Philosophical Magazine

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tphm20

Tensorial permeability microstructure model

considering crystallographic texture and grain
size for evaluation of magnetic anisotropy in
polycrystalline steels

Jun Liu & Claire Davis

To cite this article: Jun Liu & Claire Davis (2021) Tensorial permeability microstructure
model considering crystallographic texture and grain size for evaluation of magnetic
anisotropy in polycrystalline steels, Philosophical Magazine, 101:10, 1224-1244, DOI:
10.1080/14786435.2021.1892229

To link to this article: https://doi.org/10.1080/14786435.2021.1892229

Published online: 05 Mar 2021.

Submit your article to this journal

Article views: 54

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tphm20
PHILOSOPHICAL MAGAZINE
2021, VOL. 101, NO. 10, 1224–1244
https://doi.org/10.1080/14786435.2021.1892229

Tensorial permeability microstructure model considering

crystallographic texture and grain size for evaluation of
magnetic anisotropy in polycrystalline steels
Jun Liu and Claire Davis
Advanced Steel Research Centre, Warwick Manufacturing Group, University of Warwick, Coventry, UK

ABSTRACT ARTICLE HISTORY

A finite element microstructure model with permeability Received 3 September 2020
tensors that considers crystallographic texture and grain Accepted 12 February 2021
size based on magnetic domain theory has been
KEYWORDS
developed for the evaluation of magnetic anisotropy in Permeability tensor; finite
polycrystalline steels. The model has proved capable of element; microstructure;
capturing the crystallographic texture, the grain size and crystallographic texture;
the vector induction effects on the effective permeability steels
behaviours for typical textures in steels. The predicted
magnetic properties as a function of the magnetic field
direction enables a quantitative characterisation of the
magnetic anisotropy. The predicted effective permeability
maps can serve as a visual indication of the
crystallographic texture from magnetic values. These
features have been experimentally validated against a
commercial grain oriented electrical steel featuring strong
texture and magnetic anisotropy.

1. Introduction
Iron and steel crystal structures are magnetically anisotropic due to the align-
ment of magnetic dipoles in a crystal cell [1]. It has also been experimentally
confirmed that the cube edges (〈100〉) and the cube diagonals (〈111〉) are the
easiest and the hardest directions of magnetising respectively in iron [2] and
silicon-iron [3] single crystals. This fundamental magnetic anisotropy is inher-
ited by each grain in polycrystalline steels. If grains are randomly orientated, the
anisotropy effect averages out and, as a result, the steels exhibit isotropic beha-
viours. If there are preferred crystallographic orientations present, often
referred to as crystallographic texture, the overall average properties have a
certain anisotropy associated with the texture. This simple yet useful averaging
approach has been applied to predict anisotropic mechanical properties of poly-
crystalline materials, e.g. elastic modulus, based on the corresponding single-
crystal properties, with differences in specific approximations including

CONTACT Jun Liu liuj118@cardiff.ac.uk School of Engineering, Cardiff University, Queen’s Buildings, The
Parade, Cardiff, CV24 3AA, UK
© 2021 Informa UK Limited, trading as Taylor & Francis Group
 PHILOSOPHICAL MAGAZINE 1225

Reuss [4], Voigt [5], Hill [6] and finite element (FE) [7] models. In a similar
manner Daniel et al. [8] estimated the scalar effective magnetic permeability
of polycrystalline materials based on an empirical single crystal anisotropy
and effective medium approximations. L. Kestens [9] took a more basic and
simplified approach proposing an ‘A’ parameter, as opposed to a fundamental
magnetic property, that averages the minimum angle between the magnetisa-
tion (implicitly assumed to be homogeneous) and the closest easy direction
to characterise the so-called magnetic quality of a given texture for non-
oriented electrical steels.
Both Daniel’s and Kestens’ model overlooked some important aspects of the
microstructure, in particular, the morphology of individual grains as well as the
microstructure as a whole, which can also influence the magnetic flux beha-
viours and hence the effective permeability. The models work well for
uniform equiaxed single phase material but cannot be extended to more
complex microstructures. For example, alignment in the microstructure,
especially second phase, often occurs during steel processing, e.g. rolling [10],
and sometimes is present in the final product, e.g. in superduplex stainless
steel (banded austenite and ferrite structures), dual phase steels (banded
ferrite and martensite) or hot rolled C-Mn grades (banded ferrite and pearlite
structures) and can also give rise to magnetic anisotropy. Zhou et al. [11] pre-
dicted the effective permeability for dual-phase steel microstructures rep-
resented by digitised and processed (recognising different phases) real
micrographs by FE modelling. Whilst the approach enables studying the separ-
ate effects of aligned microstructures, phase balance, and more recently grain
size [12], their model does not consider the effect of crystallographic texture,
which may give misleading prediction and interpretation if textures also play
a significant role on magnetic properties in the measurement direction and
on anisotropy.
There is an important implicit assumption in the scalar permeability models
that the magnetic flux density B always parallels with the applied field H, which
is only a valid approximation at low and uniform fields. Some tensor per-
meability models [13–16] have been reported to be able to address this limit-
ation, which also facilitates finite element modelling [17–19] to solve
problems that involve rotational fields and complex geometry. Nevertheless,
tensor models require prior knowledge of the permeability for principal direc-
tions, along which B parallels with H, to formulate a permeability tensor. Note
the principal directions are not readily known or necessarily exist in polycrys-
talline materials. Some models [13, 14] simply took two orthogonal directions
with maximum and minimum permeability as the principal directions and their
values as the elements of a diagonal tensor [13]. This basic approach fails when
the maximum and the minimum permeability occur in non-orthogonal direc-
tions, e.g. in grain oriented electrical steels (GOES). Others went the extra
length to formulate a non-diagonal tensor and obtain the principal directions
1226 J. LIU AND C. DAVIS

and the corresponding permeability values by finding the eigenvalues and

eigenvectors of the tensor [15]. All these empirical permeability tensors can
not predict the anisotropy for given crystallographic textures but only deal
with rotational fields, which could be experimentally applied by a rotational
single sheet tester [20, 21] or be present in electrical steel components in
motors, in the presence of known magnetic anisotropy. Some vector hysteresis
models based on the Presiach model, e.g. [22], or the Jiles-Atherton model, e.g.
[23, 24], have also been reported to model anisotropy hysteresis behaviours
associated with rotational fields. Again, none of these hysteresis models can
predict the magnetic anisotropy associated with crystallographic textures.
There are no reports of a permeability tensor for a cubic single crystal that
can fully describe the observed anisotropy and symmetry. According to the
Neumann’s principle, the tensor representing any physical property of a
crystal should be invariant with regard to the symmetry operation of the
crystal class. In the case of cubic crystals such as electrical steel, the per-
meability tensor that satisfies all the symmetries must reduce to a scalar
[25]. It follows that the corresponding magnetic properties should be isotro-
pic, which would be inconsistent with experiments [2, 3]. This paradox ren-
dered the tensorial approach inapplicable as far as the cubic crystallographic
texture is concerned and thereby make people resort to empirical approaches
using scalar permeability e.g. [8]. The fundamental reason is that the magnetic
structure of, say, α–iron does not have all the symmetries of the crystal struc-
ture. Magnetic domains exist in ferromagnetic materials; there are more than
one direction of magnetic domains even in a single crystal. Their magnetic
structure has a lower symmetry than the crystal structure itself does as illus-
trated in Figure 1 due to the directionality of the magnetic spin. In this paper,
we propose a solution to this paradox by formulating the fundamental per-
meability tensor at the magnetic domain level without violating the general
Neumann’s principle and then extend it to single crystals and then polycrys-
talline grains in turn. Thus, the aforementioned averaging approach based on
single domain properties can be used to predict polycrystalline ones using the
tensor approach.
We have developed a new FE model based on the permeability tensors incor-
porating both microstructure and crystallographic texture and hence enabling a
more accurate and robust prediction of the anisotropic behaviours of effective
permeability. Moreover, our model considers the crystallographic orientation
of each individual grain, as opposed to statistics, i.e. orientation distribution
function (ODF) as usually seen in the literature e.g. [8], and hence is capable
of capturing any local anisotropy (the effects of grain boundary misorientation
on the electromagnetic interactions between adjacent grains and/or spatial dis-
tribution of the specific crystal orientations) as well as global anisotropy (the
effects of texture on the effective permeability anisotropy for the microstructure
as a whole).
 PHILOSOPHICAL MAGAZINE 1227

Figure 1. Schematic of magnetic structure of α–iron showing the magnetic moment directions
of each atom in a crystal cell in its crystal reference frame represented by [100], [010] and [001]
directions. The six magnetic easy directions are marked as numbers.

2. Model
2.1. Formulation of permeability tensors
Assume a cubic crystal is composed of a large number (N ) and equal size of
elementary magnetic domains that can only orientate along one of the magnetic
easy directions, i.e. cubic edges or the 〈100〉 directions. The magnetic structure
of the elementary domains orientated along direction 1 is illustrated in Figure 1.
When a magnetic field h is applied along the direction 1, the induction of a
consequential elementary domain along the direction 1 will be B = m0 mc hê1 ,
where m0 is the permeability of free space; mc is the scale constant defined as
relative elementary permeability, by analogy to the continuum counterpart,
the relative permeability, for the elementary domain along direction 1; ê1 is
the unit vector for direction 1. When h is applied along the other orthogonal
directions, i.e. the direction 3 and 5, the induction will be B = m0 hê3 and
B = m0 hê5 respectively. In other words, the direction 1, 3 and 5 are three prin-
cipal directions along which relative elementary permeability values are mc , 1
1228 J. LIU AND C. DAVIS

and 1 respectively. Therefore, the relative elementary permeability tensor for

direction 1 can be represented by
⎡ ⎤
mc 0 0
m1 = ⎣ 0 1 0 ⎦ (1)
0 0 1

The relative elementary permeability tensors for the other easy directions can
be easily obtained by symmetry and orientation rotation:
⎡ ⎤ ⎡ ⎤
1 0 0 1 0 0
m2 = m1 , m3 = m4 = ⎣ 0 mc 0 ⎦, m5 = m6 = ⎣ 0 1 0 ⎦ (2)
0 0 1 0 0 mc

where the subscript denotes the six easy directions as shown in Figure 1.
The following assumptions, after Bozorth [26], are applied:

(1) When the crystal as a whole is not magnetised, all the domains orientate
along the six easy directions by equal probability.
(2) When an external magnetic field is applied, the crystal is magnetised by re-
distributing the numbers of the domains across the six directions rep-
resented by N = {N1 , N2 , . . . , N6 }, which will be referred to as the
domain configuration, favouring those closest to the external field
direction.
(3) The resulting magnetisation must have a component along the given field
direction.

Heisenberg originally made the first two assumptions in 1930s, which have
since become widely accepted as part of domain theory. Mathematically, the
most probable domain configuration for a single crystal has already been
solved by Bozorth [26]:

N1 = ea+bgx N3 = ea+bgy N5 = ea+bgz

(3)
N2 = ea−bgx N4 = ea−bgy N6 = ea−bgz
where α and β can be determined from the following equations
gx sinh (gx b) + gy sinh (gy b) + gz sinh (gz b) Bh
= (4)
cosh (gx b) + cosh (gy b) + cosh (gz b) Bs

2ea ( cosh (gx b) + cosh (gy b) + cosh (gz b)) = N (5)

N = N1 + N2 + N3 + N4 + N5 + N6 (6)
where Bh is the component of the induction B along the applied field direction
defined by the direction cosine (gx , gy , gz ) with respect to crystal direction 1, 3
and 5; Bs denotes the saturation induction. The effective permeability tensor for
 PHILOSOPHICAL MAGAZINE 1229

a single crystal with the domain configuration N can be obtained by tensor

addition as follows

1 6
msc = Ni mi (7)
N i=1

We now have formulated permeability tensors for an ideal single crystal in its
own crystal reference frame. The relative permeability tensor for an arbitrary
orientation with respect to the specimen reference frame, which is convention-
ally chosen to consist of the rolling direction (RD), transverse direction (TD)
and normal direction (ND) as three axes, can be given as
mg = g−1 msc g (8)

where g is the crystal orientation represented by an orientation matrix (refer to

[27] for the defintion of g and more details on relevant crystallography). Now
consider a grain in a polycrystalline microstructure with orientation g and grain
diameter d. The grain size effect needs considering. Assume the elementary
domains on grain boundaries are not orientated along any easy directions, N
will decrease proportionally with the volume fraction of the grain boundaries
given by 6t/d, where t denotes the grain boundary thickness. One can correlate
t with the misorientation of the grain boundaries to consider the local aniso-
tropy. The domain configuration in the present model, as a statistical represen-
tation of domain directions, does not consider the locality and morphology of
the domains within a grain. The closure domains that are expected to be present
near grain boundaries can be considered as two groups of elementary domains:
one that parallels with any of the easy directions and the other that does not.
The effects of the former are already taken into account as presumably less
favoured easy directions; those of the latter are considered not to contribute
to the permeability tensor. For simplicity in the present paper we consider
the overall effects of the loss of the unparallel elementary domains by modifying
each element in the domain configuration per unit volume in the polycrystal-
line grain as follows
 cg 
Ni′ = Ni 1 − (9)
d
where cg is a material parameter that can be measured experimentally. Note Ni′
reduces to Ni when d approaches infinity, which is equivalent to a stand-alone
single crystal. Combining Equations (7), (8) and (9) one obtains the per-
meability tensor for the grains in polycrystalline microstructures as a function
of the crystallographic orientation, grain size and the domain configuration:

1 6  cg 
m′g = g−1 Ni 1 − m g (10)
N i=1 d i
1230 J. LIU AND C. DAVIS

2.2. Finite element microstructure model

A FE microstructure model based on the above permeability tensors was devel-

oped in MATLAB. The model considers a magnetostatics problem that involves
a uniform static field applied to the microstructure. Substituting the constitu-
tive equation
B = m0 mr H (11)
into the Maxwell’s equations for magnetostatics and choosing the Columb
gauge condition, ∇ · A = 0, one obtains the governing partial differential
equation
1
−∇ · ∇A = J (12)
m0 mr
where A is the vector potential, J the external current density, mr denotes the
relative permeability for the materials, which would be 1 for air and m′g for
the microstructure. To simulate uniform applied fields a Dirichlet boundary
condition of uniform magnetic flux density, Bb , is applied to the model. The
vector potential A can be broken down into two parts as
A = Ar + Ab (13)
where Ar is the reduced vector potential and Ab denotes the vector potential
that satisfies
Bb = ∇ × A b (14)
One solution of Ab can be given as
⎡ ⎤
−0.5yBbz
Ab = ⎣ 0.5xBb z ⎦ (15)
yBb x − xBb y

where x and y are the coordinates; Bb x , Bb y and Bb z are the three components of
Bb . No external current density is applied, i.e. J=0.
The geometry of the model consists of the microstructure and a surrounding
circular region of air as shown in Figure 2. The diameter of the air region is set
to five times the maximum dimension of the micrograph. The microstructure is
composed of a number of entities representing the grains each drawn as a
polygon rather than a single-entity micrograph or digital image. The model
accepts either virtual microstructures together with simulated crystallographic
texture data or measured Electron Backscatter Diffraction (EBSD) data. A
boundary condition of
n × A = n × Ab (16)
is applied to the outer edge of the air region, Γ, which is considered to be far
 PHILOSOPHICAL MAGAZINE 1231

Figure 2. Geometry and meshing of finite element microstructure model for a 1 mm×1 mm
virtual microstructure consisting of 500 grains. The mesh is generated by MATLAB Partial Differ-
ential Equation Toolbox.

away from the microstructure to simulate the external magnetic flux density,
where n denotes the unit normal vector.
It is important to note that m′g is not a constant tensor but dependent on the
induction B and hence the FE solution, A. It follows that the FE model is non-
linear and hence tends to be complex and computationally costly to solve. For
simplicity and computational efficiency, we recursively solve the average B
across the whole microstructure, as opposed to at all nodes, at each iteration
step, as illustrated in the flow chart, Figure 3. The Patternsearch algorithm in
the MATLAB Global Optimization Toolbox, is used and the model usually con-
verges to a very small residual (,0.0001Bs ), typically within 20 iterations. In
each Patternsearch iteration, the permeability tensor for each grain, m′g , is cal-
culated for the current guess on the Bh value, referred to as Bhs . Now that m′g is
known the FE model is linear. The model is solved using MATLAB’s Partial
Differential Equation (PDE) Toolbox. From the model solution the Bh value
for the microstructure, Bhm , is then calculated. The cost function for the Pat-
ternsearch optimisation is F(Bhs ) = |Bhm − Bhs |. Adjust Bhs according to the
Patternsearch algorithm and repeat. The solution of the linear FE model at
1232 J. LIU AND C. DAVIS

Figure 3. A flow chart illustrating the recursive approach for solving the non-linear FE model. ξ
denotes the tolerance for the Patternsearch algorithm.

the end of the optimisation process is taken, at a first approximation, as the sol-
ution to the non-linear problem.
The weak form of the governing equation for the above linear FE model in a
solvable form by the PDE Toolbox is

−∇ · (c∇A) + aA = f (17)

where the coefficients c, a and f are specified as follows as per the rules set out in
[28]. For the air domain, c = m1 , f = Ab and a is set to 1; for the whole micro-
0
structure a is set to 0 and f a 3 × 1 zero vector; for each individual grain,
⎡ ⎤
n11 0 n12 0 n13 0
⎢ 0 n11 0 n12 0 n13 ⎥
⎢ ⎥
⎢ n21 0 n22 0 n23 0 ⎥
c=⎢
⎢ 0
⎥ (18)
⎢ n21 0 n22 0 n23 ⎥
⎥
⎣ n31 0 n32 0 n33 0 ⎦
0 n31 0 n32 0 n33

where n is the inverse of m′g for that grain and the subscripts denote the index of
the element in n.
 PHILOSOPHICAL MAGAZINE 1233

The effective permeability of the microstructure, m

 can be evaluated from the
FE solution by
B
m
 def
= (19)
m0 H
where B and H are the magnitude of B and H for the microstructure. The
angle between the B and H vectors, Θ, is calculated by
B·H
Q = arccos (20)
BH

2.3. Microstructure and texture data

Virtual microstructures of targeted grain size and shape are simulated by the
open-source software Neper [29]. To simulate texture data, ODFs were
created given a mode of orientations and corresponding distribution kernel
functions and half width for the spread, in the open-source MTEX toolbox
for MATLAB [30]. In this paper, the default values, i.e. the de la Valee
Poussin function and 10◦ half width, were used; cubic crystal symmetry and
orthorhombic specimen symmetry are also applied to all texture data. Crystal
orientations were then generated from the ODFs and allocated randomly to
each grain in the microstructure.
A separate MATLAB code was developed to convert the raster EBSD data
into polygons with continuous and smoothed grain boundaries, as opposed
to discontinuous segments available in commercial EBSD software packages,
representing each grain ready for the FE geometry. The average grain orien-
tations for each grain were calculated using the Aztec software package.

3. Modelling results
3.1. Effects of crystallographic textures

To predict the separate effects of crystallographic texture on magnetic aniso-

tropy, we allocated different textures to a virtual microstructure consisting of
500 grains whilst keeping the material parameters constant. mc is set to 1000,
which, as a rule of thumb, will give predicted effective values of around 333
for random textures at a very small field (which has been measured experimen-
tally for fully ferritic steel [31]). cg is set to 0 to exclude the grain size effect.
Figure 4 shows the predicted effective permeability for some typical textures
in steels including the fibre texture with 〈100〉 in parallel with ND, notated as
〈100〉  ND and also known as θ fibre, the γ fibre (〈111〉  ND), the η fibre
(〈100〉  RD), the α fibre (〈110〉  RD) and the Goss texture ({011}〈100〉)
along a series of directions swept from RD by 10◦ interval to 180◦ . All the
1234 J. LIU AND C. DAVIS

Figure 4. Predicted effective permeability for some important textures using a same virtual
microstructure as a function of the directions of the applied field with respect to the RD.

curves are symmetrical with respect to the 90◦ axis as expected of the cubic
crystal and specimen symmetry. The highest m  values occur at RD for the η
fibre and Goss texture which are both 〈100〉 directions, i.e. the magnetic easy
direction; and the lowest value occurs at approximately 54.7◦ from RD for
the Goss texture, which parallels with the 〈111〉, i.e. the magnetic hard direc-
tion. The γ fibre and the θ fibre exhibit isotropic permeability within the micro-
structure plane, which would be consistent with their in-plane random
orientations, as a ND fibre, averaging out.
Figure 5 shows the effective permeability maps for some selected textures
and background field directions. The γ fibre (Figure 5(c)) and the Goss
texture (Figure 5(i)) exhibit distinctive m
 maps despite very similar average m
values (see Figure 4). The former features more or less random m  values
across the microstructure indicating random in-plane orientations whilst the
latter shows much less variation corresponding to the single texture (note the
simulated uni-mode ODF for the single texture is not an ideal single crystal
but has a 10◦ spread half width). Similarly, the ND fibres, e.g. the γ fibre
(Figure 5(b)), exhibits more variation than the RD fibre, e.g. the α fibre
(Figure 5(d)), for the RD field direction. By comparison, the θ fibre map has
systematically higher values than the γ fibre one and similar randomness
across the microstructure. The Goss texture map features predominately
uniform colours for each selected directions but are distinctive between each
other.
 PHILOSOPHICAL MAGAZINE 1235

Figure 5. Predicted effective permeability maps for selected textures and field directions. The
colours are mapped to the permeability values using a same virtual microstructure. (a) θ fibre
along TD, (b) γ fibre along RD and (c) TD, (d) α fibre along RD, (e) η fibre along RD and (f) TD, and
Goss texture along (g) RD, (h) 55◦ from RD and (i) TD.

The consistency in all these permeability behaviours demonstrates that the

present model is capable of capturing the crystallographic texture effects on
magnetic anisotropy. The predicted permeability curves serve as a quantitative
characterisation of the magnetic anisotropy associated with texture. In addition,
the permeability maps serve as an enhanced visual and quantitative indication
of the textures as a supplement to inverse pole figure (IPF) maps.
Thanks to the tensorial permeability, as opposed to scalar ones, the model is
also capable of predicting the angle between the B and the H vector, Θ. Figure 6
shows the predicted average Θ values for the different textures. Similar to the
effective permeability behaviours, the in-plane isotropy of the ND fibres includ-
ing the γ and the θ fibres also manifests itself in the Θ behaviours. It is worth
noting that Θ is without regard to the rotation axis direction. Thus, Θ values
do not average out to be zero despite the in-plane isotropy as a whole. The Θ
1236 J. LIU AND C. DAVIS

behaviours of the RD fibres and the Goss texture exhibit more undulated ani-
sotropy than their effective permeability behaviours. The troughs appear to
occur near the 〈100〉 direction, e.g. the RD for Goss and η fibre, the 〈110〉 direc-
tion, e.g. the RD for α fibre, as well as the 〈111〉 direction, e.g. 55◦ from RD for
the Goss texture and the η fibre. This behaviour would be consistent with the
literature [32, 33] reporting that these directions are the principal directions
where B  H.

3.2. Effects of uniform applied field strength

The uniform applied field magnitude, Bb , normalised against Bs , is set to 0.1
for all the above modelling and the average B values of the microstructure, B, 
eventually converge at 0.3–0.31. Owing to the non-linearity of the present
model the predictions are also dependent on B.  Figure 7 shows the m  and Θ

values as a function of B for the α fibre and the Goss texture for the Bb direc-
tions along which the maximum and the minimum m  values occur respectively.
The predicted m  values for all the conditions increase from approximately 333,
i.e. one third of the mc value, at different rates, by power laws as illustrated by
 by power
the fitting lines. Similarly, the predicted Θ values also increase with B
law. The order of both values for different conditions remain unaffected
throughout the modelled range. The differences between the different Bb direc-
tions increase steadily indicating the anisotropy intensifies with the increase of
the normalised B field up to 0.4.

Figure 6. The average angle between the B and H of the virtual microstructure for some typical
textures in steels.
 PHILOSOPHICAL MAGAZINE 1237

3.3. Effects of grain size

Figure 8 shows the predicted m  as a function of the average equivalent grain

 for a series of virtual microstructures with different number of
diameter, d,
grains in a 1 mm×1 mm square and hence different d,  with and without con-
sidering the grain size effects. Random grain orientations were allocated to

Figure 7. The effective permeability m (a) and the angle between B and H (b) as a function of B
for the α–fibre texture with the Bb along RD and 45◦ from RD and for the Goss texture with Bb
along RD and 54.7◦ from RD with a virtual microstructure consisting of 500 grains. The dashed
lines are power law fitting.
1238 J. LIU AND C. DAVIS

all the microstructures. The predicted m  fitting perfectly

 values increase with d
well with the power law by
6.95
m
 = 362.2 1 −  (21)
d
for cg = 6.95 m whilst remain constant at approximately 362.2 for cg = 0. Note the
remarkable similarity of Equations (21) to Equation (9). This behaviour also agrees
well with the literature reporting the magnetic permeability values increase with
the ferrite grain size by a similar inverse or inverse square root relationship in
extra-low carbon steels [31, 34] or in non-orientated electrical steels [35]. The
results prove that the present model has captured the grain size effects by consid-
ering the loss of elementary domains to the grain boundaries through introducing
the parameter cg . It is interesting and perhaps slightly counterintuitive at first view
that the interactions of magnetic flux with the grain boundaries in the FE model, as
manifested in the transition region near grain boundaries in the permeability maps
as shown in Figure 9, does not capture the grain size effect. Note where there is a
decrease in the m values on one side of the grain boundaries, as compared to the
bulk of the grain, there is increase on the other side cancelling it out. As a result,
the effective m for the microstructure as a whole remains unchanged.

4. Measurements
4.1. Experimental details

A commercial grade GOES featuring strong texture was selected for experimen-
tal validation of the present model. EBSD data were collected across a large area

Figure 8. Predicted effective permeability as a function of the average grain size for a series of
virtual microstructures.
 PHILOSOPHICAL MAGAZINE 1239

Figure 9. Predicted effective permeability maps for a series of virtual microstructure consisting
of different number of grains, ngs = (a) 100, (b) 200, (c) 500, (d) 1000, (e) 1500, (f) 2000.

of approximately 12 × 11 mm2 at a step size of 10 m. Figure 10 shows the

inverse pole figure maps exhibiting strong Goss texture and coarse grains as
expected of this steel grade.
A small (32 × 15 × 16 mm3 ) U-shaped electromagnetic (EM) sensor that
can apply a relatively low magnetic field was used to measure an A4-sized
thin (0.25 mm) GOES sheet at a series of angles (w) with respect to the RD.
The relative permeability values were then indirectly extracted by non-linear
least square regression. More details about the measurement system and the
finite element modelling approaches can be found elsewhere [36].

4.2. Identification of the model parameters

The grain size parameter cg was set to 0 for simplicity considering the
predominantly coarse grains and hence expectedly insignificant effects of cg
on the permeability values. The unknown material parameter mc were ident-
ified by Patternsearch optimisation algorithm fitting the predicted
effective permeability values with the measured ones and at the same time
 Figure 11 shows the optimised predictions of the per-
recursively solving B.
meability values as a function of w agreeing reasonably well with the
measurements.
The permeability behaviours are also generally consistent with the predic-
tions using the generated Goss texture data and virtual microstructure
described in Section 3.1. mc has been identified to be 1497 with B,  for
example, for w = 54.7, having converged at approximately 0.6 and the Θ at
1240 J. LIU AND C. DAVIS

Figure 10. Inverse pole figure maps of the grain oriented electrical steel sample for the (a)
rolling direction (RD) (b) transverse direction (TD) and (c) normal direction (ND) pole.

approximately 24.9◦ . Figure 12 shows the predicted m  maps for the identified
model parameters and the background field along RD, TD, ND and 54.7◦
with respect to RD. These maps visualise the following main characteristics
of the magnetic anisotropy associated with the Goss texture corresponding to
the IPF maps shown in Figure 10. First, the RD map shows generally highest
permeability values indicating the RD being close to 〈100〉 directions.
Second, the TD and ND ones are similar indicating these directions are close
to the same crystal direction. The map for 54.7◦ from RD shows predominantly
low permeability values indicating it is close to 〈111〉 directions.
It should be noted that the prediction of Θ values and the effects of B on the
effective permeability anisotropy cannot be fully validated using the present
measurement technique, which only measures scalar permeability (hence
B  H) and is not capable of changing the applied field strength (which is deter-
mined by the sensor geometry). A sensor system that can measure multiple B
and H components is needed and being developed. The present tensor per-
meability model is fully capable of modelling 3D microstructures and any

Figure 11. Predicted permeability as a function of the field direction using the identified model
parameters compared with the measured ones for the grain oriented electrical steel.
 PHILOSOPHICAL MAGAZINE 1241

Figure 12. Predicted permeability maps for the grain oriented electrical steel sample for
different background field directions. (a) RD, (b) TD, (c) ND, (d) 54.7◦ from RD.

direction in 3D space although only 2D microstructures have been modelled in

this paper. The 2D microstructure in-plane directions are often of more interest
and probably more accurate as far as the microstructural effects are concerned.
The present model is anticipated to be used to provide the permeability aniso-
tropy for predicting sensor measurements for low field EM sensors used for
monitoring steel quality during processing (e.g. [37]) and/or for the interpreting
the EM sensor signals to infer the texture of the steel. The model could also be
used to predict anisotropic effective permeability values that can be input into
other macroscopic FE electromagnetic models as the material property, say,
the permeability of electrical steel components of an electric motor, to consider
the steel’s microstructure and texture. As a FE microstructural model the present
model may potentially be coupled with other microstructural models, e.g. the
microstructure-based crystal plasticity models, for multi-physics modelling.

5. Conclusion
A tensorial permeability finite element microstructure model that considers
crystallographic textures based on magnetic domain theories has been
1242 J. LIU AND C. DAVIS

developed for evaluation of magnetic anisotropy of polycrystalline steels. The

model can predict consistent and logical effective permeability behaviours
and the angle between B and H for some selected typical textures that are
important and common in steel manufacturing. The model has proved
capable of capturing the crystallographic texture, the grain size and the back-
ground field effects on the magnetic anisotropy of steels based on the magnetic
domain theory. The predicted effective permeability curves as a function of the
magnetic field directions and the permeability maps can serve as a quantitative
characterisation of the magnetic anisotropy as well as an enhanced visual indi-
cation of the crystallographic texture from magnetic values. These capabilities
have been initially validated against a commercial grain oriented electrical steels
featuring strong Goss texture and magnetic anisotropy.

Acknowledgments
The authors would like to thank Dr Frenk van den Berg from Tata Steel Europe for the
useful discussion about the work.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
This project has received funding from the Research Fund for Coal and Steel under grant
agreement No. 847296.

ORCID
Jun Liu http://orcid.org/0000-0002-4549-2833

References
[1] R.E. Newnham, Magnetic phenomena, in Properties of Materials: Anisotropy,
Symmetry, Structure, chap. 14, Oxford University Press, New York, 2005, pp. 122–146.
[2] K. Honda, S. Kaya, and Y. Masuyama, On the magnetic properties of single crystals of
iron, Nature 117 (1926), pp. 753–754.
[3] H.J. Williams, Magnetic properties of single crystals of silicon iron, Phys. Rev. 52 (1937),
pp. 747–751.
[4] A. Reuss, Berechnung der Fließgrenze von Mischkristallen auf Grund der
Plastizitätsbedingung für Einkristalle, ZAMM – Z. Angew. Math. Mech. 9 (1929),
pp. 49–58.
[5] W. Voigt, Lehrbuch der Kristallphysik, Vieweg+Teubner Verlag, Wiesbaden, 1966.
 PHILOSOPHICAL MAGAZINE 1243

[6] R. Hill, The elastic behaviour of a crystalline aggregate, Proc. Phys. Soc. Sect. A 65
(1952), pp. 349–354.
[7] M. Kamaya, A procedure for estimating Young’s modulus of textured polycrystalline
materials, Int. J. Solids Struct. 46 (2009), pp. 2642–2649.
[8] L. Daniel and R. Corcolle, A note on the effective magnetic permeability of polycrystals,
IEEE Trans. Magn. 43 (2007), pp. 3153–3158.
[9] L. Kestens and S. Jacobs, Texture control during the manufacturing of nonoriented elec-
trical steels, Texture Stress. Microstruct. 2008 (2008), pp. 1–9.
[10] A. Bükki-Deme, I. Szabó, and C. Cserháti, Effect of anisotropic microstructure on mag-
netic Barkhausen noise in cold rolled low carbon steel, J. Magn. Magn. Mater. 322
(2010), pp. 1748–1751.
[11] L. Zhou, C. Davis, P. Kok, F. Van Den Berg, S. Labbé, A. Martinez-de Guerenu, D.
Jorge-Badiola, and I. Gutierrez, Magnetic NDT for steel microstructure characterisation
– modelling the effect of ferrite grain size on magnetic properties, in 19th World Conf.
Non-Destructive Test. (WCNDT 2016), Vol. 21, Munich, Germany, NDT.net, 2016.
[12] L. Zhou, J. Shen, A. Martinez-De-Guerenu, D. Jorge-Badiola, I. Gutiérrez, P. Kok, F.
van den Berg, S. Labbé, A. Skarlatos, C. Reboud, and P. Lombard, Micro and macro
microstructure modelling for the influence of grain size on magnetic properties in struc-
tural steel and modelled commercial sensor outputs, in 12th ECNDT, Jun., Gothenberg,
2018, pp. ECNDT–0228–2018.
[13] W. Brix and K. Hempel, Tensorial description of the rotational magnetization process in
anisotropic silicon steel, J. Magn. Magn. Mater. 41 (1984), pp. 279–281.
[14] M. Birkfeld, Investigation of the permeability tensor of electrical steel sheet, IEEE Trans.
Magn. 34 (1998), pp. 3667–3673.
[15] R. Langman, Prediction and measurement of rotation of magnetization in an anisotro-
pic polycrystalline ferromagnetic material, IEEE Trans. Magn. 17 (1981), pp. 1159–
1168.
[16] M. Enokizono, S. Kanao, and K. Yuki, Permeability tensor of grain-oriented silicon steel
sheet, J. Magn. Magn. Mater. 133 (1994), pp. 209–211.
[17] F. Martin, D. Singh, P. Rasilo, A. Belahcen, and A. Arkkio, Model of magnetic aniso-
tropy of non-oriented steel sheets for finite-element method, IEEE Trans. Magn. 52
(2016), pp. 1–4.
[18] J. Bastos and G. Quichaud, 3D modelling of a non-linear anisotropic lamination, IEEE
Trans. Magn.21 (1985), pp. 2366–2369.
[19] G. Shirkoohi and J. Liu, A finite element method for modelling of anisotropic grain-
oriented steels, IEEE Trans. Magn. 30 (1994), pp. 1078–1080.
[20] H. Pfutzner, Rotational single sheet testers for multi-parametric testing of soft magnetic
materials – a review, in 11th IMEKO TC4 Symposium on Trends in Electrical
Measurements and Instrumentation and 6th IMEKO TC4 Workshop on ADC
Modelling and Testing 2001, 2001, pp. 422–426.
[21] U. Aydin, F. Martin, P. Rasilo, A. Belahcen, A. Haavisto, D. Singh, L. Daniel, and A.
Arkkio, Rotational single sheet tester for multiaxial magneto-mechanical effects in steel
sheets, IEEE Trans. Magn. 55 (2019), pp. 1–10.
[22] M. Carpentieri, S. Vergura, V. Puliafito, and G. Finocchio, Vector hysteresis model to
describe micromagnetic structures, in 2016 IEEE 2nd International Forum on Research
and Technologies for Society and Industry Leveraging a Better Tomorrow (RTSI),
Sep., IEEE, 2016, pp. 1–4.
[23] P. Rasilo, D. Singh, U. Aydin, F. Martin, R. Kouhia, A. Belahcen, and A. Arkkio,
Modeling of hysteresis losses in ferromagnetic laminations under mechanical stress,
IEEE Trans. Magn. 52 (2016), pp. 1–4.
1244 J. LIU AND C. DAVIS

[24] J.B. Padilha, P. Kuo-Peng, N. Sadowski, and N.J. Batistela, Vector hysteresis model
associated to FEM in a hysteresis motor modeling, IEEE Trans. Magn. 53 (2017), pp.
1–4.
[25] J.F. Nye, ed., Physical Properties of Crystals: Their Representation by Tensors and
Matrices, Clarendon Press, Oxford, 1985.
[26] R.M. Bozorth, The theory of the ferromagnetic anisotropy of single crystals, Phys. Rev.
42 (1932), pp. 882–892.
[27] H.J. Bunge, Orientation of individual crystallites, in Texture Analysis in Materials
Science, H.J. Bunge, ed., chap. 2, Butterworth-Heinemann, Berlin, 1982, pp. 3–41.
[28] Mathworks, Setting up your PDE, in Partial Differential Equation Toolbox User’s
Guide, r2020b ed., chap. 2, Mathworks, Natick, 2020, pp. 69–88.
[29] R. Quey, P. Dawson, and F. Barbe, Large-scale 3D random polycrystals for the finite
element method: Generation, meshing and remeshing, Comput. Methods Appl.
Mech. Eng. 200 (2011), pp. 1729–1745.
[30] F. Bachmann, R. Hielscher, and H. Schaeben, Texture analysis with MTEX – free and
open source software toolbox, Solid State Phenom. 160 (2010), pp. 63–68.
[31] L. Zhou, C. Davis, and P. Kok, Steel microstructure – magnetic permeability modelling:
The effect of ferrite grain size and phase fraction, J. Magn. Magn. Mater. 519 (2021),
p. 167439.
[32] R.M. Bozorth, Determination of ferromagnetic anisotropy in single crystals and in poly-
crystalline sheets, Phys. Rev. 50 (1936), pp. 1076–1081.
[33] H. Vande Sande, F. Henrotte, K. Hameyer, and L. Froyen, A hybrid method for deter-
mining the reluctivity tensor components of Goss textured ferromagnetic materials,
COMPEL – Int. J. Comput. Math. Electr. Electron. Eng. 23 (2004), pp. 579–589.
[34] J. Liu, J. Wilson, C.L. Davis, and A. Peyton, Magnetic characterisation of grain size and
precipitate distribution by major and minor BH loop measurements, J. Magn. Magn.
Mater. 481 (2019), pp. 55–67.
[35] M. Shiozaki and Y. Kurosaki, The effects of grain size on the magnetic properties of non-
oriented electrical steel sheets, J. Mater. Eng. 11 (1989), pp. 37–43.
[36] M. Aghadavoudi-Jolfaei, J. Shen, A. Smith, L. Zhou, and C. Davis, Non-destructive
measurement of microstructure and tensile strength in varying thickness commercial
DP steel strip using an em sensor, J. Magn. Magn. Mater. 473 (2019), pp. 477–483.
[37] F. van den Berg, P. Kok, H. Yang, M. Aarnts, J.J. Vink, W. Beugeling, P. Meilland, T.
Kebe, M. Stolzenberg, D. Krix, A. Peyton, W. Zhu, A. Martinez-De-Guerenu, I.
Gutierrez, D. Jorge-Badiola, K. Gurruchaga, P. Lundin, A. Volker, M. Mota, J.
Monster, H. Wirdelius, C. Mocci, G. Nastasi, V. Colla, C. Davis, L. Zhou, R.
Schmidt, S. Labbé, C. Reboud, A. Skarlatos, T. Svaton, V. Leconte, and P.
LOMBARD, In-line characterisation of microstructure and mechanical properties in
the manufacturing of steel strip for the purpose of product uniformity control, in 19th
World Conference on Non-Destructive Testing, Jun., Munich, Germany, 2016.