Speed up Aerodynamic Optimization Thanks to Deep Learning Physics: Stellantis RAM Truck Example

NAFEMS World Congress 2023, Tampa, Florida, USA


Optimizing the shape of an object in a given fluid requires the analysis of the physical flow fields around it. To that aim engineers use Computational Fluid Dynamics (CFD), a method that has been proven to work but is highly time consuming. The Deep Learning Physics technology developed by Extrality enables to leverage on some past data to deliver very fast high-fidelity 3D field predictions. This paper presents an example of its application to a RAM truck (Stellantis) whose training is based on a dataset of 20 CFD simulations, showing positive results. The AI-model achieved less than 5% mean relative error in comparison to the CFD on the drag coefficient, while delivering the full surface and volume fields, and keeping the genericity of the predictions. The major benefits lie in the easy-to-use Extrality platform, accessible from anywhere, allowing any engineer to create a Deep Learning Physics model and perform high fidelity simulations without solver knowledge. Hence, it offers a drastic reduction of turn-around time. Less than 5 minutes are needed from the upload of a geometry to the post-processed results. Besides, it is possible to massively reduce the pre-processing time linked to CFD mesh requirements as the model can ingest the native CAD mesh as input while maintaining the aerodynamic coefficient trends. Since the simulation physical knowledge is already embedded in the AI-model, high-fidelity simulations can be done by any stakeholder involved in the project, simplifying the end-to-end optimization workflow. Moreover, the Extrality technology predicts the physical volume and surface flow fields in a format readable by mainstream visualization softwares. Experimentation conducted on the RAM truck example shows that the Extrality technology has its place in today's industry workflow. The results of the AI-model have been compared to the internal DoE tool used by Stellantis and showed that the AI-model achieves, overall, better predictions and is able to correctly identify the car giving the optimum drag coefficient. In another use case, a tonneau bed cover length optimization, the Extrality technology gave the optimal length within 1 day end-to-end, instead of 3 weeks needed to obtain the same information using a classical CFD method. This opens the door to extensive usage of high-fidelity prediction in early design stages and eventually, to shorter design cycles in the industries.


1/ Extrality, 75002 Paris, France

2/ Stellantis, Michigan, United States


/ Other published papers

INFINITY: Neural Field Modeling for Reynolds-Averaged Navier-Stokes Equations
Louis Serrano, Leon Migus, Yuan Yin, Jocelyn Ahmed Mazari, Patrick Gallinari
Learn more
Multi-Objective Hull Form Optimization with CAD Engine-based Deep Learning Physics for 3D Flow Prediction
Jocelyn Ahmed Mazari, Antoine Reverberi, Pierre Yser, Sebastian Sigmund
Learn more
Deep Learning Physics for the Hydrodynamics of Trading Vessels
Jonas Verrière, Jocelyn Ahmed Mazari, Antoine Reverberi, Francis Hueber
Learn more
AirfoilRANS: High Fidelity Computational Fluid Dynamics Dataset for Approximating Reynolds-Averaged-Navier–Stokes Solutions
Florent Bonnet, Jocelyn Ahmed Mazari, Paola Cinnella, Patrick Gallinari
Learn more
An extensible Benchmarking Graph-Mesh dataset for studying Steady-State Incompressible Navier-Stokes Equations
Florent Bonnet, Jocelyn Ahmed Mazari, Thibaut Munzer, Pierre Yser, Patrick Gallinari
Learn more
Multi-scale Physical Representations for Approximating PDE Solutions with Graph Neural Operators
Léon Migus, Yuan Yin, Jocelyn Ahmed Mazari, Patrick Gallinari
Learn more
Need to hit production with short go-to-market and better control?
Cookie Consent

By clicking “Accept”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts.