Study On Application Of Morphology In Machinery Fault Diagnosis

The key issue of machinery fault diagnosis is how to extract fault feature from fault vibration signals, signal analysis methods are most widely used in fault feature extraction. The vibration signals are usually possess non-stationary and nonlinear signals when mechanical fault occurs. Therefore, how to extract fault feature from non-stationary and nonlinear signals is one of the important research issue in the field of mechanical fault diagnosis. However, the traditional time-frequency analysis method such as Short Fourier Transform(SFT), wavelet transform and Hilbert-Huang Transform(HHT) have their own limitations. Therefore, it is necessary to apply novelty signal processing theory and method to improve efficiency and level of fault diagnosis. Mathematical morphology, as a nonlinear analysis method, is recently developed and has began to be used in machinery fault diagnosis. It has also been proven to be effective in some areas of fault diagnosis. Setting rotor system, gear and rolling element bearing as research objects, this thesis studied deeply on fault diagnosis methods based on mathematical morphology. The main contents are as follows:(1) Fault principle of common mechanical parts, such as rotor system, gear and rolling element bearing were introduced, fault signal feature analysis were also carried out to supply basis for latter chapters.(2) In order to overcome the problem of random selection in traditional morphological filter, an adaptive multi-scale compound morphological filter (AMCMF) method was proposed. Types and size of structure element could be determined adaptively according to signal local characteristic and noise. Correlation dimension was applied to classify rotor system fault types after the signal was filtered by AMCMF. Test results indicate effectiveness of the method.(3) Aiming at the problem of fuzzy edge in traditional morphological detection, a multi-structure multi-scale morphological edge detection method was proposed. Four direction structure element were used to extract edge in time-frequency chart. Its effectiveness could be demonstrated by comparing with traditional morphological edge detection. Feature could be extracted by calculating gray co-occurrence matrix. Gear fault could be classified through LSSVM. (4) In order to overcome limitation of constant lifting operator in traditional lifting morphological wavelet, an adaptive lifting morphological wavelet(ALMW) denoising method was proposed. Lifting operator could be determined by signal local characteristic. Fault energy feature vector was defined after fault feature was extracted by ALMW. Gray relation was used to diagnosis gear fault. The test result indicated that faults on different parts could be classified while different severity fault could not.(5) To avoid decreasing of the signal length in morphological wavelet decomposition, a new multiscale morphological undecimated wavelet decmopositon(MUWD) based on differential morphological filter was proposed. Morphological undecimated wavelet energy feature vector and energy entropy were defined. Based on morphological undecimated wavelet and process, information fusion fault diagnosis methods were proposed. The experiment indicated that both faults on different part and different severity faults could be distinguished.