Optimizing 3D printing parameters for enhanced tensile strength and efficiency using machine learning models

Abstract Reliable prediction of mechanical properties in fused deposition modelling (FDM) remains challenging because of the strong dependence of tensile performance on processing conditions. This study investigated the influence of printing temperature, layer height, and print speed on the tensile strength of PLA samples printed according to ASTM-D638 (Type IV). An initial Taguchi L9 experimental design provided baseline measurements, which were subsequently augmented to form a 125-sample dataset for machine learning analysis. Multiple regression models—including random forest, linear regression, support vector regression, decision tree, and XGBoost—were developed and evaluated using 10-fold cross-validation to ensure stronger generalizability. The novelty of this work lies in combining these individual models into a stacking ensemble framework and validating model behavior using SHAP interpretability. XGBoost emerged as the top performer, with an R 2 ² of 0.91, an MSE of 1.58, an RMSE of 1.083 and an MAE of 0.948, whereas the stacking ensemble (R 2 of 0.89, an MSE of 1.94, an RMSE of 1.093 and an MAE of 1.009) further improved stability with reduced variance across folds. SHAP analysis confirmed printing temperature as the most influential factor, followed by layer height and print speed, aligning model predictions with the underlying material-processing physics. Overall, the integrated approach provides an effective pathway for the data-driven optimization of the tensile strength of FDM-printed PLA components.