Construction of grinding wheel decision support system using random forests for difficult-to-cut material

Abstract

The grinding wheels comprise three elements (abrasive grain, bonding material, and pores) and five factors (grain size, grain type, grade, structure, and bonding strength). However, owing to their impact on the material's surface quality and grinding efficiency, optimal combinations must be identified for each work material. The correct setting of the elements and factors is challenging and requires in-depth knowledge and engineering expertise. In this study, a random forest data-mining method was used to construct a system that can determine the abrasive grain, grain size, and bonding strength from various combinations of material property values. The verification of the developed system was conducted through grinding experiments on difficult-to-cut materials using the general-purpose grinding wheels recommended by the system. This system facilitates highly accurate interpolation prediction by adding an interpolative property value set in 10% increments to the learning data. A grinding experiment using Inconel 718, a difficult-to-cut material that did not exist in the learning database, resulted in a 12% reduction in wear on the recommended grinding wheels. Thus, the usefulness of the system constructed using the random forest method was established.