July 2026 Summaries

Open-source AI datasets for AI Physics model training are now available on Rescale, facilitating quicker development of surrogate models. Engineers can utilize the DrivAerML dataset, a high-fidelity public dataset focused on automotive aerodynamics, which includes 500 parametrically morphed variants. This resource enables users to commence AI Physics model training without the need to generate their own simulation datasets, offering a streamlined entry point for those interested in AI Physics.

Agentic Digital Engineering, recently introduced by Rescale, aims to revolutionize AI-first product development by enhancing AI-assisted workflows, operationalizing product development, and optimizing trade-offs between speed, throughput, and cost within digital engineering. This innovation includes simulation-native agents that automate processes such as input validation and failure diagnosis, and an AI physics operating system that transforms simulation data into surrogate models, enhancing cost-efficiency and design evaluation. Rescale's collaboration with U.S. national laboratories is set to leverage agentic AI for manufacturers, and their presence at various international events highlights the integration of AI into engineering workflows without added complexity. Additionally, updates to Rescale's platform and partnerships are expanding access to high-performance computing architectures and the latest simulation software, supporting the growing demand for AI-driven engineering solutions.

GeoTransolver is a geometry-aware transformer architecture that integrates into the Rescale AI Physics platform as part of the NVIDIA PhysicsNeMo library, enhancing the development of high-accuracy AI surrogate models for Computational Fluid Dynamics (CFD) and Finite Element Analysis (FEA). Utilizing GALE attention, it effectively models relationships across intricate 3D geometries and unstructured meshes, allowing it to adapt to varying operating conditions and comprehend transient, geometry-aware physical behaviors. This makes GeoTransolver particularly valuable for predicting aerodynamic, structural, or deformation behavior across different design variants, thereby offering significant advancements in surrogate model development within the realm of AI-driven physics applications.