NeuralWing

Real-time neural simulation and design optimization of transonic aircraft aerodynamics

github

huggingface

paper

NeuralWing Demo

Modify wing geometry and test optimizations to maximize aerodynamic efficiency

The largest 3D transonic wing dataset with real-time neural simulation

We created a dataset of 30,000 steady-state CFD simulations of a 3D wing in the transonic regime, varying four geometry parameters and two inflow conditions. Utilizing our AB-UPT surrogate model trained on this data, the demo shows how design parameters can be optimized in seconds to maximize KPIs such as the lift-to-drag ratio.

Dataset size

30,000 simulations

Dataset variability

4 geometry parameters, speed, angle of attack

Simulation accuracy

AB-UPT achieves 99.5% accuracy on pressure fields

Simulation & Optimization speed

Real-time prediction & design parameter optimization

1000x

Simulation speedup

99.8%

Drag & Lift Data Agreement

30 seconds

Design Parameter optimization

Architecture & Performance

Model Architecture

Type

AB-UPT

Input

Geometry mesh (STL), speed, angle of attack

Output

fields: pressure, friction, velocity
integral forces: lift, drag

Training Data

26,000 CFD simulations

Geometry generation for inference

Geometry mesh (STL) is created in real-time from 4 design parameters in a differentiable manner

Performance Metrics

Numerical CFD Simulation (RANS)

4 CPU hours
(not optimized)

AB-UPT

100 ms for surface fields,
1 sec for volume fields

Design-parameter Optimization

50 ms per gradient-free step,
150 ms per gradient step

Pressure coefficient comparison

GFD Ground Truth

AB-UPT

Prediction Error

Friction profile

Pressure profile

The NeuralWing AI model enables real-time simulation and design optimization

Ready to accelerate your development?

Explore how AI powered models are transforming physics simulations, from thermodynamics to fluid dynamics, and learn how to get started building or integrating your own models.