Machine Learning Models in Drilling of Different Types of Glass-Fiber-Reinforced Polymer Composites

Abstract

The aim of the research presented in this paper was to simulate the relationship between selected technological drilling parameters (cutting speed, vc, and feed per tooth, fz) and cutting forces and the delamination in machining of a new glass-fiber-reinforced polymer (GFRP) composite. Four different types of new materials were manufactured with the use of a specially designed pressing device and differed in the fiber type (plain and twill woven materials) and weight fraction (wf) ratio, but they had the same number of layers and the same stacking sequence. A vertical machining center Avia VMC800HS was used for drilling holes with a two-edge carbide diamond coated drill. Measurements of the cutting force Fz in the drilling process conducted with variable technological parameters were carried out on a special test stand, 9257B, from Kistler. The new ink penetration method, involving covering the drilled hole surface with a colored liquid that spreads over the inner surface of the hole showing damage, was used to determine the delamination area. The cause-and-effect relationship between the drilling parameters was simulated with the use of five machine learning (ML) regression models (Linear Regression; Decision Tree Regressor; Decision Tree Regressor with Ada Boost; XGBRF Regressor; Gradient Boosting Regressor). Gradient Boosting Regressor results showed that the feed per tooth had the greatest impact on delamination—the higher the feed was, the greater the delamination became. Push-out delamination factors had higher values for materials that were made of plain woven fibers. The lowest amplitude of the cutting force component Fz was obtained for the lowest tested feed per tooth of 0.04 mm for all tested materials, with the lowest values obtained for the materials with twill fibers.