Sustainable additive manufacturing with FDM: Taguchi-based parameter optimization and AI-driven prediction of mechanical and environmental metrics

Abstract

Additive manufacturing (AM), particularly fused deposition modeling (FDM), has emerged as a transformative technique for producing customized, lightweight, and environmentally sustainable components. However, identifying optimal process parameters that balance mechanical performance with environmental impact remains a critical challenge. In this study, a comprehensive experimental and data-driven approach is employed to investigate the influence of key FDM parameters—infill density (ID) (30%, 60%, 90%), layer thickness (LT) (200, 300, 400 µm), printing speed (PS) (50, 100, 150 mm/s), and infill pattern (IP) (Cubic, Triangles, Concentric)—on the mechanical and sustainability-related performance of acrylonitrile butadiene styrene (ABS), polylactic acid (PLA), and carbon fiber-reinforced polylactic acid (Cf/PLA) specimens. A Taguchi L9 orthogonal array was used to design the experiments, and the resulting samples were subjected to tensile testing, surface roughness measurement, energy consumption monitoring, and carbon emission (CO 2 ) estimation. The maximum tensile strength was observed in Cf/PLA samples (up to 53.31 MPa), while PLA samples displayed a strong balance between surface quality (Ra ≈ 8.5–9.1 µm) and mechanical integrity. Energy consumption and carbon emission data revealed a direct correlation with ID and material type, with Cf/PLA exhibiting the highest environmental footprint. ANOVA revealed that ID was the most statistically significant factor influencing all target properties, contributing more than 60% to the total variance in experimental results across all material types. Scanning electron microscopy (SEM) analyses further illustrated microstructural differences and layer adhesion quality across different process settings. To enhance predictive capability and reduce experimental effort, three machine learning (ML) models—artificial neural networks (ANN), random forest regression (RFR), and gaussian process regression (GPR)—were developed to forecast mechanical and environmental outputs. The ANN yielded the highest accuracy across all metrics, with R 2 > 0.995 and MAPE < 10.5%. Validation test results confirmed the generalizability of the models, showing that the average prediction error remained below 2% across all ML algorithms. These findings demonstrate the effectiveness of integrating Taguchi-based optimization and AI-driven prediction for achieving sustainable, high-performance FDM manufacturing, providing a robust framework for future design and process optimization in AM applications.

Topics & Concepts