Machine Learning Prediction of Raster Angle Effects on Mechanical Properties of Extrusion-Based Additively Manufactured Conductive Thermoplastic Polyurethane Composites

Abstract

Machine learning (ML) is frequently used for modeling complex relationships between material properties and processing conditions in additive manufacturing (AM). In this study, we investigated how fused filament fabrication (FFF) of conductive thermoplastic polyurethane (TPU) is affected by raster angle (RA). Nineteen different RA configurations (0°-90°) were tested and Young's modulus (E), ultimate tensile strength (UTS), break strain (BS), and strain energy density (SED), were measured. The results reveal anisotropic behavior, with RA = 45° yielding the best overall performance (E = 83.45 MPa, UTS = 6.47 MPa, BS = 89.85%, and SED = 4.368 MJ/m3), according to a composite desirability optimization. To capture and predict these trends, 35 supervised regression algorithms were implemented and compared for various metrics. High-order polynomial regression (Poly6) and support vector regressors with polynomial kernels (SVR-Poly6) achieved the best predictive accuracy, yielding a test R2 of up to 0.957. Moreover, top ML models predicted intermediate RAs (7.5°, 47.5°, 72.5°) within ±5% of the experimental values. This validated, data-driven framework enables optimization for flexible, load-bearing, and electrically functional 3D-printed composites.