Predictive optimization of hole quality in CFRP drilling using machine learning-assisted multi-criteria decision analysis

Drilling of carbon fiber reinforced polymer (CFRP) composites is a damage-prone machining operation in which defects such as delamination, poor surface integrity, and geometric inaccuracy can degrade structural performance. Achieving high-quality holes requires careful selection of process parameters while accounting for multiple, often conflicting, quality criteria. This study presents a predictive optimization framework for CFRP drilling that integrates machine-learning-based response prediction with multi-criteria decision analysis. Drilling experiments were conducted on a CNC milling center using coated and uncoated drills under different cutting conditions, and hole quality metrics including delamination, surface roughness, and circularity error were evaluated. A Random Forest regression model was trained to predict drilling-induced defects ( Fd , Ra , C ) from process parameters, while CoCoSo, MABAC, and TOPSIS methods were applied to the experimentally measured decision matrix to rank the 27 tested parameter combinations. A hybrid Analytic Hierarchy Process entropy weighting method was adopted to balance subjective and objective data. The TiN-coated drill at 2250 rpm and 0.025 mm/rev, which yielded the lowest surface roughness (Ra = 5.598 µm) among the tested conditions, was consistently identified as the top-ranked alternative by CoCoSo and MABAC and second-ranked by TOPSIS. The TiN-coated drill at 3200 rpm and 0.025 mm/rev, which yielded the lowest delamination factor ( Fd = 1.046), was ranked first only by TOPSIS. Predicted responses showed close agreement with experiments ( R 2 = 0.82–0.88), and sensitivity analysis confirmed the stability of these rankings under varying criteria weights.