Fault Diagnosis Method Study Of Analog Circuit Based On Semi-Supervised Learning

With the coming of big data era,the speed of development of electronic information technology has put forward higher requirements in the aspects of the reliability and real-time performance of products and production processes.However,considering the cost of maintenance,the problem of fault diagnosis of analog circuits has been listed as a key research topic in the field of integrated circuit engineering.It has confirmed that the fault classification and diagnosis of analog circuits are of great importance in many fields.Based on the three learning forms of machine learning,this paper proposes three methods to optimize the fault diagnosis algorithm of analog circuits: supervised learning,semi-supervised learning and unsupervised learning,respectively.On the basis of Support Vector Machine(SVM)as the foundation and under the realistic in risk minimization principle,to respond effectively to the nonlinear,variable analog circuit fault types,data itself is flawed,the characteristics of the fault diagnosis is difficult to deal with such classic for different fault diagnosis method is put forward the improvement and optimization.The specific work of this paper is as follows:(1)Aiming at the problem of online health monitoring and real-time performance evaluation of analog circuits,the method of enhance the robustness of target function is proposed which can help to get more support vectors.The method of supervise algorithm not only could optimize the kernel function of least square support vector regression(LSSVR),but also could update the weight value of the objective function with the gradient descent method iterative.In the meantime,design increment and decrement of interactive learning mode is employed to assess and monitor the performance of the analog circuit online.It is easy to provide convenient time conditions for subsequent fault processing.(2)Focusing on the issue of transductive SVM(TSVM)with low efficiency of processing data under the condition of the analog circuit with small sample data,and the traditional supervised learning method could not balance tag data cost and accuracy,a semi-supervised laplace transuctive LSSVM(Lap-T-LSSVM),which introduces the Laplacian and constructing Laplacian matrix,to solve the problem of easy to fall into local minimum that traditional algorithm hard to deal with.The effectiveness of the method is verified by simulation experiments.(3)For the problem of(1),although it has the ability to realize accuracy classification,it is still easy to reduce application scope as for its dependency of marked category.In traditional SVM modeling,mining the characteristics of unmarked samples structure by introducing the thought of unsupervised clustering(UC)which could make a small number of mark samples play a more important role in guiding classification.In this thesis,a semi-supervised fuzzy c-means,(SS-FCM)algorithm to assist the SVM modeling process based semi-supervised clustering(SSC)theory combined with similarity coefficient is proposed.It can reduce the time cost effectively by optimizing the running time under the condition of hold accuracy,and certainly meet the requirement of robust learning.