Modeling UAVs using CFD and Machine Learning Methods

To design the control systems needed for aerial vehicles various low order aerodynamic and dynamics models are developed to capture the dominate handling characteristics of the vehicle. These models however can generally take on the form of simplified look-up tables which provide static aerodynamic load data and in more advanced cases loads due to oscillation in a particular degree of freedom. Therefore, these methods do not capture much of the complex nonlinear behaviors that occur outside of static and quasi-steady behavior. In this work a nonlinear dynamic model of a UAV will be developed using controlled flight responses from select unsteady CFD simulations. Machine learning techniques are employed to capture discrepancies and compare against a simple linear model and the CFD results. Overall, this work will demonstrate a method to create fast medium-fidelity reduced order dynamic models of aerial vehicles with complex geometries using small sets of costly high-fidelity CFD models suitable for control system design.