物理信息神经网络及其在空天飞行器中的应用展望

Aerospace vehicles have been recognized as a strategic focus globally due to their unique advantages of low cost and reusability. However, the dynamics involving complex multi-physics effects and wide flight conditions pose severe challenges to modeling and control design. Physics-Informed Neural Networks (PINNs), emerging as a method fusing data-driven approaches with physical constraints, provide a new pathway to address these challenges. The PINN method and its application prospects in aerospace research are systematically reviewed. The basic principles and frameworks are expounded, and the mainstream improved algorithms along with research progress are analyzed. Furthermore, the application potential and implementation paths in key sectors, including multi-physics modeling, aerodynamic identification, control system design, and fault diagnosis, are reviewed and prospected. The study provides valuable theoretical and engineering references for the development of intelligent technologies for aerospace vehicles.