Research on AGV fault diagnosis based on 1DSE-SENET

Considering the frequent sudden failures of industrial AGVs and the delay problem of traditional maintenance response, this study proposes a 1DCNN-SENet (1D-SE) method for AGV fault diagnosis. Considering the non-smooth characteristics of fault state data, a multimodal processing flow is designed: 1D time domain data is generated based on variational modal decomposition (VMD) to highlight the details of the fault region; 1D convolutional network (1DCNN) and temporal attention mechanism (SENet) are fused, and the temporal dependency is captured by the 1DCNN, and the key fault channels are adaptively amplified by the SENet to achieve the accurate classification of rotor faults and bearing aging and the accurate composite fault classification. The results show that the 1D-SE model proposed in this paper has the highest accuracy and stability with an accuracy rate of 96.7%, which verifies the effectiveness of the method.