Knowledge enhanced modeling of low-pressure turbine profile loss by combining physical-based and data-driven methods

Appropriate modeling methods are essential for establishing models of energy systems, especially when the physical mechanisms are complex. A modeling method organically combined physical-based and data-driven is developed to ensure accuracy, generalization and interpretability while keeping a reasonable amount of required data. Utilizing the method, a comprehensive loss model for low-pressure turbine profile is established. The model considers freestream turbulence intensity and periodic wakes and accounts for flow modes of separation-reattachment and separation-non-reattachment. The model not only predicts loss at design and off-design points but also provides transition and reattachment locations. For physics-based modeling, the flow is spatially and temporally divided and modeled. In the data-driven part, over 450 sets of experimental data are utilized for coefficients optimization and validation. The model demonstrates high accuracy, with a relative error of less than 8 % for trailing edge momentum thickness and less than 3 % and 4 % for transition and reattachment locations, respectively. In the context of low-pressure turbine profile design guidelines, a front-loaded and high leading-edge load design is recommended. However, excessive leading-edge load can lead to increased loss, the leading-edge load integral (LEI) is suggested not to exceed 0.6. An appropriate front-loaded and high LEI design can achieve optimal performance.