| With the rapid development of electronic information technology, the integration of circuits increases, the performance and structure also becomes more complex, which make an urgent request of the automate circuit fault diagnosis. After 30 years development, the circuit fault diagnosis technology has become the focus in the field of circuit research. However, most of the existing circuit fault diagnosis methods can only diagnose hard fault, such as open or short-circuit of components. There are few methods for the soft fault, in which the component parameters are beyond the tolerance, and such research is limited to the situation, such as the fault components parameters are limited deviation value. The continuous deviation of fault components parameters is not considered in this case. So this situation of the analog circuits fault diagnosis would be studied in this paper.For the analog circuits soft fault diagnosis, this paper presents a Wavelet Analysis and Error Correction Radial Basis Function (ECRBF) network combining diagnostic program. Compared to the signal analysis approach in time domain and frequency domain, the Wavelet Analysis has better characteristics in time-frequency domain, which can effectively extract the signal character in time and frequency domain. Therefore, in this paper Wavelet Analysis is used to decompose the signal to extract the energy of each wavelet as the character, and the normalized energy character would be the input of the neural network. The RBF network as a classifier has great advantages in the non-linear mapping, classification ability and convergence speed. This paper combines the traditional RBF network and error correction coding techniques of communication, making the improved network the ability of wrong correction. The structure of multi-input and multi-output is usually used for classification in the traditional RBF network. The ECRBF network, as a classification system, is composed of multiple single-output RBF networks, and the number of sub-networks is based on error correcting codes of classification number. In the RBF network learning algorithms, we use the K-means algorithm for clustering samples to determine the cluster centers. Finally, the actual circuit will be used to extract fault feature and fault diagnosis. The simulation and results show that the program can achieve the fault and position of fault component, and achieve higher fault diagnosis rate. |
