Research Of Feature Extraction Based On Independent Componentanalysis Indiesel Vibration Signal

Diesel engine as an important tool of modern industry, in promoting social andeconomic development plays an important role. In order to improve the safety andreliability of the diesel power plant, it’s necessary to do fault diagnosis of the diesel,but traditional signal analysis and processing technology is difficult to extract thefault feature from the mixed signals. Independent component analysis as a blindsource separation method with the ability to recover source signals from the observedsignal, is widely used in the field of speech processing, image processing,communications, signal processing, and medical signal processing. This paper basedon independent component analysis theoretical, explored diesel engine vibrationsignal feature extraction scheme with independent component analysis and faultdiagnosis with support vector machine. The main contents are as follows:(1) Researched the basic theory and algorithm of principal component analysisand independent component analysis, the corresponding data preprocessing methods.Researched theory of fast fixed-point algorithm based on negative entropy.(2) Researched diesel engine vibration signal separation based on independentcomponent analysis. Done the engine vibration signal separation research based onalgorithm of instantaneous hybrid model and convolution hybrid model ofindependent component analysis. Analyzed the time domain and frequency domaincharacteristic of the separated results comparatively.(3) On the basis of the main excitations of diesel engine vibration and vibrationcharacteristics on the surface, using wavelet packet decomposition analyzed andcompared energy spectral distribution of cylinder head vibration signals underdifferent conditions. Explored feature extraction method based on principalcomponent and independent component, and verified the effective of the methods inextracted feature of diesel vibration signal.(4) Researched the basic theory of pattern recognition and the method toestablish an identifier with support vector machine. Designed the diesel engine faultdiagnosis fault classifier, and optimized parameters through cross-classification test.Studied the effects of feature dimension to diesel engine fault recognition. Using design classifier to fault identification research for the features extracted by waveletpacket decomposition, principal component analysis and independent componentanalysis. The result shows that the independent component analysis has goodfeasibility and applicability of the diesel engine fault diagnosis.