Physics-Informed Neural Networks-based Uncertainty Identification and Control for Closed-Loop Attitude Dynamics of Reentry Vehicles

This paper investigates the application of Physics-Informed Neural Network (PINN) technique into the uncertainty identification and control issue for reentry vehicles (RV) attitude dynamics. The PINN methodology proves to obtain the solution of ordinary differential equations through incorporating the physical and data knowledges. The existing articles mainly focus on using the PINN to the inverse problem, which actually deals with the open-loop dynamics and neglects the control design. Different from that, this paper instead introduces the Euler iteration-augmented Physics-Informed Neural Networks (Euler-PINNs) to estimate unknown parameters in the RV closed-loop dynamics, meanwhile, the classical PID is designed to ensure the fine attitude track of desired commands. Furthermore, the framework of the proposed method is described in detail. Then, the sampled simulated time-series data of attitude dynamics with time-varying uncertainties is utilized for the neural network learning, and simulation results show its effectiveness. Also, the influence of control parameters on the PINN algorithm effect is discussed.