Multi-Objective Prediction of Mechanical and Thermal Responses of FDM-Printed Wood-Filled PLA Using Artificial Intelligence

Abstract

This study examines the effects of key fused deposition modeling (FDM) process parameters on the mechanical and thermal behavior of wood-filled polylactic acid (PLA/wood) using a multi-objective approach, while also proposing an artificial intelligence-based framework for prediction and optimization. Nozzle temperature (NT), printing speed (PS), and layer thickness (LT) were selected as three-level control factors. Flexural strength was measured experimentally, and the thermal decomposition onset temperature (T-onset) was determined using thermogravimetric analysis (TGA) under nitrogen at a heating rate of 10 degrees C/min over 25-600 degrees C, applying the tangent-onset method. The maximum strength was 84.04 MPa, whereas the highest T-onset reached 311.31 degrees C. ANOVA revealed that PS was the most influential factor for flexural strength, followed by LT and NT (p < 0.001), while NT dominated the variation in T-onset (p < 0.001). For prediction, artificial neural network (ANN) and support vector regression (SVR) models were developed. ANN provided superior accuracy for both outputs (flexural strength: R-2 = 0.99823; thermal decomposition onset temperature: R-2 = 0.99906). Model validity was further assessed using independent validation tests, where ANN and SVR predictions were compared with experimental results at five additional parameter sets. ANN consistently achieved lower absolute errors than SVR for both flexural strength and thermal decomposition onset temperature. The best-performing ANN and SVR models were coupled with a genetic algorithm (GA) to optimize their hyperparameters and further improve predictive accuracy.