Research On Fault Diagnosis Of Analog Circuit Based On Optimized Artificial Intelligence Algorithm

In the analog-mixed-signal circuits(AMS)test,the analog circuit test cost accounts for more than 95% of the total test cost,and most of the faults occur in the analog part.At present,the method of digital circuit fault diagnosis in AMS has matured and has been applied in practice.However,due to the complexity and nonlinearity of analog circuits,the development of analog circuit fault diagnosis is slow and has become a difficult problem in AMS testing field.With the improvement of circuit integration and complexity,the traditional analog circuit fault diagnosis method has been difficult to meet the test requirements,and the artificial intelligence method has become a research hotspot.In this paper,the artificial intelligence method as the means,the Sallen-key bandpass filter and CTSV filter circuit as the measured objects,we do a research on analog circuit fault diagnosis.The main contents of the paper are as follows:1.Research on feature extraction method for analog circuits based on wavelet packet kurtosis and Neighborhood preserving embedding(NPE).This paper studies three parts of the analog circuit feature extraction method,including feature extraction method,characteristic parameters and dimension reduction method,and different parts are combined.Based on this,wavelet packet kurtosis and NPE feature extraction method is proposed.The method uses wavelet packet transform to extract circuit feature,kurtosis as characteristic parameter,neighborhood preserving embedded compressed data.thus avoiding circuit noise interference,maintaining the internal manifold of the data,and has better ability of circuit fault extraction.The experimental results show that the wavelet packet kurtosis and NPE method can improve the diagnostic accuracy by at least 15% compared with traditional methods,which proves the superiority and universality of the proposed method for analog circuit fault extraction.2.Analog circuit fault diagnosis based on cloud model optimize probabilistic neural network(PNN).For large sample classification problem of analog circuits,this paper chooses PNN to solve the problem,and multi-dimensional normal cloud model is used to optimize the parameters of probabilistic neural network.Training samples under different fault modes are constructed as multi-dimensional normal cloud models through the peak cloud transformation,which is used as the mode layer neuron of the probabilistic neural network.Secondly,the weight between the mode layer and the summation layer is obtained by using the peak value.Finally,the optimized probabilistic neural network is used for fault classification.The method determine the number of neurons in the hidden layer,and use the peak value in the peak cloud transform to obtain the connection weight,optimize the three parameters of the PNN and simplify the training process.The diagnostic example show that the diagnostic accuracy of this method has reached more than 99%.3.Analog circuit fault diagnosis based on cloud based evolution algorithm optimize Support Vector Machine(SVM).For small sample classification problem of analog circuits,this paper chooses SVM to solve the problem,and Cloud Based Evolutionary Algorithm(CBEA)is used to optimize the parameters of SVM.The cloud based evolutionary algorithm uses expectation(Ex)as the parent,entropy(En)and hyper entropy(He)to control the search direction,and through a series of operations such as local refinement,local variation and mutation,it searches for the optimal parameters of the support vector machine.The diagnostic examples show that the method avoid the problems of the traditional optimization algorithms such as local optimization and random walk,and it has higher convergence ability,and improving the accuracy of fault diagnosis.The diagnosis examples show that the diagnostic accuracy is over 98%,which is about 8% higher than the traditional support vector machine diagnosis method.