You are on page 1of 6

CURRENT LIMITATION IN THE ADAPTIVE NEURAL SPEED CONTROL OF A DC MOTOR

M Minkova', D Minkov2, J L Rodgerson' and R G Harley' 1- Department of Electrical Engineering, 2- Department of Electronic Engineering University of Natal, Private Bag X10 Dalbridge, 4014, South Africa Corresponding author: Dr.D .Minkov: email:dminko@elaine.ee.und.ac.za
ABSTRACT
A simple design is proposed for a current limitation in the adaptive neural speed control of a DC motor. The design is based on using a current feedback from the output of the motor to the output of the ANN, and a simple logical operation between the end of the current loop and the input of the motor. Simulations are made of the performance of the controller with a current limitation and without a current limitation, for different types of reference speed trajectories. The control simulations are made for a DC motor with given parameters, and white noise is superimposed on both the rotor speed and the armature current at the output of the motor.

o(k+l)=A,o(k)+A,o(k-l)+ (3) +& ( k ,k- 1 ) +A4V( k) i(k)=A,w(k)+A,o(k-i)+ +A,( k- 1 ) +A8V( k)

(4)

where k indicates the k-th discrete time moment, A,, A,, A,, A,, A,, A,, A,, A,,, and A,, are real constants, and A,, A,, and A,, are real parameters which depend on the load of the motor. There are two approaches for application of artificial neural networks (ANNs) for plant control. For certain uncomplicated plants, it is possible to train the ANN off-line, before its use in non-adaptive controller. In this case, pre-control training of the ANN is performed, but the connection weights of the ANN are not updated during the control action. In the alternative approach, the connection weights of the ANN are updated on-line, during the control action, and the controller is adaptive [2]. The non-adaptive neural control of a DC motor has been investigated by ElSharkawi et a1 [1,3,4]. An adaptive neural speed controller of a DC motor has been proposed by Buja et a1 [SI. It is assumed in [5] that the motor parameters K and J are known, but determination of these parameters is not indicated. An adaptive neural speed controller of a DC motor which does not require knowledge of any motor/load parameters has been proposed by Minkova et a1 [6]. The idea for the design of this adaptive speed controller of a DC motor is based on the work of Henaff and Milgram [7] for adaptive robot control by a feedforward ANN trained by the online backpropagation algorithm. It can be deduced from [7] that in cases where the desired output of the ANN is not known, the error at the output of the subsequent

INTRODUCTION
The dynamics of a DC motor is described by the following equations [l]:

Ko(i)

-Ri(t)-L-+v(i) di

di( t)

(1)

Ki( i )

J d o ( f )+ D o (t)+ T,( t) dt

(2)

- rotor speed (rad/s) - armature resistance ( Q ) - armature inductance (H) - armature voltage (V) - armature current (A) - load torque (N.m) - rotor inertia (kg.m2) - torque and back emf constant (Vdrad) - viscous friction coefficient (N.m.s/rad)
In order to digitally control a plant, a discrete time model of the plant is required. The following discrete time models of a DC motor can be derived from eqn.(l) and eqn.(2):

.--r
0-7803-3019-6/96/$3.00 1996 IEEE. 6

837

plant can be backpropagated through the ANN for updating its connection weights, The ANN for the adaptive controller from [6] is a feedforward ANN with one hidden layer, three inputs & @ + I ) , U@), and U@-I), and one output OR). The ANN has three input nodes, nine hidden nodes, one output node, unit offset nodes in the input layer and the hidden layer, and activation of the hidden nodes and the output node by the sigmoidal activation function F. The output x(k) after the activation of the output node lies within the interval [0.4,0.6/ with a length li=O.2, and is mapped onto the ANN output OR)which lies within the interval [-vr,v J by the following linear transformation:

according to the theory for the ANN training by the online backpropagation algorithm [8]. The connection weights of the ANN are updated once at the start of every control step for only one training epoch. The adaptive speed controller of a DC motor proposed in [6] uses the following second order reference model:

ad(k+ 1 ) = 0.6ad( ) + k +0.2G)#( - 1 ) + r ( k) k

(11)

where v, is the rated armature voltage. The inverse mapping of the ANN output OR) onto the output x(k) after the activation of the output node is performed by the transformation:

where r@) is the input to the reference model, and the rotor speed of the DC motor has to follow the known reference speed trajec:tory wJk) defined in the time interval [0, td. The rotor speed is predicted in the &+I)-th moment as :

& ( k + l )= 0 . 6 a ( k ) + + 0 . 2 9 ( k - l )+ r ( k )

(12)

which is used during the ANN training. The ANN for the controller from [6] is trained during the control process by the online backpropagation algorithm. The initial values of the connection weights of the ANN are randomly distributed in the interval [-0. I , 0.I ] . The update A W of the connection weights ~ wij of the ANN is:

and 0.6w(k)+0.2w(k-I)=P. The ANN has three inputs: & @ + I ) , a ) and u k - l ) , and one output OR). The @, armature voltage O(k) is estimated by the ANN as:

O( k) = N [ &l 1 ) ,a(k),a(k-1 )] (k+

( 13)

where 6; is the error signal for the i-th node at the k-th discrete time moment, the learning rate is 7=0.25, and a is the momentum parameter. The concept of Henaff and Milgram [7] and its application to the controller from [6] lead to the : following expression for the error signal 6 for the output node of the ANN:

where N[4 (k+ I ) , w (k), w @-I)] is the output of the ANN, in accordance with the discrete time model of the motor introduced by eqn.(5). The motor is controlled by the armature voltage O(k), and the actual rotor speed w(k) is sampled at the output of the motor. The input to the reference model r(k) and the actual rotor speed w(k) at the output of the motor are sampled for the same sampling period T in the discrete time moment t=kT. It is shown in [6] how to optimise the sampling period T for achievement of the best performance of the controller for a particular reference speed trajectory. It is known that an armature current surge could develop in the armature circuit of a DC motor even for relatively slow motor dynamics 191. Considering that a current surge could damage the motor, some current limitation has to be performed during the speed control of a DC motor. The importance of this problem increases when the control of the motor is adaptive, in which case a small inaccuracy of the control voltage could result in a large armature current. Nevertheless, the problem for current limitation in adaptive neural speed controllers off a DC motor has not been investigated.

where U , @ is the known reference speed in the k-th ,-) discrete time moment. The error signal 6; for a hidden node of the ANN is:

0-7803-3019-6/%/$3.001996 IEEE. Q

83 8

In this paper a current limitation is proposed for the adaptive neural speed controller of a DC motor developed by Minkova et a1 [6]. The performance of the controller with a current limitation is simulated for different reference speed trajectories. A comparison is made between the armature current of the motor for the controller with a current limitation and the controller without a current limitation.
All the computations in this paper are made for a DC motor with the following parameters: R = 1.2 Q L=25mH J = 0.208 kg.m2 K = 1.21 V.s/rad D = 0.008 N.m.s/rad (14) T, = 0 i.e the DC motor is used without a load v, = 240 V P, = 5 kW i, = P,/v, = 20.83 A U , = 1800 rev/min = 188.5 rad/s where v, is the rated voltage, P, is the rated power, i, is the rated armature current, and U, is the rated rotor speed. CURRENT LIMITATION IN THE ADAPTIVE SPEED CONTROLLER The computed rotor speed trajectory and armature current trajectory are shown in Fig.1 for a particular armature voltage trajectory applied to the DC motor considered. It is seen that the changes of the armature voltage over the time result in fast changes of the armature current. This indicates that if the armature current at the output of the motor reaches too large amplitude in a given discrete time moment, then the armature current amplitude could be limited in the following moment by applying of an armature voltage with the same direction and with a slightly smaller amplitude, compared to the moment considered, to the input of the motor. The topology of the adaptive neural speed controller of a DC motor, proposed by Minkova et a1 [6], and equipped with a current limitation feedback is given in Fig.2. Fig.3 shows a flowchart of the controller with current limitation. The current limitation in this controller of a DC motor is performed by implementation of the following logical operation:

where i,w is the maximum allowed amplitude of the armature current, i.e. ,i gives the current limitation. Eqn.(l5) shows that if the amplitude of the armature current at step k-I becomes larger than 80%i,, then the armature voltage applied to the input of the motor in the following step k has a prescribed value of 98% of its value at step k-I . It turns out that such a decrease of the amplitude of the armature voltage for such amplitudes of the armature current limits the amplitude of the armature current below iM independently from the reference speed trajectory. If the amplitude of the armature current is less than 80 %,i then the armature voltage is supplied from the output of the ANN to the input of the motor. The topology of the ANN used in the controller is given in Fig.4. The choice of the parameters y,a,and Zi of the ANN training depends on the reference speed trajectory and the DC motor parameters, and is discussed in [ 6 ] .It should be noted that the ANN of the controller does not require any preferential training before the beginning of the control process, and the controller does not require any knowledge of the motor/load parameters. COMPUTED RESULTS Control processes are simulated for the DC motor with parameters introduced by eqn.( 14). White noise with a magnitude of 0.5 rad/s is superimposed on the rotor speed at the output of the motor, and white noise with a magnitude of 0.1 A is superimposed on the armature current at the output of the motor in all control simulations. The performance of the controller with current limitation of iM=5 A is illustrated in Fig.5 for a sinusoidal type of reference speed trajectory RSTl which leads to slow motor dynamics. The corresponding variation of armature current as a function of the time is shown in Fig.6. The variation of the armature current appears in Fig.7 for the corresponding controller without current limitation (the controller from [6]), and the reference speed trajectory RSTl. The comparison between Fig.6 and Fig.7 shows that the armature current reaches an amplitude of approximately 9 A for the controller without current limitation, and I i(t)I < iM=5 A for the controller with current limitation. Fig.8 illustrates the performance of the controller with current limitation of i 12 A for another sinusoidal , = type of reference speed trajectory RST2 which leads to fast motor dynamics. The change of the armature current with the time is shown in Fig.9 for the same control simulation. Fig. 10 gives the dependence of the

0-7803-3019-6/96/S3.00 0 1996 IEEE.

839

armature current as a function of the time for the corresponding controller without current limitation, and the reference speed trajectory RS72. The comparison between Fig.9 and Fig. 10 indicates that the introduction of current control decreases the amplitude of the armature current from 1 i(t)I = 45 A for the controller without current limitation to 1 i(t) I < iM=12 A for the controller with current limitation. The performance of the controller with current limitation of i,=4 A is illustrated in Fig.11 for a sigmoidal type of reference speed trajectory RST3. The corresponding change of the armature current with the time is given in Fig.12. The comparison between Fig.5, Fig.8, and Fig.11 shows that the controller with current limitation has a good performance for all the reference speed trajectories considered. CONCLUSIONS The fast changes of the armature current of a DC motor and the use of an adaptive neural speed controller lead to a significant amplitude and noise of the armature current of the motor. It indicates that current limitation has to be performed during the control process. A design is proposed which limits the amplitude of the armature current of a DC motor below a prescribed value during the adaptive neural speed control of the motor. The current limitation is achieved by using of a current feedback from the output of the motor to the output of the ANN, and a simple logical operation between the end of the current feedback and the input of the motor. The current limitation functions well, and the controller with current limitation has a good performance, independently of the speed of the motor dynamics. ACKNOWLEDGEMENT The financial support of the Foundation for Research and Development F R D is greatfully acknowledged.

trajectory controller for a DC motor, ZEEE Tranr.Energy Conversion, 1993, 8, pp. 107-113. El-Sharkawi; M A and El-Samahy, A A: High performance drive of DC brushless motors using neural network, IEEE Trans.Energy Conversion, 1994, 9, pp.317-322. Bertoluzzo, M; Buja, G S and Todesco, F: Neural network adaptive control of a DC drive. Proceedings of the Conference IEEE Industrial Electronics Society, Italy, 1994, pp. 1232-1236. Minkova, M; Minkov, D; Rodgerson, J L and Harley, R G: Adaptive neural speed controller of a DC motor, Submitted for publication to IEEE

Proceedings Electric Power Applications.


Henaff, P and Milgram, M: Adaptive neural control with backpropagation algorithm. Proceedings of he IMACS International Symposium on Signal Processing, Robotics and Neural Networh, France, 1994, pp.395-398. Krose, B J A and van der Smagt, P P: An introduction to neural networks, (University of Amsterdam, Netherlands, 1993), pp.30-37. Ogata, K: ]Modem control engineering, (Rentice-Hall, USA, 1990), pp. 118-120. LIST OF ]FIGURE CAPTIONS

-t,

REFERENCES
[l] Weerasooriya, S and El-Sharkawi, M.A: Identification and control of a DC motor using back-propagation neural networks, ZEEE Trans.Energy Conversion, 1991, 6, pp.663-669.
I

0.1

1(8)

.s

2s

[2]

Antsaklis, P. Atherton, D P and Warwick, K: J; Neural networks for control and systems (Short Run Press, UK, 1992), pp.31-68. Weerasooriya, S and El-Sharkawi, M A : Laboratory implementation of a neural network

Fig.1

Computed rotor speed trajectory (- - -), and armature current trajectory (-), for an armature voltage trajectory ( ) applied to the DC motor.

[3]

0-7803-3019-6/96/$3.00 1996 IEEE. D

840

U /\\

*
mapping
HIODEW

INPUT

LAYER

LAYER

Fig.2

A schematic diagram of the adaptive neural speed controller of a DC motor with current limitation.

Fig.4

Topology of the ANN of the controller.

INTRODUCE THE REFERENCE SPEED TRAJECTORY W ( k ) DEFINED IN THE INTERVAL [ 0 t c ] FOR A . SAMPLING PERIOD T AND CALCULATE THE INPUT r(k) TO THE REFERENCE MODEL FROM eqn.11. THE NO OF THE CONNECTION WEIGHTS VECTOR A f THE ANN IS INSIDE THE SPHERE WITH A RAOUlS 0.1 IN THE WEIGHT SPACE. k=1

,
t
THE DISCRETE TIME IS t=kT. MEASURE THE ROTOR SPEED w(k) AND THE ARMANRE CURRENT i(k-1) OF THE MOTOR. CALCULATE THE CONTROL ERROR t ( k ) 0 U,&) dk)

BACKPROPAGATE THE ERROR %(k) THROUGH THE ANN AN0 UPDATE ITS CONNECTION WEIGHTS FOR ONE TRAlNlNC EPOCH

CALCULATE THE ROTOR SPEED a k + l ) IN THE DISCRUE TIME MOMENT t=(k+l)T FROM eqn.12 FOR A j z 0 . 6 AN0 Avr0.2

Fig.5

RECALL THE ROTOR SPEED FOP. THE LAST Two DlSCREE TIME MOMENTS AND SUPPLY Q(k+l).w(k) AND d k - 1 1 TO THE THREE INPUTS O THE ANN F

Performance of the controller with current limitation for the reference speed trajectory RSTI. T=200ms, r=0.25, (r=0.7, and li =0.2.

1
W R E THE CONTROL VOLTAGE V(k) FROM THE OUTPUT O F THE ANN

4l
3
I

THE INPUT TO THE

4
Fig.6
Variation of the armature current for the controller with current limitation and RSTI. T=200ms, y=0.25, a=0.7, and Ii=0.2.

Fig.3 A flowchart of the operation of the adaptive neural speed controller of a DC motor with
current limitation.
___ 0-7803-3019-6/96/$3.00 1996 IEEE. 0
* --T1

84 1

10

61
4

!
"

--

"It
-0
-10

I,+

301

20
10

S o -2
4

1
-10;

20

40

Bo
1

80

100

120

1
1

(a

Fig.7

Variation of the armature current for the controller without current limitation and RSTI . T=200ms, y=0.25, a=0.7, and fi=0.2.

Fig.10 Variation of the armature current for the controller without current limitation and RST2. T=40ms, y=0.25, (r=0.25, and li=0.6.
--I

-loot -1%

-zoo 0

U 1
50
100

150

t (@

Fig.8

Performance of the controller with a current limitation for the reference speed trajectory R S E . T = 4 0 m , y=0.25, ~ ~ ~ 0 . and , 2 5 li =O .6.

Fig.11 Performance of the controller with current limitation for the reference speed trajectory RST3. T=20Oms, y=0.25, a=0.7, and li=0.2.

2
1

9 0
-1

-2

-lot

-151 0

5
t (*)

10

"I

-3
I

1s

io

Fig.9

Variation of the armature current for the controller with current limitation and RSl2. T = 4 0 m , y = 0 . 2 5 , a = 0 . 2 5 , and Ii=0.6.
842

Fig.12 Variation aif the armature current for the controller with current limitation and RST3.T=200m, y = 0 . 2 5 , a=0.7, and k=0.2.

0-7803-3019-6/96/$3.001996 IEEE. 0

You might also like