An Automated Bearing Fault Diagnosis Using a Self-Normalizing Convolutional Neural Network

Abstract

A Self-normalizing Convolutional Neural Network (SCNN) algorithm can have a much faster convergent rate than that of a traditional Convolution Neural Network (CNN) as the former does not require a Batch Normalization (BN) during the network training process. SCNN is employed in this study for an automated bearing fault diagnosis based on the frequency domain fault feature extracted from acoustic emission signals acquired from a bearing test rig. In the process, a fast Fourier transform is employed first to transform the time domain signal into the frequency domain, the spectra are then used as the samples to train the SCNN model. The trained network is utilized to identify various bearing conditions under both constant and varying speeds. It is shown that the proposed technique can achieve a 100 percent recognition rate in the constant speed case and a 99.4 percent accuracy in the varying speed case. It is also showed in a comparison study that SCNN can have a much faster fault recognition rate than the traditional CNN.