A machine learning–integrated framework for mechanical property prediction of FDM–printed PLA

Additive manufacturing has transformed modern production by enabling the fabrication of complex geometries while minimizing material waste. Fused deposition modeling with polylactic acid is widely adopted among its various techniques due to its accessibility and sustainability. However, accurately predicting mechanical properties, such as ultimate tensile strength, remains critical for extending its industrial applicability. This study presents an integrated framework combining machine learning, experimental validation, and finite element analysis to predict and explain the tensile behavior of fused deposition modeling-printed polylactic acid components. A standardized dataset of 422 samples was compiled from multiple sources, and ten machine learning models, including AdaBoost, Bayesian Ridge, CatBoost, decision tree, Elastic Net, k-nearest neighbors, linear regression, random forest, support vector regression, and TabNet, were developed and evaluated. TabNet achieved the highest individual prediction accuracy, while ensemble strategies combining multiple models improved performance, achieving a coefficient of determination (R²) of 99.09%. External validation with 309 independent samples confirmed the model’s generalization ability, with prediction errors remaining below 10%. Complementary finite element analysis simulations were conducted using ANSYS Explicit Dynamics, simulating tensile tests on ASTM D638 Type I specimens and incorporating experimentally derived stress–strain behavior. The simulations visualized stress pathways and accurately predicted fracture modes, which varied with print orientation. This work advances predictive modeling for additive manufacturing, providing a validated, scalable approach to optimize material performance and support smarter, material-efficient design in aerospace, biomedical, and automotive applications.