Research And Realization Of Circuit Fault Diagnosis Based On Neural Network

At present, neural network, which is a typical representation of computational intelligences, provides a powerful way to diagnose faults of circuits. In order to improve the shortcomings of the test program, which is used in the Automatic Test System, this paper puts forward a novel algorithm of BP neural network based on Wavelet Analysis and Genetic Algorithm (GA). In this novel algorithm, Wavelet Analysis is used to abstract the fault features of circuits, and GA is used to optimize the leaning of BP neural network's connection and thresholds. The main works and creations of this paper are described as follows:This paper studies the principle of the BP neural network's application in circuit fault diagnosis, explains the principle and frame of the BP network, and puts forward a circuit fault diagnosis model based on the BP neural network.In the circuit fault diagnosis, the methods of extracting the finite sampling signal can not reflect the characteristics of the various failure modes, so that they can not separate all the failure modes. Therefore, in this paper, a method of the"energy - failure"based on wavelet packet decomposition is proposed, which uses wavelet transform to extract the energy characteristics of fault signals by decomposing the circuit output signal. Then, the normalization energy characteristics are used to train the BP neural network. The method improves the BP neural network's ability of identifying fault types.After trained, BP neural network can determine the type of fault by the current measurement data. But the optimization of BP neural network is not unique; it is susceptible to fall into the local minimum, which maybe results in miscarriage of justice. GA can search the objective function in the parallel space, and check a number of possible solutions at one time. It can search the global optimal or suboptimal solution in the solution space; it has an excellent ability of global searching. However, compared with the BP neural network, GA has the shortcoming in the local searching. In order to overcome their shortcomings, this essay will effectively combine the two algorithms to form a hybrid algorithm, so that it can achieve the purpose of optimizing the network. After using GA to achieve the optimization of weights and threshold in BP neural network to narrow the search bounds, this algorithm start the course of training.Finally, the novel algorithm is applied to the fault diagnosis of WJ8615P. This essay compares the novel algorithm with the former method. The results show that the novel algorithm has the advantage of less test steps and higher ratios of the fault coverage and fault-diagnosis, and it provides a new way for circuit fault diagnosis.