Vous êtes sur la page 1sur 130

Non-Newtonian Flows

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

We describe the relationship between the shear stress


follows.
1

and shear rate

as

= 
Here,

is called the apparent viscosity of the fluid, and is a function of the

Apparent Viscosity

shear rate. In the above example, a plot of


looks like this.

as a function of the shear rate

shear-thinning fluid

Shear Rate

Many shear-thinning fluids will exhibit Newtonian behavior at extreme shear


rates, both low and high. For such fluids, when the apparent viscosity is plotted
against log shear rate, we see a curve like this.

log Apparent Viscosity

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 = exp ( a ) . Instead of b we commonly use ( n 1) for the exponent

and write a result for the apparent viscosity as follows.

= 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

Examples of viscoplastic fluids are drilling mud, nuclear fuel slurries,


mayonnaise, toothpaste, and blood. Also, some paints exhibit a yield stress.
Of course, this is not an exhaustive discussion of non-Newtonian behaviors. For
instance, some classes of fluids exhibit time-dependent behavior. This means
that even under a given constant shear rate, the viscosity may vary with time.
The viscosity of a thixotropic liquid will decrease with time under a constant
applied shear stress. However, when the stress is removed, the viscosity will
gradually recover with time as well. Non-drip paints behave in this way. The
opposite behavior, wherein the fluid increases in viscosity with time when a
constant shear stress is applied is not as common, and such a fluid is called a
rheopectic fluid.
Another important class of fluids exhibits viscoelastic behavior. This means that
these fluids behave both as solids (elastic) and fluids (viscous). Viscoelastic
fluids exhibit strange phenomena such as climbing up a rotating shaft, swelling
when extruded out of a dye, etc. An example of a common viscoelastic liquid is
egg-white. You have probably noticed that when it flows out of a container, you
can use a quick jerking motion to snap it back into the container. Several
industrially important polymer melts and solutions are viscoelastic. You can
learn more about these and other behaviors from several fluid mechanics
textbooks and from Perrys Chemical Engineers Handbook.

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 )

are used. We assume steady incompressible

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 .

( )

Similarly, the shear stress

rz = rz ( r ) .

The shell shown in the figure is

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.

Now, the volumetric flow rate of fluid into the shell is

Q = 2 r r vz
If the density of the fluid is

, the associated mass flow rate 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

Because this is independent of


right side is exactly the same.
Rate of efflux of momentum

z , the rate of efflux of momentum from the face on the

= 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

g is the magnitude of the (vector) acceleration due to gravity g .

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

shell at the location

exerts a stress

rz ( r )

on the fluid below, i.e. at smaller

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 )

Add the forces, set the sum to zero, and divide by

2 rL to obtain the following result.

p ( 0 ) p ( L ) ( r + r ) rz ( r + r ) r rz ( r )
r g cos +
=0
+
L

r 0 . The first term is


unaffected while the second term is simply the derivative of r rz with respect to r . This
leads us to the first order differential equation for the shear stress rz ( r ) .

Now we take the limit of the terms in the above equation as

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

This velocity distribution describes a parabola. It is sketched below.

vz ( r )

We see that the maximum velocity occurs at the centerline, which is at


is

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

dQ = vz ( y )W dy . This can be integrated across the cross-section as follows.


h

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

Q by the cross-sectional area 2Wh .

P h2
=
3 L

We can see that the ratio of the average to the maximum is


between wide parallel plates.

2 / 3 for steady laminar flow

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 ) .

Therefore, we can write

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

equations obtained using the same principles, known as the Navier-Stokes


equations, and simplify them for the specific problem under consideration.

10

Elementary Aspects of Two-Phase Flow in Pipes


R. Shankar Subramanian

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

Wavy Stratified Flow


Further increase in VG
leads to higher waves
Gas
Liquid
Slug 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

Annular mist flow

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.

Laminar flow that starts from a state of rest


in a straight circular tube
R. Shankar Subramanian
Here, we consider the problem in Homework # 7. We shall use a cylindrical polar
coordinate system r , , z in which z represents the direction of flow, and r and

stand for the radial and angular directions, respectively.

P0

PL

We make the following simplifying assumptions.


1. Incompressible flow: Continuity reduces to v = 0
2. Newtonian flow with constant viscosity
v
3. Neglect end effects:
= 0 (fully developed flow)
z
v
4. Symmetry in (known as axial symmetry):
= 0 ; v = 0

The incompressible version of the equation of continuity, in conjunction with


assumptions 3 and 4, yields

( 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,

The Navier-Stokes equation, subject to assumptions 1 and 2, can be written in component


form in cylindrical polar coordinates. The components in the r and -directions yield
the result that the dynamic pressure P is uniform in those directions. Therefore, the

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)

By rearranging this equation as

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

where we have defined

Therefore, Equation (2) can be rewritten as

vz P vz
=
+

r
L
r r r
t

(4)

The initial and boundary conditions are written as follows.


Initial Condition:
Boundary Conditions:

(No Slip)

vz ( 0, r ) = 0

(5)

vz ( t ,0 ) is finite

(6)

vz ( t , R ) = 0

(7)

Now, we proceed to scale this problem by defining the following dimensionless


variables.

T=

t
R

;
2

y=

r
;
R

V (T , y ) =

vz
( PR 2 / 4 L )

In the above, is the kinematic viscosity

( / ).

You may wonder about the choice

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

characteristic velocity is of the order

PR 2 / ( L ) .

The factor

is used to make the

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)

along with the boundary conditions

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)

The solution of Equations (16) through (19) is accomplished by separation of variables.


If we substitute a trial solution of the form Vt T , y = G T y into Equation (16)

we find that the functions

( ) ( )

G (T ) and ( y ) must satisfy the following differential

equations.

dG
+ 2G = 0
dT
1 d d
+ 2 = 0
y

y dy dy

(21)
(22)

The solution of Equation (21) is

G = Ce T
2

(23)

where C is an arbitrary constant of integration. The general solution of Equation (22)


can be obtained by using Frobenius series. For our purposes, we simply use the final
result without going through the solution process.

= 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

Yp ( y ) of integer order p are the two linearly independent solutions of Bessels

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) ,

For our purposes, it is important only to note that the functions

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

where the product of the two arbitrary constants

B . We
Vt (T , y )

and

(26)

C1 is written as a new constant

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

j0,s may be found in Table 9.5, page 409, in Abramowitz

and Stegun (1965). Youll see that

j0,20 j0,19 . Therefore, additional roots can

be generated by approximating the interval between each by .

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

All that remains is the determination of the constants


applying the initial condition.

Bn . We can obtain them by

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)

The procedure is as follows. Begin with Equation (29) in the form

Vs ( y ) = Bn J 0 ( n y )
n =1

(34)

Multiply both sides by


to

from

yJ 0 ( m y ) where m is some integer, and integrate with respect

y = 0 to 1. Use Equations (30) and (31) to obtain

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 )

Substitution of this result after replacing the index


complete solution from Equation (12) as follows.

with

permits us to write the

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.

Introduction to Integral Balances and Conservation of Mass


R. Shankar Subramanian
This is a brief introduction to the subject of integral balances. These balances include
statements of conservation of mass, energy, and momentum, and will prove useful in a
variety of problems. For example, conservation of mass allows us to estimate the rate of
change of the level of liquid in a process vessel or the rate at which the amount of gas left
in a tank decreases due to leakage. Conservation of mass and energy allow us to size
pumps and turbines, and help in the evaluation of flow rates using flow measurement
devices. Conservation of momentum is used in calculating the forces on the supports
used for pipe bends and the forces flange bolts need to withstand. Also, by using these
balances together, we can calculate the losses in a sudden expansion, design a jet ejector,
and make calculations involving pipe manifolds.
The idea behind these balances is that a certain entity is conserved. For instance, if we
select an arbitrary volume fixed in space, called a control volume, and abbreviated as
CV, conservation of mass would be stated as follows:
Rate of increase of mass of material within the control volume = Net rate at which
material enters the control volume.
An arbitrary control volume V is shown in the sketch.

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=

V , we can obtain the total mass of

dV

(2)

dM
. Therefore,
dt
dM
d
=
dV
dt
dt V

The time rate of change of this mass is then

(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

Rate of net inflow of mass


into 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)

m = m 1 = m 2 is constant. If the density is

constant, the above relationship reduces to Q1 = Q2 . The assumption of constant density


is known as the assumption of incompressible flow. It is always a good assumption to
make in liquids. In the case of gases flowing at velocities that are small compared with
the speed of sound in the gas, it continues to be a good assumption.

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

Therefore, we can write minor

is called the loss coefficient.

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

The losses would be smaller if the expansion or contraction is gradual.


When a pipe empties into a reservoir, all the kinetic energy in the fluid coming in is dissipated,
so that you can treat this as a sudden expansion with the ratio

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 the velocity of the disk,

is the density of the fluid, and

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

0.01 < Re 1.5 ,

CD =

64
1 + 10 x )
(
Re

x = 0.883 + 0.906log10 Re 0.025(log10 Re) 2

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 ,

and is constant at a value of

1.17

for

CD becomes insensitive

Re > 1,000 ,

and that there is some

CD passes through a minimum of about 1.03 for Re  400 . But the


authors go on to state that most data are correlated within 10% by the result valid for
1.5 < Re 133 , with CD = 1.17 for Re > 133 .
indication that

Terminal Settling Velocity


From page 148-149, we learn that a disk in free motion will move steadily with its axis
vertical so long as ReT (meaning the Reynolds number at terminal settling velocity)

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

CD Re is used to characterize transition to various regimes of unsteady motion of


2
T

the settling disk.


Apparently, when the transition from steady to unsteady motion occurs, a disk executes
regular oscillations about a diameter as it settles. At larger

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.

Drag on spherical particles and steady settling velocities


R. Shankar Subramanian
Most textbooks present results for the dependence of the drag coefficient for a smooth
sphere, CD , on the Reynolds number, Re , in the form of a curve. Such curves are
difficult to read accurately especially without a fine grid. Also, the determination of the
terminal settling velocity of a sphere in a fluid using this curve is an iterative process.
Clift, Grace, and Weber (Bubbles, Drops, and Particles, Academic Press, 1978) provide
correlations based on experimental data that can be used to calculate either the drag
coefficient or the settling velocity directly. I have reproduced their results below for your
use in solving problems. The standard definitions of the Reynolds number and the drag
coefficient are given below.

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

260 Re 1.5 103 ,

log10 CD = 1.6435 1.1242 w + 0.1558w2

For1.5 10

Re 1.2 10 4 ,

log10 CD = 2.4571 + 2.5558w 0.9295w2 + 0.1049 w3


For1.2 10

Re 4.4 10 4 , log10 CD = 1.9181 + 0.6370 w 0.0636 w2

For

4.4 104 Re 3.38 105 , log10 CD = 4.3390 + 1.5809 w 0.1546 w2

For

3.38 105 Re 4 105 , CD = 29.78 5.3w

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

immediately evaluate the terminal settling velocity. In the equations, W

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

73 < N D 580; 2.37 < Re 12.2 ,

log10 Re = 1.7095 + 1.33438 W 0.11591W 2


For

580 < N D 1.55 107 ; 12.2 < Re 6.35 103

log10 Re = 1.81391 + 1.34671 W 0.12427 W 2 + 0.006344W 3


For 1.55 10

< N D 5 1010 ; 6.35 103 < Re 3 105

log10 Re = 5.33283 1.21728 W + 0.19007 W 2 0.007005 W 3

The Graetz Problem


R. Shankar Subramanian
As a good model problem, we consider steady state heat transfer to fluid in steady flow through a
tube. The fluid enters the tube at a temperature T0 and encounters a wall temperature at Tw ,
which can be larger or smaller than T0 . A simple version of this problem was first analyzed by
Graetz (1883). A sketch of the system is shown below.

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.

4. Negligible viscous dissipation

Velocity Field
Poiseuille Flow

vr = 0;

v = 0

r2
vz ( r ) = v0 1 2
R

v0 : Maximum velocity existing at the centerline

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

and therefore, simplified to

1 T
r 2 T
v0 1 2
=
r
R
z
r
r

2
T
+

2
z

where = k / ( C p ) is the thermal diffusivity of the fluid.

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

The Pclet Number


The Pclet number plays the same role in heat transport as the Reynolds number does in fluid
mechanics. First, we note that the Pclet number is the product of the Reynolds and Prandtl
numbers.
Pe =

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

Rate of energy transport by convection


Rate of energy transport by conduction

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

= 5 . Therefore, the Reynolds number is Re = 500 , which is comfortably

in the laminar flow regime.


Prandtl number Pr =

The final version of the scaled energy equation is

(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 )

This leads to the following differential equation for W ( X ) .


X

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

is an arbitrary multiplicative constant. The function M ( a, b, X ) is the confluent

hypergeometric function, or Kummer function, and is discussed in Chapter 13 of the Handbook


of Mathematical Functions by M. Abramowitz and I. A. Stegun, It is an extension of the
exponential function, and is written in the form of the following series.
a ( a + 1) X 2
a
M ( a, b, X ) = 1 + X +
+ "
b
b ( b + 1) 2!
+

a ( a + 1)" ( a + n 1) X n
+"
b ( b + 1)" ( b + n 1) n !

You can see that when a = b ,


M ( a, a, X ) = e X

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 )

The first few eigenvalues are reported in the table.

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

designate {n } as that set for convenience.


The most important property of a proper Sturm-Liouville system is that the eigenfunctions are
orthogonal with respect to a weighting function that is specific to that system. In the present
case, the orthogonality property of the eigenfunctions can be stated as follows.
1

(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

The Heat Transfer Coefficient


The heat flux from the wall to the fluid, qw ( z ) is a function of axial position. It can be
calculated directly by using the result
T
( R, z )
r
but as we noted earlier, it is customary to define a heat transfer coefficient h ( z ) via
qw ( z ) = k

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

where b is the dimensionless bulk average temperature.

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

Dimensionless Axial Position Z

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.

The Lvque Approximation


The orthogonal function expansion solution obtained above is convergent at all values of the
axial position, but convergence is very slow at the inlet is approached. The main reason for this
2
is the assistance provided by e n Z in accelerating convergence for sufficiently large values of
Z . Lvque (1928) considered the thermal entrance region in a tube and developed an
alternative solution, which is useful precisely where the orthogonal function expansion
converges too slowly.

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

This means that the

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

By choosing ( 0 ) = 0 , this condition collapses into the condition F ( ) = 1 obtained already


from the boundary condition on the scaled temperature field as X . Summarizing the
boundary conditions on F ( ) and ( Z ) , we have
F ( 0 ) = 0, F ( ) = 1 , and

(0) = 0
Integration yields the following solution for the scaled boundary layer thickness ( Z ) .
1/ 3

9
(Z ) = Z
2

The solution for F ( ) is

F ( ) =

d
d

3
1
e d
=

( 4 / 3) 0

Here, ( x ) represents the Gamma function, discussed in the Handbook of Mathematical


Functions by Abramowitz and Stegun. The numerical value of ( 4 / 3) 1.120 .
Heat Transfer Coefficient

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

Transforming to dimensionless variables, and defining a Nusselt number Nu = 2hR / k , we


can write
Nu ( Z ) = 2

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.

By substituting for ( Z ) and

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

problems, and is given the name Grashof Number, abbreviated as Gr.

Gr =

g L3

The coefficient of volumetric expansion of a fluid is defined as

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.
=

We can finally rewrite the definition of the Grashof number as follows.

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 )

Nuaverage = 0.68 + 0.670 ( Ra )

(1 + 1.6 10

1/ 4

1/ 4

Ra 109
8

Ra )

1/12

109 Ra < 1012

In these equations is a function of the Prandtl number, defined as follows.


16 / 9

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.

Nuaverage = 0.52 Ra1/ 4

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

1. Mills, A.F., Heat Transfer, Prentice-Hall, New Jersey (1999).


2. Bird, R.B., Stewart, W.E., and Lightfoot, E.N., Transport Phenomena, Wiley, New York
2007).
3.Gebhart, B., Jaluria, Y., Mahajan, R.L., and Sammakia, B., Buoyancy-Induced Flows and
Transport, Hemisphere, Cambridge (1988).

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.

Molecular transport or Conduction


Molecular transport rates are adequately described for moderate temperature gradients by
a linear relationship between the heat flux and the temperature gradient. The
phenomenological relationship
q = k T

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.

Convective or Bulk Transport


Thermal energy also is transported by the physical movement of an element of fluid from
one place to another. This is known as convective transport. In our simplified picture,
the (vector) convective flux of thermal energy at a given point can be written as
Flux = C p v (T Tref

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

The assumption of thermodynamic equilibrium at the interface leads to the boundary


condition

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.

The heat transfer coefficient


The heat transfer coefficient h and its dimensionless counterpart, the Nusselt Number
Nu , are in common use in engineering work. Here, I discuss how the heat transfer
coefficient is defined in typical situations.
Consider a fluid at a temperature T flowing at a uniform velocity U that encounters a
rigid wall, which is maintained at a uniform temperature Tw . Let us assume that Tw > T
for the sake of definiteness; however, the results given below are equally valid when
Tw < T .

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

This definition of h ( x ) holds regardless of the sign of (Tw T ) .


The Nusselt number Nu is a dimensionless version of the heat transfer coefficient. It is
defined in the above problem as follows.
Nu =

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

Tw . A sketch of the system is given below.

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

Here, V ( r ) represents the velocity field.

The heat transfer coefficient in this system is defined as follows.


qw = h ( z )(Tw Tb )

so that we can evaluate h ( z ) using


T
( R, z )
h ( z ) = r
(Tw Tb )
k

if we know the detailed temperature distribution in the fluid.


We see from these two examples that the heat transfer coefficient will depend on
position, the system parameters, and on time in unsteady state problems. This concept of
a heat transfer coefficient is extended to many practical heat transfer situations.
Typically, in any given system, heat transfer rates and suitably defined driving forces are
both measured. The ratio of the flux of thermal energy to the driving force expressed as a
temperature difference is reported as the heat transfer coefficient.
h=

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

We already have defined a dimensionless group commonly used in engineering practice,


namely, the Nusselt number Nu . Considering it to be an average value for a given heat
transfer setting such as heat transfer to fluid flowing through a circular pipe of diameter
D , we can use dimensional analysis to identify the dimensionless groups on which Nu
will depend. Thus, we obtain
L

Nu = Nu Re, Pr,
D

hD
UD

, the Reynolds number Re =


, and the Prandtl number Pr = . The
k

velocity U appearing in the Reynolds number is a characteristic velocity in the system


such as the average velocity of flow, and and are the kinematic viscosity and the
thermal diffusivity of the fluid, respectively. The thermal diffusivity = k / ( C p ) .

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

The Prandtl number

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 =

Ability of a fluid to transport momentum by molecular means


Ability of that fluid to transport energy by molecular means

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

Small Prandtl number, Pr  1 (  )

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 ) .

Flow through Packed Beds and Fluidized Beds


R. Shankar Subramanian
Chemical engineering operations commonly involve the use of packed and fluidized beds. These
are devices in which a large surface area for contact between a liquid and a gas (absorption,
distillation) or a solid and a gas or liquid (adsorption, catalysis) is obtained for achieving rapid
mass and heat transfer, and particularly in the case of fluidized beds, catalytic chemical
reactions. You will find a good deal of information about flow through packed and fluidized
beds in the book by McCabe, Smith, and Harriott (2001) and Perrys Handbook (1997). Here,
only a brief summary is given. First, let us consider flow through a packed bed.

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

Volume of the particle


Surface area of the particle

: Density of the fluid


: Dynamic viscosity of the fluid
Vs : Superficial velocity ( Vs =

Q
where Q is the volumetric flow rate of the fluid and A is the
A

cross-sectional area of the bed)

: 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

the pores of the column.


Two simpler results, each obtained by ignoring one or the other term in the Ergun equation also
are in use. One is the Kozeny-Carman equation, used for flow under very viscous conditions.
fp =

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

Vf: Minimum Fluidization


Velocity
O

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

Balancing the two yields


p = (1 ) ( p f ) Lg
By using an expression relating p to the superficial velocity, which is the fluidization velocity
at this point, we can obtain a result for the latter.
Typically, for a bed of small particles ( D p 0.1 mm), the flow conditions at this stage are such
that the Reynolds number is relatively small ( Re 10 ) so that we can use the Kozeny-Carman
Equation, applicable to the viscous flow regime, for establishing the point of onset of
fluidization. This yields
Vf =

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.

Kinematics of Fluid Motion


R. Shankar Subramanian
Kinematics is the study of motion without dealing with the forces that affect motion. The
discussion here is of limited scope and for more details, the reader is encouraged to
consult any of the references listed at the end.
Our focus here is on fluid motion. We shall use rectangular Cartesian coordinates
( x, y, z ) , along with the associated basis set of mutually orthogonal unit vectors ( i , j, k ) .
The position vector is labeled x .
Imagine a tiny line element dx , labeled PQ in the sketch, at some instant of time. After
a small amount of time dt , the two ends have moved to new locations because of fluid
motion, and the new line element is labeled PQ .

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

and the (antisymmetric) vorticity tensor is given by


=

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 .

From Stokess theorem, we see

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

theorem, we can see that this is the circulation

v v t ds

where C is any closed curve

around the vortex tube, t is a unit tangent vector to the curve at any point, and ds is a
line element.

Rate of Strain or Rate of Deformation Tensor E


From the above discussion of the vorticity tensor, you can see that the role of that tensor
is to describe the instantaneous angular velocity of a fluid element, but that it contributes
nothing to deformation of elements. Now, we move on to discuss the significance of the
rate of strain tensor, which contains all the information about the deformation.
The rate of strain tensor is a symmetric tensor. We can write its components in terms of
the components of the velocity gradient as follows.

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

The diagonal elements of E


Consider a line element dx with a length ds .
d
d
ds 2 ) = ( dx dx )
(
dt
dt

Using the fact that

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

The second term in the far-right-side is zero because is an antisymmetric tensor. To


see this, we write

dx dx = ij dxi dx j = ji dx j dxi = ij dxi dx j so that ij dxi dx j = 0 .

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.

From the above, we can see that lim


V 0

The off-diagonal elements of E

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

Dividing both sides by ds ds yields

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.

Boundary Conditions in Fluid Mechanics


R. Shankar Subramanian
The governing equations for the velocity and pressure fields are partial differential equations that
are applicable at every point in a fluid that is being modeled as a continuum. When they are
integrated in any given situation, we can expect to see arbitrary functions or constants appear in
the solution. To evaluate these, we need additional statements about the velocity field and
possibly its gradient at the natural boundaries of the flow domain. Such statements are known as
boundary conditions. Usually, the specification of the pressure at one point in the system
suffices to establish the pressure fields so that we shall only discuss boundary conditions on the
velocity field here.

Conditions at a rigid boundary


It is convenient for the purpose of discussion to identify two types of boundaries. One is that at
the interface between a fluid and a rigid surface. At such a surface, we shall require that the
tangential component of the velocity of the fluid be the same as the tangential component of the
velocity of the surface, and similarly the normal component of the velocity of the fluid be the
same as the normal component of the velocity of the surface. The former is known as the no
slip boundary condition, and has been found to be successful in describing most practical
situations. It was a subject of controversy in the eighteenth and nineteenth centuries, and was
finally accepted because predictions based on assuming it were found to be consistent with
observations of macroscopic quantities such as the flow rate through a circular capillary under a
given pressure drop. If we designate the velocity of the rigid surface as V and that of the fluid as
v , and select a unit tangent vector to the surface as t , the no-slip boundary condition can be
stated as

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.

Conditions at a fluid-fluid interface


Sometimes, we encounter a boundary between two fluids. A common example occurs when a
liquid film flows down an inclined plane. The surface of the liquid film in contact with the
surrounding gas is a fluid-fluid interface. Other examples include the interface between a liquid
drop and the surrounding continuous phase or that between two liquid layers. It is convenient to
designate the two fluid phases in contact as phase I and phase II .

n
I
t

II
The unit normal vector n points into phase
vector to the interface t .

here and the sketch also shows a unit tangent

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.

n [TI TII ] = 2 H n s at a fluid-fluid interface


2

(jump condition on the stress)

In the stress boundary condition, the symbols

TI and TII represent the stress tensor in each

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

n ( n ) . That is , we remove the part of the gradient vector that

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 represents this stress component, we can write

t = t
I

at a fluid-fluid interface

(tangential stress balance)

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

at the free surface

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

at the free surface

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

at the free surface

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.

Elements of Prandtls Boundary Layer Theory


R. Shankar Subramanian
The failure of potential flow (incompressible irrotational flow) theory to
predict drag on objects when a fluid flows past them provided the impetus
for Prandtl to put forward a theory of the boundary layer adjacent to a rigid
surface. Prandtls principal assumptions are listed below.
Assumptions
1. When a fluid flows past an object at large values of the Reynolds number,
the flow region can be divided into two parts.
(i) Away from the surface of the object, viscous effects can be considered
negligible, and potential flow can be assumed.
(ii) In a thin region near the surface of the object, called the boundary layer,
viscous effects cannot be neglected, and are as important as inertia.
2. The pressure variation can be calculated from the potential flow solution
along the surface of the object, neglecting viscous effects altogether, and
assumed to be impressed upon the boundary layer.

U
y

x
laminar

turbulent

Transition from laminar to turbulent flow in the boundary layer on a flat


plate occurs at

Re x 5 105 , where Re x =

kinematic viscosity of the fluid.


1

xU

Here, is the

The assumptions can be used to establish the order of magnitude of the


boundary layer thickness.
A typical inertia term in the Navier-Stokes equation in rectangular Cartesian

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

Because the viscous force in the boundary layer is of comparable order to


the inertia force, these two order estimates must be comparable.

U 2
U
L

2 or 2
, which can be recast as
L

1
Re L

where the Reynolds number based on the length of the plate Re L =

LU

This type of argument is called a scaling analysis. It is a valuable tool in


dealing with transport problems. You can see that it provides not only an
idea of the variables on which key quantities depend, but also the form of
this dependence without having to solve the partial differential equations
involved.
In a like manner, we can find a scale estimate of the drag as well. The shear
stress at the plate surface is w =
this quantity as w ( x ) =

u
( x,0 ) . We can estimate the order of
y

. Because the shear stress is a local quantity,

we should use an order of magnitude of the variation of the boundary layer


thickness with x . From the order of magnitude argument used earlier, we

x
=
Re x

can estimate it as ( x )

. If the width of the plate in the

z direction is w , the drag on the plate surface is given by


L

D = w w ( x ) dx = w
0

U 3/ 2

1/ 2

dx
.
x

Ignoring the numerical factor of 2 that appears after performing the


integration (because we are only estimating the order of magnitude), we can
write

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

Order of magnitude analysis of the continuity and Navier-Stokes


Equations
Now, we shall go through an order of magnitude analysis of the twodimensional Navier-Stokes equations for steady incompressible Newtonian
laminar flow over a flat plate and simplify them using Prandtls ideas. For
more details, you can consult Schlichting [1].
We shall use scaled variables, using L as a reference length, and U as a
reference velocity. The symbols x and y are used for the scaled
counterparts of the physical coordinates in the sketch, and the symbols u
and v are used for the dimensionless counterparts of the physical velocity
components in the x and y directions, respectively.
The scaled
incompressible version of the continuity equation is

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

conclude that the derivative


equation, we see that

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

words, is the scaled boundary layer thickness. Because the 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

the boundary layer, in a distance of the order . Therefore, the second

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

( )

the thickness of the boundary layer p O 2 . This is very small, and


can be ignored, which is Prandtls assumption 2 listed on page 1. Because
the pressure change across the boundary layer is negligible, the pressure
distribution along the surface of the object, evaluated from the potential
flow, can be assumed to be impressed on the boundary layer. This means

that

p
x

in the

x component momentum equation is a known

inhomogeneity, and we can simply ignore the y component momentum


equation because all the terms are small.
Summarizing the above, we have found from the scaling analysis that the
viscous term in the main direction of flow ( x ) is negligible compared with
the viscous term in the direction normal to the solid surface. Furthermore,
the pressure gradient in the x component momentum equation is
established from potential flow theory and evaluated along the surface of the
object, and the y component momentum equation is neglected. Thus, we
have two equations for the two unknown velocity components.
Even though our analysis assumed a flat plate, you can see that for a thin
boundary layer, the effects of curvature of the surface would be negligible at
leading order. Therefore, as long as we define x and y as distance
coordinates along and normal to a surface, respectively, the same equations
can be written for flow past an object with a curved surface. For
convenience, Prandtls steady two-dimensional boundary layer equations for
incompressible Newtonian flow are written in physical variables below. To
avoid clutter, we have retained the same symbols for the velocities and
coordinates as those used earlier for scaled variables, but this should not be a
source of confusion.
Continuity

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

The boundary conditions are written as follows.

u ( 0, y ) = U
u ( x,0 ) = 0
v ( x,0 ) = 0

u ( x, y ) U

Commonly, the last condition is replaced with u ( x, ) = U .


Note that
1. The important nonlinear (inertial) terms have been retained.
2. The number of differential equations has been reduced from three to two,
consistent with the simplification that the pressure distribution is known
from potential flow theory.
3. Because the variation of pressure across the boundary layer is negligible
to this order of approximation, the potential flow pressure distribution can be
evaluated right at the solid surface and used as a known inhomogeneity in
the boundary layer equations.
Reference
1. H. Schlichting, Boundary Layer Theory, McGraw-Hill, New York, 1968.

Engineering Bernoulli Equation


R. Shankar Subramanian
The Engineering Bernoulli equation can be derived from the principle of
conservation of energy. The textbook provides such a derivation in detail.
Here, I have simply summarized the important forms of this equation for
your use in solving problems. Whenever you use this equation, be sure to
clearly mark the datum for measuring heights on a sketch. When setting a
term to zero, indicate the reason for doing so. For example, when the free
surface of the liquid in a tank is exposed to the atmosphere, or when it is
issuing as a free jet into the atmosphere, the pressure at that location is set
equal to zero gage. When liquid is taken out of a vessel through a pipe of
cross-sectional area that is small compared with that of the vessel, the
velocity of the free surface will be relatively small, and the kinetic energy
term associated with that velocity can be set equal to zero without much
error.

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

loss: losses per unit mass flowing


ws: shaft work done by fluid per unit mass flowing

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

= g : Specific weight of fluid


We can define the head developed by a pump as

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

Sometimes, we express loss as a certain number (N) of velocity


V2
.
heads. In this case, loss = N
2

Stokes Flow Invariant Representation of Solutions


R. Shankar Subramanian
The term invariant representation implies that the results are written in a
form that is independent of the coordinate system or, equivalently, the basis
set used for expressing the components of vectors.
In an early homework assignment, we learned about solutions of the Laplace
equation written in such invariant notation involving only the position vector
xi and its length r . These are called vector harmonics even though the
solutions are tensors, and are ordered such that at each level, the order
increases by 1. There are two sets of such solutions. One is the set of
decaying harmonics. The general decaying harmonic is given by

( 1)
1
( n+1) =

,



1 3 5 ( 2n 1) n times r
n

n = 0,1, 2,...

The first few are written below.

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

You can see that these solutions approach 0 as r . Growing harmonics,


in contrast, grow with distance from the origin. These are obtained from the
decaying harmonics as r 2 n +1 ( n+1) . The first few growing harmonics are given
below.
r2
r2
1, xi , xi x j ij , xi x j xk ( xi jk + x j ki + xk ij ) ,
3
5

Absolute and Pseudotensors


Well find these sets of harmonic functions very useful in constructing
solutions of Stokes problems. The important idea in each case will be that
well be looking for solutions of Laplaces equation in one case for a scalar
field (pressure) and in the other for a vector field (homogeneous solution for
the velocity). The pressure and velocity fields are entities that are called
absolute tensors. Absolute tensors are unaffected by whether the coordinate
system (and therefore the basis set) is right or left-handed. On the other
1

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

Also, an incompressible velocity field must satisfy u = 0 . If we now take


the divergence of the Stokes equation, ( 2 u p ) = 0 , which can be
rearranged to 2 ( u ) p = 0 . The first term is zero because the flow
is incompressible, leading to the result that 2 p = 0 . Therefore, the pressure
field in every Stokes problem must be a harmonic function, that is, a
function which satisfies the Laplace equation.
We can construct a solution of the Stokes equation by adding a
homogeneous solution and a particular solution. The homogeneous solution
u( H ) must also satisfy Laplaces equation 2 u(H ) = 0 . The particular solution
u( P ) , which satisfies the same equation as u , can be written as u (j P ) =

Let us verify that this is indeed correct.


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

Now, take the divergence.


( u ( P ) ) =

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

satisfies Laplaces equation


= 0 . We see that the particular solution
xi xi

is indeed correct. Note that we still must satisfy u = 0 . Therefore,


x
1
u( H ) = u( P ) =
p =
[3 p + x p ]
2
2
x
x
p
p
because
p =
( xi p ) = i p + xi = ii p + xi = 3 p + x p .
xi
xi
xi
2 xi

Constructing solutions using the invariant representation


Now, we are ready to construct solutions of Stokes problems using invariant
representations of the pressure and velocity fields. To do this, we use the
following ideas.
1. The pressure field is an absolute scalar, and the velocity field is an
absolute vector.
2. The pressure field and the homogeneous part of the velocity field are
harmonic functions. If the flow is exterior to a body and the fluid is
quiescent far from a body, we can use the decaying harmonics. If the flow is
interior to an object, such as a drop, the solution must be finite at the origin
which is usually chosen to lie within the object. In this case, we can use
growing harmonics.
3. If the object is moving with a velocity U i or rotating with an angular
velocity i , because of the linearity of the Stokes equation, the fields must
also be linear in these vectors.
4. The pressure field and the velocity field can only depend on the position
vector and the scalars and vectors arising from the boundary conditions and
the shape of the object.

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.

Notes on the Solution of Stokess Equation for Axisymmetric


Flow in Spherical Polar Coordinates
R. Shankar Subramanian
The details of the solution of Stokess Equation for the streamfunction
E 4 = 0 in spherical polar coordinates ( r , , ) by the method of separation
of variables are given in the book Low Reynolds Number Hydrodynamics
by J. Happel and H. Brenner. The solution is obtained as the sum

( ) ( r , ) + (
1

2)

( r , ) where (1) satisfies


E 2 ( ) = 0
1

and

( 2)

(1)

is the particular solution of

E 2 ( ) = ( )
2

(2)

The solution, in the form of infinite series, is specialized to typical problems


in the spherical geometry in a domain that includes at least a portion of the
symmetry axis by requiring that the velocity field be bounded along that
axis, which corresponds to = 0, . The result is given below.

( r , ) = ( An r n + Bn r n+1 + Cn r n+2 + Dn r n+3 ) Cn1/ 2 ( )

(3)

n =2

Here, = cos , and Cn1/ 2 ( ) are the Gegenbauer Polynomials of order n


and degree 1/ 2 . Their properties are discussed in Sampson (1891), and in
Abramowitz and Stegun (1965). The Gegenbauer Polynomials are solutions
of the second order ordinary differential equation

d 2
(1 ) d 2 + n ( n 1) = 0
2

(4)

where n is an integer. The second linearly independent solution of this


equation is a function that becomes unbounded along the symmetry axis.
The Gegenbauer Polynomials are closely related to the Legendre
Polynomials Pn ( ) , which satisfy Legendres equation.

d
d
1 2 )
+ n ( n + 1) = 0
(

d
d

(5)

Many important properties of the Legendre Polynomials can be found in


MacRobert (1967) and in Abramowitz and Stegun (1965).
Some useful relationships involving the Gegenbauer Polynomials are given
below.

Pn2 ( ) Pn ( )
2n 1

Cn1/ 2 ( ) = Pn1 ( s ) ds

Cn1/ 2 ( ) =

(6)
(7)

The first few Gegenbauer and Legendre Polynomials are

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 ( ) =

The Gegenbauer Polynomials satisfy the orthogonality property

+1

Cm1/ 2 ( ) Cn1/ 2 ( )

d =

for values of m, n 2.

n ( n 1)( 2n 1)

mn

(8)

The function Qn ( ) used by Leal in the textbook is related to the

Gegenbauer Polynomials through Qn ( ) = Cn+1/1 2 ( ) .

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

As we already know, a skew-symmetric tensor Aij can be formed from a vector ak by


writing Aij = ijk ak . The vector associated with the vorticity tensor in this manner is
1
1
. Therefore, ij dx j = ijk dx j k or in Gibbs notation,
2
2
1
1
dx = dx = dx
2
2
This means that the relative motion at a point an infinitesimal distance away from a
reference point in a fluid that is contributed by the vorticity tensor is equivalent to that
1
caused by a rigid 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, 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 .

From Stokess theorem, we see

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 .

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 = u are ( x , y , z ) in a basis set obtained from
rectangular Cartesian coordinates, 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

theorem, we can see that this is the circulation

v u t ds where

C is any closed curve

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.

Viscous Dissipation term


When the dot product of each term in the Navier-Stokes equation is taken with the
velocity vector u , the result can be cast in the form of an equation for the time rate of
change of the kinetic energy of the fluid per unit volume. An important term that appears
in the result for this quantity is the rate at which the work done against viscous forces is
irreversibly converted into internal energy. This is known as viscous dissipation. The
viscous dissipation per unit volume is written as : u = v where v for a
Newtonian fluid is given below in different coordinate systems.

Rectangular Cartesian Coordinates ( x, y, z )

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

Cylindrical Polar Coordinates ( r , , z )

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

Spherical Polar Coordinates ( r , , )


2
2
u 2 1 u

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

Solution of Partial Differential Equations


Combination of Variables
R. Shankar Subramanian
Introduction and Problem Statement
We encounter partial differential equations routinely in transport phenomena. Some
examples are unsteady flow in a channel, steady heat transfer to a fluid flowing through a
pipe, and mass transport to a falling liquid film. Here, we shall learn a method for
solving partial differential equations that complements the technique of separation of
variables. We shall also learn when the method can be used. We consider the same
model problem, namely the motion induced in fluid contained between two long and
wide parallel plates placed with a distance b between them as shown in the sketch below.

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

Using separation of variables, we obtained a solution of these equations that can be


written as follows.

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)

Note that to be precise, we must write Equation (6) as vx ( t , y ) 0 , and the


equation must be read to imply only such a meaning.
A Speculation

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

new variable = y / ( t ) , can we expect the velocity field vx ( t , y ) to become a function


U ( ) that depends only on the single new variable? This speculation is shown in the

sketch below.

The transformation to is known as a similarity transformation and the variable is


termed a similarity variable.
Solution by Combination of Variables

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)

The fact that a quiescent condition is approached as y , described by Equation (6),


becomes

() = 0

(16)

The initial condition, given in Equation (2), transforms to

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)

where erfc means complementary error function. This function is defined as


follows.
erfc ( ) = 1 erf ( )

(21)

where the error function erf is defined as

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

sudden movement initiated at the boundary.


Summary

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

The problem of unsteady one-dimensional heat conduction in a semi-infinite solid slab


(or a quiescent liquid layer) in the y direction, when the temperature at the surface
y = 0 is changed to a new value at time zero, is described by the same governing
equations and boundary conditions. The assumptions are that there are no sources or
sinks, heat transport occurs only by conduction with a constant thermal conductivity, the
density and specific heat of the material are constant, and that the slab is very long and
very wide so that end effects and edge effects can be neglected. By analogy, it can be
seen that the same equations also describe unsteady diffusion in a similar situation. All
of these cases can be handled by the same solution method. Note that unlike separation
of variables, combination of variables does not require the system of governing equation
and boundary conditions to be linear. This method has used successfully in solving the
Navier-Stokes equations including inertia (and therefore non-linear) in forced boundary
layer flows, and also in solving problems of natural convection in boundary layers
wherein the fluid mechanics and heat transport problems lead to coupled non-linear
governing equations.
Reference

1. M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions, Dover, New


York, 1965.

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:

Density of the liquid


viscosity of the liquid
Diameter of the tank
Diameter of the agitator
acceleration due to gravity
Rotation speed
Power

By using dimensional analysis, we can show that the Power number

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

can be important in some situations where the

objective is to emulsify a mixture; in that case,


two liquid phases.

is the interfacial tension between the

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

Assuming we use the same fluid,

Now, consider the equality

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

So, we must simultaneously satisfy


2
N m D Ap
=
=
N p DA2m

DAp
DAm

. The only way to do this is to set

DAm = DAp .

This means that

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.

Power Requirement for a Mixer


If we restrict attention only to non-vortex forming systems, the data for a given type of
mixer can be fitted to

N P = cRe a
This can be recast as

log N P = c + a log Re
where the new constant

c = log c .

If the flow in the mixer is laminar, it is found that

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.

Plate Frame Heat Exchanger

Plate Frame Heat Exchanger

Exchanger Matrix

PIPE FLOW

Re =

DV

Laminar, Transition, Turbulent

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

Friction Factor Laminar Flow

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

Friction Factor - Turbulent Flow

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

Surface roughness (, ft)


0.000005
0.00015
0.0004
0.0005
0.00085
0.001 0.01
0.003 0.03

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

From the chart

f = 0.00475

Using the equation:

10 6
f = 0.0013751 + 20,000(0.0002 ) + 5

10

f = 0.0047

Calculate:
Head Loss

Q=

BHP of the Pump

2000 gal / min


= 4.46 ft 3 / s
448.8( gal / min/ ft 3 / sec)

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

For cast iron =0.00085 ft


/D = 0.00085/0.667 = 0.0013

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

KL is called the loss coefficient


Flow which is dominated by inertia
rather than viscous effects,
pressure drop and head losses
correlate directly with the
dynamic pressure.

V2
2g

KL = (geometry, Re)
However for most practical cases:

K L p = 1 / 2 V 2

KL = (geometry)
Dynamic pressure

Minor losses are sometimes given in terms of an equivalent length leq


Head loss through a component is expressed in terms of equivalent
length that would produce the same head loss as the component.

leq V 2
D 2g

hL = f

D and f are based on the pipe containing the


component.

leq =

KLD
f

Entrance flow conditions Re-entrant KL=0.8

Sharp edge KL = 0.5

Slightly rounded entrance KL = 0.2


Well rounded KL = 0.04

Vena contracta

Flow separation

V2>V3

Ideal full recovery of kinetic energy

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

KL1 = 0.25 rounded entrance


KL2 =0.9 elbow

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

Between sections 1 and 2

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

Volumetric flow rate Q = (D2/4) x 122= (0.156)2/4 x 122=2.33 m3/s

10

Vous aimerez peut-être aussi