Modeling Unsteady Flows for Aeroelasticity Through Machine Learning and Data-Driven Methods

Aeroelasticity studies the interaction of a flexible structure immersed in unsteady fluid flows. The aeroelastic analysis involves coupling the structure model with the aerodynamic model to evaluate the instability and response of the coupled dynamic system. However, simulating unsteady flows for aeroelasticity is computationally expensive and limits the engineering application of modern aeroelastic analysis tools. The emergence of data-driven methods has given rise to a new research paradigm in fluid mechanics, particularly in the modeling of unsteady flows. Through developing data-driven methods from different sources of data, data-driven aerodynamic models maintain reasonable accuracy while largely increasing the efficiency of aeroelastic analysis. This short review summarizes some of the techniques to model unsteady aerodynamics for the analysis and prediction of various aeroelastic problems, mainly based on the recent effort from the authors. These techniques include system identification for integrated aerodynamic loads, feature extraction for field and distributed variables, and data fusion, i.e., multifidelity modeling and data assimilation, for flow data from multiple sources.