Research On Fault Feature Extraction Of Rolling Bearing Based On LLE

Rolling bearing is a key component connecting shaft and shaft seat in rotating machinery.When it is running at high speed and high load for a long time in the industrial field,it is easy to increase the failure probability of bearing components,resulting in property losses and casualties.Bearing operation will produce vibration signals,in which the key information is extracted and analyzed,which is closer to the real engineering situation,so as to improve the accuracy of bearing fault identification,which has very important practical engineering value.Bearing vibration signal contains a lot of redundant information,which is easy to cause "dimension disaster" when feature extraction,which leads to high computational complexity and affects the accuracy of bearing fault diagnosis.To solve this problem,this paper proposes an improved LLE algorithm based on the Local Linear Embedding(LLE)algorithm in the nonlinear dimensionality reduction algorithm,and combines it with the Support Vector Machine(SVM)algorithm to propose a complete bearing fault diagnosis method.The main research work is as follows:(1)LLE algorithm is sensitive to nearest neighbor parameters when mining essential structures in high-dimensional space,and fixed neighborhood parameters can easily lead to topology instability,which directly affects the effect of LLE algorithm in extracting significant features.To solve this problem,a Local Linear Embedding algorithm Based on Adaptive Neighborhood(AN-LLE)is proposed in this paper.First,the nearest neighbor of the sample is selected by the cosine distance,and then,according to each sample and its two nearest neighbor points,the Schmidt orthogonal basis is constructed,and the similarity between the orthogonal basis formed by other nearest neighbors and the above-mentioned Schmidt orthogonal basis is evaluated,and the adaptive neighborhood of the sample is constructed by selecting the nearest neighbors with similar orthogonal bases.Finally,the adaptive neighborhood is applied to the neighborhood selection of LLE,so as to improve the accuracy of neighborhood selection and the stability of topology,and realize the full mining of the essential structure of high-dimensional data.A large number of experiments are carried out on the standard bearing data set of case Western Reserve University(CWRU),and the experimental results show that the algorithm can effectively extract the significant features of bearing faults.(2)In order to solve the problem that it is difficult to mine the local structure of highdimensional data by considering only single structure information in LLE algorithm,an Adaptive Local Linear Embedding algorithm Based on Weight Fusion(AN-WFLLE)proposed.Firstly,the algorithm uses double weights to reconstruct the sample adaptive neighborhood weight matrix in AN-LLE algorithm by using Gaussian kernel function and minimum reconstruction error function,and then fuses the two weight matrices by adjusting parameters.finally,the algorithm uses local geometric structure to extract data features and keep the structure embedded in low-dimensional space to extract data features effectively.Experiments are carried out on CWRU data sets,and the experimental results show that the algorithm is effective.(3)Combining the above two improved methods with SVM algorithm respectively,a complete set of rolling bearing fault diagnosis method is proposed to complete the fault identification task of bearing data.The method is applied to the fault testing platform of rotating machinery for experimental verification.The experimental results show that the proposed two methods show a higher level of accuracy than other similar algorithms,indicating that the algorithm has better application value in practical engineering.