Methods Of Rotating Machinery Fault Diagnosis Based On Mathematical Morphology And Local Mean Decomposition

Rotating machinery, such as hydraulic pump, rolling bears and gear, is an important transmission and control part in epuiment and unit. They are widely applied in the fields of metallurgical machinery, engineering machinery, precision machine tools, precision instruments, aeronautics and astronautics, automobile manufacturing, shipbuilding, petroleum chemical industry and so on. Thus the machinery is often faced to high temperature, pressure, speed and precision, which lead to their rapid healthy deterioration and even breakdown.Aiming at the problems, new methods based on mathematical morphology and local mean decomposition are proposed in the study. Both of mathematical morphology and local mean decomposition all work based on morphological features of a signal, thus they are all highly adaptive and data driven.Main novel works and achievements of this study are listed as follows:(1) Aiming at experience and randomness of flat structure element length selection, a method based on power spectral entropy and feature energy ratio is proposed. Firstly, a signal is filtered by morphological difference operators with different flat structure element lengths. Secondly, power spectral entropies of the filtered signals are computed. Lastly, according to power spectral entropy, the length corresponding to the minimum one of the power spectral entropies is the optimal one. The simulated signal and fault signals of hydraulic pump test and verify effectiveness of the method.(2) Problem of slipper wear of hydraulic pump condition assessment is studied based on a method of morphological difference operator and morphological index. Firstly, slipper wear fault vibration signals of four deterioration degrees are filtered by morphological difference operator to get clear feature information. Secondly, morphological index is extracted from the filtered signals. Lastly, sensitivity of morphological index to slipper wear fault and its deterioration degrees is analyzed. Signals of hydraulic pump test and verify effectiveness of the method.(3) Problem of early fault diagnosis of inner and outer race is studied based on a method of morphological difference operator and difference entropy. Firstly, slight and severe fault signals of inner and outer race fault are filtered by morphological difference operator. Secondly, abrupt information of signal is extracted by difference entropy, and the uncertainty and complexity of abrupt information are measured by difference entropy. Lastly, the faults are diagnosed successfully based on mind of the periodic time interval of abrupt point coincides with the periodic impulsive time intervals of inner and outer race fault. The simulation signal and the two fault signals of rolling bears test and verify effectiveness of the method.(4) Aiming at demodulation of hydraulic pump fault signals, experience and randomness of structure element length selection in single scale morphological analysis, a fusion method based on local mean decomposition and improved adaptive multiscale morphology analysis is proposed. Firstly, a signal of hydraulic pump is decomposed by to get several product functions. Secondly, some which are full of richest fault feature information are selected as data source. Lastly, the data source is demodulated by improved adaptive multiscale morphology analysis with plat structure element, semi-circle structure element and triangle structure element, and coefficient range of effective demodulation(where improved adaptive multiscale morphology analysis can work better than original adaptive multiscale morphology analysis) and the best demodulation coefficient can be got. Moreover, the demodulation results are contrasted with those obtained by adaptive multiscale morphology analysis, Hilbert transform, Teager Kaiser Energy operator and local mean decomposition. Signals of hydraulic pump test and verify effectiveness of the method.(5) Aiming at gear fault diagnosis, a method of local mean decomposition and generalized morphological fractal dimensions is proposed. Firstly, a signal is decomposed by local mean decomposition into several product functions. Secondly, some product functions which contain the richest feature information of original signal are used as data source. Thirdly, generalized morphological fractal dimensions are extracted, and some which can quantitatively and comprehensively characterize nonlinear information of gear running states are adopted as feature vectors. Lastly, gear faults can be diagnosed by kernel fuzzy C-means. In addition, influences of signal length, speed and torque on generalized morphological fractal dimensions are analysed. Gear signals test and verify effectiveness of the method.(6) Problem of selection of product function as data source is studied based on a method of kurtosis, energy and standard deviation. Firstly, original signal of hydraulic pump is decomposed by local mean decomposition into several product functions. Secondly, the three indices of kurtosis, power and standard deviation are extracted from original signal and product functions, and vectors are constituted by the three indices respectively. Then the Euclidean distance between original signal vector and each product function vector is computed. Lastly, some product functions corresponding to some smallest Euclidean distance values are selected to be reconstructed as data source. Signals of hydraulic pump test and verify effectiveness of the method.