Research On Method Of Fault Diagnosis Of Analog Circuits Based On Support Vector Machine

The research on method of analog circuit fault diagnosis, not only is of great significance in circuit theory, but also it has a very important value in electronic devices, integrated circuits and other practical applications and so on. There are many traditional analog circuit fault diagnosis's methods, and the diagnosis of these methods depends largely on the success of the optimization of analog circuit fault diagnosis model, and just once established the model is difficult to change, and it gives rise to the lack of appropriate flexibility, and till the end there may be unable to get equation of analog circuit fault diagnosis which can not effectively diagnose them. So more and more people began to gradually try to apply the artificial intelligence techniques in analog circuit fault diagnosis, but the traditional artificial neural network has input sample with higher demands in self-learning, local optima, under-learning, over-learning and and model structure which is difficult to determine and so on, while the analog circuit fault diagnosis method based on Support Vector Machine (SVM) can solve these problems.In this paper, it investigates the general methods and procedures for fault diagnosis of analog circuits based on support vector machine, and analyzes the difficult and critical technologies in some of the fault feature extraction and the building of support vector machine;It describes the basic theory of analog circuit fault diagnosis and support vector machines; It studies analog circuit fault feature extraction methods; finally, a typical low-pass filter circuit is selected to test these methods and process by using Pspice and Matlab software. The simulation results show that support vector machines can more accurately classify and identify the single soft faults in analog circuits, while the properties of analog circuit fault diagnosis based on Support Vector Machine (SVM) are better than these based on BP neural network in diagnosis accuracy and time of training-testing; The results also show that, the two methods of fault feature extraction based on effective sampling point and wavelet packet are both practical, the properties of analog circuit fault diagnosis based on wavelet packet are better than these based on effective sampling point in diagnosis accuracy and time of training-testing. In short, the method of analog circuit fault diagnosis based on support vector machine, which can solve the problem of the single soft faults in analog circuits, is practical.