Application Research On The Method Of Intelligent Fault Diagnosis Of Analog Circuit

According to statistics, most of the faults in electronic equipments were caused by the faults in analog circuits. Therefore, the importance of the fault diagnosis technology of analog circuits became more and more obvious. It is of vital significance to the normal operation and reliability design of electronic equipments and systems. After 1970s, the study on the fault diagnosis technology of analog circuits has yielded many exciting outcomes world-widely and gradually evolved into systematic theories. It has become the third major branch of network theories, following the theory of network analysis and network synthesis.Compared with the conventional diagnostic methods,modern computational intelligence technology , represented by neural network, provided a new way for the fault diagnosis of analog circuits . It is unnecessary to establish precise mathematical model for diagnostic objects and has the properties of self-organizing and self-study. These features made it an effective method for fault diagnosis, particularly for fault diagnosis of analog circuits. In this paper, the general guideline and diagnostic procedures of neural network-based fault diagnosis of analog circuits were studied, as well as the structural design of BP network, training algorithm and its advantages in fault diagnosis.The conventional BP algorithm has its inherent flaws,such as slow convergence,easily caught in the local minimum value. This paper proposed a combined algorithm of genetic algorithm and LMBP algorithm to overcome the inherent flaws of traditional BP algorithm. In this algorithm, the genetic algorithm was used to search the optimal weights in the overall weights space to overcome the flaw of being easily caught in the local minimum value, and then LMBP algorithm is used to perform local fine-tuning. Because LMBP algorithm can achieve fast convergence performance,it is quite helpful for the promotion of diagnostic speed in analog circuits.Finally, the hybrid algorithm (GA_LMBP)mentioned above was tested on linear power supply circuit and 2-T band elimination filter by simulation. The experimental result showed that this solution could achieve relatively accurate diagnosis for the faults of analog circuits and reach the anticipated objective. It has good performance in diagnostic speed and accuracy, as well as universality.The result obtained in this paper needs to be further improved because of the gap between experimental result and real engineering application.