Unstable Vibration Signal Analysis And Intelligent Fault Diagnosis Of Diesel Engine

Diesel engine is the pivotal power assembly of vehicles. It is greatly significant of us to monitor diesel engine working performance and to provide fault forecasting.A lot of faults in the engine steady-state operation are not easily to be found in the severe noise, which would lead to feature extraction failure. To solve this problem, the dissertation proposes the idea of using vibration signal in unstable state for fault diagnosis of diesel engine. And the study is based on sample of Cummins diesel engines EQ6BT, combined with typical engineering research, developing diesel engine unstable vibration signal analysis and intelligent diagnosis research.In the base of vibration mechanism research of diesel engine, a unstable vibration signal acquisition device is designed in good reproducibility and stability; wavelet fractal method, blind source separation-bispectrum method and EMD-AR spectral method are proposed by views of non-steady, nonlinear, and non-Gaussian, which are effective for unstable vibration signal analysis; variable precision rough set theory is applied to realize the automatic extraction and optimizational options of the fault feature; applying dual-level fusional frame for multi-sources information and multi-signal processing methods, the intelligent mechanical fault diagnositic system of diesel engine is developed and embedded to field vehicle intelligent diagnositic instrument; realizes the pattern recognition of single fault and double fault mode of location and severity effectively. Summing up the research works of dissertation, the following conclusions are introduced:①Unstable vibration signals contains a lot of characteristic information which stable signal does not have. By using reasonable technical method to analysis unstable vibration signals, it could effectively diagnose fault which is not easily diagnosed by stable vibration signals.②The single fault mode feature can be effectively extracted by wavelet fractal method, blind source separation-bispectrum method and EMD-AR spectral method. And the blind source separation-bispectrum method and EMD-AR spectral method is much better than the wavelet fractal method to extract the double fault mode feature.③The principal component of unstable vibration signal is separated by nonlinear PCA algorithm. And the features extracted by BBS-Bispectrum method is clearer than ones by sole bispetrum method.④EMD-AR spectral method is fully facilitating the advantage of HHT method to process non-stationary, nonlinear signals, and overcome the errors at both ends of the demodulated signal spectrum caused by the windowing effect in Hilbert separation algorithm. The smooth clear spectrum is obtained, which can effectively recognise the fault pattern in wide frequcency band and show good stability.⑤As improper classifying in some extent is allowed in analysis of variable precision rough set, anti-jamming capability has been improved; more robust performance is showed than the classical rough set; automatic feature extraction is realized and more stable fault feature is obtained.⑥The intelligent mechanical fault diagnositic system of diesel engine that carred out by combining SVM and D-S theory, integrated with the advantages of multi-source information and multi-signal processing methods; solved the single SVM diagnostic issue of low precision and weak generalization ability and diagnostic accuracy and stability could be significantly improved.