Application of Artificial Neural
Network to Robust Speed Control
of Servodrive
Tomasz Pajchrowski and Krzysztof Zawirski, Senior Member, IEEE
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS,
VOL. 54, NO. 1, FEBRUARY 2007
PPT製作:100%
指導教授 : 龔應時
學
生 : 顏志男
Outline







Abstract
INTRODUCTION
CONCEPT OF THE ROBUST SPEED CONTROLLER
SIMULATION RESULTS
EXPERIMENT RESULTS
CONCLUSION
REFERENCES
2
Abstract

This paper deals with the problem of robust speed control of electrical
servodrives.

A robust speed controller is developed using an artificial neural network
(ANN), which creates a nonlinear characteristic of controller. An original
method of neural controller synthesis is presented.

The synthesis procedure is performed in two stages. The first stage consists
in training the ANN and at the second stage controller settings are adjusted.
3
INTRODUCTION(1/5)

One of the well-known methods to achieve robustness is the sliding mode
control (SMC) [1], [6], [12], [13].

However, SMC creates some problems in practical applications due to
chattering effects, such as high-frequency oscillation of motor torque,
additional mechanical stresses, and noise during operation.

Many solutions have been proposed [1]–[3], [5], [6],[12], [13] to
circumvent the problem mentioned above, among them one of the best is a
continuous SMC [6], [12], [13].
4
INTRODUCTION(2/5)

The problem of servodrive control robustness seems to be the actual
research task.

In this filed, interesting solutions can be obtained by the application of
computational intelligence methods like fuzzy logic control (FLC) or
artificial neural network (ANN) control [2],[8]

In this paper, an ANN controller is proposed which due to its specific
feature and training ability enables an elaborate transparent synthesis
procedure.
5
INTRODUCTION(3/5)

The robust speed controller proposed in this paper is based on the
introduction of ANN into the structure of the PI speed controller[7].

Robustness is achieved by creating proper nonlinear characteristics of the
controller by ANN.

The original feature of this synthesis procedure consists of its
decomposition into two stages and consideration of declared range of
parameter variation in training procedure and controller settings adjustment.
6
INTRODUCTION(4/5)

The first stage consists in training the ANN to form the proper shape of the
control surface, which represents the nonlinear characteristic of the
controller.

At the second stage, the PI controller settings are adjusted by means of the
random weight change (RWC) procedure [4] which optimizes the control
quality index formulated in this paper.

Fig. 1 presents the structure of the servodrive control system.
7
INTRODUCTION(5/5)
8
CONCEPT OF THE ROBUST SPEED
CONTROLLER(1/10)
A.Controller Structure

The proposed robust speed controller consists of three blocks:a module
of input signal conversion, an ANN module, and a module of output
signal generation.

The controller structure is shown in Fig. 2.
9
CONCEPT OF THE ROBUST SPEED
CONTROLLER(2/10)
A.Controller Structure

The process of neural controller synthesis is decomposed into two stages:(1)
synthesis of an ANN (selecting structure and weight coefficients in
activation functions);(2) adjusting PI controller settings.

The structure shown in Fig. 3 was obtained by trial and error during
numerous simulation tests.
10
CONCEPT OF THE ROBUST SPEED
CONTROLLER(3/10)
A.Controller Structure
11
CONCEPT OF THE ROBUST SPEED
CONTROLLER(4/10)
B.Training Procedure

Weight coefficients in the activation function are selected during the
process of training the ANN.

For the training, a pattern speed response of the controlled system is
assumed.

This pattern speed response [Fig. 4(a)] is created as a step response of a
simplified model of the servodrive speed control system presented in Fig.
4(b).
12
CONCEPT OF THE ROBUST SPEED
CONTROLLER(5/10)
B.Training Procedure
13
CONCEPT OF THE ROBUST SPEED
CONTROLLER(6/10)
B.Training Procedure

For the block diagram presented in Fig. 4(b), one can introduce an
equivalent gain coefficient kω, which is described as

where kωmax and kωmix are the assumed maximum and minimum values
of the variable parameter
14

CONCEPT OF THE ROBUST SPEED
CONTROLLER(7/10)
B.Training Procedure

The pattern waveform of the transient process is divided into 200 time steps.
At each step(k) the speed value and its derivative is calculated, which
results in a pair of input signals e(k),de/dt(k) for the ANN.

The required output controller signal, the value diq/dt(k), is calculated from
(3) to (5) where JΣ/Ko is equal to 1/kω according to (1).
15
CONCEPT OF THE ROBUST SPEED
CONTROLLER(8/10)
B.Training Procedure

The learning error QL,determined from (6), reaches a smaller value than the
arbitrary selected acceptable value
16
CONCEPT OF THE ROBUST SPEED
CONTROLLER(9/10)
C. Controller Settings Adjustment

At the second stage of the controller synthesis, an adjustment of two
controller settings: the gain coefficient kp and the integral time constant Ti
is performed.

The criterion Qc (quality index) assumed for controller optimization,
described by formula (10), contains a sum of three components
17
CONCEPT OF THE ROBUST SPEED
CONTROLLER(10/10)
C. Controller Settings Adjustment

All three components characterize the step response of speed control,
which is shown in Fig. 5.
18
SIMULATION RESULTS(1/10)
A. Model Description

The simulation technique was used for training the ANN and testing the
control system operation. As mentioned above, the PMSM was assumed as
a good representative of servodirve.

The model is written in the d-q rotating frame
19
SIMULATION RESULTS(2/10)
B. Influence of Quality Index Components on Controller Operation

The model of the PMSM and the control system was performed in
MATLAB. Data of the investigated drive are given in Table III.

These tests were performed for different values of kω. For simplicity of
PVR determination, only variation of the total moment of inertia was
assumed.

The simulation results are presented in Figs. 6–8.
20
SIMULATION RESULTS(3/10)
B. Influence of Quality Index Components on Controller Operation
21
SIMULATION RESULTS(4/10)
B. Influence of Quality Index Components on Controller Operation
22
SIMULATION RESULTS(5/10)
B. Influence of Quality Index Components on Controller Operation
23
SIMULATION RESULTS(6/10)
B. Influence of Quality Index Components on Controller Operation

Such analysis can be made on the basis of results collected in Table I. For
transient process, the same is presented in Figs. 6–8, the ISE(Qo) value was
measured for different inertia.

The value of ISE calculated in reference to its assumed maximum value
using formula (12) is treated as an index(Qo) which helps to evaluate
obtained robustness
24
SIMULATION RESULTS(7/10)
B. Influence of Quality Index Components on Controller Operation
25
SIMULATION RESULTS(8/10)
C. Influence of PVR Value

The proposed method gives a chance to make a synthesis of the robust
speed controller with assumption of different range of drive parameter
variation determined by the value of PVR.

To analyze the influence of assumed different values of PVR on control
properties some additional test were done.

The robust controller settings were adjusted during synthesis process for
following values of PVR: 2, 8, and 32. Figs. 9
26
SIMULATION RESULTS(9/10)
C. Influence of PVR Value
27
SIMULATION RESULTS(10/10)
C. Influence of PVR Value

This effect is better visible comparing values of index Qo and settling time
collected in Table II.
28
EXPERIMENT RESULTS(1/4)

The laboratory setup consists of a PMSM supplied from a PWM transistor
inverter and controlled by the floating-point signal processor ADSP–21061.

Some results of the laboratory tests are shown in Figs. 10 and 11. During
the tests, the total value of the moment of inertia of the drive was changed
from small to large.
29
EXPERIMENT RESULTS(2/4)
30
EXPERIMENT RESULTS(3/4)
31
EXPERIMENT RESULTS(4/4)
32
CONCLUSION

The presented investigation results confirm that the proposed speed
controller of PI type with proper nonlinear characteristics allows to obtain
the robustness of speed control against the variations of the moment of
inertia and torque constant.

These nonlinear controller characteristics were effectively achieved using
an ANN application.

Finally, it must be remarked that the cost of obtained robustness is visible
in the reduction of dynamic properties of the drive.
33
REFERENCES(1/2)








[1] H. Z. Akpolat, G. M. Asher, and J. C. Clare, “A practical approach to the design of robust
speed controllers for machine drives,” IEEE Trans. Ind. Electron., vol. 47, pp. 315–324, Apr.
2000.
[2] J. Arellano-Padilla, G. M. Asher, and M. Sumner, “Robust fuzzysliding mode control for
motor drives operating with variable loads and predefined system noise limits system,” in
Proc. 9th Eur. Conf. Power Electron. Appl., Graz, Austria, Aug. 27–29, 2001.
[3] S. Brock, J. Deskur, and K. Zawirski, “Modified sliding mode speed controller for servo
drives,” in Proc. IEEE Int. Symp. Ind. Electron., Bled-Slovenia, 1999, pp. 635–640.
[4] B. Burton, F. Karman, R. G. Harley, T. G. Habetler,M. A. Brooke, and R. Poddar,
“Identification and control of induction motor station currents using fast on line random
training neural networks,” IEEE Trans. Ind. Appl., vol. 33, no. 3, pp. 697–704, Jun. 1997.
[5] S. K. Chung, J. H. Lee, J. S. Ko, and M. J. Youn, “Robust speed control of brushless
direct-drive motor using integral variable structure control,”Proc. IEEE Electron. Power Appl.,
vol. 142, no. 6, Nov. 1995.
[6] K. Erbatur, O. M. Kaynak, and A. Sabanovic, “A study on robustness property of slidingmode controllers: A novel design and experimental investigations,” IEEE Trans. Ind.
Electron., vol. 46, no. 5, pp.1012–1017, Oct. 1999.
[7] M. Norgaard, O. Ravn, N. K. Poulsen, and L. K. Hansen, Neural Networks for Modelling
and Control of Dynamic Systems. Berlin, Germany: Springer-Verlag, 2000.
[8] T. Orlowska-Kowalska and K. Szabat, “Optimization of fuzzy logic speed controller for
DC drive system with elastic joints,” IEEE Trans. Ind. Appl., vol. 40, no. 4, pp. 1138–144,
Jul./Aug. 2004.
34
REFERENCES(2/2)








[9] T. Pajchrowski, K. Urbanski, and K. Zawirski, “Robust speed control of PMSM servodrive
based on ANN application,” in Proc. 10th Eur.Conf. Power Electron. Appl., Toulouse, France,
Sep. 2–4, 2003.
[10] T. Pajchrowski and K. Zawirski, “Robust speed control of PMSM with neuro and fuzzy
technique application,” in Proc. 12th Int. Power Electron.Motion Control Conf., Ryga, Sep.
2–4, 2004.
[11] ——, “Robust speed control of servodrive based on ANN,” in Proc. Int. Symp. Ind.
Electron., Dubrovnik, Croatia, Jun. 20–23, 2005.
[12] M. H. Park and K. S. Kim, “Chattering reduction in the position control of induction
motor using the sliding mode,” IEEE Trans. Power Electron.,vol. 6, Jul. 1991.
[13] M. Rodic, A. Sabanovic, and K. Jezernik, “Sliding mode observer scheme for induction
motor control,” EPE Journal, vol. 11, no. 3, pp.29–34, Aug. 2001.
[14] B. Shamin and M. Hassul, Control System Design Using Matlab. Englewood Cliffs, NJ:
Prentice-Hall, 1993.
[15] P. Vas, Sensorless Vector and Direct Torque Control. London, U.K.: Oxford Univ. Press,
1998.
[16] J. Vitek, T. Baculak, S. J. Dodds, and R. Perryman, “Near-time-optimal control of
electrical drives with permanent magnet synchronous motor,” Elect. Power Electron., 2003,
Toulouse, CD-Rom.
35