Torque Ripple Optimization of Switched Reluctance Motor Using Two-phase Model and

Optimization Search Techniques

Nitin Bhiwapurkar, Ted K.A. Brekken and Ned Mohan

Department of Electrical and Computer Engineering
University of Minnesota
Minneapolis, MN-55455, USA
Email: [email protected], [email protected], [email protected]

Abstract— Torque ripple optimization scheme for SRM drive, II. SWITCHED RELUCTANCE MOTOR MODEL
based on fibonacci and exhaustive search method and 2-phase
excitation model of SRM has been presented in this paper. The
2-phase excitation model of SRM used here takes into account ω
the nonlinear effects because of saturation and mutual flux
linkages caused by multi-phase excitation. In the proposed A' A
scheme the incoming and outgoing phase currents are D B
modulated to minimize the torque ripple. Because of the highly
nonlinear nature of torque, fibonacci and exhaustive search D' B'
methods are used for optimizing the torque ripple, by
C' C'
simulating the model multiple times. Modulation of phase
current is done by addition, subtraction and multiplication of
constants and variables. This can be very easily achieved in DSP C C
micro-controllers and FPGAs. Simulation and hardware results
B
are presented for proposed optimization scheme and compared D

with constant current operation. It has been shown by

simulations, that the proposed optimization scheme reduces the B' D'
A' A
torque ripple by more than 50% as compared to traditional
constant current scheme. The proposed optimization scheme Fig. 1: Cross-section of 4-phase 8/6 SRM
has been experimentally validated. The experimental results
show that the torque ripple is reduced by 49% to 45% for a
reference current of 8A and 10A as compared to the constant
A cross-sectional diagram of 4-phase 8/6 SRM used here is
current case. shown in fig. 1. The generalized SRM model presented in [1]
is used here. The model takes into account, saturation and
mutual coupling of phases. In normal operation of 8/6
I. INTRODUCTION 4-phase SRM more than one phase can be active. In case of
The Switched reluctance motor (SRM) drive are rugged multi-phase excitation, the flux linkage of any phase, as
and low cost, this makes it very attractive for industrial represented by equation 1, depends on the current in all the
applications. However SRMs also have certain problem phases and the rotor position. Torque is computed by taking
areas regarding high torque ripple and noise, which makes it the derivative of the co-energy associated with all the phases
unsuitable for many applications. w.r.t. the rotor position, as shown in equation 3. The
One of the methods for reducing the high torque ripple in co-energy calculation is given by equation 4.
SRMs is to shape the phase current in an optimal way.
Optimal phase current shaping has been studied extensively. λi = λi (ia , ib , ic , id ,θ ) (1)
Most of the studies take the single phase model of SRM and dλ
vi = ii Ri + i (2)
apply the various optimization techniques to it. Because of dθ
the nonlinearity involved in SRM operation, single phase ∂Wco
model is not accurate enough to describe its operation, also it T= = T (ia , ib , ic , id , θ ) (3)
∂θ const ( ia , ib , ic , id )
does not include mutual inductance effect because of
( ia , ib , ic , id )
multi-phase excitation. In this paper a generalized SRM
model for two-phase excitation [1] is used for optimization.
Wco = ∫
(0,0,0,0)
[λa , λb , λc , λd ] • [dia , dib , dic , did ] (4)
Torque ripple is optimized, by using optimal search
techniques on the data obtained by Finite Element Analysis
(FEA) for two phase operation of SRM. Current shaping Here λi , ii , vi and Ri are the flux linkage, current, input
developed for optimized torque ripple is very easy to voltage and resistance for phase i , respectively. Wco is the
implement in hardware. co-energy, θ is the rotor position and T is the torque
developed in motor.
 of phases is A-B-C-D-A, then phase A will be outgoing
Vi λi Current
ii Torque ω phase and phase B will be incoming phase. From fig. 4 and 5
+− ∫
Look-up Table Look-up Table Mechanical
ii (λa ,λb , λc ,λd ,θ ) Ti (ia , ib , ic , id ,θ ) System it can be seen that the torque region can be clearly divided
θ into three main regions. Region-I is the period, when mainly
outgoing phase is generating torque, Region-II is the period
Ri
when both incoming and outgoing phases are generating
Fig. 2: Multi-phase simulation block diagram torque and the Region-III is the period when mainly
incoming phase is generating torque. In Region-I the
Fig. 2 shows the simulation block diagram for such a incoming phase is just switched ON and as θon is near to
model. ii (λa , λb , λc , λd ,θ ) and Ti (ia , ib , ic , id , θ ) represents unaligned position so the torque contribution by this phase is
the lookup table for current and torque respectively. The data very less. In this region the torque is mainly produced by the
for these look-up tables is obtained by finite element analysis outgoing phase. In Region-II the incoming phase starts
of SRM. In normal two-phase operation of SRM both short producing significant torque. As both the phases (incoming
and long flux path are involved depending on the polarity of and outgoing) produce torque so the instantaneous torque is
excitation of each phase [1]. Here in the above model only higher than the average torque which causes a ripple in
short flux path is assumed and long flux path is neglected. torque. The outgoing phase is turned OFF in this region; this
Unaligned position of rotor w.r.t. a phase will be referenced decreases the contribution of outgoing phase to torque
as 0° position and aligned position will be referenced as 30°. resulting in torque ripple. In Region-III the contribution of
Further more it is assumed that the phase current is outgoing phase to torque is less as the current is less and as it
commutated such that it does not generate any negative is near to its aligned position. As it can been seen that major
torque. As the currents are periodic and active region (i.e. torque ripple is in the Region-II. The Region-II is more
positive torque region) of phase A and B corresponds to complicated as the incoming phase starts contributing
inactive (i.e. negative torque region) of phase C and D significant torque and then the outgoing phase is
respectively and vice-versa, we can effectively simulate commutated. The torque ripple for various constant current
4-phase SRM, by two adjacent phases with rotor period of 0° operations is tabulated in Table I. The torque profile for
to 15° for simultaneous excitation of 2-phases of SRM. These constant current operation of 14A, 12A and 10A is similar
assumptions reduce the total data requirement from FEA of whereas torque profile for constant current operation of 8A,
SRM. The modified equations governing the 2- phase SRM 6A and 4A is similar.
model is given by equations 5—8. The simulation block
Phase currents (A)

16
diagram for the two-phase model of SRM is shown in fig. 3. 14
12
10
8
λi = λi (ii , i j ,θ ) (5) 6
4
2 Incoming phase

dλ 0
Outgoing phase

vi = ii Ri + i (6) 0 2 4 6 8 10 12 14
dθ Rotor position θ (°)
12
∂Wco
= T (ii , i j , θ )
Torque (Nm)

T= (7) 10

∂θ const ( ii ,i j ) 8

( ii , i j ) 6

∫
Region-I Region-II Region-III
Wco = [λi , λ j ] • [dii , di j ] (8) 4
0 2 4 6 8 10 12 14
(0,0) Rotor position θ (°)
Fig. 4: Constant current operation for 14A
where, the subscript i and j denote the two conducting TABLE I
phases. TORQUE RIPPLE FOR CONSTANT CURRENT OPERATION
 λi   ii  Current Average Torque Torque Ripple % Torque
λ  i  (Nm) (Nm) Ripple
Current Torque
 Vi   j  j 14 A 7.804 2.335 29.919
∫
T
V  +− Look-up Table Look-up Table
12 A 6.492 1.719 26.48
 j i (λi , λ j ,θ ) T (ii , i j , θ )
10 A 5.156 1.474 28.588
θ 8A 3.868 1.159 29.964
 Ri 0 6A 2.496 0.99 39.67
0 R j 
 4A 1.199 0.548 45.74
Fig. 3: Two-phase simulation block diagram

III. CONSTANT CURRENT OPERATION OF SRM

Constant current operation of SRM is shown in fig. 4 and
fig. 5. Here θon = 0°, θoff = 23°, DC link voltage (Vdc) = 100V.
If A and B phases are conducting and sequence of excitation
 and decreased by adding and subtracting constant value from
Phase currents (A)
12
10
Incoming phase
Outgoing phase reference current. The modulated current in this region is
8
6
given by equations 12--14.
4
2
0 I mod = I ref − g ( I in ,θ ) (12)
0 2 4 6 8 10 12 14
Rotor position θ (°) g ( I in , θ ) = − g 2 ( I in ) for θ1 < θ < θ 2 (13)
6
g ( I in , θ ) = g3 ( I in ) for θ3 < θ < θ 4 (14)
Torque (Nm)

where, θ1 = 6°, θ3 = 7.75°, g2(Iin) and g3(Iin) are current

Region-I Region-II Region-III
2 dependent values and θ2, θ4 depends on the phase current
0 2 4 6 8 10 12 14
Rotor position θ (°) magnitude, and remains to be optimized. The rotor position
Fig. 5: Constant current operation for 8A θ1 and θ3 can be obtained from the torque profile data.
Though the incoming phase is modulated by giving a step
IV. OPTIMIZATION SCHEME increase or decrease, the actual current will have linear slope
SRMs are designed to operate in saturation and the mutual as shown in fig. 6, because of the phase inductance and finite
flux linkages because of multi-phase excitation are also input voltage. For phase current of 8A and below it can be
significant, because of this the torque generated by motor is a seen that the torque monotonously increases and then
nonlinear function of θon, θoff, phase currents I and rotor decreases. In this case a constant value is subtracted from the
position θ as given by equation 9. reference current. The modulated current in this region is
given by equation 15, where θ5 = 7° and θ6 = 11°, θ5 and θ6
T = f (θ on , θ off , I ,θ ) (9) can be determined from torque profile data.

Highly non-linear nature of SRM, makes it is difficult to

Phase current (A)
16
14
come up with equations describing its behavior, and apply 12
10
formal optimizations techniques for optimizing the torque 8
6
ripple. One of the schemes to reduce the torque ripple is to 4
2
Incoming Phase Constant current
Modulated Current
0
modulate the current where there is significant torque ripple. 0 2 4 6 8 10 12 14

In Region-I, to reduce the torque ripple the outgoing phase is 16

Rotor position θ (°)
Phase current (A)

modulated as it is a significant source of torque and the 14

12
Constant current
Modulated Current
torque because of incoming phase is very less (as it is near its 10
8
unaligned position), so modulation in incoming phase will 6
4 Outgoing Phase
have little effect on the torque profile. In this region the 2
0
outgoing phase is given a slight decreasing slope as shown in 0 2 4 6 8 10 12 14
Rotor position θ (°)
fig. 6(a) and (b), so that it compensates for the increase in 6(a)
torque because of incoming phase. Torque generated by the
16
incoming phase depends on the current magnitude and rotor
Phase current (A)

14 Incoming Phase Constant current

12 Modulated Current
position. So the torque ripple in this region is a function of 10
8
incoming phase current and rotor position as given by 6
4
equation 10. To compensate for this torque ripple the 2
0
outgoing phase current is modified according to equation 11, 0 2 4 6 8 10 12 14
Rotor position θ (°)
where g1(Iin) is current dependent gain and Iref is the reference 16
Phase current (A)

for constant current operation. 14 Outgoing Phase Constant current

12 Modulated Current
10
8

Tripple = f ( I in , I out ,θ )
6
(10) 4
2

I mod = I ref − g ( I in ,θ ) = I ref − g1 ( I in )θ (11) 0

0 2 4 6 8 10 12 14
Rotor position θ (°)
6(b)
In Region-II the incoming phase current is modulated as it Fig. 6: Phase current modulation: (a) For current > 8A and
starts producing significant torque and, as the outgoing phase (b) For current ≤ 8A
is commutated, so the outgoing current can not be modulated.
For phase currents of above 8A, the incoming phase current I mod = I ref − g 4 ( I in , θ ) = I ref − g 4 ( I in )θ
is boosted first to compensate for the dip in the torque and (15)
for θ5 < θ < θ 6
then decreased to compensate for the peaking of torque as
shown in fig. 6(a). The incoming phase current is increased
In Region-III the incoming phase is not modulated
because of the relatively flat torque profile in this region. given in equation 20-22.
For current > 8A, speed = 500 rpm and Vdc = 100 V:
A. Optimization search
Torque ripple is a non-linear function of phase current and g1 = a0 + a1 I ref ;
rotor position, so g1, g2, g3, g4, θ2 and θ4 are optimized by (20)
using optimization search techniques. For constant current a = [0.015, 0.01]
less than equal to 8A, the torque ripple in Region-I is θ 2 = c0 + c1 I ref + c2 I ref
2
;
optimized by parameter g1, which corresponds to modulation (21)
of outgoing phase current as shown in equation 11, and c = [5.962, 0.256, -0.0156]
torque ripple in Region-II is optimized by parameter g4, θ 4 = d 0 + d1 I ref + d 2 I ref
2
;
which corresponds to modulation of incoming phase current. (22)
d = [8.925, 0.5125, -0.0312]
As g1 and g4 are independent parameters, so they can be
optimized by either Fibonacci search or Exhaustive search
method. The advantage of fibonacci search method is that it V. SIMULATION RESULTS
requires less storage space as compared to exhaustive search.
In order to compare the results obtained by the two search

Phase currents (A)

16
14
12
methods, the parameters g1 and g4 are determined by both 10
8
methods. 6
4
Constant current
2 Incoming Phase Opt imized current
0

1) Fibonacci search: 0 2 4 6 8 10
Rotor position θ (° )
12 14

Phase currents (A)

The number of experiments N = 15 and the minimum 16
14 Constant current
12 Opt imized current
separation ε = 0.001 is used for the above fibonacci search 10
8

algorithm. The values of g1 and g4 thus obtained for various 6

4
2 Outgoing Phase
reference currents (i.e. 4A, 6A, and 8A) are fitted into 0
0 2 4 6 8 10 12 14
polynomial equation using regression analysis. The 6
Rotor position θ (° )

polynomial equations thus obtained is given in equation 16 Constant current

Torque (Nm)

Opt imized current

and 17. 4

For current ≤ 8A, speed = 500 rpm and Vdc = 100 V:

2
0 2 4 6 8 10 12 14
Rotor position θ (° )
g1 = a0 + a1 I ref + a I 2
2 ref ;
(16) Fig. 7: Fibonacci Search: current = 8A
a = [0.037, - 0.0055, 0.0015]
g 4 = b0 + b1 I ref + b2 I ref
2
Phase currents (A)

; 16
14
(17) 12

b = [-1.1116, 0.6186, -0.0408]

10
8
6
4
Constant c urrent
2 Incoming Phase Optimized current
0
2) Exhaustive search: 0 2 4 6 8 10 12 14
Rotor position θ (°)
The polynomial equations obtained for g1 and g4 for
Phase currents (A)

16
14 Constant c urrent

current ≤ 8A is given in equation 18 and 19. The interval in 12

10
Optimized current

8
this case is 0.01. 6
4

For current ≤ 8A, speed = 500 rpm and Vdc = 100 V: 2

0
Outgoing Phase
0 2 4 6 8 10 12 14
Rotor position θ (°)
4

g1 = a0 + a1 I ref + a2 I ref
2
; Constant c urrent
Torque (Nm)

Optimized current
(18)
a = [0.03, - 0.0025, 0.00125] 2

g 4 = b0 + b1 I ref + b2 I ref
2
;
(19) 0 2 4 6 8 10 12 14

b = [-1.0, 0.575, -0.0375] Rotor position θ (°)

Fig. 8: Exhaustive Search: current = 8A

3) Current greater than 8A:
Schematic of two phase model of SRM used for simulation
In this case the variables to be optimized are g1, g2, g3, θ2
is shown in fig. 3. Flux linkages and torque data of single
and θ4. In order to reduce the complexity involved because of
phase and two phase characterization of SRM obtained by
5 variables, g2 and g3 are fixed to some constant value, by
FEA, is used to get the Current Look-up table and Torque
simulating with some initial guesses. The variable to be Look-up table respectively. The torque ripple is optimized
optimized here are g1, θ2 and θ4. As more than one variable for SRM operation with speed=500 rpm and DC link voltage
needs to be searched so exhaustive search method is used to Vdc = 100 V. The phase current is modulated by applying Vdc
find the optimized values for 10, 12 and 14A. The and 0 V to phase winding (i.e. using soft switching). Phase
polynomial equation obtained after regression analysis are
current is commutated by applying – Vdc to phase.
Fig. 7 and fig. 8 shows the simulation results for reference TABLE III
HARDWARE RESULTS
current of 8A with fibonacci search and exhaustive search
Current Average Torque Torque Ripple % Torque Ripple
respectively. Fig. 9 shows simulation results for torque ripple (Nm) (Nm)
optimization with reference current of 14A. The torque ripple Constant Current
obtained using the optimized variable (optimized current 8A 3.147 2.632 83.63
10 A 4.476 4.360 97.40
profile) is tabulated in Table II.
Fibonacci Search
8A 2.923 1.336 45.69
Exhaustive Search
Phase currents (A) Phase currents (A)

16
14 8A 2.890 1.404 48.55
12
10 Current greater than 8A
8
6 10 A 4.136 2.400 58.02
4
Constant c urrent
2 Incoming Phase Optimized current
0
0 2 4 6 8 10 12 14
Rotor position θ (°)
16
14
12
Constant c urrent va
Optimized current
10
8
6
4
2 Outgoing Phase
0
0 2 4 6 8 10 12 14

12
Rotor position θ (°) torque
Constant c urrent
Torque (Nm)

Optimized current
10
ib ia
8

4
0 2 4 6 8 10 12 14
Rotor position θ (°)

Fig. 9: Current = 14A

TABLE II
TORQUE RIPPLE FOR OPTIMIZATION SCHEME Fig 10: Constant current operation of 8A
Current Average Torque Torque Ripple % Torque Ripple
(Nm) (Nm)
va
14 A 7.429 0.6766 9.108
12 A 6.159 0.545 8.851
10 A 4.83 0.635 13.145
Fibonacci Search
8A 3.572 0.4186 11.718 torque
6A 2.2647 0.3246 14.335 ib ia
4A 1.087 0.203 18.69
Exhaustive Search
8A 3.57 0.417 11.69
6A 2.265 0.3216 14.198
4A 1.087 0.208 19.138

From the simulation results tabulated in Table II it can be

seen that torque ripple reduction achieved by fibonacci and
exhaustive search methods is almost similar. In case of Fig 11: Optimized current operation of 8A with Fibonacci search
exhaustive search all the experimental results have to be
stored in memory, so that they can be compared at the end of using dSPACE and FPGA. Torque is measured using a
search. This takes lot of memory if the number of torque transducer. Fig. 10 and fig. 13 shows the results
experiments is more. In case of fibonacci search only two obtained with constant current operation at 8A and 10A
experimental results have to be stored. This reduces the respectively. Fig. 11 and fig. 12 shows the results obtained
memory requirement as compared to exhaustive search. So it with optimized current profile of 8A, for fibonacci search and
is preferable to use fibonacci search in order to minimize the exhaustive search respectively. Fig 14 shows the result for
hardware cost. optimized current profile of 10A. Trace 2 is the torque signal
and trace A is the average of trace 2 over 20 sweeps. Table III
VI. EXPERIMENTAL VALIDATION tabulates the experimental results. From Table III it can be
Optimization scheme developed is validated on 2KW 8/6 seen that the pk-pk torque ripple in optimized current case is
4-phase SRM. The DC link voltage Vdc is kept at 100V and almost half of that in constant current case. It can be seen that
the motor is operated around 500 rpm in open loop, by the proposed optimization scheme results in torque ripple
adjusting the load. Optimization scheme is implemented reduction of 49% with fibonacci search and 46.6% with
exhaustive search for 8A current. In case of 10A the the torque ripple is reduced by 49% with fibonacci search and
optimization scheme results in torque ripple reduction of 46.6% with exhaustive search for a reference current of 8A
45% as compared to traditional constant current case. and 45 % for a reference current of 10A as compared to the
constant current case.
The proposed scheme uses the 2-phase excitation model of
va
SRM, hence nonlinear effects because of saturation and
mutual flux linkages because of multi-phase excitation are
automatically taken into account. The data for the 2-phase
torque excitation model of SRM is obtained by FEA analysis. The
accuracy of this method for practical SRM, depends on how
ib ia close the FEA data, matches with the real SRM.
Modulation of phase current is done by addition,
subtraction and multiplication of constants and variables.
This can be very easily achieved in DSP micro-controllers
and FPGAs.
Torque ripple reduction obtained by fibonacci search and
exhaustive search is almost similar. But as fibonacci requires
Fig 12: Optimized current operation of 8A with Exhaustive search less memory storage space, so fibonacci search is preferable
over exhaustive search.
va
ACKNOWLEDGMENT
We gratefully acknowledge the resources of
Supercomputing Institute at University of Minnesota that we
torque
used to carry out the Finite Element Analysis in ANSYS.

ia id REFERENCES
[1] Amit K. Jain and Ned Mohan, “Modeling and experimental
characterization of SRMs for simultaneous two phase excitation”,
Industrial Electronics Society, 2003. IECON '03, Vol 2, pp 1027 –
1032.
[2] “Electronic Control of Switched Reluctance Machines” Edited by T J E
Miller, Newnes, 2001.
[3] David A. Wismer and R. Chattergy, “Introduction to Nonlinear
Fig 13: Constant current operation of 10A Optimization”, North-Holland, Newyork, 1978.
[4] ANSYS manual, http://www.ansys.com
[5] Patrick L. Chapman and Scott D. Sudhoff, “Design and Precise
Realization of Optimized Current Waveforms for an 8/6 Switched
va Reluctance Drive”, IEEE TRANSACTIONS ON POWER
ELECTRONICS, VOL. 17, NO. 1, pp 76-83, JANUARY 2002.
[6] J.M. Stephenson, A. Hughes and R. Mann, “Online torque-ripple
minimization in a switched reluctance motor over a wide speed range”,
torque IEE Proc.-Electr. Power. Appl. , Vol 149, No 4,pp 261-267, July 2002.
[7] Aleksander M. Stankovic, Gilead Tadmor, “On Torque Ripple
Reduction in Current-Fed Switched Reluctance Motors”, IEEE
ia id TRANSACTIONS ON Industrial ELECTRONICS, VOL. 46, NO. 1,
pp 177-183, FEBRUARY 1999.
[8] H.C. Lovatt, J.M. Stephenson, “Computer-optimised smooth-torque
current waveforms for switched-reluctance motors”, IEE Proc.-Electr.
Power. Appl. , Vol 144, No 5, pp 310-316, September 1997.
[9] J.M. Stephenson, A. Hughes and R. Mann, “Torque ripple
minimization in a switched reluctance motor by optimum harmonic
current injection”, IEE Proc.-Electr. Power. Appl. , Vol 148, No 4,pp
322-328, July 2001.
Fig 14: Optimized current operation of 10A [10] Ismail Agirman, Aleksandar M. Stankovic, Gilead Tadmor and Hanoch
Lev-Ari, “Adaptive Torque-Ripple Minimization in Switched
Reluctance Motors”, IEEE TRANSACTIONS ON Industrial
ELECTRONICS, VOL. 48, NO. 3, pp 664-672, June 2001.
VII. CONCLUSION [11] N.C.Sahoo, J.X.Xu and S.K.Panda, “Determination of current
From the simulation results presented it can be concluded, waveforms for torque ripple minimisation in switched reluctance
motors using iterative learning- an investigation”, IEE Proc.-Electr.
that the proposed optimization scheme reduces the torque Power. Appl. , Vol 146, No 4, pp 369-377, March 1999.
ripple by more than 50% as compared to traditional constant [12] H.C. Lovatt, J.M. Stephenson, “Computer-optimised current
current scheme. The proposed scheme has been waveforms for switched-reluctance motors” IEE Proc.-Electr. Power.
experimentally validated. The experimental results show that Appl. , Vol 141, No 2, pp 45-51, March 1994.