The GARTEUR AD/AG60 project focused on the application of machine learning (ML) and data-driven approaches to aerodynamic analysis and uncertainty quantification, particularly for the Airbus XRF1 aircraft configuration. The research explored various ML techniques, including deep learning, surrogate modeling, and multi-fidelity data fusion, to enhance the prediction of aerodynamic behaviors in industrial applications. The study aimed to integrate heterogeneous data sources, such as Computational Fluid Dynamics (CFD) simulations, wind tunnel tests, and flight data, to improve model accuracy and efficiency. Key areas of research included aerodynamic feature prediction, data fusion, and uncertainty management, with contributions from multiple European institutions. Techniques like Proper Orthogonal Decomposition (POD), Isomap, autoencoders, and graph neural networks (GNNs) were employed to optimize computational performance and enhance prediction reliability.
The project yielded promising results, showing that ML-based models can provide accurate and fast aerodynamic predictions, particularly in capturing complex flow phenomena such as shock waves and boundary layer separation. The study demonstrated that ML techniques outperform traditional reduced-order models (ROMs) in predicting pressure distributions on aircraft surfaces, especially when using graph neural networks (GNNs) and physics-based regularization. Additionally, the report highlighted the importance of multi-fidelity data fusion to improve prediction robustness, addressing challenges related to data scarcity and variability in industrial applications. These findings underscore the potential of machine learning-driven aerodynamic modelling to reduce computational costs while maintaining high accuracy, paving the way for future advancements in aerospace engineering.
Some of these advances have been disseminated in several conferences and journals:
D. Quagliarella, ”An Aerodynamic Shape Design Optimization Procedure Based on Machine Learning”, EUROGEN 2023, 15th International Conference on Evolutionary and Deterministic Methods for Design, Optimization and Control.
Hines Chaves, Derrick Armando and Bekemeyer, Philipp (2022) ”Data-driven reduced order modeling for aerodynamic flow predictions”. Eccomas Congress 2022, 5-9. Juni, 2022, Oslo, Norwegen. doi: 10.23967/eccomas.2022.077. https://elib.dlr.de/189313/
Hines, D., and Bekemeyer, P. (2023). ”Graph neural networks for the prediction of aircraft surface pressure distributions”. Aerospace Science and Technology, 137, 108268. https://www.sciencedirect.com/science/article/pii/S1270963823001657
Hines Chaves, D. A., Dias Ribeiro, M., and Bekemeyer, P. (2023). ”Physics-based Regularization of Neural Networks for Aerodynamic Flow Prediction”. In EUROGEN 2023 15th ECCOMAS Thematic Conference on Evolutionary and Deterministic Methods for Design, Optimization and Control (pp. 22-39). https://doi.org/10.7712/140123.10188.18859
- Peter, and M. Lippi, ”Comparison of data-driven methods for the prediction of flows about a wing”, Abstract submitted to ECCOMAS 2024.
- Gorgues, R. Castellanos, J. Nieto-Centenero and E. Andr´es, ”Application of a deep neural network for Cp prediction on multiple wing geometries in a transonic regime”, Aerospace Europe Conference 2023 – 10th EUCASS – 9th CEAS, Lausanne, Switzerland, July 2023.
- Castellanos, J. Nieto-Centenero, A. Gorgues, S. Discetti, A. Ianiro, and E. Andres, ”Towards Aerodynamic Shape Optimisation by Manifold Learning and Neural Networks”, 15th ECCOMAS Thematic Conference on Evolutionary and Deterministic Methods for Design, Optimization and Control (Eurogen), Chania, Crete, Greece, June 2023.
- Nieto-Centenero, R. Castellanos, A. Gorgues and E. Andres, ”Fusing Aerodynamic Data using Multi-Fidelity Gaussian Process Regression”, 15th ECCOMAS Thematic Conference on Evolutionary and Deterministic Methods for Design, Optimization and Control (Eurogen), Chania, Crete, Greece, June 2023.
- Castellanos, J. Bowen-Varela, A. Gorgues and E. Andres, ”An assessment of reduced-order and machine learning models for steady transonic flow prediction on wings”, 33rd congress of the International Council of the Aeronautical Sciences (ICAS), Stockholm, Sweden, September 2022.
- Gorgues, R. Castellanos, J. Bowen-Varela and E. Andres, ”Tree-based comparative prediction of steady transonic flows over a wing”, 8th European Congress on Computational Methods in Applied Sciences and Engineering (ECCOMAS), Oslo, Norway, June 2022.
- Bowen-Varela, R. Castellanos, E. Farzamnik, A. Gorgues, A. Ianiro, and E.Andres, ”Nonlinear interpolation of steady transonic flows via manifold learning and neural networks”, 1st Spanish Fluid Mechanics Conference, Cadiz, Spain, June, 2022.
