Experimental Investigation and Machine Learning-Based Optimization of Abrasive Water Jet Drilling in GFRP Laminates

Abrasive water jet drilling (AWJD) has emerged as a promising technique for producing fastener holes in glass fiber reinforced polymer (GFRP) laminates, where conventional drilling is often limited by severe tool wear, delamination, and dimensional inaccuracies. This study systematically investigates the influence of nozzle diameter (d), stand-off distance (SOD), water jet pressure (WJP), and abrasive flow rate (AFR) on five critical responses: material removal rate (MRR), circularity error (CE), overcut (OC), taper angle (θ), and delamination factor (Fd). Controlled experiments were conducted using Taguchi’s L18 orthogonal array for two practical hole diameters (2 mm and 4 mm). The experimental results were analysed using analysis of variance (ANOVA), Pearson’s correlation analysis, and entropy-based weighting to quantify parameter significance and response interdependencies. To address the inherent multi-response and conflicting nature of AWJD performance, an unsupervised machine learning framework based on k-means clustering was employed for optimization. Unlike conventional regression or response surface methodology (RSM) approaches, the clustering technique effectively captured nonlinear relationships and identified stable operating regimes. The results reveal that nozzle diameter and WJP predominantly govern productivity, while AFR and SOD strongly influence geometric accuracy and damage characteristics. The derived probability-density surfaces highlight pronounced trade-offs between productivity and hole quality, underscoring the necessity of multi-objective optimization. The proposed framework provides practical and scalable guidelines for optimizing AWJD parameters to achieve a balanced combination of accuracy, integrity, and throughput in GFRP composite machining