High Fidelity Heat Transfer Analysis for Cooled Turbines Using GPU Accelerated CFD

Turbine cooling has long been a challenging task for industrial applications. The turbine blades are subject to high pressure and temperature combustion gas, often exceeds 1,500 C, and require cooling to avoid structure failure. The intricacy of a cooled turbine blade is shown in Figure 1 [1], where several hundred cooling holes of different shapes and sizes are placed on hub, shroud, and blade surface to lower blade temperature. The high temperature environment has also heightened the requirement for stronger material to constantly sustain harsh operating conditions. The unique balance of effective cooling while retain minimum weight has become an integral concept in turbine blade designs. The current design method is iterative in nature, based on years of engineering experience. The turbine blade cooling analysis includes mean flow and cooling flow as well as solid heat transfer, which constitutes a conjugate heat transfer (CHT) situation. Turbine cooling analysis is generally modeled using a two-equation RANS turbulence model in most, if not all, in industry to have a reasonable turnaround time to fit in a product development cycle. While RANS turbulence models provide necessary resolution in determining turbulent heat transfer, one of the most common assumptions is “turbulent Prandtl number” when calculating turbulent energy fluctuation term. The turbulent Prandtl number is assumed to be a constant 0.9, used universally in commercial CFD software. Recent research of a simple heated cylinder using LES has demonstrated the turbulent Prandtl number is not a constant. The best method to remove the constant turbulent Prandtl number assumption in RANS turbulence models is to use LES, where the turbulent energy fluctuation hypothesis no longer exists. Instead, turbulent Prandtl number is an outcome of an LES analysis. However, the turnaround time for an LES turbine blade cooling analysis by using traditional CPU based solvers can easily surpass weeks, if not months to complete one calculation. AeroDynamic Solutions Inc. (ADS) has developed an efficient algorithm by using GPUs instead of CPUs. The GPU solver has shown revolutionary speed-ups compared to using traditional CPUs. The GPU solver is validated for different types of application in turbomachinery and aerospace industries where speed-ups range from 20x to 50x. The revolutionary GPU technology has enabled ADS to take on turbine cooling LES analysis, where the turn-around time can be reduced from months to a day or two. ADS proposes to apply GPU accelerated LES analysis for cooled turbine analysis to determine turbulent Prandtl number distribution to effectively remove the turbulent Prandtl number assumption widely used in RANS turbulent models for cooled turbine design and analysis. This will be done across a large number of cases and compared to experimental data so that a database for turbulence modeling can be created.