Abstract

Despite the recent success of deep-learning-based fault diagnosis of rotating machinery, to enable accurate and robust diagnosis models, existing approaches proceed with the assumption that training and test data follow the same distribution. However, in practical industrial settings, variations in operating conditions and environmental noise can cause changes in the characteristics of the training and test data, called domain shift, resulting in performance degradation of the test data. To deal with these issues, this paper proposes an asymmetric inter-intra domain alignments (AIIDA) approach for fault diagnosis under various levels of domain shift. First, inter-domain alignment is conducted by minimizing the maximum mean discrepancy loss and domain adversarial loss. Next, intra-domain alignment is performed by adjusting the inconsistency loss. This approach allows the proposed AIIDA method to learn features that have lower inter-domain distance and higher intra-domain distance; thus, the fault diagnosis performance in the target domain can be significantly improved. Extensive experimental assessment that examines various scenarios across three bearing datasets is performed to validate the effectiveness of the proposed approach. Furthermore, a study comparing the proposed method with other existing methods demonstrates that the proposed method outperforms other methods.