Fault Diagnosis of Rolling Bearings by Integrating Laplace Wavelet Residual Network With Cauchy Kernel Maximum Mean Discrepancy Method

Abstract

Rolling bearings are critical components extensively applied in modern industries and play a significant role in ensuring the safety of high-speed rotating machinery systems. Accurate fault recognition of rolling bearings is essential for ensuring the safety of mechanical systems. This study proposes a method for diagnosing rolling bearing faults, utilizing a Laplace wavelet residual network integrated with a spatial attention mechanism (SAM) model and the Cauchy kernel-induced maximum mean discrepancy (CK-MMD) method (LRSDAN) to address the problems of complex variations in operational environments, different defect types, and differences in the distribution of collected data in real-world environments. This method incorporates a wavelet convolutional layer to comprehensively extract the shock components associated with bearing faults, employs a residual network (ResNet) to increase the model depth, and utilizes a SAM to extract the key vibration information. Subsequently, CK-MMD was applied as a metric to reduce the interdomain distribution differences and domain shift phenomena by an unbiased estimation technique based on the MMD. The model was validated on the datasets characterized by time-varying speed and variable load conditions. The investigation outcomes corroborate the reliability and superiority of the LRSDAN model in fault diagnosis performance through two publicly available datasets.