A Fault Diagnosis Method for Dry Vacuum Pump Bearing Based on Finite Element Simulation With Deep Transfer Learning

Dry vacuum pumps play a vital role in chip manufacturing, and failure of their core components, bearings, can lead to severe downtime and economic losses. Traditional fault diagnosis methods face challenges due to difficult data collection and inconsistent working conditions. In this paper, a dry vacuum pump bearing fault diagnosis method based on finite element simulation and deep transfer learning is proposed. First, finite element simulation is used to model the bearing dynamics and create a fault dataset to address data collection challenges and verify the feasibility of the method. Secondly, an attention mechanism is introduced in the feature extraction layer, combined with a deep transfer learning framework based on domain adversarial networks, to optimize feature extraction and align source and target domain features, thereby improving the accuracy of fault diagnosis and model generalization ability. Experimental results show that the proposed method performs well on multiple datasets, especially when there is a large speed difference between domains, as shown in the PU dataset, and the accuracy is significantly improved compared with the deep transfer learning methods DANN and DDTLN. The t-SNE visualization technique is used to further confirm the advantages of the method in cross-conditional fault diagnosis.