You are on page 1of 10

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

A NUMERICAL INVESTIGATION OF THE 3D SWIRLING FLOW IN A PIPE WITH


CONSTANT DIAMETER. PART 2: TURBULENT COMPUTATION

Sebastian MUNTEAN, Senior Researcher* Ivana BUNTIĆ, PhD student


Center of Advanced Research in Institute of Fluid Mechanics and Hydraulic Machinery
Engineering Sciences University of Stuttgart
Romanian Academy - Timisoara Branch
Albert RUPRECHT, Head of Fluid Mechanics Group Romeo SUSAN-RESGA, Prof.
Institute of Fluid Mechanics and Hydraulic Machinery Department of Hydraulic Machinery

*Corresponding author: Bv Mihai Viteazu 24, 300223, Timisoara, Romania


Tel.: (+40) 256 403692, Fax: (+40) 256 403700, Email: seby@acad-tim.tm.edu.ro
Ui [m/s] filtered velocity
ABSTRACT P [Pa] averaged pressure
This paper presents a numerical investigation of the τ ij [Pa] Reynolds stresses
3D turbulent and incompressible swirling flow in a α [-] model constant
constant diameter pipe. In order to evaluate the ∆ [m] resolved length scale
influence of different turbulence models the 3D ∆ t [s] time step
2 3
computations were performed. The 3D ε [m /s ] dissipation rate
computational domain corresponds to a pipe with ν 2
[m /s] kinematic viscosity
constant diameter. Two turbulence models νt [m2/s] turbulent viscosity
implemented in FENFLOSS and one model in
Subscripts and Superscripts
FLUENT are investigated. In order to validate our
r radial direction
methodology and to assess the accuracy of the
u tangential direction
numerical results, extensive comparisons with
z axial direction
available experimental data are performed. The
vorticity and the helicity are used to analyze the
ABBREVIATIONS
flow field.
in, out inlet section, outlet section
KEYWORDS
1. INTRODUCTION
swirling flow, 3D turbulent computation, helicity
The design and construction traditionally are
NOMENCLATURE forced to rely heavily on experience, experimental
U, V, W [m/s] mean velocity components data and empirical expressions. However, such
(axial, radial and tangential) approach used to the exclusion of available
Ui [m/s] averaged velocity computational or analytical techniques would soon
prove both expensive and incapable of assimilating
R [m] radius
the vast amount of design information. Thus,
D = 2R [m] diameter
computer modeling is becoming increasingly
L [m] lenght
attractive as a complementary and supplementary
g [m/s2] gravity
tool to aid design process.
k [m2/s2] turbulent kinetic energy
A research task emphasized in both modeling and
f [-] filter function
experimental communities in recent years is that of
hmax [m] local grid size
swirling flow. A huge amount of work concerning
k [m2/s2] turbulent kinetic energy
the prediction of turbulent flows is reported in the
L [m] Kolmogorov length scale
literature. However, very few is directly applicable
Pk [-] production term
to the problem of the relaxation of disturbance in
u [m/s] local velocity
pipe flow. Contrary to the majority of the work for
engineering purposes, the geometry of this problem measurements are available at follows: 4.3, 7.7,
is very simple. However, the accuracy requirement 11.5, 21.3, 28.1, 34.9, 48.5, 75.6 (see Figure 1,
for the prediction of the pipe flow is stronger than [11]). The maximum allowed deviation of the mean
the commonly required accuracy for engineering velocity components from the nominal values
purposes. Since the numerical prediction of is/was ±1%. All data are available in non-
turbulent flows involves modeling of certain aspects dimensional form: mean velocities are scaled with
of the flow, this modeling has to be considered very the bulk velocity, Reynolds stresses are scaled with
carefully for its effect on the accurate prediction of the bulk velocity squared, radial positions
the swirling flows. (measured from the pipe axis) are scaled with the
The purpose of this paper is to discuss the pipe radius R, while the axial positions of the
application of various turbulence models in the measurement planes are scaled with the pipe
prediction of confined, swirling flows. Evaluation diameter D. The data are expressed in a cylindrical
of the various turbulence models implemented in coordinate system, with the velocity components in
FENFLOSS and FLUENT codes was performed by the axial, circumferential and radial direction
comparison of the numerical predictions with denoted by U, W and V, respectively.
experimental data. The turbulent computation of a The measurements were performed in a
swirling flow in a pipe is investigated. Additionally, hydraulically smooth pipe with diameter D=70
the boundary layer grid refinement and turbulence [mm] for the concentrated vortex (CV) with initial
models are evaluated. The objective of this study is swirl intensity S0 ≈ 0.18. All measurement series
to evaluate the capability of turbulence models to were performed for Reynolds number of 50.000 and
simulate this type of flow. The preliminary results 300.000. The data were obtained with a 2-
are included in this paper due to the fact that the component laser-Doppler system. The total velocity
turbulent computations are extremely time vector and Reynolds stress tensor were acquired by
consuming (at least one order of magnitude more performing three measurements in each point, with
time consuming than inviscid computations). the laser-Doppler system aligned under three
The paper is organized as follows. Section 2 different angles within the azimuthal plane, i.e. the
presents the experimental database. Sections 3 and 4 plane perpendicular to the pipe axis.
present the computational domains, equations and
boundary conditions when a turbulent computation 3. COMPUTATIONAL DOMAIN
is considered. In Section 5 and 6 the numerical The 3D computational domain is a pipe with the
methods, turbulence models and numerical results constant diameter D=0.32 m in order to ensure a
obtained with FENFLOSS code and FLUENT Reynolds number of 300.000 when the bulk
package are presented, respectively. In order to velocity U=0.9375 [m/s] and water kinematics
validate our methodology and to assess the accuracy viscosity are considered. The first section of the
of the numerical results, extensive comparisons measurements is used as the inlet section. Based on
with available experimental data are performed. the 3D Euler numerical investigations performed by
Next, the numerical results are analyzed and the Muntean et al. [11] the following parameters and
helicity quantity is used to investigate the flow boundary conditions are chosen: a) for FLUENT
field. The paper conclusions, as well as the computations the domain length L=15D and the
perspectives for this work are summarized in the radial equilibrium condition at the outlet section; b)
last section. for FENFLOSS the computational length L=30D.

2. EXPERIMENTAL DATA BASE 4. EQUATIONS AND BOUNDARY


The experimental data used for evaluation of the CONDITIONS
computation in this work are provided by Dr.
Wiendelt Steenbergen, Eindhoven University of 4.1. Equations
Technology. The measurements are included in his In this work an incompressible fluid with
PhD Thesis [23], in which all details of constant properties is considered. The governing
experimental setup can be found. The experimental equations describing this incompressible, viscous
data are also included in the ERCOFTAC database and time dependant flow are the Navier-Stockes
(Test case 72: Turbulent Pipe Flow with Swirl). equations. They express the conservation of mass
The following experimental data are available: and momentum. In the RANS approach, the same
the distributions of mean velocities and Reynolds equations are time or ensemble averaged leading to
stresses. Eight measurement sections of the well known RANS equations:
∂U i ∂U i ∂P ∂τ ij 1.5
+U j =− + ν∇ 2U i − (1)
∂t ∂x j ∂x i ∂x j
1
∂U i
=0
∂x i U
(2) 0.5 W

U, W
In RANS τij expresses the Reynolds stress tensor
which is unknown and has to be modelled. The task 0

of turbulence modelling is in the formulation and


determination of suitable relations for Reynolds −0.5
stresses.
It is generally believed that swirling flows cannot −1
be predicted accurately by a turbulence model using 0 0.2 0.4
r/R [−]
0.6 0.8 1

a scalar eddy viscosity, for fundamental reasons.


Most of the evidence for this failure of scalar eddy Figure 1. The radial distribution of the axial U (●)
viscosity models is collected either in shear flows and circumferential W (■) mean velocities imposed
with a longitudinal streamline curvature, or in at the inlet section according to the experimental
strongly swirling flows. The experimental results data at Re=3ooooo, [23].
obtained by Dr. Steenbergen [23] show that the 0.2

physical shortcomings of the scalar eddy viscosity


concept also apply to the flows containing a weak
0.15
swirl. It appears that the additional production of
the tangential shear stress vw induced by streamline
uu, vv, ww

curvature, is still significant at low swirl numbers. 0.1 uu


The observed development of the circumferential
axis

wall
vv
velocity distribution towards almost linear or ww

positively curved shapes even suggests, that the 0.05


adequacy of any eddy-viscosity model, either scalar
or tensorial, decreases with a decreasing swirl
number. The reason for the latter seems to be rooted 0
0 0.2 0.4 0.6 0.8 1
in the large contribution of diffusive radial transport r/R [−]
rather than in curvature induced production of vw,
Figure 2a. The radial distribution of the Reynolds
[23].
stresses uu (●), vv (■) and ww (♦) imposed at the
For these reasons, it is plausible that accurate
inlet section according to the experimental data,
predictions of the velocity field can only be realized
[23].
with models in which the dynamics of the Reynolds
stress is as much as possible represented. 0.1

Consequently, the RSM model available in


FLUENT commercial code is investigated, [3], as
well as two models implemented in FENFLOSS 0.05
code, [8], [9].
uv, vw, uw

uv
axis

wall

4.2. Boundary conditions vw


uw
At the inlet section the velocity field (three
0
velocity components) and the turbulent quantities
are imposed.

−0.05
0 0.2 0.4 0.6 0.8 1
r/R [−]

Figure 2b. The radial distribution of the Reynolds


stresses uv (×), vw (*) and uw (▲) imposed at the
inlet section according to the experimental data,
[23].
In FLUENT computations, the individual approximation of 2nd order. For the velocity
Reynolds stresses are imposed according to the components and the turbulence quantities a trilinear
measurements [16], [23], see Figures 2a-b. approximation is applied. The pressure is assumed
Additionally, the turbulence dissipation rate ε is to be constant within element. For advection
required. The radial equilibrium condition is dominated flow a Petrov-Galerkin formulation of
imposed at the outlet section displaced at 15D 2nd order with skewed upwind orientated weighting
downstream from the inlet section while on the wall function is used.
a no-slip condition is considered.
In FENFLOSS computations, the k and ε values
are required to be imposed at the inlet section.
Accordingly, these quantities are computed using
the following formulae:

k=
1 2
2
(
u + v 2 + w2 . ) (3)

and
3

ε = Cµ ( ) 3
4
k2
D
. (4)

with Cµ = 0.09 and D = 0.32 m . Figure 3 presents


the distribution of turbulent quantities k and ε
computed using the Reynolds stresses. At the outlet
section the constant pressure is imposed.
0.15

0.1

Figure 4. FENFLOSS flow chart


axis
k, eps

wall

k
eps
For the solution of the momentum and continuity
0.05 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
0
0 0.2 0.4 0.6 0.8 1 gradient method BICGSTAB2 of van der Vorst [24]
r/R [−] with an incomplete LU decomposition (ILU) for
Figure 3. The radial distribution of the turbulent preconditioning. The pressure is treated with the
kinetic energy (●) and the dissipation rate of modified Uzawa pressure correction scheme [26].
turbulence (■) computed using experimental data The pressure correction is performed in a local
with equations (3) and (4). iteration loop without reassembling the system
matrices until the continuity error is reduced to a
5. FENFLOSS NUMERICAL METHOD given order.
After solving the momentum and continuity
The computations are performed using the equations, the turbulence quantities are calculated
program FENFLOSS (Finite Element based and a new turbulence viscosity is gained. The k and
Numerical FLOw Simulation System) which is ε - equations are also linearised and solved with
developed at the Institute of Fluid Mechanics and BICGSTAB2 algorithm with ILU preconditioning.
Hydraulic Machinery, University of Stuttgart. The whole procedure is carried out in a global
It is based on the Finite Element Method. For iteration until convergence is obtained. For
spatial domain discretisation 8-node hexahedral unsteady simulation the global iteration has to be
elements are used. Time discretisation involves a
three-level fully implicit finite difference
performed for each time step. FENFLOSS flow Simulations and their main aim is to overcome the
chart is shown in Figure 4. computational costs and capacity problems. These
The code is parallelised and computational methods try to keep computational efficiency of
domain is decomposed using double overlapping RANS and the potential of LES to resolve large
grids. In that case the linear solver BICGSTAB2 has turbulent structures. Although they can be
a parallel performance and the data exchange performed on coarser grids, the simulations are
between the domains is organised on the level of the strongly dependent on the modelling.
matrix - vector multiplication. The preconditioning Above mentioned methods slightly differ in
is then local on the each domain. The data exchange filtering techniques, applied model and
uses MIP (Message Passing Interface) on the interpretation of the resolved motion, but broadly
computers with distributed memory. On the shared speaking they all have a tendency to solve complex
memory computers the code applies OpenMP. For unsteady turbulent flows at high Reynolds number
more details on the numerical procedure and implying a principle “solve less – model more”, see
parallelisation the reader is referred to [7], [8]. Figure 5 and Table 1. It means that the relevant part
of the flow (unsteadiness) is resolved and the rest is
5.1. Very Large Eddy Simulation
modelled.
Lately several hybrid methods are proposed in the
literature: 5.2. Adaptive turbulence model
Very Large Eddy Simulation (VLES) Classical turbulence models, which are usually
Semi-Deterministic Simulations (SDS) applied for solving engineering flow problems,
Coherent Structure Capturing (CSC) model the whole turbulent spectrum. They show
Detached Eddy Simulation (DES) excessive viscous behaviour and very often damp
Hybrid RANS/LES down unsteady motion quite early. Therefore they
Limited Numerical Scales (LNS) are not completely successful for some flow cases.
VLES is used for resolving at least one part of
turbulence spectrum and thus getting more precise
picture of the flow behaviour. Depending on the
type of the flow and grid size applied model should
automatically adjust to one of the modelling
approaches schematically shown in Figure 6.

Figure 5. Modelling approach in VLES

Table 1. Resolution in DNS, LES and VLES [16]


Model Resolution

Direct numerical All turbulent scales


simulation (DNS) are resolved
Large eddy Grid size and filtering Figure 6. Adjustment for adaptive model
simulation with are sufficient to Therefore an adaptive model is developed. Its
near-wall resolve 80% of the advantage is that with increasing computer power it
resolution energy can be afforded that a larger part of spectrum is
Large eddy Grid size and filtering resolved (due to a finer computational grid). As a
simulation with are sufficient to result the accuracy of the calculation improves.
near-wall resolve 80% of the For distinguishing resolved and modelled
modelling energy distant from turbulence spectrum (see Figure 7), the adaptive
the wall, but not in model uses a filtering technique. There are several
the near-wall region of them described in the literature [6], but the
Very large eddy Grid size and filtering applied technique is similar to Willems [25]. The
simulation (VLES) are not sufficiently smallest resolved length scale ∆ used in filter is
fine to resolve 80% according to Magnato and Gabi [6] dependant on
of the energy the local grid size or the computational time step
All of them are based on the same idea to and local velocity.
represent a link between RANS and LES. Generally
they all can be classified as Very Large Eddy
k3/ 2
L= . (11)
ε
Modelled length scales and turbulent viscosity are
kˆ 3 / 2 (12)
Lˆ =
εˆ
kˆ 2 (13)
νˆt = c µ ⋅
εˆ
with c µ = 0.09.
Figure 7. Distinguishing of turbulence spectrum by The filtering procedure leads to the final equations
VLES
∂k ∂k ∂  νˆ  ∂k  ˆ
The basis of the adaptive model is the k-ε model +U j = ν + t 
 ∂x  + Pk − ε (14)
of Chen and Kim [1]. It is chosen due to its ∂t ∂x j ∂x j  σk  j 
simplicity and capacity to better handle unsteady
∂ε ∂ε ∂  νˆ  ∂ε 
flows. Its transport equations for k and ε are given +U j = ν + t  +
as ∂t ∂x j ∂x j  σ ε  ∂x j 

  ∂k  ε ˆ ε2  Pˆ 
∂k ∂k ∂ ν c1ε Pk − c2ε + c3ε  k  ⋅ Pˆk (15)
+U j = ν + t 
 ∂x  + Pk − ε (5) k k
∂t ∂x j ∂x j  σk  j   k 
with the production term
∂ε ∂ε ∂  ν t  ∂ε 
+U j = ν + 
 + ) )  ∂U i ∂U j  ∂U i
∂t ∂x j ∂x j  σ ε  ∂x j 
 Pk = ν t  +  . (16)
 ∂x j ∂x i  ∂x j
2  
ε ε P 
c1ε Pk − c 2ε + c 3ε  k  ⋅ Pk (6)
k k  k 43
4 For more details of the model and its characteristics
142 4
additional term the reader is referred to [20].
with following coefficients: 5.3. FENFLOSS numerical results
σ k = 0.75, σ ε = 1.15, c1ε = 1.15 , c2ε = 1.9 and Hexahedral mesh with 390k elements was used
c 3ε = 1.15 . with refinements close to the symmetry axis and
Additionally they need to be filtered. According to near the wall. Dimensionless y+ value for the wall
the Kolmogorov theory it can be assumed that the function is about 40.
dissipation rate is equal for all scaled. This leads to Figure 8 shows the comparison between
computed axial and circumferential mean velocities
ε = εˆ (7) computed with VLES and Kim-Chen models and
It is not acceptable for turbulent kinetic energy. experimental data [23]. It can be seen that the
Therefore it is need filtering experimental data are inadequately predicted with
Kim-Chen model. The main conclusion is that the
  ∆  Kim-Chen model is still dissipative even with its
kˆ = k ⋅ 1 − f   . (8)
  L  additional production term (eg. (6)). The main
deficiency is apparent in relation to the two
As a suitable filter equations model. The problem is the isotropic
 0 for ∆ ≥ L character of the turbulent viscosity, or alternatively,
 2/3 an invariant coefficient in the equation for the eddy
f =  ∆ (9)
1−   for L>∆ viscosity. The isotropic eddy viscosity concept is
  L

not suitable for complex flows or turbulent flows
is applied where which are influenced by body forces acting in a
preferred direction, such as swirling flows.
 u ⋅ ∆t  ∆V for 2D
∆ = α ⋅ max  with hmax = 3 (10)
 hmax  ∆V for 3D

contains model constant α in a range from 1 to 5.


Then the Kolmogorov scale L for the whole
spectrum is given as
1.25
viscous near to the wall. In all cases investigated the
1
core dimensionless variable for wall function is chosen
smaller than 15 (y+<15).
0.75 Based on our numerical investigations [11] the
0.5
<U> Re=3e5 [Ste95] following solver parameters are selected: the time-
<U>, <W> [−]

advancing scheme is 2nd-order implicit; the


<W> Re=3e5 [Ste95]
VLES

momentum equation is discretized with a 2nd-order


Kim−Chen
0.25

upwind scheme; the pressure discretization uses a


0
PREssure Staggering Option (PRESTO) scheme,
−0.25 with a Pressure-Implicit with Splitting of Operators
boundary
layer boundary (PISO) pressure-velocity coupling.
−0.5
layer
S1 6.1. RSM Approach
−0.75
−1 −0.75 −0.5 −0.25 0 0.25 0.5 0.75 1 The Reynolds stress model (RSM) is the most
r/R [−]
elaborated turbulence model that FLUENT code
1.25
viscous provides [3]. Abandoning the isotropic eddy-
core
1 viscosity hypothesis, the RSM closes the Reynolds-
averaged Navier-Stokes equations by solving
0.75
transport equations for the Reynolds stresses,
<U> Re=3e5 [Ste95]
0.5 <W> Re=3e5 [Ste95] together with an equation for the dissipation rate.
<U>, <W> [−]

VLES
Kim−Chen This means that seven additional transport equations
0.25 must be solved in 3D problems. Since the RSM
0
accounts for the effects of swirl and rapid changes
in strain rate in a more rigorous manner than one-
−0.25
boundary
equation and two-equation models, it has greater
layer potential to give accurate predictions for complex
−0.5
S2 boundary
layer flows. However, the fidelity of RSM predictions is
−0.75 still limited by the closure assumptions employed to
−1 −0.75 −0.5 −0.25 0 0.25 0.5 0.75 1
r/R [−] model various terms in the exact transport equations
Figure 8. The radial distribution of the axial <U> for the Reynolds stresses. The modeling of the
(●) and circumferential <W> (■) mean velocities in pressure-strain and dissipation-rate terms is
sections S1 (above) and S2 (below), respectively. particularly challenging, and often considered to be
Comparison experimental data [23] against 3D responsible for compromising the accuracy of RSM
unsteady numerical results computed with two predictions.
turbulent models implemented in FENFLOSS code: The Reynolds stress model involves calculation
── VLES model and ─ ─ Kim-Chen model. of the individual Reynolds stresses, using
differential transport equations. The individual
Alternatively, the flow field computed with Reynolds stresses are then used to obtain closure of
VLES model predicts the measurements better than the Reynolds-averaged momentum eq (1). The exact
the Kim-Chen model. However, the tangential form of the Reynolds stress transport equations may
velocity component seems to be underestimated be derived by taking moments of the exact
while the axial component is overestimated. A momentum equation. This is a process wherein the
possible explanation for this high discrepancy is exact momentum equations are multiplied by a
that the grid is still too coarse for VLES. fluctuating property, the product then being
Consequently, further investigations with fine grid Reynolds-averaged. Unfortunately, several of the
(especially in axial direction) are needed in order to terms in the exact equation are unknown and
compute more accurately the swirling flow with modeling assumptions are required in order to close
both Kim-Chen and VLES model. the equations. More details of the RSM and its
characteristics are referred to [3].
6. FLUENT NUMERICAL METHOD Compared with the two equations models, the
The 3D unsteady turbulent flow simulations are RSM requires additional memory and CPU time
performed with the FLUENT 6.0.12 commercial due to the increased number of the transport
software [3], using the hexahedral mesh generated equations for Reynolds stresses. On average, the
with Gambit [4]. The grid has 245k cells, and it was RSM in FLUENT requires 50-60% more CPU time
refined in the symmetry axis neighbourhood and
per iteration compared to the two equations models. the Euler computations. In other hand, further
Furthermore, 15-20% more memory is needed. investigations are needed in order to compute more
accurately the turbulent swirling flows.
6.2. FLUENT numerical results
A good agreement between numerical results 7. NUMERICAL RESULTS ANALYSIS
computed with RSM and experimental data is
In order to understand better the physics of
obtained in free vortex region, see Figure 9.
1.25 swirling flow a deeply analysis is performed.
viscous
core
First, the vorticity field is computed in order to
1
compare the numerical results against theoretical
0.75 results fitted with so called Q-vortex, [11].
<U> Re=3e5 [Ste95]
Vorticity is defined mathematically as,
0.5
<U>, <W> [−]

<W> Re=3e5 [Ste95]


inviscid
RSM
ω = ∇ ×U . (17)
0.25
and informally as the local component of rotation in
0
the flow.
−0.25 70 viscous
boundary boundary
core
layer layer
boundary
−0.5 60
layer
S1
−0.75 50
−1 −0.75 −0.5 −0.25 0 0.25 0.5 0.75 1
r/R [−]

vorticity
1.25 40
viscous
core
1 30

0.75 20
<U> Re=3e5 [Ste95]
0.5 <W> Re=3e5 [Ste95] 10
<U>, <W> [−]

inviscid
RSM
0.25 0
−0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r/R [−]
0
Figure 10. The radial distribution of the vorticity
−0.25
boundary (●) at the inlet section.
layer
−0.5
S2 boundary
layer
Figure 10 shows the radial distribution of
−0.75 vorticity computed at the inlet section. The Rankine
−1 −0.75 −0.5 −0.25 0 0.25 0.5 0.75 1
r/R [−] vortex is built using the asymptotic behavior of the
Figure 9. The radial distribution of the axial <U> Burgers' vortex for large and small radius with
(●) and circumferential <W> (■) mean velocities in respect to the vortex characteristic radius Rc . The
sections S1 (above) and S2 (below), respectively. vortex characteristic radius Rc = 0.177 (see Table
Comparison experimental data [23] against 3D 1, [11]) marked with black solid line in Figure 10
unsteady numerical results computed with FLUENT represents the hypothetical border between viscous
code: ── RSM and ─ ─ Euler model. core region and free vortex region, respectively. As
Contrary, a discrepancy between RSM results and a result, this border suggests the limit where the
measurement is observed in viscous core region. local component of flow rotation is negligible.
Moreover, the tangential velocity component is However, the vorticity value at the vortex
evaluated using a numerical scheme too dissipative characteristics radius is still significant. In
whilst the axial component is contrary predicted. consequence of this observation the vorticity
The same conclusion is emerged from numerical remains significant in the transition region between
investigations obtained with VLES model. the two regions. Therefore, the Euler model
However, the numerical results computed with presents small discrepancies refer to the
Euler model predict the measurements better than measurements in the transition region.
the RSM approach, see Figure 8. Additionally, the Next, the helicity is defined [9] as
Euler computation requires less CPU time and H = ω ⋅ U = (∇ × U ) ⋅ U . (18)
memory than the RSM with one order of
magnitude. Actually, for engineering applications The helicity of a fluid particle is defined as dot
with complex geometries we strongly recommend product of vorticity and velocity vector. The helicity
is a scalar quantity and changes its sign due to the 8. CONCLUSIONS
change from a right-handed to a left-handed frame This paper presents a methodology for computing
of reference. It is therefore important to specify the the 3D incompressible and turbulent swirling flow
used frame, [10]. Right-handed cylindrical in a pipe with constant diameter.
coordinate system is used in our case. It is well For prediction of the confined, swirling flow
known that the dot product is zero only if the different turbulence models are investigated. Both
vectors are orthogonal one to another or one of turbulence models implemented in FENFLOSS
them is zero. code predicts inadequately the experimental data.
5
viscous However, the VLES model offers a better solution
core
than the Kim-Chen model. The RSM implemented
0
in FLUENT lead to the significantly better
numerical results using turbulence models. Further
−5 investigations are required to improve the numerical
helicity

results computed with all turbulence models.


−10 Finally, the vorticity and the helicity quantities
are used to evaluate the velocity field.
−15
boundary ACKNOWLEDGMENTS
layer

−20
This work has been supported by the EC-funded
−0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 project HPC-Europa contract number 506079 and
r/R [−]
Romanian National University Research Council
Figure 11. The radial distribution of the helicity (●) Grant (CNCSIS) 33/2005. Numerical computations
at the inlet section. were performed at the High Performance
Computing Center (HLRS), University of Stuttgart.
The first author would like to thank Institute of
Fluid Mechanics and Hydraulic Machinery from
University of Stuttgart where he performed this
work as Visiting Researcher in period January-
March 2005. Special gratitude to Dr. Wiendelt
Steenbergen for his assistance with the experimental
database.
Figure 12. The helical filaments generated in the
viscous core region. REFERENCES
The helicity distribution at the inlet section is 1. Chen Y.S., Kim S.W. (1987) Computation
presented in Figure 11. The values of helicity are of turbulent flows using an extended k-
maxima in viscous core region, especially on the ε turbulence closure model. NASA CR-
axis. In this case, the fluid particle injected on the 179204
axis has the maximum trajectory while the 2. Dahlhaug O.G., (1997) A study of swirl
trajectories of the fluid particles injected in viscous flow in draft tubes, PhD Thesis, Norwegian
core region at larger radii are shorter and shorter in University of Science and Technology,
axial direction, see Figure 12. Contrary, the Trondheim, Norway. Report ITEV:1997:05
trajectories of the fluid particles injected in free 3. Fluent Inc. (2001) FLUENT 6. User’s
vortex region at larger radii are longer and longer in Guide, Fluent Incorporated, Lebanon USA.
axial direction, see Figure 13. 4. Fluent Inc. (2001) Gambit 2. User’s Guide,
Fluent Incorporated, Lebanon USA.
5. Kito O., (1991) Experimental Study of
Turbulent Swirling Flow in a Straight Pipe,
J. Fluid Mech., 225:445-479.
6. Magnato F., Gabi M. (2000) A new
adaptive turbulence model for unsteady
flow fields in rotating machinery.
Proceedings of the 8th International
Figure 13. The helical filaments generated in the Symposium on Transport Phenomena and
free vortex (potential) region.
Dynamics of Rotating Machinery Ph.D. thesis, University of Stuttgart,
(ISROMAC 8) Germany.
7. Maihöfer M., Ruprecht A. (2003) A Local 20. Ruprecht A. (2005) Numerische
Grid Refinement Algorithm on Modern Strömungssimulation am Beispiel
High-Performance Computers. Proceedings hydraulischer Strömungsmaschinen,
of Parallel CFD 2003, Elsevier, Habilitation thesis, University of Stuttgart,
Amsterdam. Germany.
8. Maihöfer M. (2002) Effiziente Verfahren 21. Sloan D.G., Smith P.J., Smoot D., (1986)
zur Berechnung dreidimensionaler Modeling of swirl in turbulent flow
Strömungen mit nichtpassenden Gittern. systems, Prog. Energy Combust. Sci.,
Ph.D. thesis, University of Stuttgart, 12:163-250.
Germany. 22. Speziale G.C., (1991) Analytical Methods
9. Moffatt H.K., (1969) The degree of for the Development of Reynolds-Stress
knottedness of tangled vortex lines, J. Fluid Closure in Turbulence, Annu. Rev. Fluid.
Mech., 44:705-719. Mech. 23:107-157.
10. Moffatt H.K., Tsinober A., (1992) Helicity 23. Steenbergen W., (1995) Turbulent Pipe
and Laminar and Turbulent Flow, Annu. Flow with Swirl, PhD Thesis, Eindhoven
Rev. Fluid. Mech., 24:281-312. University of Technology, Eindhoven, The
11. Muntean S., Ruprecht A., Resiga R., (2005) Netherlands.
A Numerical Investigation of the 3D 24. van der Vorst H.A. (1994) Recent
Swirling Flow in a pipe with Constant Developments in Hybrid CG Methods. In:
Diameter. Part 1: Inviscid Computation. In Gentzsch W., Harms U. (eds) High-
Proceedings of the Workshop on Vortex Performance Computing and Networking,
Dominated Flows - Achievements and vol. 2: Networking and Tools, Lecture
Open Problems, June 10-11 2005, Notes in Computer Science, 797, Springer,
Timisoara, Romania. (submitted) Berlin-Heidelberg-New York, pp.174–183.
12. Nallasamy M., (1987) Turbulence Models 25. Willems W. (1997) Numerische Simulation
and their Appliactions to the Prediction of turbulenter Scherströmungen mit einem
Internal Flows: A Review, Computer and Zwei-Skalen Turbulenzmodell, Ph.D.
Fluids, 15(2):151-194. thesis, Shaker Verlag, Aachen, Germany.
13. Parchen R. R., (1998) Steenbergen W., 26. Zienkiewicz O.C., Vilotte J.P., Toyoshima
Experimental and Numerical Study of S., Nakazawa S. (1985) Iterative method
Turbulent Swirling Pipe Flows, J. Fluid for constrained and mixed finite
Eng., 120(1):54-61. approximation. An inexpensive
14. Parchen R.R., (1993) Decay of Swirl in improvement of FEM performance.
Turbulent Pipe Flows, PhD Thesis, Comput. Methods Appl. Mech. Eng., 51:3-
Eindhoven University of Technology, 29.
Eindhoven, The Netherlands.
15. Patankar S.V., (1980) Numerical Heat
Transfer and Fluid Flow, Hemisphere,
Washington, D.C., USA.
16. Pope S.B. (2000) Turbulent flows,
Cambridge University Press, Cambridge
17. Rhie C.M., Chow. W.L., (1983) Numerical
Study of the Turbulent Flow Past an Airfoil
with Trailing Edge Separation, AIAA
Journal, 21(11):1525-1532.
18. Rodi W., (1976) a New Algebraic Relation
for Calculating the Reynolds Stresses,
ZAMM 56:219-221.
19. Ruprecht A. (1989) Finite Elemente zur
Berechnung dreidimensionaler turbulenter
Strömungen in komplexen Geometrien,

You might also like