Optimization of turbulence models for internal cooling systems of gas turbine blades using CFD-integrated machine learning

• Applicant: Professor Dr.-Ing. Bernhard Weigand
• Subject Area: Hydraulic and Turbo Engines and Piston Engines
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 567311830

The main objectives in the development of industrial gas turbines and gas turbines for propulsion are to reduce the fuel consumption and the pollutant emissions. This can be achieved, e.g., by increasing the thermal efficiency of the gas turbine by increasing the process temperature. However, the increased combustion chamber temperature and the higher turbine inlet temperature already exceed the melting temperature of the blade materials. This makes it necessary to develop very efficient internal turbine blade cooling strategies. Various internal cooling concepts are currently being intensively investigated, such as ribbed channels, pin fins, impingement cooling, dimples and cyclone cooling chambers. The computation of the flow and heat transfer in such complex, three-dimensional cooling channels, is a major challenge that can only be mastered to a limited extent by state-of-the-art turbulence models. Improved turbulence models are, therefore, required to optimise such internal cooling systems. Due to the appearing higher Reynolds numbers and the very complex three-dimensional internal cooling geometries of blades, RANS-based methods must still be used to calculate the flow and heat transfer in such systems. In recent years, great hope has been placed in the application of machine learning methods to improve traditional turbulence models, especially in this area. However, this has so far only been demonstrated to a limited extent for highly complex internal cooling systems. As part of this research project, jointly calibrated, data-driven momentum and temperature closure models are now to be developed for the first time, especially for the use in the area of the internal cooling of gas turbines. These models consist of general equations for the Reynolds stresses and the scalar-flux vector and of transport equations for the turbulent scales. To adapt the models, all coefficients contained therin should depend on the local flow state and the associated functions should be described by a single neural network. Since an a priori optimisation of such a neural network is not puposeful, a CFD-integrated training method will be used in this project to ensure the stability of the model and, thus, also its a posteriori quality. The models obtained will be trained and validated on various test cases of varying complexity. The quality of the modelling is then tested on complex two-pass cooling systems for which sufficient experimental reference data are already available for comparison. The models obtained here and the method developed are then transferable to other complex 3D cooling problems, if similarity exists in the geometry and the Reynolds- and the Prandtl number.

• DFG Programme: Research Grants