Generative Recipes And Parametric HPC-simulation for Robotic Additive Manufacturing

Short description of the innovation study

GRAPH-RAM brings generative AI to robotic metal additive manufacturing. Today, engineers often rely on costly trial-and-error to find process parameters that can produce a stable multi-layer build. GRAPH-RAM turns this into an inverse-design problem: starting from a target geometry, deposition plan and process constraints, it will generate multiple candidate welding recipes and screen them before expensive physical trials. The project combines large-scale Simufact Welding simulations on HPC, a transformer-based model called RecipeFormer, deterministic physics checks and FEM verification, all orchestrated through RobTrack. The expected result is a complete "generate, validate, simulate, review" workflow, together with a physics-grounded training dataset, a trained generative model and clear KPIs showing how quickly feasible recipes can be identified.

Organisations involved: 

Main SME: 3D-Components AS

Technology expert: SimEvolution A/S