Research On Analog Circuit Fault Diagnosis Methods Based On Neural Network

With the development of modern electronic technology, electronic circuits have more larger scale, more integrated degree and more complex topologies. Although the increase of accuracy and decrease of failure rate of electronic circuit, the failure is inevitable and has become more difficult to detect and diagnose. Electronic devices circuit is divided into digital and analog circuits, analog circuits are more prone to failure than digital circuits, but also because of the difficulty of diagnosis of soft failure and other reasons more difficult. It has become an urgent need of practical engineer application how to use smarter means to take analog circuit fault diagnosis faster, accurately and effectively.This paper studies on the problems of the real-time of circuits, soft fault diagnosis and improving diagnosis accuracy, and so on. The main research work and achievements are as follows:(1) Embedded system based on BP neural network and its application in fault diagnosis of analog circuits. For real-time analog circuit fault detection problem, solutions of use of the embedded system to achieve BP neural network is proposed. Four hypothesis of fault detection of analog circuit is proposed; The structure and diagnostic procedure of diagnosis system is analyzed; An experiment using the system for analog circuit fault di agnosis is done. Experimental results show that BP neural network based on embedded systems can diagnose the fault of analog circuit effectively.(2) Research on Fault Diagnosis of Analog Circuits using neural network based on deep learning theory. For analog circuit fault diagnosis problem of soft breakdown, method of using deep neural network is proposed and an experiment is done using a stacked neural network which proved that the solution is effective.(3) Research on Fault Diagnosis of Analog Circuits using loose wavelet integration neural network. when a man classifies an object, he makes a comprehensive judgment after plurality of feature extraction and classification. Based on the phenomenon and in accordance with the theory of one-dimensional neural network, the concept and theory of loose wavelet integration neural network is put forward, which is applied in the field of analog circuit fault diagnosis.Firstly wavelet is used for feature extraction and principal component analysis is applied for dimension reduction. Then loose wavelet integration neural network is added as a classifier to classify. Finally, decider is used for final judging categories. Experimental results show that the proposed network architecture used for fault diagnosis of analog circuits can effectively improve the accuracy of diagnosis.In summary, this paper studies on the problem of real-time diagnose, soft breakdown of circuit, and improving accuracy based on the classical theory of BP neural network,deep learning theory, loose wavelet integration neural network combined with other technique like wavelet theory. Experiments show that the effectiveness of neural network to diagnose analog circuits.