Nonlinear analog circuit fault diagnosis has attracted attention of scholars asits important role in circuit testing field. Pattern recognition is the core of thediagnosis in intelligent fault diagnosis. The original feature values obtained bycircuit pattern recognition have a lot of redundant information, if all these featurevalues are used for fault diagnosis, then the boundary of fault classification isfuzzy and the data calculation is large. Therefore it becomes an urgent problem inhow to extract the circuit feature information. For this problem, the paper usesWiener kernel to describe nonlinear analog circuit and extract Wiener kernel faultfeatures using improved particle swarm simulated annealing algorithm to achievethe efficient and accurate circuit fault diagnosis.This paper discusses the Wiener kernel generic function description method ofnonlinear analog circuit, and researches the acquisition method of Wiener kernelwhich is based on Wiener kernel and intelligent diagnosis principle of BP neuralnetwork. At the same time, this paper gives the design method of improved BP neuralnetwork by adjusting adaptive learning rate.We need to choose some frequency points which are significantly different fromother fault state kernel when we use the Wiener kernel to describe the non-linearanalog circuit. Take these frequency points as the feature parameters can we havehigh accuracy in the diagnosis. The paper studies the improved particle swarmsimulated annealing algorithm for choosing and extracting Wiener kernel featurevalues. Through three testing functions and Wiener kernel feature extractionexample, we can verity that IPSSAO is superior to the other two algorithms in bothoptimization accuracy and speed in the same initial population number and samemax generation, and we can also verity that the method of Wiener kernel selection and extraction based on particle swarm simulated anneal algorithm is feasible and itcan extract the best fault feature vectors which can reflect fault features.After theory study and simulation, this paper designs a fault diagnosis systembased on ATmega128microcontroller, it contains signal acquisition unit, datastorage unit and data communication unit and so on. The PC application softwareis also designed to realize the acquisition of Wiener kernel, kernel featureextraction and neural network training. Use the established system to measure theactual fault circuit, we can accurately get the fault devices and fault types, and verifythe correctness and effectiveness of the feature extraction optimization method. |