Analysis of drilling conditions by a catalog mining method based on Fuzzy c-means algorithm

Abstract

To support the engineer in the selection of drills and their machining parameters, we propose a novel catalog mining system based on data mining techniques applied on drill catalogs by using twice the Fuzzy c-means method along with the Maximum Information Coefficient (MIC). We first perform the clustering algorithm Fuzzy c-means that returns the membership degree of every point to each cluster on the parameters defining a series of tools. We then use the maximum information coefficient (or MIC, index measuring the correlation between two parameters) to find the drill or material properties that have the most influence on the drilling conditions in each of the clusters in order to realize a second clustering that includes the drilling conditions. The Davies-Bouldin index (DBI), which evaluates the dispersion inside the clusters, is used to assess the result of the second clustering and find its optimal parameters. Finally, a multi linear regression is used to find the equations predicting the drilling conditions in each sub-cluster. The mean squared error indicator is used to validate the result of the prediction. A new flexible index based on the membership degree value computed by the Fuzzy c-Means algorithm is proposed to filter the points and clarify the borders of the clusters in order to optimize the data used in the regression.

Keywords

• data mining
• fuzzy c-means
• maximum information coefficient
• davies-bouldin index
• tool catalog
• machining conditions
• optimal parameters
• drill
• computer aided manufacturing