Neural Network–Driven Reentry Reference Trajectory Design and Tracking Using Sliding Mode Control

This paper introduces a novel method for designing the optimal reentry trajectory for the hypersonic winged reusable launch vehicle, X-33. The proposed method utilizes a cascaded neural network structure to determine crucial parameters for generating the reference trajectory during reentry. A significant quantity of reference entry trajectory data is generated offline to support this approach, leveraging a linear-logarithmic-linear (LLL) polynomial–based formulation and improved search space reduction (ISSR) optimization to derive the essential parameters. The generated data are then used to train the cascaded artificial neural network. Once trained, the proposed neural network accurately provides the required parameters onboard, eliminating the need for iterative optimization. The reference reentry trajectory generated by this method remains entirely within the reentry corridor, adhering to all hard and terminal constraints. An analytical expression for the reference bank angle is employed, and a sliding mode control (SMC) is developed to adjust the reference bank angle further, ensuring that the trajectory meets all path and terminal constraints even in the presence of disturbance. The effectiveness of the proposed method is validated through nominal case and Monte Carlo simulations, which account for variations in initial reentry conditions. Additionally, a real-time simulator demonstrates the method’s effectiveness.