Development of ANN for Diagnosing Induction Motor Bearing Failure

Abstract

Today, various diagnosis methods exist to diagnose bearing failure in industries such as artificial neural network (ANN), fuzzy logic, genetic algorithm and statistical analysis. This paper proposes the development of ANN model of induction motor bearing (IMB) fault diagnosis using different network types and structures. In this case Feedforward Neural Network (FFNN) and Elman Network types with Multiple Input Single Output (MISO) and Multiple Input Multiple Output (MIMO) structures were studied. The raw data used in this work was obtained from the Case Western Reserve University website in form of vibration signal. During pre-processing stage, Fast Fourier Transform (FFT) and enveloping techniques were applied to raw data before it was fed to features extraction stage. A total of 16 features were extracted from both time and frequency domain respectively. Subsequently, a distance evaluation technique was used for features selection, where 9 salient features were selected for ANN fault diagnosis. In the development of ANN fault diagnosis, FFNN and Elman Network were utilized with training algorithm of Levenberg Marquart Backpropagation. The result indicates the performance of classified IMB fault by using MIMO Elman structure which was better compared to other combination structures.