This paper presents a method to minimize torque ripple of a Vtype IPMSM using the PSO (Particle Swarm Optimization) method with FEM. The proposed algorithm includes one objective function and three design variables for a notch on the surface of a rotor. The simulation model of the Vtype IPMSM has 3phases, 8poles and 48 slots with 2 notches on the onepole rotor surface. The arcangle, length and width of the notch are optimized to minimize the torque ripple of the motor. The cogging torque of the model is reduced by 55.6% and the torque ripple is decreased by 15.5 %. Also, the efficiency of the motor is increased by 15.5 %.
1. Introduction
An Interior Permanent Magnet Synchronous Motor (IPMSM) has several advantages compared with a Surface Permanent Magnet Synchronous Motor (SPMSM). The IPMSM has a mechanically stable structure due to permanent magnets embedded in the rotor. It also has a slightly higher torque density per unit volume than other motors since the reluctance torque that results from an inductance difference between the daxis and qaxis is combined with the magnetic torque. In addition, it is highly efficient and has a wide range of velocity control for low magnetic field operations. Thus, the IPMSM appears to be an excellent candidate for vehicle propulsion systems.
Unfortunately, the IPMSM produces significantly large torque ripple due to magnetic air gaps of similar length and mechanical structure. Therefore, the reduction of torque ripple that may cause vibration and acoustic noises becomes an increasingly critical issue in IPMSM. Currently, there are many studies on reducing such torque ripples, as well as cogging torque
[1

3]
. There are also a number of suggestions made on how to achieve this such as employing skew, adjusting of the number of slots per pole, controlling of the magnetizing region of the permanent magnet, and regulating the input current
[4

5]
.
In this study, a 50 kW Vtype IPMSM is modeled to reduce torque ripple by introducing two notches in the onepole rotor surface as shown in
Fig. 2
. In the proposed algorithm, the position and size of the notch are optimized by an optimization algorithm, which uses a Latin Hypercube Sampling (LSH) strategy with a Response Surface Method (RSM) based on the multiobjective Pareto Optimization procedure. For obtaining an effective sampling point, PSO is also employed in the algorithm. To verify the proposed method, the simulation model is applied to and analyzed by the FE program
[6

7]
.
2. Simulation Model and Specifications
Table 1
describes the main specifications of the simulation model of the IPMSM. The capacity of the motor is 50 kW with a speed of 1200~1500 rpm, a rated voltage of 500 V, 3 phases, 8 poles and 48 slots. The current has a wide range according to its operational conditions. In the FE simulation, 200 A is used as the rated current.
Fig. 1
shows a crosssection of the simulation model (one pole) and the FE meshes. The total number of meshes is approximately 7,400. A doublelayer mesh is used in an air gap to calculate the force and torque of the model using a rotating air gap and external circuit.
Fig. 2
shows the optimization model of the notch, which has two design variables (i.e., notch angular length,
l_{n}
, and pitch angle between notches,
θ_{N}
) and one constraint. The notch angular length,
l_{n}
, is constraint between 0.5 and 3[°], the pitch angle between notches,
θ_{N}
, is constraint between 8 and 20 [°] while the notch depth,
d_{n}
, is kept at 0.5[mm].
Fig. 2(a)
shows an FEM model of the motor (one pole) while
Fig. 2(b)
shows the FE meshes.
Specifications of the 50 kW IPMSM model
Specifications of the 50 kW IPMSM model
Simulation model of a Vtype IPMSM (one pole) with FE meshes
Notch model of a rotor’s surface for a Vtype IPMSM
Fig. 3(a)
shows the flux density distribution of the optimized model around the PM and the notch.
Fig. 3(b)
shows the flux density profiles along the center line of the air gap with/without the notch on the rotor’s surface. The horizontal axis of
Fig. 3(b)
is ‘Rotor position’; that is one pole of PM(0 to 180 degree). In the case of the notch, the average flux density is slightly lower than for the case of a rounded surface, as expected.
(a) Flux density distribution around the notch; (b) Flux density profiles along the center line of the air gap (the average flux density is 0.703[T])
3. Optimization Algorithm with PSO
 3.1 Cogging torque calculation
Torque perturbations arise from the variations of magnetic energy associated to a change in airgap length, as the rotor rotates without load current. Since the energy change in a PM and iron is negligible compared to that of air, the magnetostatic energy is given as
[5]
,
Where
µ
_{0}
,
P
(
θ
) and ,
F_{m}
(
θ
,
a
) represent the permeability of air, the air gap permeance function, and the air gap MMF function, respectively. The cogging torque for a PM brushless motor can be expressed as
where
L_{s}
,
R_{s}
,
R_{m}
and
G_{nNL}
,
B_{nML}
denote the stack length, the stator bore radius, the magnet outer radius, and the corresponding Fourier coefficients of the relative air gap permeance and the flux density function, respectively
[5]
. Eq. (2) reveals that the cogging torque is first governed by
N_{L}
,
G_{nNL}
and
B_{nNL}
with the fundamental period of
2π /N_{L}
, where
N_{L}
is the least common multiple of number of stator and rotor poles.
 3.2. LHS and RMS
Latin Hypercube Sampling is a “space filling” experimental design strategy
[6]
. The RSM with a multiquadric radial basis function is widely used for global interpolations. It is also very efficient due to its smoothness and fitting ability when given a limited number of sampling points. As such,
where x is the design parameter vector, c
_{i}
is the coefficient corresponding to the
i
th sampling point x
_{i}
, g(x) is the multiquadric radial basis function, and
r
is so called “shape parameter” whose purpose is to control the curvature of the single basis function near the center point
[7

8]
.
 3.3. PSO (Particle Swarm Optimization Method)
Particle Swarm Optimization is a heuristic search technique that simulates the movements of a flock of birds which aim to find food. This method is advantageous in terms of relative simplicity, quick convergence requiring less computing, and effective locating of all local optimum points. These points in turn are motivated by the clustering PSO algorithm
[9

10]
.
Fig. 4
shows the procedure of the proposed optimization program with PSO. The PSO algorithm has five steps to determine an optimal sampling point: 1) identify main particle movements, 2) locate main particles that form a couple, 3) isolate a nonstop couple, 4) eliminate nearby couples, and 5) assess movement stopping criterion. The algorithm ends when the iteration number is reached at a preset figure, when an optimized position is found, or when the criteria of the objective function satisfy a predefined value.
Flow chart of PSO based optimization algorithm
4. Simulation and Results
Two design variables,
l_{n}
,
θ_{N}
, of the notch are constrained as shown in
Fig. 4
. Using the proposed optimization algorithm with PSO, 125 sampling points are selected through 5 iterations. The objective function to minimize the torque ripple is described as Eq. (4).
where τ
_{i+1}
is the torque ripple at (
i
+1)th iteration and τ
_{i}
is the average torque at
i
th iteration.
Fig. 5
shows three cases of the notch shape at a different length and arc angle;
Fig. 5(a)
θ_{N}
=20[°],
Fig. 5(b)
θ_{N}
=8[°] and
Fig. 5(c)
at
θ_{N}
=14.75[°]. As observed,
Fig. 5(c)
is the optimized model.
Table 2
describes the optimized notch length,
l_{n}
, the arc angle of
θ_{N}
and the torque ripple at each step of the optimization program. It should be noted that as the step increases, the torque ripple converges to 96.3 N.m.
Fig. 6
shows the average torque ripple of the optimized model as a function of the iteration steps. After the 5
^{th}
iteration step, the optimized arc angle,
θ_{N}
, is 14.74[°] and the notch angle,
l_{n}
, is 2.38[°].
Optimized variable and torque ripple at each step
Optimized variable and torque ripple at each step
Simulation model of the notch on the rotor’s surface
Average torque ripple of the optimized model as a function of the iteration steps
As shown in
Fig. 7
, the optimized cogging torque is reduced by 56.4% from 7.30 [N.m] to 3.19 [N.m]. Also, the torque ripple is reduced by 4.41 % as seen in
Fig. 8
. With the notch, the optimized
T_{max}
is 437.6[N. m],
T_{min}
is 340.8 [N.m], and
T_{avg}
is 389.1 [N.m]. However, without the notch (i.e., for a rounded surface),
T_{max}
is 448.9 [N.m],
T_{min}
is 335.1 [N.m], and
T_{avg}
is 389.2 [N.m].
Comparison of cogging torque between the round and the notched surface model(The final value of θ_{N} =14.74[°] and l_{n} =2.38 [°])
Comparison of torque profiles between the round and the notched surface model model(The final value of θ_{N} =14.74[°] and l_{n} =2.38 [°])
The torque ripple ratio can be calculated as in Eq. (5),
where
T_{rr}
,
T_{max}
,
T_{min}
,
T_{avg}
denote the torque ripple ratio, the maximal torque, the minimal torque, and the average torque, respectively.
5. Conclusions
In this study, an optimization algorithm is proposed to minimize the torque ripple of a Vtype IPMSM using a particle swarm optimization method with FEM. In the algorithm, an LHS strategy is incorporated with RSM based on a multiobjective Pareto Optimization procedure. PSO is also used to obtain an effective sampling point.
In order to verify the effectiveness of the proposed algorithm, the notch shape on the rotor’s surface of a 50 kW IPMSM is optimally formed with 2 design parameters and several constraints for minimizing torque ripple. The simulation model of the Vtype IPMSM has 3phases, 8poles and 48 slots with 2 notches on the onepole rotor surface. The arcangle, length and width of the notch are optimized to minimize the torque ripple of the motor. In so doing, the cogging torque of the optimized model is thus reduced by 56.4% and the torque ripple is decreased by 4.41 %. In addition, the efficiency of the motor is increased by 15.5 %.
Acknowledgements
This research was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Education (2012R1A1A4A01010937), and partially by the International Science and Business Belt Program through the Ministry of Science, ICT and Future Planning (2013K000468)
BIO
Yong Bae Kim He received B.S and M.S. degrees in electrical engineering from Hongik University, 2009 and 2011. He is a Ph.D student in Department of Electrical Engineering of Hongik University.
Ho Youn Kim He received B.S and M.S. degrees in electrical engineering from Hongik University, 2011 and 2013. He works at Keyang Electric Company as a researcher.
Pan Seok Shin He received B.S. in electrical engineering from Seoul National University, and received Ph.D in electric power engineering from Rensselaer Polytechnic Institute in 1989. He works at Hongik University as a professor since 1993.
Kisoumarsi A.
,
Moalle M.
,
Fashimi B.
2005
“Midigation of Torque Ripple in Interior Permanent Magnet Motors by Optical Shape Design”
IEEE Transactions on Magnetics
42
(11)
3706 
3711
Kang G.H.
,
Hong J.P.
,
Kim G.T.
,
Park J.W.
2000
“Improved parameter modeling of interior permanent magnet synchronous motor based on finite element analysis,”
IEEE Transactions on Magnetics
36
(4)
1867 
1870
DOI : 10.1109/20.877809
Thomas S.H.
,
Jahns T.M.
,
Soong W.L.
2007
“Torque Ripple Reduction in Interior Permanent Magnet Synchronous Machines Using the Principle of Mutual Harmonics Exclusion,”
Industry Applications Conference, 2007. 42nd IAS Annual Meeting. Conference Record of the 2007 IEEE
558 
565
Hendershot J. R.
,
Miller TJE
1994
“Design of Brushless PermanentMagnet Motors”
Magna Physics Publishing and Clarendon Press
Oxford
Hwang S. M.
,
Eom J. B.
,
Hwang G. B.
,
Jeong W. B.
,
Jung Y. H.
2000
“Cogging Torque and Acoustic Noise Reduction in Permanent Magnet Motor by Teeth Pairing”
IEEE Trans. on Magnetics
36
3144 
3146
DOI : 10.1109/20.908714
Zhang Yanli
,
Yoon H.S.
,
Koh C.S.
2007
“Study on a Robust Optimization Algorithm Using Latin Hypercube Sampling Experiment and Multiquadric Radial Basis Function”
Proceeding of KIEE EMECS Annual Spring Conference
162 
164
Shin P. S.
2007
“Shape Optimization of a Large Scale BLDC Motor Using an Adaptive RSM Utilizing Design Sensitivity Analysis,”
IEEE Trans. on Magnetics
43
(4)
1653 
1656
DOI : 10.1109/TMAG.2006.892260
Shin P.S.
,
Woo S.H.
,
Koh C.S.
2009
“An Optimal Design of Large Scale Permanent Magnet Pole Shape Using Adaptive Response Surface Method With Latin Hypercube Sampling Strategy”
Io00pEEE Transactions on Magnetics
45
(3)
1214 
1217
DOI : 10.1109/TMAG.2009.2012565
Kennedy James
,
Eberhart Russell
“Particle Swarm Optimization.”
1995
IEEE
Piscataway, NJ
proceedings of the IEEE International Conference on Neural Networks
4
1942 
1948
Kim Y. B.
,
Choi Hong Soon
,
Koh C. S.
,
Shin P. S.
2011
“A Back EMF Optimization Double Layered Large Scale BLDC Motor by Using Hybrid Optimization Method”
IEEE Transactions on Magnetics
45
(5)
998 
1001