Fault Diagnosis And Prediction Of Analog Filter Based On Deep Learning

Analog circuits are widely used in many important fields,such as aerospace,automatic control and communication.The analog circuits have higher requirements for fault feature extraction and diagnostic prediction due to their nonlinear and tolerance effects.Traditional analog circuit fault diagnosis and prediction methods mostly require manual feature extraction and many expertise in the field of electronic circuits,and have gradually failed to meet the health management requirements of complex analog circuits.Therefore,this paper takes analog filter as a typical analog circuit as the research object,and proposes an incipient fault diagnosis and fault indicator(FI)construction method based on deep learning method,and the remaining useful performance(RUP)prediction for analog filter is further researched based on obtained FI.The main research contents of this paper are as follows:(1)An incipient fault diagnosis method for analog filters based on Deep Neural Network(DNN)is proposed.The method makes full use of DNN's powerful adaptive feature extraction ability and nonlinear function processing ability,and completes the deep extraction of analog filter circuit fault features by stacking autoencoders,and uses SOFTMAX classifier to complete fault classification.(2)A method based on deep learning to construct the fault indicator of analog filter is proposed.The Deep Neural Network is used to realize the direct mapping from the circuit time domain response data to the circuit degradation state and construct the FI curve.(3)Based on the constructed analog filter fault indicator,the particle filter algorithm(PF)is used for RUP prediction.(4)Using the classic Sallen-Key circuit and Leapfrog circuit as the verification circuit,the above methods were tested and analyzed through simulation experiment and actual verification.The experimental results were compared with the related literature results.The experimental results show that the early fault diagnosis method based on DNN can obtain higher fault recognition rate.This method relies less on manual participation and does not require expertise in the field of electronic circuits and signal processing knowledge.The DNN-based unsupervised FI construction method can obtain a smooth degradation curve with better correlation and monotonicity.On the basis of constructing the FI curve,the PF algorithm is used to the RUP prediction of the analog filter,and the prediction result has a smaller prediction error.The validity of the FI curve based on DNN and the advantages of the proposed method are further verified.