Drilling performance prediction using general regression neural networks

Abstract

Mechanics of cutting approach to drilling performance prediction is based on the three-dimensional oblique cutting theory and simpler orthogonal cutting data bank. The quantitative reliability of such models depend on numerous process variables and quantitative accuracy of the data bank for a given work material. In this paper architecture of General Regression Neural Network is proposed, that use process variables such as tool geometry and operating conditions to estimate thrust and torque in drilling. The developed networks are tested over a range of process variables to estimate thrust and torque. The quantitative accuracy of thrust and torque predictions using GRNN is found to be superior compared to the conventional methods. It is shown in this work that using the GRNN architecture the drilling forces are predicted within 3% of the experimental values.