Fault Diagnosis Of Analog Circuits Based On Machine Learning

Continuity, nonlinearity and tolerance are still difficult to solve in analogous circuit diagnosis until now, which restricts the development of analogous circuit fault diagnosis technology. The existing circuit fault diagnosis technology can hardly go with the development speed of the circuit, which leads to the high cost of the analogous circuit test and maintenance. In circuit systems, the failure rate of analogous circuits is relatively high and more than 80% faults occurred in analogous circuits. In most chips, analogous circuit accounts for about 10% of the chip circuits, but the latter maintenance cost accounts for more than 90% of the running cost of the whole chip. Therefore, the research of analogous circuit fault diagnosis technology is of great value. this paper focuses on the following tasks:(1) Firstly, this paper analyzes the development of analogous circuit fault diagnosis,compares the circuit fault diagnosis algorithms such as pre simulation and post test simulation and studies BP neural network and support vector machine (SVM) in machine learning algorithm. Starting with the selected test circuit, the selection of fault sets, the Monte Carlo simulation and the application of machine learning algorithms are discussed in detail. In order to closely compare the effect of the two algorithms in analogous circuit fault diagnosis,several standards for model evaluation are introduced. Then the diagnosis model of fault data is established by SVM and BP neural network algorithm and the simulation experiments are carried out on Active Filte circuits.(2) Combining with the actual simulation, a set of complete and effective diagnostic criteria for fault diagnosis of analogous circuits based on machine learning is proposed, which includes feedback rate, precision rate, accuracy, time consumption and other evaluation indicators. BP neural network algorithm and SVM algorithm are studied respectively to evaluate the performance of analogous circuit fault diagnosis with the evaluation results saying that the evaluation system is comprehensive and effective, providing a new way for the application of machine learning in other areas.(3) Through the analysis and comparison of the diagnosis results, we can conclude that the effect of BP neural network on nonlinear problems shows very good. For diagnosing problems with complicated internal mechanism, just like analog circuit internal faults, BP neural network algorithm is very effective. However, it is slow in calculating the convergence rate, which is one of the shortcomings. The accuracy of support vector machine algorithm has a high degree of dependence on a few samples of support vectors, and these samples affect the decision function of support vector machine. The more support vectors, the more complex the computational complexity is. In the course of training, a lot of useless samples are removed,which makes the algorithm of SVM is simple.