Bearing Fault Diagnosis Using Structure Optimized Deep Convolutional Neural Network under Noisy Environment

Abstract

In recent years, deep learning has been gradually applied in bearing fault diagnosis thanks to its powerful learning and representation ability. However, deep learning-based methods are usually computationally intensive and have redundant parameters. In this paper, we propose an end-to-end structure optimized deep convolutional neural network for bearing fault diagnosis using the local sparse structure. The width and depth of the network are increased and the redundant parameters are reduced while maintaining the representation capabilities and performance of the network. The local sparse structure increases the computational efficiency and reduces the risk of overfitting. The proposed method directly employed the raw signals as input without extra feature extraction with the ability that could distinguish both the bearing fault types and the corresponding severity. Experiments show that the proposed method achieves similar performance compared to the original method by using 46.47% parameters as used in the original study, even under the noisy environment.