Cooled turbine blades represent a major component of gas turbine technology due their potential for enhancing overall cycle performance. Due to the difficulty of performing detailed experiments on these complex internal configurations, the only option is to rely intensively on high fidelity CFD modeling and subsequent optimization to enhance the heat transfer coefficient (HTC) by exploration of a broad design space.

The optimization strategy relies on the following steps: parametric modeling of cooling passages; automatic grid generation; validation of the most reliable turbulence models and boundary conditions; optimization process based on genetic algorithms.

The parametric blade modeler AutoBlade™ has been upgraded to generate CAD geometries of cooling channels, including ribs, different section shapes, U-turns.

From those geometries, the full hexahedral mesher HEXPRESS™ is able to automatically generate high quality unstructured meshes with the possibility to insert viscous sublayers to provide adequate resolution in boundary layer regions. The mesh generator can also deal with multi-domain problems allowing conjugate heat transfer (CHT) method.

CFD options were analyzed in order to obtain numerical results as accurate as possible for the optimization process. It appears that low Reynolds grids were necessary to reproduce the thermal effects due to the strong temperature gradient near solid walls. The preconditioning technique seems to be essential for density based flow to obtain realistic results for flows at low Mach number. The anisotropic EARSM turbulence model appears to provide good heat transfer predictions. In addition, the question to include or not CHT is raised: the study also compares the heat transfer obtained in case of imposed static temperature conditions on the coolant to the ones obtained by means of CHT. The results highlight that CHT is more realistic.

Optimization of HTC is performed on three baseline configurations by variation of geometric parameters including the aspect ratio of the channel cross section, the shape, position and size of the ribs. The main objective consists in maximization the heat fluxes while limiting the head loss.

The optimization process starts from an initial database of high-fidelity CHT simulations obtained by design of experiment methods. To predict a potential optimum, an evolutionary algorithm is run on an artificial neural network for a very quick evaluation of the objectives from the database. The process is iterative: starting from potential candidates, the CFD process is launched to validate the guess and populate the database.

Results and optimized geometries will be presented in the full paper.

This content is only available via PDF.
You do not currently have access to this content.