Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
AbstractIn this paper the authors present a comparison the so-called, torque sharing functions (TSFs) [3], [4].
between optimised versions of three torque sharing functions, One of the advantages of EVs as compared to the ICE
with different number of degrees of freedom, used for improving based vehicles is the high torque capability at low speed
the efficiency in Switched Reluctance Motors. Starting from
the basic rectangular variation, the ascending and descending [9]. Moreover, for EV applications, it is important to reduce,
flanks of the function are modeled using a cosine, exponential or completely eliminate the torque ripple at low speed, as
and a more general approach (further referred to as piecewise they generate low-order frequencies which along with the
cubic sharing function). The degrees of freedom varying between eigenfrequencies of the drivetrain can lead to mechanical
two and an arbitrary number (for the general approach) are resonance. Therefore, producing high torque for low ripple
determined so as to minimize losses, when operating in smooth
torque conditions. Based upon an optimisation procedure with content is one of the main requirements of an electric motor
both linear and nonlinear constraints, the total (copper and used in EV propulsion [10].
iron) losses are minimized considering a limited available DC The total torque can be produced by maximum two neigh-
bus voltage. Moreover, a smooth model is presented so as to bour phases in SRMs with four-phases or less. At each rotor
be able to separate the two main causes for the torque ripple angular position, the share of each phase can be determined
and it is shown that, neglecting the chopping ripple, a ripple-free
torque can be obtained. Extensive studies have been conducted for using an appropriate function for modelling the ascending and
determining the optimal number of degrees of freedom (DoFs) descending flanks of the basic rectangular variation. One of the
for the piecewise cubic function. The proposed methods were first application of such a function is reported in [5] where an
successfully implemented on a 8/6, 30kW peak power, SRM both exponential variation is considered, followed by the authors
numerically and experimentally-wise. in [6] where a cosine function is introduced. From the same
category of nonlinear sharing functions, authors in [7] use a
I. I NTRODUCTION
cubic variation. Although the ripple content is significantly
Due to their undeniable advantages, SRMs represent a reduced there is no attention paid to other quantities such as
serious competitor for AC permanent magnet synchronous and iron or copper losses. In [8] the functions presented in [5]
induction machines used in hybrid and full electric vehicles, to [7] are optimised for two objective functions: the rate of
(HEVs/EVs), propulsion. Their main drawbacks are exten- change of flux linkage and the copper losses.
sively mentioned in the literature [1] - namely, noise vibration In this paper the optimised versions of the functions pro-
and harshness problems (NVH), low efficiency and high torque vided in [5], [6] are compared with generalized version of the
ripple. This paper deals with the latter two problems. cubic function (with an arbitrary number of DOFs), further
Besides the improvements that can be done at the machine referred to as piecewise cubic. Furthermore, the latter men-
design level, the control technique that is used plays a fun- tioned sharing function is tested by adding the iron losses in
damental role in overcoming the two mentioned drawbacks. the optimisation. Thus, in this paper as performance indicators,
Moreover, for EV applications, wide speed range, highly- copper and iron losses are used along with the extension of
dynamic, controllers are required, which in regular basis the torque-speed region from the torque-speed plane in which
combine two torque control techniques embedded in a speed smooth torque can be produced.
control loop [2]. One example would be an optimal firing- It is shown that the proposed piecewise cubic method allows
angle based control scheme, where at low speed the ripple of for the production of smooth torque for a wider torque-
the torque is minimized by means of current profiling using speed range than other presented methods, as well as an
III comprises descriptions of the implemented torque sharing
provided in Section V and validated by measurements in
Section VI. Finally, conclusions are drawn in the last section
along with suggestions for future work.
II. SRM S MOOTH M ODEL
A. Mathematical model Fig. 1. Maximum-torque-per-current ratio
In an SR drive the switching of the transistors and the
transition from one active phase to the next phase are the root
cause of the torque ripple. In order to quantify the share of two neighbor phases (ftsf < 1). If ftsf = 1, then the peak-to-
each one, the two causes are separated. Hereafter a smooth peak torque ripple, Tpp , can be approximated with (4) whereas
model (1) is proposed which allows for the switching ripple if ftsf < 1, then solely the maximum value of the Tpp can be
to be neglected. By using this model, instead of the discrete deduced (5).
voltage (Vdc or 0) an average value (Vav ) is applied to the
Tpp = 2Ih RTi (4)
phases. An additional advantage of using a smooth model is
a significant reduction of the computation time.
Tppmax = 2(I(h)ph1 R(Ti)ph1 + I(h)ph R(Ti)ph ) (5)
&
Vav,k = Rph Ik + ,
dt
Ik = Iref,k B. Optimisation problem formulation
(1)
The objective of the optimization is to minimize the losses
with k = 1 , 2 , ..mph , Ik , Vav,k , k , Rph the phase current,
(copper losses, PCu , iron losses PFe , or the total losses) for
voltage, flux linkage and resistance, respectively, Iref,k , Vdc
a ripple-free torque (in the smooth model) and considering a
the reference current and DC bus voltage.
limited available voltage, Vdc . As stated before three different
Copper losses, which are further used in the composition of
TSFs are employed, each with a different number of degrees of
the objective function, are expressed as follows
freedom (DoFs). Each DoF represents a fraction of the stroke
Nph
Ts ' angle.
1
%
PCu = Rph Ik2 (2) Let [x] be the vector with the DoFs, of length nd , f (x)
Ts 0 k=1 the function to be minimized and s , the stroke angle. The
with Ts = Tc /4 the stroke period, where Tc is the fundamental nonlinear optimisation problem with linear and boundary
period calculated as follows constraints can be formulated as follows, regardless of the
number of DoFs
60
Tc = (3)
NR
[A][x]T
(
s
with the speed (in rpm) and NR the number of rotor poles. min{f (x), such that (6)
[l] [x] [u]
To make the model as general as possible, it is transformed in
a spatial-dependent model in which the sharing functions are where f (x) is straightforwardly computed based on the model
represented in terms of number of strokes (Fig. 9 to Fig. 5). given by (1). For the considered number DoFs, [A] =
Eventually, the performance of the chopping model (addi- [1 1 1 . . .] has the same length as [x]. The validity of the first
tional torque ripple) is estimated by means of current hys- inequality will ensure that minimization function will discard
teresis band (4-5). If Ih is denoted as half of the current the [x] values whose sum will exceed the stroke length. Vectors
hysteresis band width, RTi as the ratio between maximum [l], [u] represent the lower and upper boundaries associated
torque and the phase current (Fig. 1) and ftsf as the value of with the DoFs given by [x]. However in this paper a different
the sharing function, then there are two possibilities: either the approach is considered. The model is simplified by means of
required torque is produced by one phase (ftsf = 1), either by considering only the second part of the model given by (1)
357
Fig. 2. Cosine TSF and the associated DoFs Fig. 3. Exponential TSFs and the associated DoFs
while the optimization problem becomes more complex as a B. Exponential sharing functions
nonlinear constraint is added in the formulation Starting from [5] a 3-DoFs TSF is obtained
C(x) = 0 (7)
0, (0, i )
with
(f , 2s )
1, (mi , mf )
Nph % Ts
1
mf 2
' ) *
C(x) = max(|Vav,k | Vdc , 0)dt (8) fexp () =
Ts 0 kexp
k=1
exp f ) mf [mf , f ]
*2
III. T ORQUE SHARING FUNCTIONS
i
kexp
1 exp mi mf
A. Cosine sharing functions [i , mi ]
(11)
Starting from [6] a 2-DoFs TSF is obtained, which is
As in the previous case, i , mi , mf and f are the angles
characterized by the equations given in the following
shown in the Fig. 11 whereas 1 , 2 and 3 are fractions of the
stroke. Yet, as compared to the previous situation an additional
0, (0, i ) DoF is introduced (kexp ) which affects the sharing functions
(f , 2s )
in the manner shown in Fig. 4.
fcos () = 1, (mi , mf )
0.5 0.5 cos(k( i ))
[i , mi ] C. Picewise cubic sharing functions
0.5 + 0.5 cos(k( mf ))
[mf , f ] In this subsection a generalized TSF with an arbitrary
(9) number of DoFs is developed, which yields minimum losses
where i , mi , mf , f are defined in Fig. 9 as well as the and enhance the ripple-free torque production capability, con-
considered DoFs (1 and 3 ) and k = (mi i )/5. sidering a limited voltage. By applying this method, one stroke
Applying a particular form (considering only the equality) is divided into nd fractions. The value of the function will
of the linear constraint defined in (6), 2 will become increase with 1/(nd 1) over one fraction. An example of
such a variation is shown in Fig. 5, for nd = 6. If the fractions
2 = s 1 3 (10) of the stroke are denoted as k then the following condition
must be fulfilled nd
Finally i and mi are obtained in terms of 1 and 2 '
(i = s 2 3 , mi = s 3 ) and they will fully k 1 (12)
define the sharing function that models the descending profile k=1
of the phase torque. The function for the ascending profile, where a supplementary fraction k+1 appears when (12) is a
delimited by mf and f is straightforwardly obtained, as strict inequality (7 in Fig. 5).
the two functions are complementary (mf = s + 1 and The linear variation between two consecutive points is
f = 2s 3 ). interpolated using a cubic spline method (shown with red line
358
IV. I NCORPORATION OF IRON LOSSES
Core losses computation is a challenging task for SRMs
due to the highly distorted character of magnetic flux density
waveform as well as the fact that different segments in
the machine are subject to different frequencies [11], [12].
Eddy-currents (14) and hysteresis (13) components [13] are
Tc *2
d2
)
1
%
dB
Pecl = dt (14)
12 m Tc 0 dt
Fig. 4. Exponential TSFs and the DoFs with fph the fundamental frequency (in Hz), kh , and ,
material-dependant parameters (obtained by fitting), Bm the
magnitude of major loop of magnetic flux density (in T), d the
lamination thickness (in m), m the density of the material
(in kg/m3 ) and the electrical resistivity (in m).
The B waveforms in the stator poles have the same funda-
mental frequency, fph = 1/Tc and equal with that given by the
cycle duration. For a SRM in long flux paths configuration of
the winding, there is always a stator yoke segment only with
359
Fig. 8. Required and available average dc voltage and their effect on the
instantaneous torque at 90Nm and 1500rpm
(c) Stator yoke SY4 (d) Rotor poles
(e) Rotor yoke
Fig. 7. Typical magnetic flux density waveforms in different segments of
the machine
an optimized version of the Levenberg-Marquardt method for
minimization. Considering the specific losses dependence on
the frequency and magnetic flux density, Ps (f , B ), for the Fig. 9. Comparison between current and torque obtained with smooth and
chopping model for two conducting phases using the optimised shape of the
M800-60A electrical steel-sheet, the next values are obtained piecewise cubic TSF
for the coefficients, = 1.777, = 0.192, kh = 0.0541. The
material has the following properties: d = 0.5mm, m = 7800
kg/m3 , = 25 108 m.
V. S IMULATION RESULTS
By combining the two logics given in (1) it can be seen
that the entire torque-speed range can be covered. However it
must be stated that when |Vav | |Vdc | a dip in torque will
occur, its severity being proportional with the speed (Fig. 8).
This is not discussed in detail, as the purpose of this paper
is to perform a study on the highest torque and speed that
can be reached for a ripple-free torque (in the smooth model
- Fig. 10) and minimum losses. Yet, in Fig. 10 there is still
an error (0.03%) which is the result of the inverting the FE
obtained table, T = f (, i), so as to obtain, i = f (, T ).
The validity of the (4-5) can be checked based upon Fig. 9
and Fig. 1 taking into consideration that the simulation was
conducted for a switching frequency, fsw = 15kHz which Fig. 10. Torque produced for peak current at 300rpm, when the TSF is
corresponded to a value for the current hysteresis band width, optimised using 5-DoF
360
Ih = 3.6A.
Simulations have been carried out in order to determine the
performance of each of the three TSFs. The next performance
indicators are used
copper losses
the peak of the torque-speed region in the torque-speed
plane with smooth torque production capability
variation with both the speed (Fig. 13) and reference torque
(Fig. 12) is studied.
Fig. 13. Optimal 5DoF piecewise cubic TSF variation with speed
variation (8% at rated torque and 12.2 % at peak torque).
As for the 300rpm, another important issue is reported -
the maximum value of the torque cannot be reached in the
limit of the available voltage when cosine and exponential
TSFs are used. Therefore for the cosine, as well as for the
exponential function, the maximum ripple-free torque that can
be produced is 150Nm and 165Nm, respectively, whereas for
Fig. 11. Torque sharing function and copper losses variation with the number the piecewise cubic TSF, still, the maximum torque of 175Nm
of DoF considered in optimisation can be delivered (see Fig. 10). The copper losses dependence
361
with reference torque, is shown in Fig. 15.
Iron losses are very often neglected when considering an
objective function for improving the efficiency. In this paper,
for the proposed piecewise TSF, tests are conducted varying
the speed between 1 to 900rpm and the effect of including the
Fig. 16. Influence of the objective functions on the total losses Fig. 17. Overview of the test-rig components
362
Fig. 18. Comparison between measured torque using the classical optimal
firing angle approach and current profiling
Fig. 20. Measured and reference current at 250rpm and 30Nm reference
torque using piecewise cubic TSF
Fig. 19. Measured instantaneous torque and phase currents at 250rpm and
30Nm reference torque
363
performance of the chopping model was estimated based on
the smooth model. It was found that the generalized version of
the cubic variation provides the lowest copper losses as well
as an extension of the torque-speed region in which smooth
torque can be produced.
Despite their remarkable performances at very low speed,
for the cosine and exponential TSFs, the difference in copper
ACKNOWLEDGEMENT
This research was financially supported in the frame of
the project Doctoral studies in engineering sciences with
364
[10] S. Faid, P. Debal, S. Bervoets, Development of a Switched Reluctance [12] Y. Hayashi, T. J. E. Miller, A New Approach to Calculating Core Losses
Motor for Automotive Traction Applications, The 25th World Bat- in the SRM, IEEE Trans. on Ind. Appl., vol. 31, no. 5, pp. 10391046,
tery, Hybrid and Fuel Cell Electric Vehicle Symposium & Exhibition September/October 1995.
Shenzhen, China, 5-9 November, 2010. [13] J. Gyselinck, M. De Wulf, L. Vandevelde, J. Melkebeek, Incorporation
[11] P. N. Materu, R. Krishnan, Estimation of Switched Reluctance Motor of Vector Hysteresis and Eddy Current Losses in 2D FE Magnetody-
Losses, IEEE Trans. on Ind. Appl., vol. 28, no. 3, pp. 668679, namics, Proc. ELECTRIMACS, Vol, 3, pp. 37-44, 4-16 Sept. 1999,
May/June 1992. Lisbon, Portugal.
365