Adaptive morphing control method for variable-sweep aircraft based on deep reinforcement learning

This study proposes an adaptive morphing framework that leverages Deep Reinforcement Learning (DRL) to achieve autonomous sweep-angle control through active flow-field perception. To overcome the prohibitive computational cost of high-fidelity Computational Fluid Dynamics (CFD)-in-the-loop training, a deep neural network-based aerodynamic surrogate model is developed. This surrogate model accurately maps flight envelopes to aerodynamic coefficients and surface pressure distributions, enabling the DRL agent to derive optimal control policies within a physics-aware training environment. The proposed framework is rigorously validated via a closed-loop DRL-CFD co-simulation using high-fidelity unsteady Reynolds-Averaged Navier-Stokes (URANS) solver. The results demonstrate that the agent can accurately respond to transient aerodynamic loads and maintain optimal performance by dynamically adjusting the wing geometry. This research highlights the potential of combining physics-informed intelligence with traditional CFD for the real-time control of morphing structures in complex fluid environments, providing a reliable methodology for enhancing the aerodynamic efficiency of future morphing aircraft.