Adaptive control of hypersonic vehicles with unknown dynamics based on dual network architecture. (April 2022)

• Record Type: Journal Article
• Authors: Cheng, Lin; Wang, Zhenbo; Gong, Shengping
• Is Part Of: Acta astronautica. Volume 193(2022)
• Page Start: 197
• Page End: 208
• Publication Date: 2022-04
• Subjects: Unknown dynamics -- Extended state observation -- Iterative model learning -- Iterative control learning -- Dual network architecture; Astronautics -- Periodicals; Outer space -- Exploration -- Periodicals; Astronautics; Periodicals; 629.405
• DOI: 10.1016/j.actaastro.2021.12.043
• Languages: English
• ISSNs: 0094-5765

Abstract

Abstract: The difficulty of obtaining accurate dynamical models of hypersonic flight has been generally recognized, and modeling inaccuracy can severely deteriorate the performance of the flight control systems. To address this issue, an adaptive control approach using two neural networks (NNs) is proposed with the aim of achieving precise and robust control for hypersonic flight when unknown dynamics are involved. Different from the existing adaptive control methods, two NNs are developed in this paper to learn the forward and inverse dynamics of hypersonic flight with guaranteed convergence. Particularly, this study focuses on the following three contributions. First, an iterative model learning algorithm is proposed to train the first NN to approximate the unmodeled system dynamics and achieve accurate observations of flight responses and unknown dynamics. Second, an iterative controller learning algorithm is proposed to guide the second NN to learn the control inputs from prior flight data and improve the dynamic performance of the adaptive controller. Third, an adaptive NN-based controller for trajectory tracking is developed combining the above two NNs. Simulations are provided to substantiate the effectiveness of the proposed techniques and demonstrate the excellent adaptability and robustness of the controller. Highlights: An iterative model learning algorithm is proposed to learn the unmodeled system dynamics. An iterative controller learning algorithm is proposed to learn the control inputs from prior data. A dual neural networks architecture is proposed to achieve intelligent adaptive control.