Simultaneously Learning UAS Aerodynamic and Propulsion Performance Models

Identifying aircraft aerodynamic and installed propulsion performance models from flight data is a frequent objective of flight testing. Especially for small UAS these models are frequently identified separately using test conditions designed to simplify analysis and disambiguate potentially confounding correlations between the models. We examine loss function regularization, imposing physical constraints on learning, and the effect of noise on simultaneous identifiability of these two models using simulated datasets for a small UAS. Using a parametric model we identify approaches that result in high quality model reconstruction which are then applied to learning nonparametric models of the aircraft.