An Incipient Fault Diagnosis Method Based on Spatio-Temporal Center Network for Analog Circuits

With the rapid development of electronic technology, accurately identifying the incipient faults of analog circuits has become an important measure to improve the reliability and safety of electronic equipment. In recent years, deep learning is extensively applied to fault diagnosis because of its powerful feature mining ability. Therefore, a method based on spatio-temporal center network (STCN) is proposed to identify incipient faults for analog circuits, which includes a feature extraction module and a classification module. In the former, a spatio-temporal backbone network is designed to comprehensively mine the effective feature representation, including multi-scale spatial information and temporal information in the response signals of analog circuits. In the classification module, the spatio-temporal feature representation is imported into the Softmax layer for fault identification. Finally, in addition to the commonly used cross entropy loss, the central loss is also constructed for the STCN model. By reducing the intra class distance among similar feature representations, the discrimination of feature representation is further improved. In order to assess the effectiveness of the proposed method, the Sallen-key bandpass filter circuit is selected for experimental verification. Experimental results indicate that STCN is superior to some typical fault diagnosis approaches in incipient fault diagnosis of analog circuits.