AI for Science and Engineering

Researchers across the basic and applied sciences have begun investigating how artificial intelligence could simplify and accelerate traditional simulation methods. For example, AI can help to quickly identify simulation parameters, generate preliminary solutions without the need to run many large-scale simulations, or replace time-consuming functions or equations within large simulations. HLRS's Department of Converged Computing is developing methods and workflows for integrating traditional simulation methods using high-performance computing and data-based methods using machine learning and AI. HLRS staff has also been working closely with scientists to develop applications using physics informed neural networks, which integrate known physical information into data-driven algorithms to ensure that their results accurately represent physical reality.

Foto von Dr. Celalettin Karadogan, der ein Experiment beobachtet, bei dem ein dünnes Stück Metall getestet wird.

Integrating simulation and AI for materials research

With the help of HLRS's supercomputer, a research team at the University of Stuttgart ran billions of neural network simulations to develop models of material forming processes.

Abbildung zeigt den Relexi-Arbeitsablauf.

Machine learning in computational fluid dynamics

Researchers from the University of Stuttgart developed a method called Relexi, which uses reinforcement learning to define turbulence models as an integrated component of a CFD solver.

Bild eines Roboters, der Gegenstände in einer Schachtel ergreift

Reinforcement learning in robotics research

Investigators at the Fraunhofer IPA and University of Stuttgart are developing methods for quickly training robots to perform tasks efficiently. The approach could enable greater customization in automated manufacturing.

Return to Artificial Intelligence.