Application Of Characteristic Dimension Reduction Method In Fault Diagnosis Of Rolling Bearing Under High Dimensional Variable Condition

As one of the key parts of rotating equipment,rolling bearing has a large load and is easy to be damaged.Especially in the process of starting and stopping when the load and speed change dramatically,it is easy to cause serious accidents,which leads to great risks and hidden safety hazards in operation.In addition,bearing fault characteristics under variable working conditions are mostly shown as time-varying non-stationarity,which will provide a large amount of data in continuous monitoring of bearing conditions,increasing the difficulty of operation and analysis of fault diagnosis.Therefore,in the process of rolling bearing fault diagnosis under non-stationary working conditions,the research on the method of characteristic dimensionality reduction for mining the key information in the data is a crucial part to ensure the normal operation of the equipment.Based on multi-body dynamics simulation load bearing,analysis the characteristic of variable condition signals,determine the needs from various angles,in the form of multiple domain feature description under the variable condition bearing state,avoids the single domain characteristics is difficult to accurately describe the state of the bearing,and then verified by simulation experiments,determined the necessity of multiple domain feature set.The classical nonlinear dimension reduction method KPCA does not consider the influence of the similarity between features on the computational complexity and the separation effect in dimension reduction,which limits the improvement of the real-time performance and effectiveness of calculation and the improvement of the classification effect.Therefore,based on KPCA and clustering algorithm,a cluster KPCA method is proposed.The idea of clustering algorithm is used to cluster the similar features in the extracted multi-domain feature set to reduce the complexity of subsequent KPCA calculation.Then,KPCA is used to reduce the feature dimension after clustering,and the high-dimensional features are mapped to a feature space with high classification degree.Under the variable working conditions,the bearing operation is more complex,and the common fault diagnosis methods lack good generalization and adaptability under the variable working conditions,and the non-periodicity and non-linearity are more prominent.An improved method combining poly KPCA and supervised LDA was proposed.Firstly,the nonlinear feature of rolling bearing data was solved,and the feature dimension was reduced.Secondly,the possible problems of edge data classification and "small sample" were solved,and finally,support vector machine was used for diagnosis and recognition.The experimental results show that the accuracy of poly KPCA combined with supervised LDA method is 99.11%,56.61% higher than that of KPCA,and 11.25% higher than that of poly KPCA.It can be seen from the experimental results that the application effect of the method of poly KPCA combined with supervised LDA adopted in this paper to the rolling bearing fault diagnosis under the variable working conditions of nonlinear non-periodic outburst.This method not only reduces the computational complexity,improves the real-time performance and effectiveness of calculation,achieves a higher identification and diagnosis accuracy,has a stronger practicability,and effectively improves the efficiency of bearing fault diagnosis,so it has a strong practical application value.