Cross-Domain Diagnosis of Mechanical System Typical Faults Based on Pretrain-Finetune Machine Learning Model

In recent years, machine learning driven intelligent methods have achieved excellent performance in mechanical fault diagnosis. However, the domain shift problem leads to severe diagnostic accuracy degradation, limiting the application and development of machine learning intelligent diagnosis. In this study, we propose a machine learning cross-domain diagnosis method based on Pretrain-Finetune. It achieves diagnostic knowledge migration by freezing part of the feature extraction layer of the pre-trained network, followed by fine-tuning to achieve data feature alignment on a small sample target domain. The study also focuses on the effect of the number of frozen network layers on cross-domain diagnostics. The proposed method is validated in a cross-condition dataset based on vibration data from piston aero-engines. The results show that it exhibits good generalization ability and reduces the dependence on the number of samples in the target domain, showing good potential for engineering applications.