Control of grid rudder with transmission clearance based on active learning surrogate model

To address the recurrent necessity for clearance compensation within the grid fin actuator drive train, angular displacement sensors are strategically installed at multiple locations throughout the servo system. A dedicated load simulation test system for the grid fin is established, wherein the angular position of the grid fin shaft and the extension signal of the servo are acquired via sensors mounted on the drive mechanism. Utilizing the collected experimental data, an optimized surrogate model of the servo drive train clearance is developed based on a Kriging approach augmented with active learning. This surrogate model is subsequently integrated into the closed-loop control system of the grid fin to facilitate precise compensation for drive mechanism clearance. Simulation outcomes demonstrate that the Kriging model enhanced with active learning can achieve high-accuracy angular deviation predictions using a limited number of training samples. Incorporation of this predictive model into the grid fin control architecture significantly reduces angular errors between the actual position of the grid fin shaft and the internal feedback provided by the servo system. The methodology enables accurate estimation of the drive train clearance and enhances the overall control precision of the grid fin mechanism.