Multi-objective optimization of CO₂ laser cutting for PMMA using machine learning and gray relational analysis integration

Laser cutting is a non-contact manufacturing method valued for its precision, speed, and cost-effectiveness, though its performance remains strictly contingent on the complex interplay of process parameters. Consequently, this study explores the feasibility of utilizing an integrated model to predict and optimize these variables within a defined experimental window, particularly in PMMA cutting applications. This study develops a combined predictive and multi-objective optimization framework to simultaneously minimize kerf width, surface roughness, and machining time. A full factorial design (3 3 ) was employed to systematically map the design space across three control factors: laser power, cutting speed, and material thickness. Additionally, to model the underlying non-linear interactions within the orthogonal dataset, shallow Artificial Neural Network (ANN) and Random Forest (RF) models were developed and compared against analytical Ridge Regression (RR). Signal-to-Noise (S/N) ratio and regression analyses identified cutting speed as the most dominant factor governing both productivity and cut quality. It exhibited that using high laser power and high cutting speed on thicker materials can produce a narrower kerf and shorter cutting time; however, it also reduces surface quality. Within the limited dataset of 27 experimental runs, the compact ANN model demonstrated reasonable prediction capabilities compared to RF and RR. The validation using independent experimental runs yielded an acceptable average error rate of 6.19%, indicating its feasibility for initial process estimation. Furthermore, Grey Relational Analysis (GRA) identified the balanced optimal condition (A3B3C3), which achieved a reduced kerf width and shorter machining time with an acceptable surface finish. The findings demonstrated that a compact ANN model trained on orthogonal experimental data provides a reliable, cost-effective tool for predictive modeling and parameter tuning in PMMA laser cutting.