Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

ARTICLE IN PRESS

NDT&E International 43 (2010) 343–347

Contents lists available at ScienceDirect

NDT&E International
journal homepage: www.elsevier.com/locate/ndteint

An integral equation model for the magnetic flux leakage method


A.A. Snarskii a, M. Zhenirovskyy b, D. Meinert c,, M. Schulte c
a
Department of General and Theoretical Physics, National Technical University of Ukraine ‘‘KPI’’, Prospekt Pobedy 37, 252056 Kiev, Ukraine
b
Bogolyubov Institute for Theoretical Physics, 14-b, Metrolohichna Str., 03143 Kiev, Ukraine
c
Rosen Technology and Research Center, Am Seitenkanal 8, D-49811 Lingen/Ems, Germany

a r t i c l e in fo abstract

Article history: This paper describes a new method for the calculation of the MFL field from a given defect shape and
Received 2 October 2009 induced magnetisation. The field is described by means of three-dimensional integral equations,
Received in revised form providing a view similar to a standard MFL corrosion detection tool. This method allows the quick
28 January 2010
calculation of the MFL field on a standard PC.
Accepted 29 January 2010
& 2010 Elsevier Ltd. All rights reserved.
Available online 6 February 2010

Keyword:
Magnetic flux leakage method
Integral equation method
Finite element method

1. Introduction Here we will consider a so-called ‘‘linear ferromagnetic


material’’. In the range of magnetic field intensities used in MFL
Magnetic flux leakage (MFL) technique is one of the basic practice, i.e. 10–20 kA/m, the magnetic permeability m in such a
methods of non-destructive testing (NDT) used for the in-line material can be approximated as a constant value.
inspection (ILI) of pipelines and other objects made from A most interesting problem for NDT is the ‘‘inverse problem’’.
materials with high magnetic permeability [1–3]. This is the problem of finding geometrical parameters of a defect
The MFL method uses the fact that near a defect in an from a measured MFL field. According to [4] such a problem is an
externally magnetised material the field is forced out of the metal incorrectly formulated problem, in particular it has no unique
(‘‘leaking’’), which can be detected by field sensors. This ‘‘leak- solution. For its analysis and to get a physically sensible solution
field’’ provides information about the location and size, in some some sort of regularisation must be used. In this paper we use
cases also the shape, of the defect, see Fig. 1. that given by Tikhonov [4]. From the application point of view it is
The problem to define the MFL field from the known shape of a necessary to define a parametrisation of the defect shape, and to
defect is called a ‘‘direct problem’’. It is a problem of mathematical define numerical values for these parameters such that the
physics. Considering the situation presented in Fig. 1 it can be solution of the direct problem is possibly close to the measured
described by MFL field. Future work may consider the effect of measurement
noise on the results found here.
=mH ¼ 0; =H¼0 ð1Þ For a more complex shaped defect see, for example [5], the
analytical solution of a direct problem for a geometrically
Hð 7 1Þ ¼ H0 ð2Þ simple defect shape may only serve as a more or less successful
initial approximation. In practice the direct problem must be
where m is the magnetic permeability of the pipeline steel, H is the solved many times, with different geometrical parameters, until
magnetic field intensity, and H0 is the applied external magnetic the solution closely resembles the measured MFL field. So the
field far from the defect. On the boundary between the pipe calculation speed for the solution of the direct problem and an
material and air standard continuity conditions, for normal automatic algorithm for the consecutive parameter selection
component of flux density and tangential component of field become the prime targets of the investigation.
intensity are applied. It appears that the integral equation method is the best choice
for this case. Its capabilities have been shown previously [6] in the
solution of the direct problem for the two-dimensional case of
 Corresponding author. Tel.: + 49 591 9136 493; fax: + 49 591 9136 121.
semi-infinite space with a surface defect.
E-mail addresses: asnarskyy@gmail.com (A.A. Snarskii),
mzhenirovskyy@gmail.com (M. Zhenirovskyy),
For real objects, the wall thickness is always finite, and addi-
dmeinert@roseninspection.net, dmeinert@web.de (D. Meinert), tionally it is impossible to consider the defect as two-dimensional.
mschulte@roseninspection.net (M. Schulte). In the present paper the method of integral equations is applied to

0963-8695/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ndteint.2010.01.008
ARTICLE IN PRESS
344 A.A. Snarskii et al. / NDT&E International 43 (2010) 343–347

n(r)
ϕ2(r) S1
r−r’

ϕ1(r)

ϕ3(r) S2
r r’

x s X

Fig. 2. Sketch of a surface defect.

and the continuity conditions at the surfaces S1 and S2 between


different materials (see Fig. 2) are
@j1 @j2
m ¼ ; j1 ¼ j2
@n @n

@j1 @j3
m ¼ ; j1 ¼ j3 ð4Þ
@n @n
Here, index 1 denotes the ferromagnetic medium with relative
permeability m ¼ const:, indices 2 and 3 the external medium
(above and below the plate, see Fig. 2), @=@n is the normal
derivative to the surface. Changes of m with the varying MFL field
are not considered.
Using the properties of a flat-distribution potential, the
problem can be reformulated in terms of integral equations. This
allows the determination of the magnetic charge densities
s1 ðx; yÞ; s2 ðx; yÞ corresponding to the surfaces S1, S2 of the
magnetised object [10]
Z
l cosðnr ; rr01 Þ
s1 ðrÞ ¼ dS1 s1 ðr01 Þ
2p S1 jrr01 j
Z 0
l cosðn r ; rr2 Þ l @j0
þ dS2 s2 ðr02 Þ 
2p S2 jrr02 j 2p @nr
Z
l cosðnr ; rr02 Þ
s2 ðrÞ ¼ dS1 s1 ðr01 Þ ð5Þ
2p S1 jrr02 j
with l ¼ m1=m þ1, j0 the potential of the external field, and nr
the normal vector to the surface at point r (see Fig. 2). Similar
integral equations are often used for the calculation of electro-
magnetic fields in inhomogeneous media [9].
The potential jðrÞ is expressed through sðrÞ in a standard
manner [10]:
Z Z
s ðr0 Þ s ðr0 Þ
jðrÞ ¼ dS1 1 10 þ dS2 2 20 þ j0 ð6Þ
S1 jrr1 j S2 jrr2 j

Fig. 1. Principle of magnetic flux leakage (MFL) technique: a magnetic field is The magnetic field can be calculated as
coupled into the test specimen (top). At a defect the field is forced out of the metal,
depending on the location of the defect (middle, bottom). This leak field can be
H ¼ rjðrÞ ð7Þ
measured and provides information on the location and size of the defect. The
Let us describe the surface profile as an analytical function pðx; yÞ.
circles mark the area modelled in this paper.
In this case the integral equation (5) may be rewritten in the
the solution of the direct problem for a three-dimensional surface following form [6]:
ZZ
defect in a plate of finite thickness. l
x1 ðx; yÞ ¼ x ðs; tÞK11 ðx; y; s; tÞ ds dt
2p x0 ;y0 1
ZZ
l l @pðx; yÞ
2. The method of integral equations þ x ðs; tÞK12 ðx; y; s; tÞ ds dt H0
2p x0 ;y0 2 2p @x
ZZ
As it is well known [7–9] the scalar potential of a magnetic l
x2 ðx; yÞ ¼ x1 ðs; tÞK21 ðx; y; s; tÞ ds dt ð8Þ
field in a linear homogeneous ferromagnetic medium (‘‘linear 2p x0 ;y0
case’’) satisfies Laplace’s equation
where H0 is the intensity of the external magnetic field, applied in
DjðrÞ ¼ 0 ð3Þ the positive x direction, x1;2 ðx; yÞ are connected to the magnetic
ARTICLE IN PRESS
A.A. Snarskii et al. / NDT&E International 43 (2010) 343–347 345

charge densities s1;2 ðx; yÞ by


sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   
@pðx; yÞ 2 @pðx; yÞ 2
x1 ðx; yÞ ¼ s1 ðx; yÞ 1 þ þ
@x @y

x2 ðx; yÞ ¼ s2 ðx; yÞ ð9Þ


and the integral kernel equations are
@pðx; yÞ @pðx; yÞ
pðx; yÞpðs; tÞ ðxsÞ ðytÞ
@x @y
K11 ðx; y; s; tÞ ¼
½ðxsÞ2 þðytÞ2 þ ðpðx; yÞpðs; tÞÞ2 3=2

@pðx; yÞ @pðx; yÞ
pðx; yÞ ðxsÞ ðytÞ
@x @y
K12 ðx; y; s; tÞ ¼
½ðxsÞ2 þ ðytÞ2 þ pðx; yÞ2 3=2

pðs; tÞ
K21 ðx; y; s; tÞ ¼
½ðxsÞ2 þ ðytÞ2 þ pðs; tÞ2 3=2

K22 ðx; y; s; tÞ ¼ 0 ð10Þ


Thus we have a system of Fredholm integral equations of the
second kind to determine the magnetic charge density. With the
aid of this charge density we are able to calculate the MFL field as Fig. 4. Example for the x-component of the iterated MFL field.
ZZ
ðxsÞx1 ðs; tÞ
Hx ðx; y; zÞ ¼ 2 2 2 3=2
ds dt ZZ
s;t ½ðxsÞ þ ðytÞ þ ðzpðs; tÞÞ  þ 1Þ l
ZZ xðn
2 ðx; yÞ ¼ xðnÞ
1 ðs; tÞK21 ðx; y; s; tÞ ds dt ð14Þ
ðxsÞx2 ðs; tÞ 2p x0 ;y0
þ 2 2 3=2
ds dt ð11Þ
s;t ½ðxsÞ þðytÞ þ z2  As initial solution the following approach similar to [6] is taken
l @pðx; yÞ
ZZ xð0Þ
1 ðx; yÞ ¼  H0
ðytÞx1 ðs; tÞ 2p @x
Hy ðx; y; zÞ ¼ ds dt
s;t ½ðxsÞ2 þðytÞ2 þ ðzpðs; tÞÞ2 3=2
ZZ xð0Þ
2 ðx; yÞ ¼ 0 ð15Þ
ðytÞx2 ðs; tÞ
þ 2 2 3=2
ds dt ð12Þ Fig. 3 shows the results of some number of iterations on
s;t ½ðxsÞ þðytÞ þ z2 
equation set (Eq. (14)). Apparently the system converges very
quickly, only 15–20 iterations are required for a sufficient result.
ZZ
ðzpðs; tÞÞx1 ðs; tÞ In Fig. 4 a three-dimensional image of the calculated MFL field’s
Hz ðx; y; zÞ ¼ ds dt
½ðxsÞ2 þ ðytÞ2 þðzpðs; tÞÞ2 3=2
s;t x-component of Hðx; y; z ¼ 1Þ is shown. For the calculation the
ZZ defect surface is characterised by the parametric form
zx2 ðs; tÞ
þ 2 2 3=2
ds dt ð13Þ
s;t ½ðxsÞ þðytÞ þ z2  a
pðx; yÞ ¼ h ð16Þ
1 þ b2  xn1 þ c2  yn2
To solve the system of integral equations (8) we use an iterative
where h is the wall thickness, a the metal loss or defect depth, and
procedure. This procedure may be described by the following set
b, c, n1,2 are defect shape parameters.
of equations
ZZ
l
x1ðn þ 1Þ ðx; yÞ ¼ xðnÞ ðs; tÞK11 ðx; y; s; tÞ ds dt
2p x0 ;y0 1
ZZ 3. Comparison with FEM calculations
l l @pðx; yÞ
þ xðnÞ ðs; tÞK12 ðx; y; s; tÞ ds dt H0
2p x0 ;y0 2 2p @x
In order to compare the results of the MFL fields calculated by
the integral equation method with those of an approved
commercial FEM engine the Vector Fields Opera 3D software
has been used.
6 Here as well the scalar potential of the magnetic field is used
and has to fulfil the Laplace’s equation

4 DjðrÞ ¼ 0 ð17Þ
n=20
For the stationary case the magnetic field can be represented
2 n=2 using the scalar potential

H ¼ rjðrÞ ð18Þ
−40 −20 0 20 40 n=1 The boundary conditions at the outer boundaries used to apply
an external field are shown in Fig. 5. Here you can imagine the
−2 elements produced for simulation as well as the different
materials included in the model together with the boundary
Fig. 3. Dependence of the calculated MFL field from system equation (14) on the conditions applied in the magnetisation direction. The air is
number of iterations. shown in grey and the steel plate green.
ARTICLE IN PRESS
346 A.A. Snarskii et al. / NDT&E International 43 (2010) 343–347

The magnetic field becomes (FEM) using the three-dimensional vector fields FEM engine.
j1 j0 These calculations have also been restricted to the linear case
H0 ¼ ð19Þ m ¼ const. To achieve this, we replace the real, nonlinear function
l
The finite element model now splits the geometries within
mðHÞ by a linear approximation m ¼ mðH0 Þ ¼ const:, where H0 is the
applied magnetic field. In practice, as mentioned in Section 1,
little pieces called finite elements and gives one partial differ-
H0  10220 kA=m.
ential equation for each element. These all have to fulfil the
As a first result Fig. 4 shows the x-component of the MFL
equation (17) to get a continuous solution. The calculations have
field as a three-dimensional distribution. You can see a high peak
also been restricted to the linear case having about half a million
maximum together with two smaller minima which is character-
linear tetraeder elements.
istic for such an MFL signal.
The steel plate again is shown in Fig. 6 together with the
In the following the different defect signals calculated by the
resulting magnetic flux density. You can see the influence of the
two different methods are compared in one dimension. As can be
defect in the plate which causes a big resistance for the magnetic
seen in Fig. 7 the method of integral equations described in this
flux.
paper yields results comparable to those from FEM calculations.

4. Result
5. Discussion
To check the offered scheme of MFL field calculations the same
pit parametrisation has been used in finite element calculations The integral equation method allows to quickly calculate the
MFL field of a parameterised defect. The limitation to the ‘‘linear
case’’ m ¼ const: restricts its direct use to shallow features where
the magnetic permeability does not change significantly. For these
defects the method provides a solution to the inverse problem,
since the computation times are much shorter than for a full-
fledged MFL-model. ‘‘Shallow’’ in this case means defect depths
less than 25% wall thickness, for which we find good agreement
between nonlinear FEM calculations and the integral equation
model described here.
For deeper defects the assumption of a constant m is not valid.
Still, the integral equation method can be used to improve the
solution of the inverse problem in an indirect way since it allows
to create a large database of defect signals, which would allow to
test several methods for the solution of the inverse problem. In
particular, it allows the training and testing of methods based on
artificial neural networks, which demand rather large sets of
training data. Using an artificially created database it is possible to
decide upon a general architecture and layout of a neural network
appropriate for this kind of problem. After this, training the
network with measured MFL data is only a small additional step.

6. Conclusions

Using the integral equation method presented in this paper it


is possible quickly to calculate the MFL field of a given defect of
Fig. 5. Picture of the FEM model. You can see the elements produced for any parameterised shape on a standard PC, without expensive
simulation as well as the different materials included in the model together with
the boundary conditions applied. The air is shown grey and the steel plate green.
software. Thus the presented method allows the iterative solution
(For interpretation of the references to colour in this figure legend, the reader is of the inverse problem for MFL data at least for shallow defects
referred to the web version of this article.) where the assumption m ¼ const: is allowed.

Fig. 6. Picture of the steel plate after simulation. You can see the influence of the defect in the plate which causes a big resistance for the magnetic flux.
ARTICLE IN PRESS
A.A. Snarskii et al. / NDT&E International 43 (2010) 343–347 347

15 4 11 1
14 10.8
2 0.5
13 10.6
12 0 10.4 0
11 10.2
-2 -0.5
10 10
9 -4 9.8 -1
-50 0 50 -50 0 50 -50 0 50 -50 0 50
16 6 11.5 1.5
4 1
14 11
2 0.5
12 0 10.5 0
-2 -0.5
10 10
-4 -1
8 -6 9.5 -1.5
-50 0 50 -50 0 50 -50 0 50 -50 0 50

Fig. 7. Examples of MFL fields calculated by FEM (lines) and the integral equation method described in this paper (dots), for the linear case in m, m ¼ 135. Displayed are
slices of the x- and z -components through the field maximum. The parameters used in Eq. (16) are, for (a): h=7, a= 6, n1 =4, n2 = 2, b =c = 0.27, and for (b): h= 7, a =4, n1 = 2,
n2 = 4, b = c= 0.27.

Also, the presented method can be used to create a large [2] Hwang K, Mandayam S, Udpa SS, Udpa L, Lord W, Atzal M. Characterization of
database of defect signals which still contain important para- gas pipeline inspection signals using wavelet basis function neural networks.
NDT E Int 2000;33:531–45.
meters of a measured MFL signal, which would allow to test [3] Ryu KS, Atherton DL, Clapham L. Effect of bulk stress and pit depth on
several methods for the solution of the inverse problem. calculated far and near-side magnetic flux leakage pit signals. Insight
2002;44(5):285–8. and references cited therein.
[4] Tikhonov AN, Arsenin VYa. Solutions of ill-posed problems. New York:
Winston; 1977. ISBN 0470991240.
Acknowledgements [5] Förster F. On the way from the know-how to know-why in the magnetic
leakage field method of nondestructive testing, parts i and ii. Mater Eval
1995;43:1154. see also p. 1398.
This work was supported by the head of Rosen Technology and [6] Lukyanets S, Snarskii A, Shamonin M, Bakaev V. Calculation of magnetic
Research Center (RTRC), P. Rosen. A.S. and M.Z. want to thank the leakage field from a surface defect in a linear ferromagnetic material: an
analytical approach. NDT E Int 2003;36:51–5.
staff of RTRC’s Physics Department and Dr. S. Lukyanets for [7] Tozoni OV, Mayergoz ID. Calculation of three-dimensional magnetic fields.
helpful discussions. Kiev: Tekhnika; 1974. (in Russian).
[8] Shur ML, Zagidulin RV, Shcherbinin VE. Theoretical problems of the field
formation from a surface defect. Defektoskopiya 1988;3:14–25. (in Russian).
References [9] Minkov D, Lee J, Shoji T. Study of crack inversions utilizing dipole model
of a crack and hall element measurements. J Magn Magn Mater 2000;217:
207–15.
[1] Jansen HJM, Festen MM. Intelligent pigging developments for metal loss and [10] Tikhonov AN, Samarskii AN. Equations of mathematical physics. New York:
crack detection. Insight 1995;37(6):421–4. Dover Publications; 1990. ISBN 0486664228.

You might also like