The Research On Genetic Algorithm And ANN Based Theory And Methods For Fault Diagnosis Of Analog Circuits

For several decades, fault diagnosis of analog circuits, the forefront of modern circuit theory, has a series of diagnosis theory and methods. However, complicated theory and poor practicability of these methods make these methods' application foreground far from expectation of people. Artifical Neural Networks(ANN) have gained quick development and have been widely applied in many areas recently. With the development of micro-electronics technology especially mixed-signal circuits and VLSI technology, test and fault diagnosis technology of large-scale analog circuits with tolerances are increasingly important and imminent.This paper mainly researches ANN based methods for fault diagnosis of analog circuits with tolerances and presents some improved ANN based methods which can decrease diagnosis time and enhance diagnosis efficiency for analog circuits with tolerances.This paper presents some new fault diagnosis methods on the basis of improved error back propogation Neural Network (BP). The algorithm considers not only the impact of change trend of error, but also ensures that the network studies at the maximal study speed. Artificial Neural Network has associational and memorial abilities, strong study ability, strong robust and non-linear mapping ability, which make this diagnosis method better than traditional fault dictionary methods in the way of diagnosis effect and diagnosis time.This paper combines genetic algorithm, fuzzy logic with ANN and presents a fault identification method for non-linear analog systems. In the method, a Fuzzy Neural Network is developed based on the improved fuzzy weighted reasoning method. The training of network weights and optimization of membership functions are conducted employing genetic algorithms. Fuzzy rules can be automatically obtained according to .the weights of the network. The availability of the method are examined by simulatedtests.This paper deeply researches network decomposition algorithms of large-scale analog circuits and presents some improved multistage decomposition algorithms for diagnosing subnetworks in large-scale analog circuits. This algorithm computes the voltages of inaccessible nodes of decomposition in linear subnetworks or non-linear subnetworks. Simulated tests for large-scale analog circuits show these methods have good diagnosis effect and feasibility.