Multi-model Analog Circuit Fault Diagnosis Based On Particle Swarm Optimization Optimization Support Vector Machine

With the development of electronic equipment manufacturing technology in modern society,the complexity and integration of modern electronic equipment are increasing,and the demand and difficulty of fault diagnosis of analog circuit are also increasing.However,these analog circuits have few measurable nodes,coupled with the widespread existence of device tolerance and nonlinear devices,the complexity and diversity of analog circuit faults are becoming more and more prominent,making the development of analog circuit fault diagnosis slow.This paper mainly studies the analog circuit fault diagnosis based on support vector machine(SVM)classifier.SVM's own penalty parameter C and kernel parameter g have an important impact on the SVM classification performance,Therefore,an improved particle swarm optimization algorithm(IPSO)was proposed to optimize the SVM parameters.Based on this,an IPSO-SVM model was established,and a variety of analog circuits were used to simulate and verify the model.Then,in order to further improve the model's ability to solve complex diagnosis problems,the structure of the model is improved by combining with the parallel ensemble learning technology.The two-layer IPSO-SVM model is established,and the simulation verification of two kinds of analog circuits is conducted on the model.The main research contents of this paper are as follows:Firstly,the whole process and application method of fault diagnosis of analog circuit are studied.By analyzing the characteristics of analog circuit faults and the requirements of SVM for optimization of penalty parameter C and kernel parameter G,the wavelet packet decomposition is used to extract the features of diagnostic signals,and the advantages and disadvantages of particle swarm optimization(PSO)for SVM parameters are analyzed.Then,the particle swarm optimization algorithm(PSO)was upgraded to IPSO by dynamically setting inertia weights,adding individual and global extremum perturbations,and the particle mutation operation was easy to fall into precocity convergence,and the IPSO-SVM model was built based on the SVM with a single IPSO optimization parameter(IPSO-SVM).Three UCI public data sets with different data dimensions and three analog circuits with different circuit types and complexity were used to verify its classification ability.The results show that the IPSO-SVM model has better classification performance than the default parameter SVM model and PSO-SVM model,and has stronger global convergence ability.Finally,in order to further improve the model's ability to diagnose complex combined analog circuits,a IPSO-SVM based two-layer model with a sub-model processing layer and weighted voting layer is proposed by introducing the model framework of parallel ensemble learning technology.In the SVM sub-model processing layer,there are several independent SVM sub-models based on IPSO-SVM.The weighted voting layer will give corresponding weights to each SVM sub-model.When the model is diagnosing the analog circuit,it will vote according to the prediction result of every SVM sub-model and their corresponding weight to complete the fault diagnosis.Three UCI common data sets and two analog circuits were used to verify the classification effect.The results show that the IPSO-SVM two-layer model has a further improvement in the classification ability compared with the IPSO-SVM model,and has a more obvious advantage in the diagnosis of complex analog circuits.