错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Uncertainty involved drag divergence characteristic predicting method based on VAE

  • Wei Liu,
  • Hairun Xie,
  • Jing Wang,
  • Jixin Xiang,
  • Zenghui Huang,
  • Miao Zhang

摘要

Effective access to obtain the drag divergence characteristic of an airfoil is crucial for improving the economy, safety, and comfort of the aircraft. Current methods face challenges in providing satisfactory performance concerning efficiency, reliability and generalizability. In this study, an uncertainty-based network is developed, aiming to realize the prediction of the drag characteristic curves and drag divergence characteristics for different supercritical airfoil. To be more specific, a VAE-based network is established, with an encoder to convert the airfoil geometry into low-dimensional latent variables. These latent variables are then simultaneously fed into both the decoder and the drag characteristic prediction network. The decoder is responsible for geometric reconstruction, ensuring that the latent variables maintain a correlation with the geometric features of the airfoil. The prediction network maps the latent variables to the drag characteristic curve. Once trained, the uncertainty distribution of the drag can be realized through sampling from the distribution of latent variables. The analysis of results indicates the model’s outstanding performance, with a mean absolute error of \(7.407\times 10^{-5}\) 7.407 × 10 - 5 for the airfoil geometric reconstruction and a mean absolute error of 0.455 count for the drag coefficient prediction. The mean error of the drag divergence Mach number prediction reaches \(2.805\times 10^{-4}\) 2.805 × 10 - 4 . The results show that the model has considerable potential for engineering applications.