Research On Clustering Analysis Of Rotor Fault Data Set

Due to mechanical of strongly nonlinear and complex factors such as work environment,the mechanical fault characteristics are complexity.To achieve satisfactory result of fault diagnosis,using a single diagnosis technology has been unable to meet the needs of existing.New ideas and ways must be sought.At the same time,in the process of machine learning,the number of features often will be many.And there are redundant or irrelevant features.The redundancy and irrelevant features not only increase the complexity of the algorithm,but also reduces the computing accuracy.As an important tool,cluster analysis is used in many fields.It has become a research hotspot in recent years.In the face of data object is more and more complex,clustering research is faced with many new contents and challenges with the advent of the era of "big data".Based on the double span rotor test-rig,this paper started a related research by using feature weighting technique,nuclear method,clustering evaluation index and fuzzy clustering algorithm.In this paper,the work mainly includes the following:(1)Based on searching a lot of domestic and foreign literature,current situation and existing problems of clustering analysis in fault diagnosis are analyzed.The necessity of feature weighting and feature weighting methods are studied and discussed systematically.(2)Aiming at the problem of fault identification based on clustering algorithm the commonly used clustering algorithms were summarized by learning clustering basic theory.Based on kernel method and fuzzy C-means,a fault pattern recognition method was designed.Analyzed the data which was collected from the double span rotor test-rig,time-frequency domain features can be got.Then those features will be used in the algorithm which has been raised.Analysis shows that combined the kernel method and fuzzy C-means,the differences between the samples were increased by introducing the kernel technology.It can reduce the clustering error and realized clustering of the fault types.(3)Aiming at the problem of fuzzy C-mean algorithm:(1)The algorithm appears powerless to those data which with noise and outliers,unequal sample distribution of super sphere.(2)It doesn't differentiate among different sample characteristics on the contribution of clustering and ignores the different importance of clustering between the typical and ambiguous samples.(3)Generally the cluster number is need to be given in advance.The introduction of kernel methods is to solve the problem No.1.For solving problem No.2,feature weighting gives samples different weight.By clustering evaluation index searches for the optimal clustering number adaptively can settle the problem No.3.The proposed algorithm is applied to the simulation data set and double span rotor test-rig data sets.Analysis shows that the algorithm not only takes into account the advantage of the KFCM algorithm in dealing with noise data,but also pays attention to the influence of different sample characteristics of clustering.It can get accurate clustering number and improve the accuracy of clustering.The proposed algorithm is a kind of effective clustering method.(4)A fault identification system based on MATLAB GUI is designed.Four clustering algorithms are embedded into the system.It can realize the rotor fault data reading,feature processing and fault identification.The effectiveness of the system is proved by experiments.Clustering analysis technology has brought many new ideas and ways in the field of fault diagnosis.At the same time it is also in perfecting and developing.The study of clustering technology will promote the development of mechanical information technology to the data-driven science.