| Abstract Scope |
Welding and metal additive manufacturing (AM) involve intense, localized heating and cooling that generate residual stresses and distortion. While physics-based simulation can predict these effects, conventional finite element methods are often too computationally expensive, requiring months or years for large-scale analyses. DR-Weld, developed at Oak Ridge National Laboratory, addresses this limitation with an adaptive time-acceleration algorithm integrated into a hybrid explicit–implicit framework optimized for GPU computing. This approach overcomes stability constraints while maintaining accuracy and scalability, while supporting both solid and shell elements for a wide range of structures. DR-Weld achieves speedups of 100× to over 2000× compared to conventional solvers, reducing simulation times from months to hours or days. This advancement makes high-fidelity, full-scale welding and AM simulation practical for industrial applications. |