Académique Documents
Professionnel Documents
Culture Documents
R. Shankar Subramanian
Fluids such as water, air, ethanol, and benzene are Newtonian. This means that a
plot of shear stress versus shear rate at a given temperature is a straight line with
a constant slope that is independent of the shear rate. We call this slope the
viscosity of the fluid. All gases are Newtonian. Also, low molecular weight
liquids, and solutions of low molecular weight substances in liquids are usually
Newtonian. Some examples are aqueous solutions of sugar or salt.
Shear Stress
Any fluid that does not obey the Newtonian relationship between the shear
stress and shear rate is called non-Newtonian. The subject of Rheology is
devoted to the study of the behavior of such fluids. High molecular weight
liquids which include polymer melts and solutions of polymers, as well as
liquids in which fine particles are suspended (slurries and pastes), are usually
non-Newtonian. In this case, the slope of the shear stress versus shear rate curve
will not be constant as we change the shear rate. When the viscosity decreases
with increasing shear rate, we call the fluid shear-thinning. In the opposite case
where the viscosity increases as the fluid is subjected to a higher shear rate, the
fluid is called shear-thickening. Shear-thinning behavior is more common than
shear-thickening. Shear-thinning fluids also are called pseudoplastic fluids. A
typical shear stress versus shear rate plot for a shear-thinning fluid looks like
this.
shear-thinning fluid
Shear Rate
as
=
Here,
Apparent Viscosity
shear-thinning fluid
Shear Rate
Newtonian Region
Power-Law Region
shear-thinning fluid
Newtonian Region
log Shear Rate
The regions where the apparent viscosity is approximately constant are known
as Newtonian regions. The behavior between these regions can usually be
approximated by a straight line on these axes. It is known as the power-law
region. In this region, we can approximate the behavior by
log = a + b log
which can be rewritten as
= Kb
where
= K n1
Upon using the connection among the shear stress, apparent viscosity, and the
shear rate we get the power-law model.
= K n
where n is called the power-law index. Note that n = 1 corresponds to
Newtonian behavior. Typically, for shear thinning fluids, n lies between 1/ 3
and 1/ 2 , even though other values are possible.
Examples of shear-thinning fluids are polymer melts such as molten polystyrene,
polymer solutions such as polyethylene oxide in water, and some paints. You
can see that when paint is sheared with a brush, it flows comfortably, but when
the shear stress is removed, its viscosity increases so that it no longer flows
easily. Of course, the solvent evaporates soon and then the paint sticks to the
surface. The behavior of paint is a bit more complex than this, because the
viscosity changes with time at a given shear rate.
Some slurries and pastes exhibit an increase in apparent viscosity as the shear
rate is increased. They are called shear-thickening or dilatant fluids. Typical
plots of shear stress versus shear rate and apparent viscosity versus shear rate
are shown next.
Shear Stress
shear-thickening fluid
Apparent Viscosity
Shear Rate
shear-thickening fluid
Shear Rate
Some examples of shear-thickening fluids are corn starch, clay slurries, and
solutions of certain surfactants.
Most shear-thickening fluids tend to show
shear-thinning at very low shear rates.
Another important type of non-Newtonian fluid is a viscoplastic or yield stress
fluid. This is a fluid which will not flow when only a small shear stress is
applied. The shear stress must exceed a critical value known as the yield stress 0
for the fluid to flow. For example, when you open a tube of toothpaste, it would
be good if the paste does not flow at the slightest amount of shear stress. We
need to apply an adequate force before the toothpaste will start flowing. So,
viscoplastic fluids behave like solids when the applied shear stress is less than
the yield stress. Once it exceeds the yield stress, the viscoplastic fluid will flow
just like a fluid. Bingham plastics are a special class of viscoplastic fluids that
exhibit a linear behavior of shear stress against shear rate. Typical viscoplastic
behaviors are illustrated in the next figure.
Shear Stress
viscoplastic fluids
Bingham Plastic
Shear Rate
Shell Balances
R. Shankar Subramanian
When fluid flow occurs in a single direction everywhere in a system, shell balances are
useful devices for applying the principle of conservation of momentum. An example is
incompressible laminar flow of fluid in a straight circular pipe. Other examples include
flow between two wide parallel plates or flow of a liquid film down an inclined plane.
In the above situations, fluid velocity varies across the cross-section only in one
coordinate direction and is uniform in the other direction normal to the flow direction.
For flow through a straight circular tube, there is variation with the radial coordinate, but
not with the polar angle. Similarly, for flow between wide parallel plates, the velocity
varies with the distance coordinate between the two plates. If the plates are sufficiently
wide, we can ignore variations in the other direction normal to the flow which runs
parallel to the surfaces of the plates. If we neglect entrance and exit effects, the velocity
does not vary with distance in the flow direction in both cases; this is the definition of
fully-developed flow.
A momentum balance can be written for a control volume called a shell, which is
constructed by translating a differential cross-sectional area (normal to the flow) in the
direction of the flow over a finite distance. The differential area itself is formed by
taking a differential distance in the direction in which the velocity varies and translating it
in the other cross-sectional coordinate over its full extent. We shall see by example how
these shells are formed. The key idea is that we use a differential distance in the direction
in which velocity varies. Later, we consider the limit as this distance approaches zero and
obtain a differential equation. Typically this is an equation for the shear stress. By
inserting a suitable rheological model connecting the shear stress to the velocity gradient,
we can obtain a differential equation for the velocity distribution. This is then integrated
with the boundary conditions relevant to the problem to obtain the velocity profile. Once
the profile is known, we can calculate the volumetric flow rate and the average velocity
as well as the maximum velocity. If desired, the shear stress distribution across the crosssection can be written as well. In the case of pipe flow, we shall see how this yields the
well-known Hagen-Poiseuille equation connecting the pressure drop and the volumetric
flow rate.
Regarding boundary conditions, the most common condition we use is the no slip
boundary condition. This states that the fluid adjacent to a solid surface assumes the
velocity of the solid. Also, when necessary we use symmetry considerations to write
boundary conditions. At a free liquid surface when flow is not driven by the adjoining
gas dragging the liquid, we can set the shear stress to zero. In general at a fluid-fluid
interface the velocity and the shear stress are continuous across the interface. Even
though the interfacial region is made up of atoms and molecules and has a non-zero
thickness, we treat is as a plane of zero thickness for this purpose. This means that the
1
value of the velocity at a given point on the interface in one fluid is the same as the value
of the velocity at the same point on the interface in the second fluid. The same is true of
the shear stresses in the two fluids except when the interfacial tension varies with
position. This can happen if the temperature varies along the surface or if surface active
chemicals are present. Now, we shall proceed to construct a shell momentum balance
for steady laminar flow through a straight circular tube.
Steady Fully-Developed Incompressible Laminar Flow in a Straight Circular Tube
r
Flow
Shell
L
Cylindrical polar coordinates
( r , , z )
fully-developed laminar flow in the z-direction. The only non-zero velocity component
is vz and it only depends on the radial coordinate r . Therefore, vz = vz r .
( )
rz = rz ( r ) .
L units long
r units thick in the radial direction. Its inner cylindrical surface is at a radial
location r and the outer cylindrical surface is at r + r .
and
At steady state, the rate of accumulation of momentum in the fluid contained in the shell
is zero. Momentum enters at the left face and leaves at the right face with the flowing
fluid. The shell balance reads as follows.
Rate of entry of momentum into shell - Rate of efflux of momentum from the shell
+ Sum of all the forces acting on the fluid in the shell = 0.
Q = 2 r r vz
If the density of the fluid is
m = 2 r r vz
and the rate of entry of momentum, which is the product of the mass flow rate and the
velocity, is written as follows.
Rate of entry of momentum
= m vz = 2 r r vz2
= 2 r r vz2
and the two terms cancel each other. So the shell balance reduces to the statement that
the sum of the forces on the fluid in the shell is zero. The forces on the fluid consist of
the pressure force, the gravitational force, and the viscous force. We shall account for
each in turn.
The pressure force, given as the product of the area and the pressure acting on it, is
written as follows.
Pressure force
= 2 r r p ( 0 ) p ( L )
The force of gravity on the fluid in the shell, given as the product of the mass of the fluid
in the shell and the (vector) acceleration due to gravity g acts vertically downward. We
need to use the component of this force in the z -direction. This can be obtained as
Contribution in the flow direction from the gravitational force
where
= 2 r rL g cos
We now calculate the viscous force acting on the shell. The fluid at r + r exerts a
shear stress on the fluid in the shell in the positive z direction which we designate as
rz r + r . This notation means shear stress evaluated at the location r + r .
Multiplied by the surface area of the shell at that location this yields a force
2 r + r L rz r + r in the z direction. In the same way, the fluid within the
exerts a stress
rz ( r )
r.
As a
result, the fluid below exerts a reaction on the shell fluid that has the opposite sign. The
resulting force is 2 rL rz r . Therefore, the total force arising from the shear
( )
stresses on the two cylindrical surfaces of the shell fluid is written as follows.
Force from viscous stress
= 2 L ( r + r ) rz ( r + r ) r rz ( r )
p ( 0 ) p ( L ) ( r + r ) rz ( r + r ) r rz ( r )
r g cos +
=0
+
L
1 d
P
( r rz ) =
r dr
L
where we have introduced the symbol
P = p ( 0 ) p ( L ) + gL cos
for convenience. The quantity P is known as the hydrodynamic pressure or simply the
dynamic pressure. Its gradient is zero in a stationary fluid and therefore it is uniform
when the fluid is not in motion.
We can integrate the differential equation for the shear stress immediately by noting that
the right side is just a constant while the left side is the derivative of r rz . The result of
this integration is
P r2
r rz =
+ C1
L 2
where C1 is an arbitrary constant of integration that needs to be determined. We can
rewrite this result as
rz =
P r C1
+
L 2 r
According to this result, the shear stress should approach infinity as the radial coordinate
r 0 . This is not physically possible. Therefore, we must choose the arbitrary
constant of integration C1 = 0 and write the shear stress distribution in the tube as
follows.
rz =
P
r
2L
The shear stress is zero at the centerline and its magnitude is a maximum at the tube wall.
It varies linearly across the cross-section. We can explain why it is negative. Recall that
a force acting in the positive z direction is taken by convention here to be positive.
The fluid at the wall is stationary, however, and exerts a viscous force in the negative
z direction on fluid in the adjacent layer which is trying to slide past it. This is why
the shear stress, according to our convention, is negative. A sketch of the shear stress
distribution is given below.
rz ( r )
r
z
R
To find the velocity distribution, we need to introduce a relationship between the shear
stress and the velocity gradient. Recall the Newtonian constitutive model
rz =
dvz
dr
where is the dynamic viscosity of the fluid. Use of this constitutive model in the
result for the shear stress leads to
dvz
P
==
r
2 L
dr
This equation can be integrated immediately to yield
v z ( r ) = C2
P 2
r
4 L
where a new arbitrary constant of integration C2 has been introduced. To determine this
constant we must impose a boundary condition on the velocity field. This is the no slip
boundary condition at the tube wall r = R .
vz ( R ) = 0
Use of this condition leads to the result
P R2
C2 =
4 L
so that we can write the velocity distribution as follows.
P R2
vz ( r ) =
4 L
r2
1 R 2
vz ( r )
vmax
r = 0 . Its value
P R2
=
4 L
We can calculate the volumetric flow rate through the tube by multiplying the velocity at
a given r by the differential cross-sectional area 2 r dr to yield dQ and then
integrating across the cross-section.
P R2 R r 2
Q = dQ = 2 r vz ( r ) dr =
r 1 2 dr
2
L
R
0
0
0
Q
P R4
=
8 L
This is a result known as the Hagen-Poiseuille equation for the relationship between the
flow rate and the applied pressure drop for laminar flow through a circular tube. Now,
we can calculate the average velocity in the tube, defined as the volumetric flow rate per
unit area of cross-section.
vavg
P R2
Q
=
=
8 L
R2
It can be seen that in a circular tube, the average velocity is one-half the maximum
velocity.
Steady Fully-Developed Incompressible Laminar Flow Between Parallel Plates
A similar development can be made for flow between wide parallel plates that are
separated by a distance 2h . If the coordinate y is measured from the centerline toward
one of the plates, and z in the flow direction, we find that
P h2
vz ( y ) =
2 L
y2
1 h 2
This also is a parabolic velocity profile that is symmetric about the centerline. The
maximum velocity occurs at the centerline, y = 0 .
vmax
P h2
=
2 L
Let the plates be W units wide in the x direction. Then, the differential crosssectional area for flow is W dy and the differential volumetric flow rate would be
Q = W vz ( y ) dy = 2 W vz ( y ) dy
h
y2
0 1 h 2 dy
2 h
Ph
= 2W
2 L
yielding
2 PWh3
Q=
3 L
The average velocity is obtained by dividing
vavg
P h2
=
3 L
The laminar flow of a liquid film down an inclined plane can be modeled in the same
manner as flow between parallel plates. In this case, if we measure the y coordinate
from the free surface of the liquid, the above parabolic velocity profile applies directly.
This is because the flow between parallel plates is symmetric about the centerline y = 0 ,
which leads to the velocity gradient being zero along the centerline. For the flow of a
liquid film driven by gravity, the shear stress at the free liquid surface is negligible. This
means that the velocity gradient is negligible at y = 0 . This is why the velocity profile
in a falling liquid film in laminar flow is represented by one-half of the parabola that is
obtained for flow between parallel plates. The ratio of the average velocity to the
maximum velocity is the same as that for flow between parallel plates, namely 2 / 3 . If
we use h for the film thickness, we can immediately write the volumetric flow rate in the
film as
PWh3
Q=
3 L
The free surface of the liquid film is exposed to the atmosphere everywhere so
that there can be no pressure gradient in the direction of the flow. This means
that
p ( 0) = p ( L ) .
P
= g cos
L
which yields
g Wh3 cos
Q=
3
The utility of this relationship between the volumetric flow rate and the film
thickness lies in the fact that it can be used to determine the film thickness for a
given volumetric rate of flow.
Summary
These notes introduce you to a simple approach for modeling one-dimensional
flow situations. The main idea is to construct a shell of fluid that is long in the
flow direction, but of differential thickness in the direction along which the
velocity varies. By accounting for all the forces on the fluid contained in this
shell, and permitting the thickness of the shell to approach zero, we obtain a
differential equation for the shear stress distribution in the direction normal to
that of the flow. Introducing a constitutive model, such as the Newtonian model,
permits us to develop a differential equation for the velocity distribution. The
shear stress and velocity distributions can be obtained by integrating the
applicable differential equations. The arbitrary constants, which must be
introduced when the integrations are performed, can be evaluated by the
application of boundary conditions based on physical principles. When the
velocity distribution has been obtained, it can be multiplied by a differential area
element and integrated across the cross-section to obtain the volumetric flow
rate. The average velocity can be obtained as the volumetric flow rate divided by
the cross-sectional area.
The shell balance approach is limited to one-dimensional flow situations and
straight geometries. In the more general case, we must use partial differential
10
Chemical engineers frequently encounter the flow of a mixture of two fluids in pipes. Liquid-gas
or liquid-vapor mixtures are encountered in condensers and evaporators, gas-liquid reactors, and
combustion systems. Also, the transport of some solid materials in finely divided form is
accomplished by making a slurry of the solid particles in a liquid, and pumping the mixture
through a pipe. Liquid-liquid mixtures are encountered when dealing with emulsions as well as
in liquid-liquid extraction.
Two-phase flow is a difficult subject principally because of the complexity of the form in which
the two fluids exist inside the pipe, known as the flow regime. Toward the end of these notes,
well see some examples of these regimes. It is difficult to construct a model from first
principles in all but the most elementary situations. Dimensional analysis is used to establish the
relevant groups to aid in designing suitable experiments. Most available empirical results are
applicable only to gas-liquid two-phase flow. A little reflection will convince you that the
orientation of the pipe makes a difference in the flow regime because of the role played by
gravity and the density difference between the two fluids.
In two-phase flow, the concept of hold-up is important. It is the relative fraction of one phase in
the pipe. This is not necessarily equal to the relative fraction of that phase in the entering fluid
mixture.
The usual question for the engineer is that of calculating the pressure drop required to achieve
specified flow rates of the gas and the liquid through a pipe of a given diameter. To make design
calculations involving two-phase flow, Perrys Handbook is a useful resource. It summarizes
correlations that are currently used in industry. Also, an informative chapter in Holland and
Bragg (1995) is devoted to gas-liquid two-phase flow. Here, we only consider the qualitative
features of gas-liquid flow in a horizontal pipe to give you an appreciation of the complexity of
two-phase flow when compared with the flow of a single fluid phase.
The sketches on the following pages depict various regimes of co-current two-phase flow in a
horizontal pipe through which a gas-liquid mixture is flowing. We use VL and VG to designate
the superficial velocity of the liquid and gas phases, respectively. We first consider a situation
where the ratio VL / VG is large, and then the regime where the ratio is lower.
Consider a mixture that is almost completely liquid. This means that the ratio VL / VG is large.
In this case, the gas is present in the form of small bubbles that will rise to the top portion of the
pipe as depicted in the sketch on the left. As the gas flow rate is increased, the bubbles become
larger as shown on the right. Further increase in VG leads to annular mist flow through the
intermediate stage of slug flow; these new terms are defined later.
High VL/VG
As VG
Bubble Flow
Plug Flow
Lower VL/VG
As VG
Gas
Gas
Liquid
Liquid
Stratified Flow
Now, consider lower values of the ratio VL / VG . The sketches show what happens in this
situation. Now, there is enough gas to form a layer at the top of the pipe. This type of flow is
called Stratified Flow. As VG is increased, waves begin to appear on the surface of the liquid,
2
and this type of flow is called Wavy Stratified Flow. Further increase in the gas velocity leads
to the formation of higher wave crests that contact the pipe wall. The gas then flows in the form
of slugs, leading to the term Slug Flow.
Eventually, at large values of the gas velocity, we encounter a flow regime that is known as
annular mist flow. In this case, the liquid flows in the form of a thin film on the wall of the
tube, with the gas flowing in the core, thus forming an annulus. Most of the liquid is entrained in
the gas in the form of small drops; this is the reason it is called a mist. In the figure, the liquid
drops entrained in the gas are exaggerated in size.
Gas
References
F.A. Holland and R. Bragg, Fluid Flow for Chemical Engineers. 1995 Edward Arnold
Publishers, London.
Perrys Chemical Engineers Handbook, 7th Edition. 1997 (Ed: R.H. Perry, D.W. Green, and J.O.
Maloney), McGraw-Hill, New York.
P0
PL
( rvr ) = 0
r
(1)
rvr must be constant across the cross-section of the tube. Because it is zero
at r = 0 , it must be zero everywhere. This implies that vr = 0 . Therefore, the flow is
unidirectional and vz is the only non-zero velocity component.
Therefore,
dynamic pressure can, at best, depend only on t and z . The component in the z direction yields the following governing equation for the velocity vz t , r in the tube.
( )
vz
P vz
=
+
r
t
z r r r
(2)
vz
P vz
r
=
(3)
t
z r r r
and recognizing that the left side can only depend on t and z , while the right side can
only depend on t and r , we conclude that each can, at best, depend only on t . It can be
shown that the pressure gradient becomes steady rapidly in such a system, so that we can
P
by its
z
P = P0 PL .
replace
steady representation
PL P0
P
=
L
L
vz P vz
=
+
r
L
r r r
t
(4)
(No Slip)
vz ( 0, r ) = 0
(5)
vz ( t ,0 ) is finite
(6)
vz ( t , R ) = 0
(7)
T=
t
R
;
2
y=
r
;
R
V (T , y ) =
vz
( PR 2 / 4 L )
( / ).
of the scale for the velocity, and also about the factor 4 that is introduced in that scale.
The velocity scale is obtained by considering the steady problem. In that problem, the
PR 2 / ( L ) .
The factor
steady solution appear especially simple, as you will see a bit later.
Non-dimensionalization leads to
V
1 V
=4+
y
T
y y y
(8)
The scaled velocity field satisfies the following initial and boundary conditions.
Initial Condition:
V (0, y ) = 0
(9)
V (T ,0 ) is finite
(10)
V (T ,1) = 0
(11)
Boundary Conditions:
No Slip:
From the discussion in class, we know that directly applying separation of variables to
Equations (8) to (11) will fail because of the appearance of an inhomogeneity in the
governing equation. Therefore, we first write the solution as the sum of a steady state
solution and a transient solution.
V (T , y ) = Vs ( y ) + Vt (T , y )
(12)
The steady solution must satisfy Equation (8) without the time derivative.
4+
1 d dVs
y
y dy dy
=0
(13)
and
Vs ( 0 ) is finite
(14)
Vs (1) = 0
(15)
We first substitute the solution in Equation (12) into Equations (8) to (11), and make use
of the fact that the steady solution must satisfy Equation (13) and the boundary conditions
in Equations (14) and (15). This yields the following governing equation and initial and
boundary conditions for the transient part of the solution.
Vt
1 Vt
=
y
T
y y y
(16)
Vt (0, y ) = Vs ( y )
(17)
Vt (T ,0 ) is finite
(18)
Vt (T ,1) = 0
(19)
You can see that the consequence of separating the solution into a steady and a transient
part is to yield a homogeneous differential equation for the transient contribution. The
boundary conditions in the y -coordinate are unaffected, and amenable to writing a
product class solution. The initial condition is different from that on the complete
velocity field. In fact, it is the inhomogeneity in this initial condition that produces a
non-trivial solution for the transient problem.
We can obtain the steady field by
integrating Equation (13) and applying the boundary conditions. This leads to
Vs ( y ) = 1 y 2
(20)
( ) ( )
equations.
dG
+ 2G = 0
dT
1 d d
+ 2 = 0
y
y dy dy
(21)
(22)
G = Ce T
2
(23)
= C1 J 0 ( y ) + C2Y0 ( y )
(24)
where J 0 and Y0 are known as the Bessel functions of the first and second kinds,
respectively, of order zero. Information about these functions can be obtained from
Abramowitz and Stegun (1965).
In general, the Bessel functions J p y and
equation
1 d d
+ ( 2 p 2 ) = 0
y
y dy dy
Here are sample graphs showing the behavior of the functions
followed by
Y0 ( x ) and Y1 ( x ) .
(25)
J 0 ( x) and J1 ( x) ,
Yp ( x ) are singular at
x = 0 , that is, they approach minus infinity as x approaches zero. This is not
compatible with the boundary condition stated in Equation (18) that Vt remain finite at
y = 0 . Therefore, the constant C2 in Equation (24) must be set equal to zero. This
leaves us with a product class solution
Vt (T , y ) = Be T J 0 ( y )
2
B . We
Vt (T , y )
and
(26)
know that this solution satisfies the governing differential equation for
and the condition that the solution must remain finite at y = 0 . Now, we
need to apply the remaining conditions. The no-slip boundary condition at the wall that
leads to Equation (19) can be satisfied by requiring
J0 ( ) = 0
(27)
As you can see from the sketch showing the behavior of J 0 , this equation has more than
one root. In fact, it has an infinite number of positive roots that come paired with
negative roots of the same magnitude. It is known that J 0 x = J 0 x so that we
( )
( )
can simply use the infinite set of positive roots, labeled n ( n = 1, 2,3...) . The first 20
roots of Equation (27), labeled
Corresponding to each root of Equation (27), the product form appearing in Equation (26)
with a multiplicative constant Bn will satisfy the governing equation and the two
boundary conditions in the y -coordinate. We still need to satisfy the initial condition
given in Equation (17). This can be done by using a sum of all these solutions (recall that
in a linear problem, we can use superposition), which leads us to write the solution in the
form
Vt (T , y ) = Bn e n T J 0 ( n y )
2
(28)
n =1
Vt ( 0, y ) = Vs ( y ) = Bn J 0 ( n y )
(29)
n =1
To evaluate the constants, we need to use the following results from Hildebrand (1976).
More general results can be found in Abramowitz and Stegun (1965).
1
y J ( y ) J ( y ) dy = 0,
0
mn
(30)
J1 ( n )
2
y
J
y
dy
=
provided J 0 ( n ) = 0
0 0 ( n )
2
d
y p J p ( y ) = y p J p 1 ( y )
dy
2p
J p 1 ( x ) + J p +1 ( x ) =
J p ( x)
x
2
(31)
(32)
(33)
Vs ( y ) = Bn J 0 ( n y )
n =1
(34)
from
Bm =
J1
y V ( y ) J ( y ) dy
( )
0
(35)
Evaluate the integral in Equation (35) by splitting it into the sum of two integrals. One of
these can be found immediately by using Equation (32). To find the other, youll need to
use integration by parts in conjunction with Equation (32). The final result, after
simplification using Equation (33), is
Bm =
(36)
m3 J1 ( m )
with
J 0 ( n y ) n2T
e
3
J
(
)
n =1 n
1
n
V (T , y ) = 1 y 8
2
(37)
The velocity field, calculated from Equation (37) at a few selected values of T, is
displayed below.
References
F.B. Hildebrand, Advanced Calculus for Applications, Prentice-Hall, Englewood Cliffs,
1976.
M. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, New
York, 1965.
dA
dV
V
1
We also have marked the bounding surface A of this control volume, called the control
surface (CS) and shown an element of surface area dA and the unit outward normal
(vector) to that area element, n . Let us write a mathematical representation of the
statement that the mass of fluid in the control volume is unchanged. If we designate the
total mass of material in the control volume as M and the net rate of entry of mass into
the control volume as m , conservation of mass can be written as
dM
= m
dt
(1)
Now, we need to work out suitable results for the left and right sides of the above
equation. If we consider a differential volume dV , the mass of fluid in that volume is
obtained by multiplying the volume by the local density at that point . By adding up
all the differential volumes within the total volume
fluid M . Therefore, we can write
M=
dV
(2)
dM
. Therefore,
dt
dM
d
=
dV
dt
dt V
(3)
Now, we need to develop a result for the net rate of entry of fluid into the control volume
through the control surface. For this, we consider the differential area element dA . If the
velocity vector is V , the component of this velocity that is directed into the control
volume is given by V n , because the unit normal vector n points outward from the
control volume. This result, multiplied by the area of the element dA , gives the
volumetric rate at which fluid enters the control volume through this area element,
labeled dQ .
dQ = V n dA
(4)
The corresponding rate at which mass enters the control volume through the area element
dA , labeled dm , can be written as
dm = dQ = ( V n ) dA
(5)
Adding up all the contributions from the differential area elements, which implies
integration over the entire control surface, leads to a result for the net rate of entry of
mass into the control volume.
m = ( V n ) dA
(6)
CS
Now, we can rewrite the principle of conservation of mass, given in Equation (1) as
dM
dt
(V n ) dA
(7)
CS
Rate of
increase of
mass M in
the control volume
where we have identified the physical meaning of the terms in the left and right sides of
the equation.
In a steady state situation, the time rate of change of the mass of material in a control
volume is zero. In this case, we simply obtain
(V n )
dA = 0
(8)
CS
Physically, this result implies that the influx of mass into the control volume must equal
the efflux of mass at steady state.
Here is an example of how we may use this statement of Conservation of Mass at
steady state. A fluid is in steady flow through a pipe of changing cross-section as shown
in the sketch.
2 , V2 , A2
1 , V1 , A1
Location 1 is at the inlet and location 2 is at the outlet. Assume the velocity profiles are
flat. The control volume is indicated by the dashed boundary. It is shown as being
slightly separated from the physical boundary only for clarity. In reality, its surface
coincides with the physical surface of the pipe. Note that there is no flow through most
of the control surface, that is, V n = 0 . There is flow only at the inlet (1) and exit (2).
At the inlet surface the velocity points in a direction opposite to that of the normal vector.
Therefore, V n becomes V1 . In a like manner, at the exit surface, the velocity
points in the same direction as the normal vector which leads to
V n
becoming
V2 .
Because V1 and V2 are constant across the surfaces involved, the integrals are easy to
evaluate and we get
1 V1 A1 + 2 V2 A2 = 0
(9)
1 V1 A1 = 2 V2 A2
(10)
or
The product of the uniform velocity and the area is the volumetric flow rate Q .
Therefore, we can rewrite this as
1 Q1 = 2 Q2
The mass flow rate
m = Q
so that
(11)
Minor Losses
R. Shankar Subramanian
Minor losses is a term used to describe losses that occur in fittings, expansions, contractions, and
the like. Fittings commonly used in the industry include bends, tees, elbows, unions, and of
course, valves used to control flow. Even though these losses are called minor, they can be
substantial compared to those for flow through short straight pipe segments. Losses are
commonly reported in velocity heads. A velocity head is
losses as
V2
. where K L
hm = K L
2g
V2
.
2g
Typical values of K L for some common fittings are given below. Usually, the values depend
upon the nominal pipe diameter, the Reynolds number, and the manner in which the valve is
installed (screwed or flanged). Manufacturers data should be used wherever possible.
Globe Valve (fully open): 5.5 - 14
Gate Valve (fully open): 0.03 - 0.80
Swing Check Valve (fully open): 2.0 - 5.1
Standard 45o Elbow: 0.2 - 0.4
Long radius 45o Elbow: 0.14 - 0.21
Standard 90o Elbow: 0.21 - 2.0
Long radius 90o Elbow: 0.07 - 1.0
Tee: 0.1 - 2.4
Sudden Expansion and Sudden Contraction
A sudden expansion in a pipe is one of the few cases where the losses can be obtained from the
basic balances. The expression for K L is given by
d2
K L = 1 2
D
Here,
d and D
represent the diameters of the smaller and larger pipes, respectively. For a
d
d
0.76 . For smaller values of , we
D
D
d2
.
can use the empirical relation K L = 0.42 1
2
D
In both cases, we should multiply K L by the velocity head in the pipe segment of diameter d .
sudden contraction, we can use the same result if
d
= 0,
D
yielding
K L = 1.
Drag on disks
R. Shankar Subramanian
Clift, Grace, and Weber (Bubbles, Drops, and Particles, Academic Press, 1978) provide
correlations based on experimental data that can be used to calculate the drag coefficient
on a disk moving parallel to its axis over a limited range of values of the Reynolds
number. I have reproduced their results below for your use.
The definitions of the Reynolds number and the drag coefficient are given below.
d eV
Re =
CD =
Here,
2D
a 2V 2
is its viscosity.
The symbol D stands for the drag on the sphere. The radius of the disk is
thickness is H , leading to the following result for the equivalent diameter d e .
d e = ( 6a H )
2
a and
its
1
3
Drag Coefficient
For
Re 0.01 ,
CD =
64 Re
1+
Re 2
For
CD =
64
1 + 10 x )
(
Re
where
For 1.5 < Re 133 ,
CD =
64
1 + 0.138Re0.792
Re
Clift, Grace, and Weber state that when vortex shedding begins,
to
Re ,
1.17
for
CD becomes insensitive
Re > 1,000 ,
satisfies 0.1 < ReT < 100 . For calculating the terminal settling velocity in this range
of values of the Reynolds number, the drag coefficient correlations given earlier can be
used iteratively. The steady motion gives way to more complicated tumbling motion for
ReT > 100 , even though the steady motion persists to higher ReT up to about 172
when the moment of inertia of the disk is relatively small. It appears that a parameter
1
3
1
3,
CD Re the amplitude
2
T
of the oscillation increases and the disk moves in a succession of curved arcs, a motion
1
2
called glide-tumble. At even higher CD ReT 3 , the disk executes a tumbling motion,
rotating continually about a diameter and following a trajectory that is nearly rectilinear,
but not vertical.
When the disk executes unsteady motion, the drag correlations for steady drag
cannot be used. More information regarding this situation can be found in page 149 of
the book by Clift, Grace, and Weber.
Re =
CD =
d pV
8D
d p2V 2
d p is the diameter of the sphere, V is its velocity, is the density of the fluid,
and is its viscosity. The symbol D stands for the drag on the sphere.
Here,
Drag Coefficient
The entire range of Reynolds numbers has been divided into 10 intervals and in each, the
curve for the drag coefficient versus the Reynolds number is fitted to a suitable
expression. In the results given below, w = log10 Re .
For
Re 0.01 ,
CD =
9 24
+
2 Re
For
0.01 < Re 20 ,
CD =
24
0.820.05 w )
1 + 0.1315Re(
Re
For
20 Re 260 ,
CD =
24
1 + 0.1935Re0.6305
Re
For
For1.5 10
Re 1.2 10 4 ,
For
For
For
4 105 Re 106 ,
For 10
CD = 0.1w 0.49
8 104
CD = 0.19
Re
< Re ,
You can use the above equations to evaluate the drag coefficient when needed.
Terminal Settling Velocity
If you want to calculate the terminal settling velocity of a sphere, you will find that this
velocity appears in both the drag coefficient and in the Reynolds number. Therefore, you
must perform an iterative calculation to find the answer. To avoid doing this, Clift,
Grace, and Weber also provide results that permit the direct calculation of the terminal
settling velocity. Recall that for terminal settling, the drag on the sphere is equal to its
net weight, which is the weight minus the buoyant force on the sphere. We define the
Reynolds number in the same way as before, but with the understanding that it now
applies with V = Vterminal . A new dimensionless group N D = CD Re is introduced
in which the terminal settling velocity does not appear. Therefore you can calculate this
group and use the equations given below to calculate the value of the Reynolds number
corresponding to a given value of N D . From the Reynolds number, you can
2
For
= log10 N D .
N D 73; Re 2.37 ,
Re =
ND
1.7569 104 N D2 + 6.9252 107 N D3 2.3027 1010 N D4
24
For
Fluid at
R
z
T0
T ( R, z ) = Tw
Objective
To obtain the steady temperature distribution T ( r , z ) in the fluid, and to calculate the rate of heat
transfer from the wall to the fluid
Assumptions
1. Steady fully developed laminar flow; steady temperature field.
2. Constant physical properties , , k , C p -- This assumption also implies incompressible
Newtonian flow.
3. Axisymmetric temperature field
T
0.
Velocity Field
Poiseuille Flow
vr = 0;
v = 0
r2
vz ( r ) = v0 1 2
R
Energy Equation
Subject to assumption (2), Equation (B.9.2) from Bird et al. (page 850) can be written as follows.
1
T
C p
vr = 0
+ vr
v = 0
T
T v T
+ vz
+
z
r
r
3
1 T
=k
r
r r r
2
2
1 T T
+
+
+ v
2
2
z 2
r
1 T
r 2 T
v0 1 2
=
r
R
z
r
r
2
T
+
2
z
Boundary Conditions
Inlet:
T ( r , 0 ) = T0
Wall:
T ( R, z ) = Tw
Centerline:
T ( 0, z ) is finite
or
T
( 0, z ) = 0
r
Because of the appearance of the axial conduction term in the governing differential
equation, we should write another boundary condition in the z coordinate. But actually, the
inlet condition written above is incompatible with the inclusion of axial conduction in the
problem, because conduction will lead to some of the information about the step change in wall
temperature at the inlet to propagate backward. As we shall see shortly, well neglect axial
conduction, which will obviate the need for writing a second condition in the z coordinate.
Non-Dimensionalization
We shall use the following scheme for scaling (or non-dimensionalizing) the variables.
T Tw
r
, Y= ,
T0 Tw
R
Z=
Rv
z
, where the Pclet Number Pe = 0 .
R Pe
This permits us to transform the governing differential equation and boundary conditions to the
following form.
(1 Y ) Z = Y1 Y Y Y + Pe1
2
2
Z 2
(Y , 0 ) = 1
(1, Z ) = 0
( 0, Z ) is finite
or
( 0, Z ) = 0
Rv0
Rv0
= Re Pr
The physical significance of the Pclet number can be inferred by recasting it slightly.
Pe =
v0C p T
T
k
R
Note that the numerator represents the order of magnitude of the convective flux in the main
flow direction, whereas the denominator stands for the order of magnitude of the conduction flux
in the radial direction. If we wish to compare the rates of energy transport by these two
mechanisms in the same direction, we can multiply the Pclet number by L / R where L is a
characteristic length in the axial direction.
1
For large values of Pe , we can see that
1 . Therefore, in the scaled energy equation, the
Pe 2
term involving axial conduction can be safely neglected. Physically, there are two mechanisms
for transporting energy in the axial direction, namely, convection and conduction. Because the
Pclet number is large, we are able to neglect transport by conduction in comparison with
transport by convection. On the other hand, in the radial direction, there is only a single
mechanism for transport of energy, namely conduction. By performing calculations including
conduction in the axial direction, it has been established that it is safe to neglect axial conduction
for Pe 100. To learn about how to include axial conduction, you can consult the articles by
Davis (1973), Acrivos (1980), and Papoutsakis et al. (1980).
Let us make a sample calculation of the Pclet number for laminar flow heat transfer in a tube.
m2
, and well
The thermal diffusivity of common liquids is typically in the range 107 2 107
s
use the larger limit. Choose
R = 10 mm,
v0 = 0.05
m
m2
, = 2 107
s
s
This yields, Pe = 2,500 , which is much larger than 100. We can check to see if the flow is
laminar by calculating the Reynolds number. If the fluid is water, 10
m2
, which yields a
s
(1 Y ) Z = Y1 Y Y Y
2
We can solve this equation by separation of variables, because the boundary conditions in the
Y coordinate are homogeneous. The method of separation of variables yields an infinite series
solution for the scaled temperature field.
(Y , Z ) = An e Z n (Y )
n =1
2
n
In the above solution, the functions n (Y ) are the characteristic functions or eigenfunctions of
a proper Sturm-Liouville system.
1 d d
Y
Y dY dY
2
2
+ (1 Y ) = 0
d
( 0 ) = 0 or ( 0 ) is finite
dY
(1) = 0
The above ordinary differential equation for (Y ) can be solved by applying the following
transformations to both the dependent and the independent variables (Lauwerier, 1951, Davis,
1973).
X = Y 2
W ( X ) = e 2 (Y )
d 2W
dW 1
+ (1 X )
+ W = 0
2
dX
dX 4 2
This is known as Kummers equation. It has two linearly independent solutions, but only one is
bounded at X = 0. Because ( 0 ) must be bounded, we must require that W ( 0 ) also remain
bounded. This rules out the singular solution, leaving us with the regular solution
where
W ( X ) = c M , 1, X
2 4
a ( a + 1)" ( a + n 1) X n
+"
b ( b + 1)" ( b + n 1) n !
Application of the boundary condition at the tube wall, (1) = 0 , leads to the following
transcendental equation for the eigenvalues.
1
M , 1, = 0
2 4
The above equation has infinitely many discrete solutions for , which we designate as n , with
n assuming positive integer values beginning from 1. Corresponding to each value n , there is
an eigenfunction n (Y ) given by
n ( Y ) = e
nY 2
2
Wn ( nY 2 )
1
2.7042
2
6.6790
3
10.673
4
14.671
5
18.670
Note that technically {n2 }
n2
7.3127
44.609
113.92
215.24
348.57
is the set of eigenvalues, even though we use the term loosely to
(Y ) (Y ) Y (1 Y ) dY = 0,
2
mn
Using this orthogonality property, it is possible to obtain a result for the coefficients in the
solution by separation variables.
1
(Y ) Y (1 Y ) dY
2
An =
0
1
(Y ) Y (1 Y ) dY
2
n
qw ( z ) = h ( z )(Tw Tb )
where the bulk or cup-mixing average temperature Tb is introduced. The way to experimentally
determine the bulk average temperature is to collect the fluid coming out of the system at a given
axial location, mix it completely, and measure its temperature. The mathematical definition of
the bulk average temperature was given in an earlier section.
1
Tb =
2 rV (r )T ( r , z ) dr
0
2 rV (r ) dr
0
where the velocity field V ( r ) = v0 (1 r 2 / R 2 ) . You can see from the definition of the heat
transfer coefficient that it is related to the temperature gradient at the tube wall in a simple
manner.
T
k
( R, z )
h ( z ) = r
(Tw Tb )
We can define a dimensionless heat transfer coefficient, which is known as the Nusselt number.
(1, Z )
2hR
= 2 Y
Nu ( Z ) =
b ( Z )
k
By substituting from the infinite series solution for both the numerator and the denominator, the
Nusselt number can be written as follows.
Ae
Nu ( Z ) = 2
n =1
Ae
n =1
n2 Z
2
nZ
d n
(1)
dY
Y (1 Y ) (Y ) dY
2
The denominator can be simplified by using the governing differential equation for n (Y ) , along
with the boundary conditions, to finally yield the following result.
2
d
An e n Z n (1)
dY
n =1
Nu =
n2 Z
dn
e
2 An 2
(1)
n dY
n =1
We can see that for large Z , only the first term in the infinite series in the numerator, and
likewise the first term in the infinite series in the denominator, is important. Therefore, as
Z , Nu
12
2
= 3.656 .
The sketch qualitatively illustrates the behavior of the Nusselt number as a function of
dimensionless axial position.
Nu
3.656
0
0
A similar analysis is possible in the case of a uniform wall flux boundary condition.
Extensions of the Graetz solution by separation of variables have been made in a variety of ways,
accommodating non-Newtonian flow, turbulent flow, and other geometries besides a circular
tube.
Fluid at
T0
R
z
t
T ( R, z ) = Tw
We shall now construct the Lvque solution which is built on the assumption that the thickness
of the thermal boundary layer t R . This assumption leads to the following simplifications.
1. Curvature effects can be neglected in the radial conduction term.
1 T
1 T 2T
.
derivative
r
can
be
approximated
by
R
=
r r r
R r r r 2
2. Because we are only interested in the velocity distribution within the thermal boundary layer,
we expand the velocity field in a Taylor series in distance measured from the tube wall and retain
the first non-zero term.
Defining x = R r , we can rewrite the velocity distribution as
( R x )2
x x2
x
vz ( r ) = v0 1
= v0 2 2 2v0
2
R
R
R R
Recall that a power series obtained by any method is a Taylor series. The above approach is
simpler than working out the derivatives of vz ( r ) in the x coordinate, evaluating them at the
wall, and constructing the Taylor series.
3. Because the conditions outside the thermal boundary layer are those in the fluid entering the
tube, we shall use the boundary condition T ( x ) T0 instead of the centerline boundary
condition employed in obtaining the Graetz solution.
Beginning with the simplified energy equation in which axial conduction has been neglected
already, and invoking the above assumptions, we have the following governing equation for the
temperature field.
2v0
x T
2T
= 2
x
R z
where the chain rule has been used to transform the second derivative in r to the second
derivative in x .
The temperature field T ( x, z ) satisfies the following boundary conditions.
T ( x, 0 ) = T0
T ( 0, z ) = Tw
T ( , z ) = T0
We shall work with a dimensionless version of these equations. For consistency, we scale the
temperature and axial coordinate in the same manner as before.
T Tw
T0 Tw
Z=
z
R Pe
We define a new scaled distance from the wall via X = x / R . The scaled governing equation
and boundary conditions are given below.
2X
2
=
Z X 2
( X , 0) = 1
( 0, Z ) = 0
( , Z ) = 1
10
The similarity of this governing equation and boundary conditions to those in the fluid
mechanical problem in which we solved for the velocity distribution between two plates when
one of them is held fixed and the other is moved suddenly is not a coincidence. For small values
of time in the fluid mechanical problem, we replaced the boundary condition at the top plate with
one at an infinite distance from the suddenly moved plate, and used the method of combination
of variables to solve the equations. It would be worthwhile for you to go back and review the
notes on combination of variables at this stage.
By invoking ideas very similar to those used in the fluid mechanical problem, we postulate that a
similarity solution exists for the temperature field in the present problem. That is, we assume
( X , Z ) = F ( ) where the similarity variable = X / ( Z ) . The variable ( Z ) represents the
scaled thermal boundary layer thickness, and is unknown at this stage. We make the necessary
transformations using the chain rule.
dF
d dF
X d dF
=
= 2
=
dZ d
Z
Z d
dZ d
dF
1 dF
=
=
X
X d
d
1 d dF
1 d 2F
2
1 dF 1 dF
=
=
=
=
X d d 2 d 2
X 2
X ( Z ) d X d
Using these results, the partial differential equation for ( X , Z ) is transformed to an ordinary
differential equation for F ( ) .
d 2F
d dF
+ 2 2 2
=0
2
d
dZ d
It is evident that the similarity hypothesis will fail unless the quantity inside the parentheses is
required to be independent of Z , and therefore, a constant. For convenience, we set this
constant to 3/2. Therefore, we have an ordinary differential equation for F ( ) and another for
(Z ) .
d 2F
dF
+ 3 2
=0
2
d
d
d 3
=
dZ 2
11
To derive the boundary conditions on these functions, we must go to the boundary conditions on
( X , Z ) . In a straightforward way, we see that ( 0, Z ) = 0 yields F ( 0 ) = 0 , and ( , Z ) = 1
leads to F ( ) = 1 . The remaining (inlet) condition gives
X
= 1
(0)
( X , 0 ) = F
(0) = 0
Integration yields the following solution for the scaled boundary layer thickness ( Z ) .
1/ 3
9
(Z ) = Z
2
F ( ) =
d
d
3
1
e d
=
( 4 / 3) 0
In the thermal entrance region, when the thermal boundary layer is thin, we can approximate the
bulk average temperature Tb by the temperature of the fluid entering the tube T0 . Therefore, we
define the heat transfer coefficient in this entrance region by
qw = k
T
( R, z ) = h (Tw T0 )
r
12
2 dF
( 0, Z ) =
(0)
X
( Z ) d
dF
( 0 ) , we obtain the following approximate result for the
d
Nusselt number in the thermal entrance region.
Nu ( Z ) 1.357 Pe
1/ 3
1/ 3
R
z
Comparison with the exact solution shows this result is a good approximation in the range
Pe
Pe
z
2500 R
50
References
M. Abramowitz and I. Stegun, Handbook of Mathematical Functions, Dover, New York
(1965).
A. Acrivos, The extended Graetz problem at low Pclet numbers, Appl. Sci. Res. 36, 35 (1980).
E.J. Davis, Can. J. Chem. Engg. 51, 562 (1973).
L. Graetz, Ueber die Wrmeleitungsfhigkeit von Flssigkeiten, Annalen der Physik und Chemie
18, 79 (1883).
H.A. Lauwerier, Appl. Sci. Res. A2, 184 (1951).
M.A. Lvque, Les lois de la transmission de chaleur par convection, Annales des Mines,
Memoires, Series 12, 13, 201-299, 305-362, 381-415 (1928) {as cited by J. Newman, Trans.
ASME J. Heat Transfer, 91, 177 (1969)}
E. Papoutsakis, D. Ramkrishna, and H.C. Lim, The extended Graetz problem with Dirichlet wall
boundary conditions, Appl. Sci. Res. 36, 13 (1980).
13
Natural Convection
R. Shankar Subramanian
Natural or Buoyant or Free convection is a very important mechanism that is operative in a
variety of environments from cooling electronic circuit boards in computers to causing large
scale circulation in the atmosphere as well as in lakes and oceans that influences the weather. It
is caused by the action of density gradients in conjunction with a gravitational field. This is a
brief introduction that will help you understand the qualitative features of a variety of situations
you might encounter.
There are two basic scenarios in the context of natural convection. In one, a density gradient
exists in a fluid in a direction that is parallel to the gravity vector or opposite to it. Such
situations can lead to stable or unstable density stratification of the fluid. In a stable
stratification, less dense fluid is at the top and more dense fluid at the bottom. In the absence of
other effects, convection will be absent, and we can treat the heat transfer problem as one of
conduction. In an unstable stratification, in which less dense fluid is at the bottom, and more
dense fluid at the top, provided the density gradient is sufficiently large, convection will start
spontaneously and significant mixing of the fluid will occur.
Hot
Cold
Fluid
Fluid
Cold
Hot
Stable
Unstable
You should note that density gradients can arise not only from temperature gradients, but also
from composition gradients even in an isothermal system. Here, we restrict our discussion to the
case when temperature gradients are the source of the density gradients.
The more common situation that we encounter in heat transfer is one in which there is a density
gradient perpendicular to the gravity vector. Consider a burning candle. The air next to the hot
candle flame is hot, whereas the air laterally farther from it is relatively cooler. This will set up a
natural convection flow around the candle, in which the cool surrounding air approaches the
surface of the candle, rises, and flows in a hot plume above the flame. It is this flow that causes
the visible flame to take the shape it does; by the way, the flame is simply gas at such a high
1
temperature that it radiates visible light. In the absence of gravity, a candle flame would be
spherical.
Another example is the flow of air at the tip of a lit cigarette; in this case, the smoke from the
cigarette actually traces that flow for us. In a common technique used for home heating, the
baseboard heater consists of a tube through which hot water flows, and the heater is placed close
to the floor. The tube is outfitted with fins to provide additional heat transfer surface. The
neighboring air is heated, and the hot air rises, with cooler air moving in toward the baseboard at
floor level. This natural convection circulation set up by the hot baseboard provides a simple
mixing mechanism for the air in the room and helps us maintain a relatively uniform temperature
everywhere. Clearly, the convection helps the heat transfer process here.
Natural Convection adjacent to a heated vertical surface
Consider a hot vertical surface present in a fluid. The surface is maintained at a temperature Ts ,
which is larger than the ambient temperature in the fluid Te . Here is a sketch of the momentum
boundary layer along the plate.
Ts
Te
As shown in the sketch, the cold fluid rises along the plate surface, becoming heated in the
process, and the momentum boundary layer grows in thickness with distance along the plate. A
sample velocity profile in the momentum boundary layer is shown. Note that in this type of
boundary layer, the velocity must be zero not only at the solid surface, but also at the edge of the
boundary layer. Because the profile was sketched free-hand in PowerPoint, I am unable to show
the smooth approach to zero velocity with a zero slope at the edge of the boundary layer
properly, but that is how the correct velocity profile would appear. Compare this velocity profile
with that in a momentum boundary layer that forms on a flat plate when fluid approaches it with
a uniform velocity U . You should try to make a sketch of the thermal boundary layer on the
same plate when the fluid is air, for example, and also when it is a viscous liquid with a Prandtl
number that is large compared with unity.
Now, let us consider a typical window in a home on a winter day when the outside air is at
10D F and the inside of the room is at a balmy 68D F . What will the momentum boundary
layers on either side of the window look like? Try to sketch them yourself before looking at the
sketch. The arrows in the sketch show the direction of air flow at the location where the air
enters the boundary layer on the inside as well as on the outside, and the direction of air flow
within the boundary layer. There is a slight transverse flow in each boundary layer, but on the
scale of the picture, it is difficult to use the arrows to show it; therefore, I have drawn the flow in
the boundary layers as being vertically downward or upward as appropriate.
What will the thermal boundary layers look like? Try sketching them. Also, you should make a
sketch of the temperature distribution along the interior and exterior surfaces of the window from
the bottom to the top. Will this permit you to explain why ice forms in a certain pattern on the
outside surface of a window on really cold nights?
The Grashof and Rayleigh Numbers
In natural convection situations, an important dimensionless group is the Grashof number. To
provide some physical significance to this group prior to defining it, we use a simple order of
magnitude estimate of the natural convection velocity in the above examples. When fluid with a
1
density moves at a velocity V , the kinetic energy per unit volume can be written as V 2 .
2
This must come from some other form of energy, namely, potential energy lost by the fluid.
Over a vertical distance L , the difference in potential energy between the less dense fluid in the
boundary layer and the more dense fluid outside it can be approximately expressed as g L ,
where g is the magnitude of the acceleration due to gravity, and is a characteristic density
difference between the boundary layer fluid and that far away. We can equate these two order of
magnitude estimates, and neglect the factor of 1/ 2 , because this is only an order of magnitude
analysis.
V 2 g L
Therefore, a typical order of magnitude of the velocity arising from natural convection is
gL
Let us define a Reynolds number for the flowing fluid using this order of magnitude estimate.
g L3
g L3
so that Re 2L =
. This is a dimensionless group that occurs often in natural convection
2
Re L =
LV
Gr =
g L3
1 V
1
1
where V is the specific volume, T
=
=
V T P
T P
T p
temperature of the fluid and P is its pressure. Therefore, we can write
is the
T
= T where we have used a minus sign in relating to T because
T p
both are defined as being positive, and as temperature increases, density decreases.
=
T g L3
Gr =
2
The Grashof number is related to the Reynolds number, and in heat transfer, the Prandtl number
plays a significant role. Therefore, in natural convection heat transfer, we encounter another
dimensionless group, called the Rayleigh number, abbreviated by Ra , which is the product of
the Grashof and Prandtl numbers.
Ra = Gr Pr =
T g L3
Here, is the thermal diffusivity of the fluid. The Nusselt number in natural convection heat
transfer situations is typically a function of the Rayleigh number, the Prandtl number, and aspect
ratio parameters.
For a vertical heated plate of length L , Mills (1) suggests using the following correlation for the
h
L
average Nusselt Number, Nuaverage = average (where k is the thermal conductivity of the fluid).
k
Nuaverage = 0.68 + 0.670 ( Ra )
(1 + 1.6 10
1/ 4
1/ 4
Ra 109
8
Ra )
1/12
0.492 9 /16
= 1 +
Pr
The reason for changing from one correlation to another when the Rayleigh number exceeds 109
is that the natural convection boundary layer undergoes transition to turbulence around that value
of the Rayleigh number. Mills points out that at Ra = 109 the above two correlations do not
coincide in their predictions. This is fine, because that value of the Rayleigh number is an
arbitrary cross-over point from one correlation to the other. It is fine to use the second
(turbulent) correlation for Ra = 109 .
As usual, physical properties should be evaluated at the arithmetic average temperature between
the plate and the ambient fluid.
Other natural convection flows
Mills recommends suitable correlations for natural convection flow over a horizontal heated
cylinder and a heated sphere. For other objects of arbitrary shape, he recommends a correlation
due to Lienhard.
Here the length L to be used in both the Nusselt and Rayleigh numbers is the length of the
boundary layer; for example, L = R for a cylinder or sphere of radius R . But for those two
geometries it is better to use the specific correlations given in the textbook by Mills. Mills also
provides some useful correlations for natural convection in enclosures.
Windows used in homes are termed single-pane or thermopane. A single pane window is a
glass plate that separates the inside of a room from the outside. Heat transfer between the indoor
air and the air outside occurs by conduction through the glass, and the heat transfer rate can be
large. Therefore, the thermopane window was designed to reduce the heat loss by using two
glass plates with a small gap between them. Let us assume the gap is filled with air, and for the
sake of simplicity, that the plates are wide and long, and are each maintained at a uniform
temperature. The sketch given below is taken from a textbook by Bird et al. (2); it depicts the
temperature distribution between the two plates and the resulting natural convection velocity
distribution.
Note that even though there is convection in the air, it does not influence the heat flux through
the air gap, because the temperature distribution still remains linear at this order of
approximation. The air gap significantly increases the thermal resistance of the window and
reduces the heat flux between the outside air and that inside the room. In modern thermopane
windows, the gap between the two plates is evacuated, so that the heat transfer rate is further
reduced, at least initially when the window is new. Over time, air leaks through the seals into the
gap, increasing heat loss in the winter and heat gain in the summer. It is worth noting that as the
gap width is increased, the velocity in the gap increases proportionally to the cube of the gap
width. At larger gap widths, the temperature profile is no longer linear, and the convection
actually increases the heat flux through the gap over that occurring due to pure conduction. This
is the reason for the choice of a small width of the order of 1-2 mm for the air gap in thermopane
windows.
If you wish to learn more about natural convection heat transfer, a good reference is the book by
Gebhart et al. (2).
References
Energy Transport
R. Shankar Subramanian
Transport of thermal energy in fluids occurs by three mechanisms.
Conduction or molecular transport
Convection or bulk transport
Radiation
Of these, the proper treatment of radiation is beyond the scope of our course. Therefore,
we shall only consider conduction and convection.
Conservation of Energy
The energy of a flowing fluid consists of internal and kinetic energy. The rate of increase
of the energy content of the fluid present within a control volume at a given instant is
equal to the sum of the net flux of energy into the control volume and the work done on
the fluid within the control volume by body forces and surface forces acting on it. Bird et
al. go through a careful derivation of the mathematical form of the equation of
conservation of energy in Chapter 11. The initial balance is written for the total energy
and then the part involving the kinetic energy (known as the mechanical energy balance,
obtained by taking a dot product of Cauchys equation with the velocity) is subtracted.
After using some thermodynamic relationships, the final form of the equation of
conservation of thermal energy is obtained. Various versions are given in Table 11.4-1 of
the book. The most common version that we shall use assumes that the density and
thermal conductivity k are constant, and is given below.
T
+ v T = k 2T + v + S
t
C p
In the above equation, T is the temperature, C p is the specific heat at constant pressure,
t is time, v is the (vector) velocity, v stands for the rate of irreversible conversion of
mechanical energy into internal energy per unit volume by viscous dissipation, and S
represents the rate of generation of energy per unit volume by sources such as electrical
heating. When the thermal energy equation is obtained for a multicomponent system, the
rate of generation (or consumption) of energy per unit volume due to chemical reactions
appears naturally in the energy equation when proper accounting is made of the
enthalpies of the various species. Detailed expressions for the dissipation function v in
terms of derivatives of the velocity components in common coordinate systems can be
found in Table B.7 of Bird et al. In most situations, we can set the viscous dissipation
term to zero with negligible error, the exceptions occurring when highly viscous fluids
are subjected to large velocity gradients; an example where viscous heating is important
is polymer processing.
There are two fluxes of thermal energy that appear in the energy equation (the prefix
thermal will be omitted, but implied from hereon). One is molecular flux of energy
and the other is convective flux of energy.
is known as Fouriers law. Here, q is the (vector) heat flux, and T is the temperature
field at a given point. The negative sign tells us that heat flows in the direction opposite
to that of the temperature gradient, namely from hot regions to cold regions. The
constant of proportionality in Fouriers law, k is known as the thermal conductivity and
is a material property of the fluid. The thermal conductivity depends on temperature and
pressure in general, and Bird et al. provide some information regarding this subject in
Chapter 9. Table B.2 provides results for the components of q in common coordinate
systems.
where Tref is a reference temperature that serves as a datum. A more precise accounting
of the convective flux of total energy, is given in Bird et al. in Section 9.7.
Boundary Conditions
At the interface between a solid and a fluid, or that between a fluid and another fluid, it is
reasonable to expect thermodynamic equilibrium to prevail between the two phases
adjoining the interface, except when the heat flux across the interface is extremely large.
A small portion of the interface between phases I and II is shown schematically in the
sketch.
II
I
t
TI = TII
at the interface. In addition, because the interface is assumed to have no mass, any heat
flux crossing the interface from one phase must necessarily be transmitted to the other
phase. Therefore,
n qI = n qII
at the interface. The most general way to formulate a problem is to write the governing
energy equation for each phase and the above pair of conditions at each phase interface,
along with any other applicable initial and boundary conditions. In practice, it is far more
convenient to study heat transport in controlled conditions wherein the temperature at a
solid boundary in contact with a fluid is prescribed, or the heat flux from the solid to the
fluid is prescribed. Experimentally, we can achieve a condition of prescribed uniform
wall temperature in a pipe by choosing a highly conducting wall material and
surrounding it with either a phase change system (such as condensing steam) or a
segmented electrical heating system with a controller that maintains the temperature of
the wall at a constant value. A uniform heat flux can be achieved most conveniently by
using electrical heating. These two boundary conditions represent the two extremes of a
family of boundary conditions that are common in heat transport problems.
U , T
y
x
Tw
Just as a momentum boundary layer forms at the wall and grows in thickness with
distance x along the plate, a thermal boundary layer forms at the wall; the temperature of
the fluid changes from Tw to T in the thermal boundary layer. The thickness of the
thermal boundary layer t also grows with distance x .
At the rigid wall, the normal velocity is zero and the heat flux from the wall to the fluid
consists only of the conduction flux q y . From Fouriers law, this can be written as
qw ( x ) = k
T
( x, 0 )
y
Note that qw > 0 when Tw > T , and qw < 0 when Tw < T . Even though qw is purely a
conduction flux, it is modified by flow. As the velocity of the fluid increases, the ability
of the fluid to carry away heat supplied by the plate increases, and the temperature
gradient at the wall becomes sharper, consistent with a larger heat flux. You can see that
a sharper temperature gradient at the wall is consistent with a thinner thermal boundary
layer.
If we can solve the energy equation for the temperature distribution in the flowing fluid,
the heat flux qw ( x ) can be evaluated as a function of x . A heat transfer coefficient
h ( x ) is defined for this system as follows.
qw ( x ) = h ( x )(Tw T )
Therefore, the heat transfer coefficient is seen to be directly related to the temperature
gradient at the wall.
T
( x, 0 )
y
h ( x) =
(Tw T )
k
hL
k
Here L is a characteristic length scale, which can be taken as the axial distance x or the
length of the plate, depending on our needs.
Now, consider heat transfer to a fluid flowing through a pipe. In this case, let the fluid
enter the pipe at some temperature T0 , encountering a step change in wall temperature to
Fluid at
R
z
T0
T ( R, z ) = Tw
At any given axial position z , the heat flux from the wall to the fluid is given by
qw = k
T
( R, z )
r
Do you see why a positive sign is used in the right side? It is because the heat flux from
the wall to the fluid is in the negative r direction.
For defining the heat transfer coefficient, we need a driving force at any location z .
While it is possible to use (Tw T0 ) , the more common choice is (Tw Tb ) where Tb is
known as the bulk or cup-mixing average temperature. The bulk average temperature is
experimentally determined by collecting the fluid coming out of the system at a given
axial location and mixing it completely, and then measuring its temperature. The
following mathematical definition directly follows from this physical definition.
1
Tb =
2 rV (r )T ( r , z ) dr
0
2 rV (r ) dr
0
heat flux
driving force
We can think of the heat transfer coefficient as the conductance (using an electrical
analogy) of the system. Typically, spatial averages are easier to measure and report; it is
not common in engineering design to use local values.
In problems amenable to analysis from first principles, temperature distributions and heat
fluxes can be calculated directly from the solution. There is really no need to define a
heat transfer coefficient. But, to make it convenient to report and use the results, even in
such problems, suitably defined heat transfer coefficients are calculated from the
theoretical results and reported.
Important dimensionless groups in heat transfer
Nu = Nu Re, Pr,
D
hD
UD
where Nu =
The symbol L stands for the length of the pipe. For non-circular cross-sections, the
Nusselt number will also depend on additional aspect ratio parameters.
The physical significance of the Nusselt number is simply that it represents a
dimensionless heat flux at the wall or a ratio of the actual heat transfer rate to that
prevailing in a hypothetical system in which the same driving force applied across the
characteristic distance drives a conduction flux. We can see this by recasting it as
follows.
Nu =
h T
T
k
D
To appreciate the physical significance of the Prandtl number, we note that it is the ratio
of the intrinsic transport coefficients and for molecular transport of momentum and
energy, respectively. Consider a simple one-dimensional transport situation in which
fluid flows in the x direction with a velocity vx ( y ) . In this case, Newtons law of
viscosity for the momentum flux yx (note that we are interpreting this symbol as the
negative of the shear stress) can be written as
yx =
vx
y
and in a similar one-dimensional conduction problem, Fouriers law for the heat flux
q y can be written as
qy = k
T
y
If we assume that the density and specific heat at constant pressure are constant, we can
rewrite the above results in the following form.
yx =
( vx )
y
qy =
C p {T Tref }
The product vx represents the amount of x momentum in unit volume of fluid, and
can be regarded as the concentration of x momentum. Therefore, we see that the flux
of x momentum is proportional to the gradient of the concentration of that momentum,
and occurs in the direction opposite to that gradient. The coefficient of proportionality is
the kinematic viscosity . In a like manner, the flux of thermal energy is proportional to
the gradient of the concentration of thermal energy, with a coefficient of proportionality
equal to the thermal diffusivity . So, molecular transport leads to momentum and
energy flowing downhill in the direction opposite to that of the concentration gradient
of each, with coefficients of proportionality that represent the ability of the fluid to
transport momentum or energy by molecular means. This leads to the following physical
interpretation of the Prandtl number.
Pr =
Thus, in flow over an object, the relative thicknesses of the momentum and thermal
boundary layers reflect the magnitude of the Prandtl number, as the examples given
below show.
Large Prandtl number, Pr 1 ( )
U , T
t
x
Tw
U , T
y
m
x
Tw
For gases, the Prandtl number is typically of O (1) , while for common liquids Prandtl
numbers are found to vary from 10 to 1000. The Prandtl numbers of very viscous liquids
such as polymer melts can be even larger than 1000, reaching values of 105 or greater.
Because liquid metals are great conductors of thermal energy, Prandtl numbers for liquid
metals are typically of O (102 ) .
Packed Beds
A typical packed bed is a cylindrical column that is filled with a suitable packing material. You
can learn about different types of packing materials from Perrys Handbook. The liquid is
distributed as uniformly as possible at the top of the column and flows downward, wetting the
packing material. A gas is admitted at the bottom, and flows upward, contacting the liquid in a
countercurrent fashion. An example of a packed bed is an absorber. Here, the gas contains some
carrier species that is insoluble in the liquid (such as air) and a soluble species such as carbon
dioxide or ammonia. The soluble species is absorbed in the liquid, and the lean gas leaves the
column at the top. The liquid rich in the soluble species is taken out at the bottom.
From a fluid mechanical perspective, the most important issue is that of the pressure drop
required for the liquid or the gas to flow through the column at a specified flow rate. To calculate
this quantity we rely on a friction factor correlation attributed to Ergun. Other fluid mechanical
issues involve the proper distribution of the liquid across the cross-section, and developing
models of the velocity profile in the liquid film around a piece of packing material so that
heat/mass transfer calculations can be made. Design of packing materials to achieve uniform
distribution of the fluid across the cross-section throughout the column is an important subject as
well. Here, we only focus on the pressure drop issue.
The Ergun equation that is commonly employed is given below.
fp =
150
+ 1.75
Re p
Here, the friction factor f p for the packed bed, and the Reynolds number Re p , are defined as
follows.
fp =
p D p 3
L Vs2 1
Re p =
D pVs
(1 )
The various symbols appearing in the above equations are defined as follows.
p : Pressure Drop
L : Length of the Bed
D p : Equivalent spherical diameter of the particle defined by D p = 6
Q
where Q is the volumetric flow rate of the fluid and A is the
A
: Void fraction of the bed ( is the ratio of the void volume to the total volume of the bed)
Sometimes, we may use the concept of the interstitial velocity Vi , which is related to the
V
superficial velocity by Vi = s . The interstitial velocity is the average velocity that prevails in
150
,
Re p
Re p 1
The other is the Burke-Plummer equation, used when viscous effects are not as important as
inertia.
f p = 1.75 ,
Re p 1, 000
It is suggested that the student simply use the Ergun equation. There is no need to use these
other two approximate results, even though they continue to be reported in textbooks.
Fluidized Beds
A fluidized bed is a packed bed through which fluid flows at such a high velocity that the bed is
loosened and the particle-fluid mixture behaves as though it is a fluid. Thus, when a bed of
particles is fluidized, the entire bed can be transported like a fluid, if desired. Both gas and liquid
flows can be used to fluidize a bed of particles. The most common reason for fluidizing a bed is
to obtain vigorous agitation of the solids in contact with the fluid, leading to excellent contact of
the solid and the fluid and the solid and the wall. This means that nearly uniform temperatures
can be maintained even in highly exothermic reaction situations where the particles are used to
catalyze a reaction in the species contained in the fluid. In fact, fluidized beds were used in
catalytic cracking in the petroleum industry in the past. The catalyst is suspended in the fluid by
fluidizing a bed of catalytic particles so that intimate contact can be achieved between the
particles and the fluid. Nowadays, you will find fluidized beds used in catalyst regeneration,
solid-gas reactors, combustion of coal, roasting of ores, drying, and gas adsorption operations.
Pressure
Drop
B
E
Superficial velocity
D
E
Bed
Height
Superficial Velocity
First, we consider the behavior of a bed of particles when the upward superficial fluid velocity is
gradually increased from zero past the point of fluidization, and back down to zero. Reference is
made to the figure on page 3.
At first, when there is no flow, the pressure drop zero, and the bed has a certain height. As we
proceed along the right arrow in the direction of increasing superficial velocity, tracing the path
ABCD, at first, the pressure drop gradually increases while the bed height remains fixed. This is
a region where the Ergun equation for a packed bed can be used to relate the pressure drop to the
velocity. When the point B is reached, the bed starts expanding in height while the pressure drop
levels off and no longer increases as the superficial velocity is increased. This is when the
upward force exerted by the fluid on the particles is sufficient to balance the net weight of the
bed and the particles begin to separate from each other and float in the fluid. As the velocity is
increased further, the bed continues to expand in height, but the pressure drop stays constant. It
is possible to reach large superficial velocities without having the particles carried out with the
fluid at the exit. This is because the settling velocities of the particles are typically much larger
than the largest superficial velocities used.
Now, if we trace our path backward, gradually decreasing the superficial velocity, in the
direction of the reverse arrows in the figure, we find that the behavior of the bed follows the
curves DCE. At first, the pressure drop stays fixed while the bed settles back down, and then
begins to decrease when the point C is reached. The bed height no longer decreases while the
pressure drop follows the curve CEO. A bed of particles, left alone for a sufficient length of time,
becomes consolidated, but it is loosened when it is fluidized. After fluidization, it settles back
into a more loosely packed state; this is why the constant bed height on the return loop is larger
than the bed height in the initial state. If we now repeat the experiment by increasing the
superficial velocity from zero, well follow the set of curves ECD in both directions. Because of
this reason, we define the velocity at the point C in the figure as the minimum fluidization
velocity V f . We can calculate it by balancing the net weight of the bed against the upward force
exerted on the bed, namely the pressure drop across the bed p multiplied by the cross-sectional
area of the bed A . In doing this balance, we ignore the small frictional force exerted on the wall
of the column by the flowing fluid.
Upward force on the bed = p A
If the height of the bed at this point is L and the void fraction is , we can write
Volume of particles = (1 ) AL
If the acceleration due to gravity is g , the net gravitational force on the particles (net weight) is
Net Weight of the particles = (1 ) ( p f ) ALg
f ) gD p2 3
150
1
When the superficial velocity Vs is equal to V f , we refer to the state of the bed as one of
incipient fluidization. The void fraction at this state depends upon the material, shape, and
size of the particles. For nearly spherical particles, McCabe, Smith, and Harriott (2001) suggest
that lies in the range 0.40 0.45 , increasing a bit with particle size.
For large particles ( D p 1 mm), inertial effects are important, and the full Ergun equation must
be used to determine V f . When in doubt, use the Ergun equation instead of a simplified version
of it.
Now, we consider the condition we must impose on the superficial velocity so that particles are
not carried out with the fluid at the exit. This would occur if the superficial velocity is equal to
the settling velocity of the particles. Restricting attention to small particles so that Stokes law
can be used to calculate their settling velocity, we can write
p f ) gD p2
(
Vsettling =
18
If we now use the result for the minimum fluidization velocity for the case of small particles,
given above, we see that the ratio
Vsettling 25 1
=
3 3
Vf
For lying in the range 0.40 0.45 , this yields a ratio ranging from 78 50 . McCabe et al.
suggest that it is common to operate fluidized beds at velocities as high as 30 V f , and values as
large as 100 V f are used on occasion. Recognizing that not all particles are of the same size and
that D p is only an average size, we see that fine particles are likely to be carried out with the
5
exiting fluid in such a situation. They can be recovered by filters or cyclone separators and
returned, in order to obtain the benefits of operating a bed at such large superficial velocities.
Fluidization can be broadly classified into particulate fluidization or bubbling fluidization.
Particulate fluidization occurs in liquids. As the velocity of the liquid is increased past the
minimum fluidization velocity, the bed expands uniformly, and uniform conditions prevail in the
liquid solid mixture. In contrast, bubbling fluidization occurs in gas-fluidized beds. Here, when
the bed is fluidized, large pockets of gas, free of particles, are seen to rise through the bed.
Where there are particles, the bed void fraction is approximately at the value that prevails at the
point of incipient fluidization. The bubbles grow until they fill the cross-section, and then
successive bubbles move up the column, a condition known as slugging.
The above classification should not be interpreted rigidly. Sometimes, very dense particles in a
liquid can show bubbling and gases at high pressure when flowing through beds of fine
particles, can give rise to particulate fluidization. Usually, this occurs at lower velocities, and at
higher velocities, the bed shows bubbling.
References
W.E. McCabe, J.C. Smith, and P.Harriott 2001. Unit Operations of Chemical Engineering,
McGraw Hill, New York.
Perrys Chemical Engineers Handbook, 7th Edition. 1997 (Ed: R.H. Perry, D.W. Green, and J.O.
Maloney), McGraw-Hill, New York.
Q
Q
We can see that if the velocity were to be the same at both ends of the element, it would
change neither its length, nor its orientation. Therefore, in a uniform velocity field, there
is simple translation of fluid elements with no deformation or rotation. To cause either,
the velocity v ( x ) must be non-uniform. To understand the nature of the changes in fluid
elements brought about by the flow, we must, therefore, investigate the velocity gradient,
v , which is a second order tensor.
From calculus, we know that the differential change dvx can be written as
dvx =
vx
v
v
dx + x dy + x dz
x
y
z
and similar results can be written for the changes dv y and dvz . It follows that the
differential change in the vector velocity, dv , is given by
dv = i dv x + j dv x + k dvz
v
v
v
v
v
v
v
v
v
= i x dx + x dy + x dz + j y dx + y dy + y dz + k z dx + z dy + z dz
y
z x
y
z
y
z
x
x
v
v
v
v
v
v
v
v
v
= i x + j y + k z dx + i x + j y + k z dy + i x + j y + k z dz
x
x
y
y
z
z
x
y
z
=
v
v
v
dx +
dy + dz = v dx
x
y
z
Thus, the relative velocity of a point a distance dx from any given location is given by
the dot product of the tensor v and the differential line element dx . This tensor can be
written as follows.
vx
x
v
v = y
x
vz
vx
y
v y
y
vz
y
vx
z
v y
z
vz
Any tensor can be written as the sum of a symmetric and an antisymmetric tensor. Let us
do this with the velocity gradient tensor, writing it as
v = E +
where the (symmetric) rate of strain or rate of deformation tensor E is given by
E=
1
v + v T )
(
2
1
v - v T )
(
2
The action of each of these contributions to the velocity gradient will be explored in
detail next. First, we consider the vorticity tensor.
Vorticity Tensor
The vorticity tensor is a skew-symmetric tensor. We can write its components in
terms of the components of the velocity gradient as follows.
v v
1
= y x
2 x y
1 vz vx
2 x z
1 vx v y
2 y x
0
1 vz v y
2 y z
1 vx vz
2 z x
1 v y vz
2 z y
A skew-symmetric tensor Aij can be formed from a vector ak by writing Aij = ijk ak . The
1
vector associated with the vorticity tensor in this manner is , where = v is
2
1
known as the vorticity vector. Using the relationship between and , we obtain
2
1
1
1
ij dx j = ijk dx j k or in Gibbs notation, dx = dx = dx
2
2
2
This means that the relative motion that is contributed by the vorticity tensor at a point an
infinitesimal distance away from a reference point in a fluid is that caused by a rigid
1
rotation with an angular velocity equal to .
2
Because a fluid does not usually rotate as a rigid body in the manner that a solid does, we
should interpret the above statement as implying that the average angular velocity of a
fluid element located at a point is one-half the vorticity vector at that point. To prove this
claim, consider a surface formed by an infinitesimal circle of radius a located at a point
x . Let the unit normal vector to the surface (perpendicular to the plane of the paper) be
n , and the unit tangent vector to the circle at any point be t .
a
Apply Stokess theorem to the velocity field in this circle.
( v ) n dS = v v t ds
S
Here, dS is an area element on the surface of the circle S and ds is a line element along
the circle C. Because v t is the component of the velocity along the periphery of the
1
circle, we can write the average linear velocity along the circle as
v t ds and
2 a v
C
therefore the average angular velocity as
1
2 a 2
v v t ds .
that this is equal to the average value of v n over the surface of the circle.
2
Thus, in the limit as the radius of the circle approaches zero, we find that the average
angular velocity around the circle approaches the value of one-half the component of the
vorticity vector in a direction perpendicular to the surface of the circle. We also can
show (see Batchelor, page 82) that the angular momentum of a spherical element of fluid
is equal to one-half the vorticity times the moment of inertia of the fluid, just as it is for a
rigid body.
Vorticity Vector
The vorticity = v , is an important entity in fluid mechanics. It is transported from
one place to another in a fluid by convective and molecular means, just as energy and
species are, and an appropriate partial differential equation that governs its transport can
be written. In addition, vorticity also is intensified by the stretching of vortex lines, a
mechanism that is not present in the transport of energy and species. One reason for
working with the equations of vorticity transport is that pressure is absent as a dependent
variable in those equations. It can be shown that if a fluid mass begins with zero vorticity,
and the fluid is inviscid (meaning the viscosity is zero), the vorticity will remain zero in
that fluid mass. A flow in which the vorticity is zero is known as an irrotational flow.
Vorticity is generated at fluid-solid interfaces and at fluid-fluid interfaces. Vorticity
cannot be generated internally within an incompressible fluid. This is the reason why, in
a high Reynolds number flow (implying weak viscous effects) past a rigid body, most of
the flow can be described by using the equations that apply to irrotational flow, with the
vorticity being confined to a boundary layer near the surface of the body.
Vortex Lines and Tubes
Just as a streamline is a curve to which the velocity vector is tangent everywhere, we can
define a vortex line as a curve to which the vorticity is tangent everywhere. If the
components of the vorticity = v are ( x , y , z ) , then we can write the equations
of the space curves that are vortex lines as
dx
dy
dz
The surface that is formed by all the vortex lines passing through a closed reducible curve
is known as a vortex tube. If we construct an open surface S bounded by this closed
curve C, we can define the strength of the vortex tube as
dS .
By using Stokess
v v t ds
around the vortex tube, t is a unit tangent vector to the curve at any point, and ds is a
line element.
vx
v v
1
= y + x
2 x y
1 vz + vx
2 x z
1 vx v y
+
2 y x
v y
y
1 vz v y
+
2 y z
1 v y vz
+
2 z y
vz
1 vx vz
+
2 z x
2 ds
dx
= dv , the above result can be rewritten as
dt
d
( ds ) = 2 dx dv = 2 dx ( v dx ) = 2 dx E dx + 2 dx dx
dt
In the above result, after writing the result in index notation, we first exchange the indices
i and j to obtain an intermediate result, and then use the antisymmetry property to write
ij = ji .
Therefore, we find that
ds
d
( ds ) = dx E dx from which, by dividing through by ds 2 we can write
dt
1 d
dx
dx
( ds ) = E
ds dt
ds
ds
The vector dx / ds is a unit vector pointing in the direction of the infinitesimal vector
dx . Therefore, we can think of the right side of the above result as the double
projection of the tensor E in that direction. The term projection is used in a loose
sense here. The physical meaning is clear. The rate of strain of a line element pointing in
any direction at a given point (which is the time rate of change of length, divided by the
length) is the dot product of a unit vector in that direction with the dot product of the rate
of strain tensor with the same unit vector. Let us choose the direction to be the
x direction. In this case, the rate of strain of a line element in that direction is simply
v
E11 , which is equal to x . In a like manner, the rate of strain of a line element in the
x
v y
vz
y direction is
, and that in the z direction is
. This is the physical
y
z
interpretation of the diagonal elements of the rate of strain tensor.
The sum of the diagonal elements of E , known as the trace of E is v . This is known
as the rate of dilatation of a fluid element at the given location. To see why, consider a
material body occupying a volume V enclosed by the surface S . Let us inquire how V
changes with time. We can write the rate of change of the volume of a material body
with time as the integral of dS v over the surface.
dV
= dS v = v dV by the divergence theorem.
dt S
V
1 dV
1
= lim v dV = v . So, the trace of
V
0
V dt
VV
E is the rate of increase in the volume of an infinitesimal element, divided by its
volume, and is called the rate of dilatation. When the flow is incompressible, the rate of
dilatation is zero.
Now, consider two line elements dx and dx at a given point x and let the angle
between them be .
dx
dx
Let us investigate the time rate of change of the dot product of the vectors dx and dx .
d
d
( ds ds cos ) = ( dx dx) = dvi dxi + dxi dvi
dt
dt
v
v
= i dx j dxi + i dx jdxi
x j
x j
In writing the result in the second term in the second line, we have used the fact that the
infinitesimal change dvi is the change in the velocity over an infinitesimal distance in the
direction of the vector dx . Interchanging the indices i and j in that second term
permits us to combine the two terms.
d
( ds ds cos ) =
dt
v v j
i +
dxidx j = 2 Eij dxidx j
x
x
i
j
dx dx j
dx
1 d
dx
=2
E
ds ds cos ) = 2 Eij i
(
ds ds dt
ds ds
ds
ds
So, we see that if we take the dot product of E with two different directions in
succession (the order is immaterial because E is symmetric), the result is the left side of
the above equation. Let us work out the differentiation in the left side.
1 d
1 d
d
1 d
ds ds cos ) = cos
ds ) +
ds ) sin
(
(
(
ds ds dt
ds dt
dt
ds dt
The term in square brackets in the right side is the sum of the individual rates of strain of
the two line elements. We can see that the above result reduces to the earlier result we
obtained when the two vectors dxi and dxi are the same. Let us consider the case when
the two vectors are orthogonal to each other. In this case, we obtain
7
dx
dx
1 d
E
=
2 dt
ds
ds
So, the sequential dot products of E with unit vectors in two orthogonal directions yields
one-half the rate of decrease of the angle between those directions. If we choose these
two orthogonal directions to coincide with any two coordinate directions, then the dot
products yield the off-diagonal elements of E . For example, if we use x and y
directions, the element is E12 ( = E21 ) . Similar physical interpretations can be given to the
other off-diagonal elements of the rate of strain tensor. Thus, the off-diagonal elements
describe shear deformation of the fluid.
There are three mutually orthogonal directions associated with the symmetric tensor E
that are known as its eigenvector or principal directions. We can use a basis set built
from these principal directions to describe the components of the tensor. If we do, the
tensor will be diagonal. The off-diagonal elements will be zero, so that the rate of change
of the angles between the principal directions is zero; of course the entire set of principal
1
1
axes can rotate, and in fact it does, with the angular velocity = v .
2
2
Instantaneous Deformation of a Fluid Element
Based on all of the above material on kinematics, we can conclude that in a flow, an
infinitesimal spherical element of fluid undergoes translation, rotation, and deformation
in general. It deforms into an ellipsoid whose axes are aligned with the principal axes of
the rate of strain tensor. This ellipsoid also rotates with an instantaneous angular velocity
that is equal to the one-half of the vorticity of the fluid at the given point.
Some good sources for further study are listed below.
References
1. R. Aris, Vectors, Tensors, and the Basic Equations of Fluid Mechanics, Dover
Publications, 1989, Chapter 4 (original by Prentice-Hall, 1962).
2. G.K. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press,
1967, Chapter 5.
3. C. Truesdell, Kinematics of Vorticity, Indiana University Press, 1954.
4. J. Serrin, Mathematical Principles of Classical Fluid Dynamics, Handbuch der
Physik VIII/1, Ed. S. Flugge, 1959.
5. P.G. Saffman, Vortex Dynamics, Cambridge University Press, 1992.
v t =V t
on a rigid surface
(no slip)
The equality of the normal components of the velocity at the boundary arises from purely
kinematical considerations when there is no mass transfer across the boundary. If n represents
the unit normal,
v n =V n
on a rigid surface
(kinematic condition)
As a consequence of the two conditions, we arrive at the conclusion that the fluid velocity must
match the velocity of the rigid surface at every point on it.
v =V
on a rigid surface
The no-slip condition has been found to be inapplicable in special circumstances such as at a
moving contact line when a drop spreads over a solid surface, or in flow of a rarefied gas through
1
a pore of diameter of the same order of magnitude as the mean free path of the gas molecules.
For the types of problems that we shall encounter, it is an adequate boundary condition.
n
I
t
II
The unit normal vector n points into phase
vector to the interface t .
It so happens that the velocity fields in phases I and II are continuous across the interface.
This vector condition also can be viewed as being in two parts, one on the continuity of the
tangential component of the two velocities, analogous to the no-slip boundary condition at a rigid
boundary, and the continuity of the normal component of the two velocities, a kinematic
consequence when there is no mass transfer across the interface.
Therefore, we can write
v I t = v II t
at a fluid-fluid interface
(no slip)
v I n = v II n
at a fluid-fluid interface
(kinematic condition)
and
Notice that we have two unknown vector fields v I and v II now, and therefore need twice as
many boundary conditions. Therefore, it is not sufficient to write just the above no-slip and
kinematic conditions at a fluid-fluid interface. We also need to write a boundary condition
connecting the state of stress in each fluid at the interface. The general form of this condition is
given below.
fluid, H is the mean curvature of the interface at the point where the condition is being applied,
is the interfacial tension of the fluid-fluid interface, and s is the surface gradient operator
which can be written as
is normal to the surface. The left side in the stress boundary condition is the difference between
the stress vectors in fluids I and II at the interface, or the jump in stress. This is the reason
for the choice of terminology used in describing this condition. The resulting vector is
decomposed into a part that is normal to the interface, namely the first term in the right side, and
a part that is tangential to the interface, given in the second term in the right side. Sometimes,
the condition is written as two separate scalar boundary conditions by writing the tangential and
the normal parts separately. In that case, we call the two boundary conditions the tangential
stress balance and the normal stress balance.
In the types of problems that we shall encounter, the stress boundary condition can be simplified.
The interfacial tension at a fluid-fluid interface depends on the temperature and the composition
of the interface. If we assume these to be uniform, then the gradient of interfacial tension will
vanish everywhere on the interface. This means that the tangential stress is continuous across the
interface because the jump in it is zero. Recall that the tangential stress is purely viscous in
origin. If
t = t
I
at a fluid-fluid interface
II
The normal stress jump boundary condition actually determines the curvature of the interface at
the point in question, and therefore the shape of the entire fluid-fluid interface. This shape is
distorted by the flow. In the problems that we shall analyze, we shall always assume the shape
of the interface to be the static shape and as being specified. Therefore, we shall not be able to
satisfy the balance of normal stress. In fact, fluid mechanical problems involving the application
of the normal stress balance at a boundary are complicated, and must be solved numerically
unless one assumes the shape distortion to be very small or of a particularly simple form.
At a liquid-gas interface, we can further simplify the tangential stress balance. Consider the
surface of a liquid film flowing down an inclined plane. Let us assume that the flow is steady
and that the film surface is parallel to the inclined plane. In this situation, the normal velocity at
the free surface of the liquid is zero in both the liquid and the gas. The sketch depicts the
situation.
gas
x
liquid
3
Because the normal velocity is zero at the free surface, the tangential stress balance simplifies to
the following result where the subscripts l and g represent the liquid and gas, respectively.
vx ,l
y
= g
vx , g
y
The symbol in the above result stands for the dynamic viscosity. If we divide through by the
dynamic viscosity of the liquid, we obtain
vx ,l
y
g vx , g
l y
Because the dynamic viscosity of a gas is small compared with that of a liquid, the right side of
the above equation is small, and can be considered negligible. This allows us to write
vx ,l
y
Sometimes, this condition is represented as that of vanishing shear stress at a free liquid surface.
Note that this approximation of the tangential stress condition can be used only when the
motivating force for the motion of the liquid is not the motion of the gas. When a gas drags a
liquid along, as is the case on a windy day when the wind causes motion in a puddle of liquid,
the correct boundary condition equating the tangential stresses must be used.
U
y
x
laminar
turbulent
Re x 5 105 , where Re x =
xU
Here, is the
u
2u
coordinates is u
, and a typical viscous term is 2 . Here, ( u , v ) are
x
y
the velocity components in the ( x, y ) directions, and and are the
density and the dynamic viscosity of the fluid. We can estimate the order of
magnitude of each of these terms for a plate of length L as follows.
U 2
u
u
x
L
U
2u
2 2
y
U 2
U
L
2 or 2
, which can be recast as
L
1
Re L
LU
u
( x,0 ) . We can estimate the order of
y
x
=
Re x
can estimate it as ( x )
D = w w ( x ) dx = w
0
U 3/ 2
1/ 2
dx
.
x
D w U 3 L
A rigorous calculation from boundary layer theory yields the result
D = 0.664 w U 3 L
confirming the correctness of our scaling argument.
The Displacement Thickness
The displacement thickness of the boundary layer is defined as the distance
by which the potential flow streamlines are displaced by the presence of the
boundary layer. We can construct a mathematical definition in the case of
the flat plate by recognizing that the displacement thickness 1 is that
thickness of the uniform stream that accounts for the lost flow because of
the presence of the solid surface.
U 1 = (U u ) dy
0
or
1 = 1
0
u
U
dy
u
v
+
=0
x
y
From the scaling, we know that u is O (1) . This means that the magnitude
of u lies between 0 and a number that is of the order of unity. In this
particular case, because the maximum value of the physical velocity is that
of the uniform stream approaching the plate, namely U , the maximum
value of u is, in fact, precisely, 1. But this is not necessarily the meaning
implied by the order symbol that we are using. Note that the order of
magnitude of a quantity is the same regardless of its sign.
Because the velocity u varies in the range mentioned above, while the
scaled variable x also varies from 0 to 1 (we say x O (1) ), we can
u
is O (1) as well. From the continuity
x
u
v
and
must sum to zero; this forces the
x
y
v
to be O (1) . We know that the variable y O ( ) where
y
represents the boundary layer thickness divided by the length L . In other
derivative
v
O (1) , we must conclude that the change in the scaled velocity
y
component v across the boundary layer must be of O ( ) . We know from
the kinematic condition that v = 0 at the surface y = 0 . Therefore, the
magnitude of v must of O ( ) . We note that is a very small quantity
when the Reynolds number Re L 1. We express this fact by stating
1. Therefore, the scaled velocity in the y direction in the boundary
is a very small quantity.
Now, following Schlichting [1] we proceed to use similar arguments in the
two components of the Navier-Stokes equation applicable to this situation.
First, consider the x component.
u
u
u
+ v
x
y
1 1
2u
p
1 2u
+
+
Re L x 2
y 2
x
1
2 1
2
Below each term in the equation, we have written the order of magnitude of
that term. We already have discussed the order of magnitude of u , v , and
u
u
. To estimate the order of magnitude of
, we first note that u varies
y
x
from 0 to 1 across the boundary layer, while the variable y varies from 0
u
1
O . To estimate the
to . This is the reason for the estimate that
y
order of magnitude of the second derivatives, we must use similar
2u
arguments. For example, consider the derivative
. We know that
x 2
u
O (1) . So, this quantity must change from 0 to a magnitude of the
x
order of unity in a scaled distance x that also changes from 0 to 1. This is
2u
the reason for estimating the order of
as being unity. In a like manner,
x 2
the derivative
u
1
1
O , which means that it varies from 0 to
across
y
2u
1
derivative
O 2 . The order of magnitude of the Reynolds number
2
y
was established earlier on page 2.
Comparing the two viscous terms, we see that the viscous force in the
x direction is negligible when compared to that in the y -direction. We
need to retain all the other terms in the x component momentum equation
because they are all of comparable order of magnitude.
Now, let us consider the y component of the Navier-Stokes equation.
v
v
+ v
u
x
y
1
2v
p
1 2v
+
+
Re L x 2
y 2
y
1
2
The order of magnitude of the derivatives has been estimated in the same
manner as outlined earlier. Once again, we see that the viscous transport of
y momentum in the x direction is much weaker than that in the
y direction, and can be neglected. The most important aspect of the above
equation is that all the retained terms are of O ( ) , so that the pressure
p
must necessarily be of the same order (or smaller). Because the
y
variation of pressure in the y direction in the boundary layer must occur
over a distance of O ( ) , it is evident that the scaled pressure change across
gradient
( )
that
p
x
in the
u
v
+
=0
x
y
Navier-Stokes Equation
u
u
u
+ v
x
y
1 p
2u
+ 2
x
y
For the flat plate problem, the potential flow is simply u = U . This means
that the potential flow pressure gradient is zero. Therefore, the NavierStokes equation simplifies to
u
u
u
+ v
x
y
2u
= 2
y
u ( 0, y ) = U
u ( x,0 ) = 0
v ( x,0 ) = 0
u ( x, y ) U
Energy Form
Here is the energy form of the Engineering Bernoulli Equation. Each term
has dimensions of energy per unit mass.
Vout 2
pin Vin 2
+
+ gzout =
+
+ gzin loss ws
2
2
pout
Head Form
The head form of the Engineering Bernoulli Equation is obtained by
dividing the energy form throughout by g .
Vout 2
pin Vin 2
loss ws
+
+ zout =
+
+ zin
2g
2g
g
g
pout
hp =
ws
.
g
Note that the work is negative which implies work is done on the
fluid by the pump.
loss
= hfriction = h f : loss expressed in head of fluid.
g
( 1)
1
( n+1) =
,
1 3 5 ( 2n 1) n times r
n
n = 0,1, 2,...
xx x
x + x + x
1
x xx
, 3i , i 5 j ij3 , i 7j k i jk j 5ki k ij
r
r
r
3r
r
5r
hand, we are familiar with vectors and tensors that change sign if the basis
set is changed from one that is right-handed to one that is left-handed. One
example is the cross product of two absolute vectors. Another way to obtain
a pseudovector is by taking the curl of an absolute vector field; an example
is the vorticity vector field, defined as the curl of the velocity field.
Performing a second such operation on a pseudovector such as taking a curl
will yield an absolute entity. Also, if the cross product of a pseudovector and
an absolute vector is formed, the result with be an absolute vector. You will
find a more detailed discussion of the distinction between absolute and
pseudotensors in the textbook by Leal (pages 178-179).
Solution of the Stokes equation
Now, let us begin with the Stokes equation for an incompressible flow.
2 u = p
( P)
x j 1 x j
p 1
p
uj =
p =
p + xj
=
ij p + x j
xi
xi 2 2 xi
xi 2
xi
1
p
( ij p ) +
xj
2 xi
xi xi
xj
2
p.
x j p
1 p
2 p 1 p
p 1 p 1
x
+
+
ij + ji
= p
ij
j
=
2 xi
xi xi
xi xi 2 xi
xi xi
In obtaining the above result, we have used the fact that the pressure field
2 p
Well find that there are only a limited number of ways in which the vectors
(or scalars) from the boundary conditions can be combined with the set of
decaying or growing harmonics using operations such as the cross or dot
product or simple scalar multiplication to generate scalars or vectors of order
1. It is this fact that makes it relatively easy to construct the desired
solutions in invariant form. We shall illustrate with three examples, flow
caused by a sphere rotating in place in a quiescent fluid, flow caused by a
sphere translating at a constant velocity through a quiescent fluid, and flow
caused by a point force in a quiescent fluid.
( ) ( r , ) + (
1
2)
and
( 2)
(1)
E 2 ( ) = ( )
2
(2)
(3)
n =2
d 2
(1 ) d 2 + n ( n 1) = 0
2
(4)
d
d
1 2 )
+ n ( n + 1) = 0
(
d
d
(5)
Pn2 ( ) Pn ( )
2n 1
Cn1/ 2 ( ) = Pn1 ( s ) ds
Cn1/ 2 ( ) =
(6)
(7)
C01/ 2 ( ) = 1
C11/ 2 ( ) =
C21/ 2 ( ) = 1 (1 2 )
C31/ 2 ( ) = 1 (1 2 )
P0 ( ) = 1
P1 ( ) =
2
C41/ 2 ( ) = 1 (1 2 )( 5 2 1)
8
1
1
P3 ( ) = ( 5 2 3)
3 2 1)
(
2
2
1
P4 ( ) = ( 35 4 30 2 + 3)
8
P2 ( ) =
+1
Cm1/ 2 ( ) Cn1/ 2 ( )
d =
for values of m, n 2.
n ( n 1)( 2n 1)
mn
(8)
References
M.A. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions,
Dover Publications (1965).
J. Happel, and H. Brenner, Low Reynolds Number Hydrodynamics,
Prentice-Hall (1965); republished by Noordhoff International (1973).
L.G. Leal, Laminar Flow and Convective Transport Processes,
Butterworth-Heinemann (1992).
T.M. MacRobert, Spherical Harmonics, Pergamon Press (1967).
R.A. Sampson, On Stokess Current Function, Phil. Trans. Royal Soc. A
182, 449-518 (1891).
Vorticity
R. Shankar Subramanian
The vorticity = u , is an important entity in fluid mechanics. It is transported from
one place to another in a fluid by convective and molecular means, just as energy and
species are. In addition, vorticity also is intensified by the stretching of vortex lines, a
mechanism that is not present in the transport of energy and species. One reason for
working with the equations of vorticity transport is that pressure is absent as a dependent
variable. It can be shown that if a fluid mass begins with zero vorticity, and the fluid is
inviscid (meaning the viscosity is zero), the vorticity will remain zero in that fluid mass.
A flow in which the vorticity is zero is known as an irrotational flow.
Vorticity is generated at fluid-solid interfaces and at fluid-fluid interfaces. Vorticity
cannot be generated internally within an incompressible fluid. This is the reason why, in
a high Reynolds number flow (implying weak viscous effects) past a rigid body, most of
the flow can be described by using the equations that apply to irrotational flow, with the
vorticity being confined to a boundary layer near the surface of the body.
The vorticity tensor is a skew-symmetric tensor. We can write its components in
terms of the components of the velocity gradient in a rectangular Cartesian basis set as
follows.
1 u u
= y x
y
2 x
1 u z u x
2 x z
1 u x u y
x
2 y
0
1 u z u y
z
2 y
1 u y u z
y
2 z
1 u x u z
x
2 z
Because a fluid does not usually rotate as a rigid body in the manner that a solid does, we
should interpret the above statement as implying that the average angular velocity of a
fluid element located at a point is one-half the vorticity vector at that point. To prove
this, consider a surface formed by an infinitesimal circle of radius a located at a point x .
Let the unit normal vector to the surface be n , and the unit tangent vector to the circle at
any point be t .
t
n
a
Apply Stokess theorem to the velocity field in this circle.
( u ) n dS = v u t ds
S
Here, dS is an area element on the surface of the circle S and ds is a line element along
the circle C. Because u t is the component of the velocity along the periphery of the
1
circle, we can write the average linear velocity along the circle as
u t ds and
2 a v
C
therefore the average angular velocity as
1
2 a 2
v u t ds .
that this is equal to the average value of u n over the surface of the circle.
2
Thus, in the limit as the radius of the circle approaches zero, we find that the average
angular velocity around the circle approaches the value of one-half the component of the
vorticity vector in a direction perpendicular to the surface of the circle. We also can
show (see Batchelor, page 82) that the angular momentum of a spherical element of fluid
is equal to one-half the vorticity times the moment of inertia of the fluid, just as it is for a
rigid body.
Another useful result is that dxi ij dx j = 0 . You can prove this by using the fact that
ij = ji .
dx
dy
dz
The surface that is formed by all the vortex lines passing through a closed reducible curve
is known as a vortex tube. If we construct an open surface S bounded by this closed
curve C, we can define the strength of the vortex tube as
dS .
By using Stokess
v u t ds where
around the vortex tube, t is a unit tangent vector to the curve at any point, and ds is a
line element.
Some good sources for further study are listed below.
1. G.K. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press,
1967, Chapter 5.
2. C. Truesdell, Kinematics of Vorticity, Indiana University Press, 1954.
3. J. Serrin, Mathematical Principles of Classical Fluid Dynamics, Handbuch der
Physik VIII/1, Ed. S. Flugge, 1959.
4. P.G. Saffman, Vortex Dynamics, Cambridge University Press, 1992.
2
u 2 u y 2 u 2 u
u y
x
x
z
v = 2
+
+
+
+
x
x y z y
u
u
u
2
2
u
+ y + z + z + x ( u)
y
z
3
x
z
2
ur 2 1 u ur 2 u z 2
v = 2
+ +
+
r
r
u
+ r
r r
1 ur 1 u z u
+
+
+
z
r r
2
2
u u
+ r + z ( u)
r
3
z
2
u
u
u
1
r
r
r
v = 2
+ +
+ + cot
+
r
r
r
r
r
r
sin
u
+ r
r r
1 u
1 ur sin u
+
+
+
r r sin r sin
2
1 ur
u
2
2
+
+ r ( u)
r r
3
r sin
The fluid is initially assumed to be at rest. Motion is initiated by suddenly moving the
bottom plate at a constant velocity of magnitude U in the x direction. The velocity of
the bottom plate is maintained at that value for all future values of time t while the top
plate is held fixed in place. There is no applied pressure gradient, with motion being
caused strictly by the movement of the bottom plate.
We shall assume the flow to be incompressible with a constant density and Newtonian
with a constant viscosity . We neglect edge effects in the z direction so that we can
v
v
set vz = 0 and
= 0 . Here, v
= 0 , and assume fully developed flow, implying
x
z
stands for the velocity vector, and the subscripts denote components.
It can be established from the continuity equation and the kinematic condition at one of
the walls that v y = 0 . Therefore, the only non-zero velocity component is vx ( t , y ) ,
which can be shown to satisfy the following partial differential equation.
vx
2 vx
=
t
y 2
(1)
In Equation (1), t represents time, and is the kinematic viscosity. The initial condition
is
(2)
vx ( 0, y ) = 0
and the boundary conditions are
vx ( t , 0 ) = U
(3)
vx ( t , b ) = 0
(4)
and
vx ( t , y )
2
y
= 1
U
b
n =1
exp n 2 2 2
b
n y
sin
n
b
(5)
The infinite series in Equation (5) is uniformly convergent for all values of time t . The
exponential factor plays a strong role in assuring that the terms decrease rapidly with
increasing values of n so that only a few terms are necessary to calculate an accurate
value of the velocity at moderate to large values of time, corresponding to the scaled time
t / b2 not being too small compared to unity. But, if we attempt to calculate the sum
numerically for small values of time ( t / b 2 small compared with unity) when the
exponential factor is not as helpful, we find that a large number of terms needs to be
included to obtain a sufficiently accurate answer. Therefore, in this module we seek a
solution technique that will permit us to calculate the velocity field accurately without too
much labor for small values of time.
Physically, at values of time t for which the scaled time t / b 2 is small compared to
unity, the effect of the motion of the bottom plate is only felt by the fluid up to a small
distance (depth of penetration) from the moving plate. Outside of this region of
influence, the fluid is practically stationary. Therefore, one might approximate the
system for such small values of time by another in which the top plate is absent. This
problem was first considered by Lord Rayleigh, and therefore is known as Rayleighs
problem. Mathematically, we replace the boundary condition at the top plate, given in
Equation (4), with
vx ( t , ) = 0
(6)
There is neither a natural length scale in the problem, nor a natural time scale. We can
use the reference velocity U as a natural scale for the velocity vx , but it is convenient to
work with the remaining physical variables just as they are. The solution of Equations
(1) to (3) and (6) is qualitatively sketched below for two different values of time. In the
figure, the symbol v is used to represent vx .
The solid (black) line corresponds to a small value of time, and the dashed (red) line to a
larger value of time. We can see how the change in velocity made at the bottom plate at
time zero propagates deeper into the fluid with increasing time. It is tempting to speculate
that these profiles are similar in shape. By implying similarity of shape, we mean that
scaling the distance variable with the thickness of the affected region ( t ) should lead to
these two curves and others like them collapsing into a single universal curve. In
mathematical language, if we define a certain combination of the original variables as a
sketch below.
We now proceed to state the above speculation in mathematical form and follow through
the consequences. This is the method of Combination of Variables.
Assume
vx ( t , y ) = U ( )
(7)
where
y
(t )
(8)
and ( t ) is a function that is yet to be determined. Note that we always can transform
from two independent variables to two new independent variables, but to transform to a
single new variable is not always possible. Therefore, we need to insert Equations (7)
and (8) into the governing equation and the initial and boundary conditions and see if the
process leads to a consistent mathematical framework. For this purpose, we shall use the
chain rule of differentiation as needed.
vx
d
y d d
d d
=U
= U 2
=U
t
t d
dt d
dt d
(9)
Note that when writing the derivative of with respect to , we already have assumed
that can depend explicitly only on the single variable and used the ordinary
derivative instead of the partial derivative. If our conjecture proves to be incorrect, and
were to depend explicitly on both and t , the above chain rule result will need to be
modified to include a partial derivative of with respect to time.
Let us now obtain expressions for the derivatives with respect to y .
vx
d U d
=U
=
y
y d d
(10)
2 vx vx
U d U d
=
=
2
y
y y y d y d
U d d U d 2
=
=
y d d 2 d 2
(11)
and
Substituting Equations (9) and (11) into the governing differential equation for vx
(Equation (1), leads to the following equation after slight rearrangement.
=0
(12)
In writing Equation (12), we have used the expedient of designating derivatives with
respect to the argument of each function with primes. Recall that we assumed that
explicitly depends only on the similarity variable . But, Equation (12) suggests that
time also will explicitly appear in the result for because of the presence of the timedependent term . We have not yet specified ( t ) , however. Here is our chance to
do so and eliminate the inconsistency at the same time. We choose
= C
(13)
where C is an arbitrary constant. Later, we shall see that the value of C will affect the
result for ( t ) , but will not affect the final solution for vx ( t , y ) . Therefore, for
convenience, we set C = 2 , writing Equation (12) as
+ 2 = 0
(14)
We now need to transform the initial and boundary conditions. Note that there are three
conditions on the velocity field vx ( t , y ) , but only a second order differential equation for
( ) . The specification of the arbitrary constants that arise in the integration of the
latter requires only two conditions.
First, consider the boundary condition at the bottom surface y = 0 , given in Equation (3).
This transforms in a straightforward manner to
(0) = 1
(15)
() = 0
(16)
y
= 0
0
(
)
(17)
and we see that we have not completely eliminated the original variables from appearing
explicitly in the problem statement for . To remove this inconsistency, and at the same
time select an initial condition for ( t ) , we must set
( 0) = 0
(18)
The choice in Equation (18) makes Equation (17) collapse into Equation (16); therefore,
the three conditions on vx ( t , y ) yield two conditions on ( ) and one initial condition
on ( t ) , and we have a completely consistent mathematical framework for the problems
for ( ) and ( t ) . Note that by this approach of Combination of Variables we have
reduced the solution of the original partial differential equation to that of two ordinary
differential equations for these two functions.
First, the general solution of Equation (14) can be written as
( ) = a1 + a2 e d
2
(19)
where a1 and a2 are constants of integration that must be determined by applying the
boundary conditions given in Equations (15) and (16). Use of these conditions leads to
the result
( ) = erfc ( )
(20)
(21)
erf ( ) =
e d
2
(22)
You can find out more about the error function and the complementary error function
from Abramowitz and Stegun [1].
The solution of Equation (13) with the constant C = 2 , when specialized using the initial
condition given in Equation (18), is
(t ) = 2 t
(23)
When this result for ( t ) is used in Equation (8) in which is defined, the solution for
the velocity field can be written as
y
vx ( t , y ) = U erfc
(24)
2 t
If we had made a different choice of value for the constant C that appears in Equation
(13), it would have affected the results as follows.
( t ) = 2C t =
C
2 t
2
C
2
( ) = erfc
(25)
(26)
You can see that when the definition of given in Equation (8) is used in Equation (26),
along with ( t ) from Equation (25), the factor C / 2 cancels out, leading to the same
result for the velocity field given in Equation (24). You may wonder about the
uncertainty in the value of ( t ) , which is the thickness of the affected region, caused
by the indeterminacy of the value of C . This is perfectly natural because in a diffusive
process, the influence of a change is felt everywhere in the fluid instantaneously. This
means that there can be no unambiguous definition of a finite thickness for the affected
region; only its scaling can be established uniquely. The complementary error function
assumes a value of 4.678 103 when its argument is 2 . Therefore, at a distance
y = 4 t , the velocity would be less than 0.5 % of the value at the surface of the moving
plate, and can be considered negligible for practical purposes. Because of this, the
estimate 4 t is sometimes used for the thickness of the region influenced by the
In this module, we have learned the method of combination of variables for solving
partial differential equations; it complements the method of separation of variables. First,
we identified the governing partial differential equation and boundary conditions for our
system. Then we
1. noted that the effect of a boundary condition imposed at time zero is felt in a region
near that boundary that is small in extent for small values of time and used this fact to
replace the boundary condition at the other boundary with one at infinity;
2. assumed that the dependence of the velocity field on the two independent variables can
be expressed as a dependence on a single new similarity variable;
3. traced the consequences of this similarity hypothesis mathematically, requiring that the
original independent variables not be allowed to appear explicitly in the problem posed in
the new similarity variable;
4. obtained an ordinary differential equation for the thickness of the affected region and
another ordinary differential equation for the velocity field;
5. collapsed the three boundary conditions on the velocity field into two on the velocity
field as expressed in the similarity variable, also yielding an initial condition for the
thickness of the affected region;
6. solved these ordinary differential equations to obtain results for the thickness of the
affected region and the velocity field;
7. noted that the thickness of the affected region can only be defined to within a
multiplicative arbitrary constant, whereas the velocity field is uniquely determined.
8
The important features of the method are that the domain must be semi-infinite, and the
boundary condition at infinity must be the same as the initial condition; even though the
problem we posed is linear, the method is equally applicable to non-linear problems.
Concluding Remarks
An Introduction to Mixing
R. Shankar Subramanian
Mixing is a common operation in the chemical process industry. The objectives of mixing
are to achieve one or more of the following.
(1) homogenize a mixture of fluids
(2) emulsify one fluid in another
(3) suspend solids in a liquid
A sketch of a generic mixer used for handling low to medium viscosity fluids is given
below.
Shaft
Liquid
Level
Baffle
DT
Blade
DA
Typically, a mixer is a hollow cylinder filled with the substances to be mixed to some
appropriate level. An agitator consisting of a shaft connected to a motor at one end and a
set of blades at the other is mounted as shown. In the absence of baffles on the wall, a
simple rotating flow, sometimes called a vortex flow, can result that would not cause
mixing. In addition, such a simple flow would cause a dip in the surface and this can
exacerbate the entrainment of air into the liquid. Therefore, mixers typically have baffles
that are attached to the wall to break up this vortex flow. You can learn about mixing and
mixers from McCabe et al. (2001) and Holland and Bragg (1995) as well as Perrys
Chemical Engineers Handbook. Here only a limited discussion is given, focusing on
1
dimensional analysis. One important engineering issue is how to scale up from a laboratory
scale device to pilot plant size and on to actual process scale equipment. We use
dimensional analysis to find the important groups. A limited list of the important
parameters is given below. Several other geometric parameters are omitted. In designing a
scale model, it is important to maintain geometric similarity so that all the aspect ratios and
shapes are preserved.
:
:
DT :
DA :
g:
N:
P:
N P will be some
function of the Reynolds number Re and the Froude number Fr . It also will depend on
the geometric parameters such as the ratio of the diameter of the agitator to that of the tank
and those arising from the length scales and shapes of the blades. All of these are assumed
to be the same in the model and the prototype and therefore we shall not include them in
the list of parameters on which the Power number will depend.
N P = ( Re, Fr )
Here
NDA2
Re =
P
NP =
N 3 DA5
N 2 DA3
Also, the Weber number We =
N 2 DA
Fr =
g
If both the Reynolds and Froude numbers are important, we cannot use a scale model. Let
us investigate why. We use the subscripts m and p to designate model and prototype,
respectively, in the following. For dynamic similarity,
Rem = Re p .
Therefore,
p N p DA2
m N m DA2
=
m
p
m
N m2 DAm
g
N p2 DAp
g
m p
=
.
m p
Frm = Frp .
Nm
=
Np
or
So,
2
N m D Ap
=
N p DA2m
This implies
D Ap
DAm
DAp
DAm
DAm = DAp .
the model must be of the same size as the prototype, which defeats the purpose of building
a scale model.
Baffling the system eliminates vortex formation and removes the Froude number as a
significant parameter. In that case, dynamic similarity requires that
Nm = N p
DA2p
DA2m
assuming we use the same liquid in both the model and the prototype. The implication of
this result is that the speed of rotation in the model should be substantially higher than that
in the prototype.
N P = cRe a
This can be recast as
log N P = c + a log Re
where the new constant
c = log c .
a = 1 . Therefore, for a laminar mixer, we can write the result for the power input to the
mixer as
3
P = c N 2 DA3
Laminar mixers are not common. Usually, the flow is turbulent. For a generic baffled
mixer, the constant a is found to be small, and when the flow is fully turbulent
corresponding to Re 10,000 , we can set a 0 . In this case, the power input is found
to be given by
P = 6.3 N 3 DA5
Note the significantly larger exponents on the rotation speed and the diameter of the
agitator in the case of turbulent flow. Another important distinction between the results for
the power input in the laminar and turbulent cases is the appearance of the viscosity and
not the density in the laminar flow result, and the reverse in the turbulent flow result. Can
you rationalize this from your knowledge about the mechanisms of dissipation in these two
types of flows?
References
1. F.A. Holland and R. Bragg, Fluid Flow for Chemical Engineers, Chapter 5, Edward
Arnold, London, 1995.
2. W.L. McCabe, J.C. Smith, and P. Harriott, Unit Operations of Chemical Engineering,
Chapter 9, McGraw-Hill, New York, 2001.
Exchanger Matrix
PIPE FLOW
Re =
DV
Entrance Length
Le / D = 0.06 Re for Laminar
Le / D = 4.4 Re1/6
f =
P D
1 / 2 V 2 L
f Fanning =
2
L V
P = f
D 2
Head Loss
hf =
1
f Darcy
4
2
P
L V
= f
g
D 2 g
Q=
R 4 P D 4 P D 2
=
=
V
8L
128l
4
f =2
f =
P D
= 64
2
L V
DV
64
Re
f = Re,
D
Colebrook
2.51
= 2.0 log10 D +
3.7 Re f
f
Useful Equations
f =
P D
1 / 2 V 2 L
2
L V
P = f
D 2
Q=
f Fanning =
hf =
R 4 P D 4 P D 2
=
=
V
8L
128l
4
2
P
L V
= f
g
D 2 g
f =2
f =
f = Re,
D
1
f Darcy
4
P D
= 64
L V 2
DV
64
Re
2.51
= 2.0 log10 D +
3.7 Re f
f
Description
Drawn tubing (brass, lead, glass etc)
Commercial steel or wrought iron
Asphalt cast iron
Galvanized iron
Cast iron
Concrete
Riveted steel
Q1. Read the value of friction factor for Re=105 and (/D) = 0.0002 and compare the
value to that obtained by equation:
1
10 6 3
f = 0.0013751 + 20,000 +
D Re
f = 0.00475
10 6
f = 0.0013751 + 20,000(0.0002 ) + 5
10
f = 0.0047
Calculate:
Head Loss
Q=
If Efficiency is 0.9
Given:
A=
D 2
4
Q = 2000 gpm
8 inch ID
Kinematic viscosity
V =
10-4ft2/s
Density = 1.8 slug/ft3
(0.667 ft )2 = 0.349 ft 2
Q
= 12.8 ft / s
A
Re =
DV
0.667 x12.8
= 85,500
10 4
2.51
= 2.0 log10 D +
3.7 Re f
f
f=0.0235
Head Loss
hf =
2
P
L V
= f
g
D 2 g
= 134 ft
loss = hf x g
Power supplied = loss x mass flow rate
= 62.9 HP (550 ftlb/s = 1HP)
BHP = 62.9 /0.9 = 69.9 HP
Pipe Flow
Losses occurring in straight pipes Major losses
Losses occurring in pipe systems Minor losses
hL = f
KL =
l V2
D 2g
Q hL =
hL
p
=
V / 2 g 1 / 2 V 2
p = K L 1 / 2 V 2
hL = K L
p
hL
p 2 g
p
=
=
g V 2 / 2 g g V 2 1 / 2 V 2
V2
2g
KL = (geometry, Re)
However for most practical cases:
K L p = 1 / 2 V 2
KL = (geometry)
Dynamic pressure
leq V 2
D 2g
hL = f
leq =
KLD
f
Vena contracta
Flow separation
V2>V3
p1
V32/2
V22/2
KL V32/2
Actual
x1
x2
x3
0.5
r
0.4
KL
0.3
0.2
0.1
0
0.1
0.15
0.20
0.25
r/D
Minor losses are often a result
of the dissipation of kinetic energy
Elbows, 180 0 return bends, Tees, Union threaded, Valves Table 6.5 page 387 White
Water flows from a large reservoir and discharges into the atmosphere. Determine the
volumetric flow rate m3/s?
150m
(2)
47m
16m
Standard 900 elbows
(1)
Rounded entrance
D = 0.156m
= 5 x 10-5
150m
Density of water = 103 kg/m3
Viscosity of water = 1 x 10-3 N s/m2
Write Bernoullis equation between the free surface and the entrance.
By continuity V1=V2=V
p1 V12
p
+
+ z1 = s + z s
g 2 g
g
p1 ps
V2
= ( z s z1 ) 1
g
2g
p1 V 2
p V2
fL V 2
V2
V2
+
+ z1 = 2 +
+ z2 +
+ K L1
+ 2K L2
g 2 g
g 2 g
D 2g
2g
2g
p1 p2
fL V 2
V2
V2
= ( z 2 z1 ) +
+ K L1
+ 2K L2
g
D 2g
2g
2g
fL
+ K L1 + 2 K L 2 + 1
2g D
(z s z2 ) = V
V2 =
2 g (z s z2 )
fL
+ K L1 + 2 K L 2 + 1
D
Both f and V are unknown work simultaneously with the friction factor diagram
to obtain a solution:
V (assumed)
m/s
Re
5(10)5
0.0165
115
1.8(10) 0.0150
122
1.9(10) 0.0150
V (calculated)
m/s
7
7
122
10