Research On Engine Ignition Fault Diagnosis Based On Neural Network

Along with the increasing of technology and people’s living level unceasingenhancement, the automobile more and more common into our life, at the same time, italso gives us a problem that the automobile fault rate become more and more high.However, the number of professional talents in automobile fault diagnosis aspect issmall, so, we need to find a way can quickly and accurately diagnosed automobile fault.Engine is the heart of the car, and most of the faults caused by the engine, at thesame time, almost45%-50%of the engine failure is caused by engine ignition systemfailure. The author of the article consulted a large number of literatures, take Accord thecar F23A1model as the research object, with secondary ignition voltage waveform asthe research foundation, identified for the models of the four kinds of commonmalfunction and cause malfunctions of these specific parameters.As we know, BP neural network’s hidden node number is uncertain, at the sametime, the number of BP neural network’s hidden node determines the networkconvergence speed and error. In this paper, we got four kinds of structure model, anduse the four kinds of structure model to simulation analysis, get the best network modelis4-10-2structure, then, the model structure used in fault diagnosis, got the trainingerror, the structure’s training steps is131, the mean square error is0.000834918. Thenuse GA to optimization the BP neural network weights and threshold, got the trainingerror, the structure’s training steps is71, the mean square error is0.000173742. At last,use PSO to optimization the BP neural network weights and threshold, got the trainingerror, the structure’s training steps is35, the mean square error is0.000155071.Through the contrast we can see, compared with BP neural network and the GAoptimized BP neural network, the PSO optimized BP neural network has shortertraining time and less error, applied it to the ignition system fault diagnosis has goodaccuracy.