Research And Implementation Of Fault Diagnosis In Analog Circuits

With the fast developing of IC, in order to improve product's capability, reduce chip's area and expense, we should integrate digital and analog component into one chip. From the statistic number we can learn that there are 95 percent fault resulted from analog circuit, though only occupied 5percent of chip's area. Fault diagnosis of analog circuit always a difficult problem on IC industry. Neural networks, which are typical representation of Computational Intelligence, provide a powerful way to diagnose faults of analog circuits.Drew much analog circuit fault diagnosis research, combined the latest achievements of electronic measurement with the artificial intelligence, directed toward tolerance and nonlinear of analog circuit, this thesis based on NN profoundly studies fault feature extraction and fault diagnosis methodology. The main work of the thesis concentrates on two aspects. Firstly, we studied two types of analog circuit fault feature extraction methods, rough set theory and principal component analysis. Rough set eliminate redundant components in samples, and extract the key information; the feature parameters are compressed using principal component analysis (PCA). It has many good properties, such as simplifying the structure of NN, improving the training speed and diagnostic efficiency. Secondly, we studied the method of determining the structure of BP neural networks, explored the method of determining BP neural networks parameters, analyzed the impact of changes of networks parameters on training results, and determined the practical problems of neural networks architectureIn this thesis, the experimental results show that the proposed method can perform correct diagnosis the analog circuits with tolerances, but those are not easily applied to the practical application due to the difference from theory to the practice, and it still should be studied in the future.