International Journal of Engineering Trends and Technology (IJETT) – Volume 22 Number 2- April 2015 Error feedback based speed control of DC motor drive for variable load torque using neural network Akhilesh Sharma#1; Krishanu Nath#2; Amlesh Kumar#3; Amarjit Roy#4 #1 Assistant Professor, Dept. of Electrical Engineering North Eastern Regional Institute of Science and Technology Nirjuli, Arunachal Pradesh, India. #2,3,4 B-Tech students, Dept. of Electrical Engineering North Eastern Regional Institute of Science and Technology Nirjuli, Arunachal Pradesh, India. Abstract— Due to versatile speed control characteristic, the dc motors have been inseparable from industry. The DC motors are perhaps the most widely used energy converters in the modern machine tools and robotics. With the increasing use of power semi-conductor units, the speed control of DC motor is increasingly getting sophisticated and precise. Speed of the DC motor can be controlled by armature voltage control, field control and armature resistance control methods. The introduction of MATLAB and Simulink has made the designers to simulate complex circuits and study their characteristics. In this paper, an attempt has been made to control the speed of a separately excited DC motor by armature voltage control method incorporating a neural network as the speed controller for constant as well as variable load torque. Using MATLAB, Simulink and Neural Network toolbox, a comprehensive study has been demonstrated. Keywords— DC motor; neural network; speed control; modelling I. INTRODUCTION Although surplus amount of energy is available in different forms, but it is difficult to directly harness them due to geographical and economical reasons. These energy resources are under utilized. So, it is necessary to convert available energy to fulfill one’s need and society altogether. In the energy conversion, DC Motor plays an important role; converting electrical energy into mechanical energy. In mechanical system, speed control is necessary to do mechanical work in a proper way. It makes motor to operate easily [1]. This method of speed control is independent of load on the motor and permits remote control of speed. The introduction of variable-speed drives increases the automation and productivity as well as its efficiency. In place of constant speed operation, if a variable speed drive is introduced, then efficiency of the drive can be increased from 15 to 27%. This has lot of benefits like reduction of atmospheric pollution through lower energy production, conservation of valuable natural resources and consumption [2]. The artificial neural network (ANN) also known as neural network processes information in a similar way as human brain ISSN: 2231-5381 does. The network is composed of a large number of highly interconnected processing elements called neurons, working in parallel to solve a specific problem. It learns by example hence it can be programmed to perform a specific task [4, 5]. It has many advantageous features like parallel and distributed processing, efficient nonlinear mapping between inputs and outputs and robustness, without prior knowledge of the system model [3]. Multilayer neural networks have been applied in the identification and control of dynamic systems. These are used in control systems with the help of following three typical commonly used neural network controllers: model predictive control, NARMA-L2 control, and model reference control [2]. As with most neural controllers, they are based on standard linear control architectures. There are number of articles that use ANNs applications to identify the mathematical D.C. motor model and then this model is applied to control the motor speed. They also use inverting forward ANN with input parameters for adaptive control of D.C. motor. This paper is about the study of steady-state and dynamics control of dc machine. By using speed of the DC motor as feedback, the error signal is generated. A variable load torque is applied to check the stability in speed of the motor. . II. DC MOTOR DC motor consists of two types of winding, rotor winding also known as armature placed on the armature and stationary winding placed on the stator of the DC motor. In all, DC motors, except permanent magnet brushless motors, current conducts through armature windings by passing current through carbon brushes that slide over a set of copper surfaces called a commutator, which is mounted on the rotor. Many applications require the speed of a motor to be varied over a wide range. One of the most attractive features of DC motors in comparison with AC motors is the ease with which their speed can be varied. The governing equations of DC motor are: ππ π£π = ππ + ππ π π + πΏπ ππ‘π (1) ππ = πΎ∅ππ http://www.ijettjournal.org (2) Page 69 International Journal of Engineering Trends and Technology (IJETT) – Volume 22 Number 2- April 2015 ππ = πΎ∅ππ ππ − ππΏ = π½ (3) πππ ππ‘ + π΅ππ (4) Where π£π – The armature voltage (V) ππ – The motor back emf (V) Fig 1: Closed loop model of DC motor ππ – The armature current (A) Using the above model, the Simulink model of DC motor constructed and is stored in a subsystem as shown in Fig. 2. A gain of 9.54 is applied to convert the speed of the motor from rad/sec to revolution per minute (rpm). π π – The armature resistance (β¦) πΏπ – The armature inductance (H) ∅ – The flux per pole (Wb) πΎ – The motor back emf and torque constant (NmA−1) ππ – The speed of motor (rad/s) ππ – The motoring torque (Nm) ππΏ – The load torque (Nm) π½ – The moment of inertia of motor shaft (Nm2) π΅ –The viscous friction co-efficient (Nm-s) Using (1) and (2), the back emf can be expressed as ππ ππ = π∅ππ = π£π − ππ π π − πΏπ ππ‘π (5) Rearranging the terms, speed can be expressed as ππ ππ = (π£π − πΌπ π π − πΏπ ππ‘π)/πΎ∅ Fig 2: Simulink model of DC motor (6) It is obvious that the speed can be controlled by varying ο· Flux/pole, ∅ (Flux Control) ο· Resistance π π of armature circuit (Rheostat Control) ο· Applied armature voltage π£π (Voltage Control). In this paper, voltage control method of the speed control of DC motor has been used. . A. Modeling of DC motor The modeling part of DC motor has been done keeping the fact that the applied voltage to the field is constant hence the flux per pole is constant. Taking the Laplace transform of (1) and (4) and rearranging the terms (7) and (8) are obtained: πΌπ (π ) ππ (π )−πΈ(π ) = π πΏ ππ(π ) ππ (π )−ππΏ(π ) 1 π + π π 1 = π½π +π΅ (7) (8) B. Speed control strategies The speed control strategies which are incorporated in this paper are: ο· Open-loop speed control ο· Close-loop speed control In the open loop speed control, the neural network used, acts as an armature voltage provider which directly controls the speed of the motor. This technique initially proved to be good for light loads and no-load conditions. But, in real time applications, it is necessary to have fine speed control irrespective of the nature of load torque applied. The motor has to run at specified speed or sets in range of within 5% of the set speed. To achieve this set level of speed condition, close loop system is used and the motor speed is used as a feedback signal. The feedback allows limiting the error and producing better speed control compare to the earlier case. The block diagram of the control strategies are shown in Fig.3 and Fig. 4 respectively. Using (2), (3), (7) and (8) a closed loop block diagram of DC motor is formed which is shown in Fig. 1. Fig 3: Open-loop speed control strategy ISSN: 2231-5381 http://www.ijettjournal.org Page 70 International Journal of Engineering Trends and Technology (IJETT) – Volume 22 Number 2- April 2015 Fig 4: Close-loop speed control strategy Fig 6: MATLAB model of neural network III. NEURAL NETWORK A neural network structure can be categorized as a biological neural network or artificial neural network. The neuron receives signals from the other similar neurons through dendrites, collects in the nucleus or soma, and then distributes to the neurons through the axon as shown. The input signal passes through a synapse or synaptic junction which is infinitesimal gap in the dendrites that is filled with neurotransmitter fluid that either accelerates or decelerates the flow of electrical charge. After training the neural network it is observed that the mean square error (m.s.e.) is reduced to 2.2611x10-6after 1000 iterations. The plot for m.s.e versus training iteration is shown in Fig.7. Fig 6: Plot of mean square error v/s iteration Fig 5: Nonlinear model of a neuron An artificial neural network consists of an input layer, one or more hidden layer, and output layer. Each layer consists of several neurons. The structure of an artificial neuron consists of input nodes, synaptic weight, bias, summing junction, activation function and output node. The set of synapse is characterised by set of synaptic weights. Each input gets multiplied with the synaptic weight and gets added at the summing junction. The bias which lowers or increases the level of net output also gets added at the summing junction. The output from the summing junction is applied to an activation function which is used to limit the output of the neuron[6]. There are various types of activation functions of which tan-sigmoid and linear functions have been used in this paper. The Levenberg-Marquadt algorithm is used for training the neural network. The neural network used for speed control is trained and simulated using neural network toolbox as shown in Fig. 6. The data for training is obtained from an m-file. The objective of the neural network is to provide the armature voltage according to the input reference speed (in rpm). The network has 1-5-1 structure with tan sigmoid activation function in hidden layer. A linear function activation function has been applied at the output layer. IV. SIMULATION AND RESULTS The entire simulation work has been done using MATLAB, Simulink and neural network toolbox. The neural network used is 1-5-1 structured and feed-forward back propagation type. The simulated model has been subjected to various types of loads and reference speed for both speed control strategies. The parameters of the DC motor used are provided in Table 1[7]. Table 1: Parameters of DC motor Sl no. Parameters Value 1. 2. 3. 4. 5. 6. Power Supply Voltage Speed of the motor Field Voltage Moment of inertia Torque constant 5 HP 240V 1750 RPM 150V J=0.02215Nm2 K=1.976NmA−1 7. 8. 9. Damping factor Armature resistance Armature inductance B=0.002953 Nm-s Ra=11 β¦ La=0.1215 H A. Open-loop speed control For the open loop speed control, the neural network generates the armature voltage as per the reference speed. As stated earlier, it provides good speed control only for light loads. The Simulink model of the open loop control strategy is ISSN: 2231-5381 http://www.ijettjournal.org Page 71 International Journal of Engineering Trends and Technology (IJETT) – Volume 22 Number 2- April 2015 shown in Fig.8. Initially a constant torque of 5 N-m has been applied with 1500 rpm as reference speed and later the reference speed is increased to 1750 rpm at t=3.5 sec and again then decreased to 1600 rpm at t=6 sec. On application of neural network controller, motor picks up the reference speeds. This is done by controlling the armature voltage and hence the speed of the DC motor is controlled. This system, when introduced to a nonlinear or varying load torque, does not provide satisfactory results. As the load increases, the armature current increases as shown in Fig 9 thereby reducing in motor speed as shown in Fig 10. The difference of motor speed and reference speed at full load is around 500rpm. This is not a desirable feature especially when DC motor is used for industrial applications. To compensate this speed drop a close loop strategy has been adapted. Fig 7: Plot of speed for load torque 2 Nm Fig 10: Plot of speed for variable load torque (open-loop). B. Close-loop control The close loop strategy has dual neural network, one of them gives the armature voltage required as per the reference speed.. The second neural network feeds an additional voltage to the armature which compensates the error. The Simulink model is shown in Fig. 11. The load applied in second case is reapplied to this model. From the Fig. 12, it is evident that applying the same load, better speed of the motor is achieved and error in speed is in the range of 4% of the reference speed. The armature current curve for the same is shown in Fig. 13. Initially, the reference speed is set to 1500 rpm with 5 N-m load torque. At t=1 sec, the load torque is changed from 5N-m to 10 N-m. the speed of the motor settles to 1488 rpm, with an error in speed of 0.8% . The armature current sets to 2.8A after an over shoot of 0.4A (14.28%). At t=2.5 sec, the full load torque (20 N-m approx.) is applied, with controller, the speed drops down to 1450 rpm but without controller, speed of motor sets to 1000 rpm with reduction in error from 30% to 3.33%. The armature current overshoots to 11A and sets to 10.2A in 0.17sec. Fig 8: Simulink model of open-loop sped control Fig 11: Simulink model of close-loop speed control Fig 9: Plot of armature current for variable load torque (open-loop) Fig 12: Plot of speed for variable load torque (close-loop) ISSN: 2231-5381 http://www.ijettjournal.org Page 72 International Journal of Engineering Trends and Technology (IJETT) – Volume 22 Number 2- April 2015 Fig 13: Plot of armature current for variable load torque (close-loop) This model is then applied with three more types of load sinusoidal load, saw-tooth and random function. The plots for each type of loads are given in Fig. 14, 15 and 16 respectively. It can be concluded from these curves that with the close-loop control, the motor picks up the reference speed. Fig 16: Plot of speed for random load torque (close-loop) REFERENCES [1] [2] [3] [4] [5] [6] Fig 14: Plot of speed for saw-teeth load torque (close-loop) [7] Anurag Dwivedi “Speed Control of DC Shunt Motor with Field and Armature Rheostat Control Simultaneously”,Advance in Electronic and Electric Engineering, ISSN 2231-1297, Volume 3, Number 1 (2013), pp. 77-80, © Research India Publications Amit Kumar Singh, Dr. A.K. Pandey “Intelligent PI Controller for Speed Control of SEDM using MATLAB” International Journal of Engineering Science and Innovative Technology (IJESIT) ISSN: 23195967 Volume 2, Issue 1, January 2013 page no. 180 Muammer Gokbulut and Ahmet Tekin,”An Educational tool for neural network control of brushless DC motors” Int. J. Engng Ed. Vol. 22, No. 1, pp. 197-204, 2006, © 2006 TEMPUS Publications. Oludele Awodele and Olawale Jegede,” Neural networks and its application in engineering”, Proceedings of Informing Science & IT Education Conference (InSITE) 2009. Behnam Bavarian,” Introduction to neural networks for intelligent control”, IEEE International Conference on Neural Networks, San Diego, California, June 21-24,1987 Simon Haykin,” Neural network, a compressive foundation”, second edition, Prentice Hall International, Inc. G.MadhusudhanaRao and Dr. B.V.SankerRam,” A Neural Network Based Speed Control for DCMotor “,International Journal of Recent Trends in Engineering, Vol 2, No. 6, November 2009. Fig 15: Plot of speed for sinusoidal load torque (close-loop) ISSN: 2231-5381 http://www.ijettjournal.org Page 73