Вы находитесь на странице: 1из 6

Journal of Magnetism and Magnetic Materials 320 (2008) 204209

Measurement of Barkhausen noise and its correlation with


magnetic permeability
O. Stupakov
a,b,
, J. Pala
c
, V. Yurchenko
a,d
, I. Toma s
a
, J. Bydz ovsky
c
a
Institute of Physics, ASCR, Na Slovance 2, 18221 Prague, Czech Republic
b
Institute of Fluid Science, Tohoku University, Katahira 2-1-1, Aoba-ku, Sendai 980-8577, Japan
c
Department of Electromagnetic Theory, University of Technology, Ilkovicova 3, 81219 Bratislava, Slovak Republic
d
Department of Physics, University of Oslo, P.O. Box 1048, Blindern, 0316 Oslo, Norway
Received 25 February 2007; received in revised form 14 May 2007
Available online 3 June 2007
Abstract
The paper investigates applicability of the Barkhausen noise technique as a non-destructive testing method and its correlation with the
magnetic hysteresis measurement. The hysteresis and the Barkhausen noise experiments were performed at laboratory and industrial
congurations on a model series of open at samples plastically deformed by mechanical tension. The sample magnetic eld was
measured with the help of a near-surface Hall sensor. The results proved interrelation between the magnetic differential permeability and
the Barkhausen noise envelope. However, signicant quantitative discrepancies between them were also obtained. The reasons of these
deviations as well as the main measurement problems were discussed in order to help to standardize the Barkhausen noise technique and
to extend a number of the used parameters for the non-destructive testing.
r 2007 Elsevier B.V. All rights reserved.
PACS: 75.60.Ej; 81.70.Ex; 85.70.Ay
Keywords: Barkhausen noise; Magnetic hysteresis; Magnetic non-destructive testing
1. Introduction
The Barkhausen and the hysteresis measurement meth-
ods are widely used procedures in laboratory investigations
as well as in industrial non-destructive testing (NDT)
[110]. However, no commonly accepted standard for the
Barkhausen noise (BN) technique was established yet.
A number of industrial devices available in the market utilize
specially constructed BN probes. Moreover, their design
details and the algorithms for the BN signal processing are
often a commercial information. Generally, such devices
are equipped with a miniature probe composed of an
U-shaped yoke with plain or rounded poles for the sample
magnetization and of a BN sensor between the yoke poles.
This sensor could be differently constructed to pick up the
normal or the tangential components of BN on the sample
surface [2,11]. As a result of the measurement these devices
usually present a relative number, which describes some
average amount of the detected BN.
The research in laboratories proceeds in two main
directions: an extension of number of the used parameters
for NDT [3,5,9,12] and optimization of the magnetic
examination with respect to the measurement conditions
[4,6,13,14]. With recent development of digital measure-
ment systems it became possible to investigate the BN
signal in more details. The researchers focused their
attention on new, not yet utilized in practice, BN
parameters such as a BN envelope, a number of BN
jumps, a BN spectral density, etc. They are tested for
applicability in material evaluation and for possibility to
provide an additional information to the traditional root
mean square (RMS) value of BN. However, the other issue
ARTICLE IN PRESS
www.elsevier.com/locate/jmmm
0304-8853/$ - see front matter r 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmmm.2007.05.022

Corresponding author. Institute of Physics, ASCR, Na Slovance 2,


18221 Prague, Czech Republic. Tel.: +420 26605 2879;
fax: +420 28689 0527.
E-mail address: stupak@fzu.cz (O. Stupakov).
URL: http://www.fzu.cz/stupak (O. Stupakov).
of stable denition of the BN parameters with respect to
the experimental conditions (design of the BN probes,
variation of their contact quality at installation on the
sample surfaces, dependence of the measurement results on
the applied eld rate, etc.) is still an open question.
The laboratory BN setup usually consists of an U-shaped
ferrite or soft iron yoke for the sample magnetization and a
pick-up coil wrapped around the sample or attached to the
sample surface for the measurement. Use of the sample-
wrapped coil is a classical laboratory approach, when the
BN is measured in the magnetization direction [12,14,15].
However, in most industrially preferred cases the coil is
placed on the sample surface, and it picks up the normal
component of BN [3,16,17]. These attached coils are often
equipped with magnetically soft cores amplifying the signal.
In both cases, the measured BN parameters are still mostly
referred to the magnetizing current (the current method of
the sample eld determination) [4].
The present work intends to study the above-mentioned
scantily investigated topics of the BN measurements. We
used the sample-wrapped and the attached coil congura-
tions to compare the results between the different BN-
sensing approaches and additionally to compare the BN
measurements and the corresponding hysteresis ones. For
vivid demonstration, the experiments were done at two
different eld rates on a model series of low-carbon steel
samples, which show two-peak proles of the differential
permeability and of the BN envelope [15,18,19]. These
proles were evaluated versus the sample eld, measured
directly by a near-surface Hall sensor (the surface eld
method of the sample eld determination) [5,14]. This
method was applied by analogy with our previous works,
devoted to optimization of the hysteresis measurements
[20,21]. It was shown there that control of the sample mag-
netization by the Hall sensor/s increases the results stability
and the physical accuracy of the hysteresis measurements
in comparison with the ordinary current method. In a
similar manner, the question of possible advantages of the
surface eld method for the BN technique was investigated
in this work. The second main goal was to compare the
results of the differently congured BN and hysteresis
experiments, in order to dene a way of standardization of
the BN method for NDT.
2. Experiment and data evaluation
The measurements were done on a model series of at
samples of the low-carbon steel CSN 12021 (0.070.15%C,
0.30.5%Mn, 0.170.37%Si, max. 0.15%Cr, max.
0.25%Cu, max. 0.04%P, max. 0.04%S), frequently used
for large diameter pipelines at power plants. The tested
samples with dimensions 90 50 6 mm
3
underwent uni-
axial mechanical tensions up to 0%, 8%, 10%, 13% and
15% of the plastic strain along the longest side. The speci-
mens were tested after unloading applying the magnetic
eld along the stress direction. Since the deformed samples
showed similar magnetic behavior, the specimens strained
up to 0%, 8% and 15% were chosen to present the result.
The samples were magnetized by a single-yoke (trans-
former laminated C-core) of the same width as the samples
(see Fig. 1). The magnetization windings N 2 160 were
placed around the yoke poles. Triangular waveform of the
magnetizing current dI=dt const was used. According to
the current method, the inside-sample magnetic eld H
i
was evaluated to be proportional to the magnetizing
current I: H
i
NI=L, where the magnetic path L was
determined as 58 mm in order to correspond with the
surface Hall measurements at high elds. The used eld
rates were dH
i
=dt 2:2 and 6.6 kA/m/s, which will be
referred to hereafter as the low and the high rates, respec-
tively. The magnetizing current amplitude was chosen to
achieve the maximum eld value H
i
about 5 kA/m.
At such a yoke-sample conguration the specimens were
magnetized homogeneously in the depth in the central part
between the yoke poles [20]. Therefore, the Hall sensor for
the surface eld measurements H
sur
was placed at the yoke-
free side of the samples, where the surface eld gradients are
much smaller than at the opposite side between the yoke
poles [21]. The near-surface Hall probe, developed at the
Institute of Electrical Engineering, SAS, with 0.25mm
distance of the 0:1 0:1 mm
2
sensing area from the sensor
edge and 1 mV7:9 A=m sensitivity was used. At such
conditions the difference between the Hall-measured eld
and the real inside-sample eld was neglected. The obtained
dependencies of the eld rate dH
sur
=dt, which qualitatively
agree with our previous results [21], are shown in Fig. 2(a).
The induction signal for the hysteresis measurements
was recorded by two sensing coils. The rst one was wound
directly around the samples (laboratory condition), and the
other one was wound around the yoke pole near to the
edge (NDT conguration) [20]. The number of turns n were
55 and 20 for the sample and for the yoke coils,
respectively. The magnetic induction B was obtained by
integration of the induced voltage in the coils U
ind
according to the induction law [1,21]:
U
ind
nS
dB
dt
nSm
0
m
dif
dH
dt
, (1)
ARTICLE IN PRESS
Fig. 1. Principal scheme of the setup.
O. Stupakov et al. / Journal of Magnetism and Magnetic Materials 320 (2008) 204209 205
where S is the sample cross-section and m
dif

1=m
0
dB=dH is the magnetic differential permeability.
The amplier shifts were taken into account in order to
close and symmetrize the hysteresis loops.
The used induction rates dB=dtU
ind
, measured by the
sample coil, are shown in Fig. 2(b). In case of the current
method, the m
dif
H
i
is proportional to the induction rate:
dB=dt m
0
dH
i
=dtm
dif
[19,20]. However, evaluation of
the m
dif
H
sur
requires digital derivation: dB=dH
sur
was
calculated and gently smoothed to obtain the differential
permeability. For additional comparison, the hysteresis
properties of the non-deformed homogeneous sample were
measured on magnetically closed ring-shaped specimen
with the magnetizing and the induction coils, uniformly
wound along its perimeter. The used triangular current
waveform magnetized the ring specimen with the constant
eld rate [1,21].
The BN signal was also measured by two methods: by
the same sensing coil, wrapped around the samples, and by
a specially designed attached coil (or BN read-head),
positioned between the yoke poles again at the yoke-free
side of the samples (see Fig. 1). The attached coil had 2000
ne wire turns around a 1 2 mm
2
rectangular core of 34
stacked soft amorphous ribbons (2 mm wide, 15 mm long
and 0.02 mm thick) of Co
67
Cr
7
Fe
4
Si
8
B
14
with effective
initial permeability m
in
10
3
. The BN signals as well as the
magnetizing current/induction/Hall signals were amplied
and sampled at 25 and 50 kHz for the low and the high-
eld rates, respectively. The BN signals were band-pass
ltered: the low-frequency components up to 1 kHz for the
sample-wrapped coil and up to 0.3 kHz for the read-head
were cut out. The upper cut-off frequency of 30 kHz was
far below the resonance frequencies of the used sensing
coils (50100 kHz for the read-head) [22]. The relatively
narrow bandwidth of BN was determined by limitations of
the analog-to-digital converter. After the measurement, all
the experimental signals were averaged over 500 adjacent
points (RMS averaging for the case of BN envelopes) and
smoothed to get rid of noise [21]. Threshold of disturbing
noise for BN was determined as a constant value of the
envelope for the non-deformed sample at near-saturated
elds. The obtained envelopes U
env
, shifted by this thresh-
old, were referred to H
i
as well as to H
sur
and compared
with m
dif
.
3. Results
Fig. 3 presents result of the hysteresis measurements for
the non-deformed sample at different experimental condi-
tions. The effects of the eld rate and of the eld reference
method are illustrated in Fig. 3(a) by the differential
permeability curves, measured with the sample-wrapped
coil. For the sake of simplicity the corresponding hysteresis
loops are shown in Fig. 3(b) for the ascending branches
only. They conrm our assumption about an increasing
overestimation of the measured magnetic induction with
the applied eld [20]. The results of the ring-shaped
measurements with the constant eld rates are shown for
comparison (0.6 kA/m/s corresponds to the minimum of
the eld rate dependence in Fig. 2(a)). In contrast to the
presented results for the non-deformed sample, the
magnetically harder deformed specimens already reveal a
near-quasistatic behavior at the used eld rates (see
Fig. 2(b)).
Fig. 4 presents the BN results: the effects of the eld rate
and of the eld reference method by the example of the
sample coil measurements together with the comparison of
the two used methods of the BN sensing. The scales of the
BN envelopes are given in absolute values before the
amplication. The interesting fact is that the used eld/
induction rates differ by a factor of three (see Fig. 2), but
the values of BN signal at these two rates differ nearly
twice as shown in Fig. 4(a). The same is also valid for the
read-head experiments. In addition, the sample coil and the
read-head measurements display qualitatively similar
behavior for the investigated sample series (see Fig. 4(b)).
However, quantitative variations are evident: different
sensitivity at the negative low-eld region, different peak
positions, envelope shapes.
Comparison of the differential permeability curves and
the BN envelopes in Fig. 5 proves the same origin of
the phenomena, but again with substantial quantitative
ARTICLE IN PRESS
Fig. 2. Field (a) and induction (b) rates used for the measurements versus H
sur
. The rates for the other deformed samples are very close to the presented
dependencies for the 8%-strained specimen.
O. Stupakov et al. / Journal of Magnetism and Magnetic Materials 320 (2008) 204209 206
deviations. Very interesting result is that the sensitivity of
the BN technique to the considered heavy deformations is
much higher than that of the hysteresis testing. Also it can
be seen that correlation between the permeability and the
BN proles is dependent on the used method of the eld
determination.
4. Discussion
The hysteresis measurements have been used for
investigation of ferromagnetic materials for more than
150 years [1]. As a result, several standard techniques were
introduced in practice [4]. However, the principal problem
of the methods is that the obtained hysteresis is determined
not only by the material properties but also by the
experimental conditions: geometry and spacial scale of
the magnetizing setup and the tested samples, the
magnetization frequency, the driving waveform, etc.
[2,23]. Therefore, the used measurement approach and
the experimental details should always be taken into
account at magnetic investigation. Comparison of the
results obtained by different techniques should be done
carefully, and in rather relative than in absolute scale.
The main idea of the article was to present results of the
hysteresis measurement as a basis for understanding of the
less-studied BN technique. The great advantage of the
hysteresis measurement is that the induced voltage signal is
simply proportional to the magnetic induction rate (see Eq.
(1)) [2]. It is also much higher, less noisy, and therefore
easier for further interpretation than the BN. The
hysteresis can be stabilized at low-magnetization frequen-
cies (quasistatic regime), can be veried on the requirement
of the loop symmetry or on the known maximum induction
value. However, as it was shown above, the used hysteresis
method also contains several intrinsic sources of measuring
errors.
First of all, it is the overestimation of the sample
magnetic ux due to additional measurement of the
leakage ux between the yoke poles together with the
useful signal from the tested sample (see Fig. 3(b)) [20].
Mobility and exibility of the testing device, demanded by
industrial applications, ask for positioning of the induction
coil on the yoke body rather than around the sample.
However, even at the laboratory conditions, with the
sample-wrapped pick-up coil, the measuring induction is a
little overestimated because of an isolation gap between the
ARTICLE IN PRESS
Fig. 4. BN envelopes, measured by the sample coil at different experimental conditions (a) as well as measured by the sample and the attached coils at the
low-eld rate (b).
Fig. 3. Differential permeability curves (a) and ascending hysteresis branches (b) for the non-deformed sample at different experimental conditions. The
inset illustrates the near-coercivity region in detail.
O. Stupakov et al. / Journal of Magnetism and Magnetic Materials 320 (2008) 204209 207
sample and the coil. Moreover, the sample/yoke induction
coils evaluate the magnetic ux, whichfor the bulk
specimenscan penetrate into the sample volume in a
strongly non-homogenous way [21].
The main disadvantage of the single-yoke method still
remains the general contact problem of the attached
sensor: the dependence of the measurement results on the
contact quality between the yoke and the sample. The
poorly dened air-gaps between the attached yoke and the
tested industrially shaped samples can lead to considerable
uctuation of the magnetization conditions, namely of the
eld rate and of the eld amplitude, which correspondingly
leads to changes of the induction signal. In our recent
works [20,21], it was shown that the contact problem can
be xed by precise determination of the inside-sample eld.
However, this solution is surely limited by the quasistatic
range of the measurements (see Fig. 3(a)).
The used NDT conguration with the magnetizing coils
wound around the yoke poles leads to the complex
dependence of the magnetization rate (see Fig. 2(a)). There
is a minimum in the coercivity region, which is determined
by penetration of the yoke-generated ux into the sample
volume [21]. In the ideal case, the whole range of the
applied magnetization rate should not exceed the quasi-
static threshold. The inset in Fig. 3 clearly demonstrates the
mentioned measuring errors: the increased coercivity due
to the dynamical magnetization and the increased perme-
ability/hysteresis slope due to the induction overestimation.
Therefore, it would be more physically accurate to measure
with the constant eld/induction rate [1]. This could also
change the correlation between the hysteresis and the BN
responses [2]. However, it substantially complicates the
device.
In analogy with the analysis of the hysteresis measure-
ments, the present work investigates possible advantages of
the surface eld method for the BN envelope presentation.
Despite the long history of BN measurements, evaluation
of the BN envelope is quite a new topic. The similarities
between shapes of the differential magnetic permeability
and of the BN envelope were also realized recently [11,24].
In contrast with the traditionally measured total RMS
value of BN, determination of the local RMS prole of BN
is more difcult and less stable because of the stochastic
nature of the BN emission [2,11]. However, it can give
deeper understanding of the underlying physical processes
similar to the differential permeability [19]. Another
potential benet of the BN read-head technique is the
local measurement of the stray eld variation caused by
sub-surface domain rearrangements. The magnetic ampli-
fying core of the read-head can be easily designed into any
necessary shape according to the locality requirement and
pressed to the sample surface by a spring mechanism to
reduce the contact problem [16,17]. It looks probable that
the local surface measurement of the BN envelope, referred
to the near-surface measured sample eld, could offer more
stable results than the bulk hysteresis testing accompanied
by the induction measuring errors. This would be very
attractive for practical use in the NDT eld.
The obtained results proved our expectations about
correlation between the hysteresis and the BN processes
(see Fig. 5). The BN envelopes showed similar two-peak
behavior as the differential permeability for the uniaxially
deformed samples. However, quantitative deviations be-
tween the different types of the measurements are also
evident. Moreover, the BN envelopes, measured by the
sample coil and by the read-head, are different (see
Fig. 4(b)). But the main problem is that, in contrast to
the hysteresis measurements, there is no way for compar-
ison of the different BN results. There is no normalization
law for the BN signal as for the hysteresis induction signal
(see Eq. (1) and Fig. 3). The obtained experimental BN
signals are also difcult to be interpreted theoretically,
because usually the transfer function between the magne-
tization changes in the material and the measured signal is
not known [2]. In addition, even within the quasistatic
range of the hysteresis measurements, the non-linear
dependence of the BN signal level on the magnetization/
ux rates was found (see Fig. 4(a)).
At the moment there is no clear understanding of the
correlation between the hysteresis and the BN behaviors
yet and there is no standard widely accepted scheme for the
BN measurements either. Lively discussions are devoted to
ARTICLE IN PRESS
Fig. 5. Comparison of the differential permeability curves and the BN envelopes, measured by the sample coil at the low-eld rate, with the current (a) and
the surface eld methods (b).
O. Stupakov et al. / Journal of Magnetism and Magnetic Materials 320 (2008) 204209 208
explanations of the obtained BN proles due to different
sensitivity of the measurement methods to the various
stages of the magnetization process: nucleation/annihila-
tion/movement of the 90

=180

domain walls [11,23,25].


This should be really one of the principal reasons of the
quantitative disagreement between different methods of the
BN measurements, in particular between the used methods
of the sample-wrapped coil and the BN read-head. The
head records the BN emission predominantly in the
perpendicular direction to magnetization from the stray
ux, whereas the coil picks up noises from the sample
surface mainly along the magnetization line. The sensitivity
of the coils to the normal components is supposed to be
negligible. This is proved by the fact that the read-head
signal does not contain any signicant low-frequency
hysteresis component, whereas for the sample-wrapped
coil it is problematic to accurately lter it out [26]. The
geometrical parameters of the sensing coils (inner/outer
diameters, height, number of turns, material of the core for
the read-head) are also of strong inuence to the
measurement results [22]. However, exact relations are still
unclear, and further detailed investigations are required to
establish them [2,26].
Another reason of the considered deviations, especially
between the differential permeability and the BN envel-
opes, can originate from use of the magnetizing current as
the reference value (see Fig. 5). Evidently, it is more
accurate to utilize the surface sample eld for the reference,
and in such a way to control the magnetization condition
inuencing the BN response (see Fig. 4(a)). Moreover,
despite the proved similarity of the hysteresis and the BN
results, it was also shown that the methods can illustrate
different sensitivity to the tested structural modications
(see Fig. 5). The possible reason of higher BN sensitivity is
an increased density of the surface defects due to the
sample polishing and the following mechanical deforma-
tion. Therefore, a more general approach could be the
optimum: to combine these two techniques for the material
investigations in order to obtain the complex magnetic
response and to choose the best solution for NDT (the
propagating idea of multi-parameter sensor) [10,12].
5. Conclusions
Investigation of the hysteresis and of the BN techniques,
based on the single-yoke magnetization, was performed
with the aim to nd correlations between these conjugate
measurement methods and to discuss an optimum design
of a magnetic NDT device. The experimental results
proved that despite the close relationship between the
magnetic differential permeability and the BN envelope
there are substantial quantitative differences between them,
which are additionally dependent on the setup congura-
tion. It was found that the BN emission illustrates different
sensitivity to the material modication and manifests quite
different dependence on the magnetization rate than that of
the magnetic hysteresis. It was also proved that application
of the directly measured surface eld as the reference value
for the BN envelope is more accurate than use of the
magnetizing current. It was proposed to combine the
hysteresis and the BN techniques in practice, in order to
obtain more detailed magnetic information about the
investigated materials.
Acknowledgments
The authors appreciate nancial support of the Academy
of Sciences of the Czech Republic through the project nos.
1QS100100508 and AVOZ10100520 and of the Ministry of
Education of the Slovak Republic through the project
VEGA No.1/3116/06. O. Stupakov gratefully acknowl-
edges support by JSPS postdoctoral fellowship.
References
[1] R.M. Bozorth, Ferromagnetism, D. Van Nostrand, Toronto, 1951.
[2] G. Bertotti, Hysteresis in Magnetism, Academic Press, San Diego,
1998.
[3] G. Dobmann, N. Meyendorf, E. Schneider, Nucl. Eng. Des. 171
(1997) 95.
[4] J. Sievert, J. Magn. Magn. Mater. 215216 (2000) 647.
[5] R. Baldev, T. Jayakumar, V. Moorthy, S. Vaidyanathan, Russ.
J. Nondestr. Test. 37 (2001) 789.
[6] G.S. Korzunin, V.K. Chistyakov, F.F. Rimshev, V.M. Shevnin,
Russ. J. Nondestr. Test. 37 (2001) 397.
[7] J. Anglada-Rivera, L.R. Padovese, J. Capo -Sa nchez, J. Magn. Magn.
Mater. 231 (2001) 299.
[8] I. Me sza ros, J. Proha szka, J. Mater. Process. Tech. 161 (2005) 162.
[9] G. Ve rtesy, I. Toma s , I. Me sza ros, J. Magn. Magn. Mater. 310 (2007)
76.
[10] T. Chady, R. Sikora, G. Psuj, M. Enokizono, T. Todaka, IEEE
Trans. Magn. 41 (2005) 3721.
[11] D.C. Jiles, Czech. J. Phys. 50 (2000) 893.
[12] B. Augustyniak, M. Chmielewski, M.J. Sablik, IEEE Trans. Magn.
36 (2000) 3624.
[13] V.F. Matyuk, S.A. Goncharenko, H. Hartmann, H. Reichelt, Russ. J.
Nondestr. Test. 39 (2003) 347.
[14] O. Saquet, J. Chicois, A. Vincent, Mater. Sci. Eng. A-Struct. 269
(1999) 73.
[15] C. Gatelier-Rothea, J. Chicois, R. Fougeres, P. Fleischmann, Acta
Mater. 46 (1998) 4873.
[16] V. Moorthy, B.A. Shaw, S. Day, Acta Mater. 52 (2004) 1927.
[17] J. Gauthier, T.W. Krause, D.L. Atherton, NDT&E Int. 31 (1998) 23.
[18] J.M. Makar, B.K. Tanner, J. Magn. Magn. Mater. 222 (2000) 291.
[19] O. Stupakov, J. Pala, I. Toma s , J. Bydz ovsky , V. Nova k, Mater. Sci.
Eng. A-Struct. 462 (2007) 351.
[20] O. Stupakov, I. Toma s , J. Kadlecova , J. Phys. D 39 (2006) 248.
[21] O. Stupakov, J. Magn. Magn. Mater. 307 (2006) 279.
[22] V. Stro m, Integral and local AC-susceptibility: instrumentation and
applications, Ph.D. Thesis, Royal Institute of Technology, Stock-
holm, 1999, pp. 2324.
[23] D.K. Bhattacharya, S. Vaidyanathan, J. Magn. Magn. Mater. 166
(1997) 111.
[24] M. Blaow, J.T. Evans, B.A. Shaw, Acta Mater. 53 (2005) 279.
[25] J.A. Pe rez-Benitez, J. Capo -Sa nchez, J. Anglada-Rivera, L.R. Padovese,
J. Magn. Magn. Mater. 288 (2005) 433.
[26] A.J. Moses, H.V. Patel, P.I. Williams, J. Elec. Eng. 57 (8/S) (2006)
38 (ISSN 1335-3632).
ARTICLE IN PRESS
O. Stupakov et al. / Journal of Magnetism and Magnetic Materials 320 (2008) 204209 209

Вам также может понравиться