Application Of OFBP Neural Network In The Engine Fault Diagnosis

The rapid development of automotive technology also associated with theoccur of risk, with the increasingly high demand for the car features, theperformance of the car is improved and structure becomes complex, also theprobability of occurrence of the failure increases. So the fault diagnosistechnology also faces challenges, people’s lives and property safety need moreadvanced and reliable technology to protect. In order to minimize their losses, it’snecessary to find an effective method to detect car failure in time, and evenpredict the occurrence of the fault.Running condition of car engine is very complex, it’s not easy to diagnosetheir failures accurately. There are many parts of engine can be faulted, failurephenomenon varies, and on the whole，the failure has nonlinear characteristics.For such problems, there is no sufficiently accurate model which can representthe fault system. An intelligent diagnosis system is needed, it can extract faultfeatures from a large number of samples, from their own knowledge system bylearning from the samples of the fault, to realize fault diagnosis and prediction.The development of intelligent fault diagnosis technology provides a newtheoretical basis for solving such problems, the adaptive learning ability of neuralnetwork makes fault diagnosis system has a higher degree of intelligence andability to judge the fault.The BP neural network is one of the feedforward neural networks, its back-propagation algorithm makes it become one of the most representative of allkinds of neural networks. The good nonlinear mapping ability of BP neuralnetwork can be a good application in fault diagnosis. But the traditional BPnetwork has the trend of forgetting old samples during the training process whenlearning new samples, the result of the training before did not play a role in lateneural network training, and exists the defect of low training accuracy. A neural network algorithm of increased state feedback in the output layer is designed inthis paper to solve the problem above. The improved BP algorithm, named OFBPneural network is used in the fault diagnosis of automotive engine, the indexes ofthe automobile exhaust are used as the inputs of the neural network, the outputscorresponding to the different misfire. The simulation results show the proposedalgorithm can effectively improve the BP neural network training accuracy, andachieve misfire diagnosis more accurately.