The challenge of cycle-to-cycle variability in dynamic stall modelling

This research explores cycle-to-cycle variability in dynamic stall through wind tunnel experiments conducted on a pitching NACA 0018 aerofoil at a Reynolds number of 2.8×105. Multiple cycles are considered, and different clusters are identified based on inspection of the lift time series. Experiments reveal that the blind application of a conventional phase-averaging approach can produce inadequate results, which do not represent the underlying physics; instead, it is recommended to analyse each cycle individually and use a clustering approach. The available wind tunnel measurements are employed to build two distinct aerodynamic models, i.e. a semi-empirical Goman-Khrabrov dynamic stall model and a purely data-driven model based on artificial neural networks. The work highlights that cycle-to-cycle variability in dynamic stall represents a huge challenge from a modelling perspective. The Goman-Khrabrov model cannot capture the bifurcations in the data, while the more sophisticated data-driven model is accurate but prone to instability. The paper proposes to enhance the accuracy of the models by dynamically assimilating experimental measurements using an Extended Kalman Filter. Results demonstrate that this methodology represents a valuable and versatile tool, which allows to effectively combine imperfect model predictions with experimental observations.