Adaptive Spatiotemporal Thermal Model for Real-Time Temperature Prediction in Directed Energy Deposition

Directed Energy Deposition (DED) is an advanced metal additive manufacturing process involving complex thermal dynamics that significantly impact the quality and structural integrity of fabricated components. Effective thermal management in DED requires accurate and real-time temperature predictions throughout the workpiece. However, traditional simulation methods often lack the computational efficiency necessary for real-time control applications. In this study, we introduce a novel physics-based, data-driven, and control-oriented spatiotemporal thermal model featuring adaptive meshing to enhance real-time temperature prediction accuracy and control performance. The proposed model dynamically refines the computational mesh in regions with steep thermal gradients and progressively coarsens it in areas experiencing relatively stable thermal conditions as additional deposition layers are built, substantially reducing computational requirements. Model validation against high-fidelity finite difference (FD) simulations shows mean absolute percentage errors ranging from 5.56% to 8.30% across multiple deposition layers of the entire component, and from 3.77% to 6.25% in regions near the melt pool characterized by steep thermal gradients. Remarkably, the model completed temperature predictions for five deposition layers in just 33.7 seconds, significantly faster than the 545.9 seconds required by FD simulations and well within the total printing time of 101.4 seconds, demonstrating suitability for real-time control applications. The developed framework offers a robust, scalable, and computationally efficient solution for thermal management in DED, potentially enhancing closed-loop control capabilities and promoting broader industrial adoption.