Application Of Selective Clustering Ensemble Method Based On Spectral Graph Theory In Rolling Bearing Fault Diagnosis

Rolling bearings are crucial for the stable operation of industrial production and have been widely applied in various types of machinery.However,the diagnosis and maintenance of rolling bearings face severe challenges due to the complexity and diversity of their operating environments.This paper proposes a cluster member selection method based on spectral graph theory,using modern signal processing and feature extraction methods as a foundation,to construct a bearing fault diagnosis model.This method not only reduces manual intervention,improves diagnostic accuracy and efficiency but also provides valuable references for practical engineering applications.The main contents of this paper are as follows:Starting from the basic structure of rolling bearings,various failure modes of rolling bearings are analyzed,and the fault frequency characteristics generated by different parts of the bearings are studied,providing a basis and reference for processing vibration signals of rolling bearings.By introducing spectral graph theory,cluster integration,and other relevant theoretical foundations,theoretical support is provided for the proposed bearing fault diagnosis model.To address the issues of signal processing and feature extraction,a combined improved wavelet threshold denoising method based on CEEMDAN is employed.Firstly,the CEEMDAN method is used to decompose the original signal into several modal components,and key modal components are selected using kurtosis index.Then,the selected modal components are denoised and the signal is reconstructed using the improved wavelet threshold denoising method.Finally,the denoised vibration signal is subjected to wavelet packet decomposition to extract energy features from different subbands,thereby constructing a vibration signal energy feature dataset.Experimental results on simulated and measured signals demonstrate that this method can effectively suppress noise and clearly reflect actual fault characteristic information.Regarding the pattern recognition of different fault states in rolling bearings,this paper constructs a comprehensive framework for bearing fault data mining based on a selective cluster integration algorithm.Firstly,in the cluster member generation stage,Kmeans is used as the base clustering algorithm,with five different cluster numbers set to generate diverse base cluster members.Secondly,in the cluster member selection stage,a member pruning method based on spectral clustering is employed to eliminate outlier members,and a cluster member selection method based on spectral graph theory is used to automatically group high-quality cluster members.Finally,a consensus function is used to integrate the selected high-quality and diverse cluster members to achieve the identification of bearing fault feature data.Through experimental comparisons of different member selection strategies and the number of member groups,the significant superiority of the proposed selective cluster integration algorithm in bearing fault data mining is validated.