BEARING FAULT DIAGNOSIS BASED ON DEEP LEARNING AND ARRAY STOCHASTIC RESONANCE UNDER STRONG NOISE BACKGROUND

The diagnosis of bearing faults is important in the maintenance and running of industrial machinery. As the signal collection process is often interfered by noise, the accuracy of the diagnosis is reduced. Therefore, bearing fault diagnosis is always a challenging problem under strong noise background. This paper develops a new method that utilizes a combination of deep learning and array stochastic resonance to improve the robustness of models in the context of strong noise. By introducing a multi-branch dilated convolutional structure in the residual neural network, network performance is improved. However, the model may be disturbed and lead to performance degradation due to the presence of a strong noise background. To address this problem, an array stochastic resonance is introduced to help the network better explore the potential feature space during the training process. Array stochastic resonance enhanced the robustness of the network by introducing randomness to improve the accuracy. Experimental results show that our approach can dig deeper into defect features and has strong noise resistance. In a strong noise background with SNR of −7 dB, the accuracy can be more than 97.8%. It has a higher recognition performance compared to various deep learning algorithms. Compared to various deep learning algorithms, it has higher recognition performance. © 2025, ICIC International. All rights reserved.