You are on page 1of 6

International Journal of Applied Electromagnetics and Mechanics 13 (2001/2002) 421–426 421

IOS Press

Topology optimization of switched


reluctance motor using mutual energy
method

Jin-kyu Byun∗ and Song-yop Hahn


School of Electrical Engineering #051, Seoul National University, Kwanak P.O. Box 34, Seoul 151-742,
Korea

Abstract. In this paper, the rotor part of the SRM (switched reluctance motor) is designed by topology optimization method.
The objective function is the desired system energy profile considering the rotation angle of the rotor. The topology optimization
method can design the optimal shape and topology of SRM rotor without any prior knowledge of the basic rotor layout. The
density method is used to establish the relationship between the design variables and material properties, and mutual energy
method is used to calculate the design sensitivity.

1. Introduction

Topology optimization is the optimal design method which uses the universal, gray-scale pixel-like
representation of the device [1,2]. In the topology optimization, basic topology and layout of the
device can be determined automatically during the design process, contrary to the conventional shape
optimization where the basic layout of the device is determined before optimization and where only the
boundary surfaces between materials are modified during the optimization. Topology optimization was
researched actively in the field of structural engineering in the last decade. On the other hand, there has
been relatively little work on the topology optimization in the electromagnetic system. In the mid-90’s,
Dyck et al. were the first to adopt the universal representation of the device for the optimal design of
the magnetic systems [1]. After the papers proposed by Dyck, few works have been reported on this
subject. However, in most of the papers, the applications have been limited to basic C-core design and
the comprehensive formulations and applications in practical cases were not covered.
In this paper, the topology optimization method is applied to the design of the SRM (switched reluctance
motor) considering the rotation angle of the rotor. The mutual energy method is used for design sensitivity
analysis (DSA), and density method is used to parametrize the design space and establish the relationship
between the design variables and material properties.


Corresponding author: Jin-kyu Byun, Fax: +82 2 878 1452; E-mail: aiolia@dreamwiz.com.

1383-5416/01/02/$8.00  2001/2002 – IOS Press. All rights reserved


422 J.-k. Byun and S.-y. Hahn / Topology optimization of switched reluctance motor

2. Density method

In the topology optimization method, the entire design region where the materials can exist is divided
into small cells. These cells can coincide with the finite element method (FEM) mesh. The material
properties are allowed to vary continuously in each cell. There are several techniques for associating
design variables with these material properties.
The homogenization method considers microscopic composite materials and calculates the ‘average’
material property using asymptotic expansion. The homogenization method gives a systematic and
reasonable way to interpret intermediate densities. However, the homogenization method requires
computational efforts for calculating the material constants for each finite element. It also creates a large
number of design variables for modeling the size and orientation of the microstructure.
The density method was first proposed in order to overcome these difficulties from the engineer’s point
of view. In the density method, the material density of each finite element is the design variable, and the
relationship between the material density and the material property is defined by a simple mathematical
equation.
To use the density method with the finite element method, the normalized density of material is assumed
to be identical over one finite element. The normalized density within each finite element controls the
electromagnetic properties of the material such as permeability, permittivity, etc. For example, the
relationship between the permeability and the normalized density can be written as:
µ = µ0 (1 + (µr − 1)pn ) (0  p  1, 2  n  4) (1)
where p is the normalized density, µ r is the relative permeability of the real material, and n is the exponent
which defines the relationship between material property and normalized density. The normalized density
takes a value between 0 and 1.
When n > 1, the intermediate density is penalized and as a result the normalized density is forced to
either 0 or 1. As n grows larger, the intermediate density is more heavily penalized, thus eliminating
‘gray’ materials more efficiently. But larger value of n can also limit the design space at the same time.
Usually, the value of n can be chosen between 2 and 4 considering this trade-off [3]. The density method
is widely used because it is easy to implement, has less design variables than homogenization method,
and the intermediate density can be eliminated efficiently by adjusting the value of n.

3. Design sensitivity analysis: The mutual energy method

In the topology optimization, the number of design variables grows dramatically as the number of
meshes increases to get reasonable resolution. Thus, quick and efficient calculation of the sensitivity
information considering these characteristics is essential. The method using the mutual energy concept
is presented in this section.
The two mutual energy terms in the magnetostatic system can be defined as follows:
     

L1 A2 =  
A2 · J1 dv, L2 A 1 =  1 · J2 dv
A (2)
V V

 1 is vector potential by source current density J1 , and A


where A  2 is associated with J2 .
The total energy functional in the magnetostatic system can be written in terms of mutual energy as,
1    1   1  
χ= a A 1 , A2 − L1 A2 − L2 A1 (3)
2 2 2
J.-k. Byun and S.-y. Hahn / Topology optimization of switched reluctance motor 423
      
where a A  1, A2 1 · ν ∇ × A
= V ∇×A  2 dv is an energy bilinear term and ν is the reluctivity
of the material.
For the vector potential distribution A  1 and A
 2 which satisfy Maxwell’s equation, the following
relations are obtained from the principle of virtual work,
   
a A  2 = L1 A
1, A  2 for all A
 2 in admissible space U (4)

   
a A  1 = L2 A
2, A  1 for all A
 1 in admissible space U (5)

so that the energy functional χ satisfies at equilibrium,


1    1   1   1  
χ= a A 1 , A2 − L 1 A 2 − L 2 A 1 = − L2 A1 (6)
2 2 2 2
Now let p be the design variable, and consider χ as a function of A  2 and p. The first variation of
1, A
χ at equilibrium yields,
   
  1 ∂ 2
A 1 ∂ 2
A
δχ A 1, A2, p = a A 1, δp + δA 2 − L1 δp + δA2
2 ∂p 2 ∂p
   
1 ∂ 1
A 1 ∂ 1
A
+ a A 2, δp + δA  1 − L2 δp + δA1 (7)
2 ∂p 2 ∂p
    
1  1 · ∂ν ∇ × A  2 dvδp
+ ∇×A
2 V ∂p
The first four terms in Eq. (7) can be cancelled by the principle of virtual work Eqs (4) and (5).
Therefore, the sensitivity of χ with respect to p is given by,
    
∂χ 1  1 · ∂ν ∇ × A  2 dv
= ∇×A (8)
∂p 2 V ∂p
 
From Eqs (6) and (8), the sensitivity of mutual energy term L 2 A 1 as to the design variable p is
obtained as follows,
 
∂L2 A 1     
=− ∇×A  1 · ∂ν ∇ × A  2 dv (9)
∂p V ∂p

4. Application to the switched reluctance motor design

Figure 1 shows the simple schematic picture of the SRM with 6 stator poles and stator coils. The
rotor region is made of magnetic material and marked by gray-colored central region in Fig. 1, and it
is designed by the topology optimization method. Note that the initial rotor has no salient pole shape
because the basic layout or topology of the design region is not needed in the topology optimization.
424 J.-k. Byun and S.-y. Hahn / Topology optimization of switched reluctance motor

θ
air

a stator poles

c’ b’
stator coil

Rotor design domain


with magnetic material
b c

a’

Fig. 1. The schematic picture of the initial SRM model.

In the design region, the finite elements take the shape of squares to regularize the design space. The
normalized density of material is assumed to be identical over one square to further regularize the design
space. The rotation angle of the rotor is 5 degrees, and the total number of rotations is 36 to represent
a total rotation of 180 degrees. Because of the symmetry, the rotation of 180 degrees is enough for
optimization. The desired system energy profile is given by a cosine function of the rotor rotation angle
θ when the current is applied at phase a and a’ in Fig. 1.
The objective function F in this problem can be defined as:
nr
1
F = [Wi − Wi0 ]2 (10)
2
i=1

where nr is the number of rotation angles at which the system energy is calculated, W i is the calculated
system energy at ith rotation angle, and W i0 is the desired system energy at ith rotation angle according
to the energy profile.
The sensitivity of this objective function can be written as:
 nr
dF dWi
= [Wi − Wi0 ] (j = 1, 2, . . . , np) (11)
dpj dpj
i=1

where np is the number of the design variables and p j is the j th design variable. Using the chain rule,
dWi /dpj term in Eq. (11) can be represented as:

dWi ∂Wi ∂Wi ∂ A i


= + (12)
dpj ∂pj  i ∂pj
∂A
where A i is the magnetic vector potential at ith rotor location. From the extremum condition for
energy functional, ∂Wi /∂ A  i term in Eq. (12) is zero for any potential distribution satisfying Maxwell’s
equations, so that Eq. (12) becomes:
dWi ∂Wi
= (13)
dpj ∂pj
J.-k. Byun and S.-y. Hahn / Topology optimization of switched reluctance motor 425

(a) 0 iterations (b) 5 iterations (c) 50 iterations

Fig. 2. The material distribution of SRM rotor.

This is the sensitivity of the self-energy of the system with respect to the design variable. Following
the formulations in Section 3 and noting that W i is the self-energy instead of the mutual energy, Eq. (13)
can be rewritten as:
    
dWi 1  i · ∂ν ∇ × A  i dv
= ∇×A (14)
dpj 2 V ∂pj
The ∂ν/∂pj term in Eq. (14) is given from density method as:
∂ν
= −nµ0 (µr − 1)pn−1
j /{µ0 (1 + (µr − 1)pnj )}2 (15)
∂pj
Substituting Eqs (14) and (15) into Eq. (11), the sensitivity of the objective function can be readily
obtained.
The calculated sensitivity is used in the optimization algorithm to search the optimum design variables.
The steepest descent method is adopted as an optimization algorithm because there is no constraint on
the volume of the magnetic material used for rotor in this case. The change in the design variable vector
∆[p] is defined as ∆[p] = lm̂ where the normalized search direction m̂ is given by,

−dF/d[p] 
m̂ = (16)
dF/d[p] i
and the step size l is searched by linearizing the system energy W i and given as:
nr
nr

 
lop = − ki (Wi − Wi0 ) ki2 ,
i=1 i=1
nr
(17)
dWi dF −1 dWi  dWj
ki =
· m̂ = − · [Wj − Wj0 ]
d[p] d[p] d[p] d[p]
j=1

The gray-scale pictures of the rotor at the various iteration numbers are shown in Fig. 2. It can be
seen from the results that the topology optimization method can ‘find’ the optimal shape and topology of
the salient-pole rotor without any prior knowledge of the basic rotor layout. However, the shape of the
final device is somewhat coarse because of the limited number of the design variables in the topology
optimization. The system energy as the function of rotation angle is shown in Fig. 3. It is clear that the
energy function converges to the desired cosine function from initial flat line.
426 J.-k. Byun and S.-y. Hahn / Topology optimization of switched reluctance motor

Energy [J/m]
4500
5 Iter.
4000

0 Iter.
3500

3000

2500

50 Iter.
2000

1500

1000 Objective

0 20 40 60 80 100 120 140 160 180

Rotor Rotation Angle [Degree]

Fig. 3. The system energy variation with iteration number.

5. Conclusion

The topology optimization was applied to the optimal design of SRM rotor in this paper. The optimal
topology and shape of the salient-pole rotor could be found according to the system energy profile
without prior knowledge of the basic rotor layout. However, because of the limited resolution, the
topology optimization should be used to find the basic topology and layout of the device as a pre-
processor module of the optimum design tool. And the conventional shape design method can be used
based on the result of the topology optimization to determine the detailed shape and size of the device.

References

[1] D.N. Dyck and D.A. Lowther, Automated design of magnetic devices by optimizing material distribution, IEEE Trans.
Magn. 32(3) (May 1996), 1188–1193.
[2] J. Yoo, N. Kikuchi and J.L. Volakis, Structural optimization in magnetic devices by the homogenization design method,
IEEE Trans. Magn. 36(3) (May 2000), 574–580.
[3] R.J. Yang, Multidiscipline topology optimization, Computer & Structures 63(6) (1997), 1205–1212.

You might also like