IET Electric Power Applications

Research Article

ISSN 1751-8660
Method for radial vibration modelling in Received on 31st March 2016
Revised on 20th May 2016
switched reluctance motor Accepted on 13th June 2016
doi: 10.1049/iet-epa.2016.0177
www.ietdl.org

Xiaoqiang Guo 1, 2, Rui Zhong 1, 2, Longpan Zhao 3, Jun Yin 1,2, Weifeng Sun 1,2 ✉
1
National ASIC System Engineering Research Center, Southeast University, No. 2, Sipai Lou, Nanjing, People’s Republic of China
2
Power Devices and Integration Engineering Department, Collaborative Innovation Center of IC Design and Manufacturing of Yangtze
River Delta, No. 825, Zhangheng Road, Shanghai, People’s Republic of China
3
Electrical Technology Research Development Department, Zhuzhou Times Electronic Technology Co., Ltd., No. 199, Huanghenan Road,
Zhuzhou, People’s Republic of China
✉ E-mail: [email protected]

Abstract: The wide application of switched reluctance motor has been limited due to its high acoustic noise, which is
mainly caused by radial vibration. In this study, a new radial vibration model is proposed consisting of vibration
acceleration impulse model and natural oscillation model. The vibration acceleration impulse model is developed
based on the measured sample data using least-square support vector machine method. Natural oscillation model is
also proposed based on the superposition of each frequency, which is obtained by finite element method. The
vibration model is built up in MATLAB/Simulink and various experiments have been conducted to verify the accuracy
of the proposed model.

1 Introduction sudden change of acceleration at the commutation instant, the

artificial intelligence (AI) method of least-square support vector
Switched reluctance motor (SRM) has attracted much concern in the machine (LSSVM) is used. It is a learning machine based on
electric vehicle industry for its high efficiency, no maintenance and statistical learning theory and does not need any prior knowledge,
no consumption of rare Earth resources. However, due to the thus it is easy to use. Moreover, the exact relation between
doubly salient structure of the motor and the switch form power acceleration and electrical parameters, especially the sudden
supply, which lead to unavoidable acoustic noise, it brings great change of voltage [18], can be trained by LSSVM automatically
challenges to the development of SRM, especially in noise sensitive based on the measured sample data without the complicated
applications. The main sources of acoustic noise include torque analytical derivation. Compared with other AI methods such as
ripple [1–3], stator radial vibrations caused by radial force, neural network it has strong generalisation ability and even
unbalanced magnetic and mechanical forces [4–6]. It is widely stronger learning ability [20]. The vibration process is divided into
accepted that in the case of good mechanical processing, the two stages. At the turn-off instant, the vibration impulse is
acoustic noise caused by asymmetries can be greatly reduced. In produced causing the sudden increase of acceleration. After the
addition, asymmetries and torque ripple are considerably less switches are turned off, the damped oscillation of vibration takes
significant, and therefore not significant sources of acoustic noise, place. In this paper, the first stage, i.e. vibration acceleration
which has been fully demonstrated through experiments in [7]. It is impulse model, is built up based on the measured sample data
well known that the vibration in SRM is primarily caused by the using LSSVM. The second stage, i.e. natural oscillation, is also
elliptic deformation of the stator lamination stack due to its radial considered. It is assumed as the superposition of various vibration
magnetic attraction to the rotor. It is even worst when the frequency modes [21], which can greatly improve the accuracy than that
of radial vibration coincides with the natural frequency of the motor using just one frequency in [17].
[7]. For decades, much work has been done on the determination of This paper is organised as follows. The generation mechanism of
resonant frequencies [8, 9], mode shapes [10, 11], testing [12–14] vibration is briefly highlighted in Section 2. The fundamental
and winding, load, mounting, bearings effects [15, 16]. However, analysis of vibration for modelling will be developed in Section 3.
the accurate model of radial vibration deserves more attention. Section 4 describes the modelling process. Finally, the experimental
Currently, there is some work being done to study the model of radial results obtained on a prototype of 12/8 SRM are used to verify the
vibration. On the basis of analytical method, Zhu et al. [17] predicts the proposed model.
vibration before and after using the active vibration cancellation strategy.
However, the traditional analytical method is too complicated. In
addition, only one frequency is included to approximate the whole 2 Vibration generation mechanism
vibration. In [18], statistical energy analysis is used to describe the
high-frequency response of radial vibration. However, the description The test motor in this paper is a typical three-phase 12/8 SRM. The
of relation between radial force and rotor position is not so correct, major dimensions and specifications are listed in Table 1. Fig. 1
especially the description of the sudden change of acceleration at the shows a half the cross-sectional view and a photograph of external
commutation instant. Recently, based on measurement and the state view. The cross-sectional view is only a schematic diagram of stator
space method, the relation between the acceleration of stator tooth root and rotor laminations without housing. The actual SRM is
and the change rate of input energy is deduced in [19]. The prediction fundamentally round with a squarish housing. An accelerometer is
of radial vibration is rather accurate. Nevertheless, the derivation attached on the housing facing the root of stator tooth to measure
process of the state space method is very complex and need some the radial vibration. The end cover is first removed to find the exact
prior knowledge of the relation between various variables. position of the stator tooth on the surface of housing. After that, the
In this paper, aiming at the complicated derivation of vibration in accelerometer is fixed. Finally, the end cover is reassembled and the
previous literatures mentioned above [17, 19] and the unexplained positions of rotor position signal sensors are adjusted to ensure that

IET Electr. Power Appl., 2016, Vol. 10, Iss. 9, pp. 834–842
834 & The Institution of Engineering and Technology 2016
Authorized licensed use limited to: COLLEGE OF ENGINEERING AND TECHNOLOGY. Downloaded on April 16,2020 at 11:49:03 UTC from IEEE Xplore. Restrictions apply.
 Table 1 Major specifications of test SRM
Parameter Value

number of phases 3
number of stator and rotor poles 12/8
stator outer diameter 138 mm
thickness of stator yoke 9.5 mm
rotor outer diameter 79 mm
shaft diameter 25 mm
air-gap length 1.5 mm
stator pole arc 15°
rotor pole arc 17°
length of stator/rotor 80 mm
rated voltage 60 V
rated power 1500 W
rated speed 3600 rpm
rated torque 3.98 Nm
Fig. 2 Asymmetric H-bridge converter of one phase

the SRM can be under normal operation. The sample data of

simulation are all based on such position. For simplification, all the
analysis and the following proposed model are based on the
combination of lamination and housing. The effect of the shape and
material of the housing is not included in this paper.
Take the single pulse operation of SRM for an example in order to
simply explain about vibration generation mechanism. The drive
circuit of one phase is illustrated in Fig. 2, which is well-known
configuration in SRM drives named as asymmetric H-bridge
converter, where S1 and S2 are power switches, D1 and D2 are
freewheeling diodes and U0 is the power supply. Fig. 3 shows the
applied voltage waveform, current response and resultant
acceleration of vibration under the single pulse operation. The
phase winding is energised when both the switches are turned on.
While the rotor position reaches the turn-off angle, both the
switches are simultaneously turned off and the radial force acting
on the energised stator tooth suddenly changes its inclination so
that the vibration occurs on the excited tooth [22]. The vibration
process is divided into two stages. At the turn-off instant, the
vibration impulse is produced causing the sudden increase of
acceleration. After the switches are turned off, the damped Fig. 3 Experiment of single voltage pulse response
oscillation of vibration takes place. In [7], it has been revealed that

the natural oscillation has been dominant one after both the
switches are turned off. Moreover, it will be further demonstrated
in Section 3.3 that when both the switches are turned off, natural
oscillation occurs and the influence of current to the oscillation is
not particularly significant.

3 Fundamental analysis of vibration modelling

On the basis of the test and analysis above, the vibration process is
divided into two stages: vibration impulse and natural damped
oscillation, respectively. Thus, the theory of the modelling is also
divided into two parts. First, for the vibration impulse model, the
severe vibration mainly occurs when both the switches turn off. At
this moment, the correlation of vibration impulse should be
analysed first. Then as the correlatives with vibration impulse
maybe rather complicated, measurement method is adopted to
obtain the vibration sample data and the artificial intelligent
method LSSVM is used for research on multidimensional mapping
relation for its strong learning and generalisation ability. The
detailed correlation theory is presented in Section 3.1. The
measurement method theory is based on [23] and the
corresponding measurement process is described in Section 3.2.
The modelling theory based on LSSVM has been previously
reported in [20]. Second, the natural oscillation is primarily on the
resonant frequencies of motor. Owing to the basic theory of
signals and systems, the oscillation waveform is superimposed by
Fig. 1 Test rig sine/cosine waves of each order frequency. The more orders are
a Schematic diagram of the cross-sectional view of test motor and location of attached
superimposed, the more accuracy will be obtained. However, due
accelerometer to the limited hardware resources, this is difficult to realise. Thus,
b External photograph of test motor only the main frequency components are selected to approximate

IET Electr. Power Appl., 2016, Vol. 10, Iss. 9, pp. 834–842
& The Institution of Engineering and Technology 2016 835

Authorized licensed use limited to: COLLEGE OF ENGINEERING AND TECHNOLOGY. Downloaded on April 16,2020 at 11:49:03 UTC from IEEE Xplore. Restrictions apply.
 Fig. 4 Experimental results of vibration acceleration between different currents, voltages and rotor positions
a Acceleration relation with current at fixed voltages and rotor positions
b Acceleration relation with voltage at fixed currents and rotor positions
c Acceleration relation with current and rotor positions under working voltage

the real waveform. The detailed process for natural oscillation [19], vibration is sensitive to the derivative of voltage when there
modelling is presented in Sections 3.3 and 3.4. exists a non-zero flux

dL d2 x dL di dx di d2 i dv
3.1 Correlation analysis of vibration impulse i + 2 + R + L − =0 (1)
dx dt2 dx dt dt dt dt dt
To build up the vibration acceleration impulse model, correlation of
vibration impulse should be analysed. According to Lin and Fahimi where v is the terminal voltage, i is the phase current, L is the

IET Electr. Power Appl., 2016, Vol. 10, Iss. 9, pp. 834–842
836 & The Institution of Engineering and Technology 2016
Authorized licensed use limited to: COLLEGE OF ENGINEERING AND TECHNOLOGY. Downloaded on April 16,2020 at 11:49:03 UTC from IEEE Xplore. Restrictions apply.
 Table 2 Comparison of measured peak frequencies and FE resonant
frequencies

Number Measured results FE results Relative error, %

1 643 690 (i) 7.3

2 1378 1245 (ii) −9.7
3 1637 1656 (iii) 1.2
4 2067 2263 (vi) 9.5
5 2713 2480 (vii) −8.6
6 3015 2997 (viii) −0.6

commutation instant is larger than that when only one switch turns
off under pulse-width modulation (PWM) control mode. On the
basis of above, the purpose of this paper is to find the occurrence
of the maximum vibration, to study the basic generation
mechanism of vibration and to build up the vibration model
concerning the main, i.e. the maximum vibration. The vibration
model lays the foundation of the proposal of vibration reduction
strategy.
Fig. 5 Experimental verification of natural oscillation
3.2 Correlation acquisition of vibration impulse
inductance of phase winding, x is the radial displacement of stator The rotor is first fixed at the aligned position to obtain the relation
teeth and t is time. Thus, d2x/dt 2 is the acceleration on the stator between peak value of vibration impulse and voltage and current.
frame. Equation (1) is a variable coefficient differential equation of Two comparisons under different conditions are presented.
displacement. It implies that the displacement is related to current, First, as shown in Fig. 4a, the same voltage of 10 V is applied to
voltage and inductance at the turn-off instant. Since inductance is the phase winding with different time duration at a fixed rotor
a non-linear function of current and rotor positions, vibration position. Thus, the peak values of current (Ip) must be different. It
acceleration will finally depend on current, voltage and rotor shows that the peak values of acceleration (Ap) increase along with
position. That is to say, the peak value of the radial vibration the increase of Ip. The similar experiment with the same voltage of
acceleration at the turn-off instant lies on the terminal voltage and 15 V is conducted and the same conclusion can be obtained. It can
phase current, when the rotor is mechanically fixed at a desired be observed that the peak value of vibration acceleration is linear
position. As described above and demonstrated in [13, 17, 19, 22], to the current when using the same phase voltage and fixed at the
the vibration is proportional to the change rate of phase voltage. same position.
Thus, the vibration caused when both the switches turn off at Second, different voltages (Up) are applied and last with different
time duration so as to reach the same peak current value of 10 A. In
Fig. 4b, it shows that the peak values of acceleration increase along
with the increase of Up. The similar experiment with the same
current of 20 A is conducted and the same conclusion can be
obtained. It can also be observed that there is also a linear relation
between the peak value of vibration acceleration and voltage when
reaching the same peak current at turn-off instant and fixed at the
same position.
When the rotor is fixed at a certain position, the acceleration
relation with current and voltage can be obtained by repeated
experiments, setting different peak currents and voltages. It has
been verified in many literatures that the relation with respect to
rotor position is non-linear [7, 24]. From the method above, the
acceleration relation with current and voltage at each rotor position
can be obtained. Since the purpose of this paper is a proof of
concept, the small variation of DC power supply is not considered
in this paper. Thus, the voltage at turn-off instance is supposed to
be a constant, so only current and rotor positions will affect the
acceleration. The acceleration relation with current and rotor
positions under working voltage is shown in Fig. 5.
On the basis of previous discussion, the peak value of vibration
impulse when the switches turn off is linear to phase current and
voltage and non-linear to rotor position. Thus, the AI method of
LSSVM can be used to build the vibration acceleration impulse
model due to its strong learning and generalisation ability using
just a few sample data. Moreover, it can derive the relation
between acceleration and the sudden change of voltage
automatically.

3.3 Verification of natural oscillation

Fig. 6 FE and measured resonant frequencies
a First eight vibration modes of FE results After both the switches are turned off, oscillation will appear.
(i) Mode at 690 Hz, (ii) mode at 1245 Hz, (iii) mode at 1656 Hz, (iv) mode at 1791 Hz,
(v) mode at 1821 Hz, (vi) mode at 2263 Hz, (vii) mode at 2480 Hz and (viii) mode at
However, the phase current still exists that may cause vibration.
2997 Hz To verify the possible effect of current on vibration, experimental
b Vibration acceleration spectrum verification is shown in Fig. 5. In the experiment, a large external

IET Electr. Power Appl., 2016, Vol. 10, Iss. 9, pp. 834–842
& The Institution of Engineering and Technology 2016 837

Authorized licensed use limited to: COLLEGE OF ENGINEERING AND TECHNOLOGY. Downloaded on April 16,2020 at 11:49:03 UTC from IEEE Xplore. Restrictions apply.
 Fig. 7 Simulation model
a Structure of radial vibration model
b Complete model of single phase
c Internal structure of acceleration impulse model in Fig. 7b
d Internal structure of acceleration oscillation model in Fig. 7b

IET Electr. Power Appl., 2016, Vol. 10, Iss. 9, pp. 834–842
838 & The Institution of Engineering and Technology 2016
Authorized licensed use limited to: COLLEGE OF ENGINEERING AND TECHNOLOGY. Downloaded on April 16,2020 at 11:49:03 UTC from IEEE Xplore. Restrictions apply.
 Fig. 8 Characteristics of the test motor
a Static flux-linkage characteristics
b Static torque characteristics

capacitor, which is parallel with the voltage bus, is first charged by The comparison results are listed in Table 2. It shows that the
the power supply. Then, the power supply is turned off and relative errors of FE and measured frequencies are all within 10%,
immediately followed by the turning on of the switches. The which demonstrates the validity of the proposed modal analysis
switches are kept on since the power supply is turned off and the method. However, the error can still be reduced. The main reason
current in the phase winding continues to rise until the phase for the error is that, for simplification, the FE method uses stator
voltage decreases to zero, i.e. the capacitor is fully discharged. lamination only for analysis, whereas the measured results are
However, there is still current in the phase winding but no obvious based on the combination of stator lamination and housing. Another
vibration is produced, which demonstrates that after the switches phenomenon is that in Fig. 6b, the amplitude of vibration mode at
are turned off, the vibration is natural oscillation and the current 1656 Hz is rather low and the vibration mode at 1791 Hz has not
will have little impact on the oscillation. been aroused at all, which demonstrated the conclusion that the odd
order vibration mode [shown in Figs. 6a(iii) and (iv)] is rarely excited.

3.4 Selection of resonant frequencies

To build up the natural oscillation model, resonant frequencies of the 4 Modelling of SRM radial vibration
test motor should be obtained. In this paper, an finite element (FE)
model is constructed according to the actual motor dimensions and The radial vibration model is constructed as shown in Fig. 7a. It is a
the resonant frequencies of each modal are obtained based on the cascade of flux model and vibration model. The flux model is the
finite element analysis (FEA) software ANSYS. Fig. 6a shows the basis of the whole model. It predicts the current that the vibration
first eight modes of vibration. model needs. The inputs of flux model are rotor position and flux
In addition, Fig. 6b shows the spectrum waveform of acceleration linkage while outputs are current and torque. The characteristics of
based on measurement. The corresponding frequencies of each peak the test motor (flux and torque versus current and rotor positions)
point are probably resonant frequencies of vibration modes. The are shown in Fig. 8. Along with the rotor position and voltage, the
frequencies which are close to FEA results are selected as resonant current predicted by the flux model is fed as the input of the
frequencies of natural oscillation. As mentioned above, the vibration model. The vibration model consists of vibration
vibration is primarily caused by the elliptic deformation of the acceleration impulse model and natural oscillation model.
stator lamination stack, so only low-order modes are compared and Both the current concerning rotor position and flux linkage in the
used for modelling. In Fig. 6a, the first six peak frequencies, which flux model, and the maximum acceleration concerning rotor
are numbered from 1 to 6, are selected to compare with FE results. position, current and phase voltage in the vibration acceleration
impulse model are predicted based on LSSVM due to its strong
learning and generalisation ability under the situation of only a
few sample data. The basic theories of LSSVM and the detailed
training process including obtaining raw data by the improved DC
pulse test method [23] and selection of kernel function have been
studied in our previous literature [20].
Fig. 7b shows the complete structure of single phase. It consists of
power converter module, angular transformation module, flux model,
acceleration impulse model and acceleration oscillation model. The
expanded flux model can be referred to the previous publication in
[20]. The acceleration impulse mode and acceleration oscillation
model are expanded in Figs. 7c and d, respectively.
Fig. 7c shows the simulation block of vibration acceleration
impulse model in MATLAB/Simulink. The ‘Acceleration’ block is
based on LSSVM to predict the maximum acceleration according
to the rotor position, phase voltage and current, and it is realised
Fig. 9 Experimental platform by S-function. The ‘Clock’ module is a system module to get the

IET Electr. Power Appl., 2016, Vol. 10, Iss. 9, pp. 834–842
& The Institution of Engineering and Technology 2016 839

Authorized licensed use limited to: COLLEGE OF ENGINEERING AND TECHNOLOGY. Downloaded on April 16,2020 at 11:49:03 UTC from IEEE Xplore. Restrictions apply.
 Fig. 10 Waveforms of simulated and measured under two conditions
a Waveforms of time domain under APC mode (θon = 0°, θoff = 15°, TL = 1.22 Nm and speed is 1600 rpm)
b Frequency domain of Fig. 10a
c Waveforms of time domain under CCC mode (θon = −3°, θoff = 14°, TL = 1.43 Nm and speed is 1200 rpm)
d Frequency domain of Fig. 10c

system time. The ‘Triggered Subsystem’ is used to transfer the On the basis of vibration acceleration impulse model, the peak
maximum acceleration and record the occurrence time when both value of vibration acceleration Ap at turn-off instant can be
switches turn off. The ‘OR’ logical operator is used to detect the predicted. In Fig. 6b, the frequency and corresponding amplitude
state of two switches. Only when the switches turn off of each order is obtained. As demonstrated in [17], vibration and
simultaneously, no matter from excitation state or from single acoustic noise of SRM mainly originate from the ellipse modes of
switch freewheeling state, the logical operator will send the enable the first and second vibration modals. In this paper, only six of the
signal to trigger the subsystem. The ‘Section Time’ module eight frequencies are excited. Thus, the six excited frequencies are
calculates the time variable for the acceleration oscillation model, chosen as the natural oscillation harmonic frequencies of each
which takes the occurrence time when both switches turn off order. The six harmonic oscillation amplitudes of A1, A2, …, A6
simultaneously as time datum. Fig. 7d shows the realisation of are determined as 79.2, 7.9, 2.4, 7.9, 2.4 and 0.2% of Ap.
acceleration oscillation model based on MATLAB/Simulink. According to [25], when the switches are turned off, each
harmonic oscillation Vi (t) can be expressed as (2)

Table 3 Comparison of acceleration peak value between simulation and Vi (t) = Ai · e−jvi t · cos vi t (2)
measurement under two conditions

Simulated, ms−2 Measured, ms−2 ɛmae, m/s2 ɛmre, % where ωi is the natural angular frequency and ζ is the damping factor.
As the damping factor of vibration system is usually very small, it
APC 512 480 32 6.7
CCC 409 432 23 5.3
can be estimated based on the decay ratio of the vibration from
experiment [26]. Since the vibration can be assumed as the

IET Electr. Power Appl., 2016, Vol. 10, Iss. 9, pp. 834–842
840 & The Institution of Engineering and Technology 2016
Authorized licensed use limited to: COLLEGE OF ENGINEERING AND TECHNOLOGY. Downloaded on April 16,2020 at 11:49:03 UTC from IEEE Xplore. Restrictions apply.
 superposition of various vibration modes [21], the total natural and has been demonstrated feasible. In addition, it does neither
oscillation can be expressed as (3) need any prior knowledge nor complicated derivation. The
proposed model lays significant foundation for the research of


8

8 underlying physical mechanism, and then for the study of
V (t) = Vi (t) = Ai · e−jvi t · cos vi t (3) vibration reduction strategy.
i=1 i=1

6 Conclusion
5 Simulation and experimental results This paper verifies the correlation of radial vibration with current,
voltage and rotor position at the turn-off instant. Subsequently,
Fig. 9 shows the experimental platform. It mainly consists of the based on LSSVM with ease of use and without prior knowledge, a
STM32-based controller, current hall sensor, accelerometer, charge vibration acceleration impulse model is proposed. In addition,
amplifier and dynamometer. The radial vibrations are detected natural oscillation is considered and is built up based on modal
using a piezoelectric accelerometer attached to the stator frame, superposition. The accuracy is improved in this paper than that of
then amplified by a charge amplifier, and finally sampled by just using one frequency to describe the entire oscillation. The
oscillation. The final resolution of acceleration signal output is overall radial vibration model consists of vibration acceleration
25 mV/g at 1 Hz–100 kHz. The current is measured by current hall impulse model and natural oscillation model. Various experiments
sensor whose resolution is 4 V/300 A and is sampled by are conducted to verify the proposed model under different
oscillation. The phase voltage is directly acquired by oscillation. conditions. However, there are still problems to be solved:
Two conditions with different turn-on angles, turn-off angles and
loads are set for simulation and measurement to verify the accuracy
(i) As the superposition and elimination effects of vibration under
of the SRM radial vibration model. As this paper aims at the
PWM control mode are not included in this paper, further study will
mechanism research on vibration, for simplification, the working
be developed. In addition, the influence of turn-on delay and turn-off
voltage is only 20 V (rate voltage is 60 V). In both two conditions,
delay will be also studied.
the power supply is 20 V, the chopping limit is 40 A and the
(ii) Influence of high frequencies will be investigated, though they
chopping threshold is 5 A. The simulated and measured voltage
are less significant than low frequencies.
and current waveforms are shown in Fig. 10, where θon is turn-on
(iii) The research in this paper is not generalised. The primary work
angle, θoff is turn-off angle and TL is load torque.
in this paper is to build up a 12/8 poles low-voltage low-speed SRM
In Fig. 10a, the motor is operated under angle position control
model and to verify its validity and accuracy. The future work is to
(APC) mode. In Fig. 10c, the motor is operated under current
derive and verify the physical mechanism on other kinds of SRMs
chopping control (CCC) mode. The shapes of simulated and
such as high-voltage SRM, high-speed SRM etc. On the basis of
measured waveforms are similar, and their chopping times are the
the general physical mechanism, research on vibration reduction
same. However, this figure shows that when the switches are both
strategy will be conducted.
on, the current decreases instead. That is because the speed is
1200 rpm, causing the rotational back electromotive force rather
high and even higher than the bus voltage. In addition, in the 7 Acknowledgments
proposed model of this paper, vibrations occur only when both the
switches turn off simultaneously. Therefore, when under current This work is supported by the National Nature Science Foundation
chopping regulation mode, only one switch turns off, which has of China (51277026, 61201034 project numbers) at the National
little impact on the vibration waveform. The high consistence ASIC System Engineering Research Center of Southeast
between simulated and measured waveforms also demonstrates this University in China.
conclusion.
In Figs. 10a and c, each peak value of simulated acceleration when
both switches turn off is compared with the corresponding measured 8 References
peak value. Moreover, the relative error is calculated as well as the
absolute error between simulation and measurement. Owing to 1 Brauer, H.J., Hennen, M.D., De Doncker, R.W.: ‘Control for polyphase switched
theory of error, the relative error is more reasonable than absolute reluctance machines to minimize torque ripple and decrease ohmic machine
error to evaluate the accuracy of each predicted acceleration. losses’, IEEE Trans. Power Electron., 2012, 27, (1), pp. 370–378
2 Mikail, R., Husain, I., Islam, M.S., et al.: ‘Four-quadrant torque ripple
Among all the calculated errors under both control modes, the minimization of switched reluctance machine through current profiling with
maximum relative error of vibration acceleration at turn-off instant mitigation of rotor eccentricity problem and sensor errors’, IEEE Trans. Ind.
under APC mode and CCC mode are 6.7 and 5.3%, respectively, Appl., 2015, 51, (3), pp. 2097–2104
listed in Table 3. 3 Ye, J., Bilgin, B., Emadi, A.: ‘An offline torque sharing function for torque ripple
reduction in switched reluctance motor drives’, IEEE Trans. Energy Convers.,
In addition, the amplitude and frequency of natural oscillation are 2015, 30, (2), pp. 726–735
consistent, which can be seen in the frequency-domain waveform in 4 Weiss, C.P., Qi, F., Hofmann, A., et al.: ‘Analysis and modelling of rotor
Figs. 10b and d. Both figures show that the frequencies of low eccentricity for switched reluctance machines’. Seventh IET Int. Conf. on Power
orders, especially the frequency of ellipse deformation, are Electronics, Machines and Drives (PEMD 2014), 2014, pp. 1–6
5 Tavakoli, M.R., Torkaman, H., Faradjizadeh, F., et al.: ‘Unbalanced magnetic force
extremely significant. However, the model still needs to be mitigation and torque ripple reduction in eccentric switched reluctance motors
improved in high frequencies and under non-APC control mode, using flux leakages’, Electr. Power Compon. Syst., 2015, 43, (5), pp. 533–542
though their influence is much smaller than low frequencies 6 O’Connell, T.C., Wells, J.R., Watts, O.A.: ‘A lumped-parameter off-axis-capable
and APC. dynamic switched reluctance machine model including unbalanced radial forces’,
IEEE Trans. Energy Convers., 2015, 30, (1), pp. 161–174
The accuracy in both time domain and frequency domain 7 Cameron, D.E., Lang, J.H., Umans, S.D.: ‘The origin and reduction of acoustic
demonstrates the feasibility of the proposed modelling method. noise in doubly salient variable-reluctance motors’, IEEE Trans. Ind. Appl.,
Moreover, it is far more accurate than just using one frequency to 1992, 28, (6), pp. 1250–1255
fit the entire vibration in [17]. However, the model still needs to 8 Cai, W., Pillay, P.: ‘Resonant frequencies and mode shapes of switched reluctance
motors’, IEEE Trans. Energy Convers., 2001, 16, (1), pp. 43–48
be improved. For simplification, just six vibration modes are 9 Wang, H., Tang, S., Xue, B.: ‘Rotor resonances of bearingless switched reluctance
superimposed to fit the entire vibration and as the exact expression motor with axial bias permanent magnet’. 2015 IEEE Magnetics Conf.
of oscillation is much more complicated than that proposed in this (INTERMAG), 2015, p. 1
paper, causing the accuracy are slightly lower than that in [19]. So 10 Laudensack, C., Dentler, T., Yu, Q., et al.: ‘Calculation and measurement of
eigenmodes for a canned switched reluctance machine’. 2012 15th Int. Conf. on
the resonant frequencies and the impact of the superposition of Electrical Machines and Systems (ICEMS), 2012, pp. 1–6
various vibration modes should be further studied, whereas the 11 Long, S.A., Zhu, Z.Q., Howe, D.: ‘Vibration behaviour of stators of switched
method for modelling in this paper is much easier than that in [19] reluctance motors’, IEE Proc. Electr. Power Appl., 2001, 148, (3), pp. 257–264

IET Electr. Power Appl., 2016, Vol. 10, Iss. 9, pp. 834–842
& The Institution of Engineering and Technology 2016 841

Authorized licensed use limited to: COLLEGE OF ENGINEERING AND TECHNOLOGY. Downloaded on April 16,2020 at 11:49:03 UTC from IEEE Xplore. Restrictions apply.
 12 Cai, W., Pillay, P., Tang, Z.J., et al.: ‘Vibration measurements in the switched 20 Zhong, R., Cao, Y.P., Hua, W., et al.: ‘An improved model of switched reluctance
reluctance motor’. Industry Applications Conf., 2001 36th IAS Annual Meeting. motors based on least square support vector machine’. 2013 IEEE Int. Symp. on
Conf. Record of the 2001 IEEE, 2001, vol. 1, pp. 11–17 Industrial Electronics (ISIE), 2013, pp. 1–6
13 Lin, C.J., Fahimi, B.: ‘Design considerations for reduction of acoustic noise in 21 Fiedler, J.O., Kasper, K.A., De Doncker, R.W.: ‘Calculation of the acoustic noise
switched reluctance drives’. 2014 IEEE Energy Conversion Congress and spectrum of SRM using modal superposition’, IEEE Trans. Ind. Electron., 2010,
Exposition (ECCE), 2014, pp. 1408–1413 57, (9), pp. 2939–2945
14 Girgis, R.S., Verma, S.P.: ‘Experimental verification of resonant frequencies and 22 Makino, H., Kosaka, T., Matsui, N.: ‘Digital PWM control-based active vibration
vibration behavior of stators of electrical machines. Part 1: models, experimental cancellation for switched reluctance motors’. 2014 IEEE Energy Conversion
procedure and apparatus’, IEE Proc. B, Electr. Power Appl., 1981, 128, (1), Congress and Exposition (ECCE), 2014, pp. 3834–3841
pp. 12–21 23 Carstensen, C.E., Fuengwarodsakul, N.H., De Doncker, R.W.: ‘Flux linkage
15 Castano, S.M., Bilgin, B., Fairall, E.: ‘Acoustic noise analysis of a high-speed determination for correct modelling of switched reluctance machines – dynamic
high-power switched reluctance machine: frame effects’, IEEE Trans. Energy measurement versus static computation’. IEEE Int. Electric Machines & Drives
Convers., 2015, PP, (99), pp. 1–9
Conf., 2007, 2007, pp. 1317–1323
16 Laudensack, C., Gerling, D.: ‘Vibration and noise analyzes of canned switched
24 Wu, C.Y., Pollock, C.: ‘Time domain analysis of vibration and acoustic noise in the
reluctance machines with different winding topologies’. 2014 17th Int. Conf. on
switched reluctance drive’. Sixth Int. Conf. on Electrical Machines and Drives,
Electrical Machines and Systems (ICEMS), 2014, pp. 1847–1852
17 Zhu, Z.Q., Liu, X., Pan, Z.P.: ‘Analytical model for predicting maximum reduction 1993 (Conf. Publ. No. 376), 1993, pp. 558–563
levels of vibration and noise in switched reluctance machine by active vibration 25 Sun, J.B., Zhan, Q.H., Wang, S.H., et al.: ‘Parameter-optimized three-step
cancellation’, IEEE Trans. Energy Convers., 2011, 26, (1), pp. 36–45 commutation method for the reduction of vibration and acoustic noise in
18 Shen, L., Wu, J.H.: ‘Switched reluctance motor vibration prediction: from low switched reluctance motor’. Int. Conf. on Electrical Machines and Systems,
frequency to high frequency’. IEEE Int. Electric Machines and Drives Conf., 2007. ICEMS, 2007, pp. 513–517
2009. IEMDC ‘09, 2009, pp. 971–978 26 Miller, T.J.E.: ‘Electronic control of switched reluctance machines’ (Newnes,
19 Lin, C.J., Fahimi, B.: ‘Prediction of radial vibration in switched reluctance Oxford, UK, 2001), in Chapter 3 and Chapter 8
machines’, IEEE Trans. Energy Convers., 2013, 28, (4), pp. 1072–1081

IET Electr. Power Appl., 2016, Vol. 10, Iss. 9, pp. 834–842
842 & The Institution of Engineering and Technology 2016
Authorized licensed use limited to: COLLEGE OF ENGINEERING AND TECHNOLOGY. Downloaded on April 16,2020 at 11:49:03 UTC from IEEE Xplore. Restrictions apply.