Research On Three-Layer Weighted Clustering Ensemble Technology And Its Application In Bearing Fault Detection

Bearing is one of the key parts of machine operation.If the bearing is damaged or malfunctioning,it may lead to unstable operation or sudden shutdown of the machine,bringing safety risks to the working environment and operators.Therefore,for the equipment using bearings,bearing fault diagnosis has a very important necessity.The machine learning technology can automatically identify bearing faults by analyzing bearing vibration,temperature,noise,oil and other data,and predict the type and severity of faults,greatly improving the efficiency and accuracy of detection.Among them,cluster analysis,as an unsupervised learning method,can explore the internal relationships and rules among data through cluster analysis,providing important support for subsequent analysis and decision making,and attracting more and more scholars’ research.However,when a single clustering algorithm is faced with a specific problem,it often has a variety of different situations,such as high sensitivity to initial parameters,easy to be affected by noise or outliers,and unable to process different types of data.Clustering ensemble is a method that combines the results of multiple clustering algorithms to obtain more stable and accurate clustering results.Compared with the single clustering algorithm,clustering ensemble can reduce the randomness and deviation of clustering results,so as to improve the accuracy and robustness of clustering.Weighted clustering ensemble uses the similarity between base clusterings to weight different clustering results,and assigns greater weight to clustering results with higher similarity,which can effectively improve the accuracy of clustering.It has been documented that in clustering ensemble,the three levels of point,cluster and partition can reflect the internal structure of data from different angles.Although many domestic and foreign scholars have proposed different weighted clustering ensemble methods,most of them only consider one or two of the three aspects,and there is no unified framework that comprehensively considers the three aspects of point,cluster and partition.Aiming at the problems described,this paper proposes a PCP(point-cluster-partition)framework integrating point,clustering and partition,and further proposes a three-layer weighted clustering ensemble method based on this framework,which evaluates and weights the quality of clustering from various aspects,in an attempt to obtain better partitioning results.Finally,the algorithm proposed in this paper is applied to the bearing data.The work content of this paper mainly includes:(1)This paper proposes a PCP framework with three layers of weighting for points,clusters and partitions,and designs the weighting scheme of these three layers.Experiments on multiple UCI data and text data sets show that three-layer weighted can achieve higher accuracy and stability than unweighted and any level or two-level weighted.Compared with the more authoritative 7 weighted clustering ensemble algorithms in recent years,the average ranking of the proposed algorithm under the three evaluation indexes is the first place,which verifies the effectiveness of the proposed algorithm.(2)The method proposed in this paper is applied to bearing data.This paper has done a lot of related work in fault feature extraction and fault diagnosis both aspects: 1)In fault feature extraction,this paper discusses the optimal wavelet function and the number of decomposition layers for bearing data set.2)In the stage of fault diagnosis,the three-layer weighted clustering ensemble method proposed in this paper is used to carry out experiments,so as to further improve the accuracy of clustering in bearing fault diagnosis.The experimental results show that the method proposed in this paper can achieve more than 99%accuracy in bearing fault diagnosis,indicating the validity of the above work.