关键词: 电机轴承, 故障检测, 剪切波变换, 非线性流形, 边界约束, ConvNeXt-T

Abstract: To improve the accuracy and robustness of motor bearing fault detection under complex working conditions,a motor bearing fault detection method using improved ConvNeXt-T was proposed. Firstly,a multi-scale time-frequency synergy strategy using shearlet was adopted to perform multi-scale and multi-directional time-frequency domain conversion on vibration signals,and a two-dimensional image suitable for deep convolutional network learning was constructed by extracting local transient features. Then,by introducing non-linear manifold fusion mechanism,the multi-layer features were adaptively fused,strengthening the fusion relationship between shallow and deep features,and the feature expression ability was improved. Finally,the boundary constrained classification optimization strategies and accelerated optimization strategies were proposed to enhance the discriminative power between categories,and the inference speed of the model was improved through mixed precision training and structured model pruning. The experimental results show that the proposed improved method has stronger fault feature extraction ability and higher classification accuracy. The detection accuracy,recall rate,mAP,and inference speed have reached 95.8 %,94.2 %,94.9 %,and 36.7 fps,respectively,which are 3.2 %,3.1 %,3.1 %,and 11.1 fps higher than the original ConvNeXt-T model,providing strong technical support for the automation and intelligent diagnosis of motor bearing faults under complex working conditions.

Key words: motor bearing, fault detection, shear wave transform, nonlinear manifold, boundary constraints, ConvNeXt-T