Multi-objective evaluation of friction drilling characteristics using Taguchi design and ANOVA

Friction drilling is a non-conventional technique that utilizes the heat generated by a rotating tool to penetrate material, simultaneously forming a bushing around the hole. In this study, the influence of key process parameters like spindle speed, feed rate, and material thickness on multiple output responses, including thrust force, torque, bush length, cylindricity, top diameter, and temperature, was systematically investigated using Taguchi optimization technique and analysis of variance (ANOVA). The results revealed that spindle speed was the most influential factor for thermal and force-related responses, particularly temperature and thrust force, where increasing speed from 2000 rpm to 5000 rpm reduced thrust force by nearly 22% (from ~ 1500 N to ~ 1170 N) and peak drilling temperature by about 15%. Feed rate played a secondary role, contributing significantly when coupled with spindle speed, especially in torque and bush length formation, where higher feeds (65 mm/min) increased torque by nearly 18% compared to the lowest feed setting. Material thickness exhibited a strong effect on thrust force and cylindricity, with thrust peaking at 3 mm thickness (~ 1600 N) before dropping at 4 mm (~ 1200 N), reflecting nonlinear deformation resistance. Two-way interactions, notably speed–thickness and speed–feed, dominated in responses such as torque, cylindricity, and top diameter, contributing more than 40% of total variance. Residual plots confirmed the adequacy of the ANOVA models, with coefficients of determination (R²) ranging between 91.8% and 96.2%, indicating excellent model reliability. The study provides a consolidated understanding of parameter significance, serving as a quantitative basis for process optimization in industrial applications of friction drilling, particularly in automotive, aerospace, and lightweight structural component manufacturing. The regression models have therefore been re-evaluated using additional model-adequacy indicators, including adjusted R², predicted R², residual analysis, and prediction error, where applicable. The revised manuscript also clearly distinguishes between goodness of fit and predictive capability. In addition, confirmation experiments have been considered/performed at the predicted optimum condition, and the predicted and experimental values are compared using percentage prediction error.