A framework to transfer thermo‐elastic properties of short fiber‐reinforced polymers between extrusion deposition additive manufacturing systems

Abstract This study presents a physics‐guided transfer learning framework to efficiently transfer the thermo‐elastic properties of additively manufactured short fiber‐reinforced polymers (SFRPs) across different extrusion deposition additive manufacturing (EDAM) systems. When the same material is printed on different machines, variations in microstructure lead to changes in material properties, necessitating extensive material testing for each new printer to enable accurate process simulations. The proposed framework aims to address this challenge by significantly reducing the extensive material characterization needed to create digital material cards for additive process simulations. It achieves this by first inferring the thermo‐elastic properties of the fiber and matrix from experimental measurements of a printed system's effective properties. These inferred properties, combined with microstructural descriptors, are then used to predict the effective properties of a different printing system. Notably, the transfer process only requires the fiber orientation tensor from the second system, minimizing the characterization workload. Our results show that the predicted composite coefficient of thermal expansion (CTE) for the second printing system is closely aligned with the experimental values, validating the ability of the framework to transfer information between systems. Furthermore, we perform a sequentially coupled thermo‐mechanical analysis of two different geometries in an EDAM simulation to investigate the effects of prediction errors. The predicted residual and spring‐in deformations using the transferred CTE values align well with those obtained from experimental CTE values, confirming the accuracy and efficiency of the proposed method. Highlights Physics‐guided transfer learning for thermo‐elastic properties of composites. Reduces material characterization efforts for additive manufacturing simulations. Infers properties from one printer to predict another printer's properties. Requires only fiber orientation data for the second printer.