Domain Adaptation With Multi-Adversarial Learning for Open-Set Cross-Domain Intelligent Bearing Fault Diagnosis

Adversarial domain adaptation and transfer learning have been widely applied in the field of cross-domain fault diagnosis. However, the effectiveness of existing domain adaptation-based diagnostic methods relies on the assumption that both the source and the target domain data share the same label space. In practice, it is impossible to predict the failure mode during testing, and new failure types may appear in the target domain samples. This is an open-set fault diagnosis issue. To address this problem, we propose a multi-adversarial learning domain adaption (MALDA) model for open-set cross-domain intelligent bearing fault diagnosis. The transferable features and target sample weights are obtained in adversarial learning. By introducing a transfer weight conditional adversarial network to align the joint feature-category distributions and obtain a transferable index, the identifiable predictive information from the classifier output to further adjust and optimize the model. Selective interterritory distribution alignment is achieved by weighted adversarial learning networks, and domain partition adversarial learning can accurately identify shared health states and unknown failure modes. The validity and practicality of the proposed MALDA model are validated by three experimental cases.