11th World Congress on Computational Mechanics (WCCM XI)

5th European Conference on Computational Mechanics (ECCM V)

6th European Conference on Computational Fluid Dynamics (ECFD VI)
E. Oñate, J. Oliver and A. Huerta (Eds)

COMPUTATION OF THE EFFECTIVE LAMINATION

STACK’S BEHAVIOR CONSIDERING THE CONTACT
SIMULATION WITH A MULTI-SCALE HOMOGENIZATION

V. Luchscheider∗ , M.Maidorn† and K. Willner∗

∗ Chair of Applied Mechanics
Egerlandstr. 5, 91058 Erlangen
e-mail: vera.luchscheider@ltm.uni-erlangen.de
e-mail: kai.willner@ltm.uni-erlangen.de
†Siemens AG, Industry Sector
Drive Technologies Division, Large Drives
Vogelweiherstr. 1-15, 90441 Nuremberg
e-mail: mischa.maidorn@siemens.com

Key words: Contact Mechanics, Homogenization, Lamination Stack.

Abstract. The influence of the lamination stack on the mechanical behavior of an elec-
trical machine is significant, especially for lightweight designs. Information about the
stack’s behavior is relevant for a proper calculation of the whole motor. This behavior
is dependent on the contact between the single sheets. This paper aims to show that
the normal contact behavior can be simulated with the elastic model of Bush, Gibson
and Thomas, a plastic term depending on deformation of the highest peaks and a viscous
term. The tangential slip is described through a micro-macro model. These models can
be used in a representative volume element and consequently for a multi-scale homoge-
nization. A material model of the whole lamination stack can be identified by loading the
representative volume element with different deformations.

1 INTRODUCTION
It is common to simulate the mechanical behavior of machines before they are built.
For a proper simulation the mechanical behavior of all single parts and the different
loading situations have to be known. In the case of electrical machines, the behavior of
the lamination stack is unknown in detail, but it becomes relevant with the growing focus
on lightweight construction. Therefore machine frames are reduced, for example, which
increases the influence of the lamination stack on the machine’s mechanical behavior. The
aim is to identify a material model for the whole stack. This is dependent on the contact

1
 V. Luchscheider, M. Maidorn and K. Willner

behavior of the single sheets. Thus, the contact behavior in the normal and the tangential
direction must be identified.

2 CONTACT FORMULATION
The identification and simulation of the contact behavior is usually divided into the
normal and the tangential part. It has been shown by Willner [1] that there is almost
no difference between the simulation with an elastic halfspace model with uncoupled
directions and one with coupled directions. It is thus advisable to look at the normal and
the tangential direction on their own.
The whole contact behavior is caused by the roughness peaks of the surfaces. They
and the single real contacts of the peaks are responsible for both the progressive normal
behavior and the nonlinear tangential behavior. A few simplifications and assumptions
are common for identifying models. The surface roughness is replaced by asperities [2],
with certain radii. Heights of the asperities are specified by a normal distribution in the
case of technical surface without texture [3]. The number of asperities in contact is small
in relation to the contact area and there is thus no interference between the deformation
of the single contacts [4]. In addition, it is common to simplify the contact of two rough
surfaces by using a contact of one rough surface with a rigid ideal flat surface. To simulate
the same behavior, the characteristic parameters must be combined [5].

2.1 Normal behavior

The normal behavior of the stacks is progressive, elasto-plastic and viscous. The pro-
gression is caused by the sheet contacts - the closer the sheets, the higher the number
of single contacts. As every contact has got a small stiffness, the behavior of the stack
is progressive. The plastic behavior results from the fact that the loading of single very
high asperities is larger than the yield strength σY . And the viscosity is caused by the
contact’s behavior [6] and the existence of the coreplate varnish. The models to describe
the behavior of these three effects are presented and combined in the following.
The description of the elastic progressive behavior is based on the above mentioned con-
tact asperities and was developed by Bush, Gibson and Thomas [4], using the fractal
behavior of technical surfaces. This means that a technical surface without texture cor-
relates (for different scales) with itself. The autocorrelation function of the height data
is representative for the roughness and can be rewritten as structure function [4, 7]. It
is possible for technical surfaces to approximate the structure function through a formu-
lation of Berry and Blackwell [8]. They deduced it from the theory of fractals and the
theory to describe surfaces with sinus " waves. Thestructure function
4−2·Dp #
xk
S(xk ) = 2 σz · 1 − exp − (1)
xT
is characterized by the fractal dimension Dp , the transition length xT between the fractal
and the stationary region and the root mean square roughness σz [5]. The stationary

2
 V. Luchscheider, M. Maidorn and K. Willner

1
10
2σ2z

2D
p
4−
log S(x) 0
10

Measured data
xT
−1 Approximation
10
−2 0
10 10
log x

Figure 1: Example of a structure function of a coated sheet with measured data and an approximation
according to equation 1

part is a consequence of the existence of a maximal roughness height (see Figure 1). For
creating a contact model the fact of describing a surface with sinus waves of different
wave lengths and heights is used. These waves can be identified with the autocorrelation
function [7, 9] and the probability of a certain height with the joint probability function
can be calculated [4, 7]. Bush, Gibson and Thomas [4] combined this function with the
assumptions that the asperities are paraboloids with every possible aspect ratio and that
the single contacts follow the theory of Hertz [3]. This leads to an elastic dependence of
the contact pressure
"  2 #
E ∗ · σz2 1 g
pel (g) = exp − (2)
2π · xT · g 2 σz

of the gap g, which is known as the BGT model. The equivalent Hertzian modulus
1 1 − ν12 1 − ν22
= + (3)
E∗ E1 E2
is a combination of the elasticity moduli Ei and q the Poisson constants νi of the ma-
2 2
terials in contact. The rms roughness σz = σz,1 + σz,2 and the transition length
q
xT = x2T,1 + x2T,2 are also combinations [5].
Plastic behavior is based on the fact that a single asperity withstands maximally a mean
pressure of three times the yield strength σY [10]. To identify the plastic contact behavior,
the percentage of the asperities contact in ratio to the contact area is relevant. This ratio
is equivalent to the cumulative normal distribution of the heights, which is also known as
the Abbott-Firestone curve [11]. Consequently the plastic contact pressure
Z ∞ "  2 #
1 1 x−η
ppl (g) = 3 σY √ exp − dx (4)
ς 2π g 2 ς

3
 V. Luchscheider, M. Maidorn and K. Willner

3 0.4
Measured data Measured data
2.5 Simulation Simulation
0.3
pressure [MPa]

strain [%]
1.5 0.2

1
0.1
0.5

0 0
0 0.2 0.4 0.6 0.8 1 0 5 10 15
strain [%] time [h]

Figure 2: Measurement and simulation of the Figure 3: Measurement and simulation of a

first loading and unloading of a lamination stack stack’s creeping
(E ∗ = 115385 mm N
2 , σY = 317 mm N
2 / red:
(E ∗ = N
115385 mm 2 / red: charge A,
charge A, σz = 1.46 µm, xT = 0.18 mm, σz = 1.46 µm, xT = 0.18 mm,
mm3
ς = 1.45 µm and η = 0.97 µm / blue: charge B, A = 0.00072 N , m = 1 and n = 0.137 h1 /
σz = 1.93 µm, xT = 0.68 mm, ς = 4.73 µm blue: charge B, σz = 1.93 µm, xT = 0.68 mm,
3
and η = 1.90µm ) A = 0.00079 mm 1
N , m = 1 and n = 0.190 h )

is identified by Tsukizoe-Hisakado. ς is the standard deviation and η the expected value

of the normal distribution.
Effects of creeping and relaxation are often simulated with models, which contain springs
and frictional elements in different combinations [12]. The disadvantage of these models
is that they are only able to simulate creeping or relaxation. As both are caused by the
same reason, it is possible to calculate them with the formulation of the velocity of the
creeping strain
ε̇cr = A · σ(t)m · n · exp(−n · t) (5)
by Rust [13]. A, m and n are characteristic constants, t is the time since creeping or
relaxation started and σ is the actual stress.
To simulate the actual normal behavior, the elastic and the plastic gap have to be added
and the creeping strain has to be superposed. As can be seen in Figure 2 it is possible
to simulate the progressive behavior of loading and unloading a lamination stack. In
particular, the simulation of the elastic behavior after a first load is simulated quite well.
The discrepancies between measurement and simulation during the loading are caused by
effects such as the sheets waviness or the deformation of edges, which are not part of the
model. The simulation of the creeping is shown in Figure 3 and fits quite well. Knowing
the parameters of the creeping the relaxation is also identified.

2.2 Tangential behavior

The behavior in tangential direction is also based on the roughness and the single
contacts, even though it is often simulated just by analogous models out of springs and

4
 V. Luchscheider, M. Maidorn and K. Willner

friction elements. These may be Jenkin or Masing models [12] and the friction is usually
characterized by the Coulomb law. These models work well, but the theoretical back-
ground is minimal. For that reason a theory based on the contact behavior is used in this
paper [14, 15].
Tangential behavior is dependent on the deformation and the adhesion of the single con-
tacts. The tangential load FT is consequently proportional to the number of contacts.
Because this number is caused by the mean pressure the nominal contact area isn’t rele-
vant. This independence is the basis for the well-known contact theory of Amonton [16].
During the tangential loading, there is a small elastic deformation of the roughness peaks
and a small movement in the normal direction to loosen the single connections. For this
loosening the surface energy of the adhesion has to be overcome. The tangential load
FT = σA · A + µ · FN (6)

is thus dependent on the critical shear stress of the adhesion σA and the Amonton law
µ · FN [17]. Because the contact area A of technical surfaces is very small, the first term is
often neglected. For this case, the Amonton term remains, where the friction coefficient
µ is affected by the materials, the duration of the contact and the velocity. The static
friction coefficient is a bit larger than the dynamic one due to the small normal lift [16, 18].
Bowden and Tabor [10] found that the friction coefficient can be estimated by the shear
stiffness σY and the yield pressure pY :
FT σY · A τ
µ= = = = 0.2 . (7)
FN pY · A 5τ

For the formulation of the elasto-plastic behavior, the model of Olofsson et al. [14, 15]
is used. That model is based on the assumptions already mentioned in section 2 and is
based on the behavior of a single asperity like the model of Bush, Gibson and Tabor and
the one of Tsukizoe and Hisakado. The contact of single asperities is a Hertz contact for
the normal and tangential direction. With using the Hertzian shear modulus [14]
1 2 − ν1 2 − ν2
∗
= + (8)
G G1 G2
as a combination of the shear moduli Gi and the Poisson constants νi of the two partners,
the tangential stiffness of a single asperity is known. Slip occurs when the tangential force
reaches
FT,asp = µ · FN,asp , (9)

which equates to the Amonton law. As the tangential force FT,asp. and the normal force
FN,asp. are known from the Hertz theory, a maximal elastic asperity deformation is com-
putable. Integrating along the asperities’ heights and frictional situation, the total tan-
gential force

5
 V. Luchscheider, M. Maidorn and K. Willner

0.3 µ
0.25
0.2
k
FT/FN
t
0.15
0.1
0.05 Measured data
Simulation
0
0 0.5 1 1.5
−3
Displacement [mm] x 10

Figure 4: Measurement and simulation of the tangential sheet’s slipping (as example for sheets of
charge A)

"  52 #
√ ∗

8 5 4G δ
FT (δ) = µCAE ∗ R g 2 − g − (10)
15 µE ∗

is described as a function of the tangential displacement δ. C is the number of contact

asperities per unit area and R the radius of the asperities.
The microscopic contact stiffness kt and the friction coefficient µ can be detected quite
well in measurements. Tangential stiffness kt is the stiffness of all contacts, without any
gliding and therefore without an hysteretic effect. For that it is equivalent to the gradient
at the start of the Olofsson model
8 3√
kt = CG∗ g 2 R . (11)
3
The friction coefficient follows the maximal tangential force as shown in Amoton’s law.
Tangential movement in the micro-slip zone is not easily detectable in the case of single
sheets. Two of the main problems are the tangential fixation of the sheets and the effects
of the sheet’s edges slipping over each other. In Figure 4 measured data of stiffness and
friction coefficient are used to identify the micro slip behavior with the Olofsson model.
The friction coefficient of 0.28 is a bit larger than expected, based on the theory of Bowden
and Tabor (equation 7).

3 MULTISCALE FORMULATION
For the simulation of a whole lamination stack, a modeling of the single contacts is too
expensive in time and working memory. Because of that, a multiscale homogenization
is used to evolve a material model of the whole stack (Figure 5). That homogenization

6
 V. Luchscheider, M. Maidorn and K. Willner

computational model macroscopic model

representative volume element

Figure 5: Principle of homogenization of a lamination stack

is based on representative volume elements (RVE) used at every integration point [19].
They contain every mechanical behavior of relevance for the stack. That is, in the case
of the lamination stack, the sheet’s steel and the nonlinear normal and tangential contact
behavior. Another issue is that the RVEs have to fit together without any jumps in stress
or strain. To achieve this, a correct volume must be chosen and periodic boundary con-
ditions applied [19, 20]. In the stack’s case, the volume has the height of a single sheet
and a plane contact area in between. The periodic boundary conditions must ensure that
the fitting of the RVEs is possible during the loading with a deformation gradient. In the
3-D case, there are four nodes at the RVEs’ corners to which the deformation gradient is
applied. All other surface nodes are connected to these four nodes in a way that all edges
and surfaces are deformed like the ones on the opposite side. It is thus possible for every
deformation gradient to fit the RVEs together.
To calculate with homogenization, every integration point of the macroscopic model is
calculated with the RVE. Therefore, the Taylor hypothesis [19] postulates the mean mi-
croscopic deformation gradient
Z
1
Fm = F dV0 = F M (12)
V0 V0 m
to be the macroscopic deformation gradient F M at the integration point. That can be done
because the periodic boundary conditions lead to zero for the gradient’s term of fluctuation
∇→ −
ω of integrating it along the surface Γ0 of the RVE [20]. With the periodic deformation,
the stress field and the surface forces p at the RVE are calculated. As the nonlinear contact
is also part of the calculation, the resulting stresses are contact-dependent. To identify the
stress situation at the integration point, a resizing and homogenization to the macroscopic
scale is necessary. The mean first Piola-Kirchhoff stress tensor of the microscopic scale
Z
1
Pm = p ⊗ xm dA (13)
V0 Γ0

7
 V. Luchscheider, M. Maidorn and K. Willner

a) Normal displacement [mm] and a) Tangential displacement [mm] and

stress [MPa] of the RVE stress [MPa] of the RVE
3 0,3
Simulation FE
2.5 Simulation rve
pressure [MPa]

2 0,2

FT/FN
1.5

1 0,1
Simulation FE
0.5 Simulation rve
0
0 0 2 4 6 8
0 0.2 0.4 0.6 0.8 1 Displacement [mm] −4
strain [%] x 10

b) Loading and unloading of an FE-model (blue) b) Tangential loading of an FE-model (blue) and
and the RVE (red) the RVE (red)
Figure 6: Normal load (example for sheets Figure 7: Tangential load (example for sheets
charge A) charge A)

is dependent on the surface forces p and the displacement xm . The Hill-Mandel hypoth-
esis [20] equalizes the average power density of the RVE and the integration point
Z
1
P M : δF M = P : δF m dV . (14)
V0 V0 m
Combined with the Gaussian integration rule, the dependencies of the deformation gra-
dients (equation 12) and the mean first Piola-Kirchhoff stress tensor (equation 13), it can
be seen that [19]
Pm = PM . (15)

The identification of a stack’s material model is performed by applying different loading

situations of pressure and shear onto the RVE. Due to the non-linearity of the contact,
the created stiffness and damping matrices are also non-linear.
The RVE in use is depicted in Figure 6 a and Figure 7 a. At first it is loaded with a
normal displacement (Figure 6 a) left), what causes a unique normal stress (Figure 6 a)
right) and tangential displacements and stresses to keep constant volume. Afterward,
a small tangential displacement (Figure 7 a) left) is superposed. The consistent shear
stress, which has only small numerical variations, is depicted on the right of Figure 7 a).

8
 V. Luchscheider, M. Maidorn and K. Willner

Because periodic boundary conditions are used, the tangential displacement must be
smaller than the width of an FE element in the RVE. Otherwise the surface of the RVE
volume changes too much and new periodic boundary conditions must be introduced.
The actual calculation complies with this small displacement, because the micro-slip zone
is very small and it is sufficient to reach macro-slip. The congruence of the normal and
tangential behavior of the RVE (Figure 6 b and Figure 7 b) with the simulations of
Figure 2 and Figure 4 illustrates that all mechanically relevant characteristics are part
of the RVE. With that, it is feasible to calculate the behavior of a whole stack by using
multi-scale homogenization.

4 CONCLUSION
This paper presents a possible way to simulate the behavior of lamination stacks. The
basis is the simulation of the sheet’s contacts. They are motivated by the surface roughness
and the behavior of single peaks in contact. Both the number and behavior of the single
contacts are known and consequently models describing the behavior of two surfaces in
contact have been developed [4, 11, 15]. For a proper use of these models and to reduce
the working memory space, a multi-scale homogenization is utilized in this paper. With
this homogenization, it is possible to simulate the stack’s behavior for every reasonable
loading situation, therefore enabling a material model for the stack to be created. The
material model permits an easy and fast calculation of the mechanical behavior of the
whole stack and a better calculation of the whole electric motor.

REFERENCES
[1] Willner, K. Fully Coupled Frictional Contact Using Elastic Halfspace Theory. J.
Tribol. (2008) 130: 3.
[2] Archard, J. F. Elastic Deformation and the Laws of Friction. Proc. R. Soc. Lond. A
(1957) 243: 190-205.
[3] Greenwood, J. A. and Williamson, J. B. P. Contact of Nominally Flat Surfaces. Proc.
R. Soc. Lond. A (1966) 295: 300-319.
[4] Bush, A. W., Gibson, R. D. and Thomas, T. R. The elastic contact of a rough surface.
Wear (1975) 35: 87-111.
[5] Willner, K. Influence of Surface Parameters on the Elastoplastic Contact Behavior
of Fractal-Regular Surfaces. J. Tribol. (2008) 130: 2.
[6] Al-Bender, F., Lampaert, V. and Swevers, J. A novel generic model at asperity level
for dry friction force dynamics. Tribol. Lett. (2004) 16: 81-93.
[7] Nayak, P. R. Random Process Model of Rough Surfaces. J. of Lubrication Tech.
(1971) 398-407.

9
 V. Luchscheider, M. Maidorn and K. Willner

[8] Berry, M. V. and Blackwell, T. M. Diffractal Echoes. J. Phys. A: Math. Theor. (1981)
14: 3101-3110.
[9] Longuet-Higgins, M. S. The Statistical Analysis of a Random, Moving Surface. Phil.
Trans. R. Soc. A. (1957) 249: 321-387.
[10] Bowden, F. P. and Tabor, D. The Friction and Lubrication of Solids. Oxford Uni-
versity Press (1950).
[11] Tsukizoe, T. and Hisakado, T. On the Mechanism of Contact Between Metal Surfaces
- The Penetrating Depth and the Average Clearance. J. Basic. Eng. (1965) 87: 3,
666-674.
[12] Kolsch, H. and Ottl, D. Simulation des mechanischen Verhaltens von Bauteilen mit
statischer Hysterese. Forschung im Ingenieurwesen (1993) 59: 66-71.
[13] Rust, W. Nichtlineare Finite-Elemente-Berechungen: Kontakt, Geometrie, Material.
Vieweg+Teubner Verlag, Second Edition (2011).
[14] Olofsson, U. Cyclic micro-slip under unlubricated conditions. Tribol. Internat. (1995)
28: 207-217.
[15] Sellgren, U. and Olofsson, U. Application of a constitutive model for micro-slip in
finite element analysis. Comput. Methods Appl. Mech. Engrg. (1999) 170: 65-77.
[16] Besson, U. Historical Scientific Models and Theories as Resources for Learning and
Teaching: The Case of Friction. Sci & Educ (2012).
[17] Berman, A., Drummond, C. and Israelachvili, J. Amontons’ law at the molecular
level. Tribol. Lett. (1998) 4: 95-101.
[18] Gitis, N. V. and Volpe, L. Nature of static friction time dependence. J. Phys. D:
Appl. Phys. (1992) 25: 605-612.
[19] Miehe, C. Computational micro-to-macro transitions for discretized micro-structures
of heterogeneous materials at nite strains based on the minimization of averaged
incremental energy. Comput. Methods Appl. Mech. Engrg. (2003) 192: 559-591.
[20] Kouznetsova, V. G., Geers, M. G. D. and Brekelmans, W. A. M. Multi-scale second-
order computational homogenization of multi-phase materials: a nested finite element
solution strategy. Comput. Methods Appl. Mech. Engrg. (2004) 193: 5525-5550.

10