Multi-physical fields prediction model for turbine cascades based on physical information neural networks

The flow field information within the cascade is crucial for turbine design. Currently, the physical field data in the cascade is mostly obtained through numerical simulation, which is accurate but time-consuming. To enable fast and accurate prediction of the physical fields within the cascade, this study proposes a Physics-Informed Fourier Neural Operator (PIFNO) model. Compared to pure data-driven surrogate models, PIFNO incorporates partial physical information into the modeling process through a physics-head correction approach during training, which not only improves prediction accuracy but also enhances model interpretability to some extent. To expand the applicability of PIFNO, this paper proposes a Transfer Learning-based multi-physics fields prediction model (TL-PIFNO) that can predict the physical fields within the cascade under different operating conditions using limited training data. Experiments show that PIFNO has relative errors within 2% for pressure and temperature field prediction, and maximum relative errors within 5% for velocity field prediction. TL-PIFNO can achieve similar accuracy to PIFNO using only 3/10 of the data volume and 1/10 of the training time, showing great potential for engineering applications.