Predicting aerodynamics forces from sensor data
To advance state-of-the-art Machine Learning capabilities when creating precise surrogate models of transient nonlinear physical phenomena related to aerodynamics with known (quantified probability density function describing it) and traceable uncertainty. Create the capability to reconstruct such complex physical transient phenomenon with a minimum amount of discrete real sensorial input of finite and calibrated precision.
PI / Partners
Geodetic Engineering Lab (EPFL)
Unsteady Flow Diagnostics Lab (EPFL)
Chair of Computational Mathematics and Simulation Science (EPFL)
Federal Institute of Metrology (METAS)
Nowadays, autonomous navigation of aircraft vehicles relies mostly on systems like GPS providing a signal to determine the vehicle’s position. To avoid an uncontrolled flight in the case of signal interruption, robust fallback algorithms are required. Data-driven strategies should allow for accurate navigation by using information that remains available, such as sensor measurements on board.
Despite the notable advances in artificial intelligence, recently developed algorithms frequently fail to fulfil robustness requirements to employ data-driven systems in the real world.
Therefore, to mindfully tackle the question of autonomous navigation, in the SenseDynamics project, the SDSC
supports a highly multidisciplinary research collaboration established between three traditionally unrelated laboratories at EPFL (TOPO – topography and navigation, UNFOLD – unsteady experimental aerodynamics, MCSS – computational mathematics and simulation science), and the Swiss Federal Bureau of Meteorology and Standardisation METAS. Each of these groups is essential in the effort towards unmanned aerial vehicles that would be also GPS-free, navigating through air only by a machine learning model which uses data from a limited number of sensors.
In achieving these objectives, the SenseDynamics project uses high-fidelity data recorded in an experiment from diverse movements of an airfoil of a wing in a wind tunnel, provided by the UNFOLD group. Additional simulated data is provided by the MCSS, which is cheaper, less accurate, but indeed matches closely the expected physical behaviour under diverse unseen setups. The SDSC takes on the challenge to combine these multi-fidelity data sources, and design a machine learning model able to predict the necessary forces. At the same time, the model should fulfil the desired reliability and robustness criteria, which are finally being validated by a pipeline set up in METAS.
In a broader perspective, the SenseDynamics project should demonstrate how machine learning and data science – guided by domain expertise – can provide effective and practical solutions for real world challenges.
Figure 1. An illustration of a stall development over an airfoil, as recorded in the experimental data. In the first row, each panel A – E represents one of five different snapshots of the airflow. Every state is characterized by a significantly different behaviour from the others (some exhibiting more turbulence as C and D, compared to A and B) which makes the prediction task challenging. Namely, as depicted in the second row, the machine learning model should be able to accurately predict a volatile output (compare grey and dashed red line), along with a confidence envelope around it (red band) which will faithfully capture the variability in the data.
Donner Christian, Tagasovska Natasa, He Guosheng, Mulleners Karen Shimazaki Hideaki, Obozinski Guillaume “Learning interpretable latent dynamics for a 2D airfoil system“, presented as a RobustML workshop paper at ICLR 2021
Deparday, J., Mulleners, K. (2019). Modeling the interplay between the shear layer and leading edge suction during dynamic stall Physics of Fluids 31(10), 107104. https://dx.doi.org/10.1063/1.5121312
Mulleners, K., Raffel, M. (2013). Dynamic stall development Experiments in Fluids 54(2), 1469 1477. https://dx.doi.org/10.1007/s00348-013-1469-7
Mulleners, K., Raffel, M. (2012). The onset of dynamic stall revisited Experiments in Fluids 52(3), 779 793. https://dx.doi.org/10.1007/s00348-011-1118-y
Leishman, J. (2002). Challenges in modelling the unsteady aerodynamics of wind turbines Wind Energy 5(2-3), 85 132. https://dx.doi.org/10.1002/we.62
Alsalman, M., Colvert, B., Kanso, E. (2018). Training bioinspired sensors to classify flows. Bioinspiration & biomimetics 14(1), 016009. https://dx.doi.org/10.1088/1748-3190/aaef1d
Colvert, B., Alsalman, M., Kanso, E. (2017). Classifying vortex wakes using neural networks arXiv 13(2), 025003. https://dx.doi.org/10.1088/1748-3190/aaa787
Tuna, B., Kurelek, J., Yarusevych, S. (2019). Surface-Pressure-Based Estimation of the Velocity Field in a Separation Bubble AIAA Journal 57(9), 3825-3837. https://dx.doi.org/10.2514/1.j0580