← Technology Evidence engine

AeroFlow

AI aerodynamic assessment from 2D views or sketches, before CAD.

What it solves

It compares concept proposals in seconds and shows which ones deserve a 3D model and CFD. It does not replace high-fidelity simulation: it directs it.

Today it works on the type of vehicle it has been trained on. We adapt it to your range with your simulations.

From sketch to simulation

From a sketch to the flow field, in seconds.

Fig. 1

The flow field is a 2D reference simulation, not an AeroFlow output.

Illustrative
How the flow was computed

Lattice Boltzmann method with a Smagorinsky turbulence model and a moving ground, at a Reynolds number of about 20,000. It is a 2D slice: a real car at 100 km/h is around 10⁷ and its flow is three-dimensional. It shows the shape of the flow; it does not give the car’s drag.

Stage of development

TRL 5 · Open to pilots

Area

Automotive and transport

Line

Physics & aerodynamics

Evidence and limits

What we know, and what we still need to show.

0.00456

Cd mean absolute error

About 4.6 drag counts. Internal evaluation on automotive reference cases.

≤ 2 s

prediction per view

Plus view preparation, 20 to 90 seconds per view.

40.8 s

from 3D geometry to drag

Current 24 GB GPU endpoint.

Evaluated to date
  • Cd prediction on reference cases
  • Image-based inputs
  • Any 3D geometry
Evidence still needed
  • Close design variants
  • New design domains
  • End-to-end error from a hand sketch
Next
  • Evaluation in each customer's design domain, against their own reference evidence
Before building the 3D

The shape is chosen before the aerodynamic evidence.

Conventional aerodynamic analysis needs a developed 3D model. AeroFlow answers while the concept is still a sketch.

See the typical case: a hundred sketches, ten selected, five 3D models →

Compress a flow simulation up to 30,000 times and keep what happens next to the surface.

In the paper, on a bridge deck. In AeroFlow we apply it to cars.

NeVF · Computer-Aided Civil and Infrastructure Engineering, 2026 · See the publication

Flow visualisation around the profile: more points where the flow changes and near the surface.

Innovation
  • Aerodynamic assessment before CAD, from 2D views or sketches
  • Compares concept proposals in seconds
  • Shows which proposals deserve a 3D model and CFD
  • Adapts to each range with the customer's simulations
How to collaborate
  • Joint pilot
  • Fine-tuning of the model for internal use

fluid.deepdesignsystems.com

Fields of application
  • Conceptual vehicle design
  • Selecting proposals before 3D modelling
  • Prioritising CFD campaigns

Interested in this technology?

Write to us at weare@deepdesignsystems.com

We bring intelligence to the start of the process.

For product teams, industrial partners and institutions.

weare@deepdesignsystems.com