You are on page 1of 5

Proceedings of the 19th IAHR-APD Congress 2014, Hanoi, Vietnam

ISBN xxx-xxxx-xx-x

EDDY VISCOSITY TURBULENCE MODEL FOR INCOMPRESSIBLE FLUID FLOW IN CLOSED


CONDUITS
PRASHANTH REDDY HANMAIAHGARI 1,c, SOUMEN MAJI 2
1Departmentof Civil Engineering, Indian Institute of Technology, Kharagpur, WB, India
e-mail: hpr@civil.iitkgp.ernet.in
2Department of Civil Engineering, Indian Institute of Technology, Kharagpur, WB, India

e-mail: soumen@civil.iitkgp.ernet.in

ABSTRACT
Several methods have been proposed to account for unsteady friction effects during pressure surge oscillations.
Convolution integral methods are an alternative within the one-dimensional models category for laminar flows. These
methods use the past local accelerations and weighting functions in which computations are time consuming and require
large computer memory. The convolution integral method was extended to turbulent flow for smooth and for rough
pipes. These solutions provide acceptable approximations at the expense of numerical accuracy due to the approximation
of the convolution integral by a limited number of weighted coefficients. The above mentioned one-dimensional unsteady
friction models have been tested only for a very narrow set of unsteady flow conditions and the present state of
knowledge does not guarantee their performance in general transient flow situations. The assumptions in which these
models are based are likely to fail in many practical applications. In this research, equations of continuity and motion with
quasi two-dimensional eddy viscosity turbulence closure model is applied to an elastic circular pipe. Since the flow is
axisymmetric, the momentum and continuity equations are solved in cylindrical coordinates. The effect of roughness is
included in the form of Reynolds stresses in the momentum equation and model is tested for a rapid valve closure data at
different Reynolds numbers available in the literature. The pressure damping produced by the proposed model is quite
accurate.

Keywords: unsteady friction; RANS equations, turbulence; eddy viscosity; closed conduit flow

1. INTRODUCTION convolution based and instantaneous acceleration based


(Hanmaiahgari et al. 2012) one-dimensional unsteady
Let us consider a case of a piping system in which a fluid friction models had been tested only for a very narrow set
is flowing with a certain initial velocity. If the valve setting of unsteady flow conditions and the present state of
is changed, the velocity also changes leading to a pressure knowledge does not guarantee their performance in
wave travelling in the upstream direction from the valve general transient flow situations. Advanced turbulence
end. In such cases of unsteady flow situations, wall models need of the hour to model unsteady friction in all
friction influences both the magnitude and shape of types of transients in pipe flows.
transient waves in the pipes. The accurate prediction of
pressure waves is important for the design and analysis of The eddy viscosity turbulent closure model is
water pipelines, natural gas pipelines and pressurized implemented in this paper. The eddy viscosity model was
sewage pipeline systems. The presented method models developed by Silva-Araya and Chaudhry (1997) to
unsteady wall shear and turbulence behavior in pipe simulate unsteady friction during slow valve closure. The
transients. model computes velocity profiles in iterative method and
convergence was subjected to satisfying the continuity
Several methods had been proposed to account for equation. Energy dissipation is computed at every instant
unsteady friction effects during pressure surge during transient flow conditions and compared with
oscillations. With the exception of pezzingas model, most steady state energy dissipation and the ratio between
of the models are only applicable to hydraulically smooth them gives the energy dissipation factor. Tthhe non-
pipes. Nevertheless, in practical applications, rough pipes dimensional energy dissipation factor is time variant and
are common and several parameters in the fluid flow are space variant. The product of energy dissipation factor
affected by roughness, for example velocity profile, shear and steady state friction factor at any instant gives the
stress at the wall and the rate of energy dissipation during actual friction factor. The actual friction factor was used in
transient flow. Quasi-two dimensional models (Silva- the method of characteristics equations to compute flow
Araya and Chaudhry 1997; Pezzinga, 1999) are variables in time and space domain. The objective of the
computationally intensive and have not been extended present paper is to test Silva-Araya and Chaudhry (1997)
beyond simple pipe systems and simple boundary eddy viscosity model to simulate sudden valve closure
conditions. The exact Vardy Brown unsteady friction was experiments at different Reynolds numbers conducted by
used to model shear stress in transient turbulent pipe Adamkowski and Lewandowski (2006).
flows and the resulting water hammer equation was
2. METHODOLOGY
solved by the method of characteristics. However, the
approximate Vardy-Brown equation is more Generally frictional losses in unsteady pipe flow are
computationally efficient. The above mentioned estimated by using steady state friction factor formula,

1
1
which is known as a quasi steady approximation. This + = (1 + (2 + )2 + 1) [13]
2
approximation accurately predicts maximum pressure or Mixing length Equation:
minimum pressure in the absence of the two phase flow. + +
+ = + + (1 1 ) [14]
However, this method is not accurate for the prediction of +
Where, 1 = mixing length damping factor for flows with
the time history of pressure profile. The pressure damping zero pressure gradient.
in a real system is much higher as compared to pressure Turbulence model (smooth pipe):
profile computed by using the quasi steady 1+ =
+
[15]
+ + 1
approximation. The accurate calculation of attenuation ln[1
+ +
]
due instantaneous transient velocities is important for
long-term simulations to estimate water losses in the 2.5.1 Turbulence model for hydraulically smooth flows
system. To this end, eddy viscosity turbulence model was Fig.1 shows an inner region turbulence model for different
used here to simulate unsteady flows in closed conduits. roughness conditions. Sand grain roughness is given by
Detailed methodology is given in this paper. Prandtl-Schlisting (White 1992)
1
1 = ln(1 + 0.3 + ) [16]
Continuity and Momentum equations given in Chaudhry

(2014) are presented below. Where, + = and k is the roughness height of pipe wall.

2.1. Continuity equation: = 5.5 for smooth surfaces..

2
+ =0 [1]

2.2. Momentum equation:

4
+ + =0 [2]

where,
1
= 2 [3]
8
2.3. Energy Dissipation Factor:


= [4]

3
Where, = (2 1 ) [5]
2

= 2 2 0 [6]
1

= (
) [7]

2.4. Computational grid:


The velocity gradient is greater at the wall in turbulent
Figure 1. Inner region turbulence model
flow. It is important to have small grid steps at the wall. So
a geometrically expanding grid is required. The ratio of 2.5.2 Turbulence model (smooth-rough transition):
thickness of successive grid steps is constant and grid a) for (5.5 > 1 > 0.25)
The smooth pipe equation is given in following described
expands in geometric progression.
Eqs. [14] and [15]
Stepsize at ith interval given ratio of two successive grid
sizes is b) for (1.35 < 1 < 0.25)
= 1 [8] Granville (1985) presented + vs 1 , however, Silva and
Chaudhury (1997) obtained as
Distance of from pipe wall: 1+ = 5.025 + 3.1181 0.9 < 1 < 0.25 [17]
1 1+ = 2.07 11.1251 7.0812 1.35 < 1 < 0.9 [18]
= ( ) , = 1,2 . [9]
1
and mixing length may be computed from Eq. [14].
Total number of points in the y-direction:
ln[1+( 1)( )]
= [10] 2.5.3 Turbulence model for fully rough flows:
ln
Surface roughness is greater than the thickness of viscous
2.5. Turbulence Modelling: sublayer.
for (-8.3<C1< -1.35)
According to Prandtl mixing length theory, Reynolds stress 1+ = 0 [19]
is computed as + = + + +
+
[20]
1 1 1 2 1
+ + ) ln (2 + +
1 = (ln 4 1 + (
) = 2 | | 2 4

( [11]

2 1
The Reynolds stresses can be also computed by using eddy + + ))
2 [21]
4
viscosity concept proposed by Boussinesq. +
) = ( )
Eddy viscosity in outer region( ) is computed from
( [12]
following equations.
Eddy viscosity and mixing length models are compatible. +
= + [22]
The mixing length model is used in inner region, where, For the range 1 104 < < 2 106 the value of is
Eddy viscosity model is used for the outer region. given by (Silva Araya and Chaudhury, 1997)
Granville (1990) proposed following equation for = 0.4095 0.1390 log() + 0.0137 log()2 [23]
conversion between eddy viscosity and mixing length:
2
If < 1 104 = 0.07, > 2 106 = For steady state flow this expression reduces to,
0.075
= [34]
2
According to criterion given by Kita et. al. (1980), the outer
region starts when the following condition if fulfilled
0.154 +
+
= (1 + 1 ) [24] 2.9 Computation of the velocity profile:
0.077

Blending between inner region and outer region, eddy Mac Cormack scheme was chosen to compute velocity
viscosity is given by, profiles. Mac Cormack method was chosen for it is 2nd
+ +
+
+ = ( ) [25] order accurate and easily implementable. Mac Cormack
+ +
log( )
= + [26] method is a two-step Predictor and corrector method. In
( + )
predictor method forward finite difference is used and in
Velocity distribution in laminar flow is given by,
1 corrector method backward finite difference is used.
= ( ) (2 2 ) [27] 2.91 Predictor method:
4
Velocity distribution in turbulent flow is given by, Axisymmetric momentum equation was first solved using
+ = 0
+ +
+ [28] the predictor step. The approximations for the partial
1+ + derivatives in this step are given below.
2.6 Velocity gradient
Velocity gradient is given by, = [35]


+1 +( 2 1) 2 1 +1
[29] = [36]
(+1) +

2 2(+1 (+1) +1

)
2
= 2
[37]
(+1)( )
Where, = 1 is the ratio of two successive grid
Where, + = +1 , r - = ri -ri-1 and , = 1 .
spaces and r - = ri-1 -ri .
The subscript n and i refers to the time level and grid
The subscript i refers to i-th position inside the pipe cross
point. The asterisk indicates predicted values.
section. A second order backward approximation was
The momentum equation is solved for (Silva Araya and
used for the gradient at the wall
2 2 (1+)2 1
Chaudhury, 1997) as,
| [30] 1 +1
+1
(1+)(1 2 ) = + ( ) + ( + ) +
+

2(+1 (+1) +1

)
[38]
(+1)( )2

Outer 2.92 Corrector step:


region
The predicted values are used in the backward finite
Transition difference approximation for the first spatial derivatives
and are also used for the second spatial derivatives. The
approximations are,
Y
+ = [39]

1

= [40]


2 2(+1 (+1) +1

)
Inner region
= [41]
2 (+1)( )2
where, (**) indicates a corrected value. Therefore corrected
value is,
1 +1
+1
= + ( ) + ( + ) +
+

2(+1 (+1) +1

)
+ 2 [42]
(+1)( )
Finally, the velocity for the next time step is given by,
Figure 2. Eddy viscosity distribution +1 =

[43]
2
Where, N is the number of grid lines (the grid lines at the
wall is number N) 2.10 Eddy viscosity derivative:
The derivative of the eddy viscosity is obtained (Silva
2.7 Pressure gradient: Araya and Chaudhry, 1997) to use in computing velocity
Pressure gradient was obtained from the following profile,
+1 +( 2 +1) 2 1
expression derived from 2D axisymmetric momentum = [44]
(+1)(1 )
equation (White, 1991).
3. DETAILS OF EXPERIMENTS
1 2
= ( ) [31] In this paper, pressure profiles in water hammer
=
experiments measured by Adamkowski and
2.8 Shear stress distribution: Lewandowski (2006) for sudden valve closures had been
Shear stress obtained from the following equation, used to validate the eddy viscosity unsteady friction
+ =



+

[32] model. The experimental setup was constructed at the
0 2
Institute of Fluid-Flow Machinery of the Polish Academy
The time derivative is approximated as,
+
of Sciences in Gdansk in order to test transient pipe flows
[33] in a wide range of Reynolds numbers. The schematic of

Where, is time step chosen to solve the the experimental setup was given in Adamkowski and
characteristic equations. Lewandowski (2006). Copper pipe of L=98.11 m long with
an internal diameter, D=0. 016 m and a wall thickness of
3
e=0. 001 m was used in the experiments. An instantaneous viscosity turbulence model for the sudden closure of a
closing ball valve was installed at the downstream end of downstream valve. The piping system has a pressurized
the pipe and a high pressure tank was provided at the tank at the upstream end and the downstream boundary
upstream end. The experimental setup was provided with condition is specified by a rapid valve closure. Details of
four absolute pressure transmitters of 0.2% accuracy at two experimental runs at Reynolds numbers 10600, 15800
every quarter of pipe length. A turbine flow meter with
were given in the section of details of experiments. The
1% accuracy was used to measure the flow rate to
measured time histories of pressures at the downstream
calculate average flow velocity. Pressures were measured
end of the pipeline at two Reynolds numbers were
at a frequency of 2.5 kHz at every L/4 distance of pipe
length after rapid closure of the downstream valve. compared in Figures 3 and 4 with that of simulated
During the tests, steady water head Ho = 127.5 m was pressures obtained by using the eddy viscosity turbulence
maintained in the upstream tank. The sudden valve model. The non dimensional pressure and time wre
closure experiments were conducted at different initial represented by ( )/( ) and / respectively, in
conditions, 0 = 0.94 ( 15800) and 0 = which is instantaneous pressure head at the node; is
0.63 ( 10600) respectively. Experiments were pressure wave velocity; is time instant; and is
conducted at 22.6C and corresponding fluid properties: acceleration due to gravity. In the method of characteristics
kinematic viscosity of water = 9.493 107 2 /; bulk and were considered as 2.45 m and 0.0019 s. The ratio
modulus ( ) = 201 109 ; and density () = of two successive grid sizes ( ) = 1.15 was used. The time
1000 /3 . The pipe material properties: poissons step = 6.8 s was used to compute pressure gradients,
coefficient () = 0.35, density ( ) = 8890 /3 and velocity distribution and dissipation integral within one
Youngs modulus () = 120 109 . The pressure wave MOC time step. Simulations were carried out for 3.75 s
velocity () was estimated as 1298.4 m/s.
long. The pressure variation obtained with the proposed
eddy viscosity turbulence model matches with that of
4. RESULTS AND DISCUSSION
experiments. The proposed method was able to capture the
peaks as well as the phase of pressure profile resulting
Reynolds no = 10600
1.5 from the rapid valve closure.
Eddy viscosity model
Experimental

1
5. CONCLUSIONS
non dimensional pressure

0.5
In this paper, an eddy viscosity turbulence model was
presented for transient simulations in water pipeline
systems. It was demonstrated that this method can be used
0
to model instantaneous valve closure in water pipelines.
The performance of the eddy viscosity turbulence model is
-0.5 excellent, during rapid transients and it was proved that
the pressure profiles were accurately captured by the
proposed method. The pressure damping in the
-1
0 5 10 15 20 25 30 35 40 45 50 simulations was equal to the experiments. The proposed
non dimensional time
method was computationally intensive, which needs higher
Figure 3. Measured and computed pressure profiles at the computational time as compared to convolution based
valve for rapid closure with initial steady velocity, Vo = methods and instantaneous acceleration based methods.
0.63 m/s The advantage of the eddy viscosity model was
Reynolds no = 15800 computation of 2D velocity profiles during unsteady
1.5
Eddy viscosity model
turbulent flows. Best use of this method is to simulate
Experimental transient pressures and flow rates during more than one
1 unsteady boundary condition present in the system akin to
a water distribution system where multiple transients do
non dimensional pressure

0.5
occur simultaneously. This method can be used long term
simulations to accurately locate leakages and to estimate
unaccounted water with minimum error in water
0
distribution systems.

6. REFERENCES
-0.5

Adamkowski, A., and Lewandowski, M., (2006)


-1
0 5 10 15 20 25 30 35 40 45 50 "Experimental Examination of Unsteady Friction
non dimensional time Models for Transient Pipe Flow Simulation, Journal of
Figure 4. Measured and computed pressure profiles at the Fluids Engineering, ASME, 128(11), 1351-1363
valve for its rapid closure with initial flow conditions of Vo Chaudhary.M. Hanif, (2014), Applied Hydraulic Transients,
= 0.94 m/s 3rd edition, Springer, New York.
Granville, P.S. (1985). Mixing Length Formulations for
An experimental piping system available in the literature turbulent boundary layers over arbitrary rough
(Adamkowski and Lewandowski, 2006) was analyzed surfaces, Journal of Ship Research, 29, No. 4: 223-233.
herein to demonstrate the performance of the eddy Granville, P.S. (1989). A modified Van Driest Formula for
the mixing length of turbulent boundary layers in
4
pressure gradients, Journal of Fluids Engineering
ASME,111, 94-97.
Hanmaiahgari, P. R., Silva-Araya, W. F., and Chaudhry,
M. H. (2012). Estimation of Decay Coefficients for
Unsteady Friction for Instantaneous Acceleration
Based Models. Journal of Hydraulic Engineering, 138 (3),
260-271.
Pezzinga.G, (1999): Quasi-2D Model for Unsteady Flow in
Pipe Networks, Journal of Hydraulic Engineering,
Vol.125, pp 676-685.
Silva-Araya, W.F., and Chaudhry, M.H. (1997).
Computation of Energy Dissipation in Unsteady Flow
in Pipes. Journal of Hydraulic Engineering, American
Society of Civil Engineers, 123 (2), 108-115.
Vardy, A. and Brown, J. (2004). Efficient Approximation of
Unsteady Friction Weighting Functions. J. Hydraul.
Eng., 130 (11), 10971107.
White. F., (1992) Viscous Flow, Mc Graw Hill, 2nd edition.
Kita Y., Adachi, Y. and Hirose, K. (1980). Periodically
Oscillating Turbulent Flow in a Pipe, Bulletin of the
JSME 23, No. 179: 656-664

Copyrights
Paper(s) submitted to the IAHR-APD2014 are interpreted
as declaration that the authors obtained the necessary
authorization for publication.

You might also like