Professional Documents
Culture Documents
Abstract—This paper presents a second order sliding mode However, this type of controller gives the disadvantage of vari-
current controller for the switched reluctance machine (SRM). able switching frequency that may cause a subsonic noise in
This strategy is first analysed theoretically, then tested by SRM [8]. An alternative solution is a fixed switching frequency
simulation and on an experimental test bench. Dynamic, current
ripples and robustness of this controller are discussed. operation (PWM) with linear and nonlinear controllers. Such
controllers are widely used for variable speed/torque of AC
Index Terms—Switched reluctance machine, current control, machines. This technique makes the machine less sensitive to
second order sliding mode controller, super twisting algorithm,
experimental validation. acoustic noise, but is in general less dynamic than a hysteresis
regulator.
Several nonlinear control methods, such as the feedback
I. I NTRODUCTION linearizing [9], passivity [10], back-stepping [11] and sliding
phenomenon. It can stimulate certain modes of the system This control method is known for its robustness against
that are not necessarily taken into account in the modelling. parametric uncertainties. A proof of stability based on Lya-
Thus, in the 1980’s, the Russian school [15] has proposed new punov approach is given in [19] for the super twisting algo-
controls sliding modes that can avoid this problem. There are rithm.
based on the theory of second order sliding modes, and lead In the next section, we will use and expose this algorithm
to control laws still relatively simple to implement. for the SRM’s current control.
B. One order sliding mode III. A PPLICATION OF THE SECOND ORDER SLIDING MODE
TO THE SRM’ S CURRENT CONTROL
The implementation of a sliding mode control involves two
steps. Firstly, a surface S must be defined in the state space, A. Structure of the SRM’s control law
given by a constraint function s(t, x, u) such that, in steady
Fig.1 describes a SRM’s speed control strategy based on
sliding (locked state on this surface), the system has the desired
an average torque control [20] [21]. As with any type of
behaviour. Secondly, a discontinuous control law is defined,
electrical machine speed controlled, the speed controller’s
acting on the first derivative of the sliding variable and ensures
(Integral-Proportional) output defined the required electromag-
that the sliding surface is attractive (at least locally).
netic torque Te∗ . In this strategy, one important feature of
In the ideal case of a system without uncertainty, the com-
this classical controller is that the reference phase current is
mand which forces the system trajectories to evolve exactly
constant over one excitation period, and the reference torque
on the sliding surface is called the equivalent command Ueq
is considered as an average torque over on conducting period.
[16], and is solution of:
Thereby, this control is also called “square wave control”.
s = ṡ = 0 Three fundamental control variables, i.e., reference phase
current I ∗ , turn-on angle ψ, and conduction period θp , have to
In the general case, the sliding mode control U decomposes be adjusted. Many combinations of these control variables are
in two terms: possible to operate the SRM drive at one specific torque-speed
U = Ueq + Udisc operating point. However, one suitable combination for one
speed-torque operating point should be chosen, based on the
where Udisc is the discontinuous control which ensures con- desired optimization goal, e.g., efficiency or low torque ripple.
vergence in finite time to the surface S and the rejection of Using simulations, an optimal set of the control variables
a certain class of disturbances. Ueq represents the average of over the entire operating range can be obtained. The torque
the real command U. Therefore, the discontinuity’s amplitude translation into a current reference is located in a look-up table.
must be superior to the limits within which are evolving Linear data interpolation is performed on line to compute the
models of uncertainties and disturbances. The problem is that optimal control parameters depending on the operating point.
the first order brings up chattering on the controller output. Many classical SRM torque controllers use this approach and
rely on lookup tables of the control parameters.
C. Second order sliding mode
In the case of the second order sliding mode, and contrary to n∗ Te∗
ψ
I∗
Speed Look-up θp Commutating Current’s U SRM θ
the one order, the discontinuous control acting, not on the first n
controller tables I∗ strategy controller model
s = ṡ = 0
Different kinds of second order sliding mode algorithms B. SRM’s current controller
can be found in the literature [17]. In this work, we propose Fig. 2 presents a continuous model of the SRM’s current
to implement the super twisting algorithm [18] for the SRM’s control. Simulation and experimental results show that back-
current control. This algorithm has been developed for systems emf that acts as a strong disturbance need to be compensate so
with relative degree equal to one compared to the slip variable. that to improve the control performances. The back-emf Ê is
Therefore, the discontinuity affects the first derivative of the estimated from an analytical model of the phase’s inductance
control input U. L, dependent to the current and rotor position [22].
3302
3
The phase electric model is derived from the classical The average control voltage U is composed, like a
equation: Proportional-Integral controller, two continuous terms, that do
dΦ(i, θe ) not depend upon the first time derivative of the sliding variable.
U = Ri +
dt The discontinuity only appears in the first derivative control
where R is the phase’s resistance and Φ is the phase’s magnetic input U: U = Ui + Up with:
flux. The magnetic flux depends on the current and the electric dUi
= β sign(s) (integral term) (1)
position θe of the rotor (angle between a rotor slot and dt √
the phase’s unaligned position). Therefore, a look-up table Up = α s sign(s) (proportional term) (2)
(performed offline) give the phase’s current from the flux and
the position. with α > 0 , β > 0 and sign(s) is the sign function
(sign(s) = 1 if s > 0 ; sign(s) = −1 if s < 0).
Ê(I, θe ) Instead of integrating the error, the super-twisting algorithm
One phase electric model
∗
+ integrates the sign of the error. The proportional term depends
I + Sliding mode + U + 1 Φ I
current controller s
f (Φ, θe ) on the square root of the error. Near the reference, the error
− −
I θe is low, therefore, the proportional term varies faster than the
R
error, witch induces a more responsive controller.
The controller’s gains are α and β, and if the sufficient
conditions below are satisfied:
Fig. 2. SRM’s current control.
C0 4 C0 (KM β + C0 )
β> and α2 ≥ ,
The phase’s electric equation is: Km K2m (Km β − C0
one can obtain the convergence in finite time on the sliding
di
+E
U = R i + Linc surface S (s = ṡ = 0) [23], [24].
dt In practice, the controller is implemented on a processor.
where Linc is the phase’s incremental inductance and E the Fig. 3 shows the discrete current controller using the super
back-emf. Therefore, the first derivative of the constraint twisting algorithm. Unlike a conventional linear controller,
function s = i∗ − i where i∗ is the reference phase current there is no anti-wind-up action associated with the saturation.
and i is the measured phase current is: Indeed, the experimental results showed better performances
without this structure. This scheme include an initialization
ds di∗ U E + Ri
= − − block of the integrator when the current set point reverts to
dt dt Linc Linc
zero.
We set :
Ê
ds
= φ(t, s) + ϕ(t, s)U
dt I∗ + s + Ui + + U
βsign(s)
with − + +
E + Ri 1 Up
φ= and ϕ= − I
Linc Linc z −1 α (s)sign(s)
di∗ ∗
with the hypothesis that = 0 (i is constant because the
dt
average torque control requires slot’s current). Ui = 0 if I ∗ = 0
3303
4
Experimental tests has been performed with the mechanical response time is equal to 0.8 ms.
operating point (3.8 N.m - 500 rpm). Here, a 1103 Dspace So there is a compromise between high current dynamic and
board is used on the test bench (see Fig. 10). The sample time current ripples at steady state. To eliminate this compromise,
is set to 50 μs. Fig. 4 shows the current and voltage in one it is possible to use variables gains (α, β) that depend on the
phase of the SRM, for a reference current equal to 30 A. Gains variable s [25].
are set to α = 3.5 and β = 0.1. The static error is equal to zero |s|
α(s) = α0
and the average phase voltage does not present oscillations, so |s| + aα
the current ripple is minimal, and is only due to the inverter’s |s|
commutations. However, the response time is important and is β(s) = β0
|s| + aβ
equal to 1.4 ms. Indeed, the controller’s output is not saturated
to Udc during the initial current’s increase. Gains are greatest when the variable s is important, witch give
a high dynamic, and are reduced when the error is near zero.
(a) Reference (− −) and measured (−) one phase currents Therefore, the control is smooth near the reference. There are
40 2 additional degrees of freedom for the controller. The choice
30 of aα and aβ defines the bound where the gains α and β are
20 reduced.
A
10
Fig. 6 shows an experimental result with the variable gains.
The gains are set to α0 = 30 ; aα = 8 ; β0 = 0.1 ; aβ =
0
2 4 6 8 10 12 14 16 0. Indeed, the response time is equal to 0.8 ms, the current
−3
x 10 dynamic is high without overshoot and chattering phenomenon
(b) Average phase voltage
at steady state.
20
40
−20 30
2 4 6 8 10 12 14 16
20
A
t (s) −3
x 10
10
20
(a) Reference (− −) and measured (−) one phase currents
40
0
V
30
−20
20
A
2 4 6 8 10 12 14 16
10
t (s) −3
x 10
0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016
0
V
B. Robustness tests
−20
Experimental tests have been performed to show the influ-
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016
t (s)
ence of the speed on the controller’s performance. Speed is
set to 1000 rpm. In this test, the torque has been reduced to
Fig. 5. Experimental results : instantaneous currents and average phase maintain the current control such that the back-emf is lower
voltage - Sliding mode controller : α = 9 ; β = 0.1. than Udc . Fig. 7 shows the good robustness of this sliding
mode controller against a speed variation, and this because of
To reduce the response time, the gain α must be increase. the back-emf compensation.
Fig. 5 shows an experimental result for α = 9 and β = 0.1, Indeed, Fig. 8 shows the current response when the con-
where chattering phenomenon on the average phase voltage troller is not provided with a back-emf compensation. The
appear, thereby generating additional current ripples. Now, the increase of the back-emf results in a significant error relative to
3304
5
(a) Reference (− −) and measured (−) one phase currents (a) Reference (− −) and measured one phase currents
30 40
i1*
i1
30
20 i1 (L+30%)
i1 (L−30%)
20
A
A
10 10
0 0
2 4 6 8 10 12 14 16
4 5 6 7 8 9 10 11 12 13 −3
−3 x 10
x 10 (b) Average phase voltages
(b) Average phase voltage 30
U1
20 U1 (L+30%)
20
10 U1 (L−30%)
0
V
0
V
−10
−20
−20
−30
2 4 6 8 10 12 14 16
4 5 6 7 8 9 10 11 12 13 t (s) x 10
−3
t (s) −3
x 10
V. C ONCLUSION
the current reference. Therefore, a compensation of the back-
In this paper, a new current controller for the SRM is
emf improve the current control.
proposed. It is a second order sliding mode using the super
twisting algorithm. It has good performances, high dynamics
like the hysteresis controller and low current ripples with fixed
(a) Reference (− −) and measured (−) one phase currents
30 switching frequency operation (PWM). Moreover, it is char-
20 acterized by its good robustness properties against parametric
A
TABLE I
−20 P ROTOTYPE CHARACTERISTICS
1 2 3 4 5 6 7 8 9 10
x 10
−3 Geometric parameters
(c) Back−emf Number of rotor poles 6 Stator pole arc 19.8
30 Number of stator poles 8 Rotor pole arc 20.65
20 Stator outer diameter 143 mm Airgap lebgth 0.8 mm
V
10 Shaft diameter 23 mm
0
1 2 3 4 5 6 7 8 9 10
Electrical parameters
t (s)
x 10
−3 Number of phases 4 Nominal speed 3000
rpm
Nominal power 1.2 kW Nominal voltage 24 V
Phase’s resistance 50 mΩ
Fig. 8. Experimental results : instantaneous currents and average phase
voltage at 1000 rpm without back-emf compensation.
3305
6
3306