Mechanical Faults Feature Extraction And Diagnostic Techniques Research Based On Wavelet Analysis

In the process of the mechanical fault diagnosis, signal measuring and processing is the first important step. How to use the proper signal processing method to extract the feature information, which reflects the types of fault, is the key problem in the area of fault diagnosis, and it directly affects the efficiency and reliability of fault diagnosis. According to the target of establishing an intelligent mechanical fault diagnosis system, this thesis adopted wavelet analysis as the basic method of signal processing and feature extraction, did research on the validity of wavelet analysis in the signal de-noising and feature extraction, completed the corresponding computer programming, achieved the overall coherence to the rough sets and neural network intelligent fault diagnosis system, and finally validated the system with fault signals from rotor system.The thesis mainly works and achievements include:(1) The thesis studied the basic theory of mechanical fault diagnosis, investigated the recent research achievements in this area, got to understand the some dignosis methods, relized the importance of feature extraction of signals, found that the commonly used signal processing method in enginerring practice was FFT and there is lack of more efficient software for feature extraction. In view of this situation, the thesis aimed to find a more effective signal processing method for feature extraction, and programmed this method in software.(2) Based on the analysis of some commonly used wavelet functions, the selection of wavelet parameters, the principle of signal de-noising and feature extraction, and some examples of simulation validation, wavelet analysis was determined to be used as the basic method of signal processing and feature extraction, which satisfied the overall architecture requirements of rough set and neural network intelligent fault diagnosis system. The programming ideas of automatic de-nosing and feature extraction with wavlet analysis were put forward.(3) With Visual Basic good interface and Matlab calculation capacity, the thesis mixed Visual Basic and Matlab to program the signal processing and feature extraction software based on wavelet analysis, rich much relevant function, combined with the rough set and neural network intelligent fault diagnosis system, formed a complete diagnosis system.(4) With the laboratory rotor test system, normal, unparallel, unbalanced and rubbing fault state were simulated, wavelet analysis was used to de-noise and extract features from these fault signals, finally the signal spectrum, characteristics of wavelet coefficients and energy eigenvector were summarized and analyzed. Finally, the energy eigenvectors that extracted by feature extraction procedure we programmed were used to validated the rough and neural network intelligent fault diagnosis system.