Intelligent Fault Diagnosis Through Neural Network Optimization Using FPGA Chip

With the rapid development of the electronics industry, intelligent faultdiagnosis technology of analog circuit has become increasingly important, it hasimportant implications for the normal operation of electronics equipment and thereliability of the design. Based on the characteristics of analog circuit withtolerance, this paper uses BP neural network which is very suitable for intelligentfault diagnosis. It also uses the advantage of FPGA’s parallel processing, highreliability to construct a FPGA-based BP neural network, and designs anintelligent fault diagnosis system of analog circuit.Based on the study of the basic principles of artificial neural network, thispaper uses MATLAB software to program the learning rate automatic adjustmentmethod, BP neural network additional momentum method and LM optimizationmethod, and uses the same training sample set to train these three improvedalgorithms. The simulation graphical analysis results show that, the iteration ofLM-BP algorithm is least, its convergence speed is fastest, and its accuracy ishighest. This kind of algorithm solves the problem of slow convergence. But BPneural network algorithm is more effective in local search, and genetic algorithmhas the powerful search capability and good macro global optimization capability.Therefore, this article introduces genetic algorithm to optimize LM-BP algorithm.First, this paper uses the Protus software to simulate the " nominal value plustolerance value " of the test circuit, and gets the training sample set which areunder normal conditions and in the36kinds of hardware fault states, thendesigns a simulation of intelligent fault diagnosis circuit which is suitable for the4-11-6BP neural network model. Finally this paper uses the MATLAB softwareto design a genetic algorithm to optimize LM-BP algorithm network connectionweights and thresholds for intelligent fault diagnosis. The results show that, LM-BP algorithm of genetic algorithm optimization converges to training targets after 35trainings, the speed of its convergences is significantly faster than LM-BPalgorithm, and accelerates the convergence speed of the network further.In order to verify the feasibility and effectiveness of the established neuralnetwork, this paper uses the commonly used FPGA chip EP2C8Q208C8inintelligent fault diagnosis, and designs intelligent fault diagnosis systemhardware interface circuit. This paper also designs the A/D control module, FIFOdata input and output module, data preprocessing module, neural network moduleand LCD module of internal module function. Finally, this paper uses anintelligent diagnosis example to test this intelligent fault diagnosis system. Theresults show that this intelligent fault diagnosis system is simple, its diagnosisspeed is fast, the diagnosis accuracy rate is high, and it has a certain practicalvalue.