Analog Circuit Fault Diagnosis Technology Based On Artificial Immune Technology

Analog circuit fault diagnosis began in the1960s, so far, there have been many excellent artificial intelligence algorithms used in analog circuit fault diagnosis, such as: expert systems, fuzzy theories, neural networks, support vector machines. However, these algorithms have shortcomings of long training time and poor capability of updating knowledge dynamically, duing to the diversity of analog circuit fault phenomenon, the discrete nature of the component parameters and far-ranging nonlinearity. The artificial immune system is a highly non-linear ultra-large-scale continuous-time adaptive information processing system. And artificial immune system has diversity optimization capability, distributed storage capacity and self-learning self-organizing capacity. Its characteristics are ideal for analog circuit fault diagnosis. This thesis, by using artificial immune technology, focuses on the following:(1)improvement of clonal selection algorithm;(2) fault feature extraction technique based on wavelet analysis that is suitable for clonal selection algorithm;(3) validation of the proposed algorithm, the results show that the algorithm is effective and feasible.First, introduces three major directions of the artificial immune algorithm, such as negative-selection algorithm, immune network algorithm and clonal selection algorithm. After analyzing the principles and processes, pointes out the advantages and disadvantages of these algorithms. In these algorithms, the content of the clonal selection algorithm had been highlighted expounded. Its way of training out single-fault-centric in turn, can be effectively used to solve the problem of dynamic updating of knowledge, and the algorithm convergence speed, the training cost is small. Aiming at the shortcomings of the current clonal selection algorithm, a corresponding improvements has been proposed.Then, the optimal feature extraction method is studied for analog circuit fault diagnosis by taking four op-amp biquad high-pass filter circuit for the experiment. Two wavelets(Morlet wavelet and Haar wavelet) are adopted, and be compared in multi-aspect. The experimental results show that the more detailed fault features are extracted, the higher the diagnostic accuracy would be; meantime, the single point is as effective as multi-point sampling that explains the availability and validity of single-point output diagnosis, it has guiding significance for the selection of sampling points in fault diagnosis of large-scale analog circuit.Finally, based on improved clonal selection algorithm and wavelet technology for extracting characteristic, we have developed a comprehensive analog circuit fault diagnosis system. And we have verified the feasibility and effectiveness of the improved clonal selection algorithm, through the simulation circuit and the actual project circuit. The results show that the performance of the algorithm is superior to the neural network.