Research On Fault Feature Extraction Of Rolling Bearing Based On Vibration Signal Analysis

Rolling bearing is widely used in Mechanical engineering field.It was frequently adopted,generating complicated equipment noise,under high probability of malfunction,even stopped rolling.These were mostly caused by the failure of the rolling bearing,which brought economic losses to the industry production,even jeopardize human life and pollute environment.Thus monitoring in real time and adopting the fault diagnosis is vitally valuable to the bearings’ steady working.Feature extraction is a key technical component at fault diagnosis.When fault appears in rolling bearings,the abnormal statues information often lays in the vibration signals.This paper started with the vibration signal analysis,dived in the research of extracting the failure feature in the application.Studied the solution for the main difficulty points,which formed the rolling bearing fault feature extraction method based on LMD and LLE.The works of this thesis are as follows:1.Generated three typical simulated malfunction test signals,which are based on the single-point excitation response model on the inner ring,outer ring and the rolling element.Minimized the model error in conventional sense,and acted as the foundation of the fault feature extraction research of this paper.2.Studied the foundation of the local mean decomposition algorithm combined with the problem of the non-linearity and non-steady of the vibration signal in real world test.Compared the strong point between the LMD algorithm with other classical empirical mode decomposition algorithms.Modified the shortage of LMD when get the smooth component,eliminated the infection in feature extracting caused by the false component.3.Modified LMD algorithm,proposed the fault extract method of the first order fault feature based on LMD permutation entropy(PE),studied the description of system complexity under PE.This method utilized the sensitivity of permutation entropy caused by the minor variation of the system,aimed at the problem of weak fault information overwhelmed by noise as well as the signal component were complicated.4.Proposed the second order fault feature extraction method based on LLE and multi feature fusion.Studied related field of local linear embedding,and optimized the standard of K-nearest neighbour selecting criterion,also combined the designed sensitive parameter high dimensional feature vector construction method.This method solved the "dimensional curse" during the rolling bearing sampling phase,which introduced redundant information.