Feature Extraction from Raw Vibration Signal and Classification of Bearing Faults Using Convolutional Neural Networks

Abstract

Safe, efficient and uninterrupted continuous operation of an electric motor requires real-time condition monitoring of its rotating parts. Other than knowledge based signal analysis, fault feature extraction with statistical information or signal processing methods can be used to classify different fault patterns. But rule based feature extraction methods do not have domain adaptability, so the fault classification working in one system may not work for another system. A deep learning algorithm - Convolution Neural Networks approach is shown in this paper to classify different bearing faults and the trained network shows a good fault prediction capability for other systems.