Soft Fault Feature Extraction Method And Diagnosis System Design For Multi-port Network

With the progress of The Times,the scale of circuit integration is increasing,and the number of detectable nodes in the circuit is decreasing.There are complex problems such as tolerance and soft fault in nonlinear analog circuits,which make fault diagnosis of nonlinear analog circuits more difficult.Due to the complexity of multiple input multiple output(MIMO)nonlinear analog circuits,it is relatively difficult to establish soft fault models.At present,the fault diagnosis methods of MIMO nonlinear analog circuits are still in the theoretical stage and need to be further improved in practical application.In this dissertation,the fault diagnosis theory and key technologies of MIMO nonlinear analog circuits are studied.The purpose is to improve the soft fault diagnosis theory of nonlinear analog circuits in multi-port networks and complete the design of diagnosis system.Based on the theory of Volterra functional series,a Volterra kernel model for soft fault of MIMO nonlinear analog circuits is presented.Under the excitation signal of multi-tone test,the diagnosis system can realize the purpose of multi-point identification and improve the identification efficiency of Volterra nuclei.However,the same frequency components will occur in different orders of Volterra nuclei in MIMO nonlinear system,which makes it difficult to distinguish different orders of Volterra nuclei and affects the effect of fault diagnosis.In order to solve this problem,the excitation signals of multitone test were optimized by the optimal search theory to find the excitation signals without overlapping frequency components,and the Volterra kernel of each order of the system was solved by Vandermonde method.In the aspect of soft fault feature extraction,the theory and key technology of soft fault feature extraction of nonlinear analog circuits are studied,and an improved lenet-5 network feature extraction method is proposed.This network improves the extraction of detail features by deepening the network structure,using a smaller convolution kernel,and combining Dropout technology to prevent overfitting.Re LU function is selected as the activation function to prevent gradient disappearance.The adaptive global average pooling(GAP)layer is used to replace the full connection layer to reduce the number of network parameters.An optimization example is given.In the aspect of fault recognition,the principle and key technology of fault state recognition method of analog circuit are studied.Support vector machine(SVM)is used for fault state recognition to improve the generalization ability of soft fault diagnosis method.An improved CNN+SVM soft fault diagnosis method is proposed,and an optimization example is given.Experimental results show that the improved CNN+SVM soft fault diagnosis method has higher accuracy than the traditional CNN+Softmax method under the same number of iterations.On the basis of theoretical research,according to the demand of the diagnosis system for system design,complete system hardware platform design and components selection,to complete the design of the system software,PC software main implementation is the function of network training and fault diagnosis,and through to the mixer circuit fault diagnosis,verify the practicability and validity of system.Experiments show that the proposed fault diagnosis method is effective and feasible,the system can accurately diagnose the soft fault state of the mixer circuit,and the designed system has a certain practical value in practical applications.