AI Breakthrough: Terra Quantum and Empa Accelerate Laser Welding Simulation
Terra Quantum and Empa unveiled an AI model that transforms laser welding simulation from slow finite element runs to near-instant predictions. The model, called LP-FNO, produces 3D melt-pool forecasts in about 8 ms, a roughly 100,000x speed improvement over traditional solvers. That performance brings predictive simulation into the control loop for manufacturing systems.
Unprecedented Speed and Accuracy for Industrial Applications
LP-FNO achieves predictive accuracy suitable for production: roughly 2.5% temperature error and an intersection-over-union score near 0.9 for melt-pool geometry. At 8 ms per prediction, the model supports real-time process control, feedback-driven parameter tuning, and high-fidelity digital twins that can run synchronously with physical equipment. Manufacturers can explore larger design spaces and respond to disturbances without waiting minutes or hours for a simulation.
The Technology Behind the Leap: LP-FNO
LP-FNO stands for Laser Processing Fourier Neural Operator. It builds on the Fourier Neural Operator architecture, which learns operators mapping input fields to solution fields across scales. A quasi-steady reformulation of the welding physics reduces temporal complexity, letting the model predict steady and evolving melt-pool states efficiently. LP-FNO handles both conduction and keyhole welding regimes and is resolution invariant, meaning it generalizes across grid sizes and geometric scales while delivering full 3D melt-pool outputs.
Broader Impact for Manufacturing and Digital Twins
The collaboration pairs Empa’s materials and experimental expertise with Terra Quantum’s AI and computational know-how. Although Terra Quantum is known in the quantum technology space, this project highlights how firms in that ecosystem apply advanced AI to industrial problems today. The net result is simulation that can be embedded in control systems, enable synchronized digital twins, and unlock parameter regions previously impractical to evaluate.
For decision makers and engineers, LP-FNO signals a shift: predictive models fast enough to operate live on factory floors, reducing trial time and improving repeatability for laser welding processes.
