Abstract
This study presents an advanced approach to modeling Unmanned Aerial Vehicle (UAV) dynamics by integrating Artificial Neural Networks (ANNs) and the Output Error Method (OEM). Moreover, the research analyzes the longitudinal flight characteristics of the Multiplex FunCub R/C, using data from designed flight tests complemented by simulations incorporating sensor noise and drift effects. Furthermore, the study captures a comprehensive range of aerodynamic responses essential for precise system identification by employing a multistep elevator input signal. In addition, the OEM approach, a traditional parameter estimation method, offers robust statistical estimation by minimizing the discrepancies between measured and predicted outputs. However, due to the nonlinear complexities inherent in UAV flight dynamics, the study also explores ANNs, leveraging their capability to model intricate nonlinear behaviors without requiring predefined aerodynamic parameters. Subsequently, the performance of both methodologies is critically evaluated against measured aerodynamic coefficients, revealing ANNs’ superior adaptability in accurately predicting complex aerodynamic interactions compared to OEM. Consequently, results indicate notable reductions in relative error, particularly in challenging aerodynamic coefficients. Overall, this research not only highlights the comparative advantages of ANNs in UAV system identification but also lays an initial structure for future advancements in UAV modeling and control systems.