Multi-layer Neural Network for Servo Motor Control

Full text

International Journal of Computer Applications (0975 – 8887)
Volume 71– No.14, May 2013
32
Multi-layer Neural Network for Servo Motor Control
Lalithamma GA
Assistant Professor, Dept of E&E
SJBIT, Mandya
P.S. Puttaswamy,Ph.D
Prof, Deptt of E&E
PESCE, Mandya
Kashyap D Dhruve
Planet-I Technologies
Bangalore
ABSTRACT
AC servo systems are extensively used in robotic actuators and are
competing with DC servo motors for motion control because of
their favorable electrical and mechanical properties. This paper
presents an approach towards the control system tuning for the
speed control of an AC servo motor. An approach towards speed
control of servo motor in presence of system parameter variations is
presented. Multi-layer Artificial Neural Networks are designed and
trained to model the plant parameter variations. Improvements in
the speed control performance are presented for smaller variations
and larger variations in the motor parameters and the load
conditions.
KEYWORDS—PID, AC servo motor, multi-layer ANN, direct
torque control.
1. INTRODUCTION
The PID control algorithm is generally used to control almost
all loops in the process industries, AC servomechanisms and drive
applications. In order to use a controller, it must first be tuned to the
system requirements. The tuning synchronizes the controller with
the controlled variable, thus allowing the process to be kept at its
desired operating condition. Standard methods for tuning
controllers and criteria for judging the loop tuning have been used
for many years. Among the available methods some of them are
based on mathematical criteria such as Cohen- Coon method, some
are trial and error method, such as Continuous cycling method,
relay feedback method and Kappa-Tau tuning method. Existing
methods even though they have some advantages they have some
difficulties while defining the solution for tuning the PID controller.
Pre-selection of a controller structure can pose a challenge in
applying PID control to a system. As vendors often recommend
their own design procedure for controller structures, the tuning rules
adopted for a specific controller structure may not perform well
with other controller structure [1]. Therefore, one solution is seen to
provide support for individual structure is through software.
Detailed discussions on the use of various PID structures are
presented in [1]. Nonetheless, controller parameters are required to
be tuned such that the closed-loop control system would be stable
and meet the given objectives associated with the following
functions
1. Stability robustness.
2. Set-point following and tracking performance at transient,
including rise-time, overshoot, and settling time.
3. Regulation performance at steady-state including load
disturbance rejection.
4. Robustness against plant modeling uncertainty.
5. Noise attenuation and robustness against environmental
uncertainty.
In industrial process there are many systems having nonlinear
properties. Moreover these properties are often unknown and time
varying. The commonly used PID controllers are simpler to realize
but they suffer from poor performance if there are uncertainties and
nonlinearities in the system[2]. Therefore PID controllers have
difficulty in tuning suitable to load because of disturbance,
parameter variation and noise.To overcome these difficulties
conventional PID controllers were modified and developed lately
by using various newer techniques. The controller development
based on Fuzzy logic, neural network and Fuzzy-neural have been
proposed as the better choice. However the Fuzzy logic controller
(FLC) when compared to conventional controller, the main
advantage of FLC is that no mathematical modeling required [3].
FLC controllers have excellent ability if a simple control algorithm
is implemented. However this method has low reliability because
the control results are sensitive to change in system parameters and
do not react rapidly to parameter changes.
AC Servo systems are extensively used in robot actuator, machining
tools, positioning control servomechanisms, computer numerical
control and Industrial control [4]. AC Servo system is competing
with DC servo system for motion control because of their favorable
electrical and mechanical properties with good dynamic behavior
with high efficiency.
AC servomotor is basically a two phase reversible induction
motor and is capable of providing the desired response
characteristics, due to their smaller diameter and high resistance
rotors. The small diameter provides low inertia and helps for faster
starts, stops and speed reversals while high resistance helps to shape
the speed torque characteristics suitable for accurate control.
The evolving trend adopted by researchers for control systems
infuses the use of machine learning or intelligence along with the
conventional control techniques. Researchers have incorporated
intelligence using techniques like neural networks [6][7], genetic
algorithms [8] [17],evolutionary computing [9] and fuzzy logic
[10][14]. The use neuro-fuzzy techniques for control is also gaining
popularity [11][12] [13][15]. The use of neuro fuzzy systems is not
adopted as the combination of such complex control systems
possesses huge computational complexities for real time
applications as considered by the authors of this paper.
This paper considers a direct torque control [DTC] of the
asynchronous motors. A survey conducted has shown researchers
adopt fuzzy logic [16] for the control but the results discussed in the
paper lack accuracy of control. The use of neuro fuzzy logic
proposed by researchers in [17] [18] demonstrated good learning
capabilities, enabled to address the issue of a large torque ripple at

International Journal of Computer Applications (0975 – 8887)
Volume 71– No.14, May 2013
33
low operating speeds but could not reduce the computational
complexity of the overall system. The open research issues that
exist are clearly discussed in [17] [18] [19].A comprehensive
literature review of the techniques proposed by researcher’s in the
parameter control of motor drives is presented in [20]. This paper
presented here adopts the use of neural networks for adaptive DTC
of the ac servo motor.
2. ARTIFICIAL NEURAL NETWORKS
Some researchers have suggested that Artificial Neural Network
(ANN) is the better choice for the controller design, since one could
obtain superior and highly precise control response than a
conventional PID controller. This is because the proposed controller
can tune the conventional PID controller parameters more
accurately with neural network technology since it requires only
inputs and outputs even in Jacobian of unknown control objects.
Also neural network are best in identifying patterns or trends in
data, they are also better suited for prediction or Forecasting.
The learning ability, self adapting and fast computing futures of
ANN make it well suited for medical diagnosis, medical research,
voice recognition, machinery control, air traffic control and control
of electrical power system in many application such as electric load
forecasting, transient ability assessment, active power filter,
dynamic voltage restorer, unified power quality conditioner etc. In
learning processes neural network adjusts its structure such that it
will be able to follow the supervisors set point. The learning is
repeated until the difference between new output and the supervisor
set point is low. Neural network based PID controller
implementation and its computational task involved is so small that
the on line adaptation becomes easy.
An artificial neural network is an information processing
paradigm that is inspired by the way biological nervous system,
such as brain process information. The key element of this model is
the novel structure of the information processing system. It is
composed of a large number of highly interconnected processing
elements working in unison to solve specific problems. ANN is like
where the people learn by examples [5]. Neural network take a
different approach to problem solving than that of conventional
computing techniques. Conventional computers use an algorithmic
approach i.e., the computer follows a set of instructions in order to
solve a problem that restricts the problem solving capability. But
computers are more useful if they do things that we don’t exactly
know how to do it. Neural network processes information in a
similar way the human brain does. The network is composed of a
large number of highly inter connected processing elements
(neurons) working in parallel to solve a specific problem. Neural
network is a kind of continuous time dynamic system with high
nonlinearity and good learning ability [6].
3. MOTOR CONTROL
In this section, speed control loop for a single phase
asynchronous motor is demonstrated. The machine is fed by a
PWM inverter. The speed control loop has a PID based regulator
which produces a quadrature-axis current reference. The motor
electromagnetic torque is controlled by this quadrature current.
Hence, this method is also known as direct torque control (DTC).
The motor flux is controlled by the direct-axis current.
Transformation from d-q frame to a-b frame is used and the
resulting currents are fed to the main and auxiliary motor
windings.
The asynchronous motor has two windings: the main winding
and the auxiliary winding. The electrical part of the machine is
represented by a fourth order system whereas themechanical part is
represented by a second order system. The electrical parameters
are referred to the stator reference frame (d-q frame). The
reference frame transformation (a-b to d-q frame) is given by the
following equation 1-2. The electrical behavior of the AC motor is
modeled by equations 3-8, while the mechanical behavior is
modeled by equations 9-10.

=1 0
01

(1)

=


(2)
The flux linkage equations for the direct and the quadrature
axes are given by equations (3)-(8).


 =
+
 
 
 (3)


 =
+
 

 (4)


 =


+
 

 (5)


 =


+
 

 (6)

=
+
 (7)

=
+
 (8)
The electromechanical torque is given by
 =3
2
P
2b(ds
siqs
sqs
sids
s) (9)
The dynamics of the rotor is given by equation

 = +   (10)
Using MATLAB a SIMULINK model is used for system level
simulations and modelling. The block diagram is as shown in
Figure1.

International Journal of Computer Applications (0975 – 8887)
Volume 71– No.14, May 2013
34
Fig1: SIMULINK Model for Motor Speed Control
Fig2: Multi-layer neural network architecture
4. NEURAL NETWORK ARCHITECTURE
A multilayer neural network is shown in Fig 2. Each layer has
its own weight matrix W, its own bias vector b, a net input vector
n and an output vector a [6].

International Journal of Computer Applications (0975 – 8887)
Volume 71– No.14, May 2013
35
A layer whose output is the network output is called an output
layer while other layers are called hidden layers. Multilayer
networks are more powerful than single layer networks. In the
present simulations, feed-forward neural network architecture is
implemented in MATLAB. The neural network takes in the motor
parameter variations and the dynamic load conditions as the input
and generates the PID controller gains as the output. An input
matrix with different input parameter values is fed to the neural
network and the neural network is trained using the gradient
descent back-propagation algorithm. Back-propagation is used to
calculate derivatives of performance with respect to the weight and
bias variables X and each variable is adjusted according to gradient
descent:
 =   /
MATLAB function ( ) is used to create a feed-
forward back-propagation network. 2 hidden layers are created
with 10 neurons in each layer. The transfer function for hidden
layer neurons is ( ),while ( ) is used for the output
layer neurons. The transfer functions for the two are plotted in
Fig4. ( )is used for the neural network training. The training
function is ( )which implements neural network training
using gradient descent back-propagation algorithm.
Three different single output, multiple input neural
networks are designed to be trained for tuning the proportional,
integral and derivative gains of the PID controller based on the
motor parameter variations. Each neural network is trained for the
same set of training parameters using the gradient descent back-
propagation algorithm. The advantage of training and tuning three
different neural networks is that the mean square error parameter
used as a performance metric is better minimized when three
independent networks are tuned as compared to a single network
with multiple outputs.
5. RESULTS AND CONCLUSION
A 0.25HP AC servo motor is used for the simulation model.
The nominal motor parameters are listed in Table I.
TABLE I. Nominal motor parameters
Parameter
Value
 
0.25 
 
2.02 
 
7.4 
 
4.12 
 
5.6 

0.0146 2
Fig 4 and 5 shows the speed control response of the
motor for nominal motor parameters with different loads and
friction coefficients. The speed command of 100 rpm is tracked by
the motor controller and the motor attains the commanded speed
after some delay and slight overshoot.
Fig 6 and 7 shows the response of the motor with and
without tuning of the PID gains using neural network, for smaller
variations and for larger variations. It can be seen that the gradient
descent back-propagation algorithm is able to tune the PID gains
in-spite of the parameter variations. However the performance
improvement is more for smaller variations as compared to larger
variations.
Single input neuron
Multi-input neuron
Fig 3 Neuron models
Hyperbolic tangent
sigmoid function
Pure Linear function
Fig 4 Neuron transfer functions

International Journal of Computer Applications (0975 – 8887)
Volume 71– No.14, May 2013
36
Fig 4: Nominal Speed Response for different loads
Fig 5: Speed Response for different friction coefficients Fig 6: Comparison of response with and without neural
network tuning (lesser parameter variations)

International Journal of Computer Applications (0975 – 8887)
Volume 71– No.14, May 2013
37
Fig 7: Comparison of response with and without neural
network tuning (larger variations)
6. ACKNOWLEDGMENT
The authors would like to express their cordial thanks to
MrAshutosh Kumar of Planet-i Technologies for their much valued
support and advice.
7. REFERENCES
[1] A. Kaya and T. J. Scheib, “Tuning of PID controls of
different structures,”Control Eng., vol. 35, no. 7, pp. 62–65,
Jul. 1988.
[2] Woo-Yoog Han, Sang-Min Kim, Sung-Joong Kim, Chang-
Goo Lee sensorless vector control of induction motor using
improved self –tuning fuzzy PID controller. ICE: Annual
conference in Fukui, August 4-6, 2003.
[3] Pyoung-Ho kim, Sa-Hyun sin, Hyung-Lae baek, Geum-Bae
Cho, Speed control of AC servo motor neural networks.
[4] Principles of soft computing, S.N. Sivanandam , S.N. Deepa
[5] K.J. Hunt, D. Sbarbaro, R. Zbikowski, P.J. Gawthrop, “Neural
Networks for Control Systems: A Survey, Automatica”,pp.
1083-1122, 1992.
[6] Hagan, Martin T., Howard B. Demuth, and Mark H. Beale.
Neural
Network design. Boston London: Pws Pub., 1996.
[7] A. M. Zou and K. D. Kumar, “Neural network-based
distributed attitude coordination control for spacecraft
formation flying with input saturation,” IEEE Trans. Neural
Netw. Learn. Syst., vol. 23, no. 7, pp. 1155–1162, Jul. 2012.
[8] M B Anandaraju, Dr. P S Puttaswamy and Jaswant Singh
Rajpurohit. Article: Genetic Algorithm: An approach to
Velocity Control of an Electric DC Motor. International
Journal of Computer Applications 26(1):37-43, July 2011.
Published by Foundation of Computer Science, New York,
USA
[9] M B Anandaraju and P S Puttaswamy. Article: Modified
Interactive Evolutionary Computing for Speed Control of an
Electric DC Motor. International Journal of Computer
Applications 39(15):19-24, February 2012. Published by
Foundation of Computer Science, New York, USA
[10] M. Seera, C. P. Lim, D. Ishak, and H. Singh, “Fault detection
and
[11] diagnosis of induction motors using motor current signature
analysis and a hybrid FMM-CART model,” IEEE Trans.
Neural Netw. Learn. Syst., vol. 23, no. 1, pp. 97–108, Jan.
2012.
[12] R. J. Wai and P. C. Chen, “Intelligent tracking control for
robot
[13] manipulator including actuator dynamics via TSK-type fuzzy
neural network,” IEEE Trans. Fuzzy Syst., vol. 12, no. 4, pp.
552–559, Aug. 2004.
[14] R. J. Wai and Z. W. Yang, “Adaptive fuzzy neural network
control design via a T-S fuzzy model for a robot manipulator
including actuator dynamics,” IEEE Trans. Syst., Man,
Cybern. B, vol. 29, no. 5,pp. 583–591, Oct. 2008.
[15] Rong-Jong Wai, Rajkumar Muthusamy,” Fuzzy-Neural-
Network Inherited Sliding-Mode Control for Robot
Manipulator Including Actuator Dynamics” IEEE
Transactions On Neural Networks And Learning Systems,
Vol. 24, No. 2, February 2013
[16] G. Feng, “A survey on analysis and design of model-based
fuzzy control systems,” IEEE Trans. Fuzzy Syst., vol. 14, no.
5, pp. 676–697, May 2006.
[17] R. Sharma and M. Gopal, “A Markov game adaptive fuzzy
controller for robot manipulators,” IEEE Trans. Fuzzy Syst.,
vol. 16, no. 1, pp. 171–186, Feb. 2008.
[18] L. ROMERAL, A. ARIAS, E. ALDABAS, “Novel direct
torque control scheme with fuzzy adaptive torque ripple
reduction”. IEEE Trans on Industrial Electronics, vol. 50, pp.
487-492, 2003.
[19] Faa-Jeng Lin, Senior Member, IEEE, Po-Kai Huang, and
Wen-Der Chou, “Recurrent-Fuzzy-Neural-Network-
Controlled Linear Induction Motor Servo Drive Using
Genetic Algorithms”, IEEE Trans on Industrial Electronics,
vol. 54, pp. 1449-1461, 2007.
[20] Shu Dasong . “A Direct Torque Control of Asynchronous
Motor Based on Fuzzy Net-work”, 2011 6th IEEE Conference
on Industrial Electronics and Applications. pp 728-731. 2011
[21] Sun Xiao Hui, Zhang Zeng Ke, Han Ceng Jin. “Direct torque
control of induction motor torque ripple minimization
method”, CSEE, vol. 22, pp. 109-112,2002.
[22] E. Levi, R. Bojoi, F. Profumo, H.A. Toliyat, S. Williamson.
“Multiphase induction motor drives-a technology status
review”, IET Electr. Power Appl, vol. 1, pp. 489-516, 2007.
IJCATM : www.ijcaonline.org