- EPFL researchers developed DeepGeo, an AI tool that optimises aerodynamic shape, reducing time and effort in aircraft design.
- The model eliminates the need for manual tweaking and complex datasets, making ASO more accessible and efficient.
- DeepGeo has potential applications beyond aviation, including in the automotive and energy industries, for optimising complex designs.
Researchers at the École Polytechnique Fédérale de Lausanne (EPFL) introduced DeepGeo, a neural network tool that automates aerodynamic shape optimisation (ASO). Led by Professor Pascal Fua, head of the Computer Vision Laboratory (CVLab), the team aims to revolutionise aircraft design and drive green aviation efforts.
Traditional ASO techniques like Free Form Deformations (FFD) demand extensive manual work and trial and error. DeepGeo eliminates this by using deep learning to generate the parameters needed for optimising 3D shapes, significantly reducing time and effort. “DeepGeo handles the same task as FFD but without manual tweaking,” said Fua. “It speeds up design optimisation dramatically.”
The model also adapts volumetric meshes representing the object’s computational domain, automating mesh deformation and further lightening designers’ workloads. DeepGeo’s innovations earned it the *Best Student Paper Award* at the 2024 American Institute of Aeronautics and Astronautics Forum.
Zhen Wei, a doctoral assistant at CVLab, led the research and highlighted the model’s effectiveness in multiple case studies, including 2D airfoil and 3D Blended-Wing-Body optimisations. DeepGeo delivers the same results as FFD with far less effort. “By eliminating the need for large datasets and complex tuning, DeepGeo makes ASO more accessible and affordable,” said Wei.
Fua, a passionate glider pilot, recently flew a motorised glider from Chambéry, France, to Ouarzazate, Morocco. The trip spanned 5,000 kilometres and used minimal fuel. His eco-conscious flying experience fuels his research. “The key to sustainable aviation lies in reducing drag by modifying aircraft shapes,” Fua explained. “It’s an age-old problem, but new tools like DeepGeo can make a big difference.”
The team plans to apply DeepGeo to model gliders used in Fédération Aéronautique Internationale (FAI) competitions. These gliders must meet many design constraints, offering a perfect testbed for the tool.
DeepGeo’s potential reaches beyond aviation. The researchers foresee its application in industries like automotive and energy, where interacting parts such as car components or turbine elements require complex shape optimisations. “DeepGeo solves tough optimisation challenges across various industries, not just aeronautics,” Fua added.
By automating complex design processes, DeepGeo helps industries create energy-efficient machines more efficiently and at lower costs. As industries face growing pressure to adopt sustainable practices, tools like DeepGeo play a critical role.
“DeepGeo provides a solution for automating energy-efficient system designs, which is essential as we aim to reduce our environmental impact,” Fua concluded.
This innovation marks a significant step forward in developing tools that streamline the design of sustainable technologies with wide-ranging applications across sectors.