Fault Diagnosis Algorithm Based on Mutual Information and Deep Learning

Aiming at the fault diagnosis of noisy signals in industry, this paper propose a fault diagnosis method based on mutual information analysis(MI) and multi-scale convolutional network(MS-CNN). First, for the problem of signal denoising, an algorithm based on mutual information analysis is proposed. After decomposing the noisy signal by complete ensemble empirical mode decomposition(CEEMD), the mutual information method is used to choose out noise-dominant IMFs for denoising. Then, for the fault diagnosis problem, multi-scale convolutional layer and a bottleneck layer is introduced to optimize the convolutional neural network. Finally, this paper uses Case Western Reserve University's hearing fault data set to simulate and compare with existing methods to verity the effectiveness of the proposed method.