Research on Bearing Fault Diagnosis Methods Based on Various Convolutional Neural Network Architectures

Bearing fault diagnosis is essential for ensuring the reliable and stable operation of mechanical equipment. While Convolutional Neural Networks (CNNs) have demonstrated significant advancements in fault diagnosis, challenges such as varying operational conditions and feature extraction complexity persist. This paper introduces an innovative method that synergistically combines multiple CNN architectures to enhance the accuracy and robustness of bearing fault diagnosis. Specifically, we design five distinct convolutional network structures: SerConv, ResConv, One-Shot Aggregation Convolution (OSAConv), Cross-Stage Aggregation Convolution (CSAConv), and Multi-Domain Dual-Aggregation Convolution (MD-DAConv). These architectures integrate multi-directional, multi-scale, and residual connections to optimize feature extraction depth and breadth. Furthermore, we implement advanced feature fusion and information transmission mechanisms to bolster the model’s generalization and resilience against data variability. Experimental results on the Case Western Reserve University (CWRU) bearing dataset and SJTU-SY bearing fault diagnosis dataset demonstrate that the proposed method significantly outperforms existing approaches, achieving higher diagnostic accuracy and enhanced robustness. The findings highlight the critical role of architectural diversity and effective feature integration in advancing bearing fault diagnosis, offering a comprehensive solution to current limitations in the field.