Digital-Twin of a Quadrotor Biplane Tailsitter UAV

This paper presents the development and validation of a Digital Twin of a Quadrotor Biplane Tailsitter (QBiT) UAV to accurately capture the vehicle dynamics across multiple flight regimes. The Digital Twin integrates a baseline physics-based model with a neural network trained to predict the residual dynamics not captured by the physics model. The physics-based model consists of blade element momentum theory (BEMT) for the rotor model and lifting-line theory for the wing model. These models were validated with wind tunnel experimental data. A 12-lb QBiT UAV was developed to collect outdoor flight test data for hover, transition, and forward flight regimes, which included coordinated turns. While the physics-based model shows good qualitative agreement with experimental data, certain deviations suggest the presence of modeling inaccuracies and system parameter estimation errors. Subsequently, a neural network was trained using the flight test data to learn and generalize the corrections required to enhance the accuracy of the physics-based model. Compared to the standalone physics-based model, the resulting hybrid model (Digital Twin: physics + neural network) had significantly greater agreement with the flight test data based on the root-mean-squared error (RMSE) for the linear and angular accelerations. This result highlights the ability of the Digital Twin to compensate for unmodeled aerodynamic effects, mass property uncertainties, fabrication imperfections, and other differences between the model and the real-world aircraft. The validated Digital Twin enables accurate predictions, offering a reliable foundation for novel controller development, performance optimization, and simulation-based testing. This work demonstrates the practicality and effectiveness of fusing physics-based models with machine learning for unconventional UAVs like the QBiT.