@Article{jrm.2026.02026-0086,
AUTHOR = {Martha L. Sánchez, Juan E. Lasso, G. Capote},
TITLE = {Machine Learning-Based Prediction and Optimization of 3D-Printed PLA Biocomposites Reinforced with Vegetal Fibers},
JOURNAL = {Journal of Renewable Materials},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/jrm/online/detail/28302},
ISSN = {2164-6341},
ABSTRACT = {The behavior of 3D-printed biocomposites as substitutes for products made with plant fibers depends on the factors involved during the printing process. This study focuses on the use of a predictive model to evaluate the effect of printing parameters on the physical and mechanical properties of biocomposites printed with PLA filaments reinforced with vegetal pulverized fibers. During fabrication, the following factors were varied: reinforced filament type, build orientation, layer height, infill density, and infill pattern. The physical and mechanical properties of the printed specimens (density, absorption capacity, swelling percentage, rupture modulus, and flexural elasticity modulus) were determined following the procedures established in the ASTM D1037-20 and NTC 2261-03 standards. An analysis of variance based on response surface methodology was performed to verify the statistical significance of the printing parameters, as well as their interaction. To define the prediction model, two machine learning algorithms were implemented and compared: Extreme Gradient Boosting (XGBoost) and Gaussian Process Regression (GPR). XGBoost was selected, and SHAP analysis was performed to interpret the contribution of each parameter to the model predictions. A Non-dominated Sorting Genetic Algorithm II (NSGA-II) was implemented in order to optimize the printing parameters, enabling the identification of configurations that satisfy the minimum requirements for structural applications in humid environments. A total of sixteen optimal configurations met the standard criteria, including modulus of rupture (MOR) >16 MPa, modulus of elasticity (MOE) >2400 MPa, and swelling percentage (Sw) <10%. Two of the most representative configurations were experimentally validated to assess the optimization results. The results demonstrate that an increase of more than 75% in infill density does not result in significant changes in the mechanical behavior of the biocomposite.},
DOI = {10.32604/jrm.2026.02026-0086}
}