Research On Diesel Engine Fault Diagnosis Based On Instantaneous Speed And Vibration Signal

As the core part of naval vessel, automobile, airplane, Diesel Engine is a widely used mechanical equipment. The equipment's performance is directly influenced by the quality of Diesel Engine. It have great realistic significance to find and diagnose the fault timely and taking effective measures.Firstly, The paper introduces the methods and status of research on diesel engine fault diagnosis, Secondly, two kinds of signal processing method are introduced emphatically:Hilbert-Huang Transformation(HHT)and Wavelet analysis. Hilbert-huang transform method use empirical mode decomposition (EMD) combined with Hilbert transformation method,which deeply depicts the relationship between time and frequency ( Hilbert spectrum),and the relationship between frequency and energy(Boundary spectrum),gets the time-frequency-energy figure(The time-frequency three-dimensional graph).The frequency band is divided into multi-level and the low-frequency space is decomposed further in Wavelet analysis, so the frequency resolution is becoming more and more high, and the time-frequency resolution is improved. Then the structure of diesel engine and characteristics of vibration and instantaneous spee signal are analyzed.We test the vibration and instantaneous speed signal in a V12 diesel engine test stand.,on the basis of theoretical analysis, Due to the complexity of the actual signal and the problems of the HHT method itself, a SVD based on the improved noise reduction Hilbert - Huang method is posed to deal with the vibration signal of the cylinder head. The superiority of Hilbert-Huang Transform in time-frequency analysis of vibration signals is proved from both simulation and experiment.The db wavelet decomposition is used to get time-domain graph of the instantaneous speed signal, and the difference of peak value in single-cylinder stroke is extract as a characteristic value, the fault of the oil break in single-cylinder can be seen clearly from the time-domain curve and characteristics value curve, the characteristics values are used as samples in training the BP neural network ,which obtains a good results.