ML-Based Optimization OF CNC Machining Parameters For Complex Geometries

Abstract

This paper presents a machine-learning-based approach to optimize CNC milling parameters for workpieces with complex three-dimensional geometries. Three regression models Random Forest (RF), Support Vector Regression with radial basis function kernel (SVR-RBF), and a feedforward Artificial Neural Network (ANN) were trained on experimental data collected from an L27 orthogonal array of machining trials on Al 6061-T6 alloy. The input parameters comprised spindle speed, feed rate, depth of cut, and step-over ratio; the target responses were surface roughness (Ra) and flank tool wear (VB). Among the three models, RF achieved the highest predictive accuracy, with R² values of 0.941 for Ra and 0.923 for VB. A subsequent multi-objective optimization using NSGA-II on the trained RF surrogate produced a Pareto-optimal set of machining configurations. The best compromise solution reduced surface roughness by 18.3% and tool wear by 14.7% relative to the centre-point condition of the experimental design. These results demonstrate that data-driven surrogate models can replace computationally expensive finite-element simulations and trial-and-error approaches for parameter selection in multi-axis CNC machining of complex parts.