This paper presents a novel CFD-driven machine learning framework to develop Reynolds-averaged Navier-Stokes (RANS) models. For the CFD-driven training, the gene expression programming (GEP) method (Weatheritt & Sandberg, J. Comput. Phys., 325, 22-37 (2016)) uses RANS calculations in an integrated way to evaluate the fitness of candidate models. The resulting model, which is the one providing the most accurate CFD results at the end of the training process, is thus expected to show good performance in RANS calculations. To demonstrate the potential of this new approach, the CFD-driven machine learning is applied to develop a model for improved prediction of wake mixing in turbomachines. A new model is trained based on a high-pressure turbine training case with particular physical features. The developed model is shown to have a more compact functional form than models trained without CFD assistance. Furthermore, the trained model has been evaluated a posteriori for the training case and three additional test cases with different physical flow features, and the predicted wake mixing profiles are significantly improved in all cases. With the present framework, the model equation is explicitly given and available for analysis, thus it could be deduced that the enhanced wake prediction is due to the extra diffusion introduced by the CFD-driven model.
In this paper, we establish a benchmark data set of a generic high-pressure (HP) turbine vane generated by direct numerical simulation (DNS) to resolve fully the flow. The test conditions for this case are a Reynolds number of 0.57 × 106 and an exit Mach number of 0.9, which is representative of a modern transonic HP turbine vane. In this study, we first compare the simulation results with previously published experimental data. We then investigate how turbulence affects the surface flow physics and heat transfer. An analysis of the development of loss through the vane passage is also performed. The results indicate that freestream turbulence tends to induce streaks within the near-wall flow, which augment the surface heat transfer. Turbulent breakdown is observed over the late suction surface, and this occurs via the growth of two-dimensional Kelvin–Helmholtz spanwise roll-ups, which then develop into lambda vortices creating large local peaks in the surface heat transfer. Turbulent dissipation is found to significantly increase losses within the trailing-edge region of the vane.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.