Excerpt from the Proceedings of the COMSOL Conference 2010 Paris

Simulation of Flaw Signals in a Magnetic Flux Leakage Inspection

Procedure
O. Nemitz* and T. Schmitte
Salzgitter Mannesmann Forschung GmbH
*Corresponding author: Ehinger Straße 200, 47259 Duisburg, Germany, [email protected]

Abstract: In the inspection of steel products

with respect to flaws a magnetic flux leakage
(MFL) test procedure can be applied. In this
procedure a “horseshoe” shaped yoke is used,
whose legs are wrapped with coils through
which an alternate current with a high frequency
(3 kHz) is flowing. Hereby, a thin magnetic field
(~0.46 mm, skin effect) is induced near the
surface of the test object which leaks from the
material if a discontinuity (e.g. a flaw) is present.
This leakage is measured by a hall probe. The
aim of the simulation is to find out how different
flaw geometries and orientations influence the
Figure 1. Sketch of the MFL inspection.
measured signal. We confine our work basically
to block-shaped flaws (notches). In a first step,
we simulated a 2D-cross-section and varied Thus, a thin magnetic field (~0.46 mm, skin
width, depth and orientation of the notch. In a effect) is induced near the surface of the pipe
second step, 3D-simulations were performed. which leaks from the material if a flaw near the
Here the necessity to adequately resolve the very surface is present (see Figure 2). This leakage
thin magnetic field led to an excessive amount of between the two coils can be measured (usually
nodes. The computation on a fully meshed the x-component of the magnetic flux density,
geometry was only possible with a swept mesh Bx), e.g. by a hall probe.
and the use of symmetries. With the so called During the inspection, the yoke moves across the
impedance boundary conditions we could avoid steel object to scan its whole surface.
meshing the steel domain which saved a lot of
nodes. Hereby, the complete 3D-model could be
simulated and arbitrary notches could be
modelled. Different geometries and orientations
of the flaws were automatically generated using
Matlab scripts. As a result, the influence of
different notch geometries or orientations and a
comparison to real measured data is shown. Figure 2. Magnetic field, leaking from the material at
a notch.
Keywords: Magnetic Flux Leakage, steel
inspection, skin-effect The aim of our simulation is to find out how
different flaw geometries and orientations
1. Introduction influence the measured signal. Finally, we want
to infer the flaw geometry from the signals.
In the processing of steel products, several types However, the focus of this work is the technical
of flaws can occur, for example, cracks, holes, realization of the simulation - especially the
notches etc. To find flaws near the surface, a handling of the very thin magnetic layer in 3D.
magnetic flux leakage (MFL) inspection system
can be applied. In this procedure a “horseshoe” We limit our work mainly to block-shaped flaws
shaped yoke is used (see Figure 1), whose legs (notches) which we varied in width, depth and
are wrapped with coils with an applied alternate orientation. In 3D we also performed some
current (3 kHz). simulations with round structures (drill holes).
The paper is organized as follows: After briefly
describing the use of COMSOL Multiphysics we
consider first simulations on a 2D cross-section.
Section 4 deals with the 3D simulation and the
problem of saving degrees of freedom (DOFs).
Finally a comparison between 2D and 3D
simulation is given, followed by a validation
with real measurements.

2. Use of COMSOL Multiphysics

For our simulations we used COMSOL

Multiphysics 3.5a with MATLAB R2007b. The
situation is modeled using the “Electric and Figure 3. 2D geometry of the cross sectional model.
Induction currents” application mode contained
in the AC/DC module. We used a time harmonic 3.2 Mesh
analysis and solved for both the electric and
magnetic potential. As the yoke moves rather The high frequency of the induction current leads
slow over the steel object’s surface, the influence to a very thin magnetic field (skin effect), in our
of this movement to the induced magnetic field case the skin depth is about 0.46 mm. We have
is negligible. Hence, we set v=0 in the model. to take care that this skin depth is resolved with
at least 3-4 elements (see Figure 4 and Figure 5).
3. 2D simulation

We start with the simpler and less memory and

time consuming simulations in 2D. These
simulations were performed on a cross section of
the inspection system, such that the current
through the coils is perpendicular to the
geometry plane.

3.1 Geometry
Figure 4. The mesh adequately resolves the thin
With the 2D simulation taking place in the x-z- magnetic layer at the top surface of the steel object.
plane, we use the following axis convention:
• x-axis: the horizontal direction (this is
the direction in which the yoke moves)
• y-axis: the height direction
• z-axis: perpendicular to 2D simulation
The 2D cross section shape (x-y-plane) of the
yoke has been approximated by Beziér-curves.
Concerning the coils, we waive to model single
excitation windings. Instead, we define an
external current on the cross section area.
The steel object is a rectangular of 10 mm height
and 115 mm width. The whole inspection scene
is enclosed in a rectangular air region.

Figure 5. The complete mesh consists of about

190.000 elements.

3.3 Different notch shapes (2D)

During the inspection the yoke moves in 2) We can only model obliqueness in the
horizontal direction over the surface. This x-y-plane, but not with components in
movement is simulated by varying the x-position z-direction.
of the notch. A MATLAB script was written that Hence, simulations in 3D are necessary.
creates a notch of given width, height and angle
at a specified horizontal position (see Figure 6). 4. 3D simulation
With this script the movement of the notch could
be achieved automatically. 4.1 Geometry and initial values
This simulation was performed for different
notch width, depth and angles to analyze the The 3D geometry is defined in two steps: First,
influence of these geometric parameters on the we extrude the 2D geometry in z-direction,
magnetic flux density. afterwards we have to close the coils by some
round completions regarding the correct coil
obliqueness. Finally we have to extend the
definition of the external current, which is a little
more complicated in 3D than in 2D (see Figure
Figure 6. Different notch geometries (wide, deep, 8).
oblique).

An example result of the signal of the notches

with different width is shown in Figure 7. On the
x-axis the x-position of the probe (simulated by
x-position of the notch) is depicted, the y-axis
gives the x-component of the magnetic flux
density Bx. Each curve represents the signal of
one notch traverse, the single curves correspond
to one notch width. One can observe that the
maximum of the signal curve is split for
increasing width.

Figure 7. Resulting x-component of the magnetic flux

density (in Tesla) for notches of different depth. The
x-axis shows the position of the probe (in m). Figure 8. Top: complete geometry in 3D. Bottom:
external currents defined on the coil subdomains.
Solving a 2D system is very fast, but has two
major drawbacks: Due to the skin effect we have to choose a very
1) We do not consider any magnetic field dense mesh near the boundary. While this was a
in z-direction, possibly flowing around straightforward process in 2D, meshing the
the notch. whole 3D model with the same accuracy would
result in about 20 millions of DOFs, which is not
capable for our system. Consequently, we have
to reduce the number of DOFs.

4.2 Using symmetries

The yoke, coils, and plate are symmetric with

respect to the x-y-plane and to the y-z-plane. If
we confine our simulation to notches which are
symmetric in the z-direction, we can omit one
halve of the scene. This results in about 10
million DOFs which is still too much. Confining
to notches which are also symmetric in the x-
direction leads to a quarter model (see Figure 9).
Figure 10. A swept mesh is used to account for the
skin effect. Only the symmetric notch in the corner is
triangulated with usual tetrahedra.

By this we end up with a system of equations

with 4.7 mio DOFs which we could solve. In
Figure 11 the norm (top) and the x-component of
the magnetic flux density in a zoom (bottom) are
shown. Unfortunately, it is possible to clearly
recognize the transition between the swept mesh
and the usual tetrahedral mesh at the top surface.
For some reasons slight discontinuities
developed at this transition.
Figure 9. Top left: the complete geometry is divided
by a symmetry plane into two halves. One resulting
halve (top right) is again divided (bottom right) and
results in a quarter scene (bottom left).

4.3 Using a swept mesh

To save further nodes and to allow for the special

structure of the magnetic field we want to define
a mesh that is very dense near the boundary and
coarser far away from the boundary. For this
purpose we use a swept mesh.
One problem with this approach is the notch
which we want to be round in the front corner
with a rounded ground. Here, the swept mesh
can not be applied. Hence, we leave out a
rectangular part in the corner, apply the swept
mesh to the remaining steel plate and afterwards
use a standard tetrahedral mesh for the
rectangular part in the corner (see Figure 10).

Figure 11. Solution plotted on the surface of the fully

meshed quarter scene (top: |B|, whole scene, bottom:
|Bx|, zoom in).
4.4 Impedance boundary conditions

Using symmetries and the swept mesh as

described above, we were finally able to
compute a solution for a fully meshed 3D
situation. But we could only handle symmetric
notches, which is too restrictive for our studies.
Another way of treating the skin effect is the so
called “impedance boundary condition” (cf. [1]).
Here, the phenomena in the thin boundary layer
are “projected” to the surface – an approximation
which is very good if the thickness of the steel is
more than 4 times higher than the skin depth. In
our case the factor is approximately 21, thus, the
approximation is applicable. Figure 12. Solutions (|B|) of the complete 3D model
Using this boundary condition we can skip with impedance boundary conditions. Top: the whole
meshing the interior of the steel object and mesh model, middle: only top surface of the steel plate
(color range rescaled), bottom: zoom in to the solution
only its surface. This saves a lot of nodes in the
in the vicinity of the notch (again rescaled color
simulation (we end up with 1.7 million DOFs for range).
the whole scene). In this way we can use the
whole 3D geometry and define arbitrary notches Figure 13 shows resulting curves for notches
without taking care of any symmetry issues. In with different obliqueness angle in the x-z-plane.
Figure 12 |B| is plotted on the domain surfaces in
different zooms for different color ranges.

Figure 13. Top: Geometry with a notch of 30°

obliqueness in the x-z-plane. Bottom: curves that
represent Bx along a line parallel to the x-axis for
different obliqueness angles in the x-z-plane.
4.5 Plausibility check: Comparison of the 2D Figure 15, the curves are qualitatively quite
model with an elongated 3D notch similar.

In theory, the simulation of a notch in 2D

corresponds to a 3D simulation with an infinite
long notch. Hence, to verify our 2D simulation,
we generated a rather long (40 mm in z-
direction) notch in 3D and compared the
magnetic flux density with the flux-density of a
corresponding 2D simulation (both with a small
lift-off). As one can see in Figure 14 the two
signals are qualitatively very similar, but the
absolute values differ by a factor of about 67.
We don’t have an explanation for this
difference, so far.

Figure 15. Comparison of simulation results with

measured data. The two curves coincide quite well.

5. Conclusion

In our work we simulated the MFL inspection

system and analyzed the signal of different notch
geometries. However, the focus of this paper is
the technical realization of the model as the
magnetic field is concentrated in a thin layer near
the surface due to the skin effect.
In first 2D simulations it was uncomplicated to
resolve this layer adequately, however, in 3D the
necessary number of DOFs let us reach our
limits of memory.
By using symmetries and the swept mesh it was
possible to compute the process on a quarter
scene, but this restricted us to symmetric
notches. With impedance boundary conditions it
was possible to skip the meshing of the inner
steel object and to mesh its surface only. By this
we could take into account the complete 3D
scene and incorporate arbitrary shaped notches.
A plausibility check of the 2D simulation and a
comparison of simulation results with real
Figure 14. Comparison of a very long notch (3D, top) measurements round out the picture.
with the corresponding 2D simulation (bottom).

6. References
4.6 Validation with measured data
1. Jianming Jin, The Finite Element Method in
To validate the simulation at all, we compared Electromagnetics, 753 pages, John Wiley &
the amplitude (in dB) of our simulation results Sons, Inc., New York (2002)
with data measured on a test object with several 2. T. Schmitte, A. Koka, Using Drilled Holes as
notches of varying depth. As one can see in Reference in Magnetic Flux Leakage
Measurements: An Investigation Based on Finite
Element Modelling, www.ndt.net, 2008
3. A. Groos, S. Nitsche, T. Schmitte, Modelling
of Magnetic Flux Leakage Measurements of
Steel Pipes, www.ndt.net, 2006
4. T. Schmitte et. al., Wavelet-Filterung zur
Verbesserung der Empfindlichkeit der Streufluss-
Prüfung, www.ndt.net, 2009
5. T. Orth et. al., Wavelet signal processing of
magnetic flux leakage signals, www.ndt.net,
2009