关键词: 微特电机, 轴承故障诊断, 故障类型识别, 注意力机制, 特征金字塔, 深度学习

Abstract: As the core driving component of precision control systems,micro-special motors face significant challenges in system reliability and safety due to the concurrent fault issues in their bearings. This paper addresses the limitations of traditional diagnostic methods in feature extraction and recognition accuracy by proposing a deep learning model based on a dual-channel attention mechanism combined with a feature pyramid structure. This model enhances key fault features through spatial and temporal dual-channel attention mechanisms, achieves multi-scale feature fusion using a feature pyramid,and incorporates an adaptive noise suppression module to improve robustness in high-noise environments. Based on a self-built micro-special motor bearing test platform, multi-condition data were collected, including single and compound faults of the inner ring,outer ring,and rolling elements. Experimental results demonstrate that the proposed model achieves an accuracy of close to 100% on the test set,with significant improvements in multiple performance metrics compared to traditional methods. The study validates the effectiveness of this approach in addressing the coupling of multi-fault features in micro-special motors under complex working conditions.