'Evaluation of Kinematic, Dynamic, and Machine Learning Wind Estimation Methods Using CopterSonde UAS'

The CopterSonde, a drone developed by meteorologists and engineers at the University of Oklahoma, is designed and optimized for kinematic and thermodynamic measurements of the planetary boundary layer. The CopterSonde has been paired with mathematical models to deduce aerodynamic drag, which is then translated into estimated wind speeds. The main objective of this work is to continue testing other estimation methods in the search for the most suitable for this particular UAS. The calibration and validation dataset is a combination of Doppler wind lidars, an Oklahoma Mesonet tower, and radiosondes. Similar to a wind tunnel, these reference measurements allowed us to calculate aerodynamic drag and lift coefficients as described in the study by Wildmann et al. (2022), but with the implementation of a Differential Evolution (DE) technique. DE is an optimization technique specifically developed to minimize the root mean square error (RMSE) between the reference measurements and the estimation produced by the CopterSonde by testing the drag coefficients until meeting a defined criteria. Initial findings indicate that the DE approach combined with a Linear Extended State Observer (LESO) model of the CopterSonde yields coefficients that are reasonably accurate and highly consistent with the ideal Rayleigh equation. Additionally, LESO appears to be distinguishing wind effects and removing perturbations caused by autopilot commands more effectively. Moreover, the DE method has enabled refinement of the 3D wind estimation technique using the CopterSonde. However, given that the LESO is a simplified mathematical model based on certain assumptions, there remains room for improvement. Therefore, machine learning techniques have been implemented to enhance the accuracy and precision of 3D wind estimations. The new approach presented in this work is machine learning (ML) algorithms, which can learn intricate patterns in wind dynamics that conventional models may fail to capture. Through a combination of different ML techniques, these models can potentially capture the highly non-linear dynamics of the CopterSonde UAS and yield a better estimation of the wind flow around the platform. This work shows a thorough comparison between all the presented methods, providing guidelines for their implementation.