Research On Fault Diagnosis Method Of Rolling Bearing Based On Manifold Learning

Rolling bearing is an important component of rotating machinery and equipment,which is widely used in modern mechanized equipment.However,due to the complex working environment,operation fatigue and other factors,it is easy to cause parts wear,pitting,cracking and other failures,and then seriously damage the running state of rotating machinery and equipment.The local damage of bearing will directly affect the vibration state of bearing,so the vibration signal of rolling bearing contains a lot of important information about the state of components,which is often used for fault analysis of rolling bearing.However,the actually collected vibration signals are often nonlinear,high-dimensional and non-stationary,which makes it difficult to distinguish the types of bearing faults.Therefore,it is of great engineering significance to study a stable and efficient bearing fault diagnosis technology.In this paper,the working principle,vibration mechanism,failure mode and research status of rolling bearings are analyzed,and the local linear embedding algorithm(LLE)is deeply studied by combining manifold learning and the actual characteristics of rolling bearing vibration signals,so as to solve the existing research problems,and an improved feature extraction method is proposed and applied to bearing fault diagnosis.The research work includes:(1)Aiming at the problem that local linear embedding algorithm(LLE)relies too much on the selection of local neighborhood and ignores the role of global structure of data in feature extraction,which leads to the inability to accurately extract the essential features of rolling bearing data,a local fusion linear embedding method based on global constraints(GC-LFLE)is proposed.Firstly,this method constrains the data with low rank in the original space,and removes the data noise while constraining the global subspace structure of the data.Then,two geometric structures of the data are mined in the low rank subspace and the original space respectively,and the importance of the two structures is evaluated by reconstruction error,so as to realize the linear fusion of the two structures.Finally,the low-dimensional reconstruction function of data is constructed to obtain the essential characteristics of fault samples.(2)Aiming at the problem that LLE algorithm does not fully consider the rationality of local structure of data,which leads to the lack of data structure information,and the fixed neighborhood parameter k may lead to the distortion of neighborhood structure,an adaptive weighted distance local linear embedding method(AWLLE)is proposed.This method optimizes the internal structure of data by avoiding unreasonable neighborhood selection.AWLLE algorithm first calculates a series of "cam" parameters according to the original distribution of data to ensure the rationality of the local neighborhood structure of each data point,and then adaptively selects the neighborhood parameter k by using the systematic reconstruction error of the neighborhood data points to ensure that each data point with different distribution can select the local neighborhood that best reflects the real structure.(3)A complete fault diagnosis system of rolling bearing is established by combining the above-mentioned methods with support vector machine(SVM),and it is applied to the fault test platform of our laboratory to verify the vibration signal data of bearings under different working conditions.Finally,a software system based on Python language and QTdesigner GUI tool is developed to improve the practical engineering application value of this paper.