Physics-informed machine learning surrogate for scalable simulation of thermal histories during wire-arc directed energy deposition

Wire-arc directed energy deposition (DED) has emerged as a promising additive manufacturing (AM) technology for large-scale applications. However, the complex thermal dynamics inherent to the process present challenges in ensuring structural integrity and mechanical properties of fabricated components. Finite element method (FEM) simulations have been conventionally employed to predict thermal history during deposition. However, their high computational demand increase significantly with scale. Given the necessity of multiple repetitive simulations for heat management and the determination of optimal printing strategy, FEM simulation quickly becomes unfit. Instead, advancements have been made in using trained neural networks as surrogate models for rapid prediction. However, traditional data-driven approaches necessitate large amounts of relevant and verifiable external data, either from simulation, experimental, or analytical solutions, during the training and validation of the neural network. Regarding large-scale wire-arc DED, none of these data sources are readily available in quantities sufficient for an accurate surrogate. The introduction of physics-informed neural networks (PINNs) has opened up an alternative simulation strategy by leveraging the existing physical knowledge of the phenomena with advanced machine learning methods. However, the practical application of PINNs for real-world large-scale wire-arc DED has been rarely explored, particularly within the context of structural engineering. This study investigates one of the necessary steps for up-scaling PINN with a focus on advanced and effective sampling of collocation points — a critical factor controlling both the training time and the performance of the model. The results affirm the potential of PINNs to outperform FEM in terms of wall-clock times, while maintaining the desired accuracy and offering resolution-agnostic evaluation. Further discussion provides an outlook on the future steps for improving the PINNs for wire-arc DED simulations. • PINN was trained only on physics, needing no experimental or external data. • Advanced sampling strategy improved model efficiency and accuracy. • PINN reduced runtime by up to 98.6% compared to FEM. • PINN provides resolution-agnostic predictions in both time and space. • Applicable to large-scale, structural wire-arc DED thermal simulations.