Intra-domain self generalization network for intelligent fault diagnosis of bearings under unseen working conditions

In recent years, domain generalization fault diagnosis methods have effectively addressed the challenges of bearing fault diagnosis under unseen working conditions. Most existing approaches rely on training across multiple source domains to learn domain-invariant representations. However, collecting comprehensive fault monitoring data across various working conditions is a daunting task. This severely limits the practical application of existing methods. Faced with the common scenario where available data originates from a single working condition, this paper proposes an intra-domain adversarial network (IDAN) for bearing fault diagnosis based on self generalization. Firstly, leveraging multi-scale branches and an improved adversarial learning mechanism, a perspective sharing strategy is introduced to ensure the extraction of generalized fault representations surpassing the constraints of perspectives. In this process, semantic diagnostic knowledge inherent in multi-scale features is refined through inter-scale confusion. Additionally, a collaborative decision strategy is designed to achieve the ultimate optimization of decision boundaries. By reinforcing and aligning the classification boundaries of different branches, the model's generalization performance is further enhanced. Finally, extensive generalization diagnostic experiments conducted on three datasets validate the effectiveness of the proposed approach.