Feedback control of an aeropendulum based on a data-driven dynamic model

Abstract

This work aims to refine the aeropendulum model by improving the thrust force description and data acquisition, looking for a better closed-loop response. We propose a low-level acquisition scheme that captures previously neglected electrical quantities using an Arduino Mega. Using the parameters estimated, we develop a simulation model and apply stateof- the-art data-driven identification algorithms to refine the aerodynamic behavior of the plant. We also assess different control approaches with gradual levels of refinement regarding the actuator. Concerning the model refinement, the predicted dynamics for simulated data using the sparse identification algorithm exhibited an MSE (Mean Squared Error) on the order of 10 × 10−10 regarding the dynamics compared to the test set in simulated examples. The system performance in closed-loop indicates that the feedback linearization strategy is the most suitable approach due to the almost constant thrust generation and linear behavior.