Revolutionizing Air Travel: Deep Learning for Greener Flights

For aviation enthusiast Fua, the quest to revolutionize green aviation design extends far beyond the confines of a traditional laboratory. His passion for flying led him to embark on a remarkable journey with motorized gliders, traveling from Chambéry in France to Ourzazate in Morocco, showcasing his dedication to pushing the boundaries of aerodynamic innovation.

Aerodynamic shape optimization (ASO) stands as a pivotal technique in the realm of aerodynamic design, focusing on enhancing the physical performance of objects while adhering to specific constraints. To achieve optimal results in shaping a 3D object, it is essential to represent it using a set of parameters that can be utilized by the optimizer. Typically, Free Form Deformations (FFD) have been employed for this purpose, albeit with significant manual intervention and a trial-and-error approach, even when handled by experts.

Introducing Deep Geometric Mapping

Researchers at the Computer Vision Laboratory (CVLab) within the School of Computer and Communication Sciences (IC), in collaboration with colleagues at ISAE-SupAero in France, have introduced the DeepGeo model. This innovative approach leverages neural network technology to automate the parameterization process for complex geometries, eliminating the need for human intervention. Professor Pascal Fua, the Head of the CV Lab, explains that DeepGeo streamlines the parameterization process, making it more efficient and less time-consuming compared to traditional methods.

DeepGeo not only automates the parameterization process but also adapts the volumetric meshes that model the computational domain surrounding the object being optimized. By dynamically adjusting these volumetric meshes during the optimization process, DeepGeo simplifies the Computational Fluid Dynamics computations, reducing the workload for designers and engineers.

Recognized for its groundbreaking potential, the research paper detailing DeepGeo recently received the Best Student Paper Award at the prestigious American Institute of Aeronautics and Astronautics Forum ‘24. The paper showcases how DeepGeo, based on deep geometric learning techniques, achieves remarkable results without the need for extensive training datasets. Case studies presented in the paper, such as the optimization of 2D circle-to-airfoil, 3D CRM wing, and 3D Blended-Wing-Body aircraft, demonstrate the effectiveness and efficiency of DeepGeo compared to traditional methods like FFD.