Bearing fault diagnostics using EEMD processing and convolutional neural network methods

Abstract

The development of an intelligent fault diagnosis system to identify automatically and accurately micro-faults affecting motors continues to be a challenge for industrial rotary machinery and needs to be addressed. In this paper, we put forward a novel approach based on ensemble empirical mode decomposition (EEMD) processing for incipient fault diagnosis of rotating machinery. Accurate selection and reconstruction processes are performed to reconstruct new vibration signals with less noise through the application of EEMD processing to original vibration signals. After the rebuilt of vibration signals, manually extracted features from the reconstructed vibration signals are fed then into a multi-class support vector machine and simultaneously to the mentioned technique, generated image representations of the same raw signals are taken afterward as an input to a deep convolutional neural network (CNN) for classification and fault diagnosis. The comparison between these developed methods demonstrates the effectiveness of the deep learning approach that identifies the differences between classes automatically and can successfully classify and locate the faulty bearing status with very high accuracy for the small size of training data.