Use of Catalog Mining to Extract Valuable New Knowledge Hidden in Trivial Parameters

Abstract

We have developed a process that uses both hierarchical and non-hierarchical clustering methods to mine data in tool catalogs. Principal component regression is used for quantifying the correlation between the predictor and criterion variables, and multiple regression analysis is used for creating an end-milling condition determinant matrix for each cluster. We fixed the outside diameter of the tool shape parameter as a constant trivial value and examined the correlation between the other tool shape parameters and the end-milling conditions. We thereby extracted valuable new knowledge hidden in trivial parameters and built a hypothesis in regards to data-mining effect. We found that cutting speed is the most important of the criterion variables and that the number of determination coefficient is no less important for determining prediction accuracy of end-milling condition decision equations. End-milling condition decision determinants derived from our data-mining process are important indicators for adjusting end-milling conditions on the basis of end-milling efficiency and tool life.