Engineering Applications of Computational Fluid

Mechanics

ISSN: 1994-2060 (Print) 1997-003X (Online) Journal homepage: http://www.tandfonline.com/loi/tcfm20

Characteristics of Secondary Flow Induced by 90-

Degree Elbow in Turbulent Pipe Flow

Jongtae Kim, Mohan Yadav & Seungjin Kim

To cite this article: Jongtae Kim, Mohan Yadav & Seungjin Kim (2014) Characteristics of
Secondary Flow Induced by 90-Degree Elbow in Turbulent Pipe Flow, Engineering Applications of
Computational Fluid Mechanics, 8:2, 229-239, DOI: 10.1080/19942060.2014.11015509

To link to this article: http://dx.doi.org/10.1080/19942060.2014.11015509

Copyright 2014 Taylor and Francis Group

LLC

Published online: 19 Nov 2014.

Submit your article to this journal

Article views: 701

View related articles

View Crossmark data

Citing articles: 2 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=tcfm20

Download by: [177.192.248.16] Date: 18 April 2017, At: 14:59

 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2, pp. 229239 (2014)

CHARACTERISTICS OF SECONDARY FLOW INDUCED BY 90-

DEGREE ELBOW IN TURBULENT PIPE FLOW

Jongtae Kim^*, Mohan Yadav#, and Seungjin Kim#

^Severe Accident and PHWR Safety Research Division, Korea Atomic Energy Research Institute,
Republic of Korea
# Mechanical and Nuclear Engineering Department, The Pennsylvania State University, USA
*E-Mail: [email protected] (Corresponding Author)

ABSTRACT: The purpose of this study is to characterize the swirling secondary flow in the downstream of a
pipe bend using a numerical simulation of the flow. The CFD (Computational Fluid Dynamics) software
OpenFOAM is used to simulate the turbulent flow in pipes with elbow. Various turbulence models are benchmarked
with the existing experimental data and a comparative study is performed to select an appropriate turbulence model
for the analysis. Predictions made by the selected turbulence model are compared with the LDA (Laser Doppler
anemometer) measurements from the experiments currently conducted to find the dependency of the flows on the
Reynolds number. It is found that the swirl intensity of the secondary flow is a strong function of the radius of
curvature of the bend and a weak function of the Reynolds number. Additionally, it is found that the dissipation of
the swirl intensity is exponential in nature.

Keywords: secondary flow, 90-degree bent elbow, swirl intensity, CFD

1. INTRODUCTION cooling systems of nuclear power plants, where

many bends are used to interconnect the
A 90-degree-bent elbow is commonly used in components. The distortion in the streamwise
many piping systems to connect different velocity profile of the coolant by an elbow can
components. It is well known that during the flow affect the distribution of heat transfer rates. In the
of fluid in a pipe with elbow, a secondary motion case of a bubbly two-phase flow which may occur
of the flow is developed and it is superimposed on during an accident in a nuclear power plant, the
its primary streamwise flow. The secondary flow distribution of the bubbles in the pipe with an
is generated due to an imbalance between elbow is very different compared to that in a pipe
centrifugal force and pressure gradient near the without an elbow. Due to the inherent difference
bent-pipe wall in the radial direction of the bend in density of the two phases, the gas bubbles are
curvature. In general, the structure of the driven toward the centers of the two counter-
secondary flow is dependent on the radius of the rotating swirls generated by the elbow (Yadav,
bend curvature (Rc) and Reynolds number (Re) 2013). This accumulation of bubbles can increase
based on the pipe diameter (D) and bulk velocity the possibility of a bubble collision and
(Ub). subsequently alter the interfacial heat and mass
When the non-dimensional radius of curvature of transfers. Furthermore, when the curvature radius
the bend is greater than 1.5 (i.e. Rc/D > 1.5), the of an elbow is small enough to create a
secondary flow consisting of a pair of counter- fluctuating flow after the elbow, the downstream
rotating vortices are generated. At the same time, pipe can undergo oscillating mechanical stress,
the velocity profile of the primary streamwise which can lead to structural fatigue in the piping
flow is distorted and shifted away from the center system (Yamano et al., 2009).
of the curvature of the elbow. If Rc/D is smaller In the past, several researchers have investigated
than 1.5, the flow becomes unsteady because of a turbulent flows in pipes with elbows by means of
flow separation occurring immediately theoretical, experimental, and numerical methods.
downstream of the bend (Weske, 1948; Hellstrom Weske (1948) experimentally investigated the
et al., 2011). velocity distribution at the outlet of elbows with
The flow in the downstream region of a 90-degree various shapes of cross-sections including round,
bend is important for the primary and secondary elliptical, square, and rectangular cross sections at

Received: 22 Jul. 2013; Revised: 12 Oct. 2013; Accepted: 9 Dec. 2013

229
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

Reynolds numbers ranging from 0.2 to 0.6 106 turbulence model among the available turbulence
for the design of an aircraft duct. models in OpenFOAM CFD code.
Sudo et al. (1998) performed the experiments for
turbulent flow in a 90-degree-bent pipe with Rc/D 2. EXPERIMENTAL FACILITY AND
= 2. The study used a hot-wire anemometer to MEASUREMENT
measure streamwise velocity (Us) and
circumferential velocity (Uc) for a turbulent flow As mentioned above, the flow structure after an
with Reynolds number of 60,000. Al-Rafai et al. elbow depends on the Reynolds number and bend
(1990) performed an experimental and numerical curvature. An experiment is conducted to evaluate
study of a turbulent air flow in circular pipe bends its dependency on the Reynolds number.
to investigate the influence of Rc/D on the flow.
During the experiments, two bends with Rc/D =
3.49 and Rc/D = 6.975 were used, and the flow
had the Reynolds number of 34,132. Homicz
(2004) performed numerical simulations of the
flow through a 90-degree elbow with a Rc/D =1.4
at a Reynolds number of 5.4105 to validate the
Flow-Accelerated Corrosion (FAC) test
conducted by Korea Atomic Energy Research
Institute (KAERI). These simulations and FAC
tests are used to better characterize the flow-
accelerated corrosion/erosion in the pipe elbows. Fig. 1 Experimental facility and locations of LDA
More recently, Tanaka et al. (2009) applied a measurement.
large eddy simulation (LES) approach using a
standard Smagorinsky model with a wall function A schematic diagram of the experimental setup,
law for flows in elbows with various Rc/D at as well as the co-ordinate system used for the
several Reynolds number conditions. current study, is shown in Fig. 1. The detailed
Most of the above-mentioned studies focused description of the test facility is presented in
only on the flow structure in the immediate previous studies (Yadav, 2013) and only
vicinity of the elbow and only a limited number characteristic features of the test facility are
of studies investigated the flow in the downstream presented here. The experimental facility consists
region of the elbows (Sudo et al., 1998; Al-Rafai of both upstream and downstream sections
et Al., 1990). The knowledge of the flow structure constructed from 50.8mm inner diameter (D)
in the downstream region of the elbow is of acrylic pipes and interconnected by a 90 glass
significant importance to understand the elbow. The elbow has a radius of curvature of
dissipation characteristics of the secondary flow 152.4mm (3D), and the lengths of the upstream
and to accurately assess the development of the and downstream sections are 3.35m (L/D = 66)
flow structure. Additionally, an understanding of and 9.1m (L/D = 180), respectively. A laser
the dissipation characteristics is important in leveling system and a digital level are employed
evaluating the length required for the elbow to ensure the alignment of both the upstream and
effects to be dissipated and in modeling the downstream sections within 0.1. The flow rate
effects associated with flow restrictions. for the experiments is measured by an
In view of the above, the objectives of the current electromagnetic flow meter, with an accuracy of
study are to perform experimental and numerical 0.5% of the full-scale reading.
studies on turbulent pipe flows after 90-degree
elbows and to characterize the secondary flows Table 1 Specifications of LDA employed for
induced by the 90-degree elbows including the experiments.
dissipation characteristics. In order for that, it is
necessary to choose an appropriate turbulence Measurement Length 300 mm
model specifically for the turbulent flow in a pipe
with an elbow. Unexpectedly, there are few Beam width 2.0 mm
papers which are devoted to a numerical study of Measurement Volume 0.1 (mm)0.1 (mm)1.0 (mm)
the turbulent pipe flow through an elbow. In this
study, a quantitative assessment method based on Maximum Velocity 27 m/s
hit rate using the experimental data of Sudo et al.
Measurable Velocity Fluctuation From 0.7 m/s to 4.6 mm/s.
(1998) is applied to choose an appropriate

230
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

An integrated Laser Doppler anemometer (LDA), and curvature radius of the bend. To evaluate the
which is capable of one-dimensional velocity dependencies, the experiment of Sudo et al.
measurements, is used to measure the axial water (1998) was referenced. In their experiment, a
velocity. The light beam in the LDA is generated bend with a curvature radius of 2D was used, and
by 35 mW laser with a wavelength of 600 nm. the flow velocity was measured at a Reynolds
The principal characteristics of the LDA are listed number of 60000 based on the pipe diameter and
in Table 1. It should be noted that due to the bulk velocity. The experimental facility used in
difference in the refractive index of air, acrylic, this study is equipped with an elbow of 3D
and water, the LDA measurement volume inside curvature radius. The test condition of Run1 is
the pipe is shifted by a finite amount. This shift is chosen with a bulk velocity of 1 m/s to obtain a
accounted for by employing Snells law. The similar Reynolds number. The experiment of
tracking/seeding particles consist of Polyamide Sudo et al. (1998) and Run1 in the present
particles with a size distribution of around 20m experiments were conducted at a similar Reynolds
and specific gravity of 1.04. To ensure the number with different curvature radii of 2D and
statistical reliability of the measurements, a 3D, respectively. The conditions of other tests,
minimum of 5000 samples are measured at each Run 2 and Run 3, are chosen by doubling the bulk
local point and the validation of the burst signals velocity to scrutinize the effect of the Reynolds
was above 98%. number.
As indicated in Fig. 1, the measurements are
performed at 13D upstream of the elbow (53D 3. NUMERICAL SIMULATION OF
from inlet) to figure out a fully-developed TURBULENT FLOWS IN 90-DEGREE-
velocity profile and 3.5D, 10D, and 50D BENT PIPE
downstream of the elbow in the downstream
section. The LDA is mounted on a two- The turbulent flow in a 90-degree-bent pipe is
dimensional traversing system, which has a complicated by the secondary flow, which is
resolution of 1m in both traversing directions. generated by an imbalance of the centrifugal force
This traversing system is further attached to a and pressure gradient in the radial direction of the
plate, which is designed with mounting holes at bend curvature. Additionally, the streamwise flow
every 22.5 in the azimuthal direction. As such, is distorted by the existence of adverse pressure
the LDA along with the traversing system can be gradient along the convex wall of the pipe bend.
attached to the mounting plate to measure the These make an accurate numerical simulation of
velocity across entire pipe cross-section. Along the flow difficult, especially for turbulent flows
each azimuthal direction, the LDA is traversed to with high Reynolds number. To select an
obtain measurements at 23 radial positions. These appropriate turbulence model for the numerical
radial positions are selected such that a finer simulation of a turbulent flow in a pipe with an
measurement mesh is obtained in the near wall elbow, validation of the numerical results from
region, where a higher velocity gradient is turbulence models is very important. In the
expected. current study, the experiments conducted by Sudo
et al. (1998) on the turbulent flows in a bent pipe
Table 2 List of flow conditions investigated in current are used to validate the turbulence models. In
study. addition, the selection criteria which involve a
quantitative assessment are used to perform a
Test No. Ub [m/s] Re comparative study among the tested turbulence
models. The selected turbulence model is then
Run1 1.00 50800 used for evaluation of swirl characteristics in the
Run2 2.00 101600
downstream of an elbow.

Run3 4.00 203200 3.1 Numerical simulation of Sudo et al.s

experiment
A total of three flow conditions, as shown in Table
2, are investigated in the current study. These 3.1.1 Test conditions and numerical simulation
flow conditions are selected to study the effect of
increasing Reynolds number on the velocity Sudo et al. (1998) performed experiments for
profiles. It is expected that the swirl intensity and turbulent flow in a 90-degree-bent pipe with Rc/D
its decay characteristic in a pipe with a 90-degree = 2. A straight pipe-section with 100D of
bend are dependent on Reynolds number of flow development length was installed in the upstream

231
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

of the bend to provide a fully developed turbulent pipe cross-section is constructed with
flow to the bend. The experiments were quadrilateral cells. The constructed mesh of the
performed with air as the working fluid and flow bent pipe is shown in Fig. 3 where the surface
Reynolds number Re = 60000. The study used a mesh near the pipe exit is magnified to visualize
hot-wire anemometer to measure the streamwise the orthogonality of the mesh. The value of the
velocity (US) and circumferential velocity (UC) at non-dimensional distance from the pipe wall
several locations along the pipe. based on the friction velocity , Y+, for a near-wall
The open source CFD code OpenFOAM cell is strictly controlled to be in the range of 20 ~
(OpenCFD, 2012) is used to simulate Sudo et al.s 50, when a wall-function is used. However, the
experiments. OpenFOAM solves the Y+ value for a near-wall cell is set below 1 in the
incompressible Navier-Stokes equations case without a wall-function. The distribution of
discretized in the context of the cell-centered mesh cells along the streamwise direction is
finite volume method (FVM). Numerical selected based on iterative grid-dependency tests.
solutions are iteratively sought by pressure So, the mesh size used in this study is depending
correction method named SIMPLE algorithm. on the use of a wall function. The mesh used for a
The standard k- turbulence model is applied to high-Re turbulence model is composed of
choose a suitable convective scheme and under- 152,150 hexahedral cells. But for a low-Re
relaxation parameters of the discretized equations. turbulence model, the number of cells is 562,080.
The computational domain is composed of an Fig. 4 shows the numerical results of the Sudo et
upstream pipe with 20D length and a downstream al.s experiment by using the standard k-
pipe with 50D length connected to an elbow as turbulence model. Fig. 4a is streamwise velocity
shown in Fig. 2. The boundary conditions at pipe profiles on cut-planes near the elbow, and Fig. 4b
inlet are set as follows: Velocity is fixed with a is streamlines to show the swirling flow. From the
fully developed power-law profile which is from figures, it can be expected that the primary
the experimental data of Zagarola and Smits streamwise flow and secondary swirling flow are
(1999) for fully developed turbulence pipe flow. correlated in the elbow pipe flow.
The values of the turbulent kinetic energy and its
dissipation rate are calculated from the inlet
turbulence intensity (5% of inlet bulk velocity in
this study) and pipe diameter.

(a) (b)

Fig. 4 Numerical results for experiment of Sudo et al.

Fig. 2 Computational geometry for simulation of 90o- (1998): (a) streamwise velocity profiles, (b)
bent pipe flow. streamlines.

Fig. 3 Computational mesh near pipe exit.

The mesh for the bent pipe is generated by Fig. 5 Comparisons of streamwise velocity
stacking a two-dimensional unstructured mesh of components along symmetric lines at x/D=0,
pipe cross-section along the streamwise direction x/D=0.5, x/D=1, x/D=2, and x/D=5. Solid
to minimize the non-orthogonality of the mesh. line: standard k- turbulence model, +:
The two-dimensional unstructured mesh of the Experimental data of Sudo et al. (1998).

232
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

A comparison between the calculated and Figs. 6 and 7 show the streamwise and
measured streamwise velocity profiles along the circumferential velocity components, respectively,
symmetric lines (i.e. =90) in the downstream along the cross lines (i.e. =0) in the downstream
region of the elbow are shown in Fig. 5. In the region of the bend pipe. The circumferential
figure, the positive and negative r/D indicates the velocity component means secondary or swirling
outer and inner part from the center of the circular velocity component. The velocity profiles along
pipe section, respectively. It should be noted that the cross lines are predicted well at different
the horizontal length of the legend + used for locations in the downstream region of the elbow.
the measured data in Fig. 5 indicates a 10% error
range normalized by the bulk velocity. It is found 3.1.2 Assessment of turbulence models
that the streamwise velocity components from the
standard k- turbulence model agree very well The main objective of the present study is to
with the measured data in the outer region with characterize the secondary flow in the
positive radii. On the contrary, some downstream of the elbow through numerical
discrepancies between the numerical and simulations. As such, after selection of mesh size,
measured data in the inner region with negative suitable convective scheme and under-relaxation
radii are observed. These discrepancies may arise parameters, it is of great importance to select
due to the adverse pressure gradient near the inner appropriate turbulence model that can simulate
wall of the pipe bend. the flow across the elbow accurately.
In view of this, OpenFOAM contains various
turbulence models, which include one-equation
and two-equation models based on Boussinesq
assumption, Reynolds stress transport model
(RSTM), and large eddy simulation (LES). It is
still challenging to apply LES for a high Reynolds
number turbulent flow in a very long pipe. In this
study, the RSTM turbulence models are excluded
from the assessment because of numerical and
turbulence model problems in obtaining an
acceptable solution, and remain as future work.
For the assessment, only the linear and non-linear
eddy-viscosity turbulence models are considered.
In Table 3, the turbulence models to be used for
Fig. 6 Comparisons of streamwise velocity
the assessment are listed.
components along cross lines at x/D=1, x/D=3,
x/D=5, x/D=7, and x/D=10. Solid line: standard
k- turbulence model, +: Experimental data of Table 3 Turbulence models used for simulation of
Sudo et al.s experiment.
Sudo et al. (1998).
No Abbr. Turbulence model
1 SKE Standard k- with wall function
2 EKE Extended k- with wall function (Chen and
Kim, 1987)
3 RNGKE RNG k- with wall function (Yakhot and
Orszag, 1986)
4 RKE Realizable k- with wall function (Shih et al.,
1995)
5 KMSST k- SST with wall function (Menter, 1994))
6 SNKE Shihs nonlinear k- with wall function (Shih
and Zhu, 1993)
7 LCKE Lien cubic k- with wall function (Lien et al.,
1996)
8 LSLOW Launder-sharma low-Re k- (Launder and
Sharma, 1974)
Fig. 7 Comparisons of circumferential velocity 9 LBLOW Lam-Bremhorst low-Re k- (Lam and
components along cross lines at x/D=0, Bremhorst, 1981)
x/D=0.5, x/D=1, x/D=2, and x/D=5. Solid line: 10 LCLOW Lien cubic low-Re k- (Lien et al., 1996)
standard k- turbulence model, +: Experimental 11 LLLOW Lien-Leschziner low-Re k- (Lien and
Leschziner, 1993)
data of Sudo et al. (1998).

233
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

Turbulence models can be grouped based on comparative study is performed between

modeling of near wall flow and turbulent eddy numerical simulations and experimental
viscosity. In Table 3, turbulence models measurements. The number of points within the
numbered 1 through 7 utilize a wall-function to error bounds of the experimental data is counted.
model the near wall region and the remaining A comparison is called a hit if the numerical
models simulate the near-wall region without a result lies within 10% of the bulk velocity from
wall-function. Shihs model (No. 6 in Table 3) the measured value, otherwise it is called a
(Shih et al., 1993) and Leins models (Nos. 7 & miss.
10 in Table 3) (Lien et al., 1996; Lien and
Leschziner, 1993) are the non-linear models,
where eddy viscosity is related by more than
second moment of velocity gradient.

Fig. 10 Hit rates of turbulence models for streamwise

velocity components on cross lines of Sudo et
Fig. 8 Comparison of streamwise velocity profiles al.s bent pipe, - - -: average hit rate (92%).
from different turbulence models along
symmetric line at x/D = 0.

Fig. 11 Hit rates of turbulence models for

Fig. 9 Hit rates of turbulence models for streamwise circumferential velocity components on cross
velocity components on symmetric lines of lines of Sudo et al.s bent pipe, - - -: average
Sudo et al.s bent pipe, - - -: average hit rate hit rate (92.5%).
(74.3%).
The probing locations and measured data used for
Fig. 8 shows the numerical results from different the quantitative assessment of the turbulence
turbulence models for streamwise velocity models are plotted in Figs. 5, 6, and 7 which
profiles along the symmetric line at x/D = 0. The denote the three assessment objects, i.e.,
numerical results from some of the turbulence streamwise velocity component on the symmetric
models listed in Table 3 provide reasonable lines and streamwise and circumferential velocity
agreement compared with Sudo et al.s components on the cross lines. The number of
experimental data and Tanaka et al.s LES results. points used in each assessment is 135. Figs. 9, 10,
However, it is difficult to choose an appropriate and 11 show the hit rates for each turbulence
turbulence model from the qualitative comparison. model, where the turbulence models are sorted by
As such, a quantitative assessment method is the number of hits. In the assessment of
applied to select an appropriate turbulence model. streamwise velocity along the symmetric lines,
Each turbulence model is implemented to SNKE has the highest number of hits followed by
simulate Sudo et al.s experiments and a LCKE and RKE. It is found that along the cross

234
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

without wall functions, such as LSLOW, LLLOW,

LBLOW, and LCLOW, have a higher than average
number of hits. However, in the assessment of
circumferential velocity along the cross lines, the
low-Re number turbulence models perform poorly
and KMSST & RNGKE record the highest number
of hits.
The above assessment shows that some of the
turbulence models consistently perform below the
average. These turbulence models are eliminated
from the future assessments. Additionally, it is
not likely to find a single turbulence model with
maximum number of hits for all three assessments.
Fig. 12 Comparisons of streamwise velocity As such, two additional criteria are applied for the
components for Run1 case (Ub=1m/s) along selection of an appropriate turbulence model. The
symmetric (=90) lines at x/D=3.5, 10, and
first step utilizes the average number of hits for
50.
each assessment and the second criterion
considers the deviation from the maximum hit
rate. The deviation of each assessment, k , from
the maximum hit rate is given by:
N

x xi
2
max
k = i=1
(1)
N
where xi, xmax, and N represent the hit rate for the
ith turbulence model, maximum hit rate and the
number of turbulence models used in the
assessment, respectively
It is found that RNGKE has consistently greater
than average hit rates for the assessment of all the
Fig. 13 Comparisons of streamwise velocity three velocity profiles. Although SNKE has
components for Run1 case (Ub=1m/s) along highest number of hits for the streamwise velocity
cross (=0) lines at x/D=3.5, 10, and 50. along the symmetric lines, it does not predict
other velocity profiles satisfactorily. Therefore,
SNKE is not considered for the final calculations.
Four turbulence models namely, EKE, RNGKE,
RKE, and KMSST satisfy the criterion based on
the deviation from the maximum hit rate. In order
to select a single turbulence model out of these
four, the selection criterion is changed to reduce
the range of deviation by 2/3 of . Based on this
quantitative assessment RNGKE turbulence
model is selected for investigating the flow in the
elbow. This is consistent with the previous study
(Homicz, 2004), which has shown that RNGKE
turbulence model performs better than SKE for
Fig. 14 Dependency of streamwise velocity flow in bends. Also in the simulation of
components on Reynolds number along cross Rahimzadeh et al. (2012) of a turbulent flow over
lines (=0) at x/D=3.5, 10, and 50, : Exp. circular spillways, it is seen that RNGKE
Run1 (Ub=1m/s, Re=50,800), : Exp. Run 2
turbulence model gives very similar results to
(Ub=2m/s, Re=101,600), +: Exp. Run 3
(Ub=4m/s, Re=203,200). experiment and RSTM model.

lines, the streamwise velocity is predicted well by 3.2 Analysis of current experiment
most of the turbulence models except SNKE and The experiment on a turbulent flow in a pipe with
LCKE. The low-Re number turbulence models a 90-degree-bent elbow conducted in this study is

235
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

numerically simulated with the RNGKE

turbulence model, which was chosen through a
quantitative evaluation of the simulation results of
Sudo et al.s bend pipe. Numerical and
experimental results are shown in Figs. 12 and 13
for the case of Run1 where the bulk velocity is 1
m/s and its Reynolds number is 50,800.
Streamwise velocity distributions along vertical
symmetric lines at 3.5D, 10D, and 50D in the
horizontal pipe after the elbow are plotted in Fig.
12. It is seen in the figure that the streamwise
velocity is shifted upward just after the elbow,
and is redistributed downstream. Fig. 13 shows Fig. 15 Dependency of streamwise velocity
the streamwise velocity distributions along the components on Reynolds number along cross
horizontal cross lines at the same locations as in lines (=0) at x/D=3.5, 10, and 50, :
Fig. 12. The symmetry of the flow is easily Calc. Run1 (Ub=1m/s, Re=50,800), : Calc.
Run 2 (Ub=2m/s, Re=101,600), -: Calc.
observed from the figure. The streamwise
Run 3 (Ub=4m/s, Re=203,200).
velocity near the pipe wall is a little augmented
because of its reduction in the central region of
the pipe section by swirls at 3.5D. However, the
peaks near the pipe wall are diffused by turbulent
dissipation at 10D. It looks like the calculated
results by the currently used numerical schemes
and turbulence model match well with the
measured data. From a comparison of the velocity Fig. 16 2-D planar meshes inserted in 3-D mesh
profiles, it can be thought that the turbulence domain to get area-averaged swirl intensities
model used in this study gives mostly acceptable along pipe after elbow.
results for the flow in a bend pipe.
To evaluate the dependency of a flow after the
elbow on the Reynolds number, the streamwise
velocities normalized by the bulk velocity of each
test are compared in Fig. 14. It is shown that the
velocity profiles from the three tests are
coincident with each other, which means that
there exists a similarity in the structure of flows
after a bend with the same curvature radius if the
Reynolds number is in the range of 50000 and
200000. Furthermore, it is found that the
normalized velocity at the center of the pipe at
Fig. 17 Decay of swirl intensity along straight pipe
50D downstream of the elbow decreases with after elbow with Rc = 2D.
increasing Reynolds number. This happens
because in a fully developed turbulent pipe flow
the flattening of the velocity profile increases
with Re. This observation is also consistent with
the recent experiments in super-pipe flow facility
(Zagarola and Smits, 1997).
The streamwise velocities from a numerical
simulation of the three test cases Run1, Run2, and
Run3 are compared in Fig. 15. The velocity
profiles along the cross lines at three different
locations downstream of the elbow coincide well
with each other, as do the measured data. It is also
shown in the figure that the velocity near the
center of the pipe at 50D downstream is flattened Fig. 18 Decay of normalized swirl intensity along
more as the Reynolds number increases. pipe after elbow with Rc = 3D.

236
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

4. CHARACTERIZATION OF SECONDARY experimental facility used in this study, is

FLOW INDUCED BY 90-DEGREE BEND evaluated only based on the numerical solutions
because the swirl components of the flow
In this section, the dissipation characteristics of velocities are not measured in the experiment.
secondary flow in the downstream region of the The swirl intensities for the three test conditions
90-degree elbow are discussed. (Run1, Run2, and Run3) were calculated along
The secondary flow generated during the flow the downstream horizontal pipe. Transitions of
across the elbows consists of two counter-rotating the swirl intensities along the pipe for the three
vortices or swirls. In general, the strength of a cases (50800 Re 203200) are shown in Fig. 18.
swirl on a cut plane normal to the axial flow In the comparison of the swirl intensities among
direction is represented by the swirl number the cases, as shown in the figure, they show
(Chigier and Beer, 1964). The swirl number is the similar trends of swirl intensity decay to the case
area-averaged flux of angular momentum and it is of 2D curvature radius of elbow with a minor
used to quantify the strength of a swirl around the dependency on the Reynolds numbers of the
axis of the main flow. However, a direct flows. It is found that the dissipation of Is for all
application of the swirl number to counter- flow conditions can be expressed as an
rotating vortices such as those created by the exponentially decreasing function as
secondary flow in an elbow is not appropriate. Is L
To define the intensity of counter-rotating swirls, exp( s ) (3)
Sudo et al. (1998) used the magnitude of velocity I s0 D
tangential to a pipe section area. Eq. (2) shows the where the coefficient s represents the rate at
definition of the swirl intensity used for the elbow which the secondary flow intensity decreases
flows in this study. downstream of the elbow. The numerical value
U (U n)n) dA

of s is found to be 0.21 to obtain a best fit
2

Is = (2) between the data and correlation presented in Eq.

U dA
2
b
(3). Based on Eq. (3), I s dissipates to 10% of its
where I s is the swirl intensity, U is the vector of initial value at approximately 11D downstream of
the flow velocity and n is a unit vector normal to the elbow.
the pipe section area. To investigate the effect of other geometric
To obtain the area-averaged swirl intensity at each parameters, such as the radius of curvature, the
pipe section along the downstream pipe after the results are compared with the previous velocity
elbow, the flow solution is mapped to a two- measurement of Al-Rafai et al. (1990) with LDA
dimensional mesh with the same area as the pipe in bent pipe flow with two different radii of
section because of the use of unstructured meshes curvature, RC/D = 3.49 and 6.98, at Re = 34132.
in this study, and the tangential velocities from Additionally, numerical simulations of the flow
the mapped solution are integrated on the 2-D were performed by using the commercial CFD
mesh. Fig. 16 shows planar 2-D meshes inserted code Phoenics. Turbulent flows in the 90-degree
in the computational domain for a pipe flow. In bent pipes with the same geometric and physical
the figure, the dark-colored areas mean the 2-D conditions as in Al-Rafai et al.s experiment are
meshes used for area-averaged swirl intensities. simulated using the RNG k- turbulence model
The intensity of secondary flow calculated from and it is found that the current numerical method
Eq. (2) at different axial locations downstream of and the turbulence model give very acceptable
the elbow is normalized by the value of intensity results. Here, only the swirl intensities from the
at the exit of the elbow. The normalized swirl computed results are introduced and compared
intensity for elbow with Rc/D = 2 along the with other results. Table 4 shows the list of
downstream pipe after the elbow is depicted in experiments used to characterize the swirl
Fig. 17. It can be seen that the swirl intensity intensity of secondary flow. In the list, the radius
generated by the elbow is exponentially of the elbow curvature is in the range of 2D to 7D,
diminished along the downstream pipe. The and the Reynolds number is from 30000 to
numerical result by the RNG k- turbulence 200000.
model reproduces well the decay characteristic of In Fig. 19, the swirl intensities along the
the swirl obtained by Sudo et al.s experiment. downstream pipes of the elbows are shown for the
In the same way, the decay characteristic of the cases listed in Table 4. In the cases of curvature
secondary flow downstream of the elbow with radii lower than or equal to Al-Rafai et al.s
curvature radius of 3D, which represents the smaller bend (3.49D), the swirl intensities show a

237
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

very similar trend of decay even with different The swirl intensities at the exit of the elbows for
Reynolds numbers. However, in the case of Al- the five cases (one for Sudo et al.s study, three
Rafai et al.s larger bend (6.975D), it decays very for the current experiments, and two for Al-Rafai
quickly compared to the other cases. et al.s study) are compared in Figs. 20 and 21. It
is observed that the swirl intensity at the exit of
the elbows has a weak dependence on the flow
Reynolds number and a higher dependence on the
radius of curvature of the bend. However,
additional studies are necessary to provide a
correlation between swirl intensity and the elbow
radius of curvature.

5. CONCLUSIONS

In this study, experimental and numerical work on

Fig. 19 Dependency of decay of normalized swirl a turbulent flow in a pipe with a 90-degree elbow
intensity along pipe on radius of elbow was performed. In order to select an appropriate
curvature. turbulence model for the turbulent flow in the
bent pipe, a quantitative assessment method based
on hit rate was applied by using Sudo et al.s
experimental data. The assessment showed that
the RNG k- turbulence model gives good results
for primary streamwise velocity and secondary
swirling velocity profiles compared to other
turbulence models. The non-linear turbulence
models showed a good predictability for the
streamwise velocity components but they gave
poor results in secondary swirling velocity
components. As a future work, it is required to
Fig. 20 Reynolds number dependency of swirl study the applicability of the non-linear
intensity at exit of elbow. turbulence models for complicated swirling flows.
The experiments on turbulent flows in the pipe
with an elbow with a curvature radius of 3D were
conducted by varying the Reynolds number, and
numerically simulated using the RNG k-
turbulence model to evaluate the dependency of a
flow after the elbow on the Reynolds number. It
was found from the experimental and numerical
study that there is a strong similarity in
streamwise velocity profiles in the range of
Reynolds number from 50,000 to 200,000.
Using the numerical method and turbulence
model selected from this study, the characteristics
Fig. 21 Dependency of swirl intensity at exit of elbow of swirl intensity after an elbow were examined.
on radius of elbow curvature. It was found that the swirl intensity defined in
terms of area averaged tangential velocity
Table 4 Tests and their conditions used for study of decreases exponentially along the flow after the
swirl characterization.
elbow, dissipating quicker as the radius of the
elbow curvature is larger. It was also found that
Test Radius of elbow curvature (Rc) Reynolds number (Re)
this dissipation characteristic depends weakly on
Current 3D 50000 200000 the Reynolds number in the range of Reynolds
numbers investigated in this study and strongly on
Sudo et al. 2D 60000 the bend curvature. The swirl intensities at the
Al-Rafai et al. 3.49D, 6.975D 34132
exit of the elbows also display similar
characteristics.

238
 Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)

REFERENCE 13. Shih TH, Zhu J (1993). A Realizable

Reynolds Stress Algebraic Equation Model.
1. Al-Rafai WN, Tridimas Y D, Woolley NH NASA TM-105993, ICOMP-92-27.
(1990). A study of turbulent flows in pipe 14. Shih TH, Liou WW, Shabbir A, Zhu J (1995).
bends. Journal of Mechanical Engineering A new k- eddy-viscosity model for high
Science 204:399-408. Reynolds number turbulent flowsModel
2. Chen YS, Kim SW (1987). Computation of development and validation. Computers and
Turbulent Flows Using an Extended k- Fluids 24(3):227-238.
Turbulence Closure Model. NASA CR- 15. Sudo K, Sumida M, Hibara H (1998).
179204 USA. Experimental investigation on turbulent flow
3. Chigier NA, Beer J M (1964). Velocity and in a circular- sectioned 90-degree bend.
static pressure distribution in swirling air jets Experiments in Fluids 25:42-49.
issuing from annular and divergent nozzles. 16. Tanaka M, Ohshima H, Monji H (2009).
Journal of Basic Engineering 4:788-796. Numerical investigation of flow structure in
4. Hellstrom LHO, Sinha A, Smits AJ (2011). pipe elbow with large eddy simulation
Visualizing the very-large-scale motions in approach. Proceedings of the ASME 2009
turbulent pipe flow. Physics of Fluids Pressure Vessels and Piping Division
23:011703. Conference PVP2009, July 26-30, Prague,
5. Homicz GF (2004), Computational Fluid Czech Republic, PVP-77598.
Dynamic Simulations of Pipe Elbow Flow. 17. Weske JR, (1948). Experimental
SAND Report, SAND2004-3467, USA. Investigation of Velocity Distributions
6. Lam CKG, Bremhorst KA (1981). Modified Downstream of Single Duct Bends. NACA
form of k- model for predicting wall TN-1471, USA.
turbulence. ASME Journal of Fluids and 18. Yadav MS (2013). Interfacial Area Transport
Engineering 103:456-460. across Vertical Elbows in Air-water Two-
7. Launder BE, Sharma BI (1974). Application phase Flow. Ph.D thesis, PSU, USA.
of the energy dissipation model of turbulence 19. Yakhot V, Orszag SA (1986).
to the calculation of flow near a spinning disk. Renormalization group analysis of
Letters in Heat and Mass Transfer 1(2):131- turbulence: I. Basic Theory. Journal of
138. Scientific Computing 1(1):1-51.
8. Lien FS, Leschziner MA (1993). 20. Yamano H, Tanaka M, Ono A, Murakami T,
Computational modeling of 3D turbulent flow Iwamoto Y, Yuki K, Sago H, Hayakawa S
in S-diffuser and transition ducts. Engineering (2009). Unsteady elbow pipe flow to develop
Turbulence Modeling and Experiments 2 a flow-induced vibration evaluation
(Edited by W. Rodi and F. Martelli), p. 217, methodology. Int. Conf. on Fast Reactors and
Elsevier Science Publishers. Related Fuel Cycles, Kyoto, Japan, 291-292.
9. Lien FS, Chen WL, Leschziner MA (1996). 21. Zagarola MV, Smits AJ (1997). Scaling of
Low-Reynolds-number eddy-viscosity the mean velocity profile for turbulent pipe
modeling based on non-linear stress- flow. Physics Review Letters 78:239-242.
strain/vorticity relations. Proc. 3rd
Symposium on Engineering Turbulence
Modeling and Measurements, 27-29 May,
Crete, Greece, 91-100.
10. Menter FR (1994). Two-equation eddy-
viscosity turbulence models for engineering
applications. AIAA Journal 32(8):15981605.
11. OpenCFD (2012). OpenFOAM: The Open-
source CFD toobox, programmers guide 2.1.
12. Rahimzadeh H, Maghsoodi R, Sakardeh H,
Tavakkol S (2012). Simulating flow over
spillways by using different turbulence
models. Engineering Applications of
Computational Fluid Mechanics 6(1):100-
109.

239