ORCID

0000-0002-4377-8803

Date of Award

2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical Engineering

First Advisor

Luis Duffaut Espinosa

Abstract

Model-based control methods are seen extensively in applications due to their intuitive design and ability to provide performance guarantees. A significant drawbackin these methods comes from the difficulty of accurately capturing non-linear effects and identifying system parameters. These drawbacks have led to the advancement of learning-based controllers, which typically require heavy computational power and an extensive training process creating different problems from model-based solutions. Model-informed control is a blend of these control methodologies to provide performance guarantees but without the crux of parameter identification and gain tuning in non-linear systems while staying computationally non-complex. This thesis employs the Model-Free Control (MFC) methodology which utilizes a surrogate model for controller design via aggregating the unknown system dynamics and perturbations into a single parameter for estimation. Transforming the problem from a non-linear control problem to an estimation problem eliminates the concern for model mismatch from imprecise model identification and modeling external disturbances of highly complex non-linear systems. MFC has shown promising results for a wide variety of systems in the past decade, however, most of these results have been conducted through simulation only. A crucial drawback of the MFC methodology is the need for a high sampling period to accurately estimate the system dynamics and external disturbances. This sampling period will lead to noise amplification in the controller, requiring the choice of a sampling period to trade off avoiding noise amplification for imprecise estimation. The work in this thesis presents the hardware validation of MFC on an embedded quadcopter for position control. This implementation focuses on a newly developed Kalman Filter approach for the estimation of the system dynamics and external disturbances in real-time. The Kalman Filter approach addresses the high sampling period concern, ensuring accurate estimation and avoiding noise amplification, with control loop execution times comparable to the model-based control methods analyzed in this thesis. The adaptability of the MFC methodology is showcased through two scenarios: 1) Performance of a flight trajectory with a 33% increase in mass pushing the motors near saturation and 2) Flight performance during external wind disturbances covering parts of the flight path. The final part of this work investigates the implementation of a method for addressing a critical deficiency of MFC, algorithmically choosing a tuning parameter for the control effort that is usually selected heuristically. Estimating this parameter alongside the system dynamics and perturbations minimizes the control effort generated and mitigates oscillations in the system output.

Language

en

Number of Pages

93 p.

Download

Share

COinS