Design and sensitivity analysis of the targeted-docking model for autonomous aerial refueling based on convolutional neural networks

As the preferred method for unmanned autonomous aerial refueling, soft aerial refueling technology has drawn the attention of scholars worldwide. Existing research mainly focuses on solving the problems related to the hose-drogue assembly or on vision-based drogue positioning and tracking. However, there is currently no precise and efficient docking model or strategy that can effectively guide the docking path of the receiver. Based on earlier research findings, this paper proposes a targeted-docking approach. It constructs a dataset using the bow wave model and the Kriging interpolation method. Leveraging the advantage of convolutional neural networks in extracting deep features from data, a targeted-docking model is established for the first time. Traditional computational fluid dynamics (CFD) simulations require extensive computation time about 2 days to solve the detailed flow field characteristics, but the targeted-docking model realized the prediction of the successful docking range under any steady-flow refueling condition within 10–15 s, meeting the key requirements of the aerial refueling mission. Subsequently, by comparing with the original data obtained from CFD simulations, the accuracy and practicality of the model are verified. Finally, the Sobol method is adopted to conduct a sensitivity analysis of variables under different refueling conditions, identifying the parameters that significantly influence the refueling success rate. This result enhances the interpretability of the model and provides theoretical guidance for the design of docking schemes for unmanned autonomous refueling.