Feature Extraction And Detection System Of Non-Stationary Signals Produced By The Solid Vibration

Diesel engine is the most important power mechanical equipment of vehicles,it is greatly significant of us to monitor its condition and to diagnosis its fault. The key of the diesel engine machinery fault diagnosis is how to extract the fault feature from the non-stationary vibration signals, it is often difficult to achieve good results to various non-stationary signals when the diesel engine running from the conventional signal processing method. For these issues, taking the WD615 model diesel engine non-stationary signals as studying object, combined with the actual research, developing non-stationary signals produced by the solid vibration analysis and fault feature extraction research.Based on the analysis of solid vibration model from the microscopic point of view, with carrier of diesel engine vibration signal, studied the non-stationary vibration signal acquisition device of the wearing fault of connecting rod bearing, valve, piston, piston pin and complicating fault of piston and piston pin systematically; studied the effective method of EMD-Gabor transformation, angle area fourth-order cumulant slice spectrum, high-order cumulant image feature and polar coordinate enhancement of time-frequency diagram for signal analysis and fault feature extraction deeply; the method which combined variable precision rough set with SVM is adopted to realize fault pattern recognition of small samples. The main contents of this dissertation are organized as follows:1. Vibration of solid is produced by the coulomb interaction of atoms(molecules) outer electrons which performanced as diesel unstable vibration signal contains a wealth of information about the technical state of the internal mechanical component. The unstable vibration signal acquisition device is studyed, the unstable vibration signal acquisition project is determined according to the correlation of faulty components and measuring location and rotating speed, etc. the acquired signals have good repeatability.2. For non-stationary and non-linear characteristics of the engine vibration signal, a method based on the empirical mode decomposition(EMD) and Gabor transformation is proposed to extract fault features with different degrees about the single fault, this method could highlight the target component, inhibit the interference from the other component and noises and overcome the problem of traditional time-frequency constant resolution and false component interference,etc. The frequency band energy accumulate curve of IMFs after the EMD is used to extract frequency band energy from time-frequency distribution, the extracted feature parameter could reflect the fault characteristic of the non-stationary vibration signals analysis object effectively.3. A method based on angle area fourth-order cumulant slice spectrum is proposed to extract fault feature of non-stationary vibration signals, this method is applied to the diesel engine fault diagnosis for the first time, the non-stationary accelerating vibration signal is re-sampled and the stationary signal is obtained at the angle area by the order method, the power and peak value of angle area fourth-order cumulant slice spectrum with certain order band is used as the corresponding fault feature. This method could both analyze unstable signals and inhibit noise interference, and also simplify the computational complexity significantly, non-stationary signals of different components and different fault degree is analysed to obtaine the optimum diagnosis rotational speed and diagnosis position.4. A diesel engine fault feature extraction method based on high-order cumulant image feature is proposed which introduced the image processing technology into the field of diesel engine fault diagnosis. The de-noising ability of higher-order cumulants and all the information about its image is used to extract the image texture feature parameter based on three-order cumulant image gray level co-occurrence matrix(GLCM) of different fault components, and use different feature parameters values to describe different components fault types of diesel engine. Different mechanical fault feature of the diesel engine is easy to confused and is often drowned in other components and strong noises, so it is difficult to distinguish and extract, this method solve the above problem effectively.5. A method based on polar coordinate enhancement of time-frequency diagram is proposed for diesel engine fault feature extraction. According to the non-stationary cycling characteristic, the diesel engine vibration signal feature is mapped from the rectangular coordinate to the polar coordinate and enhancement synchronously, areas energy of polar angular frequency distribution is extracted as the fault feature. The fault feature of different location is extracted from the perspective of feature enhancement, this method is intuitive and clear which distinguish fault types effectively and diagnose the fault location accurately.6. For the feature parameters extracted from multiple fault feature extraction method, the variable precision rough set theory is applied to reduct feature parameters and eliminate redundant feature parameters, then enter into the SVM which suitable for small samples for training classification, this method obtained good diagnostic results. Finally, based on the proposed feature extraction method, developed and realized a diesel engine mechanical fault diagnosis system which embedded into a portable vehicle fault detection instrument without disassembly as a sub-module and used for the actual fault diagnosis.