ENGINEERING AND CFD

Due to the liberalization of the electricity market and an increasing share of renewable energies such as wind power and photovoltaics, hydropower plants equipped with Francis turbines are increasingly used to regulate fluctuations in the electricity grid. As a result, the share of operation at the best efficiency point is reduced and off-design operating points become more important. Particularly in these operating ranges, complex flow phenomena can occur that are for example caused by the misaligned flow in the runner. These phenomena are typically accompanied by strong pressure fluctuations in the entire hydraulic machine that can cause severe damage to turbine components.

The prediction accuracy for these operating points with RANS turbulence models that are typically used in industry is relatively poor. This includes, among other things, deviations in the prediction of the velocity profile in the draft tube cone and thus inaccuracies in the simulation of the vortex rope that is forming and the associated pressure fluctuations. Therefore it is the goal of this project to investigate possibilities to improve the simulation accuracy of Francis turbines as a reference case. The focus is on the part load operating point, which is characterized by a helically shaped vortex rope in the draft tube cone and several vortex structures of smaller scales (see figure 1).

One of the main challenges of this project is that the simulation domain that ranges from the spiral case inlet to the outlet of the downstream tank is very large. However, a reduction of the simulation domain is not advisable for highly accurate flow predictions which are one aim of this study. Furthermore, a long simulation time is necessary as on the one hand the flow phenomena need some time to develop, and on the other hand the averaging of relevant flow quantities requires a high amount of runner revolutions. For the present case up to about 80 runner revolutions were simulated. Consequently, the computational effort is tremendous and would not be feasible without the use of supercomputers.

In the scope of this project, hybrid RANS-LES turbulence modeling is applied. Even though a Large Eddy Simulation (LES) simulation would be desirable as a reference simulation, it is nowadays still computationally too expensive for Francis turbines. The reason is that a properly resolved LES simulation would need several billions of mesh cells due to the high Reynolds number of this application. Nevertheless, the use of hybrid RANS-LES models allows reducing the modeling to a minimum and resolving the majority of turbulent vortices.

In figure 2 the influence of mesh resolution and different turbulence models on the resolution of turbulent vortices is visualized. The RANS (Reynolds Averaged Navier Stokes) simulation with the SST model that is typically used in industry can only resolve large scale vortices even though a quite fine mesh resolution of 50 million (50M) cells is used. Compared to that, the hybrid RANS-LES simulation with the same mesh resolves also smaller vortices. However, at some locations the viscosity ratio is still high, which indicates still a relevant amount of modeling of turbulent structures. With a finer mesh (300 million mesh cells) the best resolution of turbulent vortices can be achieved. Furthermore, the viscosity ratio is quite low, which indicates that most turbulent structures are resolved and modeling is reduced to a minimum. All in all, a significant impact of the turbulence model and the mesh resolution can be stated that are of main importance for the highly accurate prediction of the flow.