Analog Circuit Fault Diagnosis And Its Implementation Based On The Variance And Kurtosis

Fault diagnosis is still an extremely challenging research subject in the circuit test field around the word. But the traditional analog circuit fault diagnosis methods can’t meet the diagnostic needs of the large scale integrated circuits, especially for the diagnosis of non-linear and tolerance analog circuits. At present, research on fault diagnosis of analog circuits, the first is how to extract the analog circuit fault informations, the second is how to establish an efficient diagnostic model.Taking into account that the the impulse response signals of fault circuit included time-varying and non-stationary informations, it need to preprocess with wavelet theory. In this paper, it researchs using wavelet packets to preprocess, but the reconstructed sub-signals only included the analog informations of each order sub-signals, it can’t be used as neural network’s input. Regarding that, the variance and kurtosis method is proposed to extract the fault informations of each sub-signals to solve the problem of feature extraction in a nonlinear and tolerance analog fault circuit. In Solving the small sample, nonlinear and high dimensional pattern recognition problems, compared with the neural networks, support vector machines has the advantages of global optimum, good generalization, etc. So, support vector machines is applicated for analog circuit fault diagnosis in this paper. For the randomness of selection of the DAG-SVM’s node, which easily leads to the "error accumulation", a PDAG-SVM method based on vector projection is proposed.In this paper, an analog circuit fault diagnosis system is researched. selecting Samsung’s S3C2440development board of ARM920T kernel, successing to implement fault diagnosis with support vector machines, realizing the core functions of the system. Finally, transplanting Linux system and QT to the ARM development board, and making a good man-machine interface.