A Dual-input Fault Diagnosis Model Based on Convolutional Neural Networks and Gated Recurrent Unit Networks for Analog Circuits

Abstract

To improve the reliability and safety of complex electrical systems, an end-to-end fault diagnosis method for analog circuits is proposed in this paper. First, by combining the convolutional neural networks (CNN) and the gated recurrent unit (GRU) networks, a feature extraction model based on CNN-GRU is developed to obtain information that characterizes the essential states of the circuit under test (CUT) from the its signals. Compared with traditional feature extraction methods, the CNN-GRU model can obtain the spatial features of signals while retaining the time sequence features. Then, a dual-input structure of the time domain and frequency domain is designed for the CNN-GRU model, and the time-frequency domain fusion features of the signals are obtained by using the dual-input fault diagnosis model based on CNN-GRU, thereby fully reflecting the circuit states. The Sallen-Key bandpass filter circuit in ISCAS'97 circuit set is adopted to comprehensively evaluate the proposed method. Experimental results prove that the proposed fault diagnosis method can implement the accurate identification for incipient single fault classes and double fault classes.