Research On Fault Diagnosis Method Of Analog Circuit Based On Manifold Structure

Nowsadays during the era of rapid intelligent development,analog circuits have been widely used in various fields,it is inevitable that there will be a variety of faults,so its diagnosis method is extremely important.However,due to various complex factors of analog circuit,the technical progress of analog circuit fault diagnosis is slow.Based on the development of machine learning algorithm and artificial intelligence technology,a fault diagnosis method for analog circuits using particle swarm optimization support vector machine(PSO-SVM)based on manifold structure and extreme learning machine proposed.(1)For the complexity of the analog circuit itself,the tolerance of components and the parameters of the fault are continuous,this paper uses the method of local mean decomposition and multi-scale fuzzy entropy to extract the fault feature information in the circuit.The local mean decomposition algorithm has advantages in dealing with non-stationary and nonlinear signals.Separate the pure FM signal from the envelope signal and multiply it to obtain a series of PF components,including the first three groups with the largest amount of information,and then analyze and calculate the multi-scale fuzzy entropy feature to obtain the feature vector.(2)When classifying and recognizing traditional support vector machines,only the information between data classes is considered,but the internal structure information of data is not considered.In this paper,a particle swarm optimization support vector machine based on manifold structure is proposed.The prior information of data distribution structure is fused to make the data as compact as possible when mapped to high-dimensional space on the basis of keeping the maximum spacing between data classes.At the same time,PSO and SVM are combined to optimize the parameters of the algorithm.PSO improves the iterative weight function,improves the iteration efficiency and avoids falling into local optimum,improves the classification accuracy and accelerates the rate of global convergence,which can effectively enhance the diagnosis effect of SVM.In view of the excessive number of faults,the traditional method will make the datatraining time too long and the diagnostic efficiency is relatively low.In this paper,the binary tree is combined with the extreme learning machine to construct the diagnostic model structure of the analog circuit.Compared with the traditional classification model structure,the binary tree structure can reduce the number of classifiers.At the same time,the input weight matrix and the implicit layer bias threshold of the limit learning algorithm are randomly given,and only a few hidden layer nodes of the network need to be set.The number produces a unique optimal solution.Compared with the traditional neural network and support vector machine,it not only accelerates the learning speed,but also improves the generalization ability of the model.