Fault Diagnosis and Testing of induction machine using Back Propagation Neural Network N.Rajeswaran 1 , T.Madhu 2 and M.Surya kalavathi 3 1 JNTUH/SNS College of Technology, Department of ECE, Coimbatore, Tamilnadu, India 2 Swarnandhra Institute of Engineering and Technology, Narasapur, Andhrapradesh, India 3 Jawaharlal Nehru Technological University, Hyderabad, Andhrapradesh, India ABSTRACT The recent developments with AI (Artificial Intelligence) are extremely intricate and are useful in a wide range of domestic and industrial applications. In real time environment, operating the induction motor at variable speeds is a severe constraint. The electrical and mechanical faults can impose unacceptable conditions and protective devices are therefore provided to quickly disconnect the motor from grid. In order to ensure that electrical machines receive adequate protection, extensive testing is performed to verify the high quality of assembly. Fault diagnosis and testing of induction machine is attempted under various load conditions and verified by using Field Programmable Gate Array (FPGA). Back Propagation Neural (BPN) Network is used to calculate the error and correct/regulate the induction motor. This technique has resulted in increased speed and improved fault coverage area of the induction machine. Index Terms — AI, BPN, Fault diagnosis, FPGA 1 INTRODUCTION An induction or asynchronous motor is a type of AC motor where the power is supplied to the rotor by means of electromagnetic induction. These motors are widely used in industrial drives, particularly polyphase induction motors, due to their ruggedness. Single-phase versions are used in small appliances. Although most AC (Alternating Current) motors have long been used in fixed-speed load drive service, they are increasingly being used in Variable-Frequency Drive (VFD) service. The testing and detection of faults in induction machine is not an easy task as constant speed of the motor is to be maintained. The motor speed is varied due to the inability to obtain speed below the synchronous speed, instability at higher speeds because of armature reaction and commutation difficulties. If the motor speed is increased above the rated level, the commutator will be damaged. The speed control and fault diagnosis are both performed in a comparative manner [1]. The existing methods for speed control of induction motor are pole changing, stator voltage control(V), rotor resistant control(R), stator flux(ø) and supply frequency control(f). The speed of the induction motor is tested and controlled by Space Vector Pulse Width Modulation (SVPWM) technique with various levels of switching conditions [4]. The induction motor faults are mainly classified into electrical and mechanical faults. The electrical faults are included in turn in short circuit in stator winding, open circuit in stator windings, broken rotor bar and broken end ring. The mechanical faults include bearing failures and rotor eccentricities. The statistical data shows that among variety of faults that occur in the induction motor, stator winding fault and short circuit between winding conductor and stator core is about 30-35% and rotor and shaft fault is about 10%. In this paper we focused on the stator and rotor faults of the induction motor with various load conditions for the purpose of synthesis and simulation on Field Programmable Gate Array (FPGA) using VHDL (VHSIC Very High Speed Integrated Circuits) Hardware Description Language. 2 BACK PROPAGATION NEURAL NETWORK The adaptive Artificial Neural Networks (ANN) are part of the Artificial Intelligence technique that attempts to imitate the way a human brain works with the help of a mathematical or computational model[3]. The neural network algorithm which steps back one layer (hidden layer) from the output layer and recalculates the weights of the output layer so that the output error is minimized is called as Back Propagation Network (BPN). The learning algorithm uses the gradient descent technique to minimize the output error i.e. the total squared error of the output computed by the network. Through supervised learning, the multilayered artificial neural network is systematically trained by the BPN. The obtained outputs are subtracted from the desired outputs to check whether an error signal is produced or not. If any error is found, then this error signal is sent back to the hidden layer. In each hidden processing unit the corresponding adjustment is needed to produce the correct output. The BPN is used in different areas like Image compression, problems in power system area, control system and fault detection etc., 978-1-4673-1225-7/12/$31.00 ©2012 IEEE 492