Multi-Response Optimization of FDM Process Parameters for PLA: The Strength-Energy-Roughness Trade-Off via Taguchi, ANOVA, and Machine Learning

Abstract

This study investigates the multi-response effects of five FDM process parameters on tensile strength, flexural strength, energy consumption, and surface roughness of PLA parts. A Taguchi L27 orthogonal array was employed by varying printing speed, infill density, layer height, wall thickness, and nozzle temperature at three levels. The responses were evaluated using analysis of variance (ANOVA), signal-to-noise (S/N) analysis, first-order regression, random forest, and gradient boosting models, with predictive performance assessed by leave-one-out cross-validation. Tensile strength was mainly governed by wall thickness (44.6%) and nozzle temperature (28.9%), while flexural strength was primarily influenced by nozzle temperature (28.3%). Energy consumption was dominated by wall thickness (43.5%) and layer height (28.2%), revealing a partial strength-energy trade-off. Surface roughness was strongly controlled by layer height (85.2%), while printing speed, nozzle temperature, and wall thickness also contributed. Linear regression provided reliable prediction for roughness (LOOCV R 2 = 0.956), whereas tree-ensemble models provided better predictions for strength and energy because of parameter interactions. Multi-objective Pareto and desirability analyses identified run 23 as the best overall compromise, achieving 49.2 MPa tensile strength, 96.8 MPa flexural strength, 124.6 Wh energy consumption, and 2.95 mu m roughness. The results provide a practical framework for balanced FDM parameter selection.