Intelligent Identification of Bearing Faults Using Time Domain Features

Abstract

An approach is proposed for fault diagnosis of rolling element bearings using time domain features as inputs to the artificial neural network (ANN). The time domain features are extracted from the segments of the experimental dataset for known machine conditions. The dataset has been subjected to somewhat preprocessing previous to feature extraction. The ANN consists of five input nodes, one hidden layer with five nodes and four output nodes. Each of five input nodes represents root mean square, variance, skewness, kurtosis and normalized sixth central moment of the time domain vibration signals, respectively. Four binary nodes in the output layer specify the bearing condition: normal, outer race defect, inner race defect or ball defect. The ANN is trained using back propagation algorithm with a subset of the time domain features. The ANN is tested using the remaining set of the time domain features. Training and test success are used to evaluate efficiency of the presented method. The results indicate the effectiveness of the time domain features in diagnosis of bearing failures with high accuracy and low computation.