Research On Fault Diagnosis Of Internal Combustion Engine Based On Vibration Analysis

Because of the complexity and multiple-excitation of Internal Combustion Engine(I.C.E.), there isn't a series of effective and applicable fault diagnosis methods for it at present. So it is a direction for researchers to build up a fault diagnosis system for I.C.E.This thesis mainly involved vibration diagnosis on I.C.E.. Based on the structure dynamic analysis and detailed vibration properties analysis of I.C.E., and in allusion to the time domain and frequency domain character and nonstationarity of the surface vibration signals on I.C.E., several effective feature extracting methods and feature variables of time domain, frequency domain and time-frequency domain are proposed by analyzing and contrasting fault signals with normal signals which were sampled in practice. In this section, three kinds of time-frequency analysis methods such as short time fourier transform(STFT), Wigner-Ville Distribution(WVD) and wavelet transform(WT) are emphasized to analyze and contrast. These time-frequency methods were used to diagnose the I.C.E. misfire fault, and the result shows that these time-frequency methods have more stronger capability of extracting feature from the surface vibration signals of I.C.E. than traditional Fourier analysis. The back propagation algorithm which apply to the multiplayer feedforward neural networks have been analyzed, and its conventional improving algorithms are introduced. The strategy of combining appended momentum term with adaptive learn speed adjusting is adopted to overcome the defect of the slow convergence and local extremum plunge of the standard back propagation algorithm. A fault diagnosis method for I.C.E. based on back propagation neural networks is built up, and the effectiveness of this method is validated by the practical case. On the basis of comprehensive use of those methods and theories which are presented in this thesis, a fault diagnosis system for I.C.E. is built up by using the programming method of combining Visual C++ with MATCOM. The system has been used in practice.Research results of this thesis show that it is feasible to build up an effective fault diagnosis system for I.C.E. based on back propagation neural networks through using the feature variables of time domain, frequency domain, time-frequency domain and the feature variable of wavelet packets. Meanwhile, to thoroughly solve the problem of fault diagnosis of I.C.E., it must be stated that there remains much hard work to do, especially the research of fault mechanism, the research of analyzing methods and feature extraction of nonstationary signals. A lot of work and advice is also proposed at the end of this thesis.