DARWIN
Digital Matching of Conjugate Heat Transfer (CHT) simulation models with thermal paint experimental data

The aim of the LuFo project DARWIN at GLR is to understand the influence of modeling assumptions on the quality of the temperature prediction at the turbine blades, and to develop a procedure to improve the matching between conjugate heat transfer simulations and thermal paint experiments using machine learning.

To increase the thermodynamic efficiency of aero-engines and gas turbines, the turbine entry temperature was increased far above the melting temperature of turbine blade materials. Therefore, the cooling design of the high pressure turbine is of great importance.

On the one hand it has to be ensured that the thermal loading capacity of the blades is not exceeded. On the other hand, cooling air should not be wasted to avoid a loss in efficiency. To find the optimum cooling design, numerical investigations with conjugate heat transfer (CHT) simulations of the coupled fluid domain and solid domain are performed. Due to the complex flow field in the high pressure turbine, which is dominated by turbulence, boundary layers, interaction of stationary vanes, and rotating blades, as well as mixing processes of secondary and main flow, the simulation is very complex and various modeling assumptions have to be made to reduce the computational effort. In the temperature prediction, those assumptions lead to deviations between simulation and thermal paint experiments.

The aim of this project is to understand which of those simplifications have the greatest influence on the quality of the temperature prediction, and to develop a procedure to improve the matching between conjugate heat transfer simulations and thermal paint experiments.

Methodology

Conjugate Heat Transfer: Using the commercial solver “Siemens Simcenter StarCCM+” coupled simulations of the internal and external flow as well as the solid of fully featured HPT blades are performed to predict the blade temperature distribution.

Scale-Resolving Simulation: Identification of large modelling errors in RANS simulations by performing Scale Resolving Simulations on simplified testcases.

Machine Learning: Training of a model to transfer corrections from simplified cases to industrial use case.

Key Scientific Takeaways

  • Modeling the turbulent diffusivity in RANS with a constant turbulent Prandtl number leads to large modeling errors in film cooling flows
  • Development of a model to derive a spatial distribution of the turbulent Prandtl number from scalar flow quantities

Funding and cooperation

DARWIN is a joint research program in the frame of LuFo VI Call 1. It is financially supported by the Federal Ministry for Economic Affairs and Climate Action (BMWK) under grant number 20D1911C and Rolls-Royce Deutschland. Calculations for this research were conducted on the Lichtenberg high performance computer of the Technical University of Darmstadt.