Scalable path planning and reduced order modeling for temperature optimization in Direct Energy Deposition

Abstract

Direct energy deposition (DED) processes, including laser DED and wire-arc additive manufacturing, provide high throughput and geometric flexibility, yet dimensional inaccuracies and heterogeneous properties frequently arise when sub-optimal tool paths create uneven temperature fields. Thermally aware path optimization is therefore essential but remains computationally prohibitive for complex geometries, forming the principal bottleneck in current algorithms. This study introduces an efficient planning framework that constructs a reduced order thermal model with GPyro, a machine-learning subspace technique that predicts temperature profiles only on the deposition layer. This allows swift layer-wise reductions, thereby extending the applicability of reduced-order models to arbitrary three-dimensional geometries. Additionally, the algorithm leverages the fast Fourier transform to evaluate temperature evolution efficiently, significantly reducing computational time while preserving accuracy. Compared to existing methods, the proposed approach achieves up to a 10${}^9$-fold reduction in pre-computing time and a 10${}^2$-fold acceleration in evaluating process temperature fields. Experimental validation on components with high overhang angles confirms the effectiveness of the algorithm, consistently producing high-quality, defect-free parts and demonstrating that coupling GPyro with iterative optimizers enables the optimization of deposition strategies, even for complex geometries.