Data-driven model of film cooling superposition with vortex encoding

Film cooling arrays are critical for protecting high-temperature components in advanced gas turbines. Rapid and precise prediction of superposed film cooling effectiveness is essential for effective film cooling design. However, conventional methods, such as the Sellers model, are inaccurate for closely spaced film holes. While data-driven approaches have shown improved accuracy, their ability to capture diverse hole arrangements is limited due to complex vortex interactions. In this study, a vortex-encoded AI model is proposed to accurately predict film superposition for closely spaced film holes. The introduction of vortex encoding has significantly enhanced prediction accuracy, reducing high-prediction-error areas (absolute error > 0.05) by 38%. The model integrates a four-channel U-Net with Sellers operations, greatly improving the Sellers model's predictive accuracy for closely spaced holes and extending its applicability to dense film hole arrangements. On the test set, compared to the Sellers model, our model reduces high-prediction-error areas by 76% and lowers the mean absolute error by 84%. Additionally, the proposed model is also compatible with the superposition of film cooling with compound angles.