Академический Документы
Профессиональный Документы
Культура Документы
Abstract—Direct Torque Control (DTC) is considered as one there are uncertainties existing in practical applications of
of the latest and most efficient techniques that can be used for the LIM which are usually composed of unpredictable plant
the speed and/or position tracking control problem of induction parameter variations, external load disturbances, and
motor drives. However, the main drawbacks of classical DTC unmodeled and nonlinear dynamics. Therefore, the LIM
are the variable switching frequency that could exceed the
drive system must provide high tracking performance, and
maximum allowable switching frequency of inverters and also
the ripples it has over the current and torque, especially at low high dynamic stiffness to overcome the above difficulties
speed tracking. It has been shown that applying Model [1], [5], [6].
Predictive Control (MPC) to a Linear Induction Motor (LIM) Modern control techniques have been used to control the
leads to a much better speed tracking performance. MPC speed and/or position of the induction motor drives. Among
provides the optimal 3-phase primary voltages necessary for these modern control techniques, the method of Direct
speed tracking using a Pulse Width Modulation (PWM)
Torque Control (DTC) is considered as one of the latest and
inverter. The main inherent drawbacks of the MPC strategy
are its high switching frequency and also its heavy most efficient techniques that can be used for induction
computational load which makes it inapplicable in real-time. motors [7], [9]. The basic characteristic of DTC is that the
This paper presents a new analytical approach based on the positions of the inverter switches are directly determined.
MPC strategy. The new analytical approach controls directly The advantages of the DTC strategy are fast transient
the inverter switches. Hence the PWM inverter is not needed. It response, simple configuration, and low parameter
computes the optimal position transitions sequence of the
dependence. However, the classical DTC has inherent
inverter switches to track the speed reference trajectory. The
proposed analytical nonlinear MPC controller includes an drawbacks such as variable switching frequency, high
integral action to reduce the steady state error. The proposed torque and current ripples, high noise level at low speeds
controller admits real-time implementation. Simulation results and also the difficulty to accurately control torque and flux
show that the new analytical approach has good tracking at low speeds.
properties at the same time as it reduces the average inverter Model predictive control has been applied to LIM drives
switching frequency by 93 % as compared to classical DTC.
for tracking the speed reference trajectory [14]. Based on a
linearized model of the LIM, the MPC controller calculates
I. INTRODUCTION
the optimal primary voltages while respecting the given
1940
⎡1 − 1 / 2 − 1 / 2⎤ ⎡ u1 ⎤ where ŷ is the predicted future output, w is the reference
⎡Vαs ⎤ ⎢ ⎥
⎢V ⎥ = Vdc
⎢ 3 3 ⎥ ⎢u 2 ⎥ (9) for ŷ , u is the control signal, and where Q, Pj are positive
⎣ βs ⎦ ⎢ 0 − ⎥
⎣ 2 2 ⎦ ⎢⎣u3 ⎥⎦ definite weighting matrices. This function is minimized
+Vdc/2
subject to constraints that describe the discretized dynamics
in (1)-(9).
u1=1 is
The Pj matrix should impose more penalties over the
is
LIM first time-steps than the later steps, to force the transition of
u2=0 is
u3=1 the switches to occur as late as possible. This strategy is
-Vdc/2 implemented by the following relation:
Figure 1: Three-phase inverter driving the LIM. P0 f P1 f L f PNu −1 . (12)
where Pj f Pj +1 mean that Pj − Pj +1 is positive definite.
II. ANALYTICAL NMPC CONTROLLER
To reduce the steady state error, an additional state
A. Principal ideas of Model Predictive Control xint (k ) representing integration of the control error is
The main idea of MPC is to use a model of the plant to defined as follows:
predict future outputs of the system. Based on this xint (k + 1) = xint (k ) + kint (w(k ) − y (k ) ) (13)
prediction, at each sampling period, a sequence of future
control values is computed through an on-line optimization where y is the measured output, w is the reference for y
process, which maximizes the tracking performance while and where k int is the integrator gain. To reduce the steady
satisfying constraints. Only the first value of this optimal state error, an additional penalty term including the
sequence is applied to the plant, and the whole procedure is integrator state is added to the cost function. Thus the
repeated again at the next sampling period according to what objective function (11) is redefined as:
usually is called a ‘receding’ horizon strategy [4].
∑ (yˆ (k + j k ) − w(k + j ) ) Q(yˆ (k + j k ) − w(k + j ) ) +
N T
J=
B. Analytical Nonlinear MPC Controller j =1
1941
to linearize the LIM model as done in [14]. Moreover there N u = 1 , prediction interval = 10 × Ts . The concept of
is no need to reformulate the system in the hybrid system multiple discrete models is used to reduce the number of
framework; neither as a PWA model nor as an MLD model prediction steps; first a model with sampling time Ts is used
as done in [15]. Of course operating point changes are for the first two steps, and then a model with sampling time
allowed and taken into account. equal 4Ts is used for 2 steps (i.e. the prediction interval
The developed technique still gives the good performance 10Ts is covered by 4 prediction steps). Weights on
of classical MPC [14], but in addition to that, it reduces
manipulated variable rates P = 1 , weights on the output
significantly the computational time. Moreover, one extra
signals Q = diag (0;0;0;0;1000000 ) , and sampling time
dimension of freedom through the choice of the weighting
Ts = 100 μs . Integrator parameters are: kint = 200,
matrix Pj has been added, which enables a trade-off
Pint = 0.1 .
between the average switching frequency and the speed The weights on the output currents and fluxes are set to
tracking performance. Note that reducing the torque ripple zero because the reference trajectories of those signals are
can only be achieved by increasing the switching frequency unknown, and thus they have no effect on the objective
and vice versa [16]. function J .
1942
For this simulation, the maximum number of transitions 0.4
Load Change
of the inverter switches for the classical DTC approach was Speed Response
0.2
Speed
0
25764/s, while for the analytical approach it was only -0.2 Speed Reference
Fe
500
3 20
Load Change
iabc
0
-20
2 -40
0.4 0.42 0.44 0.46 0.48 0.5 0.52 0.54 0.56 0.58 0.6
Time (s)
Analytical NMPC
Response
1 Figure 4. Simulation results obtained with the Analytical NMPC controller
at low speed.
Speed Reference
The choice of the weighting matrix on the manipulated
Speed (m/s)
with the DTC technique for such low speed. The maximum
switching frequency in this case was 1315/s with a reduction -0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)
of 93.48 % as compared to the classical DTC for the same
Figure 5. Effect of the penalty matrix P on the tracking performance.
scenario.
The robustness of the analytical NMPC controller against TABLE 2: EFFECTS OF PENALTIES ON SWITCHING FREQUENCY
Weighting Penalty Max. switching frequency / s.
parameter variations was examined. The primary resistance
P= 0.01 1096/s
detuning is considered in this test because it has significant
P= 1.0 1056/s (shown in Figure 5)
effect on the flux estimation especially at low speeds. The P= 100 852/s
value of the primary resistance Rs has been increased by P= 10000 654/s (shown in Figure 5)
50% in the LIM model, while it is kept constant at its
All constraints on fluxes and currents are satisfied in the
nominal value in the controller. As it was the case for
previously investigated scenarios as shown for example in
classical MPC [14], the proposed controller was very robust
Figure 6, where typical constraints are: secondary flux less
against this variation and almost there is no effect on the
than 0.45 and primary current value less than 50 A.
speed response.
1943
Lamda(alfa & Beta) 0.2
Load Change
classical DTC approach, and robustness against load
0.1 changes and parameter variations. With this technique there
0 is no need to use a PWM inverter, and moreover, it reduces
-0.1 significantly the computational time, which is the inherent
-0.2
0.35 0.4 0.45 0.5 0.55 0.6 0.65 drawback of the classical MPC controller. Thus a real time
implementation is possible.
40
20
Future work will include experimental works to validate
this technique practically. Finally, the same technique will
iabc
-20
be examined for other machines like the rotary induction
-40
motor, and the permanent magnet synchronous motor.
0.35 0.4 0.45 0.5 0.55 0.6 0.65
Time (s)
Figure 6. The secondary flux and primary currents are within the constraints REFERENCES
[1] Abdou G.H., and S. A. Sherif, (1991). Theoritical and experimental
A. Complexity analysis design of LIM in automated manufacturing systems. IEEE Trans.
The computational time is mainly affected by the control Indust. Appl, Vol. 27, No.2, pp.286-293.
[2] Boldea I., and S. A. Nasar, (1997). Linear electric actuators and
horizon N u , where the objective function is evaluated generators. Cambridge University Press, UK.
(6 Nu ) times at each sampling step. The proposed techniques [3] Bucci G., S. Meo, A. Ometto, and M. Scarano, (1994). The control of
LIM by a generalization of standard vector techniques, Proc. of IEEE
with N u = 1 and N = 10 (one discrete model) takes in
IAS, pp. 623-626.
average 140 μs . Applying multiple model concepts (10 [4] Dumur D. and P. Boucher, (1998). A Review Introduction to Linear
prediction interval covered by 4 steps) reduces the GPC and Applications. Journal A, 39(4), pp. 21-35.
[5] Gastli A., (1998). Compensation for the effect of joints in the
computation time to 70 μs (faster than the sampling time), secondary conductors of a linear induction motor. IEEE Trans. On
all previous results ware done with those horizon values. Energy Conversion, Vol. 13, No.2, pp. 111-116.
With control horizon N u = 2 (10 intervals – 4 steps), it [6] Gastli A., (2000). Asymmetrical constants and effect of joints in the
secondary conductors of a linear induction motor. IEEE Trans. On
takes 340 μs . All the simulations are run in Matlab 7.8 on a Energy Conversion, Vol. 15, No.3, pp. 251-256.
3 MHz PC with 2 Mram. [7] Habetler T. G., F. Profumo, M. Pastorelli, and L. M. Tolbert, (1992).
Needless to say, for the classical MPC controller the Direct torque control of induction machines using space vector
modulation. IEEE Trans. Ind. Application, Vol. 28, pp.1045-1053.
computational time because of on-line linearization and [8] ILOG, Inc., (2000). CPLEX 7.0 User Manual. Mountain View, CA,
optimization will be substantially higher, even if commercial USA; www.ilog.com.
optimization software such as CPLEX [8] is used. This [9] Lascu C., I. Boldea, and F. Blaabjerg, (2000). A modified direct
torque control of induction motor sensorless drive. IEEE Trans. Ind.
makes it impractical to implement in real time. Application, Vol. 36, pp.122-130.
Previous simulation results prove the success of the [10] Lin F.-J., P.-H. Shen, and S.-P. Hsu, (2002). Adaptive backstepping
presented technique; it has a good performance on speed sliding mode control for linear induction motor drive. IEE Proc.
Electr. Power Appl., Vol. 149, No.3, pp. 184-194.
tracking at high speed as well as at low speed. It is robust [11] Lin F.-J., and Rong-Jong Wai, (2002). Robust control using neural
against load changes and parameter variations. It is also network uncertainty observer for linear induction motor servodrive.
succeeded in reducing the average switching frequency. IEEE Trans. On power Electronics, Vol. 17, No.2, pp.241-251.
[12] Takahashi I., and Y. Ide, (1993). Decoupling control of thrust and
Moreover, the proposed controller reduces significantly the
attractive force of a LIM using a space vector control inverter. IEEE
computational time as compared to classical MPC Trans. Indust. Appl, Vol. 29, No.1, pp.161-167.
approaches, which makes it applicable in real time [13] Zhang Z., T. R Eastham, and G.E. Dawson, (1993). Peak thrust
application. operation of linear induction machines from parameter identification.
Proc. of IEEE IAS, pp. 375-379.
[14] Hassan A. and J. Thomas (2008). Model Predictive Control of Linear
V. CONCLUSIONS Induction Motor Drive. Proceedings of the 17th IFAC World
Congress, Vol. 17 Part 1, COEX, Seoul, South Korea.
This paper considers the speed tracking control problem [15] Geyer T. and G. Papafotiou (2005). Direct Torque Control for
of a linear induction motor. It presents a new analytical Induction Motor Drives: A Model Predictive Control Approach Based
NMPC controller based on the model predictive control on Feasibility. Hybrid Systems: Computation and Control, LNCS
3414, pp. 274-290.
approach. The developed controller controls directly the [16] Papafotiou G., T. Geyer and M. Morari (2004). Optimal Direct Torque
inverter switches to track the speed trajectory of the linear Control of three-phase symmetric Induction Motors. In Proceedings of
induction motor drive. The controller succeeds in tracking the 43th IEEE Conference on Decision and Control, Atlantis,
Bahamas.
the speed trajectory at high and low speed and it reduces the
[17] Itoh K. and H. Kubota (2005). “Thrust ripple reduction of linear
switching frequency in the range of about 93% as compared induction motor with direct torque control,” Proceedings of the Eighth
with classical DCT, so that inverters with a limited International Conference on Electrical Machines and Systems, ICEMS
switching frequency can be used. 2005, vol. 1, pp. 655-658, 2005.
[18] Marcos Vinicius Lazarini and Ernesto Ruppert Filho (2007).
The proposed MPC controller response has many Sensorless three-phase induction motor direct torque control using
advantages; besides being simple to construct and to sliding mode control strategy laboratory set-up for motor speed
implement, it has a very fast response, lower ripples over control teaching. International Conference on Engineering Education
– ICEE 2007.
currents and electromagnetic force in comparison to the
1944