A Bearing Fault Diagnosis Method Based on Autoencoder and Particle Swarm Optimization – Support Vector Machine

Abstract

Rolling element bearing is an important part of rotary machines. Bearing fault is a big issue because it can cause huge cost of time and money for fixing broken machines. Thus, early detecting fault of bearing is a critical task in machine health monitoring. This paper presents an automatic fault diagnosis of bearing based on the feature extraction using Wavelet Packet Analysis, feature selection using Autoencoder, and feature classification using Particle Swarm Optimization - Support Vector Machine. First, bearing vibration signals are decomposed at different depth levels by Wavelet Packet Analysis. Then the wavelet packet coefficients are used to compute the energy value of the corresponding wavelet packet node. After that, an Autoencoder is exploited to select the most sensitive features from the feature set. Finally, classification is done by using a Support Vector Machine classifier whose parameters are optimized by Particle Swarm Optimization. The effectiveness of the proposed intelligent fault diagnosis scheme is validated by experiments with bearing data of Case Western Reserve University bearing data center.