This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics
An Improved Torque Sharing Function for Torque
Ripple Reduction in Switched Reluctance Machines
Haoding Li, Berker Bilgin, Senior Member, IEEE, Ali Emadi Fellow, IEEE
Abstract—An offline Torque Sharing Function (TSF) is in-
troduced in this paper for torque ripple reduction in switched
reluctance machines (SRM). This TSF uses static flux linkage
characteristics of the machine obtained from finite element anal-
ysis (FEA) or experiments that describes the machine dynamics
to determine optimal current profiles such that the torque ripple
reduction is achieved with minimal copper losses. Due to this
feature, the proposed TSF performs well across a wide speed
range. Additionally, the objective function of the proposed TSF
uses only one weight parameter, which facilitates the use of this
TSF. In this paper, an intuitive justification for the selection of
this weight parameter is given, and the performance of this TSF
is validated in simulation and experimentally on a 5.2 kW, four
phase SRM. To baseline its performance, the proposed TSF has
been compared to the Offline TSF in literature, which shows that
it has better current tracking performance at higher speeds due
to the inclusion of flux linkage characteristics. Finally, it has been
compared to conduction angle control at speeds above the base
speed to show that it can be a viable alternative for the control
of SRM even in an operation region normally not considered for
TSF.
Index Terms—Current shaping, optimization, switched reluc-
tance motor, torque ripple minimization, torque sharing function.
I. I NTRODUCTION
I N recent years, the switched reluctance machine (SRM)
has become increasingly more viable as a suitable can-
didate for various applications, due to its simple and low-
cost construction, and robustness [1], [2], [3]. However, one
main disadvantage of SRM is the inherently high torque ripple
due to its doubly-salient structure and its torque production
mechanism using reluctance torque [4], [5]. Torque ripple in
SRM is a source of noise and vibration, as well as a source of
disturbance on the controller for the motor drive. Additionally,
it induces fatigue on the shaft over time [6], [7]. In all, these
factors reduce the performance capabilities of the machine,
along with its popularity in the industry.
In literature, there has been a number of contributions on
means of reducing the commutation torque ripple in SRM,
either as part of the SRM design process as well as using
control methods. Torque ripple in SRM is attributed primarily
to the commutation between multiple phases [8]. Torque pro-
duction in the commutation region is generally more variable
than in periods where only one phase conducts. One strategy
for torque ripple control has been to control the firing angles
to minimize the commutation period [9]. In [10], a Genetic
Algorithm (GA) is employed to find the optimal conduction
angles that minimizes the commutation torque ripple.
In other cases, torque sharing functions (TSF) have been
popularized and widely researched as an alternative to mini-
0885-8993 (c) 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.
mizing torque ripple at low speeds, in which a constant torque
command is distributed symmetrically into each phase [11]–
[21]. Classically, the torque references can be defined using
analytical expressions [11], [17]. However, other methods of
defining the torque references have been proposed in literature
as well [12], [14], [15]. Additionally, an online TSF has
been proposed in [16] which uses a proportional-integral (PI)
controller that compensates the torque tracking error. In [17],
[18], trade-offs in copper losses and the effective speed range
of the TSF has been considered.
Recently, offline TSFs have been defined using optimization
techniques to shape the current reference with more freedom
than what is possible with analytical TSFs. In [20], a family
of TSF is created in which the p-norm of the phase current is
minimized. A more sophisticated objective function is formu-
lated in [21] to consider not only the phase current, but also
the rate of change of current, using a set of weight parameters.
In both cases, the rate of change of current in the reference
can be adjusted by varying their respective weight parameters
to meet more demanding phase voltage requirements at higher
speeds. In spite of this, the current references at higher speeds
may have a rate of change that is different from the actual
current dynamics, thus leading to tracking error and lower
torque ripple minimization performances.
In this paper, a different TSF has been implemented to
explicitly incorporate the flux linkage characteristics of the
machine in the optimization problem as a constraint. Unlike
the TSFs in [20] and [21], the proposed TSF has information
on the voltage dynamics of the phase when producing the
current reference to minimize the tracking error. When all of
the constraints in the proposed TSF are met simultaneously,
the tracking performance of the TSF at a given operating point
can be maximized while still minimizing the copper losses.
For this reason, the proposed TSF can be more effective at
the base speed of the machine compared to the Offline TSF
in [21], and that it can be applied even above the base speed.
Furthermore, the objective function of the proposed TSF relies
on a single weight parameter, which simplifies its analysis.
II. T ORQUE S HARING F UNCTION (TSF)
A torque sharing function distributes a desired constant
torque command into phase torque commands for each phase,
k. A TSF can be derived analytically as shown in (1), by
choosing the functions frise and ff all .
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics

 0 θ < θon

 cmd frise (θ)
T θon ≤ θ < θon + θov


Tcmd,k = Tcmd θon + θov ≤ θ < θof f
 cmd ff all (θ)

 T θof f ≤ θ < θof f + θov

0 θof f + θov ≤ θ

In (1), Tcmd is the total torque command, and Tcmd ,k is the
torque command for phase k. Other control parameters in these
torque sharing functions include θon , θof f , and θov , which
are respectively the turn-on angle, the turn-off angle, and the
overlap angle [17]. It can be seen that during commutation,
by choosing frise and ff all to sum to unity, a constant torque
reference will be produced. The performance of the TSF can
be quantified by two metrics, which are the copper losses
associated with the current profile used, as well as the speed
range within which the TSF can operate.
The copper losses can be defined by the RMS current, given
as an integral of the current over the conduction period:
s exceed rated values. This is given by the constraints in (6).
Z θof f
1 0 ≤ ik ≤ irated
IRM S = I 2 dθ (2)
θof f − θon θon
As the machine rotates, induced voltage builds up in the
coils that opposes the rate of change of the magnetic flux
in the machine. At low speed, the impact of the induced
voltage is small. However, the induced voltage affects current
control at higher speeds. Equation (3) can be derived from
the voltage equation of SRM under steady state conditions
when the voltage drop due to phase resistance is neglected. It
quantifies the relationship between speed, ω, and the rate of
change of the flux linkage, λ. V is the dc-link voltage.
V ∂λ
= (3)
ω ∂θ
From (3), it can be understood that the rate of change of flux
linkage is limited at high speeds. For this reason, conduction
period must naturally widen as speed goes up so that the
flux linkage and current can rise to the values stipulated by
the torque sharing function. One trade-off of this necessity is
that the RMS current must increase which results in higher
copper losses. Notably, higher speeds reduce the period of an
electrical cycle. When the current might not have the time
needed to reach the necessary value as a combination of
the high speed and high induced voltage, the TSF loses its
effectiveness.
Based on the fundamental principles regarding the dynamics
of current in an SRM described above, a TSF has been
proposed in the following section. For the discussion in the
rest of this paper, a definition for torque ripple is given in (4),
which is the RMS value of the instantaneous torque ripple.
s minimization can be achieved. Dynamically, the phase voltage
Z θof f
1 2
Tripple = (T − TAvg ) dθ (4)
θof f − θon θon
III. P ROPOSED TSF
In this section, a torque sharing function is developed
as the solution to a generalized optimization problem. This
optimization problem is solved offline repeatedly to calculate
0885-8993 (c) 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.
the current reference. The formulation takes advantage of
knowledge of the flux linkage characteristics of the machine
in creating the current reference. This improves the tracking
(1) capability of the torque sharing function.
The instantaneous phase currents are used as the optimiza-
tion objectives to create a current reference to minimize copper
losses. In light of this, the objective of the proposed TSF is
formulated in (5), where ik and ik−1 are the incoming and
outgoing phase currents, respectively. A weight factor, σ, is
used to penalize the outgoing phase current, as it contributes to
non-productive negative torque. While outgoing phase current
is undesirable, it is unavoidable due to the practical issue of
the time needed to demagnetize the phase.
J = i2k + σi2k−1 (5)
Then, there are three constraints which must be considered
for optimal tracking of phase current. First of all, thermal
limitations must be considered, as the phase currents cannot
(6)
0 ≤ ik−1 ≤ irated
Next, the torque sharing function must be established, in
which the commanded torque value, Tcmd , must be the net
phase torque contributions, as shown in (7). The phase torque
is a function of rotor position, θ, and phase current, i. The
phase torque function T can be characterized beforehand using
finite element analysis (FEA) and stored as a look-up table.
Tcmd = T (ik , θk ) + T (ik−1 , θk−1 ) (7)
Finally, the full extent of the dynamics of the SRM can
be summarized by the voltage equation for the machine. For
both the incoming phase and outgoing phase, the inequality
constraints in (8)-(9) must be met to ensure tracking of current
references
∆λk (i,θ)
≤ V − ik R
∆t (8)
∆λk (i,θ)
∆t ≥ −V + ik R
∆λk−1 (i,θ)
≤ V − ik−1 R
∆t (9)
∆λk−1 (i,θ)
∆t ≥ −V + ik−1 R
where V is the supply voltage, R is the phase resistance, i
is the phase current, λ is the phase flux linkage, and ∆t is
the time sampling interval. This condition ensures that the
phase voltage, which would result from the applied current,
is within the threshold of the dc-link voltage. In this way,
the rate of change of the current reference is constrained at
each time step. While this constraint is met simultaneously
as the other constraints, the current reference can be tracked
within the tolerances of the hysteresis band, and torque ripple
can be manipulated by the motor drive so that the rate of
change of current can be matched to that of the reference.
A classical optimization solver can be used to solve this
optimization problem to produce a reference current profile
with minimum copper loss. It can be seen from (5)-(9) that
the optimization problem relies on knowledge of the machine
flux linkage and torque characteristics. For an arbitrary SRM
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics

a)
60 a)
σ =1 σ =2 σ =5 30 σ =1 σ =2 σ =5
Current [A]

Current [A]
σ =10 σ =20 σ =50 σ =100 σ =10 σ =20 σ =50 σ =100

0 0
b)
2 b) 2
Torque [Nm]

Torque [Nm]
0 0
Unaligned 0
Aligned Unaligned Aligned
Rotor Position [ electrical] Rotor Position [0 electrical]

Fig. 1: Effect of the weight factor, σ, at 1000 RPM, 2 Nm Fig. 2: Effect of the weight factor, σ, at 6000 RPM, 2 Nm
reference on a) phase current b) phase torque reference reference on a) phase current b) phase torque reference

TABLE I: Effects of σ on the dynamic performance of TSF

design, these characteristics are readily available from FEA at 1000 RPM
or from experiments. Thus, the proposed algorithm can be Speed Tcmd σ Irms Ipeak Tavg Tripple
applied to any SRM design to achieve better torque ripple [RPM] [Nm] [] [A, RMS] [A] [Nm] [Nm, RMS]
performance. Across a wide speed range, σ in (5) can be 1000 2 1 5.85 11.69 1.98 0.10
1000 2 2 5.87 11.69 1.98 0.11
adjusted to add flexibility to the selection of reference current 1000 2 5 6.27 17.67 1.98 0.13
profiles. The impacts of adjusting the weight parameter σ are 1000 2 10 7.20 26.33 1.99 0.14
discussed in the following section. 1000 2 20 9.11 36.95 1.99 0.14
1000 2 50 11.93 49.01 1.99 0.14
1000 2 100 13.20 55.52 1.98 0.14
IV. W EIGHT PARAMETER , σ
In order to achieve minimum copper losses, phases should
be excited at a point in the electrical cycle where the maximum with a delayed excitation. This would result in lower peak
possible torque can be produced. For example, phase excitation current and lower copper losses. At high speeds, the phase
near the unaligned position requires higher current to satisfy must be advanced using a higher σ to minimize negative torque
the torque command than phase excitation past it. Thus, higher production. For example, in Fig. 2, the current references at
copper losses could occur when excitation is more advanced. 6000 RPM have a much lower rate of change during the
At the same time, residual magnetic energy stored in the phase commutation period as compared to that of low speed in Fig. 1.
after the aligned position can lead to the production of negative The slopes of the currents are constrained in the optimization
torque. This shows that when choosing the timing for phase problem by the phase voltage in (8)-(9) due to the inclusion
excitation at a given operating point, a careful balance between of the flux linkage characteristics. Notably, the tail currents at
maintaining minimal copper losses and reducing negative high speeds are much longer, which can lead to negative torque
torque should be achieved. At higher speeds, achieving this production if the phase current is not sufficiently advanced
balance becomes more difficult due to the shorter excitation using a high enough value of σ.
period and the influence of induced voltage. Table I and II show the impact of changing the weight
The weight parameter σ is used in the proposed TSF to parameter, σ, on the RMS current and torque ripple. Values in
add flexibility in the choice of current references at different Table I and II corresponds to the profiles shown in Fig. 1 and
operating points. By increasing the value of σ, a higher 2, respectively. From Table I, it can be seen that at low speed,
penalty is placed on using the outgoing phase current in the there is incentive to tune σ to be below 5 in order to minimize
optimization problem, which reduces torque production near the RMS current, which has little effect on the torque ripple. At
the aligned position. In Fig. 1 and Fig. 2, the effects of high speed, tuning σ higher than 5 improves the average torque
increasing σ on the resulting current profile are shown for value as it reduces the negative torque production caused by
both low and high speeds. Since higher σ penalizes current of the outgoing phase, as shown in Table II.
the outgoing phase, it causes the incoming phase current to be It should be noted that when the phase is significantly
increased in order to meet the torque constraint in (7). Overall, advanced for higher values of σ, the peak current also in-
increasing σ results in phase advancement. Phase advancement creases, as shown in Table I and II. This is because the torque
is necessary to ensure that the TSF can be effective across production near the unaligned position is low. Therefore, other
different speeds for different requirements. than the RMS current, additional criteria might be necessary
Based on the analysis shown above, some general conclu- in selecting the optimal current profiles based on the practical
sions can be made about the selection of σ. At low speeds, limitations of the motor and specific requirements of the motor
a low value of σ should be chosen to select a current profile drive.

0885-8993 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics

TABLE II: Effects of σ on the dynamic performance of TSF A. Parameter Identification and Performance Evaluation of
at 6000 RPM the Proposed TSF
Speed Tcmd σ Irms Ipeak Tavg Tripple For the proposed TSF, a multi-objective GA has been run at
[RPM] [Nm] [] [A, RMS] [A] [Nm] [Nm, RMS]
6000 2 1 6.47 11.69 1.58 0.24 various operating points to search for suitable values of σ in
6000 2 2 5.84 11.69 1.87 0.11 an objective way. As shown in Fig. 3, the GA forms an outer
6000 2 5 5.84 11.69 1.89 0.13 optimization loop that defines different values of σ. Using
6000 2 10 6.04 12.48 1.89 0.15
6000 2 20 6.60 17.68 1.89 0.16 the GA, σ at a critical operating point can be automatically
6000 2 50 7.67 22.60 1.90 0.16 defined.
6000 2 100 8.12 24.61 1.90 0.16
ωcmd
choose Tcmd start genetic
start operation
In order to tune the appropriate value of σ at a given algorithm
point
operating point, various tools ranging from a simple parametric σ candidates
sweep to a global optimization or search algorithm can be calculate
applied. Due to having only a single optimization parameter, current
a parametric sweep is viable for the proposed TSF. However, reference
while a parametric sweep is generally simpler and computa- no
Iref
tionally efficient, a more sophisticated search algorithm, such objective/
run dynamic
as the Genetic Algorithm (GA), can lead to more precise constraints of
simulation
values of σ. For different motor topologies and different power GA satisfied?
ratings, the range of σ might differ due to the magnitude of yes
pareto
the excitation current. However, the parameter identification save current σ front
σ selection
technique shown below can be applied to analyze any SRM end reference and output σ
algorithm
design. σ

Fig. 3: A flowchart summarizing how the weight parameter at

V. S IMULATION R ESULTS
The performance capabilities of the proposed TSF has been
validated in simulation on an 8-6, four phase 5.2 kW SRM
with a base speed of 6000 RPM. The weight parameters
used in each simulation has been determined by the GA to
ensure that they are chosen optimally. For all simulations, the
hysteresis band is set to 0.5 A, and current is sampled at 100
kHz.
To highlight the performance capabilities as well as the
novelties of the proposed TSF, its performance has been
compared to the Offline TSF in [21]. Unlike the proposed TSF,
the Offline TSF simultaneously minimizes the phase currents
to reduce copper losses and the rate of change of current to
improve tracking at high speeds. It balances the two sets of
objectives with an extra weight parameter. Because of this,
the copper losses cannot be at its minimum while the rate
of change of current is considered. In contrast, the proposed
TSF employs the constraints (8)-(9) to ensure good tracking
performance. Since the rate of change of current is not a part
of the objective function, the proposed TSF can minimize the
current in every case while still maintaining good tracking
performance.
Using the Offline TSF in [21] as a baseline, this section
shows the advantages of setting the rate of change of current
as a constraint rather than an objective in improving current
tracking accuracy. Additionally, it shows the benefit of simplic-
ity in having only a single weight parameter in the objective
function. Finally, in [21], a performance comparison has been
made against classical TSFs in literature, and so by comparing
the proposed TSF to the Offline TSF, this shows implicitly the
strengths of the proposed TSF to those in literature as well.
a given operating point is identified using genetic algorithm
In Fig. 3, it can be seen that the GA calculates a set of
viable values for σ at a given operating point. For each σ
defined, the inner optimization loop for the proposed TSF is
solved, in which reference current profiles are defined. For
each current profile, a dynamic simulation has been ran to
track the current reference, and the RMS current as well as
the RMS torque ripple are evaluated. Then, the values of σ
that yield current profiles with minimal values for RMS current
and torque ripple are returned by the GA. Sample results from
the GA outer optimization loop are shown in Fig. 4 and Fig.
5.
In Fig. 4, the Pareto front from this GA optimization
procedure working at 1000 RPM, with a torque command
of 3 Nm, is shown. Similarly, a Pareto front of the TSF
performance at 6000 RPM, 10 Nm has been shown in Fig.
5 to show the performance capabilities of the proposed TSF
at high speed under high load conditions. Additionally, a
similar analysis has been conducted using the Offline TSF
presented in [21] to find suitable weight parameters for it at
the same operating points. The resulting Pareto fronts from
this set of simulations are superimposed on Fig. 4 and Fig. 5,
respectively.
By comparing Fig. 4 to Fig. 5, it can be seen that the set
of values of σ found by the GA at high speed is significantly
higher than that of low speed, which is consistent with the
analysis provided in the previous sections. As the speed of
the machine increases, phase advancement is needed to reduce
negative torque production. In the formulation of the proposed
TSF, this is achieved by increasing the weight parameter σ.
In Fig. 4 and Fig. 5, the Pareto front obtained from the

0885-8993 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics

a)
20 Phase current Current reference

Voltage [V] Torque [Nm] Current [A]

0.18 σ =1.19
Proposed TSF 10
Phase B Phase A Phase D Phase C
Offline TSF [21]
RMS Torque Ripple [Nm]

q =0.52 0
b)
r = 1.80 σ =0.96 Rotor torque Phase torque
σ =0.92 3
q =0.32
r = 3.06
σ =0.88 0
q =0.19
q =0.61 σ =0.83 c)
r = 1.56 VDC Phase induced
r = 1.25 q =0.29
σ =0.75 voltage
r = 1.68 σ =0.72 0
q =0.22
-VDC
0.12 r = 1.62
7.16 7.26 7.36 Unaligned Aligned Unaligned
RMS Current [A] Phase A Phase Position [0 electrical]

Fig. 4: Performance of the proposed TSF and Offline TSF Fig. 6: Dynamic simulation using the proposed TSF at 1000
using Pareto Optimal parameters of σ at 1000 RPM and 3 RPM and Tcmd =3 Nm showing a) Phase current and reference
Nm torque command b) Torque and phase torque c) Induced voltage when σ=0.72

a)
1.8 20 Phase current Current reference

Voltage [V] Torque [Nm] Current [A]

q =0.98 Proposed TSF
r = 12.01 10
Offline TSF [21] Phase B Phase A Phase D Phase C
Tavg=8.53 Nm q =0.63
RMS Torque Ripple [Nm]

r = 12.92 0
b)
Tavg=8.79 Nm Rotor torque Phase torque
q =0.82 3
r = 39.95
0.95 Tavg=8.70 Nm
0
c) Phase induced
VDC
voltage
σ =10.80 σ =12.55 σ =11.75
0
Tavg=8.74 Nm Tavg=8.94 Nm Tavg=8.89 Nm -VDC
0.45

Unaligned Aligned Unaligned

14.4 15.4 Phase A Phase Position [0 electrical]
RMS Current [A]

Fig. 5: Performance of the proposed TSF and Offline TSF Fig. 7: Dynamic simulation using the Offline TSF [21] at 1000
using Pareto Optimal parameters of σ at 6000 RPM and 10 RPM and Tcmd =3 Nm showing a) Phase current and reference
Nm torque command b) Torque and phase torque c) Induced voltage when q=0.19,
R=1.56

GA optimization are shown, which demonstrates the trade-off

between torque ripple and RMS current for both the Offline
TSF and the proposed TSF. Using these Pareto fronts, the
Pareto-optimal performance capabilities of these TSFs can be
compared. For clarity, weight parameters used to obtain these
points have also been included for both families of TSFs.
From the sets of points shown here, candidates are selected
based on how close the average torque value is to the torque
command. When the average torque matches closely with
the torque command, this suggests that the current reference
created by the TSF is tracked closely. At each point in Fig.
4, the average torque values obtained are all within 2% of the
torque command. In Fig. 5, there is 11% to 14% variation
between the torque command and the average torque values
obtained in simulation. This is because at 6000 RPM, some
torque ripple is introduced due to difficulty in maintaining the
current within the hysteresis band.
The performance of the two families of TSF do not dis-
tinguish drastically at 1000 RPM, as shown in Fig. 4. This
is because at low speeds, current tracking is faster due to
the low induced voltage. However, it should be noted that
the proposed TSF can achieve comparable performance to the
Offline TSF in [21] with a single control parameter, while
there are two parameters required by the Offline TSF. This
single weight parameter captures the capability of creating
phase advancement, which allows the proposed TSF to be
robustly applied across a wide speed range. However, in Fig.
5, it can be seen that there is an improvement of 0.5 Nm rms
in torque ripple reduction in the proposed TSF at 6000 RPM.
The reason of this is presented in the following section.
B. Dynamic Simulation of the Proposed TSF
Dynamic simulation results at 1000 RPM and 3 Nm torque
command is shown in Fig. 6 for the proposed TSF, and in Fig.
7 for the Offline TSF. In both cases, there is good tracking
performance of the current references.
At 6000 RPM and for a torque command of 10 Nm, the
proposed TSF can reduce the torque ripple further by 0.5 Nm
RMS over the Offline TSF without increasing copper losses, as

0885-8993 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics

a) TABLE III: Performance of the proposed TSF at simulated

Voltage [V] Torque [Nm] Current [A] 40 Phase current Current reference
operating points
20 Phase B Phase A Phase D Phase C
Speed Tcmd σ Irms Tavg Tripple
0 [RPM] [Nm] [] [A, RMS] [Nm] [Nm, RMS]
b)
Rotor torque Phase torque 1000 3 0.72 7.35 3.00 0.13
10 6000 10 12.55 15.14 8.94 0.45

TABLE IV: Performance of the Offline TSF at simulated

0
c)
operating points
VDC Phase induced voltage
Speed Tcmd q R Irms Tavg Tripple
0 [RPM] [Nm] [] [] [A, RMS] [Nm] [Nm, RMS]
-VDC 1000 3 0.19 1.56 7.34 2.98 0.12
6000 10 0.82 39.95 15.22 8.70 0.95
Unaligned Aligned Unaligned
Phase A Phase Position [0 electrical]
closely to the actual current dynamics because the current
Fig. 8: Dynamic simulation using the proposed TSF at 6000 references is distributed more appropriately. This is what leads
RPM and Tcmd =10 Nm showing a) Phase current and ref- to the improved tracking performance and the reduction of
erence b) Torque and phase torque c) Induced voltage when 0.5 Nm RMS of torque ripple. This enhancement is due to
σ=12.55 the incorporation of the flux linkage characteristics in the
a)
optimization problem, by which the slope of the current profile
Phase current Current reference in the commutation region is calculated more suitably. Thus,
Voltage [V] Torque [Nm] Current [A]

50 Phase A
0
b)
Rotor torque
10
0
c)
VDC
0 To explore the limitations of the proposed TSF, simulations
-VDC
Unaligned
Phase A Phase Position [0 electrical]
Fig. 9: Dynamic simulation using the Offline TSF [21] at
6000 RPM and Tcmd =10 Nm showing a) Phase current and
reference b) Torque and phase torque c) Induced voltage when
q=0.82, R=39.95
shown in Fig. 5. Dynamic simulations at this operating point
are shown for both the proposed and the Offline TSF in Fig.
8 and Fig. 9 respectively.
In Fig. 9a, the current reference of the Offline TSF is tracked
by the current controller with significant tracking error in the
start of the phase excitation. This results in significant torque
ripple as shown in Fig. 9b. In contrast, the current tracking
error for the proposed TSF is lower, as shown in Fig. 8a,
leading to a much smoother torque profile as shown in Fig.
8b. In this particular case, it can be seen that torque ripple
reduction is achieved by the proposed TSF through reduction
of current tracking error. This example suggests that TSF can
be used up to the base speed, despite the high induced voltage
in the phase, as shown in Fig. 8c.
A comparison between Fig. 9a and Fig. 8a shows that the
profile of the actual phase current in both cases are very
similar, due to their common conduction period. However, the
current reference of the proposed TSF in Fig. 8a matches more
Phase D Phase C Phase B the proposed TSF can be used to reduce torque ripple at an
operating point where the Offline TSF may not be effective.
For reference, Table III shows the values of the RMS value of
Phase torque current and torque ripple, as well as the average torque values,
for the proposed TSF. Table IV shows the same for the Offline
TSF.
Phase induced voltage
C. High Speed Performance of the Proposed TSF
have been carried out at 8000 RPM. Above the base speed,
Aligned Unaligned current control is difficult to achieve due to the significantly
high induced voltage. Thus, high torque commands cannot be
realized. However, this section shows that the TSF can be used
for low torque commands. Fig. 10 shows the performance of
the proposed TSF for a torque command of 1 Nm. As shown
in Fig. 10c, the induced voltage is still lower than the dc-link
voltage. A value of σ=1.91 has been chosen by the GA for
this operating point.
In Fig. 10a, it can be seen that the TSF can still be
effectively used, despite the operating speed being above the
base speed. This is because the current command is low, which
keeps the magnitude of induced voltage below the dc-link
voltage. At the same time, to reach the peak current, the
required rate of change of flux linkage is also sufficiently low,
and so (3) can be satisfied. In general, these conditions should
be met for the TSF to be effective, although it is more difficult
to satisfy these conditions at higher speeds or under higher
loads.
The proposed TSF is compared to conduction angle control
at 8000 RPM. Conduction angle control is the conventional
control strategy for SRM, in which firing angles are the only
parameters specifying the excitation timing of a phase. A
hysteresis controller is applied to maintain current at some
constant commanded value. For this comparison, conduction
angles have been optimized using torque ripple and average
torque as objectives, resulting in θon =−20 and θof f =1210 .
The peak current is chosen at 7.79 A so that the average

0885-8993 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics

a) TABLE V: Performance comparison between different control

Voltage [V] Torque [Nm] Current [A] Phase current Current reference
10 Phase B Phase A Phase D Phase C strategies at 8000 RPM
Control Method Irms Tavg Tripple
0 [A, RMS] [Nm] [Nm, RMS]
b)
Rotor torque Phase torque Proposed TSF 4.18 0.91 0.06
Cond. angles 5.89 0.90 0.06
1

0
c)
VI. E XPERIMENTAL VALIDATION
VDC Phase induced voltage
The proposed TSF has been validated experimentally on the
0 same 8-6, four phase 5.2 kW SRM, connected to an induction
-VDC machine that is running in speed loop. A torque transducer is
Unaligned Aligned Unaligned connected between the shafts to measure the instantaneous
Phase A Phase Position [0 electrical] torque, as shown in Fig. 12. Optimized current references
from the proposed algorithm at critical operating points are
Fig. 10: Dynamic simulation using the proposed TSF at 8000 calculated offline, and they are stored in a look-up table on
RPM and Tcmd =1 Nm showing a) Phase current and reference the motor controller. A hysteresis current controller samples
b) Torque and phase torque c) Induced voltage when σ=1.91 the current at 100 kHz, and tracks the current on a 0.5 A
hysteresis band.
a)
Phase current Current reference
Voltage [V] Torque [Nm] Current [A]

10 Phase A Phase D Phase C Phase B Induction Induction Torque SRM

SRM
machine drive machine transducer Drive
0
b)
Rotor torque Phase torque
1

0
c) Phase induced voltage
VDC

0
-VDC

Unaligned Aligned
Phase A Phase Position [0 electrical]
Fig. 11: Dynamic simulation using conduction angle control at
8000 RPM, θon =−20 e, θof f =1210 e, and Iref =7.79A, showing
a) Phase current and reference b) Torque and phase torque c)
Induced voltage
torque between the two control techniques are comparable.
The hysteresis band and sampling frequency have been set to
the same values as those used previously. Fig. 11 shows the
dynamic simulation using conduction angle control.
A comparison between Fig. 11a to Fig. 10a shows that
the conduction period for the conduction angle control case
is significantly more advanced. This results in less negative
torque than using TSF. However, as shown in Table V, the
TSF uses lower RMS current compared to conduction angle
control at this operating point. This suggests that even at high
speeds, while the TSF may not be applicable for higher loads,
it can still be a viable alternative control strategy. Due to the
ability to shape the current with more flexibility than just
the conduction angles and peak current, the proposed TSF
can offers certain advantages compared to conventional SRM
control strategies.
Unaligned Fig. 12: Experimental setup showing the shaft connection
between the SRM and the load, as well as their respective
drives and the torque transducer
A. Instantaneous Torque Ripple at Low Speed
Due to the low sampling frequency of the torque transducer,
the induction machine is controlled at 20 RPM to observe
the instantaneous torque ripple. Fig. 13a-b shows the phase
current and instantaneous torque at 20 RPM, for a torque
command of 1 Nm, while Fig. 14a-b shows the phase current
and instantaneous torque at 20 RPM, for a torque command
of 4 Nm. In both figures, the torque measurements have been
directly obtained from the torque transducer. Furthermore,
the performance of TSF at these points have been compared
against that of conduction angle control, shown in Fig. 13c-d
and Fig. 14c-d. For comparison purposes, all axes are scaled
identically.
In both Fig. 13c-d as well as Fig. 14c-d, the conduction
angles are chosen so that the conduction period matches that
of TSF, and the peak current is selected so that the output
torque is the same as that of TSF. Using conduction angle
control, high commutation torque ripple as well as oscillations
arise due to the interaction between the SRM, load, and the
torque transducer. These characteristics are much lower when
the proposed TSF has been used. RMS torque ripple values
also indicate that torque ripple has been reduced through

0885-8993 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics

a) Phase current Current reference

10 Phase A Phase D Phase C Phase A Phase D Phase C a)
Phase k-1 Phase k Phase k+1
Torque [Nm] Current [A]
10

Current [A]
0
b)
3
Tripple= 0.05 Nm, RMS 0
2
b)
1
Tripple=0.23 Nm, RMS

Torque [Nm]
0
c) 1
10 Phase A Phase D Phase C Phase A Phase D Phase C
Current [A]

0
0 Time [s]
d)
3 Tripple= 0.46 Nm, RMS
Torque [Nm]

Fig. 15: Experimental waveforms of TSF a) phase current b)

2
torque at 3000 RPM for Tcmd =1 Nm and σ=0.86
1
0 Phase current Current reference
a)
Time [s] Phase k-1 Phase k Phase k+1
20

Current [A]
Fig. 13: Experimental a) phase current b) torque at 20 RPM
and 1 Nm torque command with σ=0.75 using the proposed
TSF, and c) phase current d) torque using conduction angle
control with θon =710 e, θof f =1720 e and Iref =12.0A (Phase 0
current of one phase not shown) b)
Tripple=0.52 Nm, RMS
Torque [Nm]

a) Phase A Phase D Phase C Phase A Phase D Phase C 4

20
Torque [Nm] Current [A]

0
0 Time [s]
b)
Fig. 16: Experimental waveforms of TSF a) phase current b)
8 Tripple= 0.13 Nm, RMS torque at 3000 RPM for Tcmd =4 Nm and σ=0.91
4
c)
20 Phase A Phase D Phase C Phase A Phase D Phase C
Current [A]

reference and the experimental phase current at 3000 RPM for

a torque command of 1 Nm and 4 Nm. Using these current
profiles, the electromagnetic torque has been interpolated using
0 characteristics of the experimental motor. Values of σ at these
d)
Tripple= 1.41 Nm, RMS
Torque [Nm]

points have been chosen as 0.86 and 0.91. The RMS current,
8
average torque, and torque ripple for each chase are shown in
4 Table VI.
Fig. 15a and Fig. 16a shows that good tracking has been
Time [s] achieved. Torque ripple values calculated in Table VI are
comparable to values obtained in simulation. Average torque
Fig. 14: Experimental a) phase current b) torque at 20 RPM
deviated from the torque command by 7% when the torque
and 4 Nm torque command with σ=0.75 using the proposed
command is 1 Nm, and by 5% when the torque command is
TSF, and c) phase current d) torque using conduction angle
4 Nm. This suggests that the model used to create the current
control with θon =470 e, θof f =1710 e and Iref =16.84A (Phase
references predicts well the performance of the actual machine.
current of one phase not shown)
Sources of error once again may arise due to the load dynamics
and mechanical tolerances.
current profiling. Torque ripple on Fig. 13b and Fig. 14b can TABLE VI: Performance of TSF at 3000 RPM under various
be attributed to the effects of switching, as well as to torque loads
ripple from the load machine.
Tcmd Irms (reference) Irms (measured) Tavg Tripple
[Nm] [A, RMS] [A, RMS] [Nm] [Nm, RMS]
B. Performance of the TSF Under Various Loads 1 4.31 4.26 0.93 0.23
4 8.55 8.13 3.77 0.52
Next, the speed of the system is set to 3000 RPM, and
the current tracking performance at different torque commands
has been observed. Fig. 15a and Fig. 16a show the current

0885-8993 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics

C. Base Speed TSF Validation TABLE VII: Performance of TSF at 6000 RPM under various
loads
In this set of experiments, the TSF has been tested at the
base speed of 6000 RPM, for torque command of 1 Nm and 4 Tcmd Irms (reference) Irms (measured) Tavg Tripple
[Nm] [A, RMS] [A, RMS] [Nm] [Nm, RMS]
Nm. At this speed, the induced voltage in the system is high, 1 4.32 4.08 0.87 0.19
which limits the rate of change of current. This is reflected 4 9.39 8.98 3.22 0.45
in the low switching frequency shown in Fig. 17 and Fig. 18.
Table VII shows the values of the RMS current, RMS torque
ripple, and average torque from the experiments at 6000 RPM. optimization problem in which a single weight parameter is
Despite the high speed, it can be seen that there is rea- introduced in the objective function. By adjusting the weight
sonable tracking at 6000 RPM, especially when the phase is parameter at different operating points, the effective speed
demagnetizing. However, there is tracking error at the begin- range of the TSF can be improved. A method of finding
ning of phase excitation when the torque command is 4Nm. the weight parameter is shown through simulation to demon-
This results in a lower average torque than the commanded strate the effect of the weight parameter and its ability to
torque, and also higher torque ripple as compared to when the introduce a phase advancement mechanism in this family of
torque command is 1 Nm. This is possibly due to the voltage TSFs. Simulation results for a 5.2 kW SRM demonstrates the
dynamics in the machine itself, as well as possible limitations performance of the proposed TSF at the base speed of the
in sampling when operating at high speed. machine. Simulations also show that there is potential for this
TSF to be used above the base speed, subject to the machine
Phase current Current reference dynamics and the ability for current control to occur. The
a)
Phase k-1 Phase k Phase k+1 proposed TSF is then validated experimentally, at low speed
10
Current [A]

to show the instantaneous torque ripple, at 3000 RPM to show

the performance under the base speed, and at 6000 RPM to
show the performance at the base speed.
0

b) ACKNOWLEDGMENT
Tripple=0.22 Nm, RMS
This research was undertaken in part, thanks to funding from
Torque [Nm]

1
the canada Excellence Research Chairs Program and Natural
Sciences and Engineering Research Council of Canada. The
0 authors gratefully acknowledge Powersys Solutions for their
Time [s] support with JMAG software in this research.

Fig. 17: Experimental waveforms of TSF a) phase current b) R EFERENCES

torque at 6000 RPM for Tcmd =1 Nm and σ=2.31
[1] B. Bilgin, A. Emadi, and M. Krishnamurthy, “Design considerations
for switched reluctance machines with a higher number of rotor poles,”
IEEE Trans. Ind. Electron., vol. 59, no. 10, pp. 3745–3756, Oct. 2012.
Phase current Current reference [2] ——, “Comprehensive evaluation of the dynamic performance of a 6/10
a) SRM for traction application in PHEVs,” IEEE Trans. Ind. Electron.,
Phase k-1 Phase k Phase k+1
vol. 60, no. 7, pp. 2564–2575, Jul. 2013.
20
Current [A]

[3] P. C. Desai, M. Krishnamurthy, N. Schofield, and A. Emadi, “Novel

switched reluctance machine configuration with higher number of rotor
poles than stator poles: concept to implementation,” IEEE Trans. Ind.
0 Electron., vol. 57, no. 2, pp. 649–659, Feb. 2010.
[4] B. Bilgin and A. Emadi, “Electric motors in electrified transportation:
b) A step toward achieving a sustainable and highly efficient transportation
Tripple=0.45 Nm, RMS system,” IEEE Power Electronics Magazine, vol. 1, no. 2, pp. 10–17,
Jun. 2014.
Torque [Nm]

4 [5] M. Krishnamurthy, C. S. Edrington, A. Emadi, P. Asadi, M. Ehsani,

0
Time [s]
Fig. 18: Experimental waveforms of TSF a) phase current b)
torque at 6000 RPM for Tcmd =4 Nm and σ=19.96
VII. C ONCLUSION
In this paper, an offline TSF is proposed that uses the
flux linkage characteristics of the machine to create current
reference profiles that can achieve minimal torque ripple while
maintaining low copper losses. The TSF is formulated as an
and B. Fahimi, “Making the case for applications of switched reluctance
motor technology in automotive products,” IEEE Trans. Power Electron.,
vol. 21, no. 3, pp. 659–675, May 2006.
[6] S. M. Castano, B. Bilgin, E. Fairall, and A. Emadi, “Acoustic noise
analysis of a high-speed high-power switched reluctance machine: Frame
effects,” IEEE Trans. Energy Convers., vol. 31, no. 1, pp. 69–77, Mar.
2016.
[7] F. C. Lin and S. M. Yang, “An approach to producing controlled
radial force in a switched reluctance motor,” IEEE Trans. Ind. Electron.,
vol. 54, no. 4, pp. 2137–2146, Aug. 2007.
[8] J. Lin, N. Schofield, and A. Emadi, “External-rotor 6-10 switched reluc-
tance motor for an electric bicycle,” IEEE Trans. Transport. Electrific.,
vol. 1, no. 4, pp. 348–356, Dec. 2015.
[9] H. Li, E. Fairall, B. Bilgin, and A. Emadi, “Performance evaluation of a
high-speed high-power switched reluctance motor drive,” in 2015 IEEE
Applied Power Electronics Conference and Exposition, Charlotte, NC,
2015, pp. 1337–1342.

0885-8993 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2835773, IEEE
Transactions on Power Electronics

[10] J. W. Jiang, B. Bilgin, and A. Emadi, “Three-phase 24 / 16 switched
reluctance machine for a hybrid electric powertrain,” IEEE Trans.
Transport. Electrific., vol. 3, no. 1, pp. 76–85, Mar. 2017.
[11] M. Dowlatshahi, S. M. S. Nejad, and J. W. Ahn, “Torque ripple min-
imization of switched reluctance motor using modified torque sharing
function,” in 2013 21st Iranian Conference on Electrical Engineering,
ICEE 2013, Mashhad, Iran, 2013, pp. 1–6.
[12] D. H. Lee, J. Liang, Z. G. Lee, and J. W. Ahn, “A simple nonlinear
logical torque sharing function for low-torque ripple SR drive,” IEEE
Trans. Ind. Electron., vol. 56, no. 8, pp. 3021–3028, Aug. 2009.
[13] I. Husain and M. Ehsani, “Torque ripple minimization in switched
reluctance motor drives by PWM current control,” IEEE Trans. Power
Electron., vol. 11, no. 1, pp. 83–88, Jan. 1996.
[14] S. Sahoo, S. Panda, and J. Xu, “Indirect torque control of switched
reluctance motors using iterative learning control,” IEEE Trans. Power
Electron., vol. 20, no. 1, pp. 200–208, Jan. 2005.
[15] R. Mikail, I. Husain, Y. Sozer, M. S. Islam, and T. Sebastian, “Torque-
ripple minimization of switched reluctance machines through current
profiling,” IEEE Trans. Ind. Appl., vol. 49, no. 3, pp. 1258–1267, May
2013.
[16] J. Ye, B. Bilgin, and A. Emadi, “An extended-speed low-ripple torque
control of switched reluctance motor drives,” IEEE Trans. Power Elec-
tron., vol. 30, no. 3, pp. 1457–1470, Mar. 2015.
[17] X. D. Xue, K. W. Cheng, and S. L. Ho, “Optimization and evaluation
of torque-sharing functions for torque ripple minimization in switched
reluctance motor drives,” IEEE Trans. Power Electron., vol. 24, no. 9,
pp. 2076–2090, Sep. 2009.
[18] C. Choi, S. Kim, Y. Kim, and K. Park, “A new torque control method
of a switched reluctance motor using a torque-sharing function,” IEEE
Trans. Magn., vol. 38, no. 5, pp. 3288–3290, Sep. 2002.
[19] T. Kojima and R. W. De Doncker, “Optimal torque sharing in direct
instantaneous torque control of switched reluctance motors,” in 2015
IEEE Energy Conversion Congress and Exposition, no. 1, Montreal,
Canada, 2015, pp. 327–333.
[20] V. P. Vujičić, “Minimization of torque ripple in switched reluctance
motor drives,” IEEE Trans. Power Electron., vol. 27, no. 1, pp. 388–399,
Jan. 2012.
[21] J. Ye, B. Bilgin, and A. Emadi, “An offline torque sharing function
for torque ripple reduction in switched reluctance motor drives,” IEEE
Trans. Energy Convers., vol. 30, no. 2, pp. 726–735, Jun. 2015.
Ali Emadi (IEEE S’98-M’00-SM’03-F’13) received
the B.S. and M.S. degrees in electrical engineering
with highest distinction from Sharif University of
Technology, Tehran, Iran, in 1995 and 1997, respec-
tively, and the Ph.D. degree in electrical engineer-
ing from Texas A&M University, College Station,
TX, USA, in 2000. He is the Canada Excellence
Research Chair in Hybrid Powertrain at McMaster
University in Hamilton, Ontario, Canada. Before
joining McMaster University, Dr. Emadi was the
Harris Perlstein Endowed Chair Professor of Engi-
neering and Director of the Electric Power and Power Electronics Center and
Grainger Laboratories at Illinois Institute of Technology in Chicago, Illinois,
USA, where he established research and teaching facilities as well as courses
in power electronics, motor drives, and vehicular power systems. He was the
Founder, Chairman, and President of Hybrid Electric Vehicle Technologies,
Inc. (HEVT) a university spin-off company of Illinois Tech. Dr. Emadi has
been the recipient of numerous awards and recognitions. He was the advisor
for the Formula Hybrid Teams at Illinois Tech and McMaster University,
which won the GM Best Engineered Hybrid System Award at the 2010,
2013, and 2015 competitions. He is the principal author/coauthor of over 400
journal and conference papers as well as several books including Vehicular
Electric Power Systems (2003), Energy Efficient Electric Motors (2004),
Uninterruptible Power Supplies and Active Filters (2004), Modern Electric,
Hybrid Electric, and Fuel Cell Vehicles (2nd ed, 2009), and Integrated Power
Electronic Converters and Digital Control (2009). He is also the editor of
the Handbook of Automotive Power Electronics and Motor Drives (2005)
and Advanced Electric Drive Vehicles (2014). Dr. Emadi was the Inaugural
General Chair of the 2012 IEEE Transportation Electrification Conference and
Expo (ITEC) and has chaired several IEEE and SAE conferences in the areas
of vehicle power and propulsion. He is the founding Editor-in-Chief of the
IEEE Transactions on Transportation Electrification.

Haoding Li received the B.Eng.Mgt. and M.A.Sc.

degrees from McMaster University, Hamilton,
Canada in electrical engineering in 2015 and 2017,
respectively. Since 2013, he has been working with
the research group at the McMaster Institute for
Automotive Research and Technology on projects
involving electric machines and drives. His current
research interests are in the development of control
systems for advanced motor drives.

Berker Bilgin (IEEE S’09-M’12-SM’16) received

the Ph.D. degree in electrical engineering from Illi-
nois Institute of Technology, Chicago, IL, USA.
He is currently working toward the MBA degree
at the DeGroote School of Business, McMaster
University. He is the Research Program Manager
in Canada Excellence Research Chair in Hybrid
Powertrain Program at the McMaster Institute for
Automotive Research and Technology, McMaster
University, Hamilton, ON, Canada. He is managing
many multidisciplinary projects on the design of
electric machines, power electronics, electric motor drives, and electrified
powertrains. Dr. Bilgin was the General Chair of the 2016 IEEE Transportation
Electrification Conference and Expo.

0885-8993 (c) 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.