Optimization and Prediction of Mechanical Properties of Additively Manufactured PLA/GNP Composites via Response Surface Methodology and Machine Learning Models

This study investigates the optimization and prediction of mechanical properties in 3D-printed PLA composites reinforced with graphene nanoplatelets (GNP). The effects of GNP content (0, 2, and 5 wt.%), nozzle temperature (190-210 °C), print speed (20-60 mm/s), and layer thickness (0.15-0.35 mm) on tensile strength, Young's modulus, and hardness were analyzed using a central composite design, at three print orientations (0°, 45°, and 90°). Compared to pure PLA, the incorporation of 5 wt.% GNP led to a 67% improvement in tensile strength, a 205% increase in Young's modulus, and a 44% enhancement in hardness. Advanced machine learning models, such as XGBoost and Gaussian Process Regression, were employed for prediction, with R² values exceeding 0.99 and MAPE below 4%. The models were validated using K-Fold Cross-Validation (K = 5), ensuring reliable and robust predictions while preventing overfitting. SHAP (Shapley Additive exPlanations) analysis indicated that GNP composition and layer thickness were the most influential factors, with SHAP values ranging between ±0.75. The Gaussian Process model outperformed both Linear Regression and XGBoost, achieving the highest R² of 0.9900 ± 0.0021, the lowest MSE (0.6593 ± 0.1054), RMSE (0.812 ± 0.323), MAE (0.6755 ± 0.1123), MAPE (3.157% ± 0.320), and RRMSE (3.409% ± 0.513), highlighting its superior predictive accuracy and stability. This integrated methodology, combining experimental optimization, ANOVA, and interpretable machine learning, presents a promising and potentially robust strategy for optimizing the mechanical performance of GNP-reinforced PLA composites, emphasizing their potential for high-performance engineering applications.