Near-real-time prediction of full vane film cooling: A data fusion method based on iterative Fourier neural operator and secondary transfer learning

Effective monitoring of full-surface temperature field on aero-engine turbine blades is critical for performance optimization and fault detection, yet it is severely hampered by unstable inlet conditions and the extreme sparsity of sensor data in operational environments. This study overcomes these challenges by proposing a novel data fusion framework that synergistically leverages multi-source data. At its core is the innovative integration of an Iterative Fourier Neural Operator (IFNO) with a Secondary Transfer Learning (STL) strategy. This framework enables the rapid, high-resolution reconstruction of the entire vane surface's film cooling effectiveness from sparse sensor measurements—without requiring prior knowledge of the inlet conditions. Validation on the E3 NGV with film cooling, utilizing only 3% of the full field data, demonstrates exceptional accuracy: under unknown and varying coolant flow rates, the Mean Absolute Error (MAE) on the suction and pressure surfaces is less than 0.2% and 0.3%, respectively; Similarly, with unknown and varying inlet temperature distributions, the MAE is below 0.15% and 0.3%. The framework's robustness is further highlighted by an extreme case: reducing the proportion of discrete points from 3% to 0.1% increased the MAE by only 2.1%, while preserving the integrity of the predicted spatial distribution.