Comparison of Drag Characteristics of Swept Wings and Forward-Swept Wings at Different Mach Numbers and Sensor Research on Compatible Aircraft Models

This study focuses on the drag differences between swept and forward-swept wings at different Mach numbers (0.3-0.9). A unified modeling framework incorporating aerodynamic mechanisms, structure-motion coupling, and multi-sensor fusion is constructed. Furthermore, as a further development in sensing applications, a real-time drag estimation method adaptable to various wing box structures is proposed. First, the model is based on the three-dimensional compressible Navier-Stokes equations and the ALE (Arbitrary Lagrange-Euler) description. This description helps establish a unified aerodynamic drag model across Mach numbers to characterize shock wave position changes, trans-wing flow intensity, and local effective angle-of-attack correction effects under different swept/forward sweep configurations. Second, addressing the pressure measurement noise problem caused by high-energy pulsations in forward-swept wings, this study constructs a multi-source sensing system consisting of a 64-point pressure array, a strain array, and millimeter-wave displacement sensors. Simultaneously, an energy-adaptive wavelet packet thresholding method is proposed to achieve noise suppression under Mach number conditions. Experiments show that this method can reduce the high-frequency noise RMSE by$60-65 \%$in the Mach range of 0.85-0.9. Furthermore, a multi-sensor collaborative network based on graph theory topology and an RBF neural network nonlinear compensation model are used to achieve distributed drag estimation. Compared with the single-sensor integration method and the traditional Kalman filter fusion method, the proposed model has an average drag error of approximately$3.4 \%-8.7 \%$in swept wing scenarios and approximately$4. 8 \% - 1 3. 7 \%$in forward-swept wing scenarios. At the system level, the overall accuracy improvement is$35 - 55 \%$. In the sensor degradation experiments, even with a node loss of 20%, the estimation error of the proposed model is still more than 50% lower than that of the KF model. Overall experimental results show that this method maintains stable estimation accuracy under complex conditions such as shock wave movement, noise enhancement, and incomplete sensor availability.