A meta-heuristic evolutionary algorithm combined with XGBoost to predict the geometry characteristics of laser-based micro-milling in PMMA-based microchannels

Laser beam micro-milling is a non-contact advanced machining process that is highly precise, flexible, versatile, and cost-effective. In this study, microchannels in polymethyl methacrylate (PMMA) were fabricated using a CO₂ laser with varying input parameters, including laser power, cutting speed, and the number of passes. The focus was on evaluating Kerf Depth (KD) and Kerf Deviation (KDev). This study proposes a hybrid machine learning framework that combines XGBoost with Particle Swarm Optimization (PSO) for accurate prediction of KD and KDev in multi-pass CO₂ laser micromachining of PMMA. The motivation is to develop a lightweight, interpretable model that can offer reliable predictions without the need for extensive DOE. A total of 36 experimental trials were conducted under varying laser power (30–40 W), scanning speeds (20–25 mm/s), and number of passes (1–4). The collected data was used to train an XGBoost model, with PSO optimizing the hyperparameters. SHAP analysis revealed that the number of passes predominantly influenced kerf depth due to cumulative thermal effects, followed by laser power, which governed energy input and vaporization rate. Scanning speed affected dwell time and heat accumulation. The model showed high prediction accuracy for both KD and KDev (R² = 0.99, MSE ≈ 0.04). A graphical user interface (GUI) was created to allow real-time predictions and assist in process planning. The proposed framework provides a cost-effective tool for predicting kerf geometry. It has practical use in microfabrication. Future work will focus on applying this model to other materials and laser settings.