Professional Documents
Culture Documents
Fluid Flow in T-Junction of Pipes PDF
Fluid Flow in T-Junction of Pipes PDF
1
,
2
,
3
Coefcients in Vazsonyis formulas (dimensionless)
(x, t) Turbulent dissipation rate
Fluid Viscosity, Pa s
t
Turbulent viscosity, Pa s
Symmetric stress tensor
k
Turbulent Prandtl number for k
Shear stress, Pa
Kinematic viscosity of uid
Angle between main pipe and branch
viii
Mathematical Conventions
log(x) logarithm base 10 of x
e
x
exponential of x-that is, e raise to the power of x
n
i=1
a
i
the sum from i=1 to n that is, a
1
+ a
2
+ . . . + a
n
n
i=1
a
i
the product from i=1 to n that is, a
1
a
2
. . . a
n
f(x)/ x partial derivative of f with respect to x
=
_
x
1
, . . . ,
x
n
_
Vector differential operator (gradient)
=
_
2
x
2
1
, . . . ,
2
x
2
n
_
Laplace operator (nabla)
(cu) =
x
1
_
c
u
x
1
_
+ . . . +
x
n
_
c
u
x
n
_
u =
1
_
u
x
1
_
+ . . . +
2
_
u
x
n
_
_
b
a
f(x) the integral of f with respect to x
F(x;) function of x, with implied dependence upon
Mathematical Operations
equivalent to (or dened to be)
proportional to
ix
List of Figures
1.1 Water Distribution in city and industries. . . . . . . . . . . . . . . . . . . 1
1.2 Various possibilities of uid entering and leaving the junction . . . . . . . 2
2.1 Finite Element Discretization of the domain and Weak formulation . . . . 4
2.2 Control volume variants used in the nite volume method: cell-centered
and vertex-centered control volume . . . . . . . . . . . . . . . . . . . . . 6
3.1 Elemental volume used to derive the equations . . . . . . . . . . . . . . . 8
3.2 Fluid type Newtonian/conventional uids vs. non-Newtonian uids . . . . 13
3.3 Use of boundary conditions with Comsol . . . . . . . . . . . . . . . . . 19
4.1 Fluid behavior when pipe is smooth or rough from inside . . . . . . . . . 22
4.2 Moody chart for estimating Frictional factor . . . . . . . . . . . . . . . . 28
5.1 Example of ow situations and angles for combining and dividing ow . . 31
5.2 Plot of
3
(left) and Plot of and (right) . . . . . . . . . . . . . . . . . 32
5.3 Diagram for combining ow . . . . . . . . . . . . . . . . . . . . . . . . 32
5.4 T-junction as combination of an elbow and a contraction . . . . . . . . . 34
5.5 T-junction as combination of two elbows . . . . . . . . . . . . . . . . . . 35
6.1 Cross-section plot for example case of ow in T-junction . . . . . . . . . 36
6.2 Comparison of head-loss by classical formula and head loss by software
of an example cases of ow in T-junction . . . . . . . . . . . . . . . . . 37
6.3 Combining ow: Case-1 . . . . . . . . . . . . . . . . . . . . . . . . . . 37
x
6.4 Head loss for Combining ow: Case-1, Radius of branches is 0.5 cms,
Inlet velocities vary from 1 cm/sec to 3 cm/sec (25 different cases plot-
ted), Outlet pressure is 100 Pascals and classical pressure loss formula by
Andrew Vazsonyi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.5 Combining ow: Cases-2 . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.6 Head loss for Combining ow: Case-2, Radius of branches is 0.5 cms,
Inlet velocities vary from 1 cm/sec to 3 cm/sec (25 different cases plot-
ted), Outlet pressure is 100 Pascals and Classical pressure loss formula by
A. Gardel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.7 Dividing ow: Case-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.8 Head loss for dividing ow: Case-1, Radius of branches is 0.5 cms, Inlet
velocity vary from 1 cm/sec to 3 cm/sec, at both outlet pressure is 100
Pascals and Classical pressure loss formula by A. Gardel . . . . . . . . . 40
6.9 Dividing ow: Case-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.10 Head loss for dividing ow: Case-2, Radius of branches is 0.5 cms, Inlet
velocity vary from 1 cm/sec to 3 cm/sec, at both outlet pressure is 100
Pascals and Classical pressure loss formula by A. Gardel . . . . . . . . . 41
6.11 T-junction with different angles between main pipe and branch pipe . . . 41
6.12 Head loss for T-junction with angle = 91, combining ow: Case-1,
Radius of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3
cm/sec (25 different cases plotted), Outlet pressure is 100 Pascals and
Classical pressure loss formula by Andrew Vazsonyi . . . . . . . . . . . 42
6.13 Head loss for T-junction with angle = 92, combining ow: Case-1,
Radius of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3
cm/sec (25 different cases plotted), Outlet pressure is 100 Pascals and
Classical pressure loss formula by Andrew Vazsonyi . . . . . . . . . . . 43
xi
6.14 Head loss for T-junction with angle = 93, combining ow: Case-1,
Radius of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3
cm/sec (25 different cases plotted), Outlet pressure is 100 Pascals and
Classical pressure loss formula by Andrew Vazsonyi . . . . . . . . . . . 44
6.15 Head loss for T-junction with angle = 87, combining ow: Case-1,
Radius of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3
cm/sec (25 different cases plotted), Outlet pressure is 100 Pascals and
Classical pressure loss formula by Andrew Vazsonyi . . . . . . . . . . . 45
6.16 Head loss for T-junction with angle = 88, combining ow: Case-1,
Radius of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3
cm/sec (25 different cases plotted), Outlet pressure is 100 Pascals and
Classical pressure loss formula by Andrew Vazsonyi . . . . . . . . . . . 45
6.17 Head loss for T-junction with angle = 89, combining ow: Case-1,
Radius of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3
cm/sec (25 different cases plotted), Outlet pressure is 100 Pascals and
Classical pressure loss formula by Andrew Vazsonyi . . . . . . . . . . . 46
6.18 Head loss for different angle of T-junction, combining ow: Case-2, Ra-
dius of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3
cm/sec, Outlet pressure is 100 Pascals and Classical pressure loss for-
mula by A. Gardel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.19 Dividing ow: Case-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.20 Head loss for area case-1, combining ow case-1, Radius of main pipe is
branches is 0.25 cms, Radius of perpendicular pipe is branches is 1 cms,
Inlet velocity in both inlets vary from 1 cm/sec to 3 cm/sec, pressure
at outlet is 100 Pascals and Classical pressure loss formula by Andrew
Vazsonyi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.21 Dividing ow: Case-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
xii
6.22 Head loss for area case-2, combining ow case-1, Radius of main pipe is
branches is 0.3 cms, Radius of perpendicular pipe is branches is 1 cms,
Inlet velocity in both inlets vary from 1 cm/sec to 3 cm/sec, pressure
at outlet is 100 Pascals and Classical pressure loss formula by Andrew
Vazsonyi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.23 Head loss for area case-1, combining ow case-1, Radius of main pipe is
branches is 0.25 cms, Radius of perpendicular pipe is branches is 1 cms,
Inlet velocity in both inlets vary from 1 cm/sec to 3 cm/sec, pressure
at outlet is 100 Pascals and Classical pressure loss formula by Andrew
Vazsonyi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.24 Head-loss for different cross-section areas of branches of T-junction, A1 =area
of main pipe, A2 =area of branch pipe, combining ow: Case-1, Radius
of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3 cm/sec,
Outlet pressure is 100 Pascals and Classical pressure loss formula by A.
Gardel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
11.1 Buttons for drawing geometry . . . . . . . . . . . . . . . . . . . . . . . 59
xiii
List of Tables
1 Relative roughness for some common materials determined by experiments. 25
2 Reynolds Number, Nature of Flow and Friction coefcient (). . . . . . . 26
3 Minor loss coefcients for some of the most common used components
in pipe and tube systems . . . . . . . . . . . . . . . . . . . . . . . . . . 27
xiv
1 Introduction
Pipe networks are mainly used for transportation and supply of uids and gases. These
networks vary from fewer pipes to thousands of pipes (e.g. water supply network of a
large city, see in gure 1.1). In addition to pipes, the network also consists of elbows,
T-junctions, bends, contractions, expansions, valves, meters, pumps, turbines and many
other components. All these components cause loss in pressure due to change in momen-
tum of the ow caused due to friction and pipe components. This means conversion of
ow energy in to heat due to friction or energy lost due to turbulence.
Pipe networks are very common in industries, where uid or gases are to be transported
from one location to the other. The head loss (pressure loss) may vary depending on
the type of components occurring in the network, material of the pipe and type of uid
transported through the network. In industries the networks are usually large and require
very precise pressure at certain points of network. It is also sometimes essential to place
valves, pumps or turbines of certain capacity to control pressure in the network. The
placement of valves, pumps and turbines is important to overcome pressure loses caused
by other components in the network. This is one of the important reasons why this study
was conducted.
Figure 1.1: Water Distribution in city and industries.
In this work we have concentrated our attention to a very small and common component
of pipe network: T-junction (Some also refer as Tee). T-junction is a very common
component in pipe networks, mainly used to distribute (diverge) the ow from main pipe
1
to several branching pipes and to accumulate (converge) ows from many pipes to a single
main pipe. Depending on the inow and outow directions, the behavior of ow at the
junction also changes. The following gure shows some possibilities of uid entering and
leaving the junction.
Figure 1.2: Various possibilities of uid entering and leaving the junction
In present work we will numerically simulate the uid ow in T-junction of pipes with
Comsol Multiphysics and FLUENT. The results obtained by software were compared
with available classical formula and formulas constructed by assuming T-junction to be
made up of two different components. This comparison also helped in verication of
some loss coefcients used in classical formula.
In uid dynamics, head is the difference in elevation between two points in a column of
uid, and the resulting pressure of the uid at the lower point. It is possible to express
head in either units of height (e.g. meters) or in units of pressure such as Pascals. When
considering a ow, one says that head is lost if energy is dissipated, usually through
turbulence; equations such as the Darcy-Weisbach equation have been used to calculate
the head loss due to friction.
Head losses are of two types major and minor. Major head losses (also called Frictional
losses) are due to rough internal surface of pipe and occur over length of pipe. They are
mainly due to friction. Minor losses are losses due to the change in uid momentum.
They are mainly due to pipe components due to bends, valves, sudden changes in pipe
diameter, etc. Minor losses are usually negligible compared to friction losses in larger
pipe systems. Presence of additional components offer resistance to ow and turbulence.
In this work, our aim is to study behavior of uid at T-junction of pipes, head losses
caused by T-junction and change in pressure loss with change in angle of the junction.
2
2 CFD tools used
In this chapter we present an assortment of mathematical methods that we have used in
this study. This chapter includes overview of the CFD methods Finite Element method
(FEM) and Finite Volume Method (FVM).
We begin this section with a small introduction to FEM. This will include overview and
basic steps of FEM. Then, we will introduce FVM and also give basic steps of it.
2.1 Finite Element Method
The essence of the Finite Element Method (FEM) is to take a complex problem whose
solution may be difcult if not impossible to obtain, and decompose it into pieces upon
each of which a simple approximation of the solution may be constructed, and then put
the local approximate solutions together to obtain a global approximate solution. FEM is
widely used to nd approximate solutions of differential equations which are not solvable
with analytical methods or which have geometrically complex domains. There are com-
mercial software packages like Comsol Multiphysics and ANSYS available for usage.
In FEM, we divide, domain
2
of the boundary value problem into a number of
closed sub-regions called elements ({e
l
}
L
l=1
). When we do this we take following precau-
tions
1. Avoid very large and very small angles.
2. Element should be placed most densely in region where the solution of the problem
and its expected to vary rapidly.
3. High accuracy requires a ne mesh or many nodes per element.
Suppose that for a given nite element mesh there is associated with each node N
i
=
(x
i
, y
i
) a function, dened on
with certain properties (see appendix-C), this function
is called Elements basis functions. Local basis function over element e
l
is simply the
restriction of global element basis function of e
l
.
This method involves simple steps as described briey.
3
1. Discretization of the domain: Discretize the geometrically complex domain into
set of nite elements called elements. We can divide the domain into desired num-
ber of elements and desired number of nodes. These elements are non-overlapping.
It can be easily observed that the elements have simple geometrical form and are
only part of the very complex looking geometry and nodes are the points where
these elements meet. For 1-D the elements are intervals, for 2-D the elements are
triangles or quadrilaterals.
2. Weak formulation of the differential equation over elements: Multiply the equa-
tion by a weight function and integrate the equation over the domain. Distribute the
differentiation among the weight function. Use the denition of the natural bound-
ary condition in the weak form.
Figure 2.1: Finite Element Discretization of the domain and Weak formulation
3. Local Approximation of Solution: On each element let us attempt to compute the
length. We assume that the length of each arc can be approximated by the length of
the chord i.e. we approximate the arc using a straight line.
4. Assemble the Element Equations: Collect the element equations to get a repre-
sentation of the whole system. Assemble the element equations to obtain the global
system of equations.
5. Imposition of boundary conditions.
4
6. Solution of the algebraic system of equations: Obtain the Solution of standard
matrix equation by direct or indirect (iterative) method.
7. Post processing: This nal operation displays the solution to system equations in
tabular graphical or pictorial form. Other meaningful quantities may be derived
from the solution and also displayed.
The nite element solution converges to the true solution as the number of elements is
increased. FEM is easy to use and it is also easy to approximate the differential terms of
higher order. This method demands a good engineering judgment. The choice of type of
element and other basis functions can be crucial.
2.2 Finite Volume Method
The Finite Volume Method (FVM) is a numerical method based on Integral conservation
law. These methods are used for solving partial differential equations that calculates the
values of the conserved variables averaged across the volume. The integral conserva-
tion law is enforced for small control volumes dened by the computational mesh. One
advantage of FVM over FDMs is that it does not require a structured mesh (although a
structured mesh can also be used). Furthermore, FVM is preferable to other methods as
a result of the fact that boundary conditions can be applied non-invasively. This is true
because the values of the conserved variables are located within the volume element, and
not at nodes or surfaces. FVMs are especially powerful on coarse, non-uniform grids and
in calculations where the mesh moves to track interfaces or shocks.
The FVMs are very efcient in solving conservative problems. They are extensively used
in uid mechanics and many other engineering areas governed by conservative systems
that can be written in integral control volume form. The primary advantages of these
methods are numerical robustness, applicability on very general unstructured meshes,
and the intrinsic local conservation properties of the resulting schemes.
To use FVM concrete choice of control volumes, type of approximation inside them and
numerical methods for evaluation of integrals and uxes are required to be chosen care-
fully in advance. This method (Based on the control volume formulation of analytical
uid dynamics) involves simple steps as described briey.
1. In FVM, computational domain is rst tessellated into a collection of non overlap-
5
ping control volumes that completely cover the domain i.e. to divide the domain
into a number of control volumes where the variable of interest is located at the cen-
troid of the control volume. The control volumes are divided in to two categories:
cell-centered and vertex-centered control volume (See g 2.2) . In the cell-centered
nite volume method shown, the triangles themselves serve as control volumes with
solution unknowns (degrees of freedom) stored on a per triangle basis. In the vertex-
centered nite volume method shown, control volumes are formed as a geometric
dual to the triangle complex and solution unknowns stored on a per triangulation
vertex basis. The following gures give clear idea about type of control volumes in
1D, 2D and 3D.
Figure 2.2: Control volume variants used in the nite volume method: cell-centered and
vertex-centered control volume
2. Integrate the differential formof the governing equations (very similar to the control
volume approach) over each control volume.
3. Interpolation proles are then assumed in order to describe the variation of the
concerned variable between cell centroids. The resulting equation is called the
discretized or discretization equation. In this manner, the discretization equation
expresses the conservation principle for the variable inside the control volume.
6
The most compelling feature of the FVM is that the resulting solution satises the con-
servation of quantities such as mass, momentum, energy, and species. This is exactly sat-
ised for any control volume as well as for the whole computational domain and for any
number of control volumes. Even a coarse grid solution exhibits exact integral balances.
FVM is the ideal method for computing discontinuous solutions arising in compressible
ows. Any discontinuity must satisfy the Rankine-Hugoniot jump condition which is a
consequence of conservation. Since FVMs are conservative they automatically satisfy the
jump conditions and hence give physically correct weak solutions.
7
3 Governing Equations and Boundary Conditions
The ow of most uids can be mathematically described by the use of continuity equation
and momentum equation. According to continuity equation, the amount of uid entering
in certain volume leaves that volume or remains there and according to momentum equa-
tion tells about the balance of the momentum. The momentum equations are sometimes
also referred as Navier-Stokes (NS) equation. They are most commonly used mathemati-
cal equations to describe ow. In this section we shall rst derive NS equations and then
K-Epsilon model. At the end we shall also briey discuss boundary conditions used.
In this section, we shall derive Navier-Stokes equations by control volume method, the
simplest approach. These equations can be used to describe many ow situations. Being
second order, non-homogeneous, non-linear partial differential equations we require at
least two boundary conditions for obtaining solution.
3.1 Continuity equation
Consider a volume of uid in the stream with dimensions x,y and z. Consider that
the uid ow is in positive x direction. Thus, the the amount of uid that enters the
volume from face-1 is equal to product of density (), velocity of uid in x-direction (u)
and area of the face-1 (yz). Thus,
volumein
x
= uyz (3.1)
Figure 3.1: Elemental volume used to derive the equations
8
The mass leaving from face-2 is negative (its leaving the volume) product of density,
velocity of uid in x-direction and area of the face-2. But, the density and velocity of the
uid changes from u to u + u and to + . Thus,
volumeout
x
= (u + u)( + )uyz (3.2)
Similarly, for other two faces parallel to y-axis, the equations for mass entering and leav-
ing will be
volumein
y
= vxz (3.3)
volumeout
y
= (v + v)( + )vxz (3.4)
And, for other two faces parallel to z-axis, the equations for mass entering and leaving
will be
volumein
z
= wxy (3.5)
volumeout
z
= (w + w)( + )wxy (3.6)
Also, the total amount of uid accumulated in the volume xyz is
_
t
_
xyz (3.7)
This amount must be equal to the numerical sum of all the terms representing uid en-
tering the volume and uid leaving from the volume. Adding equations (3.1) to (3.7),
equating to 0 and using (fg) = fg + gf + fg, we get
_
t
_
=
((u)) uyz ((v)) vxz ((w)) wxy
xyz
(3.8)
t
_
+
(u)
x
+
(v)
y
+
(z)
z
(3.9)
And when, t 0, we can replace operator by partial differential operator.
t
+
(u)
x
+
(v)
y
+
(w)
z
= 0 (3.10)
Which is general Continuity equation for compressible uid. For incompressible uids
the Continuity Equation reduces to
u
x
+
v
y
+
w
z
= 0 (3.11)
9
Also, if the density is a function of co-ordinates x, y and z but not time then,
(u)
x
+
(v)
y
+
(w)
z
= 0 (3.12)
3.2 Navier-Stokes equation
Navier-Stokes (NS) equations are system of momentum equations for each co-ordinate
directions. We shall derive the equation only for x co-ordinate and then write for y and
z similarly. First we shall calculate Momentum Change and Flux and then calculate the
forces.
3.2.1 Momentum Change and Flux
Consider a volume of uid in the stream with dimensions x,y and z. The change in
momentum with respect to time is given by ((u)/t) xyz.
The ux of momentum in the x direction at face-1 of the volume is the product of the mass
ux (u), the x-direction velocity (u) and the area of face-1 (yz) i.e. uuyz. The
ux of momentum in the face opposite to face-1 is [uu + ((uu)) /xx] yz.
Similarly, for faces parallel to y-axis the ux of momentum in the y direction is vuxz
and the ux of momentum in the opposite to face is [vu + ((vu)) /yy] xz.
And, for faces parallel to y-axis the ux of momentum in the z direction at entering
face of the volume is wuxy and the ux of momentum in the opposite to face is
[wu + ((wu)) /zz] xy. Adding all these terms and simplifying we get,
_
(uu)
x
xyz +
(vu)
y
xyz +
(wu)
z
xyz
_
(3.13)
According to conservation of momentum law, algebraic sum of all these uxes of momen-
tum and the external forces at faces parallel to x-axis (
F
x
) should be equal to change
10
in momentum in volume with respect to time i.e.
(u)
t
xyz =
_
(uu)
x
xyz +
(vu)
y
xyz +
(wu)
z
xyz
_
+
F
x
(3.14)
Re-arranging , we get
t
(u) +
(uu)
x
+
(vu)
y
+
(wu)
z
_
xyz =
F
x
(3.15)
Applying the derivative of product rule we get,
_
u
t
+ u
t
+ u
(u)
x
+ u
u
x
+ v
(u)
y
+ v
u
y
+ w
(u)
z
+ w
u
z
_
xyz =
F
x
(3.16)
Rearranging the terms we get,
_
u
_
t
+
(u)
x
+
(u)
y
+
(u)
z
_
+
u
t
+ u
u
x
+ v
u
y
+ w
u
z
_
xyz =
F
x
(3.17)
The terms in square bracket sum up to zero because of equation of continuity. Thus, above
equation reduces to momentum equation given below
_
u
t
+ u
u
x
+ v
u
y
+ w
u
z
_
xyz =
F
x
(3.18)
Similarly, we can obtain,
_
v
t
+ u
v
x
+ v
v
y
+ w
v
z
_
xyz =
F
y
(3.19)
and
_
w
t
+ u
w
x
+ v
w
y
+ w
w
z
_
xyz =
F
z
(3.20)
3.2.2 Calculating Forces
The external force
F
x
,
F
y
and
F
z
which are external forces on the considered
volume. These forces are of two types: Body forces (acting on volume) and surface
forces (acting on surfaces).
Body forces are mostly due to gravitational forces acting on the uid. The total body
11
force acting on the volume considered is the product of component of acceleration due to
gravity in x-direction, mass of the uid in the volume i.e.
g
x
xyz (3.21)
Surface forces act on only one particular surface of the volume at a time, and arise due
to pressure or viscous stresses. The stress on a surface of the control volume acts in the
outward direction, and is given the symbol
ij
with two subscripts. The rst subscript
i indicates the normal direction of the face on which the stress acts, while the second
subscript j indicates the direction of the stress.
The force due to the stress is the product of the stress and the area over which it acts.
Thus, on the faces with normals in the x-direction (DyDz), the forces acting in the x-
direction due to the direct stresses are
xx
yz and
_
xx
+
xx
x
x
_
yz Which sum
to
_
xx
x
_
xyz.
Similarly, on the faces with normals in the y-direction (xz), the forces in the x-
direction due to shear stresses sum to
yx
x
xyz and on the faces with normals
in the z-direction (xy), the forces in the x-direction due to shear stresses sum to
_
zx
x
_
xyz.
The sum of all surface forces in the x-direction is thus
_
xx
x
+
yx
x
+
zx
x
_
xyz (3.22)
The stress
xx
includes the pressure p (negative sign because it is acting inward) and the
normal viscous stress
xx
. The stresses
yx
and
zx
include only viscous shearing stresses
yx
and
zx
. This gives the force in the x-direction as:
_
p
x
+
xx
x
+
yx
y
+
zx
z
_
xyz (3.23)
3.2.3 Newtonian/Non-Newtonian Fluids
A Newtonian uid is one whose stress at each point is linearly proportional to its strain
rate at that point. The best example of this is water. A non-Newtonian uid is one whose
12
viscosity changes with the applied strain rate. Thus, we can say that non-Newtonian uids
do not have a well-dened viscosity. The following gure can give a better idea of how
uids can be classied in Newtonian and other type of uids.
Figure 3.2: Fluid type Newtonian/conventional uids vs. non-Newtonian uids
A simple equation to describe Newtonian uid behavior is =
du
dx
. In common terms,
this means the uid continues to ow, regardless of the forces acting on it. If the uid is
incompressible and viscosity is constant across the uid, the equation governing the shear
stress, in the Cartesian coordinate system, is
ij
=
_
dU
i
dX
j
+
dU
j
dX
i
_
(3.24)
Where U = (u, v, w) and X = (x, y, z). Thus,
xx
=
_
du
dx
+
du
dx
_
= 2
du
dx
,
yx
=
_
dv
dx
+
du
dy
_
,
zx
=
_
dw
dx
+
du
dz
_
(3.25)
Substituting these values in equation obtained above, we get,
_
_
p
x
+
_
2
_
du
dx
__
x
+
_
dv
dx
+
du
dy
__
y
+
_
dw
dx
+
du
dz
__
z
_
_
xyz (3.26)
_
p
x
+ 2
2
u
x
2
+
_
2
u
y
2
+
_
dv
dx
_
y
+
2
u
z
2
+
_
dw
dx
_
z
__
xyz (3.27)
13
The terms
2
u/x
2
, (dv/dx) /y and (dw/dx) /z cancel out due to continuity equa-
tion. The terms that remain along with the body force due to acceleration due to gravity
would give the equation for the force in the x-direction,
F
x
=
_
g
x
p
x
+
_
2
u
x
2
+
2
u
y
2
+
2
u
z
2
__
xyz (3.28)
Substituting this in momentum equation, we get
_
u
t
+ u
u
x
+ v
u
y
+ w
u
z
_
xyz = g
x
p
x
+
_
2
u
x
2
+
2
u
y
2
+
2
u
z
2
_
(3.29)
Similarly, we can obtain,
_
v
t
+ u
v
x
+ v
v
y
+ w
v
z
_
xyz = g
y
p
y
+
_
2
v
x
2
+
2
v
y
2
+
2
v
z
2
_
(3.30)
and
_
w
t
+ u
w
x
+ v
w
y
+ w
w
z
_
xyz = g
z
p
z
+
_
2
w
x
2
+
2
w
y
2
+
2
w
z
2
_
(3.31)
These are the Navier-Stokes equations. There have been attempts to solve these equations
but the computational complexity involved has not allowed many but some solutions.
Navier-Stokes equation can be solved numerically, but the solutions are obtained after
only making some assumptions and some of them are not stable at high Reynolds number.
There are two important issues that arise in the solution process rst is non-linearity of
the equations and second is the coupling of the equations. In CFD the stress tensor terms
are often approximated by a turbulence model. The non-linearity makes most problems
difcult or impossible to solve and is part of the cause of turbulence.
3.3 Turbulence
Dictionary meaning of turbulence is the state of being turbulent and turbulent means dis-
turbed. When we talk about turbulence in uid dynamics it means uid ow with violent
disorder where the disorder has no specic direction or pattern. Also, its quoted as a
14
random secondary motion caused by eddies with in the uid in motion. Even though tur-
bulence is an everyday experience, it is extremely difcult to nd solutions, quantify, or in
general characterize. When the ow is turbulent, we can expect a very rapid and random
change in uid and uid motion properties like momentum diffusion, high momentum
convection, variation of pressure and velocity in space and time. Its difcult to express
turbulence mathematically for following reasons.
1. Irregularity or randomness: impossible to apply a deterministic approach.
2. Diffusivity: This characteristic causes rapid mixing and increased rate of momen-
tum, heat and mass transfer.
3. Large Reynolds number: Turbulent ow or instable laminar ow.
4. 3D Vorticity uctuations: Turbulence is 3D and rotational. Turbulence is character-
ized by high levels of uctuating vorticity.
5. Dissipation: Turbulence ows are always dissipative. Viscous shear stress performs
deformation work which increases the internal energy of the uid at expense of
kinetic energy of the turbulence. A continuous energy supply is needed to keep up
these loses. If no energy is supplied turbulence decays rapidly.
The K-epsilon model is one of the most common turbulence models. It includes two trans-
port equations to represent the turbulent properties of the ow. This allows a two equation
model to account for history effects like convection and diffusion of turbulent energy. The
rst transported variable is turbulent kinetic energy (k). The second transported variable
in this case is the turbulent dissipation (). These variables determine the scale of the tur-
bulence and energy in the turbulence. In next part, we shall derive Kappa-Epsilon model
from Incompressible NS equations.
3.4 Kappa-Epsilon Model
The K-epsilon model is most commonly used to describe the behavior of turbulent ows.
It was proposed by A.N Kolmogrov in 1942, then modied by Harlow and Nakayama and
produced K-Epsilon model for turbulence.
15
The Transport Equations for K-Epsilon model are For k,
t
(k) +
t
(ku
i
) =
x
j
__
+
t
k
_
k
x
j
_
+ P
k
+ P
b
Y
k
+ S
k
(3.32)
For ,
t
()+
t
(u
i
) =
x
j
__
+
t
k
_
x
j
_
+C
1
k
(P
k
+ C
3
P
b
)C
2
2
k
+S
(3.33)
Realizable k-epsilon model and RNG k-epsilon model are some other variants of K-
epsilon model. K-epsilon model has solution in some special cases. K-epsilon model
is only useful in regions with turbulent, high Reynolds number ows.
3.5 Derivation
K-epsilon model equations can be derived form incompressible Navier stokes equation.
(u.)u =
_
pI +
_
u + (u)
T
__
+ F (3.34)
.u = 0 (3.35)
Where, u is velocity vector eld, p is pressure eld, following are steps for deriving k-
epsilon model.
1. Apply statistical averaging to NS equation (3.35)
_
u
i
t
+
j
u
j
u
j
x
j
_
=
p
x
i
+
2
u
i
(3.36)
Where, u(x, t) represents the velocity vector eld, p(x, t) is the pressure eld. Be-
ing derived from Equations of conservation of mass, momentum and energy, we
have,
t
+
j
u
j
x
j
=
j
u
j
u
j
x
j
= 0 (3.37)
Applying statistical averaging to equation (3.36) produces Reynolds equation:
u
i
t
+
j
u
j
_
u
j
u
i
x
j
+
u
i
x
j
u
j
_
=
p
x
i
+
j
u
j
ij
x
j
(3.38)
16
With u = u + u
ij
=
_
u
i
x
j
+
u
j
x
i
_
2
u
i
=
ij
x
j
2. Multiply Navier-Stokes (3.36) by u
i
and average it.
u
i
t
u
i
+
j
u
j
u
i
x
j
u
i
=
p
x
i
u
i
+
ij
x
j
u
i
(3.39)
3. Multiply obtained Reynolds equation (3.38) by u
i
.
u
i
t
u
i
+
j
_
u
j
u
i
x
j
u
i
+
u
i
x
j
u
j
_
=
p
x
i
u
i
+
ij
x
j
u
i
(3.40)
Where,
u
i
x
j
u
j
=
_
u
i
u
j
_
x
j
or equivalently
i
t
u
i
+
j
u
j
u
i
x
j
u
i
=
p
x
i
u
i
+
j
_
ij
x
j
u
i
+
T
ij
x
j
u
i
_
(3.41)
With T
ij
= u
i
u
j
representing the components of the Reynolds stress matrix T.
4. Subtracting equation ((3.39)) from equation ((3.41)), we get.
i
t
u
i
+
j
_
u
j
u
i
x
j
u
i
u
j
u
i
x
j
u
i
_
=
p
x
i
u
i
+
j
_
ij
x
j
u
T
ij
x
j
ui
_
(3.42)
Where,
ij
x
j
u
i
=
(
ij
u
i
)
x
j
i
u
i
ij
5. Neglecting very small viscous transfer or turbulent energy, we get (3.43). Since,
the
ij
u
i
represents the viscous transfer of turbulent energy, a very small quantity
in contrast to the terms responsible for the turbulent energy in, it is neglected. Thus
17
becomes
i
t
u
i
+
j
u
i
u
i
x
j
u
j
+
j
_
i
x
j
u
i
u
j
+
u
i
u
j
x
j
u
i
+ u
j
u
i
u
i
x
j
_
=
p
x
j
u
j
_
u
i
x
j
ij
+
T
ij
x
j
u
i
_ (3.43)
6. Summing over i equation (3.43) becomes energy balance equation of turbulent ow,
with turbulent kinetic energy (K) and rate of dissipation of the turbulent energy ().
7. Using hypothesis for class of uid ow under consideration the equation of turbu-
lent energy balance reduces to For k,
k
t
=
t
_
c
k
k
x
_
(3.44)
Where, c
k
is turbulent exchange coefcient. For ,
t
=
t
_
C
x
_
U (3.45)
Where, C
_
u + (u)
T
_
n = 0
The neutral boundary condition means that no forces act on the uid and the computa-
tional domain extends to innity.
20
4 Head losses
Head is a term used to specify measure of pressure of total energy per unit weight above
a point of reference. In general, head is sum of three components; elevation head (the
elevation of the point at which the pressure is measured from above or below the arbitrary
horizontal observation point i.e. relative potential energy in terms of an elevation), veloc-
ity head (kinetic energy from the motion of water) (it is mainly used to determine minor
losses) and pressure head (equivalent gauge pressure of a column of water at the base of
the piezometer).
1
In cases where the uid is moving with very low velocity or stationary uid, we ignore the
velocity head because the uid is either stationary or moving with very low velocity and
in the cases where the uid is moving with very high velocity (cases where the Reynoldss
number exceeds 10) the elevation head and pressure head are neglected.
Head loss in uid ow in pipes means loss of ow energy due to friction or due to turbu-
lence. Head losses result in to loss in pressure at nal outlet. The pressure loss is divided
in two categories of Major (friction) losses and Minor losses. These losses are dependent
on both the type of uid and the material of the pipe.
Head loss is a measure to calculate reduction or loss in head. Head loss is mainly due
to friction between uid and walls of the duct (in our case it is pipe), friction between
adjacent layers of uid and turbulence caused by presence of pipe network components
like T-junction, elbows, bends, contractions, expansions, pumps, valves. Head losses
result in to loss in pressure at nal outlet, thus also known as pressure loss. Pressure
losses are divided in to two categories of major losses and minor losses.
Major losses: Losses due to friction between uid and internal pipe surface. These
losses occur over the length of pipe. They can be easily determined by Darcy-
Weisbach equation. Frictional loss is that part of the total head loss that occurs as
the uid ows through straight pipes
Minor losses: Losses occur at points where there is change in momentum. They
mainly occur at elbows, bends, contractions, expansions, valves, meters and similar
other pipe ttings that commonly occur in pipe networks.
1
A piezometer is small diameter water well used to measure the hydraulic head of underground water.
21
The major head loses may be large when the pipes are long (e.g. pipe network occurring
in water distribution in a city) and minor losses will also have a large contribution because
of attachments and ttings occurring in these networks. Thus, we can say that head loss
in reality are unavoidable, since no pipes are perfectly smooth to have uid ow without
friction, there does not exist a uid in which ows without turbulence.
The head loss for uid ow is directly proportional to the length of pipe, the square of the
uid velocity, and a term accounting for uid friction called the friction factor. The head
loss is inversely proportional to the diameter of the pipe. Head loss is unavoidable in pipe
networks with real uids, since there is no pipe with perfectly smooth inner surface and
there is no uid that can ow without turbulence.
Figure 4.1: Fluid behavior when pipe is smooth or rough from inside
The calculation of the head loss depends on whether the ow is laminar, transient or
turbulent and this we can determine by calculating Reynolds number.
4.1 Major head loss
There are many equations available to determine major head losses in a pipe. The most
fundamental of all is Darcy-Weisbach Equation. Major head loss (loss due to friction) is
determined by
h
major
=
_
l
d
h
__
v
2
2
_
22
This equation is valid for fully developed, steady, incompressible ow. The hydraulic
diameter (d
h
) is division on cross-section area of pipe by wetted perimeter.
d
h
=
cross sectionarea of pipe
wetted perimeter
=
4 (r
2
)
2r
= 2r = D
Thus, hydraulic diameter is the inner diameter of pipe. Therefore, major head loss formula
reduces to
h
major
=
_
l
D
__
v
2
2g
_
(4.1)
4.2 Friction Factor
Friction factor () depends on whether the ow is laminar, transient or turbulent, which
again depends on Reynolds number. Friction Factor for Laminar Flow
Consider
y = r R dy = dr
and shearing stress
=
d
dr
Where, is rate of change of velocity.
If we consider the uid to be isolated from the surrounding, the inlet will have velocity
(v
1
) and pressure (p
1
) and outlet will have velocity (v
2
) and pressure (p
2
).
Using momentum principle
2
(in uid dynamics), we get
p
1
A p
2
A + (shearing stress perimeter of pipe lengthof pipe) = Q(v
2
v
1
)
(p
1
p
2
) r
2
(2rL) = Q(v
2
v
1
)
We know that
=
p
1
p
2
2L
r
and
=
d
dr
2
The principle of conservation of momentum is an application of Newtons second law of motion to
an element of uid. That is, when considering a given mass of uid, it is stated that the rate at which the
momentum of the uid mass is changing is equal to the net external force acting on the mass.
23
Comparing both we get,
d =
p
1
p
2
2L
rdr
On integrating both sides and using = 0 at r = R and taking p
1
p
2
= p, we get
=
p
2L
_
R
2
r
2
_
The volumetric ow (Q) can be determined by
Q =
_
(2r) dr =
_
0
R
p
2L
_
R
2
r
2
_
(2r) dr
Q =
p
4L
r
4
And average velocity (V ) can be determined by
V =
Q
A
=
p
4L
r
4
1
r
2
p =
4L
R
2
V
Since, head loss equals pressure drop (p) divided by
h
major
=
p
=
4L
R
2
V
Also,
h
major
=
L
D
V
2
2g
Comparing both, we get
=
64
L
D
V D
=
64
R
e
Thus, =
64
R
e
when R
e
< 2100. This can also be conrmed from Nikuradses graph for
laminar ow.
3
Friction Factor for Transient Flow
If the Reynolds number for the ow is between 2300 and 3000 the type of ow exhibited
by the uid is known as transient ow. This is type of ow where velocity and pressure of
3
Nikuradse showed the dependence on roughness by using pipes articially roughened by xing a coat-
ing of uniform sand grains to the pipe walls. The degree of roughness was designated as the ratio of the
sand grain diameter to the pipe diameter (/D).
24
the ow are changing with time. The ow also switches between turbulent and laminar.
Because of this behavior it is difcult to determine the friction coefcient. Thus, the
friction coefcient for Transient ow can not be determined.
Friction Factor for Turbulent Flow
When the ow is turbulent, the frictional factor () can be obtained by solving the equa-
tion
1
= 2.0log
10
_
2.51
R
e
+
r
p
d
h
1
3.72
_
Where, r
p
is relative roughness of the pipe.
This equation is well known as Colebrooke equation
4
. Colebrooke equation is also graph-
ically presented by Moody Chart
5
, which can be easily used if some required parameter
values are known. The Moody chart relates the friction factor for fully developed pipe
ow to the Reynolds number and relative roughness of a circular pipe. Relative rough-
ness for some common materials can be found in the table- 1
6
below.
Surface Roughness (r
p
) 10
3
m
Copper, Lead, Brass, Aluminum (new) 0.001 0.002
PVC and Plastic Pipes 0.0015 0.007
Epoxy, Vinyl Ester and Isophthalic pipe 0.005
Stainless steel 0.015
Steel commercial pipe 0.045 0.09
Rusted steel (corrosion) 0.15
Smoothed cement 0.3 1
Ordinary concrete 0.3 0.5
Table 1: Relative roughness for some common materials determined by experiments.
Relative roughness of the pipe (r
p
) can be easily determined if we know the material of
the pipe. This value completely depends on material of pipe. These values are also easily
available on some manuals. Table-2 summarizing relation between Reynolds number
(R
e
), the type of ow and Friction coefcient ()
The Friction coefcient () can also be determined by Moody Chart. There is also a sec-
tion in this chapter that briey describes the use. An illustration is also given to understand
4
The Colebrook equation is an implicit equation which combines experimental results of studies of
laminar and turbulent ow in pipes. It was developed in 1939 by C. F. Colebrook.
5
In 1944 Lewis F. Moody, Professor, Hydraulic Engineering, Princeton University, published paper
titled Friction Factors for Pipe Flow. The work of Moody, and the Moody Diagram has become the basis
for many of the calculations on friction loss in pipes and ductwork.
6
Table for Relative roughness for some common materials was taken from website
http://www.engineeringtoolbox.com.
25
Reynolds number (R
e
) Nature of ow Friction coefcient ()
< 2300 Laminar Flow = 64/R
e
2300 4000 Transient Flow Can not be determined
> 4000 Turbulent Flow
1
= 2.0log
10
_
2.51/
_
R
e
_
+ r
p
/3.72d
h
_
Table 2: Reynolds Number, Nature of Flow and Friction coefcient ().
this more clearly.
We can summarize above discussion in these points
If the Reynolds numbers is less than about 2100 the ow will be laminar. This
indicates that the viscous force of the uid is dominating the other forces that may
disturb the ow. When owis laminar, the uid seems to move in controlled manner
with regular streamlines. It would look like very thin glass lms are sliding over
each other.
If the Reynolds number is between 2300 and 3000 the ow will be transient. This is
category between laminar and turbulent ow, where we can not determine anything
about the ow. There may also be observed a small amount of turbulence in the
ow.
If the Reynolds number is greater than 3000 which is common when the uid is
moving with high speed or with some obstacles or rough surface of duct then the
owis said to be turbulent. The owbeing turbulent indicates that the inertial forces
are more than forces due to velocity and that the streamlines are no more parallel
to each other and the ow pattern is irregular and the uid particles may cross one
point in domain more than once.
4.3 Minor head loss
Minor losses (losses due to various attachments and change in momentum) can be calcu-
lated by following formula.
p
major
= H
L
_
v
2
2g
_
Where, H
L
is loss coefcient for the pipe component and g is acceleration due to gravity.
The loss coefcients for various pipe components are available in several textbooks, man-
26
uals and supplier manuals. Table-3
7
lists minor loss coefcients for some common com-
ponents in pipe networks. These relative roughness for materials were determined by
experiments.
Type of Component or Fitting Minor Loss Coefcient (H
L
)
Flanged Tees, Line Flow 0.2
Threaded Tees, Line Flow 0.9
Flanged Tees, Branched Flow 1.0
Threaded Tees, Branch Flow 2.0
Flanged Regular 90
o
Elbows 0.3
Threaded Regular 90
o
Elbows 1.5
Threaded Regular 90
o
Elbows 0.4
Flanged Long Radius 90
o
Elbows 0.2
Threaded Long Radius 90
o
Elbows 0.7
Flanged Long Radius 90
o
Elbows 0.2
Flanged 180
o
Return Bends 0.2
Threaded 180
o
Return Bends 1.5
Fully Open Globe Valve 10
Fully Open Angle Valve 2
Table 3: Minor loss coefcients for some of the most common used components in pipe
and tube systems
As mentioned before several textbooks, manuals and supplier manuals. Values in various
sources may vary depending upon the experimental conditions, formulas and calculation
techniques used. Thus, one must rst determine if the experimental conditions of the data
are the same as the conditions of the current experiment and the other additional data
related to the same experiment are from the source.
4.4 Using the Moody Diagram
Head loss is a function of Reynolds number and relative roughness coefcient. Colebrook
developed an empirical transition
8
function for commercial pipes, which relates friction
factor and the Reynolds number. The Moody diagram is based on the Colebrook equation
in the turbulent regime. The Moody chart relates the friction factor for fully developed
pipe ow to the Reynolds number and relative roughness of a circular pipe. The frictional
factor () for head loss can be determined if Reynolds number and the relative roughness
of the pipe are known. The rougher the pipe the more turbulent the ow is through that
7
Table for Minor loss coefcients was taken from website http://www.engineeringtoolbox.com.
8
Transition is the term used by Colebrook to describe roughness of pipe. By transition he meant that
the pipes are neither too rough nor too smooth.
27
pipe. The relative roughness of a pipe is given by e/D, where e is absolute roughness of
pipe and D is diameter of pipe.
Figure 4.2: Moody chart for estimating Frictional factor
By looking at the Moody diagram it shows that the right top corner is completely tur-
bulent and the left top is laminar (smooth ow). To determine the frictional factor, nd
the relative roughness value for the pipe on the right. Then locate the pipes Reynolds
number on the bottom. Follow the relative roughness curve to where it crosses the deter-
mined Reynolds number. Now at that point project a straight line to the left, the number
determined on the left is the frictional factor.
4.4.1 Example of using Moody chart
Consider ow situation where pipe diameter (D) is 1 ft, Kinematic Viscosity is 14.1
10
6 ft
2
s
, velocity of uid is 0.141
ft
s
and e is 0.002 ft.
28
First we compute e/D and R.
e
D
=
0.002ft
1ft
= 0.002
R =
Dv
=
(1ft)(0.141
ft
s
)
14.1 10
6
ft
2
s
= 10000
Now, we consider the value of e/D and follow the curve where R is 10000. We project a
straight line to left and can see that the value is 0.034.
4.5 Total Head Loss in Serial Connected Pipes
If total head loss in a single pipe is given by
L
D
V
2
2g
+ K
L
V
2
2g
(4.2)
Then, the total head loss in several serial connected pipes is algebraic sum of all the head
losses due to pipes in the network. In network of k pipes, if i is the number of pipe the
the total head loss can be expressed as the following formula:
h =
i=1
k
_
i
L
i
D
i
V
2
i
2g
+ K
L
i
V
2
i
2g
_
(4.3)
Where, the quantities with index i is connected to i
th
pipe in the network.
29
5 Head Loss Coefcient for T-junction
The pressure loss caused by the T-junction depends on inner radius of the branches, ve-
locity of uid entering or leaving from the junction and the angle of the junction (there are
various approaches for this calculations, some cases are presented in the following text).
There are some classical formulas for pressure loss co-efcient for T-junctions. Most of
these formulas depending on angle of T-junction, inlet and outlet velocities. To compute
head loss coefcients, we have used formulas derived by A. Vazsonyi
9
, A. Gardel
10
and
V. Curic
11
.
One other idea for computing pressure loss co-efcient for T-junctions with angles, was to
consider T-junction as combination of two pipe components e.g two elbows or an elbow
and a contraction. According to this idea, we assume the T-junction to be made up of
two pipe components. The choice of components would depend on the ow conditions
i.e from which arms the ow is coming toward the junction and from which arms the ow
is leaving from the junction.
In this section we shall mention the classical formulas and the formulas that were con-
structed by assuming T-junction to be made up of other pipe components.
5.1 For dividing ows
These formulas are used for the situation where ow from a single branch ows to the
other two remaining branches. The picture in the left of gure (5.1) gives more clear idea
about such ow situations.
Various studies have been made on T-junction with dividing ow situation. Of these stud-
ies, results obtained by Andrew Vazsonyis were believed to be the closest to the available
statistical data then. Vazsonyi derived two formulas for dividing case and combining
cases (5.1). In his work he explained relation between velocity ratios, angles of the junc-
tion and loss coefcient. The formulas are the result of the comparisons made by him.
9
The reference book [2] contains Vazsonyi, Andrew: Pressure Loss in Elbows and Duct Branches.
Trans. ASME, vol. 66, no. 3, Apr. 1944, pp. 177-183 as one of the important reference for these formula.
10
The reference book [1] contains A. Gardel, Pressure drops in ows through T-shaped pipe-ttings,
Bull. Tech. Suisse Romande 9 (1957), pp. 122130 as one of the important reference for these formula.
11
Full derivations and details of the formulas derived by considering T-junction as two components are
available in [3]
30
The formula available from work of Vazsonyi is as following
Figure 5.1: Example of ow situations and angles for combining and dividing ow
K
0,1
=
1
+ (2
2
1
)
_
V
1
V
0
_
2
2
2
_
V
1
V
0
_
cos
(5.1)
Here K is depending on kinetic energy of the combined ow in branch-0, and
1
= 0.0712
0.7141
+ 0.37 for < 22.5
0
(5.2)
1
= 1.0 for 22.5
0
(5.3)
2
= 0.0592
0.7029
+ 0.37 for < 22.5
0
(5.4)
3
= 0.9 for 22.5
0
(5.5)
and
= 1.41 0.00594
2
(5.6)
The gure (5.1) shows the plots for
1
,
2
(left) and plot for
3
(right).
The other empirical formula obtained by Gardel (1957). His idea was to calculate pressure
loss coefcients separately for each inlet (loss coefcient for ow from inlet-1 to outlet-3
and loss coefcient for ow from inlet-2 to outlet-3), so for each ow situation we have
two loss coefcients (K
31
and K
32
). These formulas were derived by applying momentum
balance to the main pipe section of the junction (section abcd in g (5.3)) and equation of
continuity to the whole t-junction. Then energy balance is applied individually for each
inlet.
31
Figure 5.2: Plot of
3
(left) and Plot of and (right)
Figure 5.3: Diagram for combining ow
The formula obtained by Gardel are,
K
31
= 0.95 (1 q)
2
+ q
2
__
1.3 tan
2
0.3 +
0.4 0.1a
a
2
__
1 0.9
_
r
a
_1
2
__
+0.4q
_
1 + a
a
tan
2
_ (5.7)
K
32
= 0.03(1 q)
2
+ 0.35q
2
0.2q(1 q) (5.8)
Where, a = A
1
/A
3
and = .
It can be clearly observed that there is no effect of area ratio or radius of pipe on the loss
coefcient K
32
.
32
5.2 For combining ows
These formulas are used for the situation where ow from two branches combine in the
remaining branch. The gure (5.3) gives more clear idea about such ow situations. The
formula available from work of Vazsonyi are as follows
K
0,1
=
3
_
V
1
V
0
_
2
+ 1 2
__
V
1
V
0
__
Q
1
Q
0
_
cos
+
_
V
2
V
0
__
Q
2
Q
0
_
cos
_
(5.9)
Where, K is again depending on kinetic energy of the combined ow in branch-0. Q is
volumetric ow rate (= AV ).
3
is dened in the graph given by gure (5.1) and
(5.12)
Where V
2
= Q
2
/A
2
and V
3
= Q
3
/(A x).
And, loss coefcient for sudden contraction is
K
13
= 1
x
A
(5.13)
These values can be substituted in the following formulas to determine the pressure loss.
p
1
p
1
=
1
2
K
13
_
Q
1
x
_
2
(5.14)
and
p
2
p
2
=
1
2
K
23
_
Q
2
A x
_
2
(5.15)
The unknown x can be determine by solving the equation
p
1
p
2
=
1
2
_
_
A x
A
__
Q
1
x
_
2
__
0.61
A
2
2
_
+
_
1
(A x)
2
__
Q
2
+ 2 cos
_
Q
2
2
A
2
(A x)
_
_
(5.16)
34
T-junction as combination of two elbows
For a combining ow situation as described in gure (5.5), T-junction can be considered
as combination of two elbows. For computing the pressure loss for such combination, we
can compute pressure loss for the elbows separately and then add them. For doing so, we
have to nd the point where the two elbow are joined. For this purpose, we need to solve
equation (5.21) for x.
Figure 5.5: T-junction as combination of two elbows
For elbow-1, the loss coefcient is
K
13
= 0.61
_
A
m
x
A
_
2
+ 1 2
_
A
m
x
A
_
cos
(5.17)
For elbow-2, the loss coefcient is
K
23
= 0.61
_
x
A
_
2
+ 1 2
_
x
A
_
cos
(5.18)
These values can be substituted in the following formulas to determine the pressure loss.
p
1
p
1
=
1
2
K
13
_
Q
1
A
m
x
_
2
(5.19)
and
p
2
p
2
=
1
2
K
23
_
Q
2
x
_
2
(5.20)
The unknown x can be determine by solving the equation
p
1
p
2
=
1
2
_
0.61
_
Q
2
1
Q
2
2
A
2
_
+
Q
1
A
m
x
2
Q
2
x
2
2
Q
2
1
A(A
m
x)
cos
2
Q
2
2
A.x
cos
_
(5.21)
35
6 Computational Experiments
In this section we shall discuss observations and results obtained by experiments made
with softwares FLUENT and Comsol Multiphysics. We shall also compare the results
obtained by softwares with the results obtained from various classical head loss formulas
mentioned in last chapter. The section includes results obtained by experiments with
Figure 6.1: Cross-section plot for example case of ow in T-junction
T-junction with various diameters and inow velocities, numerical results obtained by
slightly changing the angle of the junction from 90
0
and also, we shall also explain how
the T-junction can be split in to two pipe components (e.g. two elbows) and compare
the head loss obtained by classical formula of the head loss of T-junction and formula
obtained by splitting T-junction in to two pipe components.
Figure 6.2 shows and example of comparison of head-loss by classical formula and head
loss observed by software of an example cases of ow in T-junction. The curve with data
points presented by star is the curve for head loss observed by software and the curve with
data points presented by square is the curve for head-loss obtained by classical formula.
We can clearly observe that the curves agree good for rst 3 sets of velocities but then on
the curves do not agree.
The graphs in the following section can be similarly interpreted.
36
1 2 3 4 5
4.5
5
5.5
6
6.5
7
7.5
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 (v1) in cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for example flow case
v2=1,SW3d
v2=1,CF
Figure 6.2: Comparison of head-loss by classical formula and head loss by software of an
example cases of ow in T-junction
6.1 Head loss comparison for combining ow
Case-1 This is the case where the ow in coming toward the junction from two branches
in main pipe and leaving from the junction from the perpendicular branch (See gure 6.3).
Figure 6.3: Combining ow: Case-1
From gure 6.4, it can be observed that the head loss by software and classical formulas
(using the formula by Andrew Vazsonyi) do not agree in this case. There is about 3.2 %
error between results by software and classical formula.
Case-2 This is the case where the ow in coming toward the junction from one branch
37
1 2 3 4 5
2
3
4
5
6
7
8
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 (v1) in cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Combining flow case1
v2=1,SW3d
v2=1,CF
v2=1.5,SW3d
v2=1.5,CF
v2=2,SW3d
v2=2,CF
v2=2.5,SW3d
v2=2.5,CF
v2=3,SW3d
v2=3,CF
Figure 6.4: Head loss for Combining ow: Case-1, Radius of branches is 0.5 cms, Inlet
velocities vary from 1 cm/sec to 3 cm/sec (25 different cases plotted), Outlet pressure is
100 Pascals and classical pressure loss formula by Andrew Vazsonyi
in main pipe and the branch perpendicular to it and leaving from the junction from the
remaining branch in the main pipe (See gure 6.5) [The other situation is exactly the
mirror image].
Figure 6.5: Combining ow: Cases-2
From gure 6.6, we can observe that the head loss by software and classical formulas
also do not agree in this case. There is about 5.0 % error between results by software and
classical formula.
38
1 2 3 4 5
2
3
4
5
6
7
8
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 (v1) in cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Combining flow case2
v2=1,SW3d
v2=1,CF
v2=1.5,SW3d
v2=1.5,CF
v2=2,SW3d
v2=2,CF
v2=2.5,SW3d
v2=2.5,CF
v2=3,SW3d
v2=3,CF
Figure 6.6: Head loss for Combining ow: Case-2, Radius of branches is 0.5 cms, Inlet
velocities vary from 1 cm/sec to 3 cm/sec (25 different cases plotted), Outlet pressure is
100 Pascals and Classical pressure loss formula by A. Gardel
6.2 Head loss comparison for dividing ow
Case-1 This is the case where the ow in coming toward the junction from the perpen-
dicular branch and leaving from the junction from two branches in main pipe (See gure
6.7).
Figure 6.7: Dividing ow: Case-1
Case-2 This is the case where the ow in coming toward the junction from one branch
39
1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Tjunction with diverging flow
v1=3,SW
v1=3,CF
v1=2,SW
v1=2,CF
v1=1,SW
v1=1,CF
Figure 6.8: Head loss for dividing ow: Case-1, Radius of branches is 0.5 cms, Inlet
velocity vary from 1 cm/sec to 3 cm/sec, at both outlet pressure is 100 Pascals and
Classical pressure loss formula by A. Gardel
in main pipe and perpendicular branch and leaving from the junction from the remaining
branch in the main pipe (See gure 6.9) [The other situation is exactly the mirror image]
Figure 6.9: Dividing ow: Case-2
From gure 6.8 and 6.8, we can observe that the head loss by software and classical
formulas also do not agree in this case. Though the curves, seem to get along with the in-
crease in inlet velocities, but they do not exactly match for any combination of velocities.
There is about 4.5 to 6.1 % error between results by software and classical formula.
6.3 Head loss change with change in angle of T-junction branches
In this part we shall display comparison of head loss obtained by software and classical
formulas for different angles of T-junction. The gure 6.11 cases for inow, outow and
40
1 1.5 2 2.5 3
0
1
2
3
4
5
6
7
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Tjunction with diverging flow, Case2
v1=3,SW
v1=3,CF
v1=2,SW
v1=2,CF
v1=1,SW
v1=1,CF
Figure 6.10: Head loss for dividing ow: Case-2, Radius of branches is 0.5 cms, Inlet
velocity vary from 1 cm/sec to 3 cm/sec, at both outlet pressure is 100 Pascals and
Classical pressure loss formula by A. Gardel
angle. For all the comparisons we have use formulas by Andrew Vazsonyi for combining
ow case-1 and formulas by A. Gardel for case-2.
Figure 6.11: T-junction with different angles between main pipe and branch pipe
In cases shown in gure-6.12 to gure-6.17, we have calculated and compared head loss
suffered by T-junction with angle = 91, = 91, = 93, = 89, = 88 and = 87.
These calculations were for combining ow case-1, where ow is coming toward the
junction from opposite pipes and leaving from the junction through perpendicular pipe
(see gure 6.3). It was observed that the head loss increases with increase in angle. Also,
for all the cases; the head loss obtained by software and classical formula were close to
each other.
41
1 1.5 2 2.5 3 3.5 4 4.5 5
2
3
4
5
6
7
8
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 (v1) in cm/sec
Graph of Headloss from software(SW) VS Headloss from classical formula(CF)
v2=1,SW3d
v2=1,CF
v2=1.5,SW3d
v2=1.5,CF
v2=2,SW3d
v2=2,CF
v2=2.5,SW3d
v2=2.5,CF
v2=3,SW3d
v2=3,CF
Figure 6.12: Head loss for T-junction with angle = 91, combining ow: Case-1, Radius
of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3 cm/sec (25 different cases
plotted), Outlet pressure is 100 Pascals and Classical pressure loss formula by Andrew
Vazsonyi
The gure 6.18 shows head loss for different angles of T-junction. These calculations
were for combining ow case-2, where ow is coming toward the junction from perpen-
dicular pipes and leaving from the junction through remaining main pipe (see gure 6.5).
It was observed that when the angle gamma () is less, head loss suffered is less. This
is because there is no signicant change in of momentum of the ow between incoming
and out going ow. It was also observed that when the angle gamma () is more, head
loss suffered is more. This is because of change in momentum of the ow while passing
through T-junction.
From the cases = 87 (g-6.15), = 88 (g-6.16), = 89 (g-6.17), = 91 (g-6.12),
= 92 (g-6.13) and = 93 (g-6.14), we can observe that the head loss by software
and classical formulas also do not agree in any case. Though for some cases and certain
inlet velocity combinations, the curves seem to get along with each other but this is not
sufcient to conclude that the head losses obtained by both the sources agree. There is
about 4.6 to 6.7 % error between results by software and classical formula.
42
1 2 3 4 5
1
2
3
4
5
6
7
8
9
10
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Tjunction with 92 degree angle
v1=1,SW
v1=1,CF
v1=1.5,SW
v1=1.5,CF
v1=2,SW
v1=2,CF
v1=2.5,SW
v1=2.5,CF
v1=3,SW
v1=3,CF
Figure 6.13: Head loss for T-junction with angle = 92, combining ow: Case-1, Radius
of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3 cm/sec (25 different cases
plotted), Outlet pressure is 100 Pascals and Classical pressure loss formula by Andrew
Vazsonyi
6.4 Head loss for T-junction with different radius of branches
Case-1: This is the case where the cross-section area of main pipe is one half of that
of perpendicular branch pipe. The ow is coming toward the junction from the opposite
branches in main pipe and leaving from the perpendicular branch (See gure 6.19).
Case-2: This is the case where the cross-section area of main pipe is one third of perpen-
dicular branch pipe. The ow is coming toward the junction from the opposite branches
in main pipe and leaving from the perpendicular branch (See gure 6.21).
Case-3: This is the case where the cross-section area of main pipe is one fourth of perpen-
dicular branch pipe. The ow is coming toward the junction from the opposite branches
in main pipe and leaving from the perpendicular branch similar to above two cases.
The gure 6.24 shows head loss for different cross-section areas of branches of T-junction
(A1 =area of main pipe, A2 =area of branch pipe). These calculations were for com-
bining ow case-1, where ow is coming toward the junction from opposite branches in
main pipe and leaving the junction from perpendicular branch pipes (see gure 6.5).
Fromgure 6.20, gure 6.22 and gure 6.23 we can observe that the head loss by software
43
1 2 3 4 5
0
1
2
3
4
5
6
7
8
9
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Tjunction with 93 degree angle
v2=1,SW3d
v2=1,CF
v2=1.5,SW3d
v2=1.5,CF
v2=2,SW3d
v2=2,CF
v2=2.5,SW3d
v2=2.5,CF
v2=3,SW3d
v2=3,CF
Figure 6.14: Head loss for T-junction with angle = 93, combining ow: Case-1, Radius
of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3 cm/sec (25 different cases
plotted), Outlet pressure is 100 Pascals and Classical pressure loss formula by Andrew
Vazsonyi
and classical formulas also do not agree. There is about 4.4 to 6.8 % error between results
by software and classical formula.
It was observed that head loss is reducing when the cross-section area of the main pipe
is reducing (for all the cases cross-section area of the perpendicular branch pipe was kept
same 1 cm.). This observations also veries claims by A. Gardel, that the head loss
increases with increase in ratio of the cross section area (A2/A1 where, A1 =area of
main pipe, A2 =area of branch pipe). These observations are for the case when the ow
is combining case-1, where the ow is coming in from opposite branches in main pipe
and leaving from perpendicular branch pipe. The observation is exactly reverse when we
consider combining ow case-2, where the ow is coming in from one branch in main
pipe and perpendicular branch pipe and leaving from remaining branch in main pipe. The
head loss suffered will increase with increase in ratio of the cross section area (A2/A1).
44
1 2 3 4 5
0
1
2
3
4
5
6
7
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Tjunction with 87 degree angle
v1=1,SW2d3d
v1=1,CF
v1=1.5,SW2d3d
v1=1.5,CF
v1=2,SW2d3d
v1=2,CF
v1=2.5,SW2d3d
v1=2.5,CF
v1=3,SW2d3d
v1=3,CF
Figure 6.15: Head loss for T-junction with angle = 87, combining ow: Case-1, Radius
of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3 cm/sec (25 different cases
plotted), Outlet pressure is 100 Pascals and Classical pressure loss formula by Andrew
Vazsonyi
1 2 3 4 5
1
2
3
4
5
6
7
8
9
10
11
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Tjunction with 88 degree angle
v1=1,SW
v1=1,CF
v1=1.5,SW
v1=1.5,CF
v1=2,SW
v1=2,CF
v1=2.5,SW
v1=2.5,CF
v1=3,SW
v1=3,CF
Figure 6.16: Head loss for T-junction with angle = 88, combining ow: Case-1, Radius
of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3 cm/sec (25 different cases
plotted), Outlet pressure is 100 Pascals and Classical pressure loss formula by Andrew
Vazsonyi
45
1 2 3 4 5
1
2
3
4
5
6
7
8
9
10
11
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Tjunction with 89 degree angle
v1=1,SW2d3d
v1=1,CF
v1=1.5,SW2d3d
v1=1.5,CF
v1=2,SW2d3d
v1=2,CF
v1=2.5,SW2d3d
v1=2.5,CF
v1=3,SW2d3d
v1=3,CF
Figure 6.17: Head loss for T-junction with angle = 89, combining ow: Case-1, Radius
of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3 cm/sec (25 different cases
plotted), Outlet pressure is 100 Pascals and Classical pressure loss formula by Andrew
Vazsonyi
1 1.5 2 2.5 3
1
2
3
4
5
6
7
8
9
10
11
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cm/sec
Graph of Headloss for different angle of Tjunction
gamma=88
gamma=89
gamma=90
gamma=91
gamma=92
gamma=93
Figure 6.18: Head loss for different angle of T-junction, combining ow: Case-2, Radius
of branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3 cm/sec, Outlet pressure
is 100 Pascals and Classical pressure loss formula by A. Gardel
46
Figure 6.19: Dividing ow: Case-1
1 1.5 2 2.5 3 3.5 4 4.5 5
0
1
2
3
4
5
6
7
8
P
r
e
s
s
u
r
e
in
P
a
s
c
a
ls
Velocity in cm/sec
Graph of Headloss from software(SW) VS Headloss from classical formula(CF)
v1=1,SW
v1=1,CF
v1=1.5,SW
v1=1.5,CF
v1=2,SW
v1=2,CF
v1=2.5,SW
v1=2.5,CF
v1=3,SW
v1=3,CF
Figure 6.20: Head loss for area case-1, combining ow case-1, Radius of main pipe is
branches is 0.25 cms, Radius of perpendicular pipe is branches is 1 cms, Inlet velocity in
both inlets vary from 1 cm/sec to 3 cm/sec, pressure at outlet is 100 Pascals and Classical
pressure loss formula by Andrew Vazsonyi
47
Figure 6.21: Dividing ow: Case-1
1 1.5 2 2.5 3 3.5 4 4.5 5
0
1
2
3
4
5
6
7
8
9
P
r
e
s
s
u
r
e
in
P
a
s
c
a
ls
Velocity in cm/sec
Graph of Headloss from software(SW) VS Headloss from classical formula(CF)
v1=1,SW
v1=1,CF
v1=1.5,SW
v1=1.5,CF
v1=2,SW
v1=2,CF
v1=2.5,SW
v1=2.5,CF
v1=3,SW
v1=3,CF
Figure 6.22: Head loss for area case-2, combining ow case-1, Radius of main pipe is
branches is 0.3 cms, Radius of perpendicular pipe is branches is 1 cms, Inlet velocity in
both inlets vary from 1 cm/sec to 3 cm/sec, pressure at outlet is 100 Pascals and Classical
pressure loss formula by Andrew Vazsonyi
48
1 2 3 4 5
0
1
2
3
4
5
6
7
8
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cm/sec
Graph of Headloss from software(SW) VS Headloss from
classical formula(CF) for Tjunction with A2=A1/2
v1=1,SW
v1=1,CF
v1=1.5,SW
v1=1.5,CF
v1=2,SW
v1=2,CF
v1=2.5,SW
v1=2.5,CF
v1=3,SW
v1=3,CF
Figure 6.23: Head loss for area case-1, combining ow case-1, Radius of main pipe is
branches is 0.25 cms, Radius of perpendicular pipe is branches is 1 cms, Inlet velocity in
both inlets vary from 1 cm/sec to 3 cm/sec, pressure at outlet is 100 Pascals and Classical
pressure loss formula by Andrew Vazsonyi
1 1.5 2 2.5 3
0
1
2
3
4
5
6
7
8
P
r
e
s
s
u
r
e
i
n
P
a
s
c
a
l
s
Velocity in inlet1 cmsec
Graph of Headloss for different crosssection areas of branches of
Tjunction, A1=area of main pipe, A2=area of branch pipe
A1=2*A2,v1=1
A1=3*A2,v2=1
A1=4*A2,v2=1
A1=2*A2,v2=1.5
A1=3*A2,v2=1.5
A1=4*A2,v2=1.5
A1=2*A2,v2=2
A1=3*A2,v2=2
A1=4*A2,v2=2
A1=2*A2,v2=2.5
A1=3*A2,v2=2.5
A1=4*A2,v2=2.5
A1=2*A2,v2=3
A1=3*A2,v2=3
A1=4*A2,v2=3
Figure 6.24: Head-loss for different cross-section areas of branches of T-junction,
A1 =area of main pipe, A2 =area of branch pipe, combining ow: Case-1, Radius of
branches is 0.5 cms, Inlet velocities vary from 1 cm/sec to 3 cm/sec, Outlet pressure is
100 Pascals and Classical pressure loss formula by A. Gardel
49
7 Discussion and future scope of the work
7.1 Discussion
From results in the previous section, we can observe that there is difference between head
loss in T-junction of pipes observed by calculations from software packages Fluent and
Comsol. Our main aim was to study the difference between the observations from CFD
softwares and classical formula by Andrew Vazsonyi, A. Gardel and formulas available
in reference [3].
In case of combining ow, the difference between observations obtained by Comsol (3D
experiments) and classical formula were in the range of 3.2 to 5.1 %. Incase of dividing
ow, this difference was in the range of 4.5 to 5.5 %. In the case, where we varied the
angle of the T-junction from 87 degrees to 93 degrees, difference between observations
by Comsol (3D experiments) and classical formula was in the range of 4.6 to 6.7 %.
One of the reasons for these errors is likely the limited capabilities of software. These
differences are as a result of softwares inability to handle complicated ow conditions.
Comsol Multiphysics (version 3.2a) can not handle ow situations with turbulence. Also,
this version of Comsol Multiphysics does not have ability to model rough inner surface
of pipes.
For all our experiments the uid was considered water with normal properties at room
temperature. Also the classical formulas are valid only for uid that is incompressible
and inviscid. The formulas reference [3], with the idea of considering the T-junction as
combination of two pipe components, is only valid for 2D case.
Our main aim was to study the difference between the observations from CFD software
and classical formula by Andrew Vazsonyi, A. Gardel and formulas available in refer-
ence [3]. The values obtained by CFD software were in certain agreement with classical
formulas both by Andrew Vazsonyi and A. Gardel but values obtained by CFD software
were better agreement with A. Gardel. It can be clearly observed that for combining ow
situations where we used Gardels formula, the difference was in range of 3.2 to 5.0 %
and for dividing ow cases where we used Vazsonyi formulas, the difference was in range
of 4.5 to 6.0 %.
Gardels formulas were as result of a systematic derivation from basic principles of mo-
50
mentum (applied to the main pipe), continuity principle to the uid in whole T-junction
and energy balance principle (individually) to ow coming from the branches. Unfor-
tunately, none of the classical formulas consider pipe roughness as factor for the head
loss. Roughness of the pipe varies from as material and it is also considered as one of
the major cause for major losses. This is where the accuracy of coefcients obtained
by classical formula can be questioned. Though, the loss due to friction between uid
and junction inner surface is very less, but theses small values can be very signicant for
precise calculations.
During this work, we also observe that the difference between observations by 2D sim-
ulations of software and classical formula were considerably larger than the difference
between observations by 3D simulation of software and classical formula. We also rec-
ommend 3D simulation for such calculations, since 3D simulation are more near to the
reality and also effect of turbulence can be modeled and observed in 3Dsimulations. Also,
3D simulation give more clear view of swirl movements, streamlines and turbulence in
the uid.
During the work we realized that Fluent is a better option for heavy and precise simu-
lations. Since, Fluent has capability to model turbulence with verity of Kappa-Epsilon
models and also because Gambit is a very handy tool to create even complicated geome-
tries. But, Fluent can be sometimes very expensive in terms of computational time. The
only advantage with Comsol Multiphysics is that we can create geometry and carry out
calculations in the same environment and the grid does not have to be exported every time
the experiments are repeated.
From our experience during this work, we would suggest to use Fluent for similar simu-
lations. There are also some higher versions of Comsol available that have capability to
handle complex ow situations. Gardels formulas were as result of a systematic deriva-
tion from basic principles of momentum (applied to the main pipe
12
), continuity principle
to the uid in whole T-junction and energy balance principle (individually) to ow coming
from the branches.
12
For this purpose he considered main pipe as a control volume and applied momentum balance principle.
51
7.2 Future scope of the work
In this work was restricted to only water at room temperature and t-junctions with smooth
inner surface. There can be more work done to generalize these results for the other uids
and T-junction with rough inner surfaces.
Also, with software our ambition was to construct a real time simulation of T-junction
with varying angle. Though this is a very lengthy process, since uent takes too much
time with dynamic mesh, but this is possible with higher versions of uent and other CFD
packages.
Unfortunately, none of the classical formulas consider pipe roughness as factor for the
head loss. Roughness of the pipe varies from as material and it is also considered as one
of the major cause for major losses. This is where the accuracy of coefcients obtained
by classical formula can be questioned. Though the loss due to friction between uid
and junction inner surface is very less but theses small values can be very signicant for
precise calculations.
During this study, we also came across an industrial problem concerning to ow of pulp
like uid in pipes. The problem was placing a valve of certain capacity for regulating
supply of pulp like material based on the pressure and velocity proles in the supply
network. Initially, pipe with elastic property was used to supply the material and a large
forceps was used to reduce the diameter of pipe where the supply was not needed or to be
regulated. Such kind of problems can be solved with similar techniques.
In this work all, we made an attempt to study effect of different radius of main pipe and
branch pipe. The range of ow parameters (ow velocity, pipe diameter and pressure)
used in our computational experiments was relatively small. It is also possible that the
difference of head loss observed and inaccuracy of the formula is even larger in broader
range of parameters. Thus we suggest that there should be more 3D computational exper-
iments done using more advanced CFD software packages.
This can play important role in verifying other claims made on basis of classical formulas.
E.g. Andrew Gardels observation that head-loss increases with increase in ratio of areas
of main pipe and branch pipe.
52
References
[1] Ward-Smith A. J, Internal uid ow the uid dynamics of ow in pipes and ducts, Clarendon
Press, Oxford (1980).
[2] Robert P. Benedict, Fundamentals of Pipe Flow, Wiley-Inter science Publications (1980).
[3] Vladimir Curic, Matematicki model ekasnog upravljanja protokom vode, Institute of Mathe-
matics, University of Novi Sad, Master thesis (2007).
[4] Boris Huber, CFD Simulation of a T-Junction, Institute of Hydraulic and Water Resources
Engineering, Department of Hydraulic Engineering, Vienna University of Technology.
[5] Ronald W. Jeppson, Analysis of ow in pipe networks, ANN Arbor Science (1976).
[6] Batchelor G. K, Introduction to uid dynamics, Cambridge : Cambridge University Press
(1990).
[7] Hozef Arif, Application of Computational Fluid Dynamics to the modeling of horizontal wells,
Stanford University (1999).
[8] Stephen Wolfram, Notes for Chapter 8: Implications for Everyday Systems Section: Fluid
Flow-A New Kind of Science, Page 996, www.wolframscience.com.
[9] H. Tennekes and J.L Lumley, A First Course in Turbulence, The MIT Press, Cambridge MA
(1992).
[10] C. Foias, O. Manley, R. Rosa and R. Teman, Navier-Stokes Equation and Turbulence, Cam-
bridge University Press, Cambridge (2001).
[11] Roger Temam, Navier-Stokes equations : theory and numerical analysis, North-Holland,
Amsterdam (1984).
[12] W.D. McComb, The Physics of Turbulence, Oxford University Press (1990).
[13] J. M. McDonough Introductory Lectures on Turbulence Physics: Mathematics and Model-
ing, Departments of Mechanical Engineering and Mathematics, University of Kentucky (2004).
[14] Davies, Alan J, Finite element method a rst approach,Oxford : Clarendon Press (1980).
[15] Versteeg, H. K, Introduction to computational uid dynamics the nite volume method, Har-
low: Longman Scientic and Technical (1995).
[16] Baker A. J, Finite elements 1-2-3, New York : McGraw-Hill (1991).
[17] Taylor C, Finite element programming of the Navier-Stokes equations, Swansea : Pineridge
Press (1981).
53
[18] Bruce Irons and Nigel Shrive, Finite element primer, Glasgow : Department of Trade and
Industry, National Engineering Laboratory (1986).
[19] Vivette Girault and Pierre-Arnaud Raviart, Finite element methods for Navier-Stokes equa-
tions theory and algorithms, Berlin: Springer, cop. (1986).
[20] O. Axelsson and V. A. Barker, Finite element solution of boundary value problems theory
and computation, Orlando,FL : Academic Press (1984).
[21] D. W. Jordan and P. Smith, Nonlinear ordinary differential equations, Oxford : Clarendon
Press (1983).
[22] John H. Mathews and Kurtis D. Flink, Numerical methods using MATLAB, Upper Saddle
River (NJ) : Pearson/Prentice Hall, cop. (2004).
54
8 APPENDIX A. ELEMENTS BASIS FUNCTIONS AND LOCAL BASIS
FUNCTIONS APPENDIX A
8 Appendix A. Elements Basis functions and Local Basis
Functions
Suppose that for a given nite element mesh there is associated with each node N
i
=
(x
i
, y
i
) a function ,dened on
with following properties.
1. The restriction of
i
to any element e
l
mesh is associated with each a polynomial
form
i
(x, y) =
T
s=1
C
il
(s)
x
p
s
y
q
s
; (x, y) e
l
where powers p
s
and q
s
, s = 1, 2, ..., T are independent of i and j.
2.
i
(N
j
) =
ij
for i, j = 1, 2, ...M
3.
i
is uniquely determined on every element edge by its value at the nodes belonging
to that edge.
4.
i
C(
)
5.
i
assumes non-zero values only in those elements to which N
i
belongs.
6. If N
i
is not on ,then
i
vanishes on the boundary of its support. If N
i
in on , then
i
vanishes on part of boundary of its support that lies in .
7. It is possible to chose a standard (or reference) element e in the x y plane with
local basis functions
1
( x, y), ...,
T
( x, y) of type
i
(x, y) =
T
s=1
C
il
(s)
x
p
s
y
q
s
and
nd for every element e
l
invertible afne variable transformation.
x = x( x, y) = f
11
x + f
12
y + b
1
y = y( x, y) = f
21
x + f
22
y + b
2
( x, y) e depends on l, such that this maps e onto e
1
(mapping nodes onto nodes)
and
r
( x, y) =
l
r
(x( x, y), y( x, y))
55
8 APPENDIX A. ELEMENTS BASIS FUNCTIONS AND LOCAL BASIS
FUNCTIONS APPENDIX A
denoting the inverse transformation by
x = x(x, y)
and y = y(x, y) thus (x, y) e
l
We can rewrite
r
( x, y) as
r
l
(x, y) =
r
l
(x, y)( x(x, y), y(x, y))
Local basis function over e
l
, dened by
r
(l)
(x, y) =
r
i
r
(l)
(x, y)
, (x, y) e
l
, r = 1, 2, ...TA local basis function is simply the restriction of some global
basis function to e
l
.
56
9 APPENDIX B. LAX MILGRAM LEMMA APPENDIX B
9 Appendix B. Lax Milgram Lemma
Consider a functional
f(u) =
_
a
b
_
1
2
p(x)(u
)
2
+
1
2
q(x)(u)
2
g(x)u
_
dx
, u V
V =
_
v C
2
[a, b]; v(a) = v(b) = 0
_
Where, p C
1
[a, b], q, p C[a, b], 0 < p
0
p(x) p
1
and 0 < q
0
q(x) q
1
for
a b with p
0
,p
1
,q
1
as constants.
a(u, v) =
_
a
b
{p(x)u
+ q(x)uv} dx
, u, v V
G(u) =
_
a
b
g(x)udx
, u V We can express f as, f(u) =
1
2
a(u, u) G(u), u V . Let, V be any arbitrary
Hilbert Space with inner product (., .)
v
and norm u
v
= (u, u)
v
1
2
; u V . Let a :
V V be a mapping with following four properties.
1. a(u + v, w) = .a(u, w) + .a(v, w), u, v, w V ,
2. a(w, u + v) = .a(w, u) + .a(w, v), u, v, w V ,
3. constant |a(u, v)| u
V
v
V
, u, v, w V i.e a is bounded.
4. constant > 0 a(u, v) u
2
v
, u V i.e a is coercive. Let G : V be a
mapping with following properties :
5. G(u + v) = G(u) + G(v) ; u, v, w V , i.e G is linear.
6. constant > 0 |G(u)| u
v
, u V i.e G is bounded.
Under these assumptions for
a
and
a
v
s
=
v
s
, where s = x
i
or s = t
vw = vw
Some consequences of these averaging rules are as following
u
i
u
j
= u
i
u
j
+ u
i
u
j
u
i
u
j
u
k
= u
i
u
j
u
k
+ u
i
u
j
u
k
+ u
j
u
k
ui + u
k
u
iuj + u
i
u
j
u
k
u
i
t
u
i
u
i
t
u
i
=
u
j
t
u
i
Some rules for derivative
1.
u
i
t
u
i
=
u
i
t
u
i
+
u
i
t
u
i
2.
p
x
i
u
i
=
p
x
i
u
i
+
p
u
i
u
i
3.
ij
x
i
u
i
=
ij
x
i
u
i
+
u
i
u
i
ij
4. u
j
u
i
x
j
u
i
u
j
u
i
x
j
u
i
= u
i
u
i
x
j
u
i
+
u
i
x
j
u
i
u
j
+ u
i
u
i
x
j
u
i
+ u
i
u
i
u
i
x
j
58
11 APPENDIX D. CREATING GEOMETRY IN GAMBIT APPENDIX D
11 Appendix D. Creating geometry in Gambit
We can create t-junction geometry by two ways, one is creating two rectangles perpendic-
ular to each other and then merging them second is creating vertex points and connecting
them by edges. We will do this by second way.
First we shall create the points that will be used to create the lines and then faces
of the domain. Operation > Geometry > Vertex > Create Vertex A(0,0), B(0,5),
C(5,1), D(0,1), E(2,1), F(3,1), G(2,3), H(3,3).
Figure 11.1: Buttons for drawing geometry
Now draw the straight lines that will complete the domain. Connect the points to
create the following line segments: AB, BC, CD, DE, EF, FG, GH, HA.
Operation > Geometry > Face > Form Face. Select all the line segments in the
drop list and create the face.
Operation > Mesh > Mesh Face. Select the face and specify the spacing or ratio.
Operation > Zones > Specify Boundary Types. Create boundary conditions as
follows: Left face = Velocity Inlet1, Right face = Velocity Inlet2, Upper face =
Pressure Outlet and all the other faces are walls.
59
11 APPENDIX D. CREATING GEOMETRY IN GAMBIT APPENDIX D
Save the Gambit le and export to the Fluent mesh.
60
12 APPENDIX D. SOLVING PROBLEM WITH FLUENT APPENDIX D
12 Appendix D. Solving problem with uent
Load the mesh into Fluent. File > Read > Case.
Check the mesh for errors. Grid > Check
For this problem, the default Solver settings will be sufcient. Ensure that the
proper viscous model is selected. Dene > Models > Viscous.
Now recall liquid water from the materials database so that it can be specied in
the boundary conditions. Dene > Materials. Enter the database by clicking on
Database. Select water liquid (h2o<l>) in the Fluid Materials list. Click Copy and
then Close. Now move the reference pressure into the ow domain.
Dene > Operating Conditions.
Boundary conditions can now be set. Dene > Boundary Conditions. Select uid
in the selection menu on the left and then click on Set. Change Material Name to
water-liquid. Now click on inlet in the Zones menu and enter the velocity-inlet win-
dow. Change Velocity Specication Method to Components and enter a velocity of
2.01e-4 m/s (liquid water at Re = 20) next to X-Velocity. Change the discretization
method to a higher order scheme.
Solve > Controls > Solution. Change the Discretization for Momentum to 2nd
Order Upwind.
The ow domain can now be initialized. Solve > Initialize > Initialize. Initialize
the ow with the inlet conditions.
Enable the plotting option for residuals and turn off automatic convergence check-
ing. Solve > Monitors > Residual.
The problem is ready to be iterated. Solve > Iterate. Start with 200 iterations.
Once Fluent has stopped iterating, we can post-process the data of our interest.
We can use Display > Contours. and view contour of velocity, pressure etc.
61