Multi-objective optimization of mechanical properties in PLA/SCG/silane composites using synthetic data and XGBoost

Abstract

Polylactic acid (PLA) composites reinforced with spent coffee grounds (SCG) and modified with a silane coupling agent (VTMS) offer a sustainable alternative for applications requiring biodegradability and enhanced mechanical performance. This study employed a data-driven approach to optimize tensile strength and Shore D hardness by varying the contents of PLA, SCG, and silane. Seventy-five composite samples were fabricated and tested, exhibiting tensile strengths of 26.5–57.9 MPa and hardness values of 77.5–80.8 Shore D. A multi-output XGBoost regression model, trained on 60% of the data and validated on the remaining 40%, achieved strong predictive accuracy (R² = 0.884, MSE = 12.64 for tensile strength; R² = 0.908, MSE = 0.071 for hardness) after augmentation with 159 synthetic samples generated via jittering, Gaussian noise, and kernel density estimation. Multi-objective optimization using NSGA-II simultaneously maximized both properties, revealing Pareto-optimal compositions dominated by higher PLA and moderate SCG and silane contents. The best formulation (1490 g PLA, 121 g SCG, 20 g silane) achieved 53.33 MPa tensile strength and 80.06 Shore D hardness. The combined XGBoost-NSGA-II framework demonstrates an efficient, data-driven strategy for optimizing bio-composite performance while minimizing experimental effort.