基于多域特征融合和迁移学习的跨机器轴承故障诊断方法OA

Domain shift induced by inconsistent data distribution among heterogeneous equipment severely re-stricts the generalization performance of cross-machine bearing fault diagnosis models.A multi-domain feature fu-sion transfer learning network was proposed to enhance the model's generalization capability on target machines through knowledge transfer.The network integrated a backbone-freezing strategy with low-rank adaptation to im-prove parameter-update efficiency while avoiding the computational overhead associated with full-network fine-tuning.A multimodal feature fusion method based on the Hilbert transform and Gramian angular fields was devel-oped to jointly capture analytic signal characteristics and steady-transient dynamics,thereby constructing more dis-criminative composite fault representations.Six cross-machine diagnostic tasks built from three independent datas-ets demonstrated that the method achieved an average accuracy of up to 98.5%,significantly outperforming several state-of-the-art approaches while reducing trainable parameters by more than 90%compared with conventional full fine-tuning.Experimental results further demonstrated that the method exhibited strong cross-machine generaliza-tion.It is well suited for industrial fault diagnosis under cross-domain conditions and provides a robust solution for predictive maintenance of manufacturing equipment.