AI-powered design comes to high-speed racing

2026-04-30 · Source: IBM Research · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Engineering & Applied Sciences · Depth: Expert, short

Summary

IBM Research and the Dallara Group have partnered to accelerate high-performance race car aerodynamic design using AI surrogate models. Their collaboration developed the Gauge-Invariant Spectral Transformer (GIST), a graph-based neural operator, to significantly speed up computational fluid dynamics (CFD) evaluations. In one experiment, GIST identified an optimal rear diffuser design for a Le Mans Prototype 2 (LMP2)-like car in approximately 10 seconds, a task that took traditional CFD several hours. Dallara estimates this AI surrogate could reduce simulation time for hundreds of geometry configurations from days to minutes. The GIST model, presented at ICLR 2026's AI & PDE Workshop, processes graph-structured simulation data based on partial differential equations, encoding mesh points and links to capture topology more accurately than previous models. Researchers also used random projections and a gauge-invariant architecture to ensure scalability and generalization across different mesh densities. The partnership also plans to explore quantum algorithms for future simulation fidelity improvements.

Key takeaway

For Machine Learning Engineers developing physics-based simulation tools, you should investigate graph-based neural operators like GIST to dramatically reduce computation time for complex aerodynamic design. Your team can achieve significant speedups, cutting evaluation from days to minutes, by training these AI surrogates on existing CFD data. Consider how encoding mesh topology and ensuring gauge-invariance can improve your model's accuracy and generalization for intricate designs.

Key insights

AI surrogate models, like GIST, can drastically accelerate complex physics simulations while maintaining high accuracy.

Principles

• Encoding mesh topology improves AI model accuracy for intricate designs.
• Gauge-invariant architectures enhance model generalization across data representations.
• Random projections can scale graph transformers for complex relationships.

Method

Train a graph-based neural operator (GIST) on physics-based CFD simulation data, encoding mesh points and links, then use it to rapidly evaluate aerodynamic designs.

In practice

• Apply GIST to optimize race car components like rear diffusers.
• Extend AI surrogate models to passenger cars and commercial planes.
• Explore quantum algorithms for future simulation fidelity.

Topics

• AI Surrogate Models
• Computational Fluid Dynamics
• Graph Neural Networks
• Aerodynamic Design
• High-Performance Racing
• Quantum Computing

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Ten Years of Trying to Learn from a Mesh That’s Too Large — Large CFD meshes pose representational challenges for surrogate models, requiring specialized architectures to capture localized physical features.
• Non-intrusive Learning of Physics-Informed Spatio-temporal Surrogate for Accelerating Design — PISTM combines physics constraints and Koopman autoencoders for generalizable spatio-temporal surrogate modeling.
• FLAME 2023: Diving into the Future of Fluid Dynamics and Machine Learning — AI techniques are transforming simulation by integrating physics-based modeling with data-driven machine learning.
• Mesh Graph Neural Network Framework for Accelerating Finite Element Simulation for Arbitrary Geometries — Mesh Graph Networks effectively generalize finite element analysis predictions across arbitrary structural geometries by encoding relative features.
• AI Models Trained on Physics Are Changing Engineering — Large physics models drastically accelerate engineering design by replacing or augmenting traditional physics simulations.
• Physics Simulation Just Crossed A Line — A CPU-based physics simulation uses domain decomposition to achieve superior speed and realism for complex deformable objects.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by IBM Research.