Large space structures (LSS) should be treated as flexible bodies due to the significant impact of their elastic deformation on attitude maneuvers. However, accurately modeling the entire infinite-dimensional system during ground tests under gravity is inherently challenging. This paper presents a model-independent framework for LSS attitude control that tunes active disturbance rejection control (ADRC) via Bayesian optimization. The tuning parameters are selected automatically by minimizing a cost function that balances tracking error, control effort, and its derivative, thereby eliminating the need for manual trial and error. The proposed approach is validated in simulation on a flexible-structure satellite model with several vibration modes. The Bayesian optimization process converges quickly, yielding a controller that achieves rapid settling without overshoot, smooth slewing, and effective spillover suppression. In comparison, data-driven single- and two-degree-of-freedom PID controllers, obtained through virtual reference feedback tuning (VRFT), achieve nearly the same response but exhibit slight residual oscillations. These results demonstrate that Bayesian-optimized ADRC is a practical and high-performance alternative for the attitude control of LSS.
Tawaraya et al. (Thu,) studied this question.