A new interdomain feature perception enhancement strategy for rotating machinery fault diagnosis

Abstract Industrial equipment operates under complex, varied conditions, and the sensor data from rotating machinery such as bearings often exhibit substantial distributional differences. These variations pose substantial challenges to the cross-domain generalization ability of fault diagnosis models. Traditional methods often rely on target-domain data or fail to address the disparities between source domains, which leads to notable performance drops when encountering unseen operational scenarios. As a result, enhancing model robustness across unknown domains has become a critical challenge in intelligent fault diagnosis. The interdomain feature perception enhancement (IFPE) strategy is proposed to address these issues. IFPE incorporates a feature perception module to adjust inputs from heterogeneous source domains dynamically and focuses on regions critical for fault detection. Additionally, IFPE employs a dynamic weighted fusion mechanism to aggregate discriminative features by evaluating the contributions of individual source domains. This strategy results in a unified representation that balances domain consistency and task relevance. Extensive experimental evaluations demonstrate the effectiveness of IFPE, which consistently outperforms baseline methods across diverse cross-domain tasks, including scenarios where target-domain data are entirely unavailable. The strategy maintains superior diagnostic accuracy and adaptability, which highlights its potential for practical applications in intelligent fault diagnosis.