Fault diagnosis of motor bearing in complex scenarios based on Mamba and Indicative Contrastive Learning

Abstract

Motors are critical components of rotating machinery, and bearings are essential elements of motors. Therefore, monitoring the health of motor bearings is crucial. This paper addresses three major challenges in motor bearing fault diagnosis: noise interference, sample imbalance, and cross-domain diagnosis. To overcome these challenges and achieve high-precision fault diagnosis, This paper propose a novel method that integrates a Global-Local Fusion Feature Extractor (GLFFE) and Indicative Contrastive Learning (ICL). First, to mitigate the issues of noise interference and sample imbalance, design a Bidirectional Dual-Head Gated Mamba model (BDMamba) and a Residual Local Feature Extraction Module (RLFEM) for extracting comprehensive global and local fault features. The dual-branch architecture significantly enhances the extraction of local features while preserving the global modeling capabilities of Mamba. Next, proposed a Feature Fusion Mamba Module (FFMM) to effectively combine global and local features, thereby enriching feature diversity and reducing redundancy. To address the challenge of cross-domain fault diagnosis, integrate contrastive learning with working condition label prompts to enhance the model’s adaptability to multiple operating conditions. This approach also addresses the limitations of traditional contrastive learning, such as the requirement for large sample sizes. Finally, extensive experiments are conducted on two datasets to evaluate the robustness of the proposed method in cross-domain fault diagnosis under noise, sample imbalance, and variations in load and speed. The results demonstrate that the proposed method achieves high fault diagnosis accuracy and robustness and exhibits strong cross-domain transfer capabilities.