Analog Circuit Fault Diagnosis Based On LMD Approximate Entropy And SVM

With the increasing integration of analog circuits,it put forward higher requirements to the diagnosis of circuit failure.The development of intelligent information makes it possible to develop analog circuit fault diagnosis technology with better performance.Combining with intelligent information,this thesis presents a method of fault diagnosis of analog circuits based on local mean decomposition approximate entropy and support vector machine.(1)Aiming at the characteristics such as the tolerance of the analog circuit components,the variety of faults and the non-stationary nonlinearity of the fault signals,a combination method based on local mean decomposition and approximate entropy is proposed to extract the characteristics of analog circuit faults.The local mean decomposition can decompose the non-stationary and nonlinear fault signals adaptively,and the frequency of the component signals is ranked from high to low.The approximate entropy can characterize the complexity of the time series.The eigenvectors obtained by calculating the approximate entropy of the component signals have the characteristics of large differences between different classes,small differences between the same classes,and well-defined.(2)Aiming at the problem of lack of generalization ability of SVM,a combined kernel LSSVM and chaotic particle swarm optimization algorithm is proposed.Firstly,using least squares support vector machine instead of support vector machine to avoid solving complex quadratic programming problem.Then,a combination of Gaussian radial basis function and polynomial kernel function is used instead of a single kernel function to balance the generalization ability and classification accuracy of the classifier.Finally,in order to find the optimal classification surface,chaos particle swarm optimization algorithm is used to optimize the parameters.It adds chaos theory to the particle swarm optimization algorithm.By using ergodicity,randomness and sensitivity to initial values of chaotic optimization,it avoids falling into the local optimum,increase the population diversity and particle search ergodicity,improve classification accuracy and accelerate global convergence.The final case analysis shows that the method derived has a certain effect in improving fault recognition rate and reducing test time.(3)When there is too many classification categories,the traditional classification structure is faced with the problems of long-time training and classification,moreover low-efficiency diagnosis,so it propose the binary tree support vector machine classification structure.Compared with the traditional one-to-one and one-to-many structure,the binary tree structure reduces the number of classifiers and repeated training samples,and improves the diagnostic efficiency.Due to randomly placing faulty classes in binary tree nodes may lead to the problem of low diagnostic accuracy,it calculate and sort the similar directions among various classes.The fault category is placed on the binary tree nodes according to the similarity between classes,with the more difficult to separate,the farther away from the root node.And it use genetic algorithm to find the best classification surface of each classifier.The final simulation shows that the method in this thesis has greatly improved the diagnostic efficiency under the premise of ensuring the diagnostic accuracy.