DOI : https://dx.doi.org/10.26808/rs.ed.i10v2.07
International Journal of Emerging Trends in Engineering and Development Issue 10, Vol.3 (Apr.-May 2020)
Available online on http://www.rspublication.com/ijeted/ijeted_index.htm ISSN 2249-6149
©2020 RS Publication, rspublicationhouse@gmail.com Page 1
DESIGN AND DEVELOPMENT OF A NEURAL NETWORK
(NN) CONTROLLER FOR A ROBOTIC ARM
Akaninyene M. Joshua Ph.D and Prof. Eneh I.I. and Ogbu, Mary N. C.
Dept of Electrical and Electronics Engineering, Enugu State of Science and Technology,
Enugu State.
ABSTRACT
Artificial Neural Networks (ANN) is an intelligent agent capable of being used in the control
of nonlinear motions such as motions of a robot arm manipulator. ANN is capable of
providing better control ability than traditional methods. The proposed controller has the
ability to effectively utilize a large number of sensory information, can process data
collectively and is adaptive by default. Using Back Propagation, the ANN is trained to imbibe
the parameters of the robot arm manipulators for improved robot stability and suppressed
vibration during robot operation. Mathematical models of the ANN are presented. Developed
Simulink model is simulatedand simulation result analyzed. Training performance result of
0.024 Root Mean Square Error (RSME) reduction at epoch 2 was achieved. The result show
that trained network robot controller is capable of minimizing the system error to almost zero.
A hybrid arrangement could be more responsive for better stability as robot arm manipulator
controller.
Keyword: Artificial Neural Networks (ANN), Back Propagation,Simulink model,Root Mean
Square Error (RSME).
1.0 Introduction
In many cases, when it is difficult to obtain a model structure for a system with conventional
system identification techniques, the Artificial Neural Network (ANN), known as the
intelligent techniques, is desired since it is capable of describing the system in the best
possible way (Mersha, Stramigioli and Carloni, 2014) for further analyses and application
possibilities.The artificial neural network systems are commonly used for modeling nonlinear
dynamic systems. The main advantages of utilizing neural network for system identification
are that they simultaneously evaluate many points in the parameter space and converge
towards the global solution (Mersha, Stramigioli, Carloni, 2014).In contrast, neural network
approaches for system identification offer many advantages over conventional ones
especially in terms of flexibility and hardware realization (Erkaya, 2012). This technique is
quite efficient in modeling nonlinear systems or if the system possesses nonlinearities to any
degree.
Generally, robots are machines that behave and carry out tasks like a human being. Lately,
the industry is moving from automation to robot system, to increase productivity, reduce
waste and to deliver uniform quality. Robots are mainly deployed to hostile environment such
as in atomic plant to handle radioactive material, construct and repair space station and
satellites, nursing and aiding patient in medical field, heavy earth moving equipment and
many more.
The fundamental reason for engaging a robot is to eliminate the human operator. This reason
is not just centered on safety but to save labor and reduce cost. Other classes of applications
of concern are situations where human interference has negative impact especially on the
products such as in food handling, semiconductor handling, pharmaceuticals and so on. Also,
there are situations whereby the product has negative impact on the human such as;