Artificial intelligence-driven modeling and optimization of micro-drilling carbon dot composites via extreme gradient boosting and sensitivity analysis

This paper presents an explainable artificial intelligence framework that leverages Extreme Gradient Boosting (XGBoost) regression to model and optimize the precision micro-drilling of a novel carbon dot-reinforced epoxy nanocomposite. The XGBoost algorithm was specifically chosen for its proven efficacy in capturing complex, non-linear relationships in high-dimensional parameter spaces, achieving high predictive accuracy for thrust force, hole roundness, and damage area (R 2 test up to 0.98). A full-factorial Analysis of Variance (ANOVA) statistically validated the significance of machining parameters, identifying tool diameter as the most dominant factor (p < 0.001) and revealing critical two-factor interactions. To further transform this predictive model into a tool for physical insight, the Extended Fourier Amplitude Sensitivity Test (EFAST) was integrated, moving beyond a black-box to provide robust, model-agnostic sensitivity analysis. This hybrid methodology quantified tool diameter and nanofiller content as dominant factors, with model predictions and statistical post-hoc analysis confirming that 1 wt% carbon dots significantly enhance machinability by reducing thrust force and damage area. These AI-driven findings were physically validated via physicochemical characterization, linking performance gains to improved matrix–filler interactions. Finally, multi-objective optimization performed on the XGBoost model identified optimal parameters (2500 rpm, 10 mm/min, 0.30 mm, 1 wt%), achieving a composite desirability of 0.971. The primary contribution is this validated AI framework, which demonstrates how XGBoost, complemented by rigorous statistical validation, serves as a powerful and interpretable core for optimizing complex manufacturing processes.