Deep learning for inverse design of low-boom supersonic configurations

Mitigating the sonic boom to an acceptable stage is crucial for the next generation of supersonic transports. The primary way to suppress sonic booms is to develop a low sonic boom aerodynamic shape design. This paper proposes an inverse design approach to optimize the near-field signature of an aircraft, making it close to the shaped ideal ground signature after propagation in the atmosphere. By introducing the Deep Neural Network (DNN) model for the first time, a predicted input of Augmented Burgers equation is inversely achieved. By the K-fold cross-validation method, the predicted ground signature closest to the target ground signature is obtained. Then, the corresponding equivalent area distribution is calculated using the classical Whitham’s F-function theory from the optimal near-field signature. The inversion method is validated using the classic example of the C608 vehicle provided by the Third Sonic Boom Prediction Workshop (SBPW-3). The results show that the design ground signature is consistent with the target signature. The equivalent area distribution of the design result is smoother than the baseline distribution, and it shrinks significantly in the rear section. Finally, the robustness of this method is verified through the inverse design of sonic boom for the non-physical ground signature target.