The Research And Development Of The Visualization Clustering System Oriented To PDM

As information technology deeply put into operation in manufacturing, Product Data Management(PDM)well connects information systems such as CAD,CAM,CAPP,CAE and so on, which are so-called "Information Isolated Island", and becomes the indispensable technology with which company can remain invincible position in society. Data volume stored in the enterprise databases of PDM system at present is very large, yet the knowledge implicit in the data has not been exploited substantially. Thus the data mining technology is used in Manufacturing Industry more and more, and cluster analysis in data mining is becoming a focus point. The research of the special clustering system oriented to PDM is now a very active subject in Informationization for Manufacturing Industry. The structure and model of the visualization special clustering system, DMINING oriented to PDM is introduced in the thesis, based on breaking traditional general data mining systems. Firstly, the improved K-means algorithm is introduced in cluster arithmetic module of this system. K-means algorithm is a classical partition method in cluster analysis widely used in many fields of scientific research. However, the algorithm heavily depends on the sequence of data-inputting and easily gets into local optimum for random selecting initial cluster centers. In addition, when the square-error criterion is applied to evaluate the clustering results, the objects in one cluster will be divided into two or more clusters for minimizing the value of it. Aiming at above disadvantage and a great deal of complicated data in PDM systems, an improved K-means algorithm based on effective techniques of multi-sampling and once-clustering to search the optimal initial values of centers is used in DMINING system. Experimentations used to validate the improved K-means algorithm show that it is much better than K-means algorithm in stability. Secondly, because we now have problems about a lot of coding and having errors easily in developing visualization data mining system, the proposed system employs the technique of software Matcom to embed the mathematics calculating function of Matlab into Visual C++, realizing database operation and data analyzing with friendly and interactive interface, which will reduce lots of programming efforts and assure program accuracy. Furthermore, calculation and visualization of clustering in data mining will be implemented compactly. This system has been operated in database of KMPDM, analyzing marketing condition of some parts in part database. 330 groups of data has been analyzed, and the result has 4 clusters with clear and visualization cluster result. DMINING has been used in anomaly detection of network intrusion in PDM system with the result of 87.9% detecting rate and 0.80% miss warning rate in the experiment of analyzing KDD Cup 1999 Data.