Bearing Fault Classification Based on Wavelet Transform and Artificial Neural Network

Abstract

In the present paper, hybrid bearing faults classification scheme based on wavelet transformation and neural network is proposed. Basically, the proposed methodology identifies four different types of bearing faults. For classification of the faults, vibration signals have been utilized. The vibration signals are first decomposed into components in different sub-bands using discrete wavelet transformation. Subsequently, variance and variance of autocorrelation value extracted from decomposed signal have been used as input features for the neural network. The time interval between the impacts of original signal is also exploited to characterize the bearing vibration signals. A neural network follows to classify the extracted feature vector. Trained neural networks are able to classify different states of the bearing with 100% accuracy. The proposed procedure requires only a few input features, resulting in simple preprocessing and faster training. Effectiveness of the proposed method is illustrated using the bearing vibration data obtained experimentally.