Numerical Simulation of a Complete Francis Turbine

including unsteady rotor/stator interactions

Ruprecht, A., Heitele, M., Helmrich, T.

Institute for Fluid Mechanics and Hydraulic Machinery
University of Stuttgart, Germany

Moser, W., Aschenbrenner, T.

Voith Hydro, Heidenheim, Germany

ABSTRACT
The unsteady simulation of an entire Francis turbine is shown. The investigated turbine consists of spiral case,
tandem cascade with 24 channels, runner with 13 blades and draft tube. The computational boundary conditions
are applied at the spiral inlet and at the draft tube outlet. In the calculation no periodicity is assumed and no
circumferential averaging is applied. Therefore all tandem cascade and runner channels have to be modeled. The
computational algorithm including parallelization and rotor/stator interface is briefly described. The rotor/stator
interface is realized by means of dynamic boundary conditions and allow non-matching grids. Selected
computational results are shown for two points of operation.

INTRODUCTION

For design and research in hydro turbines the application of computational methods is steadily
increasing. Nowadays it is fairly standard to simulate the different components separately. By
assuming steady state uniform flow conditions in circumferencial direction only one channel
of the tandem cascade (stay vanes and guide vanes) and of the runner has to be considered.
Since very often there are strong interactions between the components (especially between
guide vanes, runner and draft tube) it is inevitable to introduce this interaction into the
simulation for accurate results. Many attempts have been made lately to do this by applying
an averaging procedure in the circumferential direction. This still allows to consider only one
channel of tandem cascade and runner and to apply a steady state computation, which results
in a severe saving of computer resources. Different approximations are compared in [1]. The
obtained results agree quite well with measurements and can be applied for turbine design,
e.g. [2,3,4].

However, many problems arise due to dynamical effects and vibrations. Dynamical forces can
not be obtained by the steady state approach. Also the measurement of the dynamical forces is
quite complicate and requires a rather high measuring effort. Therefore it is desirable to
obtain it by simulation. For the calculation of these dynamical effects it is essential to apply an
unsteady calculation of flow including the dynamical stator - rotor interaction. Due to the non-
uniform inflow from the spiral case and due to the unequal pitching of guide vanes and runner
it is necessary to consider the entire turbine with all channels of the runner and of the tandem
cascade. This results in an extremely high computational effort, which can only be performed
in parallel on powerful computers with reasonable response times.

In this paper the unsteady simulation of an entire Francis turbine is shown. The investigated
turbine consists of spiral case, tandem cascade with 24 channels, runner with 13 blades and
draft tube. The computational algorithm including parallelization and rotor/stator interface is
described and computational results are shown.
BASICS AND NUMERICAL PROCEDURES

The calculations are based on the Reynolds averaged Navier-Stokes equations. The turbulence
is taken into account either by an algebraic model or by the standard k-ε model including the
“hybrid” Kato-Launder correction [5]. The spatial discretization is obtained by introducing a
Petrov-Galerkin FEM with tri-linear hexahedral elements. For the time discretization a 2nd
order three-level fully implicit scheme is applied. The implicit scheme has the advantage that
the description of the time step is only dependent on the physics and not on the numerical
mesh.

Because of the consideration of unsymmetrical flow in spiral case the inflow to tandem
cascade and consequently the runner can be unsymmetrical as well. Together with the unequal
pitching of guide vanes and runner, this leads to the fact, that no flow periodicity exists.
Therefore the complete turbine including all flow channels in the tandem cascade and in the
runner has to be considered. This represents the most general approach for the prediction of
rotor/stator interactions, but it requires huge computational grids. Other methods e. g.
applying time periodicity can reduce the computational efforts, they however include certain
restrictions.

The whole calculation domain is divided into the

four components (spiral case, tandem cascade,
runner, draft tube). For each component an
independent computational grid is generated. The
different grids are slightly overlapping. They do not
have to match at the boundaries. The calculation of
the four components is carried out simultaneously.
The connection between the components is realized
by dynamic boundary conditions. In downstream
direction the interpolated node values for velocities
and turbulence quantities are exchanged. In
upstream direction the pressure and the fluxes are
prescribed in form of surface integrals, see fig. 1.

Each of the components is also calculated in

Fig. 1: Dynamic boundary conditions Fig. 2: Solution procedure

parallel by domain decomposition, therefore the computational grids are divided into parts of
equal sizes. The parallelization is performed in the linear equation solver. The runner is
calculated in an rotating frame of reference. For details concerning the algorithm and the
performance of the parallelization the
reader is referred to [7,8]. The simulation
procedure is summarized in fig. 2. It shall
be mentioned that because of the rotating
runner the exchange of the boundary
conditions between the components leads
to a dynamical connection of the
different processors. This is
schematically shown in fig. 3.

It has to be mentioned, that because of

the complex geometry rather large grids Fig. 3: Moving grid algorithm
(millions of nodes) are necessary. Even
for comparatively coarse mesh the
unsteady simulation leads to a huge
amount of data. The postprocessing of
these data requires a very efficient and
powerful hardware equipment. At the
University Computer Center a VR-Cube
with a powerful 14-processor
workstation for postprocessing is
available. This allows a quit fast and
efficient evaluation of the simulation
data. A look in the Cube is shown in
fig. 4. In spite of the powerful tools,
however, the data evaluation is still
very time consuming. Consequently it
would be desirable to have intelligent
software routines, which automatically Fig. 4: Postprocessing in Virtual Reality
can investigate the data in order to
detect critical flow regions (e. g.
vortices). Otherwise with increasingly fine grids and finer time steps not all important
information included in the simulation data can be recognized by the engineer. This will be a
future field of research work.

INVESTIGATED GEOMETRY AND COMPUTATIONAL GRIDS

The investigated turbine is a Francis turbine consisting of the spiral case with 23 stay vanes
and a tip nose. These are followed by 24 guide vanes and the runner with 13 blades. At the
outlet a draft tube with two piers is located. The computational mesh of the turbine is shown
in fig. 5. Different colors represent different processors for the calculation. For details also
one channel of the tandem cascade is presented in the figure.
 Tandem cascade

Fig. 5: Geometry and computational grid of the turbine

The whole computational grid consists of more than 2 million grid nodes. It is divided into 96
parts with the approximately equal number of nodes. During the calculation each part is run
on its own processor. The simulation has been carried out on the CRAY T3E/512 of the
University Computer Center. The time step for the calculation was chosen to obtain 1.15
degree of runner rotation per time step.

As boundary conditions a fully developed pipe flow at the spiral case inlet and free outflow at
the draft tube outlet were assumed. The initial condition was obtained by a steady state
calculation with a frozen rotor approach. It has to be mentioned, that the grids are rather
coarse. Especially in the tandem cascade and in the runner finer grids are required.
Investigations with finer grids are in progress.

RESULTS

The calculations were carried out for two different points of operation
• part load and
• optimum.
At first the overall overview of the results are presented for the part load point of operation. In
fig. 6 the instantaneous pressure distribution in the tandem cascade and in the runner is
presented for a certain time step. Clearly the reduction from the high pressure (red) in the
spiral case to the low pressure (blue) at the runner outlet can be observed. At the runner inlet
the stagnation point at the leading edge is also clearly visible. Looking at the stay vane inlet
the unsymmetrical pressure distribution in circumferential direction can be seen, which results
from the non-uniform flow distribution of the spiral case.
 Fig. 6: Instantaneous pressure distribution for a certain time step

Looking on the pressure distribution at the stay vanes one can also observe a strong change
along the height. At the top of the stay vanes the pressure is quite high whereas at the bottom
the pressure is low. The reason is the strong turning of the flow from the radial to the axial
direction. This behavior can also be observed in the velocity distribution, shown in fig. 7.
There the magnitude of the absolute velocity for a certain time step is shown in the mid plane
of the tandem cascade and at the outlet of the stay vanes. Again the acceleration of the flow on
the bottom can be observed clearly, which results in the reduction of the pressure, fig. 6. The
unsymmetrical inflow to the tandem cascade can be seen again, which results from the spiral
case. Also very obvious are the wake behind the stay vanes. These wakes expand to the
runner and causes fluctuations of the surface pressure and consequently also fluctuations of
forces and torque.

Fig. 7: Instantaneous velocity distribution in the tandem cascade

In fig. 8 the overall flow behavior in the
whole turbine shall be presented. For
that purpose the velocity distributions
on different planes are shown for a
fixed time step. One can observe the
acceleration from the spiral case into
the tandem cascade and the runner and
the deceleration of the flow in the draft
tube. In the draft tube a very
unsymmetrical flow distribution is
obtained. The draft tube channel in the
middle shows a very low velocity
(black color) whereas the channel on
the right hand side (in flow direction) Fig. 8: Instantaneous velocity distribution for
has a high discharge (dark blue color). part load operation point

Especially the draft tube flow strongly

depends on the point of operation since
the swirl rate behind the runner very
much affects the flow behavior. In fig. 9
the instantaneous velocity distribution
for optimum condition is shown. Here
you can see that the discharge in the
middle channel is increased, whereas
the channel of the left hand side shows
a reduced flow rate.

In fig. 10 the instantaneous distribution

in the meridional mid plane is plotted
for three different time steps, which Fig. 9: Instantaneous velocity distribution for
means three different positions of the optimum point of operation
runner. Again it can be observed, that
the flow is strongly accelerated towards the runner. This leads to a very fast equalization and
consequently to a very fast fill in of the wakes behind the guide vanes. The result of this is,
that only a very weak effect of the stator-rotor interaction can be observed. This would be
much stronger in pure axial turbines (e. g. bulb turbines), where no strong flow acceleration
towards the runner exists, see e. g. [6,9].

Fig. 10: Velocity distribution for three different positions (time steps)
The relatively weak interaction of guide vanes and runner can also be seen in fig. 11. There
the variation of the axial force (fig. 11 top) and of the torque (fig. 11 bottom) on a single
runner blade are plotted versus the rotation angle of the runner. It can be observed, that the
axial force as well as the torque show fluctuations of approximately +/- 2%, which is very
low, compared for example with the axial type turbine presented in [6].

Looking at the distribution of the axial force one can see that it shows an almost periodical
behavior. Clearly two dominating frequencies can be evaluated. The lower one is caused by
the runner speed. The higher fluctuations correspond to the number of guide vanes time the
runner speed. In this case 24 small peaks can be counted during one revolution of the runner.
The fluctuations caused by the interaction with the guide vanes are quite small. The reason for
that is that the wakes behind the guide vanes vanish very early because of the accelerated
flow. This could be observed
already in fig. 10. The distribution
of the torque shows nearly the
same behavior than that of the axial
force. The amplitude of overall
fluctuation is slightly higher.
However it is still very low
compared e. g. with the axial
turbine in [6].

For both quantities, the axial force

as well as the torque, fig. 11 shows
the distribution on a single runner
blade. Looking at the total force or
torque the fluctuations are much
lower, since the forces on the 13
runner blades have different phases
and consequently the sum of the
forces (and torques) remain quite
constant.

Again it has to be mentioned, that

the presented calculations are
performed on a rather coarse mesh.
This means, that the wakes behind
the blades are equalized to early. In
addition it has to be emphasized,
that the computational effort is
quite high. Therefore it is costly to
simulate long time periods.
However this leads to a severe
problem, because the flow in the Fig. 11: Variation of discharge, axial force and torque
turbine consists of a huge spectrum with runner position
of frequencies and initial periods.
Concerning the rotor-stator interaction, for example, one obtains quite fast a “periodic”
situation. For this it is enough to calculate 2-3 rotations of the runner. In regard to the vortex
rope in the draft tube, however, the calculation period has to be much longer. It takes quite a
long time until the vortex rope forms and reaches a “periodic” circulation. This requires the
calculation of at least 15-20 revolutions of the runner, which is extremely costly. For the
above shown results the calculation time therefore was to short to obtain the vortex rope,
which appeared in the model tests. That the calculation procedure is able to predict the vortex
rope, however, is shown in [6]. To overcome this difficulty it is intended to apply a multi-
scale approach with different time steps in order to reduce the computational effort. This
approach is in progress.

CONCLUSIONS

The unsteady calculation of the flow in an entire Francis turbine from spiral case through stay
vanes, wicket gates, runner down to the draft tube is presented. No periodicity is applied
neither in the stay vanes and wicket gates nor in the runner. Therefore all flow channels are
modeled. This results in huge computational grids and computational times, which only can
be handled by parallel computing. The coupling between the different components are
performed on sliding non-matching meshes with dynamic boundary conditions, whereas the
parallelization in each component is obtained by domain decomposition and application of a
parallel solver for the linear system of equations.

The calculations were carried out for a turbine with 24 stay vanes and wicket gates and with
13 runner blades and some results are presented. These calculations are intended to be a
feasibility study rather than to obtain major quantitative results, but these and similar
calculations for an axial turbine [7] show that the simulations are feasible and meaningful and
quantitatively accurate results concerning dynamical forces and loading can be expected.
Therefore more accurate solutions on finer grids are in progress. One major difficulty in the
calculation is the presence of quite different flow phenomena, which very much distinguish in
frequencies and initial periods. Applying a single time step for all the problems results in
extensive requirements of computer time. Therefore it is worked on a multi-scale approach.

Reference
[1] Kächele, Th., 20th IAHR Symposium on Hydraulic Machinery and Cavitation, Charlotte, 2000.
[2] Riedelbauch, S., Klemm, D., Hauff, C.: Importance of interaction between turbine components in flow field
simulation, 18th IAHR Symposium Hydraulic Machinery and Cavitation, Valencia, 1996.
[3] Labrecque, Y., Sabourin, M., Deschênes, C.: Numerical simulation of a complete turbine and interaction
between components, Modelling, Testing & Monitoring for Hydro Powerplants, Lausanne, 1996.
[4] Sick, M., Casey, M. V., Galphin, P.F.: Validation of a stage calculation in a Francis turbine, 18th IAHR
Symposium Hydraulic Machinery and Cavitation, Valencia, 1996.
[5] Chen, W. L., Lien, F. S., Leschziner, M.: A Computational Modeling of Turbulent Flow in Turbomachine
Passage with Low-Re Two-equation Models, Computational Fluid Dynamics, Wiley & Sons Ltd., 1994.
[6] Ruprecht, A.:Unsteady Flow Analysis in Hydraulic Turbomachinery, 20th IAHR Symposium on Hydraulic
Machinery and Cavitation, Charlotte, 2000.
[7] Heitele, M., Helmrich, T., Maihöfer, M., Ruprecht, A.: New Insight into an Old Product by High
Performance Computing, 5th European SGI/CRAY MPP Workshop, Bologna, 1999.
[8] Ruprecht, A., Bauer, C., Gentner, C., Lein, G.: Parallel Computation of Stator-Rotor Interaction in an Axial
Turbine, ASME PVP Conference, CFD Symposium, Boston, 1999.
[9] Ruprecht, A., Bauer, C. Heitele, M.: Unsteady forces on the blading of an axial turbine caused by stator-
rotor interaction, IAHR WG "The Behaviour of Hydraulic Machinery under Steady Oscillatory
Conditions", Brno, 1999.