You are on page 1of 9

Scientific Bulletin of the Workshop on

Politehnica University of Timisoara Vortex Dominated Flows –


Transactions on Mechanics Achievements and Open Problems
Special issue Timisoara, Romania, June 10 - 11, 2005

VERY LARGE EDDY SIMULATION FOR SWIRLING FLOWS WITH APPLICATION IN


HYDRAULIC MACHINERY

Ivana BUNTIĆ, PhD student Thomas HELMRICH, Research assistant


Institute of Fluid Mechanics and Hydraulic Institute of Fluid Mechanics and Hydraulic
Machinery Machinery
University of Stuttgart University of Stuttgart
Albert RUPRECHT, Head of Fluid Mechanics Group *
Institute of Fluid Mechanics and Hydraulic Machinery
University of Stuttgart
*Corresponding author: Pfaffenwaldring 10, D-70550, Stuttgart, Germany
Tel.: (+49) 711 6853256, Fax: (+49) 711 6853255, Email: ruprecht@ihs.uni-stuttgart.de

ABSTRACT ε [m2/s3] dissipation rate


In this paper a new turbulence model for Very Large ν [m2/s] kinematic viscosity
Eddy Simulation is presented. Its main νt [m2/s] turbulent viscosity
characteristic is an adaptive filtering technique ∆V [m2 or m3] size of the local
which can distinguish between numerically resolved element
and unresolved parts of the flow. This unresolved Subscripts and Superscripts
part is then modelled with extended k-ε model of ^ modelled
Chen and Kim. VLES is applied to the simulation of i covariant indices, i = 1,2,3
unsteady vortex flows in a straight diffuser, a draft
tube and a pipe trifurcation which usually cannot be 1. INTRODUCTION
predicted with classical turbulence models. Using
It is well known that one of the fundamental
the new technique, these complex phenomena are
problems of Computational Fluid Dynamics (CFD)
well predicted.
is the prediction of turbulence. Reynolds-averaged
Navier-Stokes (RANS) equations are established as
KEYWORDS
a standard tool for industrial simulations and
Very Large Eddy Simulation, adaptive turbulence
analysis of fluid flows, although it means that the
model, straight diffuser, draft tube, trifurcation
complete turbulence behaviour has to be enclosed
within appropriate turbulence model which takes
NOMENCLATURE
into account all turbulence scales (from the largest
f [-] filter function
eddies to the Kolmogorov scale). Consequently
hmax [m] local grid size
defining a suitable model for prediction of complex,
k [m2/s2] turbulent kinetic energy
especially unsteady, phenomena is very difficult.
L [m] Kolmogorov length scale
The highest accuracy for resolving all turbulence
Pk [-] production term
scales is offered by a Direct Numerical Simulation
u [m/s] local velocity
(DNS). It requires a very fine grid resolution. Hence
Ui [m/s] filtered velocity
carrying out 3D simulations for complex geometries
Ui [m/s] averaged velocity
and flow with high Reynolds number is nowadays
P [Pa] averaged pressure time consuming even for high performance
τij [Pa] Reynolds stresses computers, see Figure 1. DNS is unlikely to be
α [-] model constant applied to flows of practical relevance in the near
∆ [m] resolved length scale future.
∆t [s] time step
Figure 1. Degree of turbulence modelling and Figure 2. Modelling approaches for RANS and LES
computational effort for the different approaches
In this paper the development of a VLES
Large Eddy Simulation (LES) starts to be a turbulence model is presented. It is based on the
mature technique for analyzing complex flows, extended k-ε Chen and Kim [1]. Applying an
although its major limitation is still expensive appropriate filtering technique the new turbulence
computational cost. In the “real” LES all anisotropic model distinguishes between resolved and modelled
turbulent structures are resolved in the computation part of the turbulence spectrum. Because of its
and only the smallest isotropic scales are modelled. adaptive characteristic it can be applied for the
It is schematically shown in Figure 2. The models whole range of turbulence modelling approaches
used for LES are simple compared to those used for from the RANS to the DNS.
RANS because they only have to describe the Here presented applications of the new adaptive
influence of the isotropic scales on the resolved turbulence model are simulation of the flow in a
anisotropic scales. With increasing Reynolds straight diffuser, a draft tube and a pipe trifurcation.
number the small anisotropic scales strongly In all cases unsteady motions are observed and
decrease becoming isotropic and therefore not computationally well predicted.
resolvable. There are many “LES” of engineering
relevant flows in the literature, although they can be 2. SIMULATION METHOD
characterised as unsteady RANS (URANS) due to
the fact that they only resolve unsteady mean flow 2.1. Governing equations
not taking into account any turbulence structure. In this work an incompressible fluid with
If there is a gap in the turbulence spectrum constant properties is considered. The governing
between the unsteady mean flow and the turbulent equations describing this incompressible, viscous
flow, ”classical” RANS i.e. URANS models can be and time dependant flow are the Navier-Stockes
applied, as they are developed for modelling the equations. They express the conservation of mass
whole range of turbulent scales, see Figure 2. It also and momentum. In the RANS approach, these
means that they are not suitable for prediction and equations are time or ensemble averaged leading to
analysis of many unsteady vortex phenomena. the well known RANS equations:
Contrary, if there is no spectral gap and even one ∂U i ∂U i ∂P ∂τ ij
part of the turbulence can be numerically resolved, +U j =− + ν∇ 2U i − (1)
∂t ∂x j ∂x i ∂x j
we can use Very Large Eddy Simulation (VLES). It
∂U i
is very similar to the LES, with the difference that a =0 (2)
smaller part of the turbulence spectrum is resolved ∂x i
and the influence of a larger part of the spectrum In RANS τij expresses the Reynolds stress tensor
has to be expressed with the model, see Figure 2. which is unknown and has to be modelled. The task
Nowadays it seems to be a promising compromise of turbulence modelling is the formulation and
for simulation of industrial flow problems with determination of suitable relations for Reynolds
reasonable computational time and costs. stresses. Details of the new VLES approach are
described in section 3.
2.2. Numerical method
The calculations are performed using the
program FENFLOSS (Finite Element based
Numerical FLOw Simulation System) which is
developed at the Institute of Fluid Mechanics and
Hydraulic Machinery, University of Stuttgart.
It is based on the Finite Element Method. For
spatial domain discretisation 8-node hexahedral
elements are used. Time discretisation involves a
three-level fully implicit finite difference
approximation of 2nd order. For the velocity
components and the turbulence quantities a trilinear
approximation is applied. The pressure is assumed
to be constant within element. For advection
dominated flow a Petrov-Galerkin formulation of
2nd order with skewed upwind orientated weighting
function is used.
For the solution of the momentum and continuity
equations a segregated algorithm is used. It means
that each momentum equation is handled
independently. They are linearised and the linear
equation system is solved with a conjugated
gradient method BICGSTAB2 of van der Vorst [12]
with an incomplete LU decomposition (ILU) for
preconditioning. The pressure is treated with the Figure 3. FENFLOSS flow chart
modified Uzawa pressure correction scheme [14].
The pressure correction is performed in a local 3. MODELLING APPROACH
iteration loop without reassembling the system
3.1. Very Large Eddy Simulation
matrices until the continuity error is reduced to a
given order. Lately several hybrid methods are proposed in
After solving the momentum and continuity the literature:
equations, the turbulence quantities are calculated - Very Large Eddy Simulation (VLES)
and a new turbulence viscosity is gained. The k and - Semi-Deterministic Simulations (SDS)
ε - equations are also linearised and solved with - Coherent Structure Capturing (CSC)
BICGSTAB2 algorithm with ILU preconditioning. - Detached Eddy Simulation (DES)
The whole procedure is carried out in a global - Hybrid RANS/LES
iteration until convergence is obtained. For - Limited Numerical Scales (LNS)
unsteady simulation the global iteration has to be All of them are based on the same idea to represent
performed for each time step. FENFLOSS flow a link between RANS and LES. Generally they all
chart is shown in Figure 3. can be classified as Very Large Eddy Simulations
The code is parallelised and computational and their main aim is to overcome the
domain is decomposed using double overlapping computational costs and capacity problems. These
grids. In that case the linear solver BICGSTAB2 has methods try to keep computational efficiency of
a parallel performance and the data exchange RANS and the potential of LES to resolve large
between the domains is organised on the level of the turbulent structures. Although they can be
matrix-vector multiplication. The preconditioning is performed on coarser grids, the simulations are
then local on each domain. The data exchange uses strongly dependent on the modelling.
MIP (Message Passing Interface) on computers with Above mentioned methods slightly differ in
distributed memory. On the shared memory filtering techniques, applied model and
computers the code applies OpenMP. For more interpretation of the resolved motion, but broadly
details on the numerical procedure and speaking they all have a tendency to solve complex
parallelisation the reader is referred to [5,6]. unsteady turbulent flows at high Reynolds number
implying a principle “solve less – model more”, see
Figure 4 and Table 1. It means that the relevant part
of the flow (unsteadiness) is resolved and the rest is
modelled.

Figure 5. Adjustment for adaptive model

Figure 4. Modelling approach in VLES For distinguishing resolved and modelled


turbulence spectrum, see Figure 6, the adaptive
Table 1. Resolution in DNS, LES and VLES [7] model uses a filtering technique. There are several
Model Resolution of them described in the literature [2, 4, 11], but the
applied technique is similar to Willems [13]. The
Direct numerical All turbulent scales are smallest resolved length scale ∆ used in filter is
simulation (DNS) resolved according to Magnato and Gabi [4] dependant on
the local grid size or the computational time step
Large eddy simulation Grid size and filtering and local velocity.
with near-wall are sufficient to resolve
resolution 80% of the energy

Large eddy simulation Grid size and filtering


with near-wall are sufficient to resolve
modelling 80% of the energy
distant from the wall,
but not in the near-wall
region

Very large eddy Grid size and filtering


simulation (VLES) are not sufficiently fine
to resolve 80% of the
energy
Figure 6. Distinguishing of turbulence spectrum by
3.2. Adaptive turbulence model
VLES
Classical turbulence models, which are usually
applied for solving engineering flow problems, The basis of the adaptive model is the k-ε model
represent the whole turbulent spectrum. They show of Chen and Kim [1]. It is chosen due to its
excessive viscous behaviour and very often damp simplicity and capacity to better handle unsteady
down unsteady motion quite early. Therefore they flows. Its transport equations for k and ε are given
are not completely appropriate for simulation of as
certain flow cases. 
∂k ∂k ∂ ν  ∂k 
VLES is used for resolving at least one part of +U j = ν + t   + Pk − ε (3)
turbulence spectrum and thus getting more precise ∂t ∂x j ∂x j  σ k  ∂x j 
picture of the flow behaviour. Depending on the
type of the flow and grid size applied model should ∂ε ∂ε ∂  νt  ∂ε 
+U j = ν + 
 ∂x +
automatically adjust to one of the modelling ∂t ∂x j ∂x j  σε  j 
approaches schematically shown in Figure 5. (4)
ε ε2 P 
Therefore an adaptive model is developed. Its c1ε Pk − c 2ε + c 3ε  k  ⋅ Pk
advantage is that with increasing computer power it k k
142  k 43
4 4
can be afforded that a larger part of spectrum is additional term

resolved (due to a finer computational grid and time with following coefficients:
step). As a result the accuracy of the calculation σ k = 0.75, σ ε = 1.15, c1ε = 1.15, c 2ε = 1.9 and
improves. c 3ε = 1.15 .
Additionally they need to be filtered. According is applied. Unsteady RANS with the standard
to the Kolmogorov theory it can be assumed that the k-ε model usually leads to a steady state solution.
dissipation rate is equal for all scaled. This leads to The vortex shedding and its unsteadiness are
ε = εˆ (5) suppressed by the too diffusive turbulence model.
More sophisticated turbulence model i.e. extended
It is not acceptable for turbulent kinetic energy. k-ε of Chen and Kim is less diffusive and therefore
Therefore it need filtering vortex shedding is gained. Further simulation with
VLES method provides slightly improved results. In
  ∆ 
kˆ = k ⋅ 1 − f   . (6) comparison to k-ε model of Chen and Kim, it
  L  proves to be less damping in the downstream flow
As a suitable filter behind trailing edge. The comparison of these two
models is shown in Figure 7.
 0 for ∆ ≥ L
 2/3
f =  ∆ (7)
1−   for L > ∆
  L

is applied where
 u ⋅ ∆t  ∆V for 2D
∆ = α ⋅ max  with hmax = 3 (8)
 hmax  ∆V for 3D
Figure 7. Pressure distribution by vortex shedding
contains model constant α in a range from 1 to 5. behind the trailing edge, comparison of extended
Then the Kolmogorov scale L for the whole k-ε model of Chen and Kim (left) and adaptive
spectrum is given as VLES (right)
k3/ 2
L= . (9) 4. APPLICATIONS
ε
4.1. Swirling flow in diffuser and draft tube
Modelled length scales and turbulent viscosity are
VLES was used for simulation of swirling flow
kˆ 3 / 2 (10)
Lˆ = in straight diffuser and elbow draft tube with two
εˆ piers. In both cases the specific inlet velocity profile
kˆ 2 corresponds to the flow at a runner outlet under part
νˆt = c µ ⋅ (11)
εˆ load conditions. It is well known that under these
with c µ = 0.09. conditions an unsteady vortex rope is formed.
Computational grid for straight diffuser had
The filtering procedure leads to the final equations 250 000 elements. Applied inlet boundary
∂k ∂k ∂  νˆ  ∂k  ˆ
conditions can be found in [10]. For elbow draft
+U j = ν + t   + Pk − ε (12) tube two grids were used (180 000 and 1 million
∂t ∂x j ∂x j  σ k  ∂x j 
elements). Computational grid and the inlet
boundary conditions (part load operational point of
∂ε ∂ε ∂  νˆ  ∂ε 
+U j = ν + t  + 93%) for the draft tube are shown in Figure 8.
∂t ∂x j ∂x j  σ ε  ∂x j 
(13)
ε ε2  Pˆ 
c1ε Pˆk − c2ε + c3ε  k  ⋅ Pˆk
k k  k 
with the production term
) )  ∂U i ∂U j  ∂U i
Pk = ν t  +  . (14)
 ∂x j ∂x i  ∂x j
 
For more details of the model and its Figure 8. Computational grid and inlet boundary
characteristics the reader is referred to [9]. conditions for the elbow draft tube
The vortex shedding behind the trailing edge, Applying an unsteady RANS with the standard
which can be considered as a convenient test case,
k-ε model leads to a steady state solution. It forms a
shows very often difficulties when unsteady RANS
recirculation region in the centre and keeps it amplitudes are higher in the experiment although
steady. On the other hand applying the extended the frequency corresponds quite well. Fourier
k-ε of Chen and Kim small unsteady vortex forms. transformation of the calculated and measured
It is too short due to the damping character of the signal at position 1 is shown in Figure 12.
turbulence model. With VLES and adaptive
turbulence model the damping of the swirl rate is
clearly reduced and vortex rope expends
downstream. Comparison of the Chen and Kim
model and VLES on example of straight diffuser is
shown in Figure 9.

Figure 11. Comparison of the pressure distribution,

Figure 9. Vortex rope in a straight diffuser, Chen &


Kim (left), VLES (right)
In a practice elbow draft tubes are usually
installed. VLES simulation shows clearly formation
of cork-screw type vortex. In Figure 10 the flow for
one time step is shown as well as velocity Figure 12. Fourier transformation of the signals at
distribution in a cross section after the bend. position 1
Disturbed velocity field can be observed. Therefore, 4.2. Flow in pipe trifurcation
the discharge through the single daft tube channels
differs significantly. In this section the flow in a pipe trifurcation of a
water power is presented. The complete water
passage consists of the upper reservoir, channel,
surge tank, penstock, trifurcation and three turbines.
The spherical trifurcation distributes the water from
the penstock into the three pipe branches leading to
the turbines, schematically shown in Figure 13.

Figure 10. Simulation of the vortex rope in elbow


draft tube with VLES
The simulation results are compared with
experiment data. The pressure distributions at two
pints, see Figure 8 are compared and shown in
Figure 11. It can be seen that the fluctuation Figure 13. Water passage with trifurcation
During the power plant exploitation severe power It is observed in the simulation as well as in the
oscillations were encountered at the outer turbines model test.
(1 and 3). Vortex instability was discovered as a
cause of these fluctuations. The vortex is formed in
the trifurcation sphere, appearing at the top and
extending into one of the outer branches. After a
certain period it changes its behaviour and extends
into opposite outer branch. Then the vortex jumps
back again. This unstable vortex motion is not
periodic and due to its strong swirling flow
produces a very high losses. These losses reduce the
head of the turbine and consequently the power
output. For better understanding and analysing of
this flow phenomenon, a computer simulation was
performed and results were compared with available
model test measurements.
Computational grid had approximately 500 000 Figure 15. Flow inside the trifurcation – vortex
elements and is shown in Figure 14. Simulation was position in the branch 1
performed in parallel on 32 processors.

Figure 16. Flow inside the trifurcation – vortex


position in the branch 3
Due to the strong swirl at the inlet of the branch
Figure 14. Trifurcation – computational grid in which the vortex is located, the losses inside this
Simulation applying unsteady RANS with the branch are much higher compared to the other two.
standard k-ε turbulence model leads to a steady Therefore, the discharge through this branch is
state solution. The obtained vortex structure extends reduced. It is obvious that the discharges i.e. losses
through both outer branches and is fully stable. through two outer branches vary successively, while
Thus vortex swirl component is severely the discharge i.e. losses in the middle branch shows
underpredicted leading to a poor forecast of the much smaller oscillations. In the reality turbines are
losses in the outer branches. It clearly shows that located at the outlet of each branch. Therefore the
unsteady RANS is not able to predict this flow discharge variation is rather small since the flow
phenomenon. rate through the different branches is prescribed by
Applying VLES with the new adaptive the turbines. In the simulation, however, a free
turbulence model, this unstable vortex movement is outflow boundary condition is applied which leads
predicted. In Figure 15 the flow inside branch 1 at a to the higher discharge variations. For comparison
certain time step is presented. The vortex is with the experiment, loss coefficients for each
represented by an iso-pressure surface and branch are calculated and presented in Figure 17.
instantaneous streamlines. After some time, see
Figure 16, vortex ”jumps” to the opposite branch.
Since the geometry is not completely symmetric,
the vortex stays longer in branch 3 than in branch 1.
distinguishes between numerically resolved and
unresolved part of the flow.
With the help of this new model the vortex
motions in a straight diffuser, a draft tube and a pipe
trifurcation are calculated. Using classical RANS
method and common turbulence models these flow
phenomena cannot be predicted. Applying VLES
with adaptive turbulence model unsteady vortex
motions were obtained due to its less damping
character. In all simulated cases the results agree
reasonably well with measurement data.

Figure 17. Loss coefficients for the three branches REFERENCES


Model tests were carried out by ASTRÖ in Graz, 1.Chen Y.S., Kim S.W. (1987) Computation of
Austria. For more details the reader is referred to turbulent flows using an extended
[3]. As shown in Figure 18, the loss coefficients for k-ε turbulence closure model. NASA CR-
each branch were calculated from the pressure and 179204
discharge measurements. 2.Constantinescu G.S., Squires K.D. (2000)
LES and DES Investigations of Turbulent
flow over a Sphere. AIAA-2000-0540
3.Hoffmann H., Roswora R.R., Egger A. (2000)
Rectification of Marsyangdi Trifurcation.
Hydro Vision 2000 Conference Technical
Papers, HCI Publications Inc., Charlotte
4.Magnato F., Gabi M. (2000) A new adaptive
turbulence model for unsteady flow fields
in rotating machinery. Proceedings of the
Figure 18. Measured loss coefficients for all three 8th International Symposium on Transport
branches Phenomena and Dynamics of Rotating
Machinery (ISROMAC 8)
Comparing the measured loss coefficients with
5.Maihöfer M. (2002) Effiziente Verfahren zur
those gained by simulation, it can be seen that the
Berechnung dreidimensionaler Strömungen
maximum values are still underpredicted, although
mit nichtpassenden Gittern. Ph.D. thesis,
general flow tendency and quantitative prediction fit
University of Stuttgart
in with measurement data reasonably well. This
6.Maihöfer M,, Ruprecht A. (2003) A Local
underprediction of the loss coefficient is assumed to
Grid Refinement Algorithm on Modern
be primarily due to the rather coarse grid and
High-Performance Computers. Proceedings
secondly due to a strong anisotropic turbulent
of Parallel CFD 2003., Elsavier,
behaviour which cannot be accurately predicted by
Amsterdam
the turbulence model based on the eddy viscosity
7.Pope S.B. (2000) Turbulent flows, Cambridge
assumption.
University Press, Cambridge
In order to solve the oscillation problem in the
8.Ruprecht A. (1989) Finite Elemente zur
hydro power plant, it was proposed to change the
Berechnung dreidimensionaler turbulenter
shape of the trifurcation. To avoid the formation of
Strömungen in komplexen Geometrien,
the vortex, upper and lower parts of the sphere are
Ph.D. thesis, University of Stuttgart
cut off in flat plates. In the meantime this
9.Ruprecht A. (2005) Numerische
modification was made, the power oscillations
Strömungssimulation am Beispiel
disappeared and no unsteady vortex was noticed.
hydraulischer Strömungsmaschinen,
5. CONCLUSIONS Habilitation thesis, University of Stuttgart
10.Skotak A. (2000) Of the helical Vortex in the
An adaptive turbulence model for Very Large Turbine Draft Tube Modelling. Proceedings
Eddy Simulation is presented. It is based on the of the 20th IAHR Symposium on Hydraulic
extended k-ε model of Chen and Kim. Introducing a Machinery and Systems, Charlotte, USA
filtering technique the new turbulence model
11.Spalart P.R., Jou W.H., Strelets M., Allmaras
S.R. (1997) Comments on the Feasibility of
LES for Wings, and on Hybrid RANS/LES
Approach. In: Liu C., Liu Z. (ed) Advances
in DNS/LES, Greyden Press, Columbus
12.van der Vorst H.A. (1994) Recent
Developments in Hybrid CG Methods. In:
Gentzsch W, Harms U (eds) High-
Performance Computing and Networking,
vol. 2: Networking and Tools, Lecture
Notes in Computer Science, 797, Springer,
Berlin, Heidelberg, New York , pp 174–183
13.Willems W. (1997) Numerische Simulation
turbulenter Scherströmungen mit einem
Zwei-Skalen Turbulenzmodell, Ph.D.
thesis, Shaker Verlag, Aachen
14.Zienkiewicz O.C., Vilotte J.P., Toyoshima
S., Nakazawa S. (1985) Iterative method
for constrained and mixed finite
approximation. An inexpensive
improvement of FEM performance.
Comput Methods Appl Mech Eng 51:3-29

You might also like