Pairing Bayesian statistics with transition networks: A data-driven approach for aerodynamic state estimation

Swimmers and flyers in nature utilize haptic sensor feedback to control the interaction between the surrounding fluid and their bodies, even in challenging environments such as wake flows or gusts. Inspired by this behavior, a series of data-driven, bio-inspired works have been proposed that utilize real-time sensor input for aerial vehicle control. To tackle unsteady, highly separated, and high Reynolds number flows and their highly non-linear dynamics with a sparse set of pressure sensors and limited training data, advanced data-driven approaches are required. In the present work, a transition network approach relying on a Markov model is employed. We extend previous transition network-based approaches through a Bayesian estimation process to account for the typically high noise levels in realistic experimental data. The flow around an accelerating elliptical plate is selected as a test case. The plate is accelerated and decelerated at various (fixed) angles of attack, and the aerodynamic loads are estimated from a set of sparse pressure measurements. Results show that the combination of transition networks with Bayes' theorem can lead to good load estimation in real-time when facing the issue of noise in the measured quantities.