关键词: 电机轴承, 故障诊断, 多尺度特征提取, 生成对抗网络, 自适应位置纠正策略, 双通道

Abstract: A dual-stream adaptive network motor bearing fault diagnosis method was proposed to address the low accuracy caused by existing methods relying on single feature transformation and basic data fusion strategies. A dual-spectrum feature transformation strategy was integrated to reconstruct high-dimensional representations of the global and local features of vibration signals through multi-scale feature extraction. A discrete dual-channel structure was designed to learn these features,and a generative adversarial training mode was adopted to achieve data augmentation and feature comprehensive analysis. Subsequently,an adaptive position correction strategy was developed to merge feature information from two channels,promoting self-correction and optimization during training. Experimental results demonstrated that the proposed method could effectively extract key features from motor bearing operation data,achieving an accuracy of 98.3 % on multi-category motor bearing fault datasets,outperforming five other mainstream fault diagnosis methods.

Key words: motor bearing, fault diagnosis, multi-scale feature extraction, generative adversarial networks, adaptive position correction strategy, dual-branch