Rolling bearing fault diagnosis based on multi-channel convolution neural network and multi-scale clipping fusion data augmentation

Abstract

Abstract Deep learning has evolved to a prevalent approach for machinery fault diagnosis in recent years. However, the high demanding for training data amount refrains its implementation. In this study, we proposed a novel rolling bearing fault diagnosis strategy based on multi-channel convolution neural network(MCNN) combining multi-scale clipping fusion(MSCF) data augmentation technique. The fault signals were augmented using MSCF before transformed to time–frequency images through short-time Fourier transform, then the multi-sensor derived image data were fused by MCNN for feature extraction and fault pattern classification. Experiments validate that the combination of MSCF and MCNN is good at making the best of the information contained in each single sensor recording, leading to a significantly improved fault pattern classification accuracy and cluster effect. The proposed approach is low complexity but effective and robust, it is well suited for bearing fault diagnosis in case limited sensor data and/or variable working condition is presented.