:1.5.2.2], no
additional information is observable as long as all these measurements are outside the
source itself. This is the case in ILI, since the sensors do not penetrate the wall. The
ambiguity of the MFL method can be approached by knowledge databases, which store
and link the source geometries with their practical occurrence probability. Since most of
the pipelines and their metal loss origins and mechanisms are relatively well known, the
chance of pitfalls is enormously reduced [Stuart, Clouston 2013]. Nevertheless, the most
probable metal loss shape is not necessarily the correct one. Thus, the most conservative
interpretation may generate false calls, eventually many. Significant efforts try reducing
their number, but will never overcome inherent methodological limitations. Already the
simple existence of a benefit from optimized interpretation as formulated in standards like
API 1163 (e.g. 8.2.6b in [API 2013]) shows already this intrinsic vagueness, reflected in
interpretational simplified boxes. Boxes are tending to over-conservatism, have often
poorly defined outer limits and hence, cause low repeatability as well as ambiguous, and
sometimes hidden, corrosion growth rates.

2.1 Approach to Overcome MFL Indirectness and Ambiguity

The industry result quality requirements continuously grow, hand in hand with increasing
MFL measurement accuracy. Therefore, outliers become less acceptable. Contributing
tool components and sizing data bases were continuously improved. Meanwhile, it is not
the scattering of involved components in the MFL process chain, but the fundamental MFL
sizing definitions and the MFL methodological nature itself that dominate the restrictions.
The agreement between the synthetic forward calculation (Fig.1c) and the real ILI
measurement data (Fig.1a) show a level of capability, which suggests to replace the
described established “interpretation” with “calculation”. Calculation is simply aiming at
high repeatability, comparability and accuracy.

________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 4 of 11
 Pipeline Technology Conference 2020, Berlin

Figure 2 shows the principal difference of the new MFL calculation approach. The complex
interpretation operator Ω (Fig.2a) is experience-based and probability-driven. Ω is based
on human experience, knowledge databases from verifications and synthetic calculations
and machine learning. It uses human expertise and artificial intelligence to find the best
sizing PIPE0.

The calculation F (Fig.2b) simply computes the difference ΔMFL of ILI measurement and
refined assumption PIPEn. As long as the difference is not satisfactory, the iterative
calculation continues. This process is called “Deep Field Analysis” (DFA). A typical
complex assumption (Fig. 2a) may serve as start model. The success speed will depend
on its quality, but the success itself does not.

Figure 2a Established MFL data interpretation Figure 2b MFL result calculation (Deep Field
applying knowledge based operator Ω Analysis) via iteration loop to minimize
ΔMFL of ILI and assumption

One of the earliest practical test examples is outlined in Figure 3a. Left hand side shows
the DFA result of conventional axial MFL (MFL-A) and on the right the combination of
MFL-A and circumferentially magnetized MFL (MFL-C). The two results differ significantly
by metal loss volume and maximum depth.

Figure 3a Feasibility test of new MFL evaluation technique DFA on complex 2cm wide axial corrosion;
test patch is 0.5m*0.2m with original 50% max. depth; comparison between stand-alone axial MFL (left)
and combined axial plus circumferential MFL (right)

________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 5 of 11
 Pipeline Technology Conference 2020, Berlin

The reason for that difference is the ambiguity described above. Even the cross-checked
three components show no difference for the two different solutions, whilst the
(circumferentially magnetizing) MFL-C allows for correct differentiation, as shown in
Figure 3b.
The MFL-A alone will find one of the mathematically correct metal loss solutions. But only
the combination of MFL-A plus MFL-C will be able to find the only real correct one. Both
solutions are displayed in Figure 3c in comparison to the laser scan.

Figure 3b Output check of all three MFL-A components and MFL-C circumferential main component

Figure 3c Comparison of DFA models calculated from MFL-A alone (left) and MFL-A plus MFL-C (right)
vs. original verification scan (center)

The concept is proven to calculate the high resolution 3D metal loss geometry from MFL-
A & MFL-C data. A first prototype of a Virtual−Dig Up (VDU) User Interface was created,
as visualized in Fig. 4.

________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 6 of 11
 Pipeline Technology Conference 2020, Berlin

Figure 4a Virtual−Dig Up – Graphical User Interface

External data, e.g. laser scans, can also be displayed in this frame as shown in Figure 4b.
A long-term perspective could be the direct von Mises stress calculation from the 3D
models, without the need of 2D profile simplification. Figure 5 provides an example in this
regard.
Figure 4b Virtual−Dig Up – Graphical User Figure 5 Long-term perspective 3D von Mises
Interface, Optional Data, e.g. laser scans Finite Element Model stress calculation

________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 7 of 11
 Pipeline Technology Conference 2020, Berlin

3 Virtual−Dig Up Blind Test

Together with a North American operator with high reputation in ILI analysis, a blind test
was carried through in 2018 [Danilov 2019]. Three locations covered complex corrosion
with general thinning, complex pitting partly close to pinholes, and slotting of varying
directions. The results are summarized here. Figure 6 shows the match of three 3D VDU
results vs. laser scans and Figure 7 the corresponding comparison of the profiles with
RSTRENG safe pressure predictions and maximum depth.
Specifically the continuous general thinning and its extent was reflected well in the VDU
MFL calculation result. These parts of complex corrosion are difficult to be picked up in
conventional MFL interpretation, but often have significant influence on pressure
calculations.Maximum depth for very small sized elements, like embedded pinholes, may
show small depth underestimation as it does even for UT pitting or laser scans itself.
The typical result observed in mainly all test scenarios was confirmed as well: the
robustness of this new ILI approach with regards to the application of the Effective Area
Method. The safe pressure of all three types of complex metal loss varied with a deviation
< 2%, whilst deviations of the established box simplifications are usually an order higher.

Figure 6 3D maps of a blind test of VDU vs. high resolution laser scan maps at three locations of complex
corrosion with general thinning, complex pitting partly close to pinholes, and slotting of varying directions

________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 8 of 12
 Pipeline Technology Conference 2020, Berlin

The MFL evaluation method presented here generates detailed three-dimensional maps
of the metal geometry, which agree well with field observations. This approach is
overcoming the principal methodological problem of ambiguity by using two independent
magnetic field directions, which allow in principal for mathematically unique solutions. The
methodological limits of the technique are then data quality or principal attributes, like the
sensor to wall loss distance. This may affect, as mentioned above, small scale structures
and specifically the depth of embedded small deep parts of complex corrosion. In principle,
this effect can have comparable impact on high-resolution UT.

The new approach is strong as it analyzes principal structures. More specifically, general
thinning is well captured. These observations imply that the method works well with the
robust Effective Area Method, which is by nature less sensitive against depth undulations
of small lateral extent, even if affecting the maximum depth eventually.

Figure 7 RSTRENG profiles of VDU blind test vs. high-resolution laser scan profiles at three locations of
complex corrosion with general thinning, pitting partly close to pinholes, and varying direction slotting

________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 9 of 11
 Pipeline Technology Conference 2020, Berlin

4 Conclusions

Current MFL observations:

 The relevance of reliable and accurate ILI safe pressure prediction increases, but
knowledge-based MFL parameterization reaches development limits.
 The established way of MFL data analysis is industry accepted. The approach is
knowledge- and experience-based, parameterizes and simplifies the output
necessarily. Together with complex corrosion, this simplification may cause
difficulties, e.g. over-conservatism.
 Parameterization with boxes produces complications in pressure calculations and
corrosion growth correlations.
 MFL interpretation may find the most probable or most conservative metal loss
geometry, but not necessarily the correct one.

MFL Deep Field Analysis observations:

 Changing the MFL sizing principle from knowledge-based interpretation to
mathematical calculation requires 3D feature description.
 Calculation allows for high MFL result repeatability and independence from sizing
model selection. Its ambiguity can be excluded by integrating two different
magnetic field directions, whilst varying measurement components or sensor
heights would not achieve this.
 Calculation together with two field directions allow to find the actually correct
complex corrosion geometry. First blind tests confirmed this.

MFL Deep Field Analysis perspectives:

 The value of the presented calculation approach is obvious, but comes along with
enormous numerical efforts, which will be reduced in the next development
phase.
 Artificial intelligence may increase the quality of the start model and herewith the
performance. Nevertheless, the calculation will remain the final step, which
guarantees the sizing result quality.
 Currently, the box sizing approach is rather tolerant with varying measurement
data quality. With this new calculation, data deficits become transparent as the
dominant source for individual remaining small deviations. This will help to identify
and justify measurement tool improvement.
 The development of a Virtual−Dig Up service will allow for a detailed analysis of
difficult-to-access areas, or areas with difficult-to-size complex corrosion and their
reliable corrosion growth assessment.
 Three-dimensional metal structures per se allow the modelling of von Mises
stress. Once the quality and expected errors are under control, this approach may
become the basis for direct safe pressure prediction via modelling.

________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 10 of 11
 Pipeline Technology Conference 2020, Berlin

5 Acknowledgements
The authors would like to thank all the ROSEN scientists involved in this development
for many years and the team for excellent technical contributions and good team work.
The partnership, excellent technical support and fruitful discussions with the North
American pipeline operator were very much appreciated as well.

6 Abbreviations
3D three-dimensional MFL-A conventional MFL with
axial magnetizing direction
API American Petroleum Institute
MFL-C transverse MFL with circum-
ASME American Society of
ferential magnetizing direction
Metallurgic Engineers
Ω MFL interpretation operator
DFA Deep Field Analysis
rstreng Remaining Strength approach
F DFA calculation operator
by Kiefner as outlined in
ILI In-line inspection ASME B31G
MFL Magnetic Flux Leakage UT ultrasonic testing
VDU Virtual−Dig Up
7 References
API STANDARD 1163, In-Line Inspection Systems Qualification, American Petroleum
Institute, 2nd ed., 2013, The American Petroleum Institute, Washington DC
ASME B31G, Manual for Determining the Remaining Strength of Corroded Pipelines,
2012, The American Society of Mechanical Engineers, New York
DANILOV, A, Palmer, J., Schartner, A., Tse, V., Computing 3D Metal Loss from MFL ILI
Data
for Reliable Safe Pressure Prediction, Rio Pipeline 2019, Rio de Janeiro, Brazil, 2019
KARIYAWASAM, S., Zhang, S., Yan, J., Huang, T., Al-Amin, M., Improving Safety and
Economy through a More Accurate and Precise Burst Pressure Model, APGA-EPRG-
PRCI, Proceedings of the 22nd Joint Technical Meeting, Brisbane, Australia, 2019
MILITZER, H., WEBER, F.; Angewandte Geophysik-Band 1 – Gravimetrie und Magnetik;
1984; Springer; Wien, New York, Berlin
MILLER, S., CLOUSTON, S., Optimizing magnetic-flux leakage inspection sizing model
performance using hi-resolution nondestructive examination data, in J.Tiratsoo (ed.),
Pipeline Pigging and Integrity Technology, 4th Edition, 2013, Clarion, Houston TX

8 Keywords
ILI In-line Inspection, MFL Magnetic Flux Leakage, safe pressure, burst pressure,
RSTRENG, 3D metal loss, granularity, high resolution profiles, complex corrosion,
boxes, clustering, axial magnetization, circumferential magnetization
________________________________________________________________________________
Calculation of a laser-scan-like 3D defect profile from conventional MFL data by J.Palmer, A.Danilov

Page 11 of 11