Flexible Canopy Deformation and Aeroelastic Interaction in Parafoil Systems

ABSTRACT

Parafoils are widely used in modern aerospace systems due to their high lift-to-drag ratio and controllability. However, unlike rigid wings, a parafoil is constructed from fabric and consequently undergoes significant flexible deformation during inflation, gliding flight, turning manoeuvres, and the bird-landing process. These deformations introduce substantial modifications to its aerodynamic behaviour. This study integrates wind-tunnel experiments with numerical simulations to investigate the influence of flexible canopy deformation on aerodynamic performance. A comparative analysis of rigid, semi-rigid, and fully flexible parafoils reveals that canopy bulging leads to a clear degradation in aerodynamic performance, while the primary parameters governing flexible deformation further affect the overall aerodynamic characteristics of the system. These findings underscore the necessity of accurately accounting for flexible deformation effects when predicting parafoil performance or designing next-generation flexible aerodynamic systems.

Physics-Informed Neural Networks with Sparse Identification of Nonlinear Dynamics for Quadcopter Force Dynamics

ABSTRACT

Accurate modeling of quadcopter force dynamics is essential for robust autonomous flight control, yet conventional modeling approaches struggle to balance the high predictive accuracy of data-driven methods with the need for physical interpretability. This work introduces a hybrid Physics-Informed Neural Network (PINN) framework integrated with Sparse Identification of Nonlinear Dynamics (SINDy) to learn quadcopter force and moment dynamics directly from experimental flight data. The neural network leverages automatic differentiation to compute temporal derivatives of the predicted forces, which are then used by SINDy to extract sparse, interpretable governing equations. These discovered equations are incorporated back into the training process as physics-based regularization, creating a bidirectional coupling in which the network benefits from data-driven learning while progressively aligning with transparent, physically meaningful dynamics. This approach aims to transform a black-box deep learning model into a more interpretable framework grounded in identifiable physical laws.