NEURAL NETWORK BASED SPEED CONTROL FOR A DC MOTOR

Samartsev I. V.

National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"

NEURAL NETWORK BASED SPEED CONTROL FOR A DC MOTOR

Abstract. This paper introduces a new concept of Artificial Neural Networks (ANNs) in estimating speed and controlling a separately excited DC motor. The neural control scheme consists of two parts. One is a neural estimator, which is used to estimate the motor speed. The other is a neural controller, which is used to generate a control signal for a converter. These two networks are developed by Levenberg-Marquardt back propagation algorithm. A standard three-layer feed forward neural network with sigmoid activation functions in the input and hidden layers and purelin in the output layer is used. Simulation results are presented to demonstrate the effectiveness and advantages of the control system of the DC motor with the ANNs in comparison with the conventional control scheme.

I. INTRODUCTION

Nowadays, the fields of an electrical power system control in general and a motor control in particular are gaining momentum. The new technologies are emerging for control scheme. One of these new technologies is Artificial Neural Networks (ANNs) that are based on the operating principle of a human being nerve neural. This method is applied to control the motor speed [1]. Inverting forward ANNs with two input parameters for an adaptive control of the DC motor [4] is used. However, these researches were not interested in the ability of forecasting and estimating the DC motor speed. The ANNs are applied broadly because of the following special qualities:

1. All the ANNs signals are transmitted in one direction, the same as in an automatically control system.

2. The ability of the ANNs to study the sample.

3. The ability to create the parallel signals in analog as well as in a discrete system.

4. The adaptive ability.

With the special qualities mentioned above, the ANNs can be trained to display the nonlinear relationships that the conventional tools could not implement. It is also applied to control complicated electromechanical systems such as DC motor and synchronous machines [5]. To train the ANNs, the input and output datasheets are to be determined first, and then the ANNs’ net is being designed by optimizing the number of hidden layers, and neurals of each layer, the number of neurals of each layer, as well as the input/output number and the transfer function. The following is to find the ANNs net learning algorithm. The ANNs are trained to rely on two basic principles: supervisor and unsupervisor. According to a supervisor, the ANNs study the input/ output data (targets) before being used in the control system. In this paper, the new ANNs’ application in speed estimating and controlling a separately excited DC motor is presented. The motor speed is controlled by a forecasting method and a forecasting task, which the ANNs undertake from the terminal voltage parameter, armature current, and a reference speed.

II. DC MOTOR CONTROL MODEL WITH ANNS

The DC motor is the obvious proving ground for advanced control algorithms in electric drives due to the stable and straightforward characteristics associated with it. It is also ideally suited for trajectory control applications as shown in the reference [1-3]. From a control systems’ point of view, the DC motor can be considered as a SISO plant, thereby eliminating the complications associated with a multi-input drive system.

Conventional control systems of the DC motor:

There are different methods to synthesize control systems of the DC motor. The ANNs authors have presented a conventional control system of the DC motor, where a current regulator and a speed regulator are synthesized by Bietrage-optimum to reduce the over-regulation [6].

In the conventional model, current and voltage sensors are very important elements that play the main role during the regulation of speed alongside with a current regulator and a speed regulator. For the speed control of the DC machine, a conventional feedback control logic approach is observed to be lower in accuracy due to direct sensor measurements. The approach is to be developed for selecting the speed parameters and providing accurate controlling to the driving circuitry. For the realization of such a controlling approach, in this paper, a neural network based control strategy is proposed. The developed approach is briefed in the following sections.

The control system of the DC motor using ANNs:

A neural network is a generalized approach to making the learning algorithm and a decision for accurate control operation in various applications. The approach of a neural network basically works on the provided priories’ information and makes a suitable decision for a given testing input based on the provided training information. This approach is analogous to the human controlling approach where all the past observations are taken as the reference information and are used as a decision variable. To obtain such estimation in the current DC motor controlling approach, the current DC motor drives are to be improved using such a learning approach. In this paper, a dual level neural network approach is designed for DC machine speed control. A dual level modelling provides faster training and converging as compared to a single level neural modelling. For the realization of a dual level neural modelling, a two-neuro architecture, namely ANN-control and ANN-train, is proposed.

The 2 models of the control system of the DC motor using the ANNs are built with the ANN-train and ANN-control unit where the network is developed to emulate a function: ANN-train to estimate the speed, ANN-control to control the terminal voltage.

The structure and the process of ANNs’ learning.

The ANNs are trained to emulate a function by presenting it with a representative set of input/output functional patterns. The back-propagation training technique adjusts the weight in all connecting links and thresholds in the nodes so that the difference between the actual output and target output are minimized for all given training patterns [1]. In designing and training an ANN to emulate a function, the only fixed parameters are the number of inputs and outputs to the ANN, which are based on the input/output variables of the function. It is also widely accepted that the maximum of two hidden layers is sufficient to learn any arbitrary nonlinearity [4]. However, the number of hidden neurons and the values of learning parameters, which are equally critical for satisfactory learning, are not supported by such well-established selection criteria. The choice is usually based on experience. The ultimate objective is to find a combination of parameters which gives a total error of required tolerance a reasonable number of training sweeps [1, 2, 3].

f1: tansig; f2:tansig; f3: purelin

Fig 1. Structure of ANN-training

The ANN1 and the ANN2 structures are shown in Fig4, and Fig5. It consists of input layer, output layer, and one hidden layer. The input and hidden layers are tansig-sigmoid activation functions, while the output layer is a linear function. Three inputs of the ANN are reference speed ωr(k), terminal voltage Vt(k-1), and armature current ia(k-1). And output of ANN1 is an estimated speed ωp*(k). The ANN2 has four inputs: reference speed ωr(k), terminal voltage Vt(k-1), armature current ia(k-1), and estimated speed ωp*(k) from ANN-1. The output of the ANN is the control signal for converter Alpha.

f1: tansig; f2:tansig; f3: purelin

Fig 2. Structure of ANN model

The ANNs are trained off-line using inputs patterns of ω_r (k), V_t (k), i_a (k) - for ANN1, and of ω_r (k), V_t (k), i_a (k), ω_p*(k) for ANN2.

The training program of the ANN is written in the Neural Network of Matlab program under m-file; and it uses the Levenberg – Marquardt back propagation. There are no references that mention the optimal number of neural in each layer, so collecting the neural networks becomes more complicated. In order to choose the optimal number of neurals, the neural network is trained by the m-file program, reducing the number of neurals in the ANNs’ hidden layer until the learning error can be accepted.

The ANNs and the training effort are briefly described by the following statistics.

Table 1

The results of the ann training

Network	ANN1	ANN2
Number of input	3	4
Number of output	1	1
Number or hidden layer	1	1
Number of hidden neurons	3	4
Number of training patterns	1215	1215
Number of training sweeps	5000	5000
Learning error	1e-7	1e-8

III. SIMULATION RESULTS

To simulate the conventional control system and the control system with the ANNs, a Simulink/Matlab program with the toolbox of Neural-network is used. The DC motor, which is used in models has the following parameter: 5HP, 240V, 1750 RPM, field 150V, J=0.02215 Nm2, KF=1.976 NmA-1, B=0.002953 Nms, Ra=11, La=0.1215 H. To compare the results of two control system schemes, different operating modes of the DC motor are considered.

Fig 3. Training observation of NN designed

Fig 4. Reference parameters of the DC motor are the same. At reference = 0.1

a) b)

Fig 5. (a,b,c) - Reference parameters of the DC motor are different.

IV. CONCLUSION

The DC motor has been successfully controlled using an ANN. Two ANNs are trained to emulate functions: estimating the speed of the DC motor and controlling it. Therefore, the ANNs can replace speed sensors in control system models. Using the ANNs, there is no need to calculate the parameters of the motor when designing the system control. It has shown appreciable advantages of a control system using the ANNs above the conventional one, when the parameter of the DC motor is variable during the operation of the motors. The satisfied ability of the system control with the ANNs is much better than the conventional controller. The ANNs application can be used in adaptive controls for machines with complicated loads.

The list of references:

1. Weerasoorya S., Al-Sharkawi M. A. Identification and control of a DC Motor using back-propagation neural networks / S. Weerasoorya, M. A. Al-Sharkawi. – IEEE transactions on Energy Conversion. – 1991. – Vol. 6. – No. 4. – Рp. 663–669.

2. Psaltis D., Sideris A., Yamamura A. A. A Multilayered Neural Network Controller / D. Psaltis, A. Sideris, A. A. Yamamura. – IEEE Control System Magazine. – 1988. – Рp. 17–20.

3. Kosko B. Neural Networks and Fuzzy Systems / B. Kosko. – Prentice-Hall International Inc. – 1992.

4. Levin E., Gewirtzman R., Inbar G. F. Neural Network Architecture for Adaptive System Modellinggand Control / E. Levin, R. Gewirtzman, G. F. Inbar. – Neural Networks. – 1991. – No. 4 (2). – Рp. 185–191.

5. Scott G. M. Knowledge-Based Artificial Neural Networks for Process Modelling and Control / G. M. Scott. – PhD. Thesis, University of Wisconsin. – 1993.

6. German-Gankin S. G. The computing modeling for power electronic systems in Matlab / S. G. German-Gankin. – 2001.

7. Rasid. High performance nonlinear MIMO field weakening controller of a separately excited dc motor / Rasid. – Electric Power Systems Research. – 2000. – Vol. 55, issue 3. – Рp. 157–164.

8. Nabil A. Ahmed Modeling and simulation of AC-DC buck-boost converter fed dc motor with uniform PWM technique / A. Ahmed Nabil. – Electric Power Systems Research. –2005. – Vol. 73, issue 3. – Рp. 363–372.

9. Figueroa J., Brocart C., Cros J., Viarouge P. Simplified simulation methods for poly phase brushless DC motors. Mathematics and Computers in Simulation / J. Figueroa, C. Brocart, J. Cros, P. Viarouge. – 2003. – Vol. 63, issues 3-5. – Рp. 209–224.

10. Santana J., Naredo J. L., Sandoval F., Grout I., Argueta O. J. Simulation and construction of a peed control for a DC series motor : мechatronics / J. Santana, J. L. Naredo, F. Sandoval, I. Grout, O. J. Argueta. – 2002. – Vol. 12, issues 9-10. – Рp. 1145–1156.

Charles I. Ume, Ward J., Amos J. Application of microcontroller for speed control of a DC motor / I. Ume Charles, J. Ward, J. Amos // Journal of Microcomputer Applications. – 1992. – Vol. 15, issue 4. – Рp. 375–385.