Research On Fault Diagnosis Of Analog Circuit Based On Improved RBF Neural Network

Since the mid-1970s, with rapid development for system-level, board-level andchip-level of analog circuits, analog circuit fault diagnosis has been a hot research field.Due to the lack of diagnostic model, component tolerances and nonlinear problems,leading to the difficulty and complexity of the analog circuit fault diagnosis. Thetraditional fault diagnosis theory and methods based on model is difficult to achievethe desired results in practical engineering, artificial intelligence represented by neuralnetwork provides a new way for analog circuit fault diagnosis, attracting thewidespread attention in academic and engineering field.Aiming at the problem of the poor accuracy of analog circuits fault diagnosisbased on classical RBF neural network, this paper presents a novel approach tooptimize the parameters of RBF neural network using genetic algorithm and K-meansclustering. The work of this paper is mainly reflected in the following aspects:①For the feature extraction problem of analog circuit fault diagnosis, the analysisof the minor changes and high frequency distortion of fault signals in analog circuits isrealized by improving wavelet packet, and thus capture the fault features of the originalfault circuit more accurately.②For the low recognition rate problem in analog circuit diagnosed by classicRBF neural network, a diagnosis approach is presented by using hidden layer variablenodes of RBF neural network, thereby improving the recognition rate of analog circuitfault diagnosis based on RBF neural network significantly. With numericalexperiments, compared with fault diagnosis by BP neural network in performance, theimproved RBF neural network has obvious advantages in diagnosis speed andrecognition accuracy, to demonstrate the efficiency③For the performance of RBF neural network analog circuit fault diagnosisaffected by the parameters, a parameter optimization method based on K-meansclustering and genetic algorithm is presented. To design the optimization problem,considering speed and accuracy of the diagnostic, and the complexity of the neuralnetworks, width and data centers of RBF are get the best settings, improving theengineering significance of RBF neural network analog circuit fault diagnosis.