Application of adversarial reciprocal point learning in open set fault diagnosis of rolling bearing

In industrial scenarios, bearing operating conditions are complex and unknown faults may occur unexpectedly, which usually reflect some new modes. This can result in the recognition failure of traditional intelligent algorithms based on the closed set assumption. To address this issue of open set fault diagnosis (OSFD), an OSFD approach of rolling bearing is proposed based on adversarial reciprocal point learning (ARPL) and efficient multi-scale attention (EMA). First, ARPL is introduced to diagnose the bearing faults under open set scenarios, which considers the deep distribution of unknown classes in learners by using an adversarial mechanism, achieving better open set recognition ability. Then, the EMA is employed to improve the open set classification performance of the ARPL model by interacting with information without channel dimensionality reduction. Finally, the effectiveness of the proposed method is evaluated on the bearing datasets. The experimental results show that the proposed ARPL-EMA model can effectively identify the unknown faults and its OSFD performance is superior to the comparative methods.