首页|Prediction of compressor blade cascade flow field based on Fourier neural operator

Prediction of compressor blade cascade flow field based on Fourier neural operator

扫码查看
原文链接
  • NETL
  • NSTL
  • Elsevier
This study employs the Fourier Neural Operator (FNO), a machine learning framework, to achieve fast prediction of cascade flow fields, accounting for variable cascade geometries and flow conditions. Based on the NACA65 airfoil, we sampled design parameters to generate 300 different cascade geometries. Subsequently, several sets of flow field data were generated by varying the inlet flow velocity and inlet angle of attack, of which 1500 sets were selected for model training and validation. For the geometric representation of the cascade, a coordinate transformation method was used, which can provide the geometric information of the airfoil more accurately and improve the prediction accuracy of the blade wall region. To further enhance the representation of geometric information, a signed distance function (SDF) was additionally introduced. Variations in cascade flow conditions were primarily reflected by adjustments to the inlet airflow velocity and angle of attack. The loss function was designed to comprehensively account for multiple factors, including norm-based loss, gradient-weighted loss, wall-layer-weighted loss, and periodic boundary constraint loss. By incorporating these constraints, the model achieved an MSE error magnitude of 10~(-6), with significantly improved prediction accuracy near the blade walls. In summary, this study demonstrates the effectiveness and high accuracy of the fourier neural operator in the prediction of the flow field of complex cascades, and provides a fast and reliable computational method for the design and optimization of cascades, which has an important value for engineering applications.

Flow field predictionNACA65Deep learningFNOCoordinate transformation

Lixiang Jiang、Xinlong Feng、Quanyong Xu

展开 >

School of Intelligence Science and Technology, Xinjiang University, Urumqi, 830046, PR China||Institute for Aero Engine, Tsinghua University, Beijing, 100084, PR China

College of Mathematics and System Sciences, Xinjiang University, Urumqi, 830046, PR China

Institute for Aero Engine, Tsinghua University, Beijing, 100084, PR China

2025

10.1016/j.ast.2025.110208

Aerospace science and technology

Aerospace science and technology

SCI
ISSN：1270-9638
年,卷(期)：2025.163(Aug.)
  • 29