RANS Calculation of Three-Dimensional Turbulent Swirling Flow

in a Short Pipe

Ms Jinli Song

Mechanical Engineering Department, Khalifa University of Science and Technology, Abu Dhabi,

United Arab Emirates

Email: [email protected]

Prof. Lyes Khezzar (Corresponding author)

Mechanical Engineering Department, Khalifa University of Science and Technology, Abu Dhabi,

United Arab Emirates

Email: [email protected]

Dr. Nabil Kharoua

Ecole Polytechnique de Constantine, Constantine, Algeria ,

Email: [email protected]

Dr. Mohamed Alshehhi

Mechanical Engineering Department, Khalifa University of Science and Technology, Abu Dhabi,

United Arab Emirates

Email: [email protected]

1
Abstract

Turbulent swirling flow inside a short pipe and interacting with a conical bluff body is simulated

using the commercial CFD code Fluent. The geometry used is akin to a liquid gas separators that

use swirl to initiate separation. Three turbulence models belonging to the Reynolds averaged

Navier-Stokes (RANS) equations frameworks are used, RNG k-ε, SST k-ω and the full Reynolds

stress model (RSM) in their steady and unsteady versions. The performance of the turbulence

models is assessed with particular focus on the RSM against experimental data and published

Large Eddy Simulation results using a full three-dimensional mesh of industrial scale. Steady and

unsteady RSM simulations show similar behavior and give comparatively the best predictions of

mean velocity profiles among the three RANS turbulence models but result in poor agreement on

turbulence field with experimental results in the core region.

The influence of Reynolds number on velocity profiles, swirl decay, and wall pressure on the bluff

body are also presented. In the case of Reynolds numbers that generate a combined Rankine

velocity profiles, the width and magnitude of flow reversal zone decreases along the pipe axis

direction and disappears downstream if the Reynolds number is small. The peak tangential

velocity increases with increasing Reynolds number. The swirl decay rate follows closely an

exponential form with decay rates that follow similar previously observed trends.

Key words: Swirling pipe flow; Conical bluff body; CFD; RANS

2
1. Introduction

Confined turbulent swirling flows are complex, highly three-dimensional, unsteady with strong

streamline curvature and turbulence anisotropy where the rotational component of the mean strain

rate plays an important. These flows take place in a broad range of industrial applications

including fluid phase separation in cyclone separators Rosa, et al. [1], heat transfer rates

enhancement in heat exchangers [2] and enhancing mixing in addition to flame stabilization in

combustion burners [3]. In view of this, confined turbulent swirling flows have received attention

and have been studied experimentally and numerically. However on the simulation side, the

calculation of such flows with turbulence modelling within the Reynolds averaged Navier-Stokes

(RANS) equations framework still remains a challenge. This work represents a contribution

towards assessing such models for a three-dimensional turbulent swirling pipe flow which

encounters a blockage of conical shape. This particular geometry possesses features of swirling

pipe flow and those that pertain to liquid-gas in-line separators [4] with a novel type of swirler

consisting of multiple radial cylindrical ports inside a hollow cylinder.

Confined turbulent swirling flows and their evolution in long pipes were studied by Kitoh [6];

Steenbergen [6]; Moene [7], Steenbergen and Voskamp [8] Pashtrapanka et al. [9] where swirl

was generated by stationary swirlers or rotating pipes. Studies of flows in the near field past

expansions include those of Dellenback et al. [10], Khezzar [11] and Mak and Balabani [12]. In

their study Pashtrapanka et al. [9] found that swirling flow in a pipe can loose its axial symmetry

as it progresses in the downstream direction and suggested that calculations for swirling pipe flow

should be done with three-dimensional model to capture the features of the flow.

3
Numerical simulations of turbulent swirling pipe flows have been conducted by several authors

using a variety of RANS turbulence models that ranged from the two-equation family to the full

Reynolds stress models. The encompassing study of Jakirlic et al. [13] used the standard k- model

and its low Reynolds number extension and three variants of the full Reynolds stress model (RSM),

the basic high Reynolds number version, the model of Speziale et al. [14] and a low Reynolds

number version of the RSM. This particular study permitted to highlight several important points:

(i) the standard k-e model fails to replicate important features of the flow resulting in solid body

rotation, this was also observed by several other authors previously such as Kobayachi and Yoda

[15] and Escue and Cui [16], (ii) The low-Reynolds number version of the RSM was superior since

it is able to mimic stress anisotropy in the near wall region, (iii) wall function approach is

unsuitable for strong swirling flows driven by near wall phenomena such as in rotating pipes or if

transition is present (iv) whereas if the extra strain rates responsible for non-equilibrium effects

arise from the inner part of the flow as in swirling flow entering pipes and if the bulk Reynolds

number is high the wall function concept can provide a reasonable alternative for modeling the

near wall region. Brennan [17] in his study of swirling flow in cyclonic separators found that the

RSM model with its linear and quadratic pressure strain correlations predict almost identical

velocity profiles. Erdal and Shirazi [18] simulated 3D swirling flow with different turbulent

models in a cylindrical cyclone. The simulations captured the general trend of the experimental

data. Simulation with standard k-ε model predicts higher rotational flow and RSM predicts a much

stronger decay of tangential and axial velocities. Wegner, et al. [3] evaluated the performance of

Unsteady Reynolds Averaged Navier-Stokes (URANS) simulations in predicting an unconfined

swirling flow with Precessing Vortex Core (PVC). The swirl number was 0.75 and Reynolds

number ranging from 10,000 and 42,000. The URANS results show good agreement of mean

4
velocities with experimental data. Ramirez and Cortes [19] simulated an unsteady single-phase

turbulent flow in a swirl combustor using the standard k–ε and Reynolds stresses models. Both

models predict a complex asymmetric flow with PVC and inner recirculation zone. The prediction

via RSM was more realistic, for which the flow lacks exact symmetry and periodicity, and exhibits

more stronger and persistent vortical motions. Large eddy simulation (LES) was used by Kharoua

et al. [4] for the same geometry used in the present study in the presence and absence of the bluff

cone. LES was able to predict the presence of the peak in the r.m.s of the axial and tangential

velocity components

Advanced turbulence models such as Large eddy simulation (LES) or direct numerical simulation

(DNS) can perform better than RANS model for such flows but remain prohibitively expensive

especially so for industrial type configurations which have usually flow passages of complex three-

dimensional nature, see [4]. Complex three-dimensional geometries such as encountered in

industrial applications will be demanding in terms of meshing and the turn-around time for design

and what-if scenarios simulations is usually excessively large.

The present work involves the simulation of turbulent confined swirling flow inside a short pipe

which encounters a downstream blockage of conical shape. In contrast to many of the previous

numerical work published on swirling pipe flow, the geometry considered in the simulation is

three-dimensional so that it will be able to capture the inherent features of the flow. In order to

obtain meaningful simulations, the complex three-dimensional swirler geometry is also included

in the simulation domain. The work is an extension of Kharoua et al. [4] and aims to evaluate the

performance of mainly the RSM turbulence model with linear pressure strain correlation and non-

5
equilibrium wall function which takes into account the influence of the pressure gradients on the

distortion of the velocity distributions. While the main focus is on the RSM model, other RANS

two-equation turbulence models are also used for comparison. Thus, the RNG k-ε with its options

of eddy viscosity and swirl dominated flow and enhanced wall treatment with pressure gradient

effects and SST k-ω with low-Re and curvature corrections are used in this study. While similar

studies using versions of the k- and RSM are numerous, this is not the case for the SST k- model.

Numerical results were benchmarked with experimental data obtained using laser Doppler

Anemometry. Since the geometry is extracted from that of a two-phase separator additional

parametric simulations are conducted to study the influence of Reynolds number on flow and swirl

decay. This is meant to be a first step to evaluate the ability of CFD to mimic the flow features of

single phase swirling flow which interacts with a bluff body before extending the work later on to

two-phase flow. It is to be noted that numerical studies on confined turbulent swirling flows

interacting with a bluff body are relatively rare in the literature [20, 21]. So this study is a

contribution towards understanding of their features. when the shapes of bluff body are the disks

Huang and Tsai [20] and spheres Atvars et al. [21]).

Section 2 of this report presents the simulation approach used, while the results are discussed in

section 3 and followed by conclusions in section 4.

2. Numerical simulation approach

The flow is assumed incompressible, turbulent and three-dimensional, under steady and unsteady

conditions. The mathematical formulation is presented first followed by the computational

domain configuration, the boundary conditions and the numerical assumptions.

6
2.1. Reynolds averaged equations and turbulence models

The flow is governed by the Reynolds-averaged Navier-Stokes equations. The continuity and

momentum conservation equations in Cartesian tensor notation are given by:

𝜕
(𝜌𝑢𝑖 ) =0 (1)
𝜕𝑥𝑖

𝜕 𝜕 𝜕𝑝 𝜕 𝜕𝑢 𝜕𝑢 𝜕
(𝜌𝑢𝑖 ) + ̅̅̅̅̅̅̅
(𝜌𝑢𝑖 𝑢𝑗 ) = − 𝜕𝑥 + 𝜕𝑥 [𝜇 (𝜕𝑥 𝑖 + 𝜕𝑥𝑗 )] + 𝜕𝑥 (−𝜌𝑢 𝑖 ′𝑢𝑗 ′) (2)
𝜕𝑡 𝜕𝑥𝑗 𝑖 𝑗 𝑗 𝑖 𝑗

̅̅̅̅̅̅̅
The Reynolds stresses −𝜌𝑢 𝑖 ′𝑢𝑗 ′ require a turbulent model in order to close the above system of

equations.

2.1.1. Reynolds Stress Model

The Reynolds Stress Model (RSM), solves the transport equations for each term of the Reynolds

stress tensor, it is recognized to give the best performance among RANS models for highly

swirling flow. The exact transport equations of the Reynolds stresses, ̅̅̅̅̅̅̅
𝜌𝑢𝑖′ 𝑢𝑗′ , may be written as

follows:

𝜕 ̅̅̅̅̅̅̅ 𝜕
(𝜌𝑢𝑖′ 𝑢𝑗′ ) + 𝜕𝑥 (𝜌𝑢𝑘 ̅̅̅̅̅̅
𝑢𝑖′ 𝑢𝑗′ )
𝜕 ̅̅̅̅̅̅̅̅̅
= − 𝜕𝑥 [𝜌𝑢 ′ ′ ′ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
′ ′ ′ 𝜕 𝜕 ̅̅̅̅̅̅
′ ′
𝜕𝑡 𝑖 𝑢𝑗 𝑢𝑘 + 𝑃 (𝛿𝑘𝑗 𝑢𝑖 + 𝛿𝑖𝑘 𝑢𝑗 )] + 𝜕𝑥 [𝜇 𝜕𝑥 (𝑢𝑖 𝑢𝑗 )] −
𝑘 𝑘 𝑘 𝑘

′ ′ 𝜕𝑢𝑗 𝜕𝑢 𝜕𝑢′ 𝜕𝑢𝑗′ ̅̅̅̅̅̅̅̅̅

The term on the left side of the above equation represents the local time derivative and convective

term, and the terms on the right hand side represent the turbulent diffusion, molecular diffusion,

stress production, pressure strain and the dissipation, respectively. The stress transport equations

solved are naturally the modelled equations. In this study, we use Linear Pressure-Strain Model.

The pressure-strain term, ∅𝑖𝑗 , is modeled according to the proposals by Fu, et al. [22]; Gibson and

Launder [23]; and Launder [24, 25].

7
2.1.2. RNG k-ε and SST k-ω Model

The two-equation turbulence models (k-ε and k-ω models) are based on the Boussinesq

hypothesis to model the Reynolds stresses:

𝜕𝑢𝑖 𝜕𝑢 2
̅̅̅̅̅̅̅
−𝜌𝑢 𝑖 ′𝑢𝑗 ′ = 𝜇𝑡 ( + 𝑗) − (𝜌𝑘)𝛿𝑖𝑗 (5)
𝜕𝑥𝑗 𝜕𝑥𝑖 3

This method requires low computational cost but it based on the assumption of an isotropic

turbulent viscosity 𝜇𝑡 .

In the RNG k-ε model a further two modelled transport equations for the turbulence kinetic energy

k and its rate of dissipation are solved. The model equations were derived using a statistical

technique called renormalization group theory [26]. Compared to the standard k-ε model, it has

an additional term in the dissipation rate ε equation that improves the accuracy for rapidly strained

flows and a modified turbulent viscosity to account for the influence of swirl.

The Shear-Stress Transport SST k-ω model Menter [27] accounts for the transport of the

turbulence shear stress in the definition of the turbulent viscosity. It is more accurate and reliable

for a wider class of flows (for example, adverse pressure gradient flows, airfoils, transonic shock

waves) than the standard k-ω model. Further details can be found in fluent documentation [26].

2.2. Geometry and meshing

The computational geometry used in this study is relevant to the one existing in a liquid-gas

separator, a Perspex pilot model of which is shown in figure 1. The separator uses swirling flow

to separate the phases. As mentioned above the present simulation is focused on single phase

liquid flow to assess the performance of RANS turbulence models in this geometry and flow and

subsequently extend it to two-phase flow modelling. The geometry defining the flow domain is

8
presented in figure 2 with the origin of the system of coordinates adopted. It consists, see figure

2a, of a short inlet pipe of inner diameter 41 mm connected to a short cylindrical housing of length

94 mm and diameter 150 mm which houses the swirler. The swirler shown in figure 2b is made

of a stainless steel cylinder concentric to the housing with external and inner diameters of 76.2 mm

and 39 mm respectively with sixty 4 mm diameter radial holes. The swirler is axially connected

to a 610 mm long outlet pipe of 41 mm in diameter. The straight entry inflow is converted into a

swirling flow through the inclined holes inside the swirl cage as shown in figure 2a. The swirling

flow develops along the outlet pipe. At the end of the pipe a conical bluff body the details of which

are given on figure 1c. The bluff body is attached to a circular flange which contains three kidney

shaped holes through which the fluids escapes in the axial direction.

The hexahedral grids were generated using ANSYS Workbench. Different computational grids

were used as illustrated in table 1. Attention is directed towards the resulting swirling flow

which takes place between the swirler outlet and the tip of the bluff body.

Cone

Swirling flow

Figure 1: Pilot Perspex model of the in-line separator (swirler removed)

9
 (a)

(b)

(c)

Figure 2: (a)- geometry of flow domain, (b)- Swirler, (c)- Bluff body

Table 1. Computational grids

10
 Mesh #1 #2 #3

Number of elements 3,631,633 6,441,292 11,236,287

Wall y+ ≤ 21 9 -18 ≤ 11

2.3. Boundary conditions

An inlet velocity condition was used while a constant pressure (zero gauge pressure) was

prescribed at the outlet. A no-slip condition was taken for the remaining boundaries.

2.4. Numerical tools and simulation strategy

The commercial software Fluent 17.1, based on the finite volume method, was used in this study

and calculations were executed under steady conditions for all turbulence models and in addition

unsteady calculations were performed with the STT k- and RSM models to investigate the effect

of steady assumption. Water was the working fluid under atmospheric pressure and ambient

temperature.

The pressure-velocity coupling is achieved by using the SIMPLE scheme. For SST k-ω and RNG

k-ε models, the pressure was discretized using second order upwind scheme while the RSM model

simulation made use of PRESTO! Scheme. The discretization of the equations for momentum,

turbulent kinetic energy, dissipation rate and Reynolds stresses is achieved by second order upwind

scheme.

During the simulation for the unsteady case, the mean axial and tangential velocity and static

pressure from two points at different axial positions within the outlet pipe are used for monitoring

purposes. Convergence is assumed when the monitoring data achieve a statistically steady state.

The transient simulation is conducted in three steps. First, a steady solution with same settings is

launched to generate a good initial instantaneous field for the transient simulation. After that, the

11
transient simulation was run to allow the instantaneous field to develop with a time step equal to

4×10-4 s. Finally, the time-averaged field was monitored at the abovementioned points until a

statistical convergence is reached. It is important to mention that the residence time of the flow,

based on the axial bulk velocity of 1 m/s and the length of the computational domain, was 0.81 s.

For stability purposes, the Courant Number was kept smaller than unity. The time integration was

based on the second order implicit scheme. After 3 s of physical flow time, data collection, was

performed during 2.76 s of flow time.

3. Results and discussion

The base case flow corresponds to a Reynolds number based on the bulk velocity and pipe diameter

of 41,000. The test fluid was water at room temperature. Grid independency tests are considered

first. Subsequently, a comparison of performance between the turbulence models considered in

this study is conducted primarily against experimental data of Zhang [28] and accessorily with

LES computational results of Kharoua et al. [4] which, it has to be stressed, were obtained with a

much larger grid and involved considerably more usage of computational time. The quantitative

comparison uses the mean axial and tangential velocity components, their corresponding normal

turbulent single point correlations and the cross correlation between the axial and tangential

components of the fluctuating velocities and the swirl number axial evolution. The third

subsection examines the effect of varying the Reynolds number on the mean flow quantities such

as axial and tangential velocities and swirl number.

3.1. Grid independency and validation

Grid independency tests were conducted based on the use of the RSM model since this is the model

at the center of this study and the pipe Reynolds number value of 41,000 and are also gaged with

12
experimental results. Three meshes of 3,631,633, 6,441,292 and 11,236,287 elements were used.

Figure 3 compares the mean axial and tangential velocity profiles for the three meshes at an axial

location x/D=10. It can be seen that the differences between mesh 2 and 3 are not large while the

results for mesh 1 are quite different and fail to a large extent to replicate certain important flow

features such as the central reverse flow region, and the Rankine vortex flow structure. The

estimated wall y+ values inside the outlet pipe and around the bluff body surface for the different

meshes are illustrated in table 1. As a compromise between accuracy and computational cost,

Mesh #2 is used for the simulation of the cases considered in this work and it is inherently assumed

that the use of the RSM as the main model used for grid convergence test will not invalidate this

conclusion when the other models are used. It is worth mentioning here that the mesh attempts to

capture all the details of the three-dimensional geometry. It therefore includes the swirler with its

intricate flow passages in the solution domain and the resulting three-dimensional structure of the

flow and hence tries to conduct a simulation and an evaluation at industrial scale under RANS

framework of modelling.

2
Mesh #1
Mesh #2
1,5
Mesh #3
Experiment
1
U/Um

0,5

-0,5
x / D = 10
-1
-1 -0,5 0 0,5 1
r/R

13
 2,5
Mesh #1
2 Mesh #2
1,5 Mesh #3
1 Experiment
0,5
W/Um

0
-0,5
-1
-1,5
-2 x / D = 10
-2,5
-1 -0,5 0 0,5 1
r/R

Figure 3: Radial profiles of mean axial and tangential velocity for different meshes at x/ D = 10.

3.2. Mean velocities and Reynolds stresses and swirl number

The radial profiles of the mean axial and tangential velocity, at three representative axial positions,

are compared with experimental and LES data as shown in Figure 4. The simulation using RNG

k-ε model is performed in a steady state while the simulations using other models under both

steady and unsteady states. The LES profiles of reference [4] are considered for comparison.

14
 2
SST k-ω steady
SST k-ω unsteady
1,5 RNG k-ε
RSM steady
RSM unsteady
1 LES
Experiment
0,5
U/Um

-0,5

-1
x/D=5
-1,5
-1 -0,5 0 0,5 1
r/R

2
SST k-ω steady
SST k-ω unsteady
1,5 RNG k-ε
RSM steady
RSM unsteady
1 LES
Experiment
0,5
U/Um

-0,5

-1
x / D = 10
-1,5
-1 -0,5 0 0,5 1
r/R

15
 2
SST k-ω steady
SST k-ω unsteady
RNG k-ε
1,5 RSM steady
RSM unsteady
LES
1 Experiment

0,5
U/Um

-0,5
x / D = 14
-1
-1 -0,5 0 0,5 1
r/R

2,5
SST k-ω steady
2 SST k-ω unsteady
RNG k-ε
1,5 RSM steady
1 RSM unsteady
LES
0,5 Experiment
0
W/Um

-0,5
-1
-1,5
-2
x/D=5
-2,5
-1 -0,5 0 0,5 1
r/R

16
 2,5
SST k-ω steady
2 SST k-ω unsteady
RNG k-ε
1,5 RSM steady
RSM unsteady
1
LES
W/Um 0,5 Experiment
0
-0,5
-1
-1,5
-2
x / D = 10
-2,5
-1 -0,5 0 0,5 1
r/R

2,5
SST k-ω steady
2 SST k-ω unsteady
RNG k-ε
1,5 RSM steady
RSM unsteady
1
LES
0,5 Experiment
0
W/Um

-0,5
-1
-1,5
-2
x / D = 14
-2,5
-1 -0,5 0 0,5 1
r/R

Figure 4: Radial profiles of mean axial and tangential velocities.

Steady and unsteady RSM simulations predict similar results suggesting that a steady RSM

simulation is sufficient to avoid a costly unsteady RSM simulation if not specifically needed. The

mean axial velocity profiles show that the central core flow reversal region, which plays an

important role in a phase separation, is predicted well with RSM and persists from the exit of the

17
swirl cage till the bluff body. The SST k- and RNG k- under steady conditions fail to capture

the mean axial velocity shape and loose symmetry starting at x/D=4 of the mean axial velocity

profile shape. The reason behind this remains unclear. The mean axial velocity remains under-

predicted by RSM in the annular region near the wall, by as much as 50%.

The profiles of the mean tangential velocity correspond to a Rankine vortex. As the flow proceeds

downstream, the mean axial and tangential velocities decay due to viscous dissipation and the peak

tangential velocity shifts towards the pipe axis which is consistent with previous findings

Ahmadvand, et al. [30]. The RNG k-ε and steady SST k-ω simulations exhibit a strong asymmetry

in the profile of the velocity at x/D= 10 and 14. Similar to the axial velocity the results of unsteady

SST k-ω simulation show great improvement compared with steady SST k-ω simulation. All the

RANS and URANS simulations fail to capture the mean tangential velocity peaks in strong

contrast with LES simulation results of [4] which were much closer to the experimental ones. In

particular all RSM calculations display a lower solid body rotation than the experiments inside the

core and hence suggest that the calculated swirl intensity is decaying faster that the experimental

one.

In swirling pipe flow, the profiles of the normal turbulent stresses differ from those inside uni-

directional conventional pipe flow by the fact that they exhibit larger values, peaks at the center

and are highly anisotropic. For this reason, the radial profiles of the axial and tangential Reynolds

stress components at three representative axial positions are shown in figure 5 and comparison is

limited to RSM, experiments and LES of reference [4]. The RSM model fails to capture the peak

along the axis of the pipe, in the core region. Similar behavior was also obtained in the

computational work of highly swirling flow in combustors investigated by Leschziner and Hogg

[30]. They concluded that the inadequate modeling of pressure strain, especially its rapid part, is

18
responsible for the inaccurate prediction of Reynolds stresses. It is well known that the model

coefficient could be adjusted to improve the results but the approach remains customized and not

universal. The LES simulation predicts the turbulence, at the core, in a better way although still

exhibiting important discrepancies suggesting that model and mesh-resolution tunings are required

in the core region due to this complex flow behavior. Outside the core region the levels of the

normal stresses remain in qualitative agreement. The same can be said for the shear stress

component also shown in Figure 5.

0,5
x/D=5 Experiment
0,45
LES
0,4
RSM unsteady
0,35
RSM steady
0,3
0,25
uu

0,2
0,15
0,1
0,05
0
-1 -0,5 0 0,5 1
r/R

0,5
x / D = 10 Experiment
0,45
LES
0,4
RSM unsteady
0,35
RSM steady
0,3
0,25
uu

0,2
0,15
0,1
0,05
0
-1 -0,5 0 0,5 1
r/R

19
 0,5
x / D = 14 Experiment
0,45 LES
0,4 RSM unsteady
0,35 RSM steady
0,3
0,25
uu

0,2
0,15
0,1
0,05
0
-1 -0,5 0 0,5 1
r/R

0,4
x/D=5 Experiment
0,35 LES
RSM unsteady
0,3
RSM steady
0,25

0,2
ww

0,15

0,1

0,05

0
-1 -0,5 0 0,5 1
r/R

20
 0,4
x / D = 10 Experiment
0,35 LES
RSM unsteady
0,3
RSM steady
0,25

0,2
ww

0,15

0,1

0,05

0
-1 -0,5 0 0,5 1
r/R

0,4
x / D = 14 Experiment
0,35 LES
RSM unsteady
0,3
RSM steady
0,25

0,2
ww

0,15

0,1

0,05

0
-1 -0,5 0 0,5 1
r/R

21
 0,3
x/D=5 Experiment
LES
0,25
RSM unsteady
uw 0,2 RSM steady

0,15

0,1

0,05

0
-1 -0,5 0 0,5 1
r/R

0,3
x / D = 10 Experiment
LES
0,25
RSM unsteady
0,2 RSM steady

0,15
uw

0,1

0,05

0
-1 -0,5 0 0,5 1
r/R

22
 0,3
x / D = 14 Experiment
LES
0,25
RSM unsteady
0,2 RSM steady

0,15
uw

0,1

0,05

0
-1 -0,5 0 0,5 1
r/R

Figure 5: Radial profiles of Reynolds stresses for different turbulence models.

The swirl number is a non-dimensional number that is used to characterize intensity of swirling
flows, which for a pipe of radius R can be expressed as, see Kitoh [5]:

R
2  r 2UWdr
S 0 (6)
R 3U b2

where 𝑈𝑏 is the bulk velocity inside the pipe. To calculate S a numerical integration is necessary

to evaluate the numerator and it is assumed that the velocities vary linearly between the closest

measured velocity to the wall and the zero value at the wall itself.

The swirl number variation along the length of the pipe is shown for the RSM model and

experimental data in figure 6. Results from the two-equation turbulence models are not shown

since they failed to capture accurately the variation of the fluid velocity along the pipe at least with

the present mesh used. The RSM variation shows a decay of swirl with downstream distance and

rate of decay almost equal to the experimental one but lower by about 10%. This is due to the fact

23
that swirl intensity was under predicted by RSM as shown with the mean tangential velocity.

However, the rate of decay predicted by the RSM is similar to the experimental one. The swirl

intensity itself will depend on accurate prediction of the tangential mean velocity profile which

itself will depend on accurate prediction of the flow inside the intricate swirler passages which are

directly responsible for swirl generation and this could be the source of the discrepancy.

1,6

1,5

1,4
Swirl number

1,3

1,2

1,1 Experiment

1 LES
RSM steady
0,9
RSM unsteady
0,8
0 3 6 9 12 15
x/D

Figure 6: Swirl decay for different turbulence models

3.3 Effect of the Reynolds number

Numerical simulations were conducted for a range of Reynolds numbers from 4,100 to 82,000 to

examine its effect on the velocity distribution, swirl number decay and pressure on the conical

body. This is discussed below.

Figure 7 depicts the mean axial and tangential velocity profiles for the cases with different inlet

Reynolds number at three representative axial positions between the exit from the swirler and

the tip of the conical bluff body. The numerical results and discussion are confined the the RSM

24
simulations. Each velocity profile is normalized by the bulk axial velocity for that profile. For

Re=4,100, the reversal flow disappears at x/D=6 and the size of forced vortex is smallest among

all the cases. The flow then develops with a positive axil velocity similar to unidirectional pipe

flow. This type of flow indicates possibly vortex breakdown. For Re=20,500, the width and

magnitude of reversal flow zone decrease along the pipe axis direction and disappear at x / D =

14 just ahead of the tip of the bluff body. The velocity profiles show slight deviation for

Re=41,000 and 82,000. The maximum reversal flow region is obtained when Re=41,000. The

peak tangential velocity in the annular region increases with Reynolds number. It is to be noted

that for Re=4,100 after the disappearance of the core reverse flow region, the flow is trying to

adjust to solid body rotation distribution, while the positive axial velocity increases rapidly in the

core of the flow. Parchen and Steenbergen [31] have demonstrated how the axial velocity

distribution is swirling pipe flow can be influenced by the swirl generator geometry so that in

some cases the central reverse flow region can be eliminated and in the work of Pashtrapanska et

al. [9] no reverse flow region was measured.

1,5

0,5
U/Um

Re=4,100
-0,5 Re=20,500
Re=41,000
Re=82,000 x/D=5
-1
-1 -0,5 0 0,5 1
r/R

25
 1,5

0,5
U/Um

0
Re=4,100
-0,5 Re=20,500
Re=41,000
Re=82,000 x / D = 10
-1
-1 -0,5 0 0,5 1
r/R

1,5

0,5
U/Um

Re=4,100
-0,5 Re=20,500
Re=41,000
x / D = 14
Re=82,000
-1
-1 -0,5 0 0,5 1
r/R

26
 2,5
Re=4,100
2
Re=20,500
1,5
Re=41,000
1
Re=82,000
0,5
W/Um

0
-0,5
-1
-1,5
-2 x/D=5
-2,5
-1 -0,5 0 0,5 1
r/R

2,5
Re=4,100
2
Re=20,500
1,5
Re=41,000
1
Re=82,000
0,5
W/Um

0
-0,5
-1
-1,5
-2
x / D = 10
-2,5
-1 -0,5 0 0,5 1
r/R

27
 2,5
Re=4,100
2
Re=20,500
1,5
Re=41,000
1
Re=82,000
0,5
W/Um

0
-0,5
-1
-1,5
-2
x / D = 14
-2,5
-1 -0,5 0 0,5 1
r/R

Figure 7: Radial profiles of mean axial and tangential velocity for different Reynolds Numbers.

Swirl decay prediction for swirling pipe flow is relevant to several engineering applications

(Steenbergen and Voskamp [8] including flow inside two-phase separation equipment, it is

therefore important to consider the influence of the Reynolds number on it. At this point it is fit

to point out that several researchers have obtained an expression for the variation of swirl number

with axial distance in a pipe Steebergen and Voskamp [8] and Najafi et al. [32]. Thus by

integrating the momentum equation in cylindrical coordinates and assuming slowly developing

flow along the axial direction, neglecting turbulent shear stress and assuming relatively low swirl

intensity the swirl number is shown to vary as:

𝑆 (𝑥−𝑥𝑟 )
= 𝑒 −𝛽 𝐷 (7)
𝑆𝑟

28
 1,2

0,8 =0.027

=0.033
S/Sr

0,6
Re=4,100
=0.058
0,4 Re=20,500
Re=41,000
0,2 Re=82,000
Experiments =0.184
0
0 5 10 15
x/D

Figure 8: Swirl decay for different Reynolds numbers

The present discussion will be limited to cases when Re>4,100 where the flow exhibits a tangential

velocity profile of the Rankine compound type. The calculated values of  were 0.058, 0.033 and

0.027 for Reynolds number values of 20,500, 41,000 and 82,000 respectively. These values are

in agreement with the findings of Najafi et al. [32], so that as  decreases almost linearly with

Reynolds number. In addition, Steenbergen and Voskamp [8] have generated a comprehensive

graph encapsulating the variation of the decay rate of swirl  with Reynolds number for a number

of previous experimental studies performed during the 1960’s till 1998. The results cover a wide

range of swirl generation methods and swirl intensities. The graph confirms the exponential nature

of the decay rate and despite a relatively wide scatter depending on the swirl number ranges and

exact experimental conditions remains however a very useful reference for smooth pipes. The

present results agree very well with the values reported in that graph. Thus the value of  for

Reynolds of 20,500, 41,000 and 82,000 agree very well with the findings of Nissan and Bresan

29
[33], Bake [34] and Kitoh [5] respectively. The value of b for Reynolds of 41,100 of 0,033

compares favourably with the experimental value of Steenbergen and Voskamp [8] for a near

enough Reynolds equal to 50,000 and a  value of 0.302.

This information is crucial when considering the optimum axial position of the bluff body inside

the real gas liquid separator.

Figure 9 depicts the pressure distributions on the bluff body surface for different inlet Reynolds

numbers. It illustrates the effect of the swirling flow, with different swirl intensities (shown in

figure 8), on the loads experienced by the bluff body. For the case with Re=4,100, the pressure

peak is observed on the bluff body apex. It means the stagnation point is at the cone apex for

low swirl since the flow for this Reynolds number is flowing the downstream direction without

flow reversal on the axis. Increasing the Reynolds number (swirl intensity) causes flow reversal

on the axis and the apex to become gradually a region with minimum pressure and the first

cylindrical surface after the cone to become the maximum pressure region. The implication for

flow separation is that if the Reynolds number is low enough, the central recirculation zone does

not extend to the tip of the cone and hence the swirl strength is reduced which would adversely

affect separation. Thus, it is very important to consider swirling flow characteristics when

investigating the optimal design of the bluff body in terms of size, shape and location for a more

efficient separation. It is worth mentioning that the hollow cone would undergo a negative drag

force, for high swirl numbers, tending to dismantle it from its supporting hollow tail pipe.

Obviously, more research work is necessary to investigate this effect in detail.

30
Re=4100

Re=20,500

Re=41,000

31
Re=82,000

Figure 9: Pressure (Pa) contours on the bluff body surface

4. Conclusions

A numerical study on turbulent swirling flow interacting with a conical bluff body inside a short

pipe was conducted. The simulations in contrast to previous ones on swirling pipe flows which

were done using two-dimensional grids, include the full complex three-dimensional geometry of

the swirler in the solution domain. The following conclusions can be made:

The RSM turbulence model presents the best performance in contrast to the two-equation

turbulence models such as RNG k-e and k-w. The RNG k-e and SST k-w in steady mode fail to

predict important features of the flow and can be considered as the worst of the models. The

SST k-w model in unsteady mode performs better than the first two but remains inferior to the

RSM.

The RSM in both steady and unsteady modes performed equally well and provided the best

performance in the sense of being able to capture the variation of the mean axial and tangential

velocity profiles and swirl decay along the pipe.

32
The two-equation models and the steady RSM are not able to capture the correct variation of the

Reynolds stresses. Especially the peak in the central core zone. Only the unsteady RSM is able

to capture the peak but unable to obtain the correct values.

The effect of the Reynolds number on the flow calculated with the unsteady RSM indicated that

for Reynolds greater than 4,100, the flow is of the Rankine compound type and that the rate of

decay of swirl is exponential. The rate of decay was found to decrease almost linearly with

increase in Reynolds number. The obtained values of the rate of decay are in agreement with

previous experimental and numerical findings published in the literature.

5. Acknowledgement

The authors are grateful to ADNOC Onshore Company. (ADCO) for the financial support of

this research project and for Khalifa University of Science and Technology for granting Jinli

Song a graduate fellowship for MSc studies.

6. References

[1] E. S. Rosa, F. A. França, and G. S. Ribeiro, "The cyclone gas–liquid separator: operation
and mechanistic modeling," Journal of Petroleum Science and Engineering, vol. 32, pp. 87-101,
2001.
[2] S. Martemianov and V. L. Okulov, "On heat transfer enhancement in swirl pipe flows,"
International Journal of Heat and Mass Transfer, vol. 47, pp. 2379-2393, 2004.
[3] B. Wegner, A. Maltsev, C. Schneider, A. Sadiki, A. Dreizler, and J. Janicka, "Assessment
of unsteady RANS in predicting swirl flow instability based on LES and experiments,"
International Journal of Heat and Fluid Flow, vol. 25, pp. 528-536, 2004.
[4] N. Kharoua, L. Khezzar, and M. Alshehhi, "The interaction of confined swirling flow
with a conical bluff body: Numerical simulation," Chemical Engineering Research and Design,
vol. 136, pp. 207-218, 2018.
[5] O. Kitoh, "Experimental study of turbulent swirling flow in a straight pipe," Journal of
Fluid Mechanics, vol. 225, pp. 445-479, 1991.

33
[6] W. Steenbergen, Turbulent pipe flow with swirl: Citeseer, 1995.
[7] A. F. Moene, Swirling pipe flow with axial strain: experiment and large eddy simulation:
Technische Universiteit Eindhoven, 2003.
[8] W. Steenbergen and J. Voskamp, "The rate of decay of swirl in turbulent pipe flow,"
Flow Measurement and Instrumentation, vol. 9, pp. 67-78, 1998.
[9] M. Pashtrapanska, J. Jovanović, H. Lienhart, and F. Durst, “Turbulence measurements in
a swirling pipe flow”,. Exp. Fluids, 41, 813-827, 2006.
[10] P. A. Dellenback, D. E. Metzger, D.E. and G. Neitzel," Measurements in turbulent swirling
flow through an abrupt axisymmetric expansion," AIAA J., 26(6), pp. 669-681, 1988.
[11] L. Khezzar, "Velocity measurements in the near field of a radial swirler," Exp. Therm.
Fluid Sci. , 16(3), pp. 230-236, 1998.
[12] H. Mak, and S. Balabani, "Near field characteristics of swirling flow past a sudden
expansion," Chem. Eng. Sci., 62(23), pp. 6726-6746, 2007.
[13] S. Jakirlic, K. Hanjalic, and C. Tropea, “Modeling rotating and swirling turbulent flows:
A perpetual challenge”, AIAA J., 40, 10, pp. 1984-1996, 2002.
[14] C. G. Speziale, S. Sarkar, and T. B. Gatski, “Modelling the pressure-strain correlation of
turbulence: An invaraint dynamical systems approach”, Journal of Fluid Mechanics, 227, pp.
245-272, 1991.
[15] T. Kobayachi and M. Yoda, “Modified k-e model for turbulent swirling flow in a pipe”,
JSME International Journal, 30, 259, pp. 66-71, 1987.
[16] A. Escue and J. Cui, "Comparison of turbulence models in simulating swirling pipe
flows," Applied Mathematical Modelling, vol. 34, pp. 2840-2849, 2010.
[17] M. Brennan, "CFD Simulations of Hydrocyclones with an Air Core: Comparison
Between Large Eddy Simulations and a Second Moment Closure," Chemical Engineering
Research and Design, vol. 84, pp. 495-505, 2006.
[18] F. M. Erdal and S. A. Shirazi, "Local velocity measurements and computational fluid
dynamics (CFD) simulations of swirling flow in a cylindrical cyclone separator," Journal of
Energy Resources Technology, vol. 126, pp. 326-333, 2004.
[19] J. A. Ramirez and C. Cortes, "Comparison of different URANS schemes for the
simulation of complex swirling flows," Numerical Heat Transfer, Part B: Fundamentals, vol. 58,
pp. 98-120, 2010.
[20] R. Huang and F. Tsai, "Observations of swirling flows behind circular disks," AIAA
journal, vol. 39, pp. 1106-1112, 2001.
[21] K. Atvars, M. Thompson, and K. Hourigan, "Modification of the flow structures in a
swirling jet," in IUTAM Symposium on Unsteady Separated Flows and their Control, pp. 243-
253, 2009.
[22] S. Fu, B. Launder, and M. Leschziner, "Modelling strongly swirling recirculating jet flow
with Reynolds-stress transport closures," in 6th Symposium on Turbulent Shear Flows, pp. 17-6,
1987.
[23] M. Gibson and B. Launder, "Ground effects on pressure fluctuations in the atmospheric
boundary layer," Journal of Fluid Mechanics, vol. 86, pp. 491-511, 1978.
[24] B. E. Launder, "Second‐moment closure and its use in modelling turbulent industrial
flows," International Journal for Numerical Methods in Fluids, vol. 9, pp. 963-985, 1989.
[25] B. E. Launder, "Second-moment closure: present… and future?," International Journal of
Heat and Fluid Flow, vol. 10, pp. 282-300, 1989.
[26] A. Fluent, "Theory Guide 17.2," Ansys Inc. USA, 2016.

34
[27] F. R. Menter, "Two-equation eddy-viscosity turbulence models for engineering
applications," AIAA journal, vol. 32, pp. 1598-1605, 1994.
[28] T. Zhang, " Experimental investigation of swirling flow interaction with a bluff body,"
Master, Khalifa University of Science and Technology, Abu Dhabi, UAE, 2018.
[29] M. Ahmadvand, A. Najafi, and S. Shahidinejad, "An experimental study and CFD
analysis towards heat transfer and fluid flow characteristics of decaying swirl pipe flow
generated by axial vanes," Meccanica, vol. 45, pp. 111-129, 2010.
[30] M. Leschziner and S. Hogg, "Computation of highly swirling confined flow with a
Reynolds stress turbulence model," AIAA journal, vol. 27, pp. 57-63, 1989.
[31] R. R. Parchen & W. Steenbergen, “An experiemnatl and numerical study of turbulent
swirling pipe flows”, Journal of Fluis Engineering, 120, 54-61, 1998.
[32] A. F. Najafi, S. M. Mousavian, and K. Amini, "Numerical investigations on swirl
intensity decay rate for turbulent swirling flow in a fixed pipe," International Journal of
Mechanical Sciences, vol. 53, pp. 801-811, 2011.
[33] A. H. Nissan, & V. P. Bresan, “Swirling flow in cylindres”, A.I.Ch.E. Journal, 7, 4, 543-
547, 1961.
[34] D. W. Baker, “Decay of swirling turbulent flow of incompressible fluids in long pipes”,
PhD thesis, University of Maryland, 1967.

35