Fault Diagnosis Method Of Analog Circuit Based On DGCCA With Multiple Feature Set Fusion

Analog circuits are an important part of integrated circuit systems,and circuit systems are the basis for the normal operation of electronic equipment,so it is necessary to efficiently diagnose and maintain analog circuit faults.However,due to the problems of tolerance,high nonlinearity and environmental interference of analog circuit components,the research progress of fault diagnosis technology is slow and cannot meet the practical requirements of today’s highly reliable and reliable electronic equipment.With the increasing scale and integration of the circuit,how to effectively and as much as possible to extract more discriminative fault characteristics is the key research direction of analog circuit fault diagnosis.In this paper,based on theories and techniques related to Empirical Mode Decomposition,Composite Multiscale Sample Entropy,Deep Generalized Canonical Correlation Analysis,and Extreme Learning Machine,we construct feature sets by two ways of fault feature extraction,and investigate fault mode classification strategies.First,the experimental circuit to be diagnosed is built to obtain fault data according to the analog circuit fault diagnosis process.In this paper,Sallen-Key band-pass filter and Four-opamp dual quadratic high-pass filter are selected as the experimental circuit in the analog filter,the excitation source is determined,the circuit simulation schematic is drawn and the nominal values and tolerances of the circuit components are set,the fault components are selected according to the circuit sensitivity analysis,the fault mode is set for each fault component according to the soft fault criteria,and the Monte Carlo analysis is performed to obtain the original fault data set.Next,the obtained raw fault dataset is subjected to fault feature extraction.In this paper,the fault features are obtained in two ways and combined to form the final fault feature set:(1)a fault feature extraction method based on Empirical Mode Decomposition combined with Composite Multiscale Sample Entropy.To address the problems of parameter continuity,tolerance characteristics and nonlinear and non-smooth fault response of analog circuit components,the EMD method is used to decompose complex fault signals adaptively and obtain a single-frequency IMF component set that can reflect the inherent change trend of the signal;then the composite multi-scale sample entropy value of each IMF component is calculated.The result(EMD＿CMSE)can more accurately characterize the self-correlation and complexity of time series at different time scales.(2)Multi-feature fusion extraction based on Deep Generalized Canonical Correlation Analysis.In order to obtain as much fault information as possible,the time and frequency domain characteristic parameter feature sets are constructed separately for the purpose of the widespread use of large scale highly integrated circuit systems today,which have few measurable points leading to the limitation of the data that can be measured.Also for a large number of more discriminative fault features,the DGCCA method is facilitated for feature extraction of the time and frequency domain feature sets,and the fused feature shared representation matrix G is obtained by transforming the fault labels as additional views into a supervised fusion and dimensionality reduction process for multiple features.Finally,the fault classification model is built for training and validated for example diagnosis.In this paper,the Extreme Learning Machine(ELM)is selected as the fault classifier for subsequent fault mode diagnosis.The model parameters of the ELM classifier are determined by parameter iteration,and the better number of hidden layer neurons is determined after several rounds of training based on the average diagnosis rate results of the model.Using the trained fault c classifier for fault diagnosis of the experimental circuit,the correct diagnosis rate is 99.16% for the Sallen-Key experimental circuit and 99.48% for the Four-opamp experimental circuit.The EMD＿CMSE matrix alone and EMD＿CMSE+Time-Frequency domain feature set are set as the control group for fault diagnosis test,and the final comparative analysis results show that the EMD＿CMSE+G feature set construction method proposed in this paper can effectively improve the fault diagnosis accuracy.