A deposition-agnostic reduced-order thermal modeling strategy for evolving geometries in directed energy deposition

Thermal history prediction in Directed Energy Deposition (DED) remains computationally prohibitive due to evolving adiabatic boundaries. Existing approaches face limitations: numerical models are generally expensive; analytical or semi-analytical formulations apply only to static boundaries; and data-driven surrogates require extensive offline training with limited generalization under dynamic boundary updates. To overcome these issues, we proposed a deposition-agnostic, boundary-aware reduced-order model based on Component Mode Synthesis (CMS) that maintains accuracy under arbitrary deposition patterns during DED processes. The build is partitioned into blocks, with fixed-interface normal modes and constraint modes pre-computed offline and incrementally assembled during arbitrary deposition in online simulation. In addition, a two-level dynamic truncation is adopted, which increases the retained modes near the moving heat source while keeping a lower order elsewhere to maintain the overall reduction efficiency. Under linear thermal assumptions, the method operates within a standard FEM workflow and delivers deposition-agnostic acceleration. Validation on three 2D DED benchmarks (single-layer, five-layer build, and a topology-optimized part) demonstrates relative errors of approximately 1%, computational speedups up to 7.7 × , and a reduction of DOF to 9.9% of the FEM baseline. Additionally, the approach preserves the numerical robustness of inactive-element paradigms and shows strong potential for extension into thermo-mechanical simulations, enabling efficient studies of deposition strategies.