Detection and classification of induction motor faults using feed-forward backpropagation network

Abstract

                  For this paper artificial neural networks technique is applied to identify broken rotor bar fault of single phase induction motor. Since artificial neural networks can be able to deal with non-linear problem, it is capable for apply to many applications in particular for recognizing patterns positively. Furthermore, fast Fourier transform (FFT) is employed for converting original stator current waveform, which is time domain, to stator current signal, which is frequency domain, that is labeled as motor current signature analysis for collecting essential data in order for sending into artificial neural networks later. Although performance of artificial neural network is related as many factors, three different multilayer neural network with several training algorithms are researched in this paper. Also three training algorithms are focused to study for exempla gradient descent algorithm, Levenberg-Marquardt algorithm and resilient back propagation algorithm. Consequently, the Levenberg Marquardt algorithm can perform very well for every different multilayer of neural networks in term the network mean squares errors (MSE) which having 5.25E-07 percent that comparing with other results of training algorithms.