Based On BP Neural Network And RBF Neural Network Of Engine Fault Diagnosis

With the development of society and economy, the car is becoming one of the necessary transportation of people’s lives. The engine as a power source of car, It’s working environment is complex, varied, and often work under high temperature and pressure environments. has a higher incidence of failure. If we can diagnose the trouble shooting in the case of the engine is not disintegration, not only will improve vehicle economy, power and security, but also reduce maintenance costs and improve efficiency. The mechanical fault diagnosis method based on vibration signal analysis with online, real-time, non-invasive, convenient, accurate, and in the mechanical fault diagnosis, that has been widely used.This paper used the engine valve clearance abnormalities and lack of fire as research subjects, for the engine mechanical failure characteristic extraction and diagnose problems, put forward the way based on wavelet packet analysis and BP neural network and RBF neural network. fault diagnosis. The topic build a engine vibration signal test system, Artificially simulated engine valve clearance failure and misfire fault condition, collect Engine vibration signal in normal conditions and fault conditions as the signal source. Due to the complex structure of the engine, and it is affected by the environment, test equipment and worker inaccurate operation in the signal acquisition process, the vibration signal of acquisition containing noise, impact the neural network training recognition accuracy, So before making neural network fault diagnosis, need to de-noise for collecting vibration signal therefore, we use the methods based on wavelet packet of threshold de-noising to process collected signal. First through hard threshold and soft threshold, semi-soft threshold for de-noising of the vibration signal was acquisitied. By comparison with the original signal, obtained semi-soft threshold noise reduction get the good results. Then using wavelet packet decomposition and reconstruction signal of noise cancellated. for normalization, extraction feature vector, the eigenvectors as part of BP and RBF neural network training input as part of BP and RBF neural network detection input.When using neural network fault diagnostic analysis, Introduction BP neural network and RBF neural network structures and algorithms, Established a vibration signal is applied to the engine failure recognition BP neural network and RBF neural network Model of intelligent recognition, Through analysis of experimental results, compared to the ability to identify BP neural network and RBF neural network, The results showed that, RBF neural network have the feature of a simple parameter adjustment, the training time is short, promotion ability bad, the same approaching capacity with BP neural network, BP neural network have the characteristics slow convergence, long training time, the ability of promote is strong.