IEEE TRANSACTIONS ON MAGNETICS, VOL. 54, NO.

3, MARCH 2018 8102304

Methodology for Cage Shape Optimization of a Permanent

Magnet Synchronous Motor Under Line Start Conditions
Cezary Je˛dryczka 1, Łukasz Knypiński 1, Andrzej Demenko1, and Jan K. Sykulski 2, Fellow, IEEE
1 Institute of Electrical Engineering and Electronics, Poznan University of Technology, 60-965 Poznań, Poland
2 Electronics and Computer Science, University of Southampton, Southampton SO17 1BJ, U.K.

This paper proposes a methodology for shape optimization of the starting cage of a line start permanent magnet synchronous motor
motor with the aim to improve its synchronization performance. The parameters of the machine are established from a field-circuit
model, where the magnetic field is simulated using a finite element method (FEM). A strategy for evaluating machine parameters
exploiting parallel computing is proposed. To facilitate the use of FEM package, bespoke procedures have been developed and model
parameterization applied with the aid of the scripting language Visual Basic. A particle swarm algorithm has been adapted for
design optimization purposes. The proposed strategy has been verified via test simulations.
Index Terms— Electromagnetic field, finite-element analysis, line start permanent magnet synchronous motor, optimization
methods.

I. I NTRODUCTION an approach described in [3]. Following many tests and

practical experience with numerical analysis, it was decided
T HE global drive for energy efficiency inspired the search
for new types of permanent magnet synchronous motors
with the ability for line start. The development of such motors
that methods exploiting particle swarm optimization (PSO)
were most appropriate for the efficient design of the LSPMSM.
depends on the availability of reliable methods for analysis, II. M ODELING S TRATEGY AND T ECHNIQUES
design, and optimization. A particular challenge is to find a In the design of LSPMSMs, various constraints must be
cage shape allowing synchronization even under high inertia met and specific requirements satisfied regarding both steady-
load. With the aim to find a reliable methodology, in this paper state synchronous operation and asynchronous starting per-
an accurate electromagnetic model has been derived in combi- formance. Lumped parameter models are typically used for
nation with a computationally efficient algorithm for solving analysis [2]. Unfortunately, such models tend to be unreliable,
the resultant equations. The field-circuit model encompasses: especially in the context of the transient state during starting;
1) electromagnetic field equations taking into account the skin hence, their usefulness is very limited for design optimization
effect in the cage bars; 2) a description of the driving circuit purposes. This limitation provided motivation to search for
including winding connection; and 3) mechanical formulation, more accurate approaches based on field models. Originally
where the electromagnetic torque is expressed in terms of field a sequential algorithm was applied, where a steady state
quantities. was reached via a simulation of a transient start-up phase.
The parameters resulting in the best starting performance In this algorithm, first the start-up and then the steady-state
of the line start permanent magnet synchronous motor parameters were established. Calculations were performed for
motor (LSPMSM) are normally different to the optimal a given value of the load inertia JL , specified with the help of
design for synchronous operation (efficiency and power a coefficient ki defined by the ratio of the moment of inertia of
factor) [1], [2], thus it is advisable to conduct a multiobjective the load system to that of the rotor of the motor itself. To assess
search to account for all conflicting requirements. This type the starting parameters of the motor, the maximum value kimax
of optimization is very time consuming if full numerical is used, above which the motor fails to synchronize. In order
modeling is used; in this paper, efficient design strategies are to establish this maximum value, it is necessary to repeat
considered as well as ways of extracting motor parameters simulations with increasing values of ki . Finding the value
from field solutions. An algorithm is developed to establish— of kimax by these repetitive simulations of the motor start-
via parallel computation—such parameters, primarily the syn- up results in unacceptably long computing times, in particular
chronous torque and the synchronizing torque during starting. as the number of necessary function calls is often unknown in
The methodology first proposed in [2] has been adapted rely- advance. Such sequential strategy may therefore be inefficient.
ing on numerical field solutions using finite-element modeling In the search for a computationally efficient approach,
at an imposed speed. it was therefore proposed to implement a parallel methodology
Various optimization methods were considered, including where the transient start-up and steady-state simulations are
performed simultaneously as two independent computational
Manuscript received June 27, 2017; revised August 29, 2017 and
September 25, 2017; accepted October 13, 2017. Date of current version processes. The relevant start-up parameters are assessed using
February 21, 2018. Corresponding author: C. Je˛dryczka (e-mail: the approach described in [4], where the parameters are
cezary.jedryczka@put.poznan.pl). evaluated based on the value of the synchronizing torque
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. T80 defined as the asynchronous torque generated by the
Digital Object Identifier 10.1109/TMAG.2017.2764680 cage winding at a speed close to the final operating speed
0018-9464 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: UNIVERSITE DE LILLE. Downloaded on June 17,2021 at 12:16:05 UTC from IEEE Xplore. Restrictions apply.
8102304 IEEE TRANSACTIONS ON MAGNETICS, VOL. 54, NO. 3, MARCH 2018

Fig. 2. Proposed concept of design optimization of LSPMSM.

Fig. 1. Finite-element mesh. (a) For starting. (b) For steady-state operation. The inclusion of the end-ring effects is essential when using
2-D finite element method (FEM) models and the results
(here taken as 80% of the synchronous speed). The value of
should be verified either by full 3-D analysis or by measure-
this torque may be found using finite-element modeling at
ments on a real machine. The 2-D FEM models developed
a prescribed speed, i.e., neglecting the mechanical transient,
for the study presented in this paper had been originally
thus significantly reducing the computational effort required to
validated by measurements on a prototype of a motor with
establish the component of the objective function representing
round bars [5], and the method of estimating the end-ring
the synchronizing parameters of the machine.
lumped parameters was then extended to include the drop-
Two possible strategies were considered: 1) based on two
shaped bars.
discretization meshes for the two states (asynchronous and
synchronous) and 2) using identical meshes. Strategy 1) bears
III. O PTIMIZATION A LGORITHM
a resemblance to an adaptive mesh refinement approach. The
mesh of the stator is the same for both modes of operation, Despite the implementation of the efficient strategy for code
but the rotor area is discretized differently with a refined mesh parallelization, the optimization process remains computation-
in the air gap region of the slot pitch (see Fig. 1). ally time consuming due to the need for repetitive simulations
Comparisons were made between the calculated start-up with the continuously varied cage parameters; hence the need
parameters using the mesh optimized for the steady-state for an effective optimization algorithm. Many alternatives
simulations and values obtained from the mesh adapted so were considered, including genetic [3], particle swarm [6] and
as to better capture the transient performance during the start- ant colony algorithms [7], as well as the less popular brain
up period at 80% of the synchronous speed. The differences storm [8], bat-inspired [9] and gray wolf optimizers [10].
were noticeable, reaching a few percent of the synchronizing Following intensive search and testing, it was decided to
torque T80 . Moreover, it was observed that using an adapted rely on the well performing particle swarm approach, with
mesh allowed for a reduction of computing times up to about a modification of an added coefficient representing the swarm
15%, depending on the level of magnetic saturation and related center.
number of iterations in the Newton–Raphson algorithm. The PSO algorithm, inspired by the flocking and schooling
In the model dedicated to the steady-state operation [syn- patterns of birds and fish, was first introduced in 1995 [11].
chronous mode (SM)] at synchronous speed, the dynamics of In the PSO, the swarm consists of particles. In the optimization
the moving elements was considered assuming a relatively process, the particles move in the n-dimensional design space,
small value of the load moment of inertia (ki = 1) and where n refers to the number of design parameters. Each
the rotor speed at the instant of switching ON equal to the i th particle in the swarm is described by its position xi and
speed of the rotating field. This has allowed shortening of velocity vi representing the possible variant of the machine.
the simulated transient associated with the switching ON of Each particle remembers its best position in previous time
the motor. A criterion was used for two simulation cycles steps xiL and the position of the leader xG , which is the fittest
not to differ by more than 1% in terms of efficiency and the particle in the swarm. In the classical PSO method in the kth
power factor for the steady state to be assumed to have been time step, the position vector of the particle is derived from
reached. A similar strategy was applied to the simulation of the the following formula:
   
asynchronous state [asynchronous mode (ASM)] with imposed xik = xik−1 + wvik−1 + c1r1 xiL − xik−1 + c2r2 xG − xik−1
speed. For both types of simulation (SM and ASM), a 2-D
(1)
formulation was used in terms of magnetic vector potential
and time stepping with the same time step t. where w is the inertia factor, xik−1 is the position vector
It should be noted that in order to accomplish the desired in the previous step k − 1; c1 and c2 are the learning
parallelization—and achieve the automation of the execution coefficients, and r1 and r2 are the random numbers from the
of the relevant tasks—dedicated scripts had to be developed range (0, 1) [12].
associating the algorithms for field modeling at start-up and Since the original introduction of the classical PSO, this
steady state. Moreover, both algorithms have been linked to population-based stochastic optimization approach has been
the optimization routines explained later in this paper. The successfully applied to many engineering problems. The more
overall concept of the design optimization of the LSPMSM is advanced applications include modifications aiming at improv-
explained in Fig. 2. The calculations presented here have been ing the convergence and accuracy [13]–[15].
performed for a thermally steady-state condition, i.e., constant In this paper, a modified version has been utilized, adapted
properties of permanent magnets and electric conductivities. to particular requirements of the design of electromechanical

Authorized licensed use limited to: UNIVERSITE DE LILLE. Downloaded on June 17,2021 at 12:16:05 UTC from IEEE Xplore. Restrictions apply.
JE˛DRYCZKA et al.: METHODOLOGY FOR CAGE SHAPE OPTIMIZATION 8102304

devices. In the proposed algorithm, a term is added to (1)

which describes the center of gravity of the swarm
 
c3r3 xmc − xik−1 (2)
where c3 is the learning factor, r3 is the random number, and
xmc is the vector of coordinates of the gravity of the swarm.
The vector xmc is given by
N  N
i i
xmc = x f (x) f i (x) (3)
i=1 i=1

where f i (x)is the objective function and N is the number of

particles in the swarm.
The effectiveness of the modification of the PSO algorithm
was first highlighted in [15]. It was demonstrated that adding Fig. 3. Design variables describing the shape of the cage bars.
a component representing the center of the mass of the swarm
circular bar radius (see Fig. 3), and z 5 = k z the coefficient
helps to improve the convergence of the optimization process
introduced above. In the optimization algorithm, the design
of the LSPMSM, thus offers practical benefits, in particular
parameters z describe different quantities with different ranges;
the shortening of computational times.
in the numerical implementation, these have therefore been
normalized [10] and the vector of actual design parameters z
IV. C ASE S TUDY
replaced by a normalized vector x.
In order to verify the effectiveness of the proposed strategy, The design of permanent magnet machines usually involves
a particular LSPMSM has been considered whose stator comes multiobjective optimization with multiple parameters and sev-
from a classical 3.5 kW general purpose induction motor; eral often conflicting constraints. Here the multiobjective opti-
details of the initial design may be found in [2]. mization problem has been transformed to a single objective
The optimization task has been defined as a search for function, combining all objectives, defined as
such a shape of the starting cage rods which—with the      
given dimensions and distribution of the rotor permanent η(x) q1 cos ϕ(x) q2 T80 (x) q3
f (x) = (4)
magnets—would maximize the previously defined parameter η0 cos ϕ0 T0
ki , while maintaining high values of efficiency and power where qi (i = 1, 2, and 3) are the weighting coefficients,
factor. Magnetic field simulation was undertaken using com- and η0 , cos ϕ0 , and T0 denote the initial values of efficiency,
mercial software Maxwell assuming planar symmetry of the power factor, and the torque T80 , respectively, assumed in the
machine, supplemented by parameterized shape descriptors first iterative step of the optimization process. The weights can
of the starting cage, and linked—via appropriate scripting— be adjusted to suit particular requirements; for the purpose of
with specially developed optimization routines. The algorithm this paper, they have been assumed to be identical, i.e., qi = 1.
includes two 2-D FEM transient field simulations: 1) a model
to calculate efficiency η and power factor (PF) under the
rated load condition and 2) calculation of the synchronizing V. R ESULTS OF THE C AGE S HAPE O PTIMIZATION
torque T80 (a torque generated at 80% of the synchronous Calculations were performed for the following values of the
speed). As mentioned above, these simulations are carried out control parameters of the swarm: N = 40, c1 = 1.2, c1 = 1.4,
in parallel. and c3 = 1.2. The number of triangular elements in the
The shape of the starting-cage bars has been described 2-D FEM models for synchronous operation and asynchronous
by six dimensions defined in Fig. 3. It will be noted that operation of the machine were equal to about 35 000 and
the introduction of the radii r1 and r2 as design parameters 39 000, respectively. The time step t for both models
overspecifies the problem and unnecessarily complicates the was equal to time period/180. The average values of η, PF,
description of cage geometry; in practice, these radii are and T80 at the start of the iterations were: 92%, 0.94, and
selected by considering the cross-sectional area of the rotor 25 Nm, respectively. These values were determined by taking
tooth at r = const so that rmin < r ≤ rmin +h sk . Consequently, an average for several runs of the PSO initiation algorithm.
in the formulated models, a parameter k z has been introduced In order to validate the proposed strategy of evaluation of
describing the angular fill of the slot pitch by the rotor bar. the synchronizing torque in the optimization of the LSPMSM,
Through simple geometrical derivation, with the assumption of field simulations of the starting process were undertaken.
rotor teeth sides being parallel under rmin < r ≤ rmin +h sk , the The shapes before and after optimization were consid-
radii r1 and r2 may be easily calculated for a given value of k z . ered (Fig. 4). Simulations of the start-up transient were exe-
As a result of expressing the radii r1 and r2 in terms cuted for various, gradually increasing, values of the moment
of k z , the number of components of the vector z of the of inertia of the load that is for different values of the
design parameters has been reduced to 5: z 1 = so the coefficient ki , up to the maximum value kimax , above which
width of the slot opening, z 2 = h o the depth of the slot the motor fails to synchronize—i.e., rotor speed oscillations
opening, z 3 = h sk defining the depth of the slot, z 4 = r pr do not fade away. Fig. 5 shows the speed waveforms for the

Authorized licensed use limited to: UNIVERSITE DE LILLE. Downloaded on June 17,2021 at 12:16:05 UTC from IEEE Xplore. Restrictions apply.
8102304 IEEE TRANSACTIONS ON MAGNETICS, VOL. 54, NO. 3, MARCH 2018

to capture the specific properties of the field distributions in

both the synchronous and asynchronous states; this improves
the accuracy and by itself reduces the computation time by
around 15%.
The second innovation is an implementation of the modi-
fied PSO algorithm, complemented by dedicated scripting to
link the design process with commercial finite-element code,
incorporating a specific strategy for assessing the starting
performance. This has resulted in a unique integrated dedicated
system for design optimization of LSPMSM motors.
Fig. 4. Initial and optimized shapes of the cage of the LSPMSM. (a) Initial
shape. (b) Optimized shape.
The effectiveness of the proposed methodology and com-
putational efficiency of the developed design algorithms were
verified by considering a particular case. It was shown that
a significant performance improvement was possible; in the
case studied, a 75% increase of the moment of inertia of
the load was made possible during starting without sacrificing
the steady-state performance.
R EFERENCES
[1] C. Ogbuka, C. Nwosu, and M. Agu, “Dynamic and steady state perfor-
mance comparison of line-start permanent magnet synchronous motors
with interior and surface rotor magnets,” Arch. Electr. Eng., vol. 65,
no. 1, pp. 105–116, 2016.
[2] Ł. Knypiński, C. Jȩdryczka, and A. Demenko, “Influence of the shape
of squirrel-cage bars on the dimensions of permanent magnets in an
optimized line-start permanent magnet synchronous motor,” COMPEL-
Fig. 5. Speed waveforms during the start-up for the initial shape of the cage.
Int. J. Comput. Math. Elect. Electron. Eng., vol. 36, no. 1, pp. 298–308,
2017.
[3] Ł. Knypiński, L. Nowak, and A. Demenko, “Optimization of the
synchronous motor with hybrid permanent magnet excitation system,”
COMPEL-Int. J. Comput. Math. Elect. Electron. Eng., vol. 34, no. 2,
pp. 448–455, 2015.
[4] C. Jedryczka, R. Wojciechowski, and A. Demenko, “Influence of squirrel
cage geometry on the synchronisation of the line start permanent magnet
synchronous motor,” IET Sci. Meas. Techn., vol. 9, no. 2, pp. 197–203,
2015.
[5] M. Barański and C. Jȩdryczka, “Analysis of the line start permanent
magnet synchronous motor performance under different values of the
supply voltage,” in Proc. Int. Symp. Electromagn. Phenomena Nonlinear
Circuits (EPNC), Pilsen, Czech Republic, Jul. 2014, pp. 27–28.
[6] A. Belahcen, F. Martin, M. E.-H. Zaim, E. Dlala, and Z. Kolondzovski,
“Combined FE and particle swarm algorithm for optimization of high
speed PM synchronous machine,” COMPEL-Int. J. Comput. Math. Elect.
Fig. 6. Speed waveforms during the start-up for the optimized cage. Electron. Eng., vol. 34, no. 2, pp. 475–484, 2015.
[7] F. Moussouni, S. Brisset, and P. Brochet, “Comparison of two multi-
agent algorithms: ACO and PSO for the optimization of a brushless DC
initial parameters, whereas Fig. 6 contains the results for the wheel motor,” in Intelligent Computer Techniques in Applied Electro-
optimized machine. It can be seen that the optimized machine magnetics. Berlin, Germany: Springer-Verlag, 2008, pp. 3–10.
can synchronize even for the value of kimax = 48, whereas [8] D. Haibin, S. Li, and Y. Shi, “Predator–prey brain storm optimiza-
tion for DC brushless motor,” IEEE Trans. Magn., vol. 49, no. 10,
the original motor would only do this for the maximum value pp. 5336–5340, Oct. 2013.
of kimax = 27. [9] T. C. Bora, L. dos S. Coelho, and L. Lebensztajn, “Bat-inspired
optimization approach for the brushless DC wheel motor problem,” IEEE
Trans. Magn., vol. 48, no. 2, pp. 947–950, Feb. 2012.
VI. C ONCLUSION [10] Y. L. Karnavas, I. D. Chasiotis, and E. L. Peponakis, “Permanent magnet
In this paper, an effective strategy has been put forward for synchronous motor design using grey wolf optimizer algorithm,” Int.
J. Electr. Comput. Eng., vol. 40, no. 3, pp. 1353–1362, 2016.
the LSPMSM design optimization. The start-up and operating [11] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proc. Int.
parameters are established using the numerical field modeling Conf. Neutral Netw., Perth, WA, Australia, 1995, pp. 1942–1948.
techniques, while the optimization relies on the use of a [12] S. Kiranyaz, T. Ince, A. Yildirim, and M. Gabbouj, “Fractional particle
swarm optimization in multidimensional search space,” IEEE Trans.
specially modified PSO approach. Syst., Man, Cybern. B, Cybern., vol. 40, no. 2, pp. 298–319, Apr. 2010.
The proposed methodology creates two independent design [13] Z. D. Zaharis and T. V. Yioultsis, “A novel adaptive beamforming tech-
tracks, both exploiting full field models of the machine. Thus, nique applied on linear antenna arrays using adaptive mutated Boolean
PSO,” Progr. Electromagn. Res., vol. 117, pp. 165–179, Aug. 2011.
both the starting and steady-state states are considered, aiming [14] A.-Q. Mu, D.-X. Cao, and X.-H. Wang, “A modified particle swarm
at achieving the best compromise between optimal operating optimization algorithm,” Natural Sci., vol. 1, no. 2, pp. 151–155, 2009.
parameters (efficiency and power factor) while delivering the [15] Ł. Knypiński, L. Nowak, and C. Jȩdryczka, “Optimization of the rotor
geometry of the line-start permanent magnet synchronous motor by the
best starting performance (T80 ). The simulations are conducted use of particle swarm optimization,” COMPEL-Int. J. Comput. Math.
in parallel, which allows the finite-element mesh to be defined Elect. Electron. Eng., vol. 34, no. 3, pp. 882–892, 2015.

Authorized licensed use limited to: UNIVERSITE DE LILLE. Downloaded on June 17,2021 at 12:16:05 UTC from IEEE Xplore. Restrictions apply.