Accelerating Simulation Speed and Efficiency Using AI Physics

The aerospace and defense industries have long been at the forefront of engineering and technological innovation across space exploration, research and warfighting domains. However, staying ahead of threats and ensuring ongoing readiness becomes increasingly complex as the global landscape grows in competitiveness and geopolitical volatility. Cutting-edge fields in AI-driven engineering and computing such as AI Physics are quickly helping leaders meet these challenges to unlock unprecedented levels of enhanced speed, accuracy, and efficiency over traditional modeling and simulation engineering methods.

The concept of AI physics merges probabilistic physics-based AI algorithms with deterministic simulation, which enhances the computer-aided engineering (CAE) based methods like computational fluid dynamics (CFD) and finite element analysis (FEA) solvers. This blend allows for making highly accurate predictions at high speeds. By using first-party data, R&D teams can train AI Physics models to produce highly accurate predictions based on their specific needs and past domain experience.

This discussion will demonstrate a workflow optimization of the NASA High lift CRM model using the FUN3D software and Graph Neural Networks and reduction in time to solution from hours to minutes. The audience will learn about the deep learning architecture used for this workflow and the performance optimization of the model training using the latest GPU architectures.

Similarly a turbomachinery CFD simulation workflow using CREAT-AV Kestrel will be demonstrated as well to drive home the possibility of 1000x acceleration in design evaluation.

PRESENTER

Kelly, Kevin (Presenting for Cook, Kyle)
kcook@rescale.com
925-303-6536

Rescale, Inc.

CO-AUTHOR

Kelly, Kevin
kpk@rescale.com

CATEGORY

Artificial Intelligence / Machine Learning usage for HPC Applications

SYSTEMS USED

Rescale Cloud and Carpenter

SECRET

No