Prediction of aerial refueling docking results by flying-wing UAV bow wave model established on CNN

Soft air refueling has become the most widely used unmanned autonomous refueling method for air forces in various countries due to its simple, reliable, and convenient advantages. The bow wave effect generated by the oil receiver is the main reason for docking failure. Establishing a bow wave model and accurately predicting the docking effect under its influence has become a key focus of research in this field. Traditional modeling methods have many problems, such as the semi-Rankine body model or vortex lattice method not being suitable for wing layout, and numerical simulation methods cannot cover a large number of flight conditions and are time-consuming. This study proposes a convolutional neural network (CNN) with transposed convolutional structures, which utilizes its ability to extract deep features from data to establish a flow field regression model from flight conditions to velocity distribution. And based on multi-rigid body dynamics, a finite element model was constructed to solve for the equilibrium position of the hose-drogue assembly (HDA). Subsequently, the Kriging interpolation method was used to train the drogue position offset model under different docking positions, thereby determining the range within which the oil receiver can successfully complete docking. This process uses flow field calculations obtained from CNN prediction and computational fluid dynamics (CFD) calculation respectively, and the results show that these two methods are very close, verifying the effectiveness of CNN method in constructing bow wave models and the feasibility of obtaining soft aerial refueling docking effects based on this model.