Research On Fault Feature Extraction Method Of Rotor Vibration Signal

With the development of science and technology,the precision and operation requirements of mechanical equipment are becoming more and more high,which will produce massive data,which lead to mechanical fault diagnosis has stepped into the era of big data.How to extract features from the data has become an urgent problem when face big data.Due to the vibration signal produced by the machine and equipment not only contains the linear vibration information,but also the nonlinear vibration signal.Therefore,it is necessary that get rid of redundant information form nonlinear vibration signal.The main work of this paper includes the following parts:(1)In order to solve the problem that the traditional Euclidean distance sometimes fails in the high-dimensional observation space,a new method is proposed to calculate the Euclidean distance of the nearest neighbor point in the local preserving projection algorithm,namely,the nearest neighbor probability distance and the distance of the nearest neighbor,which is used to replace the local preserving projection algorithm,which is used to calculate the Euclidean distance of the nearest neighbor point.A local preserving projection algorithm based on nearest neighbor probability distance(Nearby Probability Distance Locality Preserving projection,NPDLPP)and a K-nearest neighbor classifier based on nearest neighbor probability distance are proposed.The effectiveness of the improved method is proved on a two-span rotor test-bench.(2)Aiming at the difficulty of identifying bearing fault types,a novel feature extraction model is proposed.The new model includes empirical Wavelet transform(EWT)and multi-scale Permutation Entropy.Firstly,the original vibration information is decomposed adaptively by empirical wavelet transform(EWT).A series of AM-FM components are obtained,and then the redundant variables are removed by correlation analysis and their multi-scale permutation entropy is calculated according to the selected AM-FM components.Finally,the established feature set input into the GG clustering algorithm,which is verified by the rolling bearing data from the Electrical Engineering Laboratory of case Western Reserve University,USA.(3)A new fault diagnosis model that contains the empirical wavelet transform and the AR model is proposed.The AR model reflect the effective information of the equipment's running state,and the AR model's autoregressive parameters reflect the effective information of the equipment's running state.The autoregressive parameter of AR model is used as the characteristic vector to express the operation of the equipment.However,the AR model is aimed at the processing and analysis of stationary signal,therefore,it is necessary to preprocess the signal before establishing the AR model,so an AR model based on empirical wavelet transform is proposed,and the results are input into FCM clustering to verify its validity with the data of rolling bearings.