Missile aerodynamic shape optimization design using deep neural networks

It is necessary to optimize the design of the missile aerodynamic shape for better performance while meeting tactical specifications. However, current design methods for aerodynamic shape are based on manual design and physical model simulations, which are very time-consuming. Therefore, we propose an optimization framework based on conditional Wasserstein Gan-GP (CWGAN-GP), convolutional neural network (CNN), multi-task learning with multi-gate mixture-of-Experts-3D (MMoE-3D) and differential evolution (DE). This method consists of four stages. In the first step, CWGAN-GP can learn the relationship between existing missile shape designs and shape conditions, generating diverse missile shapes as required. In the second step, CNN is used for feature extraction of missile design drawing, and the missile shape generated by CWGAN-GP is transformed into missile shape parameters. In the third step, the MMoE-3D model is trained in the subsonic and supersonic ranges to efficiently generate aerodynamic data corresponding to the missile shape. In the fourth step, DE is used to select the optimal missile shape by adjusting the potential variables of CWGAN-GP. The efficiency of the proposed optimization framework is verified by optimizing the rate of change of the center of pressure and the lift-to-drag ratio, with the neural network-based optimization framework achieving almost the same optimization results in a shorter time compared to conventional optimization with DATCOM. (C) 2022 Elsevier Masson SAS. All rights reserved.

Automated summary

This paper proposes an optimization framework for missile aerodynamic shape design based on CWGAN-GP, CNN, MMoE-3D, and DE. The framework generates diverse missile shapes by learning the relationship between existing designs and shape conditions, and achieves optimization results similar to conventional methods but with shorter computation time through neural network-based optimization.