Research On Purification, Feature Extraction And Automatic Identification Of Axis Orbit Of Rotating Machinery

As a powerful diagnostic tool for rotating machinery fault, orbits contain a wealth of fault information. The effect of purification and automatic identification of the orbits feature determines the level of online monitoring and intelligent fault diagnosis for rotating machinery Therefore, expand a comprehensive research to the feature extraction and automatic identification, and development of special orbits analyzer is necessary.Firstly, under various fault causes, rotor vibration mechanism was detailed analyzed, theoretical calculation of force conditions, was carried out, both time domain waveform feature and frequency spectrum feature of axis orbits was introduced. Secondly, in the study of purification filter to original orbits signal, against non-stationary feature of orbits signal, taking into account of frequency domain localization properties of harmonic wavelet, use the harmonic wavelet to refine and decompose frequency band ,extract the data points of target band, reconstruct signal and synthesize orbit, This method can solve the "binary" feature can not. Experiments show the harmonic wavelet has the advantages and feasibility in non-stationary signals filteringThirdly, targeting at data acquisition feature of vibration signals of rotating machinery and feature of axis orbits graphics, a new rotor axis orbit purification method using mathematical morphology filters is proposed. With this method, noise interference can be eliminated and purified rotor center orbits can be obtained following filtration of vibration signals by a combination of on-off and off-on morphology filters, without requiring consideration of the rotor's vibration spectrum feature and neither any prior experience in rotor fault treatment. Simulation calculation and actual practice have proved the effectiveness of this method.Fourthly, as axis orbits are image symptoms, a recognition method based on geometrical feature and invariant moments is proposed. In this method, orbits feature are extracted in the premise of using axis graphics as binary image, the affine invariant moments features are more effective image feature. The simulation results show invariant moments do not change with the image translation stretching and rotation, Therefore, using invariant moments as the Shape feature is feasible and effective. In order to combined image feature and practical significance when identification, propose a Fourier descriptors based algorithm for feature extraction. Under different failure mode, the difference both Fourier descriptors and its cause was analyzed, studies indicate Fourier descriptors is a good representation for axis, and the presentation of this method is related to fault mechanism, associated with strong physical background.Because of the neural network has advantage for a highly adaptive learning capability, fault tolerant capability and Robustness, so it is very suitable for the problem of identification and classification. In the thesis, BP network for automatic identification of axis orbits is designed. The simulation results show the use of neural networks to identify orbit automatically has high accuracy.Finally, after implanting various processing algorithms about axis orbits in software, configuring some related test hardware, an integrated orbital analysis instrument was completed. Some experiments are carried out in State Key Laboratory of Mechanical Transmission Chongqing University and Chengdu Ningjiang Machine Tool Factory, obtained multiple sets of real orbits data, data analysis results show analysis procedure. is correct.