This action might not be possible to undo. Are you sure you want to continue?
NonNewtonian 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.
Any fluid that does not obey the Newtonian relationship between the shear
stress and shear rate is called nonNewtonian. 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
nonNewtonian. 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 shearthinning. In the opposite case
where the viscosity increases as the fluid is subjected to a higher shear rate, the
fluid is called shearthickening. Shearthinning behavior is more common than
shearthickening. Shearthinning fluids also are called pseudoplastic fluids. A
typical shear stress versus shear rate plot for a shearthinning fluid looks like
this.
Shear Rate
S
h
e
a
r
S
t
r
e
s
s
shearthinning fluid
We describe the relationship between the shear stress τ and shear rate θ
as
follows.
1
τ ηθ =
Here, η is called the “apparent viscosity” of the fluid, and is a function of the
shear rate. In the above example, a plot of η as a function of the shear rate θ
looks like this.
Shear Rate
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
shearthinning fluid
Many shearthinning 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 Shear Rate
l
o
g
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
Newtonian Region
PowerLaw Region
Newtonian Region
shearthinning fluid
2
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 powerlaw
region. In this region, we can approximate the behavior by
log log a b η θ = +
which can be rewritten as
b
K η θ =
where
( )
exp K a = . Instead of we commonly use b
( )
1 n −
Examples of shearthinning fluids are polymer melts such as molten polystyrene,
can see that when paint is sheared with a brush, it flow
for the exponent
nd write a result for the apparent viscosity as follows. a
1 n
K η θ
−
=
Upon using the connection among the shear stress, apparent viscosity, and the
shear rate we get the powerlaw model.
n
K τ θ =
where n is called the powerlaw index. Note that 1 n =
nia
2
n 1/ 3
1/
are shown next.
corresponds to
ewto n behavior. Typically, for shear thinning fluids, lies between
and , even though other values are possible.
polymer solutions such as polyethylene oxide in water, and some paints. You
s comfortably, but when
e 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
urface. The behavior of paint is a bit more complex than this, because the
ome slurries and pastes exhibit an increase in apparent viscosity as the shear
rate is increased. They are called shearthickening or dilatant fluids. Typical
lots of shear stress versus shear rate and apparent viscosity versus shear rate
N
th
s
viscosity changes with time at a given shear rate.
S
p
3
S
h
e
a
r
S
t
r
e
s
s
shearthickening fluid
Shear Rate
Shear Rate
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
shearthickening fluid
Some examples of shearthickening fluids are corn starch, clay slurries, and
solutions of certain surfactants. Most shearthickening fluids tend to show
shearthinning at very low shear rates.
Another important type of nonNewtonian 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 τ
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.
4
Shear Rate
S
h
e
a
r
S
t
r
e
s
s
Bingham Plastic
viscoplastic fluids
so, some paints exhibit a yield stress.
Another important class of fluids exhibits viscoelastic behavior. This means that
uids (viscous). Viscoelastic
ting shaft, swelling
example of a common viscoelastic liquid is
f a container, you
it back into the container. Several
d solutions are viscoelastic. You can
uid mechanics
Examples of viscoplastic fluids are drilling mud, nuclear fuel slurries,
mayonnaise, toothpaste, and blood. Al
Of course, this is not an exhaustive discussion of nonNewtonian behaviors. For
instance, some classes of fluids exhibit timedependent 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. Nondrip 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.
these fluids behave both as solids (elastic) and fl
fluids exhibit strange phenomena such as climbing up a rota
when extruded out of a dye, etc. An
eggwhite. You have probably noticed that when it flows out o
can use a quick jerking motion to snap
industrially important polymer melts an
learn more about these and other behaviors from several fl
textbooks and from Perry’s Chemical Engineers’ Handbook.
5
1
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 crosssection 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
fullydeveloped flow.
A momentum balance can be written for a control volume called a shell, which is
constructed by translating a differential crosssectional 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 crosssectional 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 cross
section can be written as well. In the case of pipe flow, we shall see how this yields the
wellknown HagenPoiseuille 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 fluidfluid
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 nonzero
thickness, we treat is as a plane of zero thickness for this purpose. This means that the
2
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 FullyDeveloped Incompressible Laminar Flow in a Straight Circular Tube
Cylindrical polar coordinates
( )
, , r z φ are used. We assume steady incompressible
fullydeveloped laminar flow in the zdirection. The only nonzero velocity component
is
z
v and it only depends on the radial coordinate r . Therefore,
( )
z z
v v r = .
Similarly, the shear stress
( )
rz rz
r τ τ = . The shell shown in the figure is L units long
and 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 + ∆ .
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.
Flow
Shell
r
z
r ∆
L
θ
R
3
Now, the volumetric flow rate of fluid into the shell is
2
z
Q r r v π ∆ = ∆
If the density of the fluid is ρ , the associated mass flow rate is
2
z
m r r v π ρ ∆ = ∆
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
2
2
z z
mv r r v π ρ = ∆ = ∆
Because this is independent of z , the rate of efflux of momentum from the face on the
right side is exactly the same.
Rate of efflux of momentum
2
2
z
r r v π ρ = ∆
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 0 r r p p L π = ∆ −
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 2 cos r rL g π ρ θ = ∆
where 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 τ + ∆ . 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
rz
r r L r r π τ + ∆ + ∆ in the z −direction. In the same way, the fluid within the
4
shell at the location r exerts a stress
( )
rz
r τ 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
rz
rL r π τ − . 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
rz rz
L r r r r r r π τ τ = + ∆ + ∆ −
Add the forces, set the sum to zero, and divide by 2 rL π∆ to obtain the following result.
( ) ( ) ( ) ( ) ( )
0
cos 0
rz rz
p p L r r r r r r
r g
L r
τ τ
ρ θ
− + ∆ + ∆ −  
+ + =

∆
\ ¹
Now we take the limit of the terms in the above equation as 0 r ∆ → . The first term is
unaffected while the second term is simply the derivative of
rz
rτ with respect to r . This
leads us to the first order differential equation for the shear stress
( )
rz
r τ .
( )
1
rz
d P
r
r dr L
τ
∆
= −
where we have introduced the symbol
( ) ( )
0 cos P p p L gL ρ θ ∆ = − +
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
rz
rτ . The result of
this integration is
2
1
2
rz
P r
r C
L
τ
∆
= − +
where
1
C is an arbitrary constant of integration that needs to be determined. We can
rewrite this result as
5
1
2
rz
C P r
L r
τ
∆
= − +
According to this result, the shear stress should approach infinity as the radial coordinate
0 r → . This is not physically possible. Therefore, we must choose the arbitrary
constant of integration
1
0 C = and write the shear stress distribution in the tube as
follows.
2
rz
P
r
L
τ
∆
= −
The shear stress is zero at the centerline and its magnitude is a maximum at the tube wall.
It varies linearly across the crosssection. 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.
To find the velocity distribution, we need to introduce a relationship between the shear
stress and the velocity gradient. Recall the Newtonian constitutive model
z
rz
dv
dr
τ µ =
r
z
R
( )
rz
r τ
6
where µ is the dynamic viscosity of the fluid. Use of this constitutive model in the
result for the shear stress leads to
2
z
dv P
r
dr L µ
∆
== −
This equation can be integrated immediately to yield
( )
2
2
4
z
P
v r C r
L µ
∆
= −
where a new arbitrary constant of integration
2
C 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 = .
( )
0
z
v R =
Use of this condition leads to the result
2
2
4
PR
C
L µ
∆
=
so that we can write the velocity distribution as follows.
( )
2 2
2
1
4
z
PR r
v r
L R µ
∆
= −
This velocity distribution describes a parabola. It is sketched below.
r
z
R
( )
z
v r
7
We see that the maximum velocity occurs at the centerline, which is at 0 r = . Its value
is
2
max
4
PR
v
L µ
∆
=
We can calculate the volumetric flow rate through the tube by multiplying the velocity at
a given r by the differential crosssectional area 2 r dr π to yield dQ and then
integrating across the crosssection.
( )
2 2
2
0 0 0
4
2 1
2
8
Q R R
z
PR r
Q dQ r v r dr r dr
L R
PR
L
π
π
µ
π
µ
  ∆
= = = −

\ ¹
∆
=
∫ ∫ ∫
This is a result known as the HagenPoiseuille 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 crosssection.
It can be seen that in a circular tube, the average velocity is onehalf the maximum
velocity.
Steady FullyDeveloped 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
( )
2 2
2
1
2
z
Ph y
v y
L h µ
∆
= −
This also is a parabolic velocity profile that is symmetric about the centerline. The
maximum velocity occurs at the centerline, 0 y = .
2
avg 2
8
Q PR
v
R L π µ
∆
= =
8
2
max
2
Ph
v
L µ
∆
=
Let the plates be W units wide in the x −direction. Then, the differential cross
sectional area for flow is W dy and the differential volumetric flow rate would be
( )
z
dQ v y W dy = . This can be integrated across the crosssection as follows.
( ) ( )
0
2 2
2
0
2
2 1
2
h h
z z
h
h
Q W v y dy W v y dy
Ph y
W dy
L h µ
−
= =
  ∆
= −

\ ¹
∫ ∫
∫
yielding
3
2
3
PWh
Q
L µ
∆
=
The average velocity is obtained by dividing Q by the crosssectional area 2Wh.
2
avg
3
Ph
v
L µ
∆
=
We can see that the ratio of the average to the maximum is 2/ 3 for steady laminar flow
between wide parallel plates.
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 0 y = ,
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 0 y = . This is why the velocity profile
in a falling liquid film in laminar flow is represented by onehalf 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
9
3
3
PWh
Q
L µ
∆
=
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
( ) ( )
0 p p L = . Therefore, we can write
cos
P
g
L
ρ θ
∆
=
which yields
3
cos
3
gWh
Q
ρ θ
µ
=
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 onedimensional
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 crosssection to obtain the volumetric flow
rate. The average velocity can be obtained as the volumetric flow rate divided by
the crosssectional area.
The shell balance approach is limited to onedimensional flow situations and
straight geometries. In the more general case, we must use partial differential
10
equations obtained using the same principles, known as the NavierStokes
equations, and simplify them for the specific problem under consideration.
Elementary Aspects of TwoPhase Flow in Pipes
R. Shankar Subramanian
Chemical engineers frequently encounter the flow of a mixture of two fluids in pipes. Liquidgas
or liquidvapor mixtures are encountered in condensers and evaporators, gasliquid 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. Liquidliquid mixtures are encountered when dealing with emulsions as well as
in liquidliquid extraction.
Twophase 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,
we’ll 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 gasliquid twophase 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 twophase flow, the concept of holdup 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 twophase flow, Perry’s 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 gasliquid twophase flow. Here, we only consider the qualitative
features of gasliquid flow in a horizontal pipe to give you an appreciation of the complexity of
twophase flow when compared with the flow of a single fluid phase.
The sketches on the following pages depict various regimes of cocurrent twophase flow in a
horizontal pipe through which a gasliquid mixture is flowing. We use V and V to designate
the superficial velocity of the liquid and gas phases, respectively. We first consider a situation
where the ratio V is large, and then the regime where the ratio is lower.
L G
/
L
V
G
1
Consider a mixture that is almost completely liquid. This means that the ratio V 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 V leads to annular mist flow through the
intermediate stage of slug flow; these new terms are defined later.
/
L
V
G
G
G
High V
L
/V
G
As V
G
Plug Flow
Bubble Flow
Lower V
L
/V
G
Stratified Flow
Gas
Liquid
Wavy Stratified Flow
Slug Flow
Gas
Liquid
Further increase in V
G
leads to higher waves
Gas
Liquid
As V
G
Now, consider lower values of the ratio V . 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 V is increased, waves begin to appear on the surface of the liquid,
/
L
V
G
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.
Annular mist flow
Gas
References
F.A. Holland and R. Bragg, Fluid Flow for Chemical Engineers. 1995 Edward Arnold
Publishers, London.
Perry’s Chemical Engineers Handbook, 7
th
Edition. 1997 (Ed: R.H. Perry, D.W. Green, and J.O.
Maloney), McGrawHill, New York.
3
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 θ in which represents the direction of flow, and and z r θ
stand for the radial and angular directions, respectively.
z
R
r
P
0
P
L
We make the following simplifying assumptions.
1. Incompressible flow: Continuity reduces to 0 ∇• = v
2. Newtonian flow with constant viscosity µ
3. Neglect end effects:
z
∂
=
∂
0
v
(fully developed flow)
4. Symmetry in θ (known as axial symmetry): ; 0 v
θ
θ
∂
= =
∂
0
v
The incompressible version of the equation of continuity, in conjunction with
assumptions 3 and 4, yields
( )
0
r
rv
r
∂
=
∂
(1)
Therefore, must be constant across the crosssection of the tube. Because it is zero
at , it must be zero everywhere. This implies that
r
rv
0 r = 0
r
v = . Therefore, the flow is
unidirectional and is the only nonzero velocity component.
z
v
The NavierStokes equation, subject to assumptions 1 and 2, can be written in component
form in cylindrical polar coordinates. The components in the r and θ directions yield
the result that the dynamic pressure P is uniform in those directions. Therefore, the
1
dynamic pressure can, at best, depend only on and . The component in the 
direction yields the following governing equation for the velocity
t z z
( )
,
z
v t r in the tube.
z
v v P
r
t z r r
µ
ρ
∂ ∂ ∂ ∂
⎛ ⎞
= − +
⎜
∂ ∂ ∂ ∂
⎝ ⎠
z
r
⎟
(2)
By rearranging this equation as
z
v v P
r
z r r r t
µ
ρ
z
∂ ∂ ∂ ∂
⎛ ⎞
= −
⎜ ⎟
∂ ∂ ∂
⎝ ⎠
∂
(3)
and recognizing that the left side can only depend on t and , 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
replace
z
P
z
∂
∂
by its steady representation
0 L
P P P
L L
− ∆
= − where we have defined
.
0 L
P P P ∆ = −
Therefore, Equation (2) can be rewritten as
z
v v P
r
t L r r r
µ
ρ
∂ ∂ ∆ ∂
⎛ ⎞
= +
⎜
∂ ∂
⎝ ⎠
z
⎟
∂
(4)
The initial and boundary conditions are written as follows.
Initial Condition:
( )
0, 0
z
v r = (5)
Boundary Conditions:
( )
, 0 is finite
z
v t (6)
(No Slip)
( )
,
z
v t R 0 = (7)
Now, we proceed to scale this problem by defining the following dimensionless
variables.
( )
( )
2
2
; ; ,
/ 4
z
v t r
T y V T y
R R
PR L
ν
µ
= = =
∆
In the above, ν is the kinematic viscosity
( )
/ µ ρ . You may wonder about the choice
of the scale for the velocity, and also about the factor that is introduced in that scale.
The velocity scale is obtained by considering the steady problem. In that problem, the
4
2
characteristic velocity is of the order
( )
2
/ PR L µ ∆ . The factor is used to make the
steady solution appear especially simple, as you will see a bit later.
4
Nondimensionalization leads to
1
4
V V
y
T y y
⎛ ⎞ ∂ ∂
= +
⎜
∂ ∂
⎝ ⎠
y
∂
⎟
∂
(8)
The scaled velocity field satisfies the following initial and boundary conditions.
Initial Condition:
(0, ) 0 V y = (9)
Boundary Conditions:
( )
, 0 is finite V T (10)
No Slip:
( ,1) 0 V T = (11)
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.
( ) ( ) ( )
,
s t
V T y V y V T y , = + (12)
The steady solution must satisfy Equation (8) without the time derivative.
1
4
s
dV d
y
y dy dy
⎛ ⎞
0 + =
⎜ ⎟
⎝ ⎠
(13)
along with the boundary conditions
( )
0 is finite
s
V (14)
and
(1) 0
s
V = (15)
3
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.
1
t
V V
y
T y y y
⎛ ⎞ ∂ ∂ ∂
=
⎜
t
⎟
∂ ∂ ∂
⎝ ⎠
(16)
( )
(0, )
t
V y V y
s
= − (17)
( )
, 0 is finite
t
V T (18)
( ,1) 0
t
V T = (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 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
nontrivial solution for the transient problem. We can obtain the steady field by
integrating Equation (13) and applying the boundary conditions. This leads to
y
( )
2
1
s
V y y = − (20)
The solution of Equations (16) through (19) is accomplished by separation of variables.
If we substitute a trial solution of the form
( ) ( ) ( )
,
t
V T y G T y φ = into Equation (16)
we find that the functions
( )
G T and
( )
y φ must satisfy the following differential
equations.
2
0
dG
G
dT
λ + = (21)
2
1
0
d d
y
y dy dy
φ
λ φ
⎛ ⎞
+ =
⎜ ⎟
⎝ ⎠
(22)
The solution of Equation (21) is
2
T
G Ce
λ −
= (23)
4
where 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.
C
( ) ( )
1 0 2 0
C J y C Y y φ λ = + λ
)
(24)
where and 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
0
J
0
Y
(
p
J y α and
(
p
Y y
)
α of integer order are the two linearly independent solutions of Bessel’s
equation
p
( )
2 2
1
0
d d
y p
y dy dy
φ
α φ
⎛ ⎞
+ − =
⎜ ⎟
⎝ ⎠
(25)
Here are sample graphs showing the behavior of the functions
0
( ) J x and
1
( ) J x ,
followed by
( )
0
Y x and .
( )
1
Y x
5
For our purposes, it is important only to note that the functions
( )
p
Y x are singular at
0 x = , that is, they approach minus infinity as x approaches zero. This is not
compatible with the boundary condition stated in Equation (18) that remain finite at
t
V
0 y = . Therefore, the constant in Equation (24) must be set equal to zero. This
leaves us with a product class solution
2
C
( ) ( )
2
0
,
T
t
V T y Be J y
λ
λ
−
= (26)
where the product of the two arbitrary constants C and is written as a new constant
1
C
B. We know that this solution satisfies the governing differential equation for
( )
,
t
V T y and the condition that the solution must remain finite at . Now, we
need to apply the remaining conditions. The noslip boundary condition at the wall that
leads to Equation (19) can be satisfied by requiring
0 y =
( )
0
0 J λ = (27)
As you can see from the sketch showing the behavior of
0
J , 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
( ) (
0 0
J x J x
)
− = so that we
can simply use the infinite set of positive roots, labeled
( )
1, 2, 3...
n
n λ = . The first 20
roots of Equation (27), labeled
0,s
j may be found in Table 9.5, page 409, in Abramowitz
6
and Stegun (1965). You’ll see that
0,20 0,19
j j π − ≈ . Therefore, additional roots can
be generated by approximating the interval between each by π .
Corresponding to each root of Equation (27), the product form appearing in Equation (26)
with a multiplicative constant will satisfy the governing equation and the two
boundary conditions in the 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
n
B
y
(28)
( ) (
2
0
1
,
n
T
t n
n
V T y B e J y
λ
λ
∞
−
=
=
∑
)
n
)
n
All that remains is the determination of the constants . We can obtain them by
applying the initial condition.
n
B
( ) ( ) (
0
1
0,
t s n
n
V y V y B J y λ
∞
=
= − =
∑
(29)
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
0 0
0
0,
n m
y J y J y dy m n λ λ = ≠
∫
(30)
( )
( )
( )
2
1
1
2
0 0
0
provided 0
2
n
n n
J
y J y dy J
λ
λ λ
⎡ ⎤
⎣ ⎦
= =
∫
(31)
( ) (
1
p p
p p
d
y J y y J y
dy
)
λ λ
−
⎡ ⎤ =
⎣ ⎦
λ (32)
( ) ( ) ( )
1 1
2
p p p
p
J x J x J x
x
− +
+ = (33)
The procedure is as follows. Begin with Equation (29) in the form
( ) (
0
1
s n
n
V y B J y
)
n
λ
∞
=
− =
∑
(34)
7
Multiply both sides by
(
0 m
yJ y
)
λ where is some integer, and integrate with respect
to from
m
y 0 y = to 1. Use Equations (30) and (31) to obtain
( )
( ) ( )
1
0 2
0
1
2
m s
m
m
B y V y J y dy
J
λ
λ
= −
⎡ ⎤
⎣ ⎦
∫
(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, you’ll need to
use integration by parts in conjunction with Equation (32). The final result, after
simplification using Equation (33), is
( )
3
1
8
m
m m
B
J λ λ
= − (36)
Substitution of this result after replacing the index with permits us to write the
complete solution from Equation (12) as follows.
m n
( )
( )
( )
2
0 2
3
1
1
, 1 8
n
n T
n
n n
J y
V T y y e
J
λ
λ
λ λ
∞
−
=
= − −
∑
(37)
The velocity field, calculated from Equation (37) at a few selected values of T, is
displayed below.
8
References
F.B. Hildebrand, Advanced Calculus for Applications, PrenticeHall, Englewood Cliffs,
1976.
M. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, New
York, 1965.
9
Intr oduction to Integr al Balances and Conser vation of Mass
R. Shankar Subr amanian
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 incr ease of mass of mater ial within the contr ol volume = Net r ate at which
mater ial enter s the contr ol volume.
An arbitrary control volume V is shown in the sketch.
dA
dV
V
A
n
1
We also have marked the bounding surface of this control volume, called the control
surface (CS) and shown an element of surface area and the unit outward normal
(vector) to that area element, . 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
A
dA
n
M and the net rate of entry of mass into
the control volume as , conservation of mass can be written as m
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 ρ . By adding up
all the differential volumes within the total volume V , we can obtain the total mass of
fluid M . Therefore, we can write
V
M dV ρ =
∫
(2)
The time rate of change of this mass is then
dM
dt
. Therefore,
V
dM d
dV
dt dt
ρ =
∫
(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 . If the
velocity vector is V , the component of this velocity that is directed into the control
volume is given by , because the unit normal vector n points outwar d 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 .
dA
− • V n
dQ
dQ dA = − • V n (4)
The corresponding rate at which mass enters the control volume through the area element
, labeled , can be written as dA dm
( ) dm dQ dA ρ ρ = = − • V n (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.
2
(6) ( )
CS
m d ρ = − •
∫
V n A
Now, we can rewrite the principle of conservation of mass, given in Equation (1) as
( )
CS
dM
dA
dt
ρ = − •
∫
V n (7)
Rate of Rate of net inflow of mass
increase of into the control volume
mass M in
the control volume
where we have identified the physical meaning of the terms in the left and right sides of
the equation.
In a steady state situation, the time rate of change of the mass of material in a control
volume is zero. In this case, we simply obtain
( ) 0
CS
dA ρ • =
∫
V n (8)
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 crosssection as shown
in the sketch.
n n
1 1 1
, , A V ρ
2 2 2
, , A V ρ
3
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, 0 • =
1
V
V n . 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, becomes − . In a like manner, at the exit surface, the velocity
points in the same direction as the normal vector which leads to
• V n
• V becoming V .
Because and are constant across the surfaces involved, the integrals are easy to
evaluate and we get
n
2
1 2
V V
1 1 1 2 2 2
0 V A V A ρ ρ − + =
2
(9)
or
1 1 1 2 2
V A V A ρ ρ = (10)
The product of the uniform velocity and the area is the volumetric flow rate .
Therefore, we can rewrite this as
Q
1 1 2
Q
2
Q ρ ρ = (11)
The mass flow rate Q m ρ = so that
2 1
m m m = = is constant. If the density is
constant, the above relationship reduces to
1 2
Q Q = . 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.
4
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
2
2
V
g
. Therefore, we can write minor
losses as
2
2
m L
V
h K
g
= . where
L
K is called the loss coefficient.
Typical values of
L
K 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 45
o
Elbow: 0.2  0.4
Long radius 45
o
Elbow: 0.14  0.21
Standard 90
o
Elbow: 0.21  2.0
Long radius 90
o
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
L
K is given by
2
2
2
1
L
d
K
D
= −
Here, and represent the diameters of the smaller and larger pipes, respectively. For a d D
1
sudden contraction, we can use the same result if 76 . 0 ≥
D
d
. For smaller values of ,
D
d
we
can use the empirical relation
− =
2
2
1 42 . 0
D
d
K
L
L
.
K In both cases, we should multiply by the velocity head in the pipe segment of diameter d .
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 , 0 =
D
d
yielding . 1
L
K =
2
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.
Re
e
d Vρ
u
=
2 2
2
D
D
C
a V πρ
=
Here, V is the velocity of the disk, ρ is the density of the fluid, and u is its viscosity.
The symbol stands for the drag on the sphere. The radius of the disk is and its
thickness is
D a
H , leading to the following result for the equivalent diameter .
e
d
( )
1
2
3
6
e
d a H =
Drag Coefficient
For , Re 0.01 ≤
64 Re
1
Re 2
D
C
π π
 
= +

\ .
For , 0.01 Re 1.5 < ≤
( )
64
1 10
Re
x
D
C
π
= +
where
2
10 10
0.883 0.906log Re 0.025(log Re) x = − + −
For 1.5 , Re 133 < ≤
0.792
64
1 0.138Re
Re
D
C
π
= +
Clift, Grace, and Weber state that when vortex shedding begins,
D
C becomes insensitive
to , and is constant at a value of 1.17 for , and that there is some Re Re 1, 000 >
1
indication that
D
C passes through a minimum of about 1.03for . But the
authors go on to state that most data are correlated within 10% by the result valid for
, with C for .
Re 400
T
1.5 Re < ≤
0.1
Re 100
T
>
133
Re
T
<
1.17
D
=
T
100 <
Re 133 >
1
2
3
Re
D T
C
1
2
3
Re
T
C
1
2
3
Re
D T
Terminal Settling Velocity
From page 148149, we learn that a disk in free motion will move steadily with its axis
vertical so long as (meaning the Reynolds number at terminal settling velocity)
satisfies . 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
, even though the steady motion persists to higher up to about 172
when the moment of inertia of the disk is relatively small. It appears that a parameter
Re
Re
is used to characterize transition to various regimes of unsteady motion of
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
D
, the amplitude
of the oscillation increases and the disk moves in a succession of curved arcs, a motion
called glidetumble. At even higher C
, 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.
2
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,
D
C , on the Reynolds number, , 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
Re
p
d Vρ
µ
=
2 2
8
D
p
D
C
d V πρ
=
Here, is the diameter of the sphere, V is its velocity,
p
d ρ is the density of the fluid,
and µ is its viscosity. The symbol D stands for the drag on the sphere.
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,
10
log Re w = .
For , Re 0.01 ≤
9 24
2 Re
D
C = +
For , 0.01 Re 20 < ≤
(0.82 0.05
24
1 0.1315Re
Re
w
D
C
− )
⎡ ⎤
= +
⎣ ⎦
For , 20 Re 260 ≤ ≤
0.6305
24
1 0.1935Re
Re
D
C ⎡ ⎤ = +
⎣ ⎦
For ,
3
260 Re 1.5 10 ≤ ≤ ×
2
10
log 1.6435 1.1242 0.1558
D
C w = − + w
1
For ,
3 4
1.5 10 Re 1.2 10 × ≤ ≤ ×
2 3
10
log 2.4571 2.5558 0.9295 0.1049
D
C w w = − + − + w
w
w
For ,
4 4
1.2 10 Re 4.4 10 × ≤ ≤ ×
2
10
log 1.9181 0.6370 0.0636
D
C w = − + −
For ,
4 5
4.4 10 Re 3.38 10 × ≤ ≤ ×
2
10
log 4.3390 1.5809 0.1546
D
C w = − + −
For ,
5 5
3.38 10 Re 4 10 × ≤ ≤ × 29.78 5.3
D
C w = −
For ,
5 6
4 10 Re 10 × ≤ ≤ 0.1 0.49
D
C w = −
For ,
6
10 Re <
4
8 10
0.19
Re
D
C
×
= −
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 . A new dimensionless group 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
terminal
V V =
2
Re
D D
N C =
D
N . From the Reynolds number, you can
immediately evaluate the terminal settling velocity. In the equations,
10
log
D
W N = .
For , 73; Re 2.37
D
N ≤ ≤
4 2 7 3 10 4
Re 1.7569 10 6.9252 10 2.3027 10
24
D
D D D
N
N N
− −
= − × + × − × N
−
2
For , 73 580; 2.37 Re 12.2
D
N < ≤ < ≤
2
10
log Re 1.7095 1.33438 0.11591 W W = − + −
For
7 3
580 1.55 10 ; 12.2 Re 6.35 10
D
N < ≤ × < ≤ ×
2 3
10
log Re 1.81391 1.34671 0.12427 0.006344 W W = − + − + W
5
W
For
7 10 3
1.55 10 5 10 ; 6.35 10 Re 3 10
D
N × < ≤ × × < ≤ ×
2 3
10
log Re 5.33283 1.21728 0.19007 0.007005 W W = − + −
3
1
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 and encounters a wall temperature at ,
which can be larger or smaller than . A simple version of this problem was first analyzed by
Graetz (1883). A sketch of the system is shown below.
0
T
w
T
0
T
Fluid at
0
T
r
z
R
( )
,
w
T R z T =
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 , , ,
p
k C ρ μ  This assumption also implies incompressible
Newtonian flow.
3. Axisymmetric temperature field 0
T
θ
∂
⇒ ≡
∂
.
4. Negligible viscous dissipation
2
Velocity Field
Poiseuille Flow
0; 0
r
v v
θ
= =
( )
2
0 2
1
z
r
v r v
R
⎛ ⎞
= −
⎜ ⎟
⎝ ⎠
Maximum velocity existing at the centerline
0
: v
Energy Equation
Subject to assumption (2), Equation (B.9.2) from Bird et al. (page 850) can be written as follows.
1 0 0
r
p
v v
T
C
t
θ
ρ
= =
∂
∂
r
v +
v
T
r
θ ∂
+
∂
2
2 2
3 4
1 1
z
T T
v
r z
T T
k r
r r r r
θ
θ
⎡ ⎤
∂ ∂
+ ⎢ ⎥
∂ ∂
⎢ ⎥
⎣ ⎦
∂ ∂ ∂ ⎛ ⎞
= +
⎜ ⎟
∂ ∂ ∂
⎝ ⎠
2
2 v
T
z
μ
⎡ ⎤
∂
+ + Φ
⎢ ⎥
∂
⎢ ⎥
⎣ ⎦
and therefore, simplified to
2 2
0 2 2
1
1
r T T T
v r
R z r r r z
α
⎛ ⎞ ⎡ ∂ ∂ ∂ ∂ ⎛ ⎞
− = +
⎜ ⎟ ⎜ ⎟ ⎢ ⎥
∂ ∂ ∂ ∂
⎝ ⎠
⎝ ⎠ ⎣
⎤
⎦
)
)
where is the thermal diffusivity of the fluid.
(
/
p
k C α ρ =
Boundary Conditions
Inlet: ( )
0
, 0 T r T =
Wall: ( ) ,
w
T R z T =
Centerline: is finite (0, T z
or ( ) 0, 0
T
z
r
∂
=
∂
3
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, we’ll neglect axial
conduction, which will obviate the need for writing a second condition in the coordinate. z −
NonDimensionalization
We shall use the following scheme for scaling (or nondimensionalizing) the variables.
0
w
w
T T
T T
θ
−
=
−
,
r
Y
R
= ,
z
Z
R Pe
= , where the Péclet Number
0
Rv
Pe
α
= .
This permits us to transform the governing differential equation and boundary conditions to the
following form.
( )
2
2
2 2
1 1
1 Y Y
Z Y Y Y Pe Z
θ θ θ ∂ ∂ ∂ ∂ ⎛ ⎞
− = +
⎜ ⎟
∂ ∂ ∂ ∂
⎝ ⎠
( ) , 0 1 Y θ =
( ) 1, 0 Z θ =
( ) 0, Z θ is finite
or ( ) 0, 0 Z
Y
θ ∂
=
∂
The Péclet Number
The Péclet number plays the same role in heat transport as the Reynolds number does in fluid
mechanics. First, we note that the Péclet number is the product of the Reynolds and Prandtl
numbers.
0 0
Re Pr
Rv Rv
Pe
ν
α ν α
= = × = ×
The physical significance of the Péclet number can be inferred by recasting it slightly.
0
Rate of energy transport by convection
Rate of energy transport by conduction
p
v C T
Pe
T
k
R
ρ Δ
= =
Δ
4
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 Péclet number by where is a
characteristic length in the axial direction.
/ L R L
For large values of , we can see that Pe
2
1
1
Pe
. Therefore, in the scaled energy equation, the
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
Péclet 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 To learn about how to include axial conduction, you can consult the articles by
Davis (1973), Acrivos (1980), and Papoutsakis et al. (1980).
100. Pe ≥
Let us make a sample calculation of the Péclet number for laminar flow heat transfer in a tube.
The thermal diffusivity of common liquids is typically in the range
2
7 7
10 2 10
m
s
− −
− × , and we’ll
use the larger limit. Choose
2
7
0
10 , 0.05 , 2 10
m m
R mm v
s s
α
−
= = = ×
This yields, , 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,
2, 500 Pe =
2
6
10
m
s
ν
−
≈ , which yields a
Prandtl number Pr 5
ν
α
= = . Therefore, the Reynolds number is Re 500 = , which is comfortably
in the laminar flow regime.
The final version of the scaled energy equation is
( )
2
1
1 Y Y
Z Y Y Y
θ θ ∂ ∂ ∂ ⎛ ⎞
− =
⎜ ⎟
∂ ∂ ∂
⎝ ⎠
We can solve this equation by separation of variables, because the boundary conditions in the
coordinate are homogeneous. The method of separation of variables yields an infinite series
solution for the scaled temperature field.
Y −
( ) (
2
1
,
n
Z
n n
n
Y Z A e Y
λ
θ φ
∞
−
=
=
∑
)
5
In the above solution, the functions ( )
n
Y φ are the characteristic functions or eigenfunctions of
a proper SturmLiouville system.
( )
2 2
1
1 0
d d
Y Y
Y dY dY
φ
λ φ
⎛ ⎞
+ − =
⎜ ⎟
⎝ ⎠
( ) ( ) 0 0 or 0
d
dY
φ
φ = is finite
( ) 1 0 φ =
The above ordinary differential equation for ( ) Y φ can be solved by applying the following
transformations to both the dependent and the independent variables (Lauwerier, 1951, Davis,
1973).
( ) ( )
2
2
X
X Y W X e λ φ = = Y
This leads to the following differential equation for ( ) W X .
( )
2
2
1
1 0
4 2
d W dW
X X
dX dX
λ ⎛ ⎞
+ − + − =
⎜ ⎟
⎝ ⎠
W
This is known as Kummer’s equation. It has two linearly independent solutions, but only one is
bounded at Because must be bounded, we must require that 0. X = ( ) 0 φ ( ) 0 W also remain
bounded. This rules out the singular solution, leaving us with the regular solution
( )
1
, 1,
2 4
W X c M X
λ ⎛ ⎞
= −
⎜ ⎟
⎝ ⎠
where is an arbitrary multiplicative constant. The function c ( ) , , 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.
( )
( )
( )
( ) ( )
( ) ( )
2
1
, , 1
1 2!
1 1
1 1 !
n
a a
a X
M a b X X
b b b
a a a n
X
b b b n n
+
= + + +
+
+ + −
+ +
+ + −
"
"
"
"
You can see that when , a b =
( ) , ,
X
M a a X e =
6
Application of the boundary condition at the tube wall, ( ) 1 φ 0 = , leads to the following
transcendental equation for the eigenvalues.
1
, 1, 0
2 4
M
λ
λ
⎛ ⎞
− =
⎜ ⎟
⎝ ⎠
The above equation has infinitely many discrete solutions for λ , which we designate as
n
λ , with
assuming positive integer values beginning from 1. Corresponding to each value n
n
λ , there is
an eigenfunction ( )
n
Y φ given by
( ) ( )
2
2
2
n
Y
n n
Y e W Y
λ
φ λ
−
=
n
The first few eigenvalues are reported in the table.
n
n
λ
2
n
λ
1 2.7042 7.3127
2 6.6790 44.609
3 10.673 113.92
4 14.671 215.24
5 18.670 348.57
Note that technically
{ }
2
n
λ is the set of eigenvalues, even though we use the term loosely to
designate { }
n
λ as that set for convenience.
The most important property of a proper SturmLiouville 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
2
0
1 0,
m n
Y Y Y Y dY m φ φ n − = ≠
∫
Using this orthogonality property, it is possible to obtain a result for the coefficients in the
solution by separation variables.
( ) ( )
( ) ( )
1
2
0
1
2 2
0
1
1
n
n
n
Y Y Y dY
A
Y Y Y dY
φ
φ
−
=
−
∫
∫
7
The Heat Transfer Coefficient
The heat flux from the wall to the fluid, ( )
w
q z is a function of axial position. It can be
calculated directly by using the result
( ) ( ) ,
w
T
q z k R z
r
∂
=
∂
but as we noted earlier, it is customary to define a heat transfer coefficient via ( ) h z
( ) ( )( )
w w
q z h z T T = −
b
where the bulk or cupmixing average temperature 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.
b
T
( )
1
0
1
0
2 ( ) ,
2 ( )
b
rV r T r z dr
T
rV r dr
π
π
=
∫
∫
where the velocity field ( ) ( )
2 2
0
1 / V r v r R = − . 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.
( )
( )
( )
,
w b
T
k R z
r
h z
T T
∂
∂
=
−
We can define a dimensionless heat transfer coefficient, which is known as the Nusselt number.
( )
( )
( )
1,
2
2
b
Z
hR
Y
Nu Z
k Z
θ
θ
∂
∂
= = −
where
b
θ is the dimensionless bulk average temperature.
8
By substituting from the infinite series solution for both the numerator and the denominator, the
Nusselt number can be written as follows.
( )
( )
( ) ( )
2
2
1
1
2
1
0
1
2
4 1
n
n
Z
n
n
n
Z
n n
n
d
A e
dY
Nu Z
A e Y Y Y dY
λ
λ
φ
φ
∞
−
=
∞
−
=
= −
−
∑
∑
∫
The denominator can be simplified by using the governing differential equation for ( )
n
Y φ , along
with the boundary conditions, to finally yield the following result.
( )
( )
2
2
1
2
1
1
2 1
n
n
Z
n
n
n
Z
n
n
n n
d
A e
dY
Nu
d e
A
dY
λ
λ
φ
φ
λ
∞
−
=
− ∞
=
=
∑
∑
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 →∞
2
1
3.656
2
Nu
λ
→ = .
The sketch qualitatively illustrates the behavior of the Nusselt number as a function of
dimensionless axial position.
3.656
Nu
Dimensionless Axial Position
0
0
Z
9
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 nonNewtonian flow, turbulent flow, and other geometries besides a circular
tube.
The Lévêque 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
is the assistance provided by
2
n
Z
e
λ −
in accelerating convergence for sufficiently large values of
Z . Lévêque (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
0
T
r
z
R
( )
,
w
T R z T =
t
δ
We shall now construct the Lévêque solution which is built on the assumption that the thickness
of the thermal boundary layer
t
R δ . This assumption leads to the following simplifications.
1. Curvature effects can be neglected in the radial conduction term. This means that the
derivative
1 T
r
r r r
∂ ∂ ⎛
⎜
∂ ∂
⎝ ⎠
⎞
⎟
can be approximated by
2
2
1 T T
R
R r r r
∂ ∂ ∂ ⎛ ⎞
=
⎜ ⎟
∂ ∂ ∂
⎝ ⎠
.
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 nonzero term.
Defining x R r = − , we can rewrite the velocity distribution as
( )
( )
2
2
0 0 2 2
1 2
z
R x
0
2
x x x
v r v v v
R R R R
⎛ ⎞
− ⎛ ⎞
= − = − ≈ ⎜ ⎟
⎜ ⎟
⎜ ⎟
⎝ ⎠
⎝ ⎠
10
Recall that a power series obtained by any method is a Taylor series. The above approach is
simpler than working out the derivatives of ( )
z
v 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 ( )
0
T x T →∞ → 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.
2
0 2
2
x T T
v
R z x
α
∂ ∂
=
∂ ∂
where the chain rule has been used to transform the second derivative in to the second
derivative in
r
x .
The temperature field satisfies the following boundary conditions. ( , T x z)
( )
0
, 0 T x T =
( ) 0,
w
T z T =
( )
0
, T z T ∞ =
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.
0
w
w
T T
T T
θ
−
=
−
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.
2
2
2X
Z X
θ θ ∂ ∂
=
∂ ∂
( ) , 0 1 X θ =
( ) 0, 0 Z θ =
( ) , 1 Z θ ∞ =
11
)
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 θ η = where the similarity variable ( ) / X Z η δ = . The variable ( ) Z δ represents the
scaled thermal boundary layer thickness, and is unknown at this stage. We make the necessary
transformations using the chain rule.
2
dF X d dF d dF
Z Z d dZ d dZ d
θ η δ η δ
η δ η δ
∂ ∂ ⎛ ⎞
= = − = −
⎜ ⎟
∂ ∂
⎝ ⎠
η
1 dF dF
X X d d
θ η
η δ η
∂ ∂
= =
∂ ∂
( )
2 2
2 2 2
1 1 1 1 dF dF d dF d F
X X Z d X d X d d d
θ η
δ η δ η δ η η δ η
⎡ ⎤
⎡ ⎤ ⎡ ⎤ ∂ ∂ ∂ ∂
= = = =
⎢ ⎥
⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦
⎣ ⎦
Using these results, the partial differential equation for ( ) , X Z θ is transformed to an ordinary
differential equation for ( ) F η .
2
2 2
2
2 0
d F d dF
d dZ d
δ
η δ
η η
⎛ ⎞
+ =
⎜ ⎟
⎝ ⎠
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 η and another for
( ) Z δ .
2
2
2
3 0
d F dF
d d
η
η η
+ =
2
3
2
d
dZ
δ
δ =
12
)
To derive the boundary conditions on these functions, we must go to the boundary conditions on
( , X Z θ . In a straightforward way, we see that ( ) 0, 0 Z θ = yields ( ) 0 F = 0 , and ( ) , 1 Z θ ∞ =
leads to . The remaining (inlet) condition gives ( ) 1 F ∞ =
( )
( )
, 0 1
0
X
X F θ
δ
⎛ ⎞
= =
⎜ ⎟
⎜ ⎟
⎝ ⎠
By choosing , this condition collapses into the condition ( ) 0 δ = 0 ( ) 1 F ∞ = obtained already
from the boundary condition on the scaled temperature field as . Summarizing the
boundary conditions on
X →∞
( ) F η and ( ) Z δ , we have
( ) ( ) 0 0, F F = ∞ 1 = , and
( ) 0 0 δ =
Integration yields the following solution for the scaled boundary layer thickness ( ) Z δ .
( )
1/ 3
9
2
Z Z δ
⎛ ⎞
=
⎜ ⎟
⎝ ⎠
The solution for ( ) F η is
( )
( )
3
3
3
0
0
0
1
4/ 3
e d
F e
e d
η
γ
η
γ
γ
γ
d η γ
γ
−
−
∞
−
= =
Γ
∫
∫
∫
Here, ( ) x Γ represents the Gamma function, discussed in the “Handbook of Mathematical
Functions” by Abramowitz and Stegun. The numerical value of ( ) 4/ 3 1.120 Γ ≈ .
Heat Transfer Coefficient
In the thermal entrance region, when the thermal boundary layer is thin, we can approximate the
bulk average temperature by the temperature of the fluid entering the tube . Therefore, we
define the heat transfer coefficient in this entrance region by
b
T
0
T
( ) (
0
,
w w
T
q k R z h T T
r
∂
= =
∂
) −
13
/ k Transforming to dimensionless variables, and defining a Nusselt number , we
can write
2 Nu hR =
( ) ( )
( )
( )
2
2 0, 0
dF
Nu Z Z
X Z d
θ
δ η
∂
= =
∂
By substituting for ( ) Z δ and ( ) 0
dF
dη
, we obtain the following approximate result for the
Nusselt number in the thermal entrance region.
( )
1/ 3
1/ 3
1.357
R
Nu Z Pe
z
⎛ ⎞
≈
⎜ ⎟
⎝ ⎠
Comparison with the exact solution shows this result is a good approximation in the range
2500 50
Pe z Pe
R
⎛ ⎞
≤ ≤
⎜ ⎟
⎝ ⎠
References
M. Abramowitz and I. Stegun, Handbook of Mathematical Functions, Dover, New York
(1965).
A. Acrivos, The extended Graetz problem at low Péclet numbers, Appl. Sci. Res. 36, 35 (1980).
E.J. Davis, Can. J. Chem. Engg. 51, 562 (1973).
L. Graetz, Ueber die Wärmeleitungsfähigkeit von Flüssigkeiten, Annalen der Physik und Chemie
18, 79 (1883).
H.A. Lauwerier, Appl. Sci. Res. A2, 184 (1951).
M.A. Lévêque, Les lois de la transmission de chaleur par convection, Annales des Mines,
Memoires, Series 12, 13, 201299, 305362, 381415 (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).
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.
Fluid
Hot
Cold
T ∇
Stable
Fluid
Hot
Cold
T ∇
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
s
T ,
which is larger than the ambient temperature in the fluid . Here is a sketch of the momentum
boundary layer along the plate.
e
T
m
δ
s
T
e
T
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 freehand 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.
∞
2
Now, let us consider a typical window in a home on a winter day when the outside air is at
and the inside of the room is at a balmy . 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.
10 F
D
68 F
D
m
δ
m
δ
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
density ρ moves at a velocity V , the kinetic energy per unit volume can be written as
2
1
2
V ρ .
This must come from some other form of energy, namely, potential energy lost by the fluid.
Over a vertical distance , the difference in potential energy between the less dense fluid in the L
3
boundary layer and the more dense fluid outside it can be approximately expressed as g L ρ Δ ,
where is the magnitude of the acceleration due to gravity, and g ρ Δ 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/ , because this is only an order of magnitude
analysis.
2
2
V g L ρ ρ ≈ Δ
Therefore, a typical order of magnitude of the velocity arising from natural convection is
V g
ρ
ρ
Δ
≈ L
Let us define a Reynolds number for the flowing fluid using this order of magnitude estimate.
3
2
Re
L
g L
LV
ρ
ρ
ν ν
Δ
= =
so that
3
2
2
Re
L
g L
ρ
ρ
ν
Δ
= . This is a dimensionless group that occurs often in natural convection
problems, and is given the name Grashof Number, abbreviated as Gr.
3
2
g L
Gr
ρ
ρ
ν
Δ
=
The coefficient of volumetric expansion of a fluid β is defined as
1 1 1
P P
p
V
V T T T
ρ
β ρ
ρ ρ
⎛ ⎞ ∂ ∂ ∂ ⎛ ⎞ ⎛ ⎞
= = = −
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
∂ ∂ ∂
⎝ ⎠ ⎝ ⎠
⎝ ⎠
where is the specific volume, T is the
temperature of the fluid and is its pressure. Therefore, we can write
V
P
p
T
T
T
ρ ρ
β
ρ ρ
Δ Δ ∂ ⎛ ⎞
= − = Δ
⎜ ⎟
∂
⎝ ⎠
where we have used a minus sign in relating ρ Δ to because
both are defined as being positive, and as temperature increases, density decreases.
T Δ
We can finally rewrite the definition of the Grashof number as follows.
3
2
T g L
Gr
β
ν
Δ
=
4
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.
3
Pr
T g L
Ra Gr
β
να
Δ
= × =
Here, α 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 , Mills (1) suggests using the following correlation for the
average Nusselt Number,
L
average
average
h L
Nu
k
= (where is the thermal conductivity of the fluid). k
( )
1/ 4
9
0.68 0.670 , 10
average
Nu Ra Ra = + Ψ ≤
( ) ( )
1/12
1/ 4
8 9
0.68 0.670 1 1.6 10 , 10 10
average
Nu Ra Ra Ra
−
= + Ψ + × Ψ ≤ <
12
In these equations is a function of the Prandtl number, defined as follows. Ψ
16/ 9
9/16
0.492
1
Pr
⎡ ⎤
⎛ ⎞
Ψ = +
⎢ ⎥
⎜ ⎟
⎝ ⎠
⎢ ⎥
⎣ ⎦
The reason for changing from one correlation to another when the Rayleigh number exceeds
is that the natural convection boundary layer undergoes transition to turbulence around that value
of the Rayleigh number. Mills points out that at the above two correlations do not
coincide in their predictions. This is fine, because that value of the Rayleigh number is an
arbitrary crossover point from one correlation to the other. It is fine to use the second
(turbulent) correlation for .
9
10
9
10 Ra =
9
10 Ra =
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.
5
1/ 4
0.52
average
Nu Ra =
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 π = 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 “singlepane” 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 12 mm for the air gap in thermopane
windows.
6
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, PrenticeHall, 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., BuoyancyInduced Flows and
Transport, Hemisphere, Cambridge (1988).
7
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 Cauchy’s 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 11.41 of
the book. The most common version that we shall use assumes that the density ρ and
thermal conductivity are constant, and is given below. k
2
p v
T
C T k T
t
ρ µ
∂ ⎛ ⎞
S + •∇ = ∇ + Φ +
⎜ ⎟
∂
⎝ ⎠
v
In the above equation, T is the temperature, is the specific heat at constant pressure,
is time, is the (vector) velocity,
p
C
t v
v
µ Φ stands for the rate of irreversible conversion of
mechanical energy into internal energy per unit volume by viscous dissipation, and
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
S
v
Φ 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
1
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
k T = − ∇ q
is known as Fourier’s law. Here, 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 Fourier’s law, 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 in common coordinate
systems.
q
k
q
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
p r
C T T ρ = − v
ef
where 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.
ref
T
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.
2
n
I
II
t
I
I
The assumption of thermodynamic equilibrium at the interface leads to the boundary
condition
I I
T T =
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,
I I
• = • n q n q
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 and its dimensionless counterpart, the Nusselt Number
, are in common use in engineering work. Here, I discuss how the heat transfer
coefficient is defined in typical situations.
h
Nu
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 T . Let us assume that T T
∞
w w ∞
>
for the sake of definiteness; however, the results given below are equally valid when
.
w
T T
∞
<
3
, U T
∞ ∞
x
y
w
T
t
δ
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 to T
w
T
∞
in the thermal boundary layer. The thickness of the
thermal boundary layer
t
δ 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 . From Fourier’s law, this can be written as
y
q
( ) ( ) , 0
w
T
q x k x
y
∂
= −
∂
Note that when , and 0
w
q >
w
T T
∞
> 0
w
q < when
w
T T
∞
< . Even though 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.
w
q
If we can solve the energy equation for the temperature distribution in the flowing fluid,
the heat flux ( )
w
q x can be evaluated as a function of x . A heat transfer coefficient
is defined for this system as follows. ( ) h x
( ) ( )( )
w w
q x h x T T
∞
= −
Therefore, the heat transfer coefficient is seen to be directly related to the temperature
gradient at the wall.
( )
( )
( )
, 0
w
T
k x
y
h x
T T
∞
∂
−
∂
=
−
This definition of holds regardless of the sign of ( ) h x ( )
w
T T
∞
− .
The Nusselt number is a dimensionless version of the heat transfer coefficient. It is
defined in the above problem as follows.
Nu
hL
Nu
k
=
4
Here is a characteristic length scale, which can be taken as the axial distance L 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 , encountering a step change in wall temperature to
. A sketch of the system is given below.
0
T
w
T
Fluid at
0
T
r
z
R
( )
,
w
T R z T =
At any given axial position , the heat flux from the wall to the fluid is given by z
( ) ,
w
T
q k 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 .
While it is possible to use (
z
)
0 w
T T − , the more common choice is ( )
w b
T T − where is
known as the bulk or cupmixing 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.
b
T
( )
1
0
1
0
2 ( ) ,
2 ( )
b
rV r T r z dr
T
rV r dr
π
π
=
∫
∫
Here, represents the velocity field. ( ) V r
5
The heat transfer coefficient in this system is defined as follows.
( )( )
w w
q h z T T = −
b
so that we can evaluate using ( ) h z
( )
( )
( )
,
w b
T
k R z
r
h z
T T
∂
∂
=
−
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.
heat flux
driving force
h =
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 . 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
, we can use dimensional analysis to identify the dimensionless groups on which
will depend. Thus, we obtain
Nu
D Nu
Re, Pr,
L
Nu Nu
D
⎛ ⎞
=
⎜ ⎟
⎝ ⎠
6
where
hD
Nu
k
= , the Reynolds number Re
UD
ν
= , and the Prandtl number Pr
ν
α
= . The
velocity appearing in the Reynolds number is a characteristic velocity in the system
such as the average velocity of flow, and
U
ν and α are the kinematic viscosity and the
thermal diffusivity of the fluid, respectively. The thermal diffusivity .
The symbol stands for the length of the pipe. For noncircular crosssections, the
Nusselt number will also depend on additional aspect ratio parameters.
( )
/
p
k C α ρ =
L
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.
h T
Nu
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 ν and α for molecular transport of momentum and
energy, respectively. Consider a simple onedimensional transport situation in which
fluid flows in the x − direction with a velocity ( )
x
v y . In this case, Newton’s law of
viscosity for the momentum flux
yx
τ (note that we are interpreting this symbol as the
negative of the shear stress) can be written as
x
yx
v
y
τ µ
∂
= −
∂
and in a similar onedimensional conduction problem, Fourier’s law for the heat flux
can be written as
y
q
y
T
q k
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.
( )
x
yx
v
y
ρ
τ ν
∂
= −
∂
7
{ } ( )
p ref
y
C T T
q
y
ρ
α
∂ −
= −
∂
The product
x
v ρ 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 ν . 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 α . 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.
Ability of a fluid to transport momentum by molecular means
Pr
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 ν α
, U T
∞ ∞
x
y
w
T
t
δ
m
δ
Small Prandtl number, ( ) Pr 1 ν α
, U T
∞ ∞
x
y
w
T
t
δ
m
δ
8
For gases, the Prandtl number is typically of ( ) 1 O , 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 or greater.
Because liquid metals are great conductors of thermal energy, Prandtl numbers for liquid
metals are typically of
5
10
( )
2
10 O
−
.
9
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 Perry’s 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 Perry’s 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 crosssection, 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 crosssection 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.
150
1.75
p
p
f
Re
= +
Here, the friction factor
p
f for the packed bed, and the Reynolds number
p
Re , are defined as
follows.
1
3
2
1
p
p
s
D
p
f
L V
ε
ρ ε
∆
=
−
( ) 1
p s
p
D V
Re
ρ
ε µ
=
−
The various symbols appearing in the above equations are defined as follows.
: p ∆ Pressure Drop
: L Length of the Bed
:
p
D Equivalent spherical diameter of the particle defined by
Volume of the particle
6
Surface area of the particle
p
D =
: ρ Density of the fluid
: µ Dynamic viscosity of the fluid
:
s
V Superficial velocity (
s
Q
A
= V where Q is the volumetric flow rate of the fluid and is the
crosssectional area of the bed)
A
: ε Void fraction of the bed ( ε is the ratio of the void volume to the total volume of the bed)
Sometimes, we may use the concept of the interstitial velocity V , which is related to the
superficial velocity by
i
s
i
V
V
ε
= . 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 KozenyCarman equation, used for flow under very viscous conditions.
150
p
p
f
Re
= , 1
p
Re ≤
The other is the BurkePlummer equation, used when viscous effects are not as important as
inertia.
1.75
p
f = , 1, 000
p
Re ≥
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.
2
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 particlefluid 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,
solidgas reactors, combustion of coal, roasting of ores, drying, and gas adsorption operations.
Superficial velocity
Superficial Velocity
V
f
: Minimum Fluidization
Velocity
D
C
C
D
B
A
A
B
E
E
O
O
Pressure
Drop
Bed
Height
3
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, we’ll 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 . 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
f
p ∆ multiplied by the crosssectional
area of the bed . In doing this balance, we ignore the small frictional force exerted on the wall
of the column by the flowing fluid.
A
Upward force on the bed p A = ∆
If the height of the bed at this point is and the void fraction is L ε , we can write
Volume of particles (1 ) AL ε = −
If the acceleration due to gravity is , the net gravitational force on the particles (net weight) is g
Net Weight of the particles ( )( )
1
p f
ALg ε ρ ρ = − −
4
Balancing the two yields
( )( )
1
p f
p Lg ε ρ ρ ∆ = − −
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 ( 0.1
p
D ≤ mm), the flow conditions at this stage are such
that the Reynolds number is relatively small ( ) 10 ≤ Re so that we can use the KozenyCarman
Equation, applicable to the viscous flow regime, for establishing the point of onset of
fluidization. This yields
( )
2
3
150 1
p f p
f
gD
V
ρ ρ
ε
µ ε
−
=
−
When the superficial velocity
s
V is equal to V , we refer to the state of the bed as one of
incipient fluidization. The void fraction
f
ε at this state depends upon the material, shape, and
size of the particles. For nearly spherical particles, McCabe, Smith, and Harriott (2001) suggest
that ε lies in the range 0.40 , increasing a bit with particle size. 0.45 −
For large particles ( mm), inertial effects are important, and the full Ergun equation must
be used to determine V . When in doubt, use the Ergun equation instead of a simplified version
of it.
1
p
D ≥
f
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
( )
2
settling
18
p f
gD
V
ρ ρ
µ
−
=
p
If we now use the result for the minimum fluidization velocity for the case of small particles,
given above, we see that the ratio
3
25 1
3
settling
f
V
V
ε
ε
−
=
For ε lying in the range 0.40 , this yields a ratio ranging from . McCabe et al.
suggest that it is common to operate fluidized beds at velocities as high as 30 , and values as
large as 100 are used on occasion. Recognizing that not all particles are of the same size and
that is only an average size, we see that fine particles are likely to be carried out with the
0.45 − 78 50 −
V
f
f
V
p
D
5
6
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 gasfluidized 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 crosssection, 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.
Perry’s Chemical Engineers Handbook, 7
th
Edition. 1997 (Ed: R.H. Perry, D.W. Green, and J.O.
Maloney), McGrawHill, 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 .
The position vector is labeled
( ) , , i j k
x .
Imagine a tiny line element dx , labeled in the sketch, at some instant of time. After
a small amount of time , the two ends have moved to new locations because of fluid
motion, and the new line element is labeled
PQ
dt
P Q ′ ′ .
P
Q
Q
′
P
′
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 must be nonuniform. To understand the nature of the changes in fluid
elements brought about by the flow, we must, therefore, investigate the velocity gradient,
, which is a second order tensor.
( ) v x
∇v
From calculus, we know that the differential change
x
dv can be written as
x x x
x
v v v
dv dx dy dz
x y z
∂ ∂ ∂
= + +
∂ ∂ ∂
and similar results can be written for the changes
y
dv and . It follows that the
differential change in the vector velocity, , is given by
z
dv
dv
1
y y y
x x x z z z
y y y
x x x z z
d dv dv dv
v v v
v v v v v v
dx dy dz dx dy dz dx dy dz
x y z x y z x y z
v v v
v v v v v
dx dy dz
x x x y y y z z z
+ +
∂ ∂ ∂ ⎛ ⎞ ⎛ ⎞ ⎛ ∂ ∂ ∂ ∂ ∂ ∂
= + + + + + + + +
⎜ ⎟ ⎜ ⎟ ⎜
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
⎝ ⎠ ⎝ ⎝ ⎠
∂ ∂ ∂ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ∂ ∂ ∂ ∂ ∂
= + + + + + + + +
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
⎝ ⎠ ⎝ ⎠ ⎝ ⎠
∂
=
x x z
v = i j k
i j k
i j k i j k i j k
z
v
⎞
⎟
⎠
∂
dx dy dz d
x y z
∂ ∂
+ + = ∇ •
∂ ∂ ∂
v v v
v x
Thus, the relative velocity of a point a distance dx from any given location is given by
the dot product of the tensor and the differential line element ∇v dx . This tensor can be
written as follows.
x x x
y y y
z z z
v v v
x y z
v v v
x y z
v v v
x y z
∂ ∂ ∂ ⎛ ⎞
⎜ ⎟
∂ ∂ ∂
⎜ ⎟
∂ ∂ ∂ ⎜ ⎟
∇ =
⎜ ⎟
∂ ∂ ∂
⎜ ⎟
⎜ ⎟
∂ ∂ ∂
⎜ ⎟
∂ ∂ ∂
⎝ ⎠
v
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 Ω
where the (symmetric) rate of strain or rate of deformation tensor E is given by
( )
1
2
= ∇ ∇
T
E v + v
and the (antisymmetric) vorticity tensor Ω is given by
( )
1
=
2
∇ ∇
T
v  v Ω
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 Ω
The vorticity tensor Ω is a skewsymmetric tensor. We can write its components in
terms of the components of the velocity gradient as follows.
2
1 1
0
2 2
1 1
0
2 2
1 1
0
2 2
y
x x z
y y
x z
y
x z z
v
v v v
y x z x
v v
v v
x y z
v
v v v
x z y z
⎛ ∂ ⎛ ⎞ ∂ ∂ ∂ ⎛ ⎞
− −
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
∂ ∂ ∂ ∂
⎝ ⎠
⎝ ⎠
⎜ ⎟
⎜ ⎟
∂ ∂ ⎛ ⎞ ⎛ ∂ ∂
⎜ ⎟
= − −
⎜ ⎟ ⎜
∂ ∂ ∂ ∂ ⎜ ⎟
⎝ ⎠ ⎝
⎜ ⎟
∂ ⎛ ⎞ ∂ ∂ ∂ ⎛ ⎞ ⎜ ⎟
− −
⎜ ⎟ ⎜ ⎟
⎜ ⎟
∂ ∂ ∂ ∂
⎝ ⎠
⎝ ⎠ ⎝ ⎠
Ω
y
⎞
⎞
⎟
⎠
A skewsymmetric tensor
ij
A can be formed from a vector by writing
k
a
ij ijk k
A a ε = . The
vector associated with the vorticity tensor in this manner is
1
2
− ω, where = ∇×v ω is
known as the vorticity vector. Using the relationship between Ω and
1
2
− ω, we obtain
1
2
ij j ijk j k
dx dx ε ω Ω = − or in Gibbs notation,
1 1
2 2
• = − × = × dx dx dx Ω ω ω
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
rotation with an angular velocity equal to
1
2
ω.
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 onehalf the vorticity vector at that point. To prove this
claim, consider a surface formed by an infinitesimal circle of radius located at a point a
x . Let the unit normal vector to the surface (perpendicular to the plane of the paper) be
, and the unit tangent vector to the circle at any point be . n t
a
t
n
Apply Stokes’s theorem to the velocity field in this circle.
( )
S C
dS ds ∇× • = •
∫ ∫ v
v n v t
3
Here, dS is an area element on the surface of the circle S and ds is a line element along
the circle C. Because is the component of the velocity along the periphery of the
circle, we can write the average linear velocity along the circle as
• v t
1
2
C
ds
a π
•
∫ v
v t and
therefore the average angular velocity as
2
1
2
C
ds
a π
•
∫ v
v t . From Stokes’s theorem, we see
that this is equal to the average value of
1
2
⎛ ⎞
∇× •
⎜ ⎟
⎝ ⎠
v n over the surface of the circle.
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 onehalf 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 onehalf the vorticity times the moment of inertia of the fluid, just as it is for a
rigid body.
Vorticity Vector
The vorticity , 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.
= ∇×v ω
Vorticity is generated at fluidsolid interfaces and at fluidfluid 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 are = ∇× v ω
( )
, ,
x y z
ω ω ω , then we can write the equations
of the space curves that are vortex lines as
x y z
dx dy dz
ω ω ω
= = .
4
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
S
•
∫
dS ω. By using Stokes’s
theorem, we can see that this is the circulation
C
ds •
∫ v
v t where is any closed curve
around the vortex tube, is a unit tangent vector to the curve at any point, and is a
line element.
C
t ds
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.
1 1
2 2
1 1
2 2
1 1
2 2
y
x x x z
y y y
x z
y
x z z z
v
v v v v
x y x z x
v v v
v v
x y y z
v
v v v v
x z y z z
⎛ ∂ ⎛ ⎞ ∂ ∂ ∂ ∂ ⎛ ⎞
+ +
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
∂ ∂ ∂ ∂
⎝ ⎠
⎝ ⎠
⎜ ⎟
⎜ ⎟
∂ ∂ ∂ ⎛ ⎞ ⎛ ∂ ∂
⎜ ⎟
= + +
⎜ ⎟ ⎜
∂ ∂ ∂ ∂ ∂ ⎜ ⎟
⎝ ⎠ ⎝
⎜ ⎟
∂ ⎛ ⎞ ∂ ∂ ∂ ∂ ⎛ ⎞ ⎜ ⎟
+ +
⎜ ⎟ ⎜ ⎟
⎜ ⎟
∂ ∂ ∂ ∂ ∂
⎝ ⎠
⎝ ⎠ ⎝ ⎠
Ε
y
⎞
∂
⎞
⎟
⎠
The diagonal elements of E
Consider a line element dx with a length . ds
( ) ( )
2
d d
ds d d
dt dt
= • x x
Using the fact that
d
d
dt
=
x
v , the above result can be rewritten as
( ) ( ) 2 2 2 2 2
d
ds ds d d d d d d d d
dt
= • • ∇ • = • • • • x v = x v x x E x + x x Ω
The second term in the farrightside is zero because Ω is an antisymmetric tensor. To
see this, we write
5
ij i j ji j i ij i j
d d dx dx dx dx dx dx • • Ω = Ω = − Ω x x = Ω so that 0
ij i j
dx dx Ω = .
In the above result, after writing the result in index notation, we first exchange the indices
and to obtain an intermediate result, and then use the antisymmetry property to write
.
i j
ij ji
Ω = − Ω
Therefore, we find that
( )
d
ds ds d d
dt
= • • x E x from which, by dividing through by we can write
2
ds
( )
1 d d
ds
ds dt ds ds
= • •
d x x
E
The vector is a unit vector pointing in the direction of the infinitesimal vector / d ds x
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
, which is equal to
11
E
x
v
x
∂
∂
. In a like manner, the rate of strain of a line element in the
direction is y −
y
v
y
∂
∂
, and that in the z − direction is
z
v
z
∂
∂
. This is the physical
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 . 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
S
d • S v over the surface.
S V
dV
d
dt
= • = ∇•
∫ ∫
S v v dV by the divergence theorem.
From the above, we can see that
0 0
1 1
lim lim
V V
V
dV
dV =
V dt V
→ →
= ∇• ∇•
∫
v v . So, the trace of
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.
6
The offdiagonal elements of E
Now, consider two line elements dx and d ′ x at a given point x and let the angle
between them be θ .
P
Q
Q′
θ
dx
d ′ x
Let us investigate the time rate of change of the dot product of the vectors dx and d ′ x .
( ) ( ) cos
i i i i
i i
j i j i
j j
d d
ds ds d d dv dx dx dv
dt dt
v v
dx dx dx dx
x x
θ
′ ′
′ ′ = • = +
∂ ∂
′ ′
= +
∂ ∂
x x
In writing the result in the second term in the second line, we have used the fact that the
infinitesimal change is the change in the velocity over an infinitesimal distance in the
direction of the vector
i
dv
′
d ′ x . Interchanging the indices and in that second term
permits us to combine the two terms.
i j
( ) cos 2
j
i
i j ij i j
j i
v
v d
ds ds dx dx E dx dx
dt x x
θ
⎛ ⎞ ∂
∂
′ ′
′ = + =
⎜ ⎟
⎜ ⎟
∂ ∂
⎝ ⎠
Dividing both sides by ds yields ds′
( )
1
cos 2 2
j
i
ij
dx
dx d d
ds ds E
ds ds dt ds ds ds ds
θ
′
′
′ = = •
′ ′
d
•
′
x x
E
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 1 1
cos cos sin
d d d
ds ds ds ds
ds ds dt ds dt ds dt dt
dθ
θ θ θ
⎡ ⎤
′ ′ = + −
⎢ ⎥
′ ′
⎣ ⎦
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 and
i
dx
i
dx
′
are the same. Let us consider the case when
the two vectors are orthogonal to each other. In this case, we obtain
7
d d
ds ds
′
• • =
′
x x
E
1
2
d
dt
θ
−
So, the sequential dot products of E with unit vectors in two orthogonal directions yields
onehalf 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 offdiagonal elements of E . For example, if we use x and y −
directions, the element is (
12 21
) E E = . Similar physical interpretations can be given to the
other offdiagonal elements of the rate of strain tensor. Thus, the offdiagonal 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 offdiagonal 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
axes can rotate, and in fact it does, with the angular velocity
1 1
2 2
∇× ω = v .
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 onehalf 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 PrenticeHall, 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.
8
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
, and select a unit tangent vector to the surface as , the noslip boundary condition can be
stated as
v t
• = • 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 represents
the unit normal,
n
on a rigid surface (kinematic condition) • = • v n V n
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
1
The noslip 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
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 fluidfluid 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 fluidfluid 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 and phase . I II
II
n
I
t
The unit normal vector points into phase here and the sketch also shows a unit tangent
vector to the interface .
n I
t
It so happens that the velocity fields in phases and 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 noslip 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.
I II
Therefore, we can write
2
t
at a fluidfluid interface (no slip) • = •
I II
v t v
and
• = •
I II
v n v n at a fluidfluid interface (kinematic condition)
Notice that we have two unknown vector fields and now, and therefore need twice as
many boundary conditions. Therefore, it is not sufficient to write just the above noslip and
kinematic conditions at a fluidfluid 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.
I
v
II
v
[ ]
2
I II s
Hσ σ • − = −∇ n T T n at a fluidfluid interface (jump condition on the stress)
In the stress boundary condition, the symbols
I
T and
II
T represent the stress tensor in each
fluid, is the mean curvature of the interface at the point where the condition is being applied, H
σ is the interfacial tension of the fluidfluid interface, and
s
∇ is the surface gradient operator
which can be written as . 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 and 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.”
(
∇− •∇ n n
)
I II
In the types of problems that we shall encounter, the stress boundary condition can be simplified.
The interfacial tension at a fluidfluid 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
τ represents this stress component, we can write
I II
t t
τ τ = at a fluidfluid interface (tangential stress balance)
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 fluidfluid 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 liquidgas 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
y
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 represent the liquid and gas, respectively. g
, , x g x l
l g
v v
y y
µ µ
∂ ∂
=
∂ ∂
at the free surface
The symbol µ in the above result stands for the dynamic viscosity. If we divide through by the
dynamic viscosity of the liquid, we obtain
, , g x g x l
l
v v
y y
µ
µ
∂ ∂
=
∂ ∂
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
,
0
x l
v
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.
4
Elements of Prandtl’s 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. Prandtl’s 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.
y
U
∞
x
laminar turbulent
Transition from laminar to turbulent flow in the boundary layer on a flat
plate occurs at , where
5
Re 5 10
x
≈ × Re
x
xU
ν
∞
= . Here, ν is the
kinematic viscosity of the fluid.
1
The assumptions can be used to establish the order of magnitude of the
boundary layer thickness.
A typical inertia term in the NavierStokes equation in rectangular Cartesian
coordinates is
u
u
x
ρ
∂
∂
, and a typical viscous term is
2
2
u
y
u
∂
∂
. Here, are
the velocity components in the
(
, u v
)
( )
, x y directions, and ρ and u 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.
2
U u
u
x L
ρ ρ
∞
∂
∂
∼
2
2 2
U u
y
u u
δ
∞
∂
∂
∼
Because the viscous force in the boundary layer is of comparable order to
the inertia force, these two order estimates must be comparable.
2
2
U U
L
ρ u
δ
∞ ∞
∼ or
2
L
U
u
δ
ρ
∞
∼ , which can be recast as
1
Re
L
L
δ
∼
where the Reynolds number based on the length of the plate Re
L
LU
ν
∞
= .
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
(
, 0
w
u
)
x
y
τ u
∂
=
∂
. We can estimate the order of
this quantity as
( )
w
U
x τ u
δ
∞
= . Because the shear stress is a local quantity,
2
we should use an order of magnitude of the variation of the boundary layer
thickness δ with x. From the order of magnitude argument used earlier, we
can estimate it as
( )
Re
x
x x
x
U
ν
δ
∞
= ∼ . If the width of the plate in the
direction is , the drag on the plate surface is given by z − w
( )
3/ 2
1/ 2
0 0
L L
U dx
x w
w
x d
x
τ u
ν
∞
= =
∫ ∫
2
3
L ρ u ∼
D w L ρ u =
( )
U U u dy
∞
1
δ dy
U
D w .
Ignoring the numerical factor of that appears after performing the
integration (because we are only estimating the order of magnitude), we can
write
D w U
∞
A rigorous calculation from boundary layer theory yields the result
3
0.664 U
∞
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
δ is that
thickness of the uniform stream that accounts for the “lost” flow because of
the presence of the solid surface.
1
0
δ
∞
= −
∫
or
0
1
u
∞
∞
 
= −

\ .
∫
3
Order of magnitude analysis of the continuity and NavierStokes
Equations
Now, we shall go through an order of magnitude analysis of the two
dimensional NavierStokes equations for steady incompressible Newtonian
laminar flow over a flat plate and simplify them using Prandtl’s 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 are used for the scaled
counterparts of the physical coordinates in the sketch, and the symbols u
and are used for the dimensionless counterparts of the physical velocity
components in the
y
v
x and directions, respectively. The scaled
incompressible version of the continuity equation is
y
0
u v
x y
∂ ∂
+ =
∂ ∂
From the scaling, we know that u is
( )
1 O . This means that the magnitude
of u lies between 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
0
∞
, 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 ), we can
conclude that the derivative
( )
1 x O ∼
u
x
∂
∂
is
( )
1 O as well. From the continuity
equation, we see that
u
x
∂
∂
and
v
y
∂
∂
must sum to zero; this forces the
derivative
v
y
∂
∂
to be . We know that the variable
( )
1 O
( )
y O δ ∼ where
δ represents the boundary layer thickness divided by the length L. In other
words, δ is the scaled boundary layer thickness. Because the derivative
4
( )
1
v
O
y
∂
∂
∼ , we must conclude that the change in the scaled velocity
component across the boundary layer must be of v
( )
O δ . We know from
the kinematic condition that 0 v = at the surface 0 y = . Therefore, the
magnitude of v must of
( )
O δ . We note that δ is a very small quantity
when the Reynolds number We express this fact by stating Re
L
1.
1. δ Therefore, the scaled velocity in the y −direction in the boundary
is a very small quantity.
2 2
1 1
2 2
2
2
1
Re
L
p
1 1 1
u u u
x y x y
δ δ
δ δ
∂
+
∂ ∂ ∂
− +
∂ ∂ ∂ ∂
u
x
∂
∂
u
y
1
O
u
y δ
∂
∂

\ .
∼
2
2
u
x
∂
∂
x
2
∂
∂
Now, following Schlichting [1] we proceed to use similar arguments in the
two components of the NavierStokes equation applicable to this situation.
First, consider the x −component.
u
u v
x
∂
+ =
∂
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, , and v
. To estimate the order of magnitude of
∂
∂
, we first note that u varies
from to 1 across the boundary layer, while the variable 0 y varies from
to
0
δ . This is the reason for the estimate that


. To estimate the
order of magnitude of the second derivatives, we must use similar
arguments. For example, consider the derivative . We know that
(
1
)
u
O
x
∂
∂
∼ . So, this quantity must change from to a magnitude of the
order of unity in a scaled distance
0
that also changes from to 1. This is
the reason for estimating the order of
0
2
u
x
as being unity. In a like manner,
5
the derivative
1 u
O
y δ
∂

∂
\ .
∼


, which means that it varies from to 0
1
δ
across
the boundary layer, in a distance of the order δ . Therefore, the second
derivative
2
2
1 u
O
y
2
δ
∂
 
∂
\ .
∼

. The order of magnitude of the Reynolds number
was established earlier on page 2.
y
∂
∂
(
O
)
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 −component of the NavierStokes equation.
2 2
2 2
2
1
Re
1
1 1
L
v v p v
u v
v
x y y x
δ δ δ δ
y
δ
∂ ∂ ∂ ∂
+ = − + +
∂ ∂ ∂ ∂
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
)
δ , so that the pressure
gradient
p
y
∂
∂
must necessarily be of the same order (or smaller). Because the
variation of pressure in the y −direction in the boundary layer must occur
over a distance of O
( )
δ , it is evident that the scaled pressure change across
the thickness of the boundary layer
(
2
p O δ ∆ ∼ . This is very small, and
can be ignored, which is Prandtl’s 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
6
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.
p ∂
∂
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, Prandtl’s steady twodimensional 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
0
u v
x y
∂ ∂
+ =
∂ ∂
NavierStokes Equation
2
2
1 u u
u v
x y x y
ν
ρ
∂ ∂
+ = − +
∂ ∂
u ∂
∂
For the flat plate problem, the potential flow is simply u U
∞
= . This means
that the potential flow pressure gradient is zero. Therefore, the Navier
Stokes equation simplifies to
7
2
2
u u
u v
u
x y y
ν
∂ ∂
+ =
∂ ∂
∂
∂
The boundary conditions are written as follows.
( )
0, u y U
∞
=
( )
, 0 0 u x =
( )
, 0 0 v x =
( )
, u x y U
∞
→ ∞ →
Commonly, the last condition is replaced with
( )
, U u x
∞
∞ = .
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, McGrawHill, New York, 1968.
8
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
crosssectional 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.
2 2
loss
2 2
out out in in
out in s
p V p V
gz gz w
ρ ρ
+ + = + + − −
loss: losses per unit mass flowing
w
s
: 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
2 2
loss
2 2
out out in in s
out in
p V p V
z z
g g γ γ
+ + = + + − −
w
g g
1
g γ ρ = : Specific weight of fluid
We can define the head developed by a pump as
g
w
h
s
p
− = .
Note that the work is negative which implies work is done on the
fluid by the pump.
friction
loss
f
h
g
= = h : loss expressed in head of fluid.
Sometimes, we express loss as a certain number (N) of velocity
heads. In this case,
2
ss
2
V
N = lo .
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
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
i
x
( )
( )
( 1)
times
1
1
, 0,1, 2
1 3 5 2 1
n
n
n
n
n r
φ
− +
−
 
= ∇∇∇⋅ ⋅ ⋅ ∇ =

⋅ ⋅ ⋅ ⋅ ⋅ −
\ .
,...
The first few are written below.
1
r
,
3
i
x
r
,
5 3
3
i j ij
x x
r r
δ
− ,
7 5
5
i j k i jk j ki k ij
x x x x x x
r r
δ δ δ + +
−
You can see that these solutions approach 0 as . Growing harmonics,
in contrast, grow with distance from the origin. These are obtained from the
decaying harmonics as . The first few growing harmonics are given
below.
r → ∞
2 1
( 1)
n
n
r φ
+
− +
( )
2 2
,
1, , ,
3 5
i i j ij i j k i jk j ki k ij
r r
x x x x x x x x x δ δ δ − − + + δ ⋅⋅ ⋅
Absolute and Pseudotensors
We’ll find these sets of harmonic functions very useful in constructing
solutions of Stokes problems. The important idea in each case will be that
we’ll be looking for solutions of Laplace’s 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 lefthanded. 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 righthanded to one that is lefthanded. 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 178179).
Solution of the Stokes equation
Now, let us begin with the Stokes equation for an incompressible flow.
2
p u ∇ = ∇ u
Also, an incompressible velocity field must satisfy . If we now take
the divergence of the Stokes equation, ∇• , which can be
rearranged to . The first term is zero because the flow
is incompressible, leading to the result that . Therefore, the pressure
field in every Stokes problem must be a harmonic function, that is, a
function which satisfies the Laplace equation.
0 ∇• = u
(
0 p − ∇ = u
)
2
u ∇
2
0 p ∇ =
( )
2
0 p u ∇ ∇• − ∇•∇ = u
We can construct a solution of the Stokes equation by adding a
homogeneous solution and a particular solution. The homogeneous solution
must also satisfy Laplace’s equation ∇ . The particular solution
, which satisfies the same equation as , can be written as
( ) H
u
( ) P
u
2 (H
= 0
)
u
u
( )
2
j P
j
x
u p
u
= .
Let us verify that this is indeed correct.
( ) ( )
1 1
2 2 2
j j P P
j j
i i i i
x x
p p
u p p x p x
x x x x
δ
u u u
∂      ∂ ∂ ∂ ∂
∇ = = = + = +
 
∂ ∂ ∂ ∂ ∂
\ . \ . \
u
ij j
i
x


.
Now, take the divergence.
( ) ( )
( )
1
2
P
ij j
i i
p
p x
x x
δ
u
  ∂ ∂ ∂
∇• ∇ = +

∂ ∂ ∂
\ .
u
i
x
2
2
1 1
2 2
j
ij j ij ji
i i i i i i i i
x
p p p p p p
x p
x x x x x x x x
δ δ δ
u u
∂ ∂ ∂ ∂ ∂ ∂ ∂
= + + = + = =
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
1 1
u u
∇
In obtaining the above result, we have used the fact that the pressure field
satisfies Laplace’s equation
2
0
i i
p
x x
 ∂
=
∂ ∂
\ .


. We see that the particular solution
is indeed correct. Note that we still must satisfy . Therefore, 0 ∇• = u
 
( ) ( )
1
3
2 2
H P
p p
u u
 
∇• = − ∇• = − ∇• = − + •∇

\ .
x
u u x p
because ( ) 3
2
i
i i ii i
i i i i
x p p
p x p p x p x p
x x x x
δ
u
  ∂ ∂ ∂ ∂
∇• = = + = + = + •

∂ ∂ ∂ ∂
\ .
x
x p ∇ .
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 or rotating with an angular
velocity , because of the linearity of the Stokes equation, the fields must
also be linear in these vectors.
i
i
Ω
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.
3
We’ll 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.
4
Notes on the Solution of Stokes’s Equation for Axisymmetric
Flow in Spherical Polar Coordinates
R. Shankar Subramanian
The details of the solution of Stokes’s Equation for the streamfunction
in spherical polar coordinates
(
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
4
0 E ψ =
)
, , r θ φ
( )
( )
( )
(
2
, , r r
)
1
ψ θ ψ θ + where
( ) 1
ψ satisfies
( ) 1
2
0 E ψ = (1)
and
( ) 2
ψ is the particular solution of
( ) ( ) 2
2
E ψ ψ =
1
π
(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 θ . The result is given below. 0, =
( ) ( ) ( )
1 2 3 1/ 2
2
,
n n n n
n n n n n
n
r A r B r C r D r ψ θ η
∞
− + + − + −
=
= + + +
∑
C (3)
Here, η , and are the Gegenbauer Polynomials of order
and degree − . 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
cos =
1/ 2
θ
( )
1/ 2
n
η
−
C n
( ) ( )
2
2
2
1
d
n n
d
η
η
Φ
− + − Φ 1 0 = (4)
where 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 , which satisfy Legendre’s equation.
n
( )
n
P η
1
( ) ( )
2
1
d d
n n
d d
η
η η
Φ
− + + Φ
1 0 = (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.
( )
( ) ( )
2 1/ 2
2 1
n n
n
P P
n
η η
η
− −
−
=
−
C (6)
(7)
( ) ( )
1/ 2
1
1
n n
P s d
η
η
−
−
−
= −
∫
C s
The first few Gegenbauer and Legendre Polynomials are
( ) ( )
( ) ( ) ( ) ( )
( ) ( )( )
1/ 2 1/ 2
0 1
1/ 2 2 1/ 2 2
2 3
1/ 2 2 2
4
1
1 1
1 1
2 2
1
1 5 1
8
η η
η η η η
η η η
− −
− −
−
= =
= − = −
= − −
C C
C C
C
η
η
( ) ( )
( ) ( ) ( ) ( )
( ) ( )
0 1
2 2
2 3
4 2
4
1
1 1
3 1 5 3
2 2
1
35 30 3
8
P P
P P
P
η η η
η η η η η
η η η
= =
= − = −
= − +
The Gegenbauer Polynomials satisfy the orthogonality property
( ) ( )
( )( )
1/ 2 1/ 2
1
2
1
2
1 1
m n
mn
d
n n n
η η
η
η
− −
+
−
=
− −
∫
C C
2 1
δ
−
.
(8)
for values of , 2 m n ≥
2
The function Q used by Leal in the textbook is related to the
Gegenbauer Polynomials through .
( )
n
η
( ) ( )
1/ 2
1 n n
Q η η
−
+
= − C
References
M.A. Abramowitz and I.A. Stegun, “Handbook of Mathematical Functions,”
Dover Publications (1965).
J. Happel, and H. Brenner, “Low Reynolds Number Hydrodynamics,”
PrenticeHall (1965); republished by Noordhoff International (1973).
L.G. Leal, “Laminar Flow and Convective Transport Processes,”
ButterworthHeinemann (1992).
T.M. MacRobert, “Spherical Harmonics,” Pergamon Press (1967).
R.A. Sampson, “On Stokes’s Current Function,” Phil. Trans. Royal Soc. A
182, 449518 (1891).
3
Vorticity
R. Shankar Subramanian
The vorticity , 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.
= ∇×u ω
Vorticity is generated at fluidsolid interfaces and at fluidfluid 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 Ω is a skewsymmetric tensor. We can write its components in
terms of the components of the velocity gradient in a rectangular Cartesian basis set as
follows.
1 1
0
2 2
1 1
0
2 2
1 1
0
2 2
y
x x z
y y
x z
y
x z z
u
u u u
y x z x
u u
u u
x y z
u
u u u
x z y z
⎛ ∂ ⎛ ⎞ ∂ ∂ ∂ ⎛ ⎞
− −
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
∂ ∂ ∂ ∂
⎝ ⎠
⎝ ⎠
⎜ ⎟
⎜ ⎟
∂ ∂ ⎛ ⎞ ⎛ ∂ ∂
⎜ ⎟
= − −
⎜ ⎟ ⎜
∂ ∂ ∂ ∂ ⎜ ⎟
⎝ ⎠ ⎝
⎜ ⎟
∂ ⎛ ⎞ ∂ ∂ ∂ ⎛ ⎞
⎜ ⎟
− −
⎜ ⎟ ⎜ ⎟
⎜ ⎟
∂ ∂ ∂ ∂
⎝ ⎠
⎝ ⎠ ⎝ ⎠
Ω
y
⎞
⎞
⎟
⎠
As we already know, a skewsymmetric tensor
ij
A can be formed from a vector by
writing
k
a
ij ijk k
A a ε = . The vector associated with the vorticity tensor in this manner is
1
2
− ω. Therefore,
1
2
ij j ijk j k
dx dx ε ω Ω = − or in Gibbs notation,
1 1
2 2
• = − × = × dx dx dx Ω ω ω
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
caused by a rigid rotation with an angular velocity equal to
1
2
ω.
1
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 onehalf the vorticity vector at that point. To prove
this, consider a surface formed by an infinitesimal circle of radius located at a point a x .
Let the unit normal vector to the surface be , and the unit tangent vector to the circle at
any point be .
n
t
a
t
n
Apply Stokes’s theorem to the velocity field in this circle.
( )
S C
dS ds ∇× • = •
∫ ∫ v
u n u t
Here, dS is an area element on the surface of the circle S and ds is a line element along
the circle C. Because is the component of the velocity along the periphery of the
circle, we can write the average linear velocity along the circle as
• u t
1
2
C
ds
a π
•
∫ v
u t and
therefore the average angular velocity as
2
1
2
C
ds
a π
•
∫ v
u t . From Stokes’s theorem, we see
that this is equal to the average value of
1
2
⎛ ⎞
∇× •
⎜ ⎟
⎝ ⎠
u n over the surface of the circle.
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 onehalf 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 onehalf the vorticity times the moment of inertia of the fluid, just as it is for a
rigid body.
Another useful result is that 0
i ij j
dx dx Ω = . You can prove this by using the fact that
.
ij ji
Ω = − Ω
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 ω are
( )
, ,
x y z
ω ω ω in a basis set obtained from
rectangular Cartesian coordinates, then we can write the equations of the space curves
that are vortex lines as
2
x y
dx dy dz
z
ω ω ω
= = .
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
S
•
∫
dS ω. By using Stokes’s
theorem, we can see that this is the circulation
C
ds •
∫ v
u t where is any closed curve
around the vortex tube, is a unit tangent vector to the curve at any point, and is a
line element.
C
t ds
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.
3
Viscous Dissipation term
When the dot product of each term in the NavierStokes equation is taken with the
velocity vector , 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 τ where for a
Newtonian fluid is given below in different coordinate systems.
u
: u ∇
v
u = Φ
v
Φ
Rectangular Cartesian Coordinates
( )
, , x y z
( )
2 2
2 2
2
2
2
2
2
3
y y
x x z
v
y
x z z
u u
u u u
x y z y
u
u u u
z y x z
∂ ∂
   ∂ ∂ ∂
   
Φ = + + + +
  
∂ ∂ ∂ ∂ ∂
\ . \ .
\ . \
∂
  ∂ ∂ ∂
 
+ + + + − ∇ •
 
∂ ∂ ∂ ∂
\ .
\ .
u
x


.
Cylindrical Polar Coordinates
( )
, , r z θ
( )
2 2 2
2
2
2
2
1
2
1 1
2
3
r r
v
r z
r z
u u u u
r r r z
u u u u
r
r r r r z
u u
z r
θ
θ θ
θ
θ θ
∂ ∂ ∂
     
Φ = + + +
  
∂ ∂ ∂
\ . \ . \ .
∂ ∂ ∂
 
+ + + +

∂ ∂ ∂
\ .
∂ ∂
+ + − ∇ •
∂ ∂
u
z
∂
∂
1
Spherical Polar Coordinates
( )
, , r θ φ
( )
2
2 2
2
2
2
2
1 1
2 c
sin
1 sin 1
sin sin
1 2
sin 3
r r r
v
r
r
u
u u u u u
r r r r r r
u
u u u
r
r r r r r
u
u
r
r r r
φ
θ θ
φ
θ θ
φ
θ
θ θ φ
θ
θ θ θ θ
θ φ
∂
  ∂ ∂
   
Φ = + + + + +
  
∂ ∂ ∂
\ . \ .
\ .
  ∂ ∂ ∂ ∂
 
+ + + +
 
∂ ∂ ∂ ∂
\ .
\ .
  ∂ ∂
+ + − ∇•

∂ ∂
\ .
u
ot
φ
2
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 between them as shown in the sketch below. b
x
y
b
U
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 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.
t
We shall assume the flow to be incompressible with a constant density ρ and Newtonian
with a constant viscosity µ . We neglect edge effects in the z − direction so that we can
set and 0
z
v = 0
z
∂
=
∂
v
, and assume fully developed flow, implying 0
x
∂
=
∂
v
. Here, v
stands for the velocity vector, and the subscripts denote components.
1
It can be established from the continuity equation and the kinematic condition at one of
the walls that . Therefore, the only nonzero velocity component is 0
y
v = ( ) ,
x
v t y ,
which can be shown to satisfy the following partial differential equation.
2
2
x
v
t y
ν
x
v ∂ ∂
=
∂ ∂
(1)
In Equation (1), represents time, and t ν is the kinematic viscosity. The initial condition
is
( ) 0, 0
x
v y = (2)
and the boundary conditions are
( ) , 0
x
v t U = (3)
and
( ) ,
x
v t b 0 = (4)
Using separation of variables, we obtained a solution of these equations that can be
written as follows.
( )
2 2
2
1
exp
,
2
1
x
n
t
n
v t y
y b
U b n b
ν
π
π
π
∞
=
⎡ ⎤
−
⎢ ⎥
⎛
⎣ ⎦
= − −
⎜
⎝ ⎠
∑
sin
n y ⎞
⎟
(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 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
n
2
/ t b ν not being too small compared to unity. But, if we attempt to calculate the sum
numerically for small values of time (
2
/ t b ν 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
2
/ t b ν 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 Rayleigh’s
problem. Mathematically, we replace the boundary condition at the top plate, given in
Equation (4), with
2
( ) ,
x
v t 0 ∞ = (6)
Note that to be precise, we must write Equation (6) as ( ) ,
x
v 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 , 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 is used to represent .
x
v
v
x
v
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 δ 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
3
new variable , can we expect the velocity field ( ) / y t η δ = ( ) ,
x
v t y to become a function
( ) Uφ η that depends only on the single new variable? This speculation is shown in the
sketch below.
The transformation to η is known as a “similarity transformation” and the variable η 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
( ) ( ) ,
x
v t y Uφ η = (7)
where
( )
y
t
η
δ
= (8)
and ( ) t δ 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
4
process leads to a consistent mathematical framework. For this purpose, we shall use the
chain rule of differentiation as needed.
2
x
v d y d d d d
U U U
t t d dt d dt d
η φ δ φ η δ φ
η δ η δ
∂ ∂
= = − =−
∂ ∂ η
(9)
Note that when writing the derivative of φ with respect to η , we already have assumed
that φ can depend explicitly only on the single variable η and used the ordinary
derivative instead of the partial derivative. If our conjecture proves to be incorrect, and
φ were to depend explicitly on both η and t , the above chain rule result will need to be
modified to include a partial derivative of φ with respect to time.
Let us now obtain expressions for the derivatives with respect to . y
x
v d U d
U
y y d d
η φ φ
η δ η
∂ ∂
= =
∂ ∂
(10)
and
2
2
2
2 2
x x
v v U d U d
y y y y d y d
U d d U d
y d d d
φ φ
δ η δ η
η φ φ
δ η η δ η
⎛ ⎞ ∂ ∂ ⎛ ⎞ ⎛ ∂ ∂ ∂
= = =
⎜ ⎟ ⎜ ⎟ ⎜
∂ ∂ ∂ ∂ ∂
⎝ ⎠ ⎝ ⎝ ⎠
⎛ ⎞ ∂
= =
⎜ ⎟
∂
⎝ ⎠
⎞
⎟
⎠
(11)
Substituting Equations (9) and (11) into the governing differential equation for
(Equation (1), leads to the following equation after slight rearrangement.
x
v
0
δδ
φ η φ
ν
′
⎡ ⎤
′′ ′ + =
⎢ ⎥
⎣ ⎦
(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 φ
explicitly depends only on the similarity variable η . But, Equation (12) suggests that
time also will explicitly appear in the result for φ because of the presence of the time
dependent term δδ′ . We have not yet specified ( ) t δ , however. Here is our chance to
do so and eliminate the inconsistency at the same time. We choose
C δδ ν ′ = (13)
where is an arbitrary constant. Later, we shall see that the value of C will affect the
result for
C
( ) t δ , but will not affect the final solution for ( ) ,
x
v t y . Therefore, for
convenience, we set , writing Equation (12) as 2 C =
5
2 0 φ ηφ ′′ ′ + = (14)
We now need to transform the initial and boundary conditions. Note that there are three
conditions on the velocity field ( ) ,
x
v t y , but only a second order differential equation for
( ) φ η . 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 0 y = , given in Equation (3).
This transforms in a straightforward manner to
( ) 0 φ 1 = (15)
The fact that a quiescent condition is approached as , described by Equation (6),
becomes
y → ∞
( ) 0 φ ∞ = (16)
The initial condition, given in Equation (2), transforms to
( )
0
0
y
φ
δ
⎛ ⎞
=
⎜ ⎟
⎜ ⎟
⎝ ⎠
(17)
and we see that we have not completely eliminated the original variables from appearing
explicitly in the problem statement for φ . To remove this inconsistency, and at the same
time select an initial condition for ( ) t δ , we must set
( ) 0 δ 0 = (18)
The choice in Equation (18) makes Equation (17) collapse into Equation (16); therefore,
the three conditions on yield two conditions on ( ,
x
v t y) ( ) φ η and one initial condition
on ( ) t δ , and we have a completely consistent mathematical framework for the problems
for ( ) φ η and ( ) t δ . 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
( )
2
1 2
0
a a e d
η
γ
φ η
−
= +
∫
γ (19)
6
where and 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
1
a
2
a
( ) ( ) erfc φ η η = (20)
where “ ” means “complementary error function.” This function is defined as
follows.
erfc
( ) ( ) erfc 1 erf η η = − (21)
where the “error function” “er ” is defined as f
( )
2
2
2
0
0
0
2
e d
erf e d
e d
η
γ
η
γ
γ
γ
η γ
π
γ
−
−
∞
−
= =
∫
∫
∫
(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 2 C = , when specialized using the initial
condition given in Equation (18), is
( ) 2 t t δ ν = (23)
When this result for ( ) t δ is used in Equation (8) in which η is defined, the solution for
the velocity field can be written as
( ) , erfc
2
x
y
v t y U
t ν
⎛
=
⎜
⎝ ⎠
⎞
⎟
(24)
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.
( ) 2
2
C
t C t 2 t δ ν = = ν (25)
( )
2
C
erfc φ η
⎛
=
⎜
⎜
⎝ ⎠
η
⎞
⎟
⎟
(26)
7
You can see that when the definition of η given in Equation (8) is used in Equation (26),
along with ( ) t δ from Equation (25), the factor / 2 C 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 δ , 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 when its argument is . Therefore, at a distance
3
4.678 10
−
× 2
4 y t ν = , 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
)
ν 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 semiinfinite, 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 nonlinear problems.
Concluding Remarks
The problem of unsteady onedimensional heat conduction in a semiinfinite solid slab
(or a quiescent liquid layer) in the y −direction, when the temperature at the surface
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
NavierStokes equations including inertia (and therefore nonlinear) 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 nonlinear
governing equations.
0 y =
Reference
1. M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions, Dover, New
York, 1965.
9
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.
Blade
D
T
Liquid
Level
Shaft
Baffle
D
A
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 Perry’s
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.
: ρ Density of the liquid
: µ viscosity of the liquid
:
T
D Diameter of the tank
:
A
D Diameter of the agitator
: g acceleration due to gravity
: N Rotation speed
: P Power
By using dimensional analysis, we can show that the Power number will be some
function of the Reynolds number and the Froude number
P
N
Re 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.
( )
,
P
N Re Fr φ =
Here
3 5 P
A
P
N
N D ρ
=
2
A
ND
Re
ρ
µ
=
2
A
N D
Fr
g
=
Also, the Weber number
2 3
A
N D ρ
We
σ
= can be important in some situations where the
objective is to emulsify a mixture; in that case, σ is the interfacial tension between the
two liquid phases.
If both the Reynolds and Froude numbers are important, we cannot use a scale model. Let
us investigate why. We use the subscripts and m p to designate model and prototype,
respectively, in the following. For dynamic similarity,
. Therefore,
m
Re Re =
p
2
2
p
m
p p A
m m A
m p
N D
N D
ρ
ρ
µ µ
=
2
Assuming we use the same fluid,
p
m
m p
ρ
ρ
µ µ
= . So,
2
2
p
m
A
m
p A
D
N
N D
= .
Now, consider the equality
m p
Fr Fr = . This implies
2
2
p
m
p A
m A
N D
N D
g g
= or
p
m
A
m
p A
D
N
N D
=
So, we must simultaneously satisfy
2
2
p
m m
p
A A
m
p A
D D
N
N D D
= =
A
. The only way to do this is to set
m p
A A
D D = . 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
m p
N N =
2
2
p
m
A
A
D
D
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 nonvortex forming systems, the data for a given type of
mixer can be fitted to
a
P
N c Re ′ =
This can be recast as
log log
P
N c a R = + e
c
where the new constant c log ′ = . If the flow in the mixer is laminar, it is found that
. Therefore, for a laminar mixer, we can write the result for the power input to the
mixer as
1 a = −
3
2 3
A
P c N D µ ′ =
Laminar mixers are not common. Usually, the flow is turbulent. For a generic baffled
mixer, the constant is found to be small, and when the flow is fully turbulent
corresponding to , we can set
a
≥ Re 10,000 0 a ≈ . In this case, the power input is found
to be given by
3 5
6.3
A
P N ρ = D
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, McGrawHill, New York, 2001.
4
1
Pl at e Fr ame Heat Exchanger
Pl at e Fr ame Heat Exchanger
2
Exchanger Matrix
3
PIPE FLOW
μ
ρ DV
= Re
Laminar, Transition, Turbulent
Entrance Length
L
e
/ D = 0.06 Re for Laminar
L
e
/ D = 4.4 Re
1/6
L
D
V
P
f
2
2 / 1 ρ
Δ
= Darcy Fanning
f f
4
1
=
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎟
⎠
⎞
⎜
⎝
⎛
= Δ
2
2
V
D
L
f P
ρ
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎟
⎠
⎞
⎜
⎝
⎛
=
Δ
=
g
V
D
L
f
g
P
h
f
2
2
ρ
V
D
l
P D
L
P R
Q
4 128 8
2 4 4
π
μ
π
μ
π
=
Δ
=
Δ
=
Friction Factor – Laminar Flow
Re
64
64 2
2
=
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
Δ
=
f
DV V
D
L
P
f
ρ
μ
ρ
Head Loss
4
⎟
⎠
⎞
⎜
⎝
⎛
=
D
f
ε
φ Re,
Friction Factor  Turbulent Flow
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
+ − =
f
D
f Re
51 . 2
7 . 3
log 0 . 2
1
10
ε
Colebrook
L
D
V
P
f
2
2 / 1 ρ
Δ
=
Darcy Fanning
f f
4
1
=
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎟
⎠
⎞
⎜
⎝
⎛
= Δ
2
2
V
D
L
f P
ρ
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎟
⎠
⎞
⎜
⎝
⎛
=
Δ
=
g
V
D
L
f
g
P
h
f
2
2
ρ
V
D
l
P D
L
P R
Q
4 128 8
2 4 4
π
μ
π
μ
π
=
Δ
=
Δ
=
Re
64
64 2
2
=
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
Δ
=
f
DV V
D
L
P
f
ρ
μ
ρ
⎟
⎠
⎞
⎜
⎝
⎛
=
D
f
ε
φ Re,
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
+ − =
f
D
f Re
51 . 2
7 . 3
log 0 . 2
1
10
ε
Useful Equations
5
Description Surface roughness (ε, ft)
Drawn tubing (brass, lead, glass etc) 0.000005
Commercial steel or wrought iron 0.00015
Asphalt cast iron 0.0004
Galvanized iron 0.0005
Cast iron 0.00085
Concrete 0.001 – 0.01
Riveted steel 0.003 – 0.03
6
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
+ + =
3
1
6
Re
10
000 , 20 1 001375 . 0
D
f
ε
( )
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
+ + =
3
1
5
6
10
10
0002 . 0 000 , 20 1 001375 . 0 f
Q1. Read the value of friction factor for Re=10
5
and (ε/D) = 0.0002 and compare the
value to that obtained by equation:
From the chart f = 0.00475
Using the equation:
f = 0.0047
( )
500 , 85
10
8 . 12 667 . 0
Re
/ 8 . 12
349 . 0 667 . 0
4 4
/ 46 . 4
sec) / min/ / ( 8 . 448
min / 2000
4
2 2
2
3
3
= = =
= =
= = =
= =
−
x DV
s ft
A
Q
V
ft ft
D
A
s ft
ft gal
gal
Q
ν
π π
For cast iron ε=0.00085 ft
ε/D = 0.00085/0.667 = 0.0013
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
+ − =
f
D
f Re
51 . 2
7 . 3
log 0 . 2
1
10
ε
f=0.0235
Calculate:
Head Loss
BHP of the Pump
If Efficiency is 0.9
Given:
Q = 2000 gpm
8 inch ID
Kinematic viscosity
10
4
ft
2
/s
Density = 1.8 slug/ft
3
7
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎟
⎠
⎞
⎜
⎝
⎛
=
Δ
=
g
V
D
L
f
g
P
h
f
2
2
ρ
Head Loss
= 134 ft
loss = h
f
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
g
V
D
l
f h
L
2
2
=
( ) ( )
( )
( )
2
2
2
2 2
2 / 1
2 / 1
2 / 1
2
2 /
V p K
g
V
K h
V K p
V
p
g V
h
K
L
L L
L
L
L
ρ
ρ
ρ
= Δ ⇒
=
= Δ
Δ
= =
Losses occurring in straight pipes – Major losses
Losses occurring in pipe systems – Minor losses
K
L
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.
K
L
= ø (geometry, Re)
However for most practical cases:
K
L
= ø (geometry)
Dynamic pressure
2 2 2
2 / 1
2
2 / V
p
V
g
g
p
g V
h
g
p
h
L
L
ρ ρ ρ
Δ
=
Δ
= ⇒
Δ
= Q
8
Minor losses are sometimes given in terms of an equivalent length l
eq
Head loss through a component is expressed in terms of equivalent
length that would produce the same head loss as the component.
f
D K
l
g
V
D
l
f h
L
eq
eq
L
=
=
2
2
D and f are based on the pipe containing the
component.
Entrance flow conditions – Reentrant K
L
=0.8 Sharp edge K
L
= 0.5
Slightly rounded entrance – K
L
= 0.2
Well rounded – K
L
= 0.04
Vena contracta
2 1 3
Flow separation
V
2
>V
3
p
1
x
1 x
2
x
3
ρV
2
2
/2
ρV
3
2
/2
K
L
ρV
3
2
/2
Actual
Ideal full recovery of kinetic energy
x
9
K
L
r/D
0 0.1 0.15 0.20 0.25
0.5
0.4
0.3
0.2
0.1
0
D
r
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
S
Rounded entrance
Standard 90
0
elbows
47m
16m
150m
150m
D = 0.156m
ε = 5 x 10
5
K
L1
= 0.25 rounded entrance
K
L2
=0.9 elbow
Density of water ρ = 10
3
kg/m
3
Viscosity of water = 1 x 10
3
N s/m
2
Water flows from a large reservoir and discharges into the atmosphere. Determine the
volumetric flow rate m
3
/s?
(1)
(2)
10
Write Bernoulli’s equation between the free surface and the entrance.
By continuity V
1
=V
2
=V
( )
g
V
z z
g
p p
z
g
p
z
g
V
g
p
s
s
s
s
2
2
2
1
1
1
1
2
1 1
− − =
−
+ = + +
ρ
ρ ρ
Between sections 1 and 2
( )
g
V
K
g
V
K
g
V
D
fL
z z
g
p p
g
V
K
g
V
K
g
V
D
fL
z
g
V
g
p
z
g
V
g
p
L L
L L
2
2
2 2
2
2
2 2 2 2
2
2
2
1
2
1 2
2 1
2
2
2
1
2
2
2
2
1
2
1
+ + + − =
−
+ + + + + = + +
ρ
ρ ρ
( )
( )
1 2
2
1 2
2
2 1
2 2
2 1
2
2
+ + +
−
=
⎟
⎠
⎞
⎜
⎝
⎛
+ + + = −
L L
s
L L s
K K
D
fL
z z g
V
K K
D
fL
g
V
z z
Both f and V are unknown – work simultaneously with the friction factor diagram
to obtain a solution:
122 0.0150 1.9(10)
7
122
V (calculated)
m/s
0.0150 1.8(10)
7
115
0.0165 5(10)
5
3
f Re V (assumed)
m/s
Volumetric flow rate Q = π (D
2
/4) x 122= π (0.156)
2
/4 x 122=2.33 m
3
/s
τ = ηθ
Here,
η
is called the “apparent viscosity” of the fluid, and is a function of the
shear rate. In the above example, a plot of looks like this.
η
as a function of the shear rate
θ
Apparent Viscosity
shearthinning fluid
Shear Rate
Many shearthinning 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.
Newtonian Region log Apparent Viscosity
PowerLaw Region
shearthinning fluid Newtonian Region log Shear Rate
2
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 powerlaw region. In this region, we can approximate the behavior by
logη = a + b log θ
which can be rewritten as
η = Kθ b
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.
η = Kθ n−1
Upon using the connection among the shear stress, apparent viscosity, and the shear rate we get the powerlaw model.
τ = Kθ n
where n is called the powerlaw 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 shearthinning 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 shearthickening or dilatant fluids. Typical plots of shear stress versus shear rate and apparent viscosity versus shear rate are shown next.
3
This is a fluid which will not flow when only a small shear stress is applied. Another important type of nonNewtonian fluid is a viscoplastic or “yield stress” fluid. it would be good if the paste does not flow at the slightest amount of shear stress. Most shearthickening fluids tend to show shearthinning at very low shear rates. We need to apply an adequate force before the toothpaste will start flowing. Once it exceeds the yield stress. when you open a tube of toothpaste. Bingham plastics are a special class of viscoplastic fluids that exhibit a linear behavior of shear stress against shear rate. 4 . clay slurries. Typical viscoplastic behaviors are illustrated in the next figure. So. For example. viscoplastic fluids behave like solids when the applied shear stress is less than the yield stress. The shear stress must exceed a critical value known as the yield stress τ0 for the fluid to flow.Shear Stress shearthickening fluid Shear Rate Apparent Viscosity shearthickening fluid Shear Rate Some examples of shearthickening fluids are corn starch. the viscoplastic fluid will flow just like a fluid. and solutions of certain surfactants.
You can learn more about these and other behaviors from several fluid mechanics textbooks and from Perry’s Chemical Engineers’ Handbook. The opposite behavior. you can use a quick jerking motion to snap it back into the container. Several industrially important polymer melts and solutions are viscoelastic. 5 . and such a fluid is called a rheopectic fluid. toothpaste. Of course. The viscosity of a thixotropic liquid will decrease with time under a constant applied shear stress. some paints exhibit a yield stress. some classes of fluids exhibit timedependent behavior. Viscoelastic fluids exhibit strange phenomena such as climbing up a rotating shaft. This means that these fluids behave both as solids (elastic) and fluids (viscous). swelling when extruded out of a dye. Nondrip paints behave in this way. However. etc. nuclear fuel slurries. and blood. wherein the fluid increases in viscosity with time when a constant shear stress is applied is not as common. An example of a common viscoelastic liquid is eggwhite.Shear Stress viscoplastic fluids Bingham Plastic Shear Rate Examples of viscoplastic fluids are drilling mud. For instance. the viscosity will gradually recover with time as well. Another important class of fluids exhibits viscoelastic behavior. mayonnaise. when the stress is removed. You have probably noticed that when it flows out of a container. the viscosity may vary with time. This means that even under a given constant shear rate. this is not an exhaustive discussion of nonNewtonian behaviors. Also.
we shall see how this yields the wellknown HagenPoiseuille equation connecting the pressure drop and the volumetric flow rate. when necessary we use symmetry considerations to write boundary conditions. we can set the shear stress to zero. the velocity does not vary with distance in the flow direction in both cases. Regarding boundary conditions. Shankar Subramanian When fluid flow occurs in a single direction everywhere in a system. the velocity varies with the distance coordinate between the two plates. we can obtain a differential equation for the velocity distribution. the shear stress distribution across the crosssection can be written as well. For flow through a straight circular tube. If we neglect entrance and exit effects. Typically this is an equation for the shear stress. This means that the 1 . Also. Other examples include flow between two wide parallel plates or flow of a liquid film down an inclined plane. At a free liquid surface when flow is not driven by the adjoining gas dragging the liquid. In general at a fluidfluid interface the velocity and the shear stress are continuous across the interface. for flow between wide parallel plates. this is the definition of fullydeveloped flow. which is constructed by translating a differential crosssectional area (normal to the flow) in the direction of the flow over a finite distance. we can ignore variations in the other direction normal to the flow which runs parallel to the surfaces of the plates. 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 crosssectional coordinate over its full extent. fluid velocity varies across the crosssection only in one coordinate direction and is uniform in the other direction normal to the flow direction. we consider the limit as this distance approaches zero and obtain a differential equation. In the above situations. Once the profile is known. A momentum balance can be written for a control volume called a shell. the most common condition we use is the “no slip” boundary condition.Shell Balances R. If desired. shell balances are useful devices for applying the principle of conservation of momentum. there is variation with the radial coordinate. This states that the fluid adjacent to a solid surface assumes the velocity of the solid. If the plates are sufficiently wide. Even though the interfacial region is made up of atoms and molecules and has a nonzero thickness. we treat is as a plane of zero thickness for this purpose. Later. Similarly. The key idea is that we use a differential distance in the direction in which velocity varies. we can calculate the volumetric flow rate and the average velocity as well as the maximum velocity. By inserting a suitable rheological model connecting the shear stress to the velocity gradient. An example is incompressible laminar flow of fluid in a straight circular pipe. In the case of pipe flow. This is then integrated with the boundary conditions relevant to the problem to obtain the velocity profile. We shall see by example how these shells are formed. but not with the polar angle.
Its inner cylindrical surface is at a radial location r and the outer cylindrical surface is at r + ∆ r . The shell shown in the figure is L units long ∆ r units thick in the radial direction. At steady state. we shall proceed to construct a shell momentum balance for steady laminar flow through a straight circular tube. Now.φ .Rate of efflux of momentum from the shell + Sum of all the forces acting on the fluid in the shell = 0. We assume steady incompressible fullydeveloped laminar flow in the zdirection. Momentum enters at the left face and leaves at the right face with the flowing fluid. 2 . the shear stress and τ rz = τ rz ( r ) . Rate of entry of momentum into shell . Steady FullyDeveloped Incompressible Laminar Flow in a Straight Circular Tube ∆r Flow r z R Shell θ L Cylindrical polar coordinates ( r . Therefore. the rate of accumulation of momentum in the fluid contained in the shell is zero. vz = vz r .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. z ) are used. ( ) Similarly. This can happen if the temperature varies along the surface or if surface active chemicals are present. The shell balance reads as follows. The only nonzero velocity component is vz and it only depends on the radial coordinate r .
So the shell balance reduces to the statement that the sum of the forces on the fluid in the shell is zero. and the viscous force. the fluid within the ( ) ( ) 3 . We need to use the component of this force in the z direction. In the same way. Rate of efflux of momentum z . This can be obtained as Contribution in the flow direction from the gravitational force where = 2π r ∆ rL ρ g cosθ g is the magnitude of the (vector) acceleration due to gravity g . The forces on the fluid consist of the pressure force. the associated mass flow rate is ∆ m = 2π r ∆ r ρ vz and the rate of entry of momentum. which is the product of the mass flow rate and the velocity. The fluid at r + ∆ r exerts a shear stress on the fluid in the shell in the positive z − direction which we designate as τ rz r + ∆ r . We shall account for each in turn. Pressure force = 2π r ∆ r p ( 0 ) − p ( L ) The force of gravity on the fluid in the shell. Rate of entry of momentum = ∆ m vz = 2π r ∆ r ρ vz2 Because this is independent of right side is exactly the same.Now. the rate of efflux of momentum from the face on the = 2π r ∆ r ρ vz2 and the two terms cancel each other. the gravitational force. The pressure force. This notation means “shear stress evaluated at the location r + ∆ r . the volumetric flow rate of fluid into the shell is ∆ Q = 2π r ∆ r vz If the density of the fluid is ρ . given as the product of the mass of the fluid in the shell and the (vector) acceleration due to gravity g acts vertically downward.” ( ) Multiplied by the surface area of the shell at that location this yields a force 2π r + ∆ r L τ rz r + ∆ r in the z − direction. We now calculate the viscous force acting on the shell. is written as follows. given as the product of the area and the pressure acting on it. is written as follows.
and divide by p ( 0 ) − p ( L ) ( r + ∆ r )τ rz ( r + ∆ r ) − rτ rz ( r ) r ρ g cosθ + =0 + L ∆r Now we take the limit of the terms in the above equation as ∆ r → 0 . The quantity P is known as the hydrodynamic pressure or simply the dynamic pressure. The result of this integration is ∆ P r2 rτ rz = − + C1 L 2 where C1 is an arbitrary constant of integration that needs to be determined. Therefore. We can rewrite this result as 4 . at smaller r. Its gradient is zero in a stationary fluid and therefore it is uniform when the fluid is not in motion. Force from viscous stress = 2π L ( r + ∆ r )τ rz ( r + ∆ r ) − rτ rz ( r ) 2π∆ rL to obtain the following result. the total force arising from the shear ( ) stresses on the two cylindrical surfaces of the shell fluid is written as follows. the fluid below exerts a reaction on the shell fluid that has the opposite sign. As a result. The first term is unaffected while the second term is simply the derivative of rτ rz with respect to r .e. The resulting force is −2π rL τ rz r . i. This leads us to the first order differential equation for the shear stress τ rz ( r ) . 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 rτ rz . Add the forces. 1 d ∆P ( rτ rz ) = − r dr L where we have introduced the symbol ∆ P = p ( 0 ) − p ( L ) + ρ gL cosθ for convenience. set the sum to zero.shell at the location r exerts a stress τ rz ( r ) on the fluid below.
we must choose the arbitrary constant of integration C1 = 0 and write the shear stress distribution in the tube as follows. The fluid at the wall is stationary. the shear stress should approach infinity as the radial coordinate r → 0 . 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. This is not physically possible. We can explain why it is negative. we need to introduce a relationship between the shear stress and the velocity gradient. is negative.τ rz = − ∆ P r C1 + L 2 r According to this result. A sketch of the shear stress distribution is given below. Therefore. Recall that a force acting in the positive z − direction is taken by convention here to be positive. however. Recall the Newtonian constitutive model τ rz = µ dvz dr 5 . τ rz ( r ) r z R To find the velocity distribution. It varies linearly across the crosssection. τ rz = − ∆P r 2L The shear stress is zero at the centerline and its magnitude is a maximum at the tube wall. according to our convention.
Use of this constitutive model in the result for the shear stress leads to dvz ∆P == − r 2µ L dr This equation can be integrated immediately to yield v z ( r ) = C2 − ∆P 2 r 4µ L where a new arbitrary constant of integration C2 has been introduced. r z R vz ( r ) 6 . vz ( R ) = 0 Use of this condition leads to the result ∆ P R2 C2 = 4µ L so that we can write the velocity distribution as follows. It is sketched below.where µ is the dynamic viscosity of the fluid. ∆ P R2 vz ( r ) = 4µ L r2 1 − R 2 This velocity distribution describes a parabola. 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 .
Now. Its value vmax ∆ P R2 = 4µ L We can calculate the volumetric flow rate through the tube by multiplying the velocity at a given r by the differential crosssectional area 2π r dr to yield dQ and then integrating across the crosssection. y = 0 . defined as the volumetric flow rate per unit area of crosssection. vavg ∆ P R2 Q = = 8µ L π R2 It can be seen that in a circular tube.We see that the maximum velocity occurs at the centerline. the average velocity is onehalf the maximum velocity. and z in the flow direction. 7 . Steady FullyDeveloped 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 . we find that ∆ P h2 vz ( y ) = 2µ L y2 1 − h 2 This also is a parabolic velocity profile that is symmetric about the centerline. If the coordinate y is measured from the centerline toward one of the plates. which is at is r = 0 . The maximum velocity occurs at the centerline. we can calculate the average velocity in the tube. π ∆ P R2 R r 2 Q = ∫ dQ = ∫ 2π r vz ( r ) dr = r 1 − dr 2µ L ∫ R 2 0 0 0 Q R π ∆ P R4 = 8µ L This is a result known as the HagenPoiseuille equation for the relationship between the flow rate and the applied pressure drop for laminar flow through a circular tube.
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 . we can immediately write the volumetric flow rate in the film as 8 . If we use h for the film thickness. The ratio of the average velocity to the maximum velocity is the same as that for flow between parallel plates. The laminar flow of a liquid film down an inclined plane can be modeled in the same manner as flow between parallel plates. the differential crosssectional area for flow is W dy and the differential volumetric flow rate would be dQ = vz ( y )W dy . the above parabolic velocity profile applies directly. This can be integrated across the crosssection as follows. This is why the velocity profile in a falling liquid film in laminar flow is represented by onehalf of the parabola that is obtained for flow between parallel plates. This is because the flow between parallel plates is symmetric about the centerline y = 0 . namely 2 / 3 . Then.vmax ∆ P h2 = 2µ L Let the plates be W units wide in the x − direction. vavg ∆ P h2 = 3µ L 2 / 3 for steady laminar flow We can see that the ratio of the average to the maximum is between wide parallel plates. Q = ∫ W vz ( y ) dy = 2∫ W vz ( y ) dy −h 0 2 h h h ∆Ph = 2W 2µ L yielding y2 ∫ 1 − h 2 dy 0 2∆ PWh3 Q= 3µ L The average velocity is obtained by dividing Q by the crosssectional area 2Wh . which leads to the velocity gradient being zero along the centerline. In this case. if we measure the y − coordinate from the free surface of the liquid.
we obtain a differential equation for the shear stress distribution in the direction normal to that of the flow. The shell balance approach is limited to onedimensional flow situations and straight geometries. and permitting the thickness of the shell to approach zero. Introducing a constitutive model. but of differential thickness in the direction along which the velocity varies. When the velocity distribution has been obtained. it can be multiplied by a differential area element and integrated across the crosssection to obtain the volumetric flow rate. can be evaluated by the application of boundary conditions based on physical principles. Summary These notes introduce you to a simple approach for modeling onedimensional flow situations. we must use partial differential 9 . The arbitrary constants. This means that p ( 0) = p ( L ) . such as the Newtonian model. The average velocity can be obtained as the volumetric flow rate divided by the crosssectional area. permits us to develop a differential equation for the velocity distribution.∆ PWh3 Q= 3µ 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. which must be introduced when the integrations are performed. Therefore. In the more general case. we can write ∆P = ρ g cosθ L which yields ρ g Wh3 cosθ Q= 3µ 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. By accounting for all the forces on the fluid contained in this shell. The main idea is to construct a shell of fluid that is long in the flow direction. The shear stress and velocity distributions can be obtained by integrating the applicable differential equations.
equations obtained using the same principles. known as the NavierStokes equations. and simplify them for the specific problem under consideration. 10 .
To make design calculations involving twophase flow. Toward the end of these notes. Dimensional analysis is used to establish the relevant groups to aid in designing suitable experiments. It summarizes correlations that are currently used in industry. Also. and pumping the mixture through a pipe. Twophase flow is a difficult subject principally because of the complexity of the form in which the two fluids exist inside the pipe. and combustion systems. 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. We first consider a situation where the ratio VL / VG is large. and then the regime where the ratio is lower. Shankar Subramanian Chemical engineers frequently encounter the flow of a mixture of two fluids in pipes. We use VL and VG to designate the superficial velocity of the liquid and gas phases. known as the flow regime. we only consider the qualitative features of gasliquid flow in a horizontal pipe to give you an appreciation of the complexity of twophase flow when compared with the flow of a single fluid phase. This is not necessarily equal to the relative fraction of that phase in the entering fluid mixture. In twophase flow. Perry’s Handbook is a useful resource. The sketches on the following pages depict various regimes of cocurrent twophase flow in a horizontal pipe through which a gasliquid mixture is flowing. It is the relative fraction of one phase in the pipe. the concept of holdup is important. Here. gasliquid reactors. Also. Liquidgas or liquidvapor mixtures are encountered in condensers and evaporators. 1 . It is difficult to construct a model from first principles in all but the most elementary situations. an informative chapter in Holland and Bragg (1995) is devoted to gasliquid twophase flow. we’ll see some examples of these regimes. 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. respectively. Most available empirical results are applicable only to gasliquid twophase flow. the transport of some solid materials in finely divided form is accomplished by making a slurry of the solid particles in a liquid.Elementary Aspects of TwoPhase Flow in Pipes R. Liquidliquid mixtures are encountered when dealing with emulsions as well as in liquidliquid extraction.
” As VG is increased. Now. Further increase in VG leads to annular mist flow through the intermediate stage of slug flow. 2 . 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.Consider a mixture that is almost completely liquid. As the gas flow rate is increased. This type of flow is called “Stratified Flow. This means that the ratio VL / VG is large. there is enough gas to form a layer at the top of the pipe. consider lower values of the ratio VL / VG . In this case. waves begin to appear on the surface of the liquid. The sketches show what happens in this situation. High VL/VG As VG Bubble Flow Lower VL/VG As VG Gas Liquid Stratified Flow Plug Flow Gas Liquid Wavy Stratified Flow Further increase in VG leads to higher waves Gas Liquid Slug Flow Now. these new terms are defined later. the bubbles become larger as shown on the right.
Holland and R. Green. D. Bragg. Fluid Flow for Chemical Engineers. Maloney). Gas Annular mist flow References F. London. In the figure.” In this case. we encounter a flow regime that is known as “annular mist flow.H. 1997 (Ed: R. Most of the liquid is entrained in the gas in the form of small drops. leading to the term “Slug Flow. with the gas flowing in the core. Perry’s Chemical Engineers Handbook.A. Perry. the liquid drops entrained in the gas are exaggerated in size.O. thus forming an annulus.” Further increase in the gas velocity leads to the formation of higher wave crests that contact the pipe wall. 1995 Edward Arnold Publishers.and this type of flow is called “Wavy Stratified Flow. and J. The gas then flows in the form of slugs.W. this is the reason it is called a mist. New York.” Eventually. 3 . 7th Edition. McGrawHill. the liquid flows in the form of a thin film on the wall of the tube. at large values of the gas velocity.
Neglect end effects: = 0 (fully developed flow) ∂z ∂v 4. r P0 z R PL We make the following simplifying assumptions. z in which z represents the direction of flow. we consider the problem in Homework # 7. in conjunction with assumptions 3 and 4. This implies that vr = 0 . 1. can be written in component form in cylindrical polar coordinates. the 1 . vθ = 0 ∂θ The incompressible version of the equation of continuity. The components in the r and θ directions yield the result that the dynamic pressure P is uniform in those directions. respectively. Therefore. yields ∂ ( rvr ) = 0 ∂r Therefore. Newtonian flow with constant viscosity µ ∂v 3.Laminar flow that starts from a state of rest in a straight circular tube R. it must be zero everywhere. (1) rvr must be constant across the crosssection of the tube. We shall use a cylindrical polar coordinate system r . and r and θ ( ) stand for the radial and angular directions. Incompressible flow: Continuity reduces to ∇ • v = 0 2. subject to assumptions 1 and 2. Shankar Subramanian Here. The NavierStokes equation. Because it is zero at r = 0 . Therefore.θ . Symmetry in θ (known as axial symmetry): = 0 . the flow is unidirectional and vz is the only nonzero velocity component.
and also about the factor 4 that is introduced in that scale. so that we can replace ∂P by its ∂z ∆P = P0 − PL . 2 y= r .dynamic pressure can. at best. ν is the kinematic viscosity (µ / ρ ). It can be shown that the pressure gradient becomes steady rapidly in such a system. R ) = 0 (5) (6) (7) (No Slip) Now. while the right side can only depend on t and r . y ) = vz ( ∆PR 2 / 4µ L ) In the above. The velocity scale is obtained by considering the steady problem. r in the tube. r ) = 0 vz ( t . depend only on t . we proceed to scale this problem by defining the following dimensionless variables. R V (T .0 ) is finite vz ( t . steady representation PL − P0 ∆P = − L L where we have defined Therefore. the 2 . In that problem. T= νt R . Initial Condition: Boundary Conditions: vz ( 0. The component in the z direction yields the following governing equation for the velocity vz t . we conclude that each can. at best. Equation (2) can be rewritten as ρ ∂vz ∆P µ ∂ ⎛ ∂vz ⎞ = + ⎟ ⎜r L r ∂r ⎝ ∂r ⎠ ∂t (4) The initial and boundary conditions are written as follows. You may wonder about the choice of the scale for the velocity. depend only on t and z . ( ) ρ ∂vz ∂P µ ∂ ⎛ ∂vz ⎞ =− + ⎜r ⎟ ∂t ∂z r ∂r ⎝ ∂r ⎠ (2) By rearranging this equation as ∂v ∂P µ ∂ ⎛ ∂vz ⎞ r = −ρ z (3) ⎜ ⎟ ∂t ∂z r ∂r ⎝ ∂r ⎠ and recognizing that the left side can only depend on t and z .
Nondimensionalization leads to ∂V 1 ∂ ⎛ ∂V ⎞ =4+ y ∂T y ∂y ⎜ ∂y ⎟ ⎝ ⎠ The scaled velocity field satisfies the following initial and boundary conditions.characteristic velocity is of the order ∆PR 2 / ( µ L ) . V (T . we first write the solution as the sum of a steady state solution and a transient solution. as you will see a bit later. y ) = Vs ( y ) + Vt (T .1) = 0 (11) From the discussion in class.0 ) is finite No Slip: (10) V (T . Initial Condition: (8) V (0. 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. (12) 4+ along with the boundary conditions 1 d ⎛ dVs y y dy ⎜ dy ⎝ ⎞ ⎟=0 ⎠ (13) Vs ( 0 ) is finite and (14) (15) Vs (1) = 0 3 . y ) = 0 Boundary Conditions: (9) V (T . The factor 4 is used to make the steady solution appear especially simple. y ) The steady solution must satisfy Equation (8) without the time derivative. Therefore.
0 ) is finite Vt (T . and amenable to writing a product class solution. This leads to Vs ( y ) = 1 − y 2 (20) The solution of Equations (16) through (19) is accomplished by separation of variables. G (T ) and φ ( y ) must satisfy the following differential ( ) ( ) ( ) dG + λ 2G = 0 dT 1 d ⎛ dφ ⎞ 2 ⎜ y dy ⎟ + λ φ = 0 y dy ⎝ ⎠ The solution of Equation (21) is (21) (22) G = Ce − λ T 2 (23) 4 . it is the inhomogeneity in this initial condition that produces a nontrivial solution for the transient problem.We first substitute the solution in Equation (12) into Equations (8) to (11). If we substitute a trial solution of the form Vt T . 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. The initial condition is different from that on the complete velocity field.1) = 0 (16) (17) (18) (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. y ) = − Vs ( y ) Vt (T . In fact. The boundary conditions in the y coordinate are unaffected. We can obtain the steady field by integrating Equation (13) and applying the boundary conditions. ∂Vt 1 ∂ ⎛ ∂Vt ⎞ = y ⎟ ∂T y ∂y ⎜ ∂y ⎠ ⎝ Vt (0. y = G T φ y into Equation (16) we find that the functions equations.
Information about these functions can be obtained from Abramowitz and Stegun (1965). of order zero. the Bessel functions J p α y and Yp (α y ) of integer order p are the two linearly independent solutions of Bessel’s 1 d ⎛ dφ ⎞ 2 2 ⎜ y dy ⎟ + (α − p ) φ = 0 y dy ⎝ ⎠ ( ) equation (25) Here are sample graphs showing the behavior of the functions followed by Y0 ( x ) and Y1 ( x ) . In general. we simply use the final result without going through the solution process. The general solution of Equation (22) can be obtained by using Frobenius series. respectively.where C is an arbitrary constant of integration. J 0 ( x) and J1 ( x) . φ = C1 J 0 ( λ y ) + C2Y0 ( λ y ) (24) where J 0 and Y0 are known as the Bessel functions of the first and second kinds. For our purposes. 5 .
labeled ( ) j0.For our purposes. in Abramowitz 6 . This leaves us with a product class solution Vt (T . that is. This is not compatible with the boundary condition stated in Equation (18) that Vt remain finite at y = 0 .) . page 409. 2.5. it has an infinite number of positive roots that come paired with negative roots of the same magnitude. In fact. labeled λn ( n = 1. Therefore. Now.3.s may be found in Table 9. this equation has more than one root. The first 20 roots of Equation (27). they approach minus infinity as x approaches zero. y ) know that this solution satisfies the governing differential equation for and the condition that the solution must remain finite at y = 0 .. it is important only to note that the functions Yp ( x ) are singular at x = 0 . The noslip boundary condition at the wall that leads to Equation (19) can be satisfied by requiring J0 (λ ) = 0 (27) As you can see from the sketch showing the behavior of J 0 . the constant C2 in Equation (24) must be set equal to zero.. y ) = Be − λ T J 0 ( λ y ) 2 (26) where the product of the two arbitrary constants C and C1 is written as a new constant B . We Vt (T . It is known that J 0 − x = J 0 x so that we ( ) can simply use the infinite set of positive roots. we need to apply the remaining conditions.
additional roots can Corresponding to each root of Equation (27). 0 n 0 m 0 1 m≠n (30) ⎡ J1 ( λn ) ⎤ ⎦ provided J λ = 0 y J 02 ( λn y ) dy = ⎣ 0( n) ∫ 2 0 d ⎡ y p J p ( λ y ) ⎤ = λ y p J p −1 ( λ y ) ⎦ dy ⎣ 2p J p −1 ( x ) + J p +1 ( x ) = J p ( x) x 1 2 (31) (32) (33) The procedure is as follows. You’ll see that be generated by approximating the interval between each by π .19 ≈ π .20 − j0. We can obtain them by Vt ( 0. y ) = − Vs ( y ) = ∑ Bn J 0 ( λn y ) n =1 (29) To evaluate the constants. ∫ y J ( λ y ) J ( λ y ) dy = 0. ∞ Bn . j0. which leads us to write the solution in the form Vt (T . More general results can be found in Abramowitz and Stegun (1965). We still need to satisfy the initial condition given in Equation (17). y ) = ∑ Bn e− λn T J 0 ( λn y ) 2 ∞ (28) n =1 All that remains is the determination of the constants applying the initial condition. This can be done by using a sum of all these solutions (recall that in a linear problem. we need to use the following results from Hildebrand (1976). we can use superposition). Begin with Equation (29) in the form −Vs ( y ) = ∑ Bn J 0 ( λn y ) n =1 ∞ (34) 7 .and Stegun (1965). 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. Therefore.
you’ll need to use integration by parts in conjunction with Equation (32). is Bm = − 3 λm J1 ( λm ) 8 (36) Substitution of this result after replacing the index complete solution from Equation (12) as follows. ∞ m with n permits us to write the V (T . One of these can be found immediately by using Equation (32). is displayed below. To find the other. The final result. 8 . y ) = 1 − y − 8 ∑ 2 J 0 ( λn y ) − λn2T e λn3 J1 ( λn ) n =1 (37) The velocity field. Use Equations (30) and (31) to obtain Bm = − ⎡ J1 ⎣ ∫ y V ( y ) J ( λ y ) dy ( λ )⎤ ⎦ 2 s 0 m m 0 2 1 (35) Evaluate the integral in Equation (35) by splitting it into the sum of two integrals.Multiply both sides by to yJ 0 ( λm y ) where m is some integer. and integrate with respect y from y = 0 to 1. calculated from Equation (37) at a few selected values of T. after simplification using Equation (33).
Dover. Advanced Calculus for Applications.References F. 9 . M. Stegun. Handbook of Mathematical Functions. PrenticeHall.B. 1965.A. New York. Englewood Cliffs. 1976. Hildebrand. Abramowitz and I.
we can calculate the losses in a sudden expansion.” 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.” and abbreviated as “CV. Shankar Subramanian This is a brief introduction to the subject of “integral balances.” These balances include statements of conservation of mass. design a jet ejector. energy. called a “control volume. The idea behind these balances is that a certain entity is “conserved. An arbitrary control volume V is shown in the sketch. and will prove useful in a variety of problems.” For instance.Introduction to Integral Balances and Conservation of Mass R. n dA dV V 1 A . by using these balances together. if we select an arbitrary volume fixed in space. and momentum. Conservation of mass and energy allow us to size pumps and turbines. Conservation of momentum is used in calculating the forces on the supports used for pipe bends and the forces flange bolts need to withstand. and make calculations involving pipe manifolds. 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. For example. and help in the evaluation of flow rates using flow measurement devices. Also.
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 ρ . By adding up all the differential volumes within the total volume fluid M . Therefore, we can write
V , we can obtain the total mass of
M=
V
∫ ρ dV
(2)
The time rate of change of this mass is then
dM . Therefore, dt dM d = ∫ ρ dV dt dt V
(3)
Now, we need to develop a result for the net rate of entry of fluid into the control volume through the control surface. For this, we consider the differential area element dA . If the velocity vector is V , the component of this velocity that is directed into the control volume is given by − V • n , because the unit normal vector n points outward from the control volume. This result, multiplied by the area of the element dA , gives the volumetric rate at which fluid enters the control volume through this area element, labeled dQ .
dQ = − V • n dA
(4)
The corresponding rate at which mass enters the control volume through the area element dA , labeled dm , can be written as
dm = ρ dQ = − ρ ( 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.
2
m = − ∫ ρ ( V • n ) dA
CS
(6)
Now, we can rewrite the principle of conservation of mass, given in Equation (1) as
dM dt
=
−
CS
∫ ρ (V • n ) dA
(7)
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
CS
∫ ρ (V • n )
dA = 0
(8)
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 crosssection as shown in the sketch.
ρ1 , V1 , A1
ρ 2 , V2 , A2
n
n
3
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
− ρ1 V1 A1 + ρ 2 V2 A2 = 0
(9)
or
ρ1 V1 A1 = ρ 2 V2 A2
(10)
The product of the uniform velocity and the area is the volumetric flow rate Q . Therefore, we can rewrite this as
ρ1 Q1 = ρ 2 Q2
The mass flow rate
(11)
m = ρQ
so that
m = m1 = m2 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.
4
14 Gate Valve (fully open): 0.0 . losses as hm = K L 2g Typical values of K L for some common fittings are given below.2. elbows.1. we can write minor 2g V2 .2.0 Long radius 90o Elbow: 0. where K L is called the loss coefficient. respectively.21 Standard 90o Elbow: 0. Usually.5 . Therefore.0.5. Even though these losses are called minor. Globe Valve (fully open): 5. Shankar Subramanian Minor losses is a term used to describe losses that occur in fittings. and the like.07 .2 . the values depend upon the nominal pipe diameter. Losses are V2 commonly reported in velocity heads. tees.14 . they can be substantial compared to those for flow through short straight pipe segments.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.21 .4 Long radius 45o Elbow: 0.03 .80 Swing Check Valve (fully open): 2. Manufacturers’ data should be used wherever possible.1 .0. Fittings commonly used in the industry include bends. valves used to control flow. 2 d and D represent the diameters of the smaller and larger pipes. and the manner in which the valve is installed (screwed or flanged). For a 1 . unions. the Reynolds number.1 Standard 45o Elbow: 0. A velocity head is .0 Tee: 0. and of course.Minor Losses R. expansions.0. The expression for K L is given by d2 K L = 1 − 2 D Here. contractions.
d d ≥ 0.76 . For smaller values of , we D D d2 can use the empirical relation K L = 0.42 1 − . D2 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.
2
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.
Re = CD =
Here,
d eV ρ
µ
2D πρ a 2V 2
is the velocity of the disk,
V
ρ
is the density of the fluid, and
µ
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 .
a and its
d e = ( 6a H )
2
1 3
Drag Coefficient
For
Re ≤ 0.01 ,
0.01 < Re ≤ 1.5 ,
CD =
64 Re 1+ π Re 2π
For
CD =
64 (1 + 10 x ) π Re
where For 1.5 < Re ≤ 133 ,
x = −0.883 + 0.906log10 Re− 0.025(log10 Re) 2 CD = 64 1 + 0.138Re0.792 π Re
for
Clift, Grace, and Weber state that when vortex shedding begins, to
Re ,
and is constant at a value of
1.17
Re > 1,000 ,
CD becomes insensitive
and that there is some
1
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 148149, 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
CD Re is used to characterize transition to various regimes of unsteady motion of
2 T
1 3
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
CD Re the amplitude
2 T
1 3,
of the oscillation increases and the disk moves in a succession of curved arcs, a motion
2 called glidetumble. 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.
2
w = log10 Re .82−0. Shankar Subramanian Most textbooks present results for the dependence of the drag coefficient for a smooth sphere. Drops. in the form of a curve.01 < Re ≤ 20 . d pV ρ µ 8D 2 π ρ d pV 2 d p is the diameter of the sphere. Re . The symbol D stands for the drag on the sphere. on the Reynolds number. V is its velocity.6435 − 1. and µ is its viscosity.6305 ⎤ ⎦ Re ⎣ For For For log10 CD = 1.Drag on spherical particles and steady settling velocities R. 260 ≤ Re ≤ 1.1558w2 1 . and Particles. CD .05 w ) ⎤ 1 + 0. I have reproduced their results below for your use in solving problems.1935Re0. the curve for the drag coefficient versus the Reynolds number is fitted to a suitable expression. For Re ≤ 0. CD = CD = CD = 9 24 + 2 Re 24 ⎡ 0. 1978) provide correlations based on experimental data that can be used to calculate either the drag coefficient or the settling velocity directly. Re = CD = Here.1242 w + 0. The standard definitions of the Reynolds number and the drag coefficient are given below. Academic Press. and Weber (Bubbles. Clift.01 . In the results given below. Grace. Such curves are difficult to read accurately especially without a fine grid. Drag Coefficient The entire range of Reynolds numbers has been divided into 10 intervals and in each.1315Re( ⎣ ⎦ Re 24 ⎡1 + 0. 0.5 × 103 . Also. the determination of the terminal settling velocity of a sphere in a fluid using this curve is an iterative process. ρ is the density of the fluid. 20 ≤ Re ≤ 260 .
9252 × 10−7 N D − 2. log10 CD = −1. Therefore. which is the weight minus the buoyant force on the sphere.6370 w − 0. but with the understanding that it now applies with V = Vterminal .4571 + 2. W = log10 N D .For1.19 − Re You can use the above equations to evaluate the drag coefficient when needed.2 × 10 For For For 4 ≤ Re ≤ 4.37 .1049 w3 For1.1546 w2 3. Grace. Clift. For N D ≤ 73. you must perform an iterative calculation to find the answer. log10 CD = −2.9181 + 0.5558w − 0. A new dimensionless group N D = CD Re is introduced in which the terminal settling velocity does not appear.4 × 104 ≤ Re ≤ 3. To avoid doing this. you can 2 immediately evaluate the terminal settling velocity. the drag on the sphere is equal to its net weight.0636 w2 4.1w − 0. 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 . 8 × 104 CD = 0. In the equations.5 × 10 3 ≤ Re ≤ 1. and Weber also provide results that permit the direct calculation of the terminal settling velocity. Re ≤ 2. From the Reynolds number.3027 × 10−10 N D 24 2 .4 × 10 4 .3w 4 × 105 ≤ Re ≤ 106 . Recall that for terminal settling.5809 w − 0. CD = 29.7569 × 10−4 N D + 6. 6 CD = 0. We define the Reynolds number in the same way as before. log10 CD = −4. you will find that this velocity appears in both the drag coefficient and in the Reynolds number.3390 + 1. Terminal Settling Velocity If you want to calculate the terminal settling velocity of a sphere.9295w2 + 0.38 × 105 ≤ Re ≤ 4 × 105 .2 × 10 4 . Re = ND 2 3 4 − 1.38 × 105 .49 For 10 < Re .78 − 5.
35 × 103 log10 Re = −1.21728 W + 0.35 × 103 < Re ≤ 3 × 105 log10 Re = 5. 6.006344W 3 For 1.55 × 10 7 < N D ≤ 5 × 1010 . 2.55 × 107 .2 .33438 W − 0.81391 + 1.11591W 2 For 580 < N D ≤ 1.12427 W 2 + 0.37 < Re ≤ 12.33283 − 1.19007 W 2 − 0.7095 + 1.2 < Re ≤ 6.34671 W − 0. log10 Re = −1.For 73 < N D ≤ 580.007005 W 3 3 . 12.
we consider steady state heat transfer to fluid in steady flow through a tube. Axisymmetric temperature field ⇒ 4.The Graetz Problem R. 3. 2. steady temperature field.This assumption also implies incompressible Newtonian flow. Shankar Subramanian As a good model problem. C p . r Fluid at T0 R z T ( R. which can be larger or smaller than T0 . A sketch of the system is shown below. The fluid enters the tube at a temperature T0 and encounters a wall temperature at Tw . Negligible viscous dissipation ∂T ≡ 0. z ) = Tw Objective To obtain the steady temperature distribution T ( r . and to calculate the rate of heat transfer from the wall to the fluid Assumptions 1. k . z ) in the fluid. ∂θ 1 . μ . Constant physical properties ρ . A simple version of this problem was first analyzed by Graetz (1883). Steady fully developed laminar flow.
z ) = 0 ∂r 2 .9. 1 vr = 0 vθ = 0 ρC p ⎢ ⎡ ∂T ⎢ ∂t ⎣ + vr ∂T ⎤ ∂T vθ ∂T + vz + ⎥ ∂z ⎥ ∂r r ∂θ ⎦ 3 ⎡ 1 ∂ ⎛ ∂T =k⎢ ⎜r ⎢ r ∂r ⎝ ∂r ⎣ 2 2 ⎞ 1 ∂ T ∂ T⎤ + 2 + 2 ⎥ + μΦ v ⎟ 2 ∂z ⎥ ⎠ r ∂θ ⎦ 4 and therefore. (page 850) can be written as follows. 0 ) = T0 T ( R. simplified to ⎛ ⎡ 1 ∂ ⎛ ∂T r 2 ⎞ ∂T v0 ⎜1 − 2 ⎟ =α ⎢ ⎜r R ⎠ ∂z r ∂r ⎝ ∂r ⎝ ⎣ 2 ⎞ ∂ T⎤ + 2⎥ ⎟ ⎠ ∂z ⎦ where α = k / ( ρ C p ) is the thermal diffusivity of the fluid. vθ = 0 ⎛ r2 ⎞ vz ( r ) = v0 ⎜ 1 − 2 ⎟ ⎝ R ⎠ v0 : Maximum velocity existing at the centerline Energy Equation Subject to assumption (2). Equation (B. Boundary Conditions Inlet: Wall: Centerline: or T ( r . z ) is finite ∂T ( 0. z ) = Tw T ( 0.2) from Bird et al.Velocity Field Poiseuille Flow vr = 0.
because conduction will lead to some of the information about the step change in wall temperature at the inlet to propagate backward. Z ) = 0 The Péclet Number The Péclet number plays the same role in heat transport as the Reynolds number does in fluid mechanics. 0 ) = 1 θ (1. Y= . Z ) is finite ∂θ ∂Y or ( 0.Because of the appearance of the axial conduction term in the governing differential equation. ∂θ 1 ∂θ 1 (1 − Y ) ∂Z = Y ∂∂ ⎛ Y ∂Y ⎞ + Pe ⎜ ⎟ Y⎝ ⎠ 2 2 ∂ 2θ ∂Z 2 θ (Y . R Pe α This permits us to transform the governing differential equation and boundary conditions to the following form. we note that the Péclet number is the product of the Reynolds and Prandtl numbers. But actually. Pe = Rv0 α = Rv0 ν × ν = Re × Pr α The physical significance of the Péclet number can be inferred by recasting it slightly. the inlet condition written above is incompatible with the inclusion of axial conduction in the problem. where the Péclet Number Pe = 0 . T0 − Tw R Z= Rv z . Z ) = 0 θ ( 0. NonDimensionalization We shall use the following scheme for scaling (or nondimensionalizing) the variables. we should write another boundary condition in the z − coordinate. Pe = ρ v0C p ΔT ΔT k R = Rate of energy transport by convection Rate of energy transport by conduction 3 . which will obviate the need for writing a second condition in the z − coordinate. First. we’ll neglect axial conduction. As we shall see shortly. θ= T − Tw r .
ν ≈ 10 Prandtl number Pr = −6 ν = 5 .05 m m2 . and Papoutsakis et al. namely. whereas the denominator stands for the order of magnitude of the conduction flux in the radial direction.500 . α = 2 × 10−7 s s This yields. If the fluid is water. it has been established that it is safe to neglect axial conduction for Pe ≥ 100. we can see that 1 . On the other hand. and we’ll The thermal diffusivity of common liquids is typically in the range 10−7 − 2 × 10−7 s use the larger limit. convection and conduction. which is comfortably α in the laminar flow regime. Physically. there are two mechanisms for transporting energy in the axial direction. Therefore. (1980). the Reynolds number is Re = 500 . which yields a s We can solve this equation by separation of variables. Z ) = ∑ An e − λ Z φn (Y ) 2 n ∞ n =1 4 . Therefore. Acrivos (1980). If we wish to compare the rates of energy transport by these two mechanisms in the same direction. m2 . Because the Péclet number is large. The final version of the scaled energy equation is ∂θ 1 ⎛ ∂θ ⎞ (1 − Y ) ∂Z = Y ∂∂ ⎝ Y ∂Y ⎠ ⎜ ⎟ Y 2 m2 . the Pe 2 term involving axial conduction can be safely neglected. we are able to neglect transport by conduction in comparison with transport by convection. Pe = 2. Let us make a sample calculation of the Péclet number for laminar flow heat transfer in a tube. which is much larger than 100. we can multiply the Péclet number by L / R where L is a characteristic length in the axial direction. Choose R = 10 mm. in the radial direction. The method of separation of variables yields an infinite series solution for the scaled temperature field. you can consult the articles by Davis (1973). To learn about how to include axial conduction. 1 For large values of Pe . v0 = 0. We can check to see if the flow is laminar by calculating the Reynolds number. θ (Y . in the scaled energy equation. there is only a single mechanism for transport of energy.Note that the numerator represents the order of magnitude of the convective flux in the main flow direction. By performing calculations including conduction in the axial direction. because the boundary conditions in the Y − coordinate are homogeneous. namely conduction.
X ⎟ ⎝2 4 ⎠ is an arbitrary multiplicative constant. or Kummer function. X d 2W dW ⎛ λ 1 ⎞ + (1 − X ) + ⎜ − ⎟W = 0 2 dX dX ⎝ 4 2 ⎠ This is known as Kummer’s equation. X ) is the confluent where c hypergeometric function. A. 1973). and is discussed in Chapter 13 of the “Handbook of Mathematical Functions” by M. X = λY 2 W ( X ) = e 2 φ (Y ) X This leads to the following differential equation for W ( X ) . 1 d ⎛ dφ ⎜Y Y dY ⎝ dY ⎞ 2 2 ⎟ + λ (1 − Y ) φ = 0 ⎠ dφ ( 0 ) = 0 or φ ( 0 ) is finite dY φ (1) = 0 The above ordinary differential equation for φ (Y ) can be solved by applying the following transformations to both the dependent and the independent variables (Lauwerier. 1. but only one is bounded at X = 0. 1951. a ( a + 1) X 2 a M ( a. This rules out the singular solution. It is an extension of the exponential function. we must require that W ( 0 ) also remain bounded. the functions φn (Y ) are the characteristic functions or eigenfunctions of a proper SturmLiouville system. Davis. b. b. X ) = 1 + X + + b b ( b + 1) 2! + You can see that when a = b . It has two linearly independent solutions.In the above solution. Stegun. a. Abramowitz and I. and is written in the form of the following series. M ( a. Because φ ( 0 ) must be bounded. leaving us with the regular solution ⎛1 λ ⎞ W ( X ) = c M ⎜ − . The function M ( a. X ) = e X a ( a + 1) b ( b + 1) ( a + n − 1) X n ( b + n − 1) n ! + 5 .
it is possible to obtain a result for the coefficients in the solution by separation variables. even though we use the term loosely to designate {λn } as that set for convenience.7042 2 6. n λn 1 2.24 348. with an eigenfunction φn (Y ) given by n assuming positive integer values beginning from 1.57 is the set of eigenvalues. ∫ φ (Y ) φ (Y ) Y (1 − Y ) dY = 0.Application of the boundary condition at the tube wall. In the present case. 1.3127 44. which we designate as λn . φ (1) = 0 . ⎛1 λ ⎞ M ⎜ − .609 113. An = ∫ φ (Y ) Y (1 − Y ) dY 2 n 2 0 0 1 ∫ φ (Y ) Y (1 − Y ) dY 2 n 1 6 .6790 3 10. there is φn ( Y ) = e − λnY 2 2 Wn ( λnY 2 ) The first few eigenvalues are reported in the table. the orthogonality property of the eigenfunctions can be stated as follows.671 5 18. leads to the following transcendental equation for the eigenvalues. The most important property of a proper SturmLiouville system is that the eigenfunctions are orthogonal with respect to a weighting function that is specific to that system.673 4 14. 2 m n 0 1 m≠n Using this orthogonality property. Corresponding to each value λn .92 215.670 Note that technically {λn2 } λn2 7. λ ⎟ = 0 ⎝2 4 ⎠ The above equation has infinitely many discrete solutions for λ .
It can be calculated directly by using the result ∂T ( R. qw ( z ) is a function of axial position. 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. z ) ∂r but as we noted earlier. z ) dr 0 1 ∫ 2π rV (r ) dr 0 1 where the velocity field V ( r ) = v0 (1 − r 2 / R 2 ) . mix it completely. z ) h ( z ) = ∂r (Tw − Tb ) We can define a dimensionless heat transfer coefficient. Tb = ∫ 2π rV (r )T ( r . which is known as the Nusselt number. The way to experimentally determine the bulk average temperature is to collect the fluid coming out of the system at a given axial location. ∂T k ( R. ∂θ (1. The mathematical definition of the bulk average temperature was given in an earlier section. 7 . and measure its temperature.The Heat Transfer Coefficient The heat flux from the wall to the fluid. Z ) 2hR = − 2 ∂Y Nu ( Z ) = θb ( Z ) k where θb is the dimensionless bulk average temperature. 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 cupmixing average temperature Tb is introduced.
∞ 2 dφ ∑ An e−λn Z dYn (1) n =1 Nu = 2 ∞ e − λn Z dφn 2 ∑ An 2 (1) λn dY n =1 We can see that for large Z . the Nusselt number can be written as follows. to finally yield the following result. Nu 3. only the first term in the infinite series in the numerator. Nu → λ12 2 = 3.656 0 0 Dimensionless Axial Position Z 8 . as Z → ∞ .656 . and likewise the first term in the infinite series in the denominator. Nu ( Z ) = − 2 4 ∑Ae λ − n =1 n ∞ 2 nZ d φn (1) dY 2 n ∑Ae n =1 n ∞ 2 − λn Z ∫ Y (1 − Y ) φ (Y ) dY 0 1 The denominator can be simplified by using the governing differential equation for φn (Y ) . The sketch qualitatively illustrates the behavior of the Nusselt number as a function of dimensionless axial position. Therefore. is important. along with the boundary conditions.By substituting from the infinite series solution for both the numerator and the denominator.
1 ∂ ⎛ ∂T ⎞ 1 ∂ ⎛ ∂T ⎞ ∂ 2T . which is useful precisely where the orthogonal function expansion converges too slowly. The main reason for this 2 is the assistance provided by e − λn Z in accelerating convergence for sufficiently large values of Z . Curvature effects can be neglected in the radial conduction term. z ) = Tw We shall now construct the Lévêque solution which is built on the assumption that the thickness of the thermal boundary layer δ t R . Because we are only interested in the velocity distribution within the thermal boundary layer.A similar analysis is possible in the case of a uniform wall flux boundary condition. Lévêque (1928) considered the thermal entrance region in a tube and developed an alternative solution. turbulent flow. This assumption leads to the following simplifications. and other geometries besides a circular tube. 1. Defining x = R − r . The Lévêque Approximation The orthogonal function expansion solution obtained above is convergent at all values of the axial position. 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 ⎠ ⎝ ⎠ 9 . but convergence is very slow at the inlet is approached. r Fluid at T0 R z δt T ( R. Extensions of the Graetz solution by separation of variables have been made in a variety of ways. accommodating nonNewtonian flow. we expand the velocity field in a Taylor series in distance measured from the tube wall and retain the first nonzero term. derivative r can be approximated by ⎜ ⎟ ⎜R ⎟= r ∂r ⎝ ∂r ⎠ R ∂r ⎝ ∂r ⎠ ∂r 2 This means that the 2.
The scaled governing equation and boundary conditions are given below. 0 ) = T0 T ( 0. we have the following governing equation for the temperature field. Z ) = 1 10 . The above approach is simpler than working out the derivatives of vz ( r ) in the x − coordinate. we scale the temperature and axial coordinate in the same manner as before. 0) = 1 θ ( 0. Z ) = 0 θ ( ∞. and constructing the Taylor series. 3. 2X ∂θ ∂ 2θ = ∂Z ∂X 2 θ ( X . z ) = Tw T ( ∞. Beginning with the simplified energy equation in which axial conduction has been neglected already. For consistency. The temperature field T ( x.Recall that a power series obtained by any method is a Taylor series. z ) satisfies the following boundary conditions. T ( x. z ) = T0 We shall work with a dimensionless version of these equations. Because the conditions outside the thermal boundary layer are those in the fluid entering the tube. and invoking the above assumptions. θ= T − Tw T0 − Tw Z= z R Pe We define a new scaled distance from the wall via X = x / R . 2v0 x ∂T ∂ 2T =α 2 ∂x R ∂z where the chain rule has been used to transform the second derivative in r to the second derivative in x . we shall use the boundary condition T ( x → ∞ ) → T0 instead of the centerline boundary condition employed in obtaining the Graetz solution. evaluating them at the wall.
a constant. d 2F dF + 3η 2 =0 2 dη dη δ2 dδ 3 = dZ 2 11 . d 2F ⎛ d δ ⎞ dF + 2η 2 ⎜ δ 2 =0 ⎟ 2 dη ⎝ dZ ⎠ dη It is evident that the similarity hypothesis will fail unless the quantity inside the parentheses is required to be independent of Z . and therefore. That is. We make the necessary transformations using the chain rule. and is unknown at this stage. For convenience. we replaced the boundary condition at the top plate with one at an infinite distance from the suddenly moved plate. we postulate that a similarity solution exists for the temperature field in the present problem. The variable δ ( Z ) represents the scaled thermal boundary layer thickness. and used the method of combination of variables to solve the equations. we set this constant to 3/2. Therefore. ∂θ ∂η dF η dδ dF ⎛ X dδ ⎞ dF = = ⎜− 2 = − ⎟ δ dZ dη ∂Z ∂Z dη ⎝ δ dZ ⎠ dη ∂θ ∂η dF 1 dF = = ∂X ∂X dη δ dη 1 ∂η d ⎡ dF ⎤ 1 d 2F ∂ 2θ ∂ ⎡ 1 dF ⎤ 1 ∂ ⎡ dF ⎤ = = = = 2 ⎢ ⎥ δ ∂X dη ⎢ dη ⎥ δ dη 2 ∂X 2 ∂X ⎣ δ ( Z ) dη ⎦ δ ∂X ⎢ dη ⎥ ⎣ ⎦ ⎣ ⎦ Using these results. Z ) = F (η ) where the similarity variable η = X / δ ( Z ) . For small values of time in the fluid mechanical problem. It would be worthwhile for you to go back and review the notes on “combination of variables” at this stage. we have an ordinary differential equation for F (η ) and another for δ (Z ) .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. Z ) is transformed to an ordinary differential equation for F (η ) . we assume θ ( X . By invoking ideas very similar to those used in the fluid mechanical problem. the partial differential equation for θ ( X .
we must go to the boundary conditions on θ ( X . and θ ( ∞. z ) = h (Tw − T0 ) ∂r 12 . this condition collapses into the condition F ( ∞ ) = 1 obtained already from the boundary condition on the scaled temperature field as X → ∞ . we define the heat transfer coefficient in this entrance region by qw = k ∂T ( R. Z ) = 0 yields F ( 0 ) = 0 . we can approximate the bulk average temperature Tb by the temperature of the fluid entering the tube T0 . Z ) = 1 leads to F ( ∞ ) = 1 . and δ (0) = 0 Integration yields the following solution for the scaled boundary layer thickness δ ( Z ) . Z ) . 0) = F ⎜ ⎜ ⎛ X ⎞ ⎟ =1 ⎟ ⎝ δ (0) ⎠ By choosing δ ( 0 ) = 0 . discussed in the “Handbook of Mathematical Functions” by Abramowitz and Stegun. The numerical value of Γ ( 4 / 3) ≈ 1. Γ ( x ) represents the Gamma function. we have F ( 0 ) = 0. Therefore.To derive the boundary conditions on these functions. The remaining (inlet) condition gives θ ( X .120 . we see that θ ( 0. Summarizing the boundary conditions on F (η ) and δ ( Z ) . F ( ∞ ) = 1 . In a straightforward way. ⎛9 ⎞ δ (Z ) = ⎜ Z ⎟ ⎝2 ⎠ 1/ 3 The solution for F (η ) is η F (η ) = 0 ∞ ∫e ∫e 0 −γ 3 dγ dγ −γ 3 1 −γ 3 = ∫ e dγ Γ ( 4 / 3) 0 η Here. when the thermal boundary layer is thin. Heat Transfer Coefficient In the thermal entrance region.
A. 36.C. Appl. Acrivos. Ueber die Wärmeleitungsfähigkeit von Flüssigkeiten. E. Lim. 35 (1980). The extended Graetz problem with Dirichlet wall boundary conditions. 184 (1951). Heat Transfer. 51. and defining a Nusselt number Nu = 2hR / k . Trans. D. and H. 36. A2. Engg. 13. L. Annalen der Physik und Chemie 18. we can write Nu ( Z ) = 2 ∂θ 2 dF ( 0. Appl. Nu ( Z ) ≈ 1. Abramowitz and I. Z ) = (0) ∂X δ ( Z ) dη By substituting for δ ( Z ) and dF ( 0 ) . Chem. Stegun. ASME J. Newman. 79 (1883). Graetz. 91. Lauwerier. Memoires. 562 (1973). Ramkrishna.357 Pe 1/ 3 ⎛R⎞ ⎜ ⎟ ⎝z⎠ 1/ 3 Comparison with the exact solution shows this result is a good approximation in the range Pe Pe ⎛z⎞ ≤ ⎜ ⎟ ≤ 2500 ⎝ R ⎠ 50 References M. Appl. 201299.Transforming to dimensionless variables. M. Dover. Handbook of Mathematical Functions. 381415 (1928) {as cited by J. New York (1965). Sci. J. H. A. Les lois de la transmission de chaleur par convection. The extended Graetz problem at low Péclet numbers. 13 (1980). Sci. Res. Sci. Series 12. Lévêque. 305362. 177 (1969)} E. Res. Res. 13 . Can. Papoutsakis.A. we obtain the following approximate result for the dη Nusselt number in the thermal entrance region. Annales des Mines.J. Davis.
in which less dense fluid is at the bottom. in which the cool surrounding air approaches the surface of the candle. Hot Cold Fluid Hot Unstable ∇T Fluid Cold Stable ∇T You should note that density gradients can arise not only from temperature gradients. and more dense fluid at the top. and we can treat the heat transfer problem as one of conduction. In one. less dense fluid is at the top and more dense fluid at the bottom. convection will start spontaneously and significant mixing of the fluid will occur. convection will be absent. a density gradient exists in a fluid in a direction that is parallel to the gravity vector or opposite to it. It is this flow that causes the visible flame to take the shape it does. Here. This is a brief introduction that will help you understand the qualitative features of a variety of situations you might encounter. the flame is simply gas at such a high 1 . Consider a burning candle. by the way. In a stable stratification. but also from composition gradients even in an isothermal system. whereas the air laterally farther from it is relatively cooler. The more common situation that we encounter in heat transfer is one in which there is a density gradient perpendicular to the gravity vector. There are two basic scenarios in the context of natural convection. and flows in a hot plume above the flame. In an unstable stratification. rises.Natural Convection R. provided the density gradient is sufficiently large. 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. In the absence of other effects. The air next to the hot candle flame is hot. we restrict our discussion to the case when temperature gradients are the source of the density gradients. Such situations can lead to “stable” or “unstable” density stratification of the fluid. This will set up a natural convection flow around the candle. It is caused by the action of density gradients in conjunction with a gravitational field.
Natural Convection adjacent to a heated vertical surface Consider a hot vertical surface present in a fluid. the convection helps the heat transfer process here. for example. and the hot air rises.temperature that it radiates visible light. the smoke from the cigarette actually traces that flow for us. Another example is the flow of air at the tip of a lit cigarette. becoming heated in the process. and also when it is a viscous liquid with a Prandtl number that is large compared with unity. in this case. 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 ∞ . the baseboard heater consists of a tube through which hot water flows. In the absence of gravity. and the heater is placed close to the floor. Clearly. Because the profile was sketched freehand in PowerPoint. Note that in this type of boundary layer. A sample velocity profile in the momentum boundary layer is shown. You should try to make a sketch of the thermal boundary layer on the same plate when the fluid is air. δm Ts Te As shown in the sketch. but also at the edge of the boundary layer. I am unable to show the smooth approach to zero velocity with a zero slope at the edge of the boundary layer properly. The surface is maintained at a temperature Ts . and the momentum boundary layer grows in thickness with distance along the plate. 2 . Here is a sketch of the momentum boundary layer along the plate. 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. which is larger than the ambient temperature in the fluid Te . The neighboring air is heated. with cooler air moving in toward the baseboard at floor level. the cold fluid rises along the plate surface. The tube is outfitted with fins to provide additional heat transfer surface. the velocity must be zero not only at the solid surface. In a common technique used for home heating. a candle flame would be spherical. but that is how the correct velocity profile would appear.
When fluid with a 1 density ρ moves at a velocity V . To provide some physical significance to this group prior to defining it. but on the scale of the picture. let us consider a typical window in a home on a winter day when the outside air is at 10 F and the inside of the room is at a balmy 68 F . 2 This must come from some other form of energy. and the direction of air flow within the boundary layer. potential energy lost by the fluid. There is a slight transverse flow in each boundary layer. I have drawn the flow in the boundary layers as being vertically downward or upward as appropriate. the difference in potential energy between the less dense fluid in the 3 . you should make a sketch of the temperature distribution along the interior and exterior surfaces of the window from the bottom to the top. What will the momentum boundary layers on either side of the window look like? Try to sketch them yourself before looking at the sketch. 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. it is difficult to use the arrows to show it. δm δm What will the thermal boundary layers look like? Try sketching them. Over a vertical distance L . therefore. an important dimensionless group is the Grashof number. 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. namely. we use a simple order of magnitude estimate of the natural convection velocity in the above examples. Also. the kinetic energy per unit volume can be written as ρV 2 .Now.
ρ V 2 ≈ g Δρ L Therefore. and is given the name Grashof Number. We can equate these two order of magnitude estimates. T ⎜ ⎟ =ρ ⎜ ⎟ =− ⎜ V ⎝ ∂T ⎠ P ρ ⎝ ∂T ⎟ P ∂T ⎝ ρ ⎠ p ⎠ temperature of the fluid and P is its pressure. This is a dimensionless group that occurs often in natural convection L ν2 problems. a typical order of magnitude of the velocity arising from natural convection is V≈ Δρ ρ gL Let us define a Reynolds number for the flowing fluid using this order of magnitude estimate. Therefore. density decreases.boundary layer and the more dense fluid outside it can be approximately expressed as g Δρ L . where g is the magnitude of the acceleration due to gravity. Δρ Re L = LV = ρ ν ν2 Δρ g L3 ρ so that Re 2 = . abbreviated as Gr. we can write β= is the ΔT ⎛ ∂ρ ⎞ = β ΔT where we have used a minus sign in relating Δ ρ to ΔT because ρ ρ ⎜ ∂T ⎟ p ⎝ ⎠ both are defined as being positive. and neglect the factor of 1/ 2 . and Δρ is a characteristic density difference between the boundary layer fluid and that far away. because this is only an order of magnitude analysis. and as temperature increases. = − Δρ We can finally rewrite the definition of the Grashof number as follows. β ΔT g L3 Gr = ν2 4 . g L3 Δρ Gr = ρ ν2 g L3 The coefficient of volumetric expansion of a fluid β is defined as 1 ⎛ ∂V ⎞ 1 ⎛ ∂ρ ⎞ ∂ ⎛1⎞ where V is the specific volume.
called the Rayleigh number. For other objects of arbitrary shape. defined as follows. Ra ≤ 109 −8 Nuaverage = 0. k Nuaverage = 0. and aspect ratio parameters. Mills points out that at Ra = 109 the above two correlations do not coincide in their predictions. physical properties should be evaluated at the arithmetic average temperature between the plate and the ambient fluid.670 ( Ra Ψ ) 1/ 4 (1 + 1.68 + 0.68 + 0. ⎡ ⎛ 0. Therefore. The Nusselt number in natural convection heat transfer situations is typically a function of the Rayleigh number. Mills (1) suggests using the following correlation for the h L average Nusselt Number. he recommends a correlation due to Lienhard. This is fine. which is the product of the Grashof and Prandtl numbers. the Prandtl number. Ra = Gr × Pr = β ΔT g L3 να Here. Other natural convection flows Mills recommends suitable correlations for natural convection flow over a horizontal heated cylinder and a heated sphere. we encounter another dimensionless group. As usual. the Prandtl number plays a significant role.6 × 10 Ra Ψ ) 1/12 . For a vertical heated plate of length L . α is the thermal diffusivity of the fluid. It is fine to use the second (turbulent) correlation for Ra = 109 . 109 ≤ Ra < 1012 In these equations Ψ is a function of the Prandtl number. 5 .492 ⎞9 /16 ⎤ Ψ = ⎢1 + ⎜ ⎟ ⎥ ⎢ ⎝ Pr ⎠ ⎥ ⎣ ⎦ 16 / 9 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. in natural convection heat transfer.670 ( Ra Ψ ) 1/ 4 .The Grashof number is related to the Reynolds number. Nuaverage = average (where k is the thermal conductivity of the fluid). because that value of the Rayleigh number is an arbitrary crossover point from one correlation to the other. and in heat transfer. abbreviated by Ra .
the gap between the two plates is evacuated. Note that even though there is convection in the air. and are each maintained at a uniform temperature. for example. L = π R for a cylinder or sphere of radius R . Over time. and the convection actually increases the heat flux through the gap over that occurring due to pure conduction. at least initially when the window is new. At larger gap widths. the velocity in the gap increases proportionally to the cube of the gap width. The sketch given below is taken from a textbook by Bird et al.Nuaverage = 0. because the temperature distribution still remains linear at this order of approximation.” A single pane window is a glass plate that separates the inside of a room from the outside. This is the reason for the choice of a small width of the order of 12 mm for the air gap in thermopane windows. it depicts the temperature distribution between the two plates and the resulting natural convection velocity distribution. the “thermopane” window was designed to reduce the heat loss by using two glass plates with a small gap between them. (2). and the heat transfer rate can be large. It is worth noting that as the gap width is increased. In modern thermopane windows. that the plates are wide and long. 6 . and for the sake of simplicity. increasing heat loss in the winter and heat gain in the summer.52 Ra1/ 4 Here the length L to be used in both the Nusselt and Rayleigh numbers is the length of the boundary layer. the temperature profile is no longer linear. air leaks through the seals into the gap. Mills also provides some useful correlations for natural convection in enclosures. Let us assume the gap is filled with air. Heat transfer between the indoor air and the air outside occurs by conduction through the glass. 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. But for those two geometries it is better to use the specific correlations given in the textbook by Mills. it does not influence the heat flux through the air gap. Therefore. so that the heat transfer rate is further reduced. Windows used in homes are termed “singlepane” or “thermopane.
Gebhart. and Sammakia. a good reference is the book by Gebhart et al. BuoyancyInduced Flows and Transport. B.N... References 1. Jaluria..E.F. R..L. and Lightfoot. 2. Bird. B.. Stewart. Y. Cambridge (1988).. PrenticeHall. 7 . Mahajan. Hemisphere. Wiley. A. Mills. New York 2007). W. New Jersey (1999).. Transport Phenomena. (2).B. R.. Heat Transfer. E. 3.If you wish to learn more about natural convection heat transfer.
Energy Transport R. 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. When the thermal energy equation is obtained for a multicomponent system. C p is the specific heat at constant pressure. Shankar Subramanian Transport of thermal energy in fluids occurs by three mechanisms. In most situations. The most common version that we shall use assumes that the density ρ and thermal conductivity k are constant.7 of Bird et al. we shall only consider conduction and convection. ρC p ⎜ ⎛ ∂T ⎞ + v • ∇T ⎟ = k ∇ 2T + µ Φ v + S ⎝ ∂t ⎠ In the above equation. obtained by taking a dot product of Cauchy’s equation with the velocity) is subtracted. Conservation of Energy The energy of a flowing fluid consists of internal and kinetic energy. the proper treatment of radiation is beyond the scope of our course. 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. Various versions are given in Table 11. and S represents the rate of generation of energy per unit volume by sources such as electrical heating. the exceptions occurring when highly viscous fluids 1 . the final form of the equation of conservation of thermal energy is obtained. The initial balance is written for the total energy and then the part involving the kinetic energy (known as the mechanical energy balance. go through a careful derivation of the mathematical form of the equation of conservation of energy in Chapter 11. v is the (vector) velocity. Bird et al. Therefore. Detailed expressions for the dissipation function Φ v in terms of derivatives of the velocity components in common coordinate systems can be found in Table B. T is the temperature. Conduction or molecular transport Convection or bulk transport Radiation Of these. µ Φ v stands for the rate of irreversible conversion of mechanical energy into internal energy per unit volume by viscous dissipation. t is time. and is given below. we can set the viscous dissipation term to zero with negligible error.41 of the book. After using some thermodynamic relationships.
A small portion of the interface between phases I and II is shown schematically in the sketch. One is molecular flux of energy and the other is convective flux of energy. in Section 9. The negative sign tells us that heat flows in the direction opposite to that of the temperature gradient. and T is the temperature field at a given point. A more precise accounting of the convective flux of total energy. except when the heat flux across the interface is extremely large. k is known as the thermal conductivity and is a material property of the fluid. q is the (vector) heat flux. Here. The constant of proportionality in Fourier’s law.7. but implied from hereon). and Bird et al. This is known as convective transport. or that between a fluid and another fluid.2 provides results for the components of q in common coordinate systems. The thermal conductivity depends on temperature and pressure in general. n II I t 2 . The phenomenological relationship q = − k ∇T is known as Fourier’s law.are subjected to large velocity gradients. 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. provide some information regarding this subject in Chapter 9. is given in Bird et al. it is reasonable to expect thermodynamic equilibrium to prevail between the two phases adjoining the interface. Boundary Conditions At the interface between a solid and a fluid. In our simplified picture. the (vector) convective flux of thermal energy at a given point can be written as Flux = ρ C p v (T − Tref ) where Tref is a reference temperature that serves as a datum. There are two fluxes of thermal energy that appear in the energy equation (the prefix “thermal” will be omitted. Convective or Bulk Transport Thermal energy also is transported by the physical movement of an element of fluid from one place to another. namely from hot regions to cold regions. Table B. an example where viscous heating is important is polymer processing.
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. n • qI = n • qII at the interface. Let us assume that Tw > T∞ for the sake of definiteness. which is maintained at a uniform temperature Tw . 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. along with any other applicable initial and boundary conditions. Here. In practice. the Nusselt Number Nu . Consider a fluid at a temperature T∞ flowing at a uniform velocity U ∞ that encounters a rigid wall. any heat flux crossing the interface from one phase must necessarily be transmitted to the other phase. 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. Therefore. are in common use in engineering work.The assumption of thermodynamic equilibrium at the interface leads to the boundary condition TI = TII at the interface. U ∞ . or the heat flux from the solid to the fluid is prescribed. Experimentally. T∞ y x Tw δt 3 . because the interface is assumed to have no mass. the results given below are equally valid when Tw < T∞ . These two boundary conditions represent the two extremes of a family of boundary conditions that are common in heat transport problems. I discuss how the heat transfer coefficient is defined in typical situations. In addition. The heat transfer coefficient The heat transfer coefficient h and its dimensionless counterpart. however.
consistent with a larger heat flux. The Nusselt number Nu is a dimensionless version of the heat transfer coefficient. The thickness of the thermal boundary layer δ t also grows with distance x . and the temperature gradient at the wall becomes sharper. As the velocity of the fluid increases. the heat transfer coefficient is seen to be directly related to the temperature gradient at the wall. ∂T ( x. Nu = hL k 4 . this can be written as qw ( x ) = − k ∂T ( x. You can see that a sharper temperature gradient at the wall is consistent with a thinner thermal boundary layer. 0 ) ∂y h ( x) = (Tw − T∞ ) −k This definition of h ( x ) holds regardless of the sign of (Tw − T∞ ) . From Fourier’s law. It is defined in the above problem as follows. the ability of the fluid to carry away heat supplied by the plate increases. 0 ) ∂y Note that qw > 0 when Tw > T∞ . A heat transfer coefficient h ( x ) is defined for this system as follows. Even though qw is purely a conduction flux. the temperature of the fluid changes from Tw to T∞ in the thermal boundary layer. the heat flux qw ( x ) can be evaluated as a function of x .Just as a momentum boundary layer forms at the wall and grows in thickness with distance x along the plate. 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 . qw ( x ) = h ( x )(Tw − T∞ ) Therefore. it is modified by flow. If we can solve the energy equation for the temperature distribution in the flowing fluid. and qw < 0 when Tw < T∞ . a thermal boundary layer forms at the wall.
z ) = Tw At any given axial position z . the heat flux from the wall to the fluid is given by qw = k ∂T ( R. consider heat transfer to a fluid flowing through a pipe. depending on our needs. Now. The following mathematical definition directly follows from this physical definition. While it is possible to use (Tw − T0 ) . let the fluid enter the pipe at some temperature T0 . For defining the heat transfer coefficient. encountering a step change in wall temperature to Tw . 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. A sketch of the system is given below.Here L is a characteristic length scale. In this case. 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. V ( r ) represents the velocity field. and then measuring its temperature. Tb = ∫ 2π rV (r )T ( r . r Fluid at T0 R z T ( R. z ) dr 0 1 ∫ 2π rV (r ) dr 0 1 Here. 5 . we need a driving force at any location z . the more common choice is (Tw − Tb ) where Tb is known as the bulk or cupmixing average temperature. which can be taken as the axial distance x or the length of the plate.
to make it convenient to report and use the results. Pr. we can use dimensional analysis to identify the dimensionless groups on which Nu will depend. The ratio of the flux of thermal energy to the driving force expressed as a temperature difference is reported as the heat transfer coefficient. Typically. it is not common in engineering design to use local values. z ) h ( z ) = ∂r (Tw − Tb ) k if we know the detailed temperature distribution in the fluid. Important dimensionless groups in heat transfer We already have defined a dimensionless group commonly used in engineering practice.The heat transfer coefficient in this system is defined as follows. In problems amenable to analysis from first principles. temperature distributions and heat fluxes can be calculated directly from the solution. we obtain L⎞ ⎛ Nu = Nu ⎜ Re. There is really no need to define a heat transfer coefficient. Typically. h= heat flux driving force We can think of the heat transfer coefficient as the conductance (using an electrical analogy) of the system. 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 see from these two examples that the heat transfer coefficient will depend on position. spatial averages are easier to measure and report. heat transfer rates and suitably defined driving forces are both measured. ⎟ D⎠ ⎝ 6 . the system parameters. even in such problems. qw = h ( z )(Tw − Tb ) so that we can evaluate h ( z ) using ∂T ( R. and on time in unsteady state problems. This concept of a heat transfer coefficient is extended to many practical heat transfer situations. namely. in any given system. suitably defined heat transfer coefficients are calculated from the theoretical results and reported. the Nusselt number Nu . But. Thus.
Nu = h ∆T ∆T k D The Prandtl number To appreciate the physical significance of the Prandtl number.hD UD ν . and ν and α are the kinematic viscosity and the thermal diffusivity of the fluid. τ yx = − ν ∂ ( ρ vx ) ∂y 7 . and the Prandtl number Pr = . For noncircular crosssections. we note that it is the ratio of the intrinsic transport coefficients ν and α for molecular transport of momentum and energy. Consider a simple onedimensional transport situation in which fluid flows in the x − direction with a velocity vx ( y ) . the Reynolds number Re = . Fourier’s 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. Newton’s law of viscosity for the momentum flux τ yx (note that we are interpreting this symbol as the negative of the shear stress) can be written as ∂vx ∂y τ yx = − µ and in a similar onedimensional conduction problem. we can rewrite the above results in the following form. where Nu = The symbol L stands for the length of the pipe. The k ν α velocity U appearing in the Reynolds number is a characteristic velocity in the system such as the average velocity of flow. We can see this by recasting it as follows. In this case. 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. respectively. The thermal diffusivity α = k / ( ρ C p ) . respectively.
we see that the flux of x − momentum is proportional to the gradient of the concentration of that momentum. the flux of thermal energy is proportional to the gradient of the concentration of thermal energy. Pr = Ability of a fluid to transport momentum by molecular means Ability of that fluid to transport energy by molecular means Thus. with coefficients of proportionality that represent the ability of the fluid to transport momentum or energy by molecular means. and occurs in the direction opposite to that gradient.qy = − α ∂ ρ C p {T − Tref } ∂y ( ) The product ρ vx represents the amount of x − momentum in unit volume of fluid. and can be regarded as the concentration of x − momentum. So. Pr 1 (ν δm δt Tw α) U ∞ . the relative thicknesses of the momentum and thermal boundary layers reflect the magnitude of the Prandtl number. In a like manner. in flow over an object. with a coefficient of proportionality equal to the thermal diffusivity α . Pr 1 (ν α) U ∞ . Large Prandtl number. T∞ y x Tw δt δm 8 . molecular transport leads to momentum and energy flowing “downhill” in the direction opposite to that of the concentration gradient of each. Therefore. This leads to the following physical interpretation of the Prandtl number. T∞ y x Small Prandtl number. The coefficient of proportionality is the kinematic viscosity ν . as the examples given below show.
Because liquid metals are great conductors of thermal energy.For gases. Prandtl numbers for liquid metals are typically of O (10−2 ) . 9 . 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. the Prandtl number is typically of O (1) . reaching values of 105 or greater.
we only focus on the pressure drop issue. only a brief summary is given. Other fluid mechanical issues involve the proper distribution of the liquid across the crosssection. contacting the liquid in a countercurrent fashion. Here. wetting the packing material. An example of a packed bed is an absorber. catalytic chemical reactions. Here. Smith. 1 . 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. distillation) or a solid and a gas or liquid (adsorption. let us consider flow through a packed bed. catalysis) is obtained for achieving rapid mass and heat transfer. and the Reynolds number Re p . are defined as follows. You can learn about different types of packing materials from Perry’s Handbook. A gas is admitted at the bottom. and flows upward. Here. the friction factor f p for the packed bed. and particularly in the case of fluidized beds. 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. fp = 150 + 1.75 Re p Here. First. Shankar Subramanian Chemical engineering operations commonly involve the use of packed and fluidized beds. 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. and Harriott (2001) and Perry’s Handbook (1997).Flow through Packed Beds and Fluidized Beds R. Design of packing materials to achieve uniform distribution of the fluid across the crosssection throughout the column is an important subject as well. The liquid rich in the soluble species is taken out at the bottom. Packed Beds A typical packed bed is a cylindrical column that is filled with a suitable packing material. The soluble species is absorbed in the liquid. The Ergun equation that is commonly employed is given below. These are devices in which a large surface area for contact between a liquid and a gas (absorption. You will find a good deal of information about flow through packed and fluidized beds in the book by McCabe. To calculate this quantity we rely on a friction factor correlation attributed to Ergun. The liquid is distributed as uniformly as possible at the top of the column and flows downward. and the lean gas leaves the column at the top. From a fluid mechanical perspective.
Two simpler results.fp = ∆p D p ε 3 L ρVs2 1 − ε Re p = (1 − ε ) µ D pVs ρ The various symbols appearing in the above equations are defined as follows. used for flow under very viscous conditions. even though they continue to be reported in textbooks. used when viscous effects are not as important as inertia. 000 It is suggested that the student simply use the Ergun equation. One is the KozenyCarman equation. 2 . each obtained by ignoring one or the other term in the Ergun equation also are in use. ∆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 ρ : Density of the fluid µ : 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 crosssectional area of the bed) ε : Void fraction of the bed ( ε is the ratio of the void volume to the total volume of the bed) Sometimes. Re p ≥ 1. fp = 150 . we may use the concept of the interstitial velocity Vi . f p = 1. The interstitial velocity is the average velocity that prevails in ε the pores of the column. which is related to the V superficial velocity by Vi = s . Re p Re p ≤ 1 The other is the BurkePlummer equation.75 . There is no need to use these other two approximate results.
you will find fluidized beds used in catalyst regeneration. drying. 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. Thus. roasting of ores.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 particlefluid mixture behaves as though it is a fluid. and gas adsorption operations. In fact. when a bed of particles is fluidized. Both gas and liquid flows can be used to fluidize a bed of particles. 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. B E C D Pressure Drop A Vf: Minimum Fluidization Velocity O Superficial velocity D Bed Height E A B C O Superficial Velocity 3 . Nowadays. the entire bed can be transported like a fluid. if desired. leading to excellent contact of the solid and the fluid and the solid and the wall. The most common reason for fluidizing a bed is to obtain vigorous agitation of the solids in contact with the fluid. solidgas reactors. fluidized beds were used in catalytic cracking in the petroleum industry in the past. combustion of coal.
the pressure drop zero. In doing this balance. when there is no flow.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. This is because the settling velocities of the particles are typically much larger than the largest superficial velocities used. If we now repeat the experiment by increasing the superficial velocity from zero. A bed of particles. When the point B is reached. but the pressure drop stays constant. the net gravitational force on the particles (net weight) is Net Weight of the particles = (1 − ε ) ( ρ p − ρ f ) ALg 4 . we define the velocity at the point C in the figure as the minimum fluidization velocity V f . and then begins to decrease when the point C is reached. the pressure drop stays fixed while the bed settles back down. After fluidization. gradually decreasing the superficial velocity. in the direction of the reverse arrows in the figure. Upward force on the bed = ∆p A If the height of the bed at this point is L and the void fraction is ε . tracing the path ABCD. We can calculate it by balancing the net weight of the bed against the upward force exerted on the bed. and the bed has a certain height. At first. 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. but it is loosened when it is fluidized. As the velocity is increased further. becomes consolidated. It is possible to reach large superficial velocities without having the particles carried out with the fluid at the exit. namely the pressure drop across the bed ∆p multiplied by the crosssectional area of the bed A . left alone for a sufficient length of time. Now. we’ll follow the set of curves ECD in both directions. we ignore the small frictional force exerted on the wall of the column by the flowing fluid. The bed height no longer decreases while the pressure drop follows the curve CEO. Reference is made to the figure on page 3. the pressure drop gradually increases while the bed height remains fixed. the bed continues to expand in height. we find that the behavior of the bed follows the curves DCE. At first. and back down to zero. This is a region where the Ergun equation for a packed bed can be used to relate the pressure drop to the velocity. the bed starts expanding in height while the pressure drop levels off and no longer increases as the superficial velocity is increased. Because of this reason. at first. if we trace our path backward. we can write Volume of particles = (1 − ε ) AL If the acceleration due to gravity is g . As we proceed along the right arrow in the direction of increasing superficial velocity. 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.
increasing a bit with particle size. and Harriott (2001) suggest that ε lies in the range 0. For large particles ( D p ≥ 1 mm). we can obtain a result for the latter. and the full Ergun equation must be used to determine V f . for a bed of small particles ( D p ≤ 0. and values as large as 100 V f are used on occasion. we refer to the state of the bed as one of incipient fluidization. The void fraction ε at this state depends upon the material. Now. Recognizing that not all particles are of the same size and that D p is only an average size. McCabe et al. the flow conditions at this stage are such that the Reynolds number is relatively small ( Re ≤ 10 ) so that we can use the KozenyCarman Equation.45 .1 mm). given above.40 − 0. this yields a ratio ranging from 78 − 50 . we can write ( ρ p − ρ f ) gDp2 Vsettling = 18µ If we now use the result for the minimum fluidization velocity for the case of small particles.45 . Smith. for establishing the point of onset of fluidization. Restricting attention to small particles so that Stokes law can be used to calculate their settling velocity. shape. suggest that it is common to operate fluidized beds at velocities as high as 30 V f . we see that fine particles are likely to be carried out with the 5 . This would occur if the superficial velocity is equal to the settling velocity of the particles.40 − 0. McCabe. and size of the particles. For nearly spherical particles. inertial effects are important.Balancing the two yields ∆p = (1 − ε ) ( ρ p − ρ f ) Lg By using an expression relating ∆p to the superficial velocity. which is the fluidization velocity at this point. 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 yields Vf = (ρ p 2 − ρ f ) gD p ε 3 150 µ 1− ε When the superficial velocity Vs is equal to V f . When in doubt. we see that the ratio Vsettling 25 1 − ε = 3 ε3 Vf For ε lying in the range 0. Typically. applicable to the viscous flow regime. use the Ergun equation instead of a simplified version of it.
in order to obtain the benefits of operating a bed at such large superficial velocities. Particulate fluidization occurs in liquids. Smith. and P. and then successive bubbles move up the column. Fluidization can be broadly classified into particulate fluidization or bubbling fluidization.H. and J. and uniform conditions prevail in the liquid solid mixture. The bubbles grow until they fill the crosssection. Maloney). very dense particles in a liquid can show “bubbling” and gases at high pressure when flowing through beds of fine particles.exiting fluid in such a situation.Harriott 2001.E. large pockets of gas. In contrast. Unit Operations of Chemical Engineering. D. J. and at higher velocities. free of particles. the bed shows bubbling. McCabe. 7th Edition. when the bed is fluidized. 6 . Here. are seen to rise through the bed. Sometimes. New York. Usually. Green. Perry. can give rise to particulate fluidization. the bed expands uniformly. bubbling fluidization occurs in gasfluidized beds. References W.W. this occurs at lower velocities. New York. 1997 (Ed: R. Perry’s Chemical Engineers Handbook. a condition known as slugging. They can be recovered by filters or cyclone separators and returned. McGraw Hill. McGrawHill. Where there are particles. the bed void fraction is approximately at the value that prevails at the point of incipient fluidization. As the velocity of the liquid is increased past the minimum fluidization velocity.C.O. The above classification should not be interpreted rigidly.
∇v . From calculus. investigate the velocity gradient. Q′ Q P P′ We can see that if the velocity were to be the same at both ends of the element. the velocity v ( x ) must be nonuniform. it would change neither its length. the two ends have moved to new locations because of fluid motion. we must. there is simple translation of fluid elements with no deformation or rotation. at some instant of time. y. Imagine a tiny line element dx . dv . labeled PQ in the sketch. It follows that the differential change in the vector velocity. the reader is encouraged to consult any of the references listed at the end. which is a second order tensor. in a uniform velocity field. 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 . To cause either. The position vector is labeled x . To understand the nature of the changes in fluid elements brought about by the flow. We shall use rectangular Cartesian coordinates ( x. Shankar Subramanian Kinematics is the study of motion without dealing with the forces that affect motion. nor its orientation. and the new line element is labeled P′Q′ .Kinematics of Fluid Motion R. Therefore. along with the associated basis set of mutually orthogonal unit vectors ( i . is given by 1 . j. k ) . The discussion here is of limited scope and for more details. z ) . Our focus here is on fluid motion. After a small amount of time dt . therefore.
We can write its components in terms of the components of the velocity gradient as follows. 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 . ⎛ ∂vx ⎜ ∂x ⎜ ⎜ ∂v ∇v = ⎜ y ⎜ ∂x ⎜ ∂vz ⎜ ⎝ ∂x ∂vx ∂y ∂v y ∂y ∂vz ∂y ∂vx ∂z ∂v y ⎞ ⎟ ⎟ ⎟ ⎟ ∂z ⎟ ∂vz ⎟ ⎟ ∂z ⎠ Any tensor can be written as the sum of a symmetric and an antisymmetric tensor. 2 .∇ v T ) 2 The action of each of these contributions to the velocity gradient will be explored in detail next. we consider the vorticity tensor. Vorticity Tensor Ω The vorticity tensor Ω is a skewsymmetric tensor. Let us do this with the velocity gradient tensor. This tensor can be written as follows. writing it as ∇v = E + Ω 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 Ω is given by Ω= 1 ( ∇v . First.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.
Using the relationship between Ω and − ω . we obtain 2 1 1 1 Ωij dx j = − ε ijk dx jω k or in Gibbs notation.⎛ 0 ⎜ ⎜ ⎜ ⎛ ∂v ⎞ ⎜ 1 ⎜ y − ∂vx ⎟ Ω= ⎜ 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 ⎠ ⎟ ⎟ ⎟ 0 ⎟ ⎠ A skewsymmetric tensor Aij can be formed from a vector ak by writing Aij = ε ijk ak . Ω • dx = − dx × ω = ω × 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 ω . and the unit tangent vector to the circle at any point be t . t n a Apply Stokes’s theorem to the velocity field in this circle. consider a surface formed by an infinitesimal circle of radius a located at a point x . To prove this claim. where ω = ∇ × v is 2 1 known as the vorticity vector. Let the unit normal vector to the surface (perpendicular to the plane of the paper) be n . The 1 vector associated with the vorticity tensor in this manner is − ω . 2 Because a fluid does not usually rotate as a rigid body in the manner that a solid does. ∫ ( ∇ × v ) • n dS = ∫ v • t ds S C 3 . we should interpret the above statement as implying that the average angular velocity of a fluid element located at a point is onehalf the vorticity vector at that point.
One reason for working with the equations of vorticity transport is that pressure is absent as a dependent variable in those equations. ω z ) . 4 . we see ⎛1 ⎞ that this is equal to the average value of ⎜ ∇ × v ⎟ • n over the surface of the circle. It is transported from one place to another in a fluid by convective and molecular means. and an appropriate partial differential equation that governs its transport can be written. If the components of the vorticity ω = ∇ × v are (ω x . We also can show (see Batchelor. dS is an area element on the surface of the circle S and ds is a line element along the circle C. ω y . we can write the average linear velocity along the circle as ∫ v • t ds and 2π a C therefore the average angular velocity as 1 2π a 2 C ∫ v • t ds . just as energy and species are. most of the flow can be described by using the equations that apply to irrotational flow. In addition. a mechanism that is not present in the transport of energy and species. Vorticity is generated at fluidsolid interfaces and at fluidfluid interfaces. This is the reason why. in the limit as the radius of the circle approaches zero. Vortex Lines and Tubes Just as a streamline is a curve to which the velocity vector is tangent everywhere. we find that the average angular velocity around the circle approaches the value of onehalf the component of the vorticity vector in a direction perpendicular to the surface of the circle. and the fluid is inviscid (meaning the viscosity is zero). A flow in which the vorticity is zero is known as an irrotational flow. in a high Reynolds number flow (implying weak viscous effects) past a rigid body. ⎝2 ⎠ Thus. From Stokes’s theorem. vorticity also is intensified by the stretching of vortex lines. It can be shown that if a fluid mass begins with zero vorticity. with the vorticity being confined to a boundary layer near the surface of the body.Here. Vorticity cannot be generated internally within an incompressible fluid. page 82) that the angular momentum of a spherical element of fluid is equal to onehalf the vorticity times the moment of inertia of the fluid. Because v • t is the component of the velocity along the periphery of the 1 circle. the vorticity will remain zero in that fluid mass. Vorticity Vector The vorticity ω = ∇ × v . we can define a vortex line as a curve to which the vorticity is tangent everywhere. is an important entity in fluid mechanics. then we can write the equations of the space curves that are vortex lines as dx ωx = dy ωy = dz ωz . just as it is for a rigid body.
we can see that this is the circulation ∫ dS • ω . the above result can be rewritten as dt 2 ds d ( ds ) = 2 dx • dv = 2 dx • ( ∇v • dx ) = 2 dx • E • dx + 2 dx • Ω • dx dt The second term in the farrightside is zero because Ω is an antisymmetric tensor. we move on to discuss the significance of the rate of strain tensor. If we construct an open surface S bounded by this closed curve C.The surface that is formed by all the vortex lines passing through a closed reducible curve is known as a vortex tube. you can see that the role of that tensor is to describe the instantaneous angular velocity of a fluid element. we can define the strength of the vortex tube as theorem. The rate of strain tensor is a symmetric tensor. which contains all the information about the deformation. We can write its components in terms of the components of the velocity gradient as follows. ⎛ ∂vx ⎜ ∂x ⎜ ⎜ ⎛ ∂v ∂v ⎞ 1 Ε= ⎜ ⎜ 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 ⎟ ⎟ ∂z ⎠ 1 ⎛ ∂vx ∂vz + ⎜ 2 ⎝ ∂z ∂x The diagonal elements of E Consider a line element dx with a length ds . S By using Stokes’s C ∫ v • t ds where C is any closed curve around the vortex tube. we write 5 . To see this. but that it contributes nothing to deformation of elements. Now. t is a unit tangent vector to the curve at any point. d d ( ds 2 ) = dt ( dx • dx ) dt Using the fact that dx = dv . Rate of Strain or Rate of Deformation Tensor E From the above discussion of the vorticity tensor. and ds is a line element.
we first exchange the indices i and j to obtain an intermediate result. In a like manner. the rate of strain of a line element in the ∂x ∂v y ∂vz y − direction is . This is known as the rate of dilatation of a fluid element at the given location. This is the physical ∂y ∂z interpretation of the diagonal elements of the rate of strain tensor. 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. the trace of V →0 V V dt V E is the rate of increase in the volume of an infinitesimal element. dt ∫ S V From the above. the rate of strain of a line element in that direction is simply ∂v E11 . In the above result. we find that ds d ( ds ) = dx • E • dx from which. we can think of the right side of the above result as the “double projection” of the tensor E in that direction. Therefore. 6 . The sum of the diagonal elements of E . Therefore. In this case. To see why. and that in the z − direction is . consider a material body occupying a volume V enclosed by the surface S . So. the rate of dilatation is zero. 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 its volume. 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. The physical meaning is clear. we can see that lim V →0 1 dV 1 = lim ∫ ∇ • v dV = ∇ • v . after writing the result in index notation. which is equal to x . Let us choose the direction to be the x − direction.dx • Ω • dx = Ωij dxi dx j = Ω ji dx j dxi = − Ω ij dxi dx j so that Ω ij dxi dx j = 0 . known as the trace of E is ∇ • v . Let us inquire how V changes with time. 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 . The term “projection” is used in a loose sense here. When the flow is incompressible. dV = dS • v = ∫ ∇ • v dV by the divergence theorem. and then use the antisymmetry property to write Ωij = − Ω ji . and is called the rate of dilatation.
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). We can see that the above result reduces to the earlier result we obtained when the two vectors dxi and dxi′ are the same. we have used the fact that the infinitesimal change dvi′ is the change in the velocity over an infinitesimal distance in the d ( ds ds′ cos θ ) = dt ⎛ ∂v ∂v j ⎞ dx ′dx = 2 Eij dxi′dx j ⎜ i + ⎜ ∂x ∂x ⎟ i j ⎟ i ⎠ ⎝ j Dividing both sides by ds ds′ yields dx ′ dx dx 1 d dx′ ( ds ds′ cos θ ) = 2 Eij i′ j = 2 ′ • E • ds ds′ dt ds ds ds ds So. In writing the result in the second term in the second line. In this case. d d ( ds ds′ cos θ ) = ( dx • dx′) = dvi dxi′ + dxi dvi′ dt dt ∂v ∂v = i dx j dxi′ + i dx j′dxi ∂x j ∂x j direction of the vector dx′ . Q dx θ dx ′ P Q′ Let us investigate the time rate of change of the dot product of the vectors dx and dx′ .The offdiagonal elements of E Now. Let us consider the case when the two vectors are orthogonal to each other. 1 d 1 d 1 d dθ ⎤ ( ds ds′ cos θ ) = cos θ ⎡ ⎢ ds dt ( ds ) + ds′ dt ( ds′ ) ⎥ − sin θ dt ds 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. the result is the left side of the above equation. Let us work out the differentiation in the left side. we obtain 7 . consider two line elements dx and dx′ at a given point x and let the angle between them be θ . Interchanging the indices i and j in that second term permits us to combine the two terms.
Ed. Tensors. the element is E12 ( = E21 ) .” Cambridge University Press. Thus.K. 1954. 1989. “Kinematics of Vorticity. Truesdell. We can use a basis set built from these principal directions to describe the components of the tensor. rotation. References 1. This ellipsoid also rotates with an instantaneous angular velocity that is equal to the onehalf of the vorticity of the fluid at the given point. P. It deforms into an ellipsoid whose axes are aligned with the principal axes of the rate of strain tensor. Batchelor. and in fact it does. and the Basic Equations of Fluid Mechanics. Similar physical interpretations can be given to the other offdiagonal elements of the rate of strain tensor. 1959. Serrin.” Cambridge University Press. an infinitesimal spherical element of fluid undergoes translation. Aris. For example. 1967. 5. of course the entire set of principal 1 1 axes can rotate. If we choose these two orthogonal directions to coincide with any two coordinate directions. the offdiagonal elements describe shear deformation of the fluid. The offdiagonal elements will be zero. 8 . C. so that the rate of change of the angles between the principal directions is zero. R. “An Introduction to Fluid Dynamics. 4. J.” Dover Publications. 1992. 2. we can conclude that in a flow. 1962).” Handbuch der Physik VIII/1. “Mathematical Principles of Classical Fluid Dynamics. Some good sources for further study are listed below. with the angular velocity ω = ∇ × v . 3. Chapter 5. G. if we use x and y − directions.” Indiana University Press. and deformation in general. 2 2 Instantaneous Deformation of a Fluid Element Based on all of the above material on kinematics. Chapter 4 (original by PrenticeHall. Saffman. the sequential dot products of E with unit vectors in two orthogonal directions yields onehalf the rate of decrease of the angle between those directions. There are three mutually orthogonal directions associated with the symmetric tensor E that are known as its eigenvector or principal directions. “Vectors.G. the tensor will be diagonal. “Vortex Dynamics. Flugge. S. then the dot products yield the offdiagonal elements of E .dx ′ dx 1 dθ •E• = − 2 dt ds′ ds So. If we do.
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. At such a surface. 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. When they are integrated in any given situation. If we designate the velocity of the rigid surface as V and that of the fluid as v .Boundary Conditions in Fluid Mechanics R. or in flow of a rarefied gas through 1 . One is that at the interface between a fluid and a rigid surface. To evaluate these. we arrive at the conclusion that the fluid velocity must match the velocity of the rigid surface at every point on it. 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. Such statements are known as boundary conditions. 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. If n represents the unit normal. v • n =V • n on a rigid surface (kinematic condition) As a consequence of the two conditions. Conditions at a rigid boundary It is convenient for the purpose of discussion to identify two types of boundaries. and has been found to be successful in describing most practical situations. and select a unit tangent vector to the surface as t . the noslip 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. and similarly the normal component of the velocity of the fluid be the same as the normal component of the velocity of the surface. It was a subject of controversy in the eighteenth and nineteenth centuries. Usually. v =V on a rigid surface The noslip 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. we can expect to see arbitrary functions or constants appear in the solution. we need additional statements about the velocity field and possibly its gradient at the natural boundaries of the flow domain. The former is known as the “no slip” boundary condition.
Other examples include the interface between a liquid drop and the surrounding continuous phase or that between two liquid layers. The surface of the liquid film in contact with the surrounding gas is a fluidfluid interface. we can write v I • t = v II • t and at a fluidfluid interface (no slip) v I • n = v II • n at a fluidfluid interface (kinematic condition) Notice that we have two unknown vector fields v I and v II now. n • [TI − TII ] = 2 H σ n − ∇ sσ at a fluidfluid interface 2 (jump condition on the stress) . we encounter a boundary between two fluids. Therefore.a pore of diameter of the same order of magnitude as the mean free path of the gas molecules. This vector condition also can be viewed as being in two parts. A common example occurs when a liquid film flows down an inclined plane. analogous to the noslip boundary condition at a rigid boundary. Therefore. The general form of this condition is given below. and the continuity of the normal component of the two velocities. We also need to write a boundary condition connecting the state of stress in each fluid at the interface. Conditions at a fluidfluid interface Sometimes. For the types of problems that we shall encounter. it is an adequate boundary condition. a kinematic consequence when there is no mass transfer across the interface. It is convenient to designate the two fluid phases in contact as phase I and phase II . I t I 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. one on the continuity of the tangential component of the two velocities. and therefore need twice as many boundary conditions. it is not sufficient to write just the above noslip and kinematic conditions at a fluidfluid interface. n II The unit normal vector n points into phase vector to the interface t .
the stress boundary condition can be simplified. σ is the interfacial tension of the fluidfluid interface.” In the types of problems that we shall encounter. and a part that is tangential to the interface. gas x y liquid 3 . fluid mechanical problems involving the application of the normal stress balance at a boundary are complicated. the normal velocity at the free surface of the liquid is zero in both the liquid and the gas.In the stress boundary condition. the condition is written as two separate scalar boundary conditions by writing the tangential and the normal parts separately. we can write at a fluidfluid interface II τt = τt I (tangential stress balance) The normal stress jump boundary condition actually determines the curvature of the interface at the point in question. and must be solved numerically unless one assumes the shape distortion to be very small or of a particularly simple form. The sketch depicts the situation. In that case. we can further simplify the tangential stress balance. the symbols TI and TII represent the stress tensor in each fluid. we shall always assume the shape of the interface to be the static shape and as being specified. H is the mean curvature of the interface at the point where the condition is being applied. given in the second term in the right side. we call the two boundary conditions the “tangential stress balance” and the “normal stress balance. This means that the tangential stress is continuous across the interface because the jump in it is zero. then the gradient of interfacial tension will vanish everywhere on the interface. In fact. At a liquidgas interface. If we assume these to be uniform. In this situation. This shape is distorted by the flow. If τ t represents this stress component. Recall that the tangential stress is purely viscous in origin. and therefore the shape of the entire fluidfluid interface. The left side in the stress boundary condition is the difference between the stress vectors in fluids I and II at the interface. This is the reason for the choice of terminology used in describing this condition. namely the first term in the right side. we shall not be able to satisfy the balance of normal stress. The interfacial tension at a fluidfluid interface depends on the temperature and the composition of the interface. That is . Therefore. The resulting vector is decomposed into a part that is normal to the interface. Consider the surface of a liquid film flowing down an inclined plane. and ∇ s is the surface gradient operator which can be written as ∇ − n ( n • ∇ ) . or the “jump” in stress. we remove the part of the gradient vector that is normal to the surface. Let us assume that the flow is steady and that the film surface is parallel to the inclined plane. Sometimes. In the problems that we shall analyze.
Because the normal velocity is zero at the free surface. respectively. and can be considered negligible.l ∂y = µ g ∂vx . the right side of the above equation is small. This allows us to write ∂vx . 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. the correct boundary condition equating the tangential stresses must be used. the tangential stress balance simplifies to the following result where the subscripts l and g represent the liquid and gas. as is the case on a windy day when the wind causes motion in a puddle of liquid. g µl ∂y at the free surface Because the dynamic viscosity of a gas is small compared with that of a liquid. If we divide through by the dynamic viscosity of the liquid.l ∂y = µg ∂vx . 4 . this condition is represented as that of vanishing shear stress at a free liquid surface. When a gas drags a liquid along.l ∂y ≈0 at the free surface Sometimes. µl ∂vx . g ∂y at the free surface The symbol µ in the above result stands for the dynamic viscosity. we obtain ∂vx .
When a fluid flows past an object at large values of the Reynolds number. (ii) In a thin region near the surface of the object. Here. The pressure variation can be calculated from the potential flow solution along the surface of the object. and are as important as inertia. called the boundary layer. viscous effects can be considered negligible. 2. where Re x = ν . U∞ y x laminar turbulent xU ∞ Transition from laminar to turbulent flow in the boundary layer on a flat plate occurs at Re x ≈ 5 × 105 . viscous effects cannot be neglected. the flow region can be divided into two parts.Elements of Prandtl’s Boundary Layer Theory R. Assumptions 1. and assumed to be impressed upon the boundary layer. and potential flow can be assumed. 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. ν is the kinematic viscosity of the fluid. 1 . neglecting viscous effects altogether. Prandtl’s principal assumptions are listed below. (i) Away from the surface of the object.
We can estimate the order of ∂y U∞ δ . We can estimate the order of magnitude of each of these terms for a plate of length L as follows. It is a valuable tool in dealing with transport problems. which can be recast as L δ2 ρU ∞ δ L ∼ 1 Re L where the Reynolds number based on the length of the plate Re L = LU ∞ ν . This type of argument is called a scaling analysis. and a typical viscous term is µ 2 . The shear stress at the plate surface is τ w = µ this quantity as τ w ( x ) = µ ∂u ( x. these two order estimates must be comparable. In a like manner. we can find a scale estimate of the drag as well. A typical inertia term in the NavierStokes equation in rectangular Cartesian ∂u ∂ 2u coordinates is ρ u . y ) directions. You can see that it provides not only an idea of the variables on which key quantities depend. v ) are ∂x ∂y the velocity components in the ( x. but also the form of this dependence without having to solve the partial differential equations involved. 2 U∞ ∂u ρu ∼ ρ ∂x L U ∂ 2u µ 2 ∼µ ∞ ∂y δ2 Because the viscous force in the boundary layer is of comparable order to the inertia force.The assumptions can be used to establish the order of magnitude of the boundary layer thickness. Because the shear stress is a local quantity.0 ) . 2 U∞ U µL ρ ∼ µ ∞ or δ 2 ∼ . Here. 2 . ( u . and ρ and µ are the density and the dynamic viscosity of the fluid.
From the order of magnitude argument used earlier. 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. x Ignoring the numerical factor of 2 that appears after performing the integration (because we are only estimating the order of magnitude). U ∞δ1 = ∫ (U − u ) dy 0 ∞ or δ1 = ∫ 1 − 0 ∞ u U∞ dy 3 . we can estimate it as δ ( x ) ∼ x = Re x L νx U∞ . we can write 3 D ∼ w ρ µ U∞ L A rigorous calculation from boundary layer theory yields the result 3 D = 0. If the width of the plate in the z − direction is w .we should use an order of magnitude of the variation of the boundary layer thickness δ with x .664 w ρ µ U ∞ L confirming the correctness of our scaling argument. We can construct a mathematical definition in the case of the flat plate by recognizing that the displacement thickness δ1 is that thickness of the uniform stream that accounts for the “lost” flow because of the presence of the solid surface. the drag on the plate surface is given by D = w ∫ τ w ( x ) dx = µ w 0 L 3/ U∞ 2 ν 1/ 2 ∫ 0 dx .
while the scaled variable x also varies from 0 to 1 (we say x ∼ O (1) ). 1. respectively. From the continuity ∂x ∂u ∂v and must sum to zero. Note that the order of magnitude of a quantity is the same regardless of its sign. For more details. because the maximum value of the physical velocity is that of the uniform stream approaching the plate. precisely. δ is the scaled boundary layer thickness. This means that the magnitude of u lies between 0 and a number that is of the order of unity. you can consult Schlichting [1]. We know that the variable y ∼ O (δ ) where ∂y δ represents the boundary layer thickness divided by the length L . But this is not necessarily the meaning implied by the order symbol that we are using. We shall use scaled variables. we can conclude that the derivative equation.Order of magnitude analysis of the continuity and NavierStokes Equations Now. we know that u is O (1) . The symbols x and y are used for the scaled counterparts of the physical coordinates in the sketch. and U ∞ as a reference velocity. Because the derivative ∂v to be O (1) . namely U ∞ . we see that derivative ∂u ∂v + =0 ∂x ∂y ∂u is O (1) as well. In this particular case. the maximum value of u is. using L as a reference length. in fact. and the symbols u and v are used for the dimensionless counterparts of the physical velocity components in the x and y directions. In other 4 . this forces the ∂x ∂y words. Because the velocity u varies in the range mentioned above. The scaled incompressible version of the continuity equation is From the scaling. we shall go through an order of magnitude analysis of the twodimensional NavierStokes equations for steady incompressible Newtonian laminar flow over a flat plate and simplify them using Prandtl’s ideas.
v . ∂u ∂u u + v ∂x ∂y 1 1 = δ 1 δ ∂ 2u ∂p 1 ∂ 2u + − + Re L ∂x 2 ∂y 2 ∂x 1 δ2 1 2 δ ∂u ∂u . ∂x 2 5 . First. To estimate the order of magnitude of . following Schlichting [1] we proceed to use similar arguments in the two components of the NavierStokes equation applicable to this situation. To estimate the to δ . So. We know from the kinematic condition that v = 0 at the surface y = 0 . This is the reason for the estimate that ∂y δ order of magnitude of the second derivatives. consider the derivative . In a like manner. we must use similar Below each term in the equation. the magnitude of v must of O (δ ) . we must conclude that the change in the scaled velocity ∂y component v across the boundary layer must be of O (δ ) . We express this fact by stating δ 1. the scaled velocity in the y − direction in the boundary is a very small quantity. We already have discussed the order of magnitude of u . Now. and ∂ 2u arguments. Therefore. 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. We note that δ is a very small quantity when the Reynolds number Re L 1. We know that ∂x 2 ∂u ∼ O (1) . we have written the order of magnitude of that term.∂v ∼ O (1) . while the variable y varies from 0 ∂u 1 ∼ O . we first note that u varies ∂y ∂x from 0 to 1 across the boundary layer. Therefore. For example. This is ∂ 2u the reason for estimating the order of as being unity. consider the x − component.
This means ( ) 6 . and can be ignored. the second ∂u 1 1 ∼ O .the derivative the boundary layer. we see that the viscous force in the x − direction is negligible when compared to that in the y direction. The most important aspect of the above ∂p must necessarily be of the same order (or smaller). which is Prandtl’s assumption 2 listed on page 1. Because the pressure change across the boundary layer is negligible. We need to retain all the other terms in the x − component momentum equation because they are all of comparable order of magnitude. can be assumed to be “impressed” on the boundary layer. The order of magnitude of the Reynolds number 2 ∂y δ was established earlier on page 2. it is evident that the scaled pressure change across the thickness of the boundary layer ∆p ∼ O δ 2 . let us consider the y − component of the NavierStokes equation. ∂v ∂v + v u ∂x ∂y 1 δ = δ 1 ∂ 2v ∂p 1 ∂ 2v − + + Re L ∂x 2 ∂y 2 ∂y 1 2 δ δ δ equation is that all the retained terms are of O (δ ) . we see that the viscous transport of y − momentum in the x − direction is much weaker than that in the y − direction. which means that it varies from 0 to across ∂y δ δ ∂ 2u 1 derivative ∼ O 2 . so that the pressure gradient The order of magnitude of the derivatives has been estimated in the same manner as outlined earlier. Once again. the pressure distribution along the surface of the object. Therefore. Comparing the two viscous terms. and can be neglected. This is very small. Now. Because the ∂y variation of pressure in the y − direction in the boundary layer must occur over a distance of O (δ ) . evaluated from the potential flow. in a distance of the order δ .
the same equations can be written for flow past an object with a curved surface. For convenience. Therefore. Thus. and we can simply ignore the y − component momentum equation because all the terms are small. Summarizing the above. Prandtl’s steady twodimensional boundary layer equations for incompressible Newtonian flow are written in physical variables below. the effects of curvature of the surface would be negligible at leading order. the potential flow is simply u = U ∞ . Furthermore. 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. Therefore. Continuity ∂p ∂x in the x − component momentum equation is a known ∂u ∂v + =0 ∂x ∂y NavierStokes Equation ∂u ∂u u + v ∂x ∂y = 1 ∂p ∂ 2u − +ν 2 ρ ∂x ∂y For the flat plate problem. and the y − component momentum equation is neglected. the NavierStokes equation simplifies to 7 . as long as we define x and y as distance coordinates along and normal to a surface. the pressure gradient in the x − component momentum equation is established from potential flow theory and evaluated along the surface of the object. respectively.that inhomogeneity. we have two equations for the two unknown velocity components. you can see that for a thin boundary layer. but this should not be a source of confusion. we have retained the same symbols for the velocities and coordinates as those used earlier for scaled variables. To avoid clutter. This means that the potential flow pressure gradient is zero. Even though our analysis assumed a flat plate.
0 ) = 0 Commonly. the potential flow pressure distribution can be evaluated right at the solid surface and used as a known inhomogeneity in the boundary layer equations. the last condition is replaced with u ( x. y ) = U ∞ u ( x. 2. New York. H. u ( 0. 3. The important nonlinear (inertial) terms have been retained. Note that 1. Reference 1. Because the variation of pressure across the boundary layer is negligible to this order of approximation.∂u ∂u u + v ∂x ∂y ∂ 2u = ν 2 ∂y The boundary conditions are written as follows.0 ) = 0 v ( x. Schlichting. y → ∞ ) → U ∞ 8 . McGrawHill. 1968. u ( x. Boundary Layer Theory. ∞ ) = U ∞ . 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.
be sure to clearly mark the datum for measuring heights on a sketch. or when it is issuing as a free jet into the atmosphere. Here. When setting a term to zero. Whenever you use this equation. For example. when the free surface of the liquid in a tank is exposed to the atmosphere. Shankar Subramanian The Engineering Bernoulli equation can be derived from the principle of conservation of energy. Energy Form Here is the “energy” form of the Engineering Bernoulli Equation.Engineering Bernoulli Equation R. pout Vout 2 pin Vin 2 loss ws + + zout = + + zin − − γ 2g γ 2g g g 1 . pout Vout 2 pin Vin 2 + + gzout = + + gzin − loss − ws 2 2 ρ ρ 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 . the pressure at that location is set equal to zero gage. indicate the reason for doing so. and the kinetic energy term associated with that velocity can be set equal to zero without much error. The textbook provides such a derivation in detail. I have simply summarized the important forms of this equation for your use in solving problems. the velocity of the free surface will be relatively small. When liquid is taken out of a vessel through a pipe of crosssectional area that is small compared with that of the vessel. Each term has dimensions of energy per unit mass.
γ = ρ g : Specific weight of fluid We can define the head developed by a pump as hp = − ws . loss = N 2 2 . g Note that the work is negative which implies work is done on the fluid by the pump. In this case. heads. we express loss as a certain number (N) of velocity V2 . loss = hfriction = h f : loss expressed in head of fluid. g Sometimes.
the basis set used for expressing the components of vectors. Shankar Subramanian The term “invariant representation” implies that the results are written in a form that is independent of the coordinate system or. xi x j − δ ij . One is the set of decaying harmonics. The general decaying harmonic is given by ( −1) 1 φ− ( n+1) = ∇∇∇ ⋅⋅⋅ ∇ . The pressure and velocity fields are entities that are called absolute tensors. The first few are written below. δ x xjx xδ + x δ + x δ 1 xi x x .. i 5 j − ij3 . in contrast.. 3 . These are called “vector harmonics” even though the solutions are tensors. we learned about solutions of the Laplace equation written in such invariant notation involving only the position vector xi and its length r . the order increases by 1.. 1 ⋅ 3 ⋅ 5 ⋅⋅⋅ ( 2n − 1) n times r n n = 0. grow with distance from the origin. Growing harmonics. There are two sets of such solutions. On the other 1 . Absolute tensors are unaffected by whether the coordinate system (and therefore the basis set) is right or lefthanded. and are ordered such that at each level. 2. The important idea in each case will be that we’ll be looking for solutions of Laplace’s equation in one case for a scalar field (pressure) and in the other for a vector field (homogeneous solution for the velocity). In an early homework assignment.1. ⋅⋅⋅ 3 5 Absolute and Pseudotensors We’ll find these sets of harmonic functions very useful in constructing solutions of Stokes problems. i 7 k − i jk j 5ki k ij r r r 3r r 5r You can see that these solutions approach 0 as r → ∞ . These are obtained from the decaying harmonics as r 2 n +1φ− ( n+1) . The first few growing harmonics are given below. equivalently. r2 r2 1. xi .Stokes Flow – Invariant Representation of Solutions R. xi x j xk − ( xiδ jk + x jδ ki + xk δ ij ) .
that is. if the cross product of a pseudovector and an absolute vector is formed.hand. Therefore. a function which satisfies the Laplace equation. defined as the curl of the velocity field. we are familiar with vectors and tensors that change sign if the basis set is changed from one that is righthanded to one that is lefthanded. Another way to obtain a pseudovector is by taking the curl of an absolute vector field. Solution of the Stokes equation Now. ∇ • ( ∇u ( P ) ) = 1 ∂ ∂ ∂p (δ ij p ) + xj 2 µ ∂xi ∂xi ∂xi 2 . The homogeneous solution u( H ) must also satisfy Laplace’s equation ∇ 2 u(H ) = 0 . Performing a second such operation on a pseudovector such as taking a curl will yield an absolute entity. One example is the cross product of two absolute vectors. an example is the vorticity vector field. can be written as u (j P ) = xj 2µ p. an incompressible velocity field must satisfy ∇ • u = 0 . the pressure field in every Stokes problem must be a harmonic function. We can construct a solution of the Stokes equation by adding a homogeneous solution and a particular solution. ∇ • ( µ ∇ 2 u − ∇p ) = 0 . take the divergence. ∇u ( P ) = ∂ ( P) ∂ x j 1 ∂x j ∂p 1 ∂p uj = p = p + xj = δ ij p + x j ∂xi ∂xi 2µ 2µ ∂xi ∂xi 2µ ∂xi Now. which satisfies the same equation as u . Let us verify that this is indeed correct. Also. which can be rearranged to µ ∇ 2 ( ∇ • u ) − ∇ • ∇p = 0 . The first term is zero because the flow is incompressible. let us begin with the Stokes equation for an incompressible flow. µ ∇ 2 u = ∇p Also. leading to the result that ∇ 2 p = 0 . You will find a more detailed discussion of the distinction between absolute and pseudotensors in the textbook by Leal (pages 178179). the result with be an absolute vector. If we now take the divergence of the Stokes equation. The particular solution u( P ) .
1. Note that we still must satisfy ∇ • u = 0 . 3. 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. the fields must also be linear in these vectors. 4. because of the linearity of the Stokes equation. In this case. and the velocity field is an absolute vector. 3 . ∂xi ∂xi ∂xi 2µ ∂xi Constructing solutions using the invariant representation Now. we have used the fact that the pressure field ∂2 p satisfies Laplace’s equation = 0 . 2. We see that the particular solution ∂xi ∂xi is indeed correct. If the object is moving with a velocity U i or rotating with an angular velocity Ωi . The pressure field and the homogeneous part of the velocity field are harmonic functions. we can use growing harmonics. If the flow is interior to an object. Therefore. such as a drop. To do this. we use the following ideas.= ∂x ∂p 1 ∂p ∂ 2 p 1 ∂p ∂p 1 ∂p 1 δ ij + j + xj δ ij + δ ji = ∇p = = 2 µ ∂xi ∂xi ∂xi ∂xi ∂xi 2 µ ∂xi ∂xi µ ∂xi µ In obtaining the above result. the solution must be finite at the origin which is usually chosen to lie within the object. If the flow is exterior to a body and the fluid is quiescent far from a body. we can use the decaying harmonics. x 1 ∇ • u( H ) = − ∇ • u( P ) = − ∇ • p = − [3 p + x • ∇p ] 2µ 2µ x ∂ ∂x ∂p ∂p because ∇ • p = ( xi p ) = i p + xi = δ ii p + xi = 3 p + x • ∇p . The pressure field is an absolute scalar. we are ready to construct solutions of Stokes problems using invariant representations of the pressure and velocity fields.
flow caused by a sphere rotating in place in a quiescent fluid.We’ll 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. We shall illustrate with three examples. and flow caused by a point force in a quiescent fluid. 4 . It is this fact that makes it relatively easy to construct the desired solutions in invariant form. flow caused by a sphere translating at a constant velocity through a quiescent fluid.
ψ ( r . π . 1 .θ ) =∑ ( An r n + Bn r − n+1 + Cn r n+2 + Dn r − n+3 ) Cn−1/ 2 (η ) n =2 ∞ (3) Here. η = cosθ . Shankar Subramanian The details of the solution of Stokes’s Equation for the streamfunction E 4ψ = 0 in spherical polar coordinates ( r . which satisfy Legendre’s equation. Happel and H.θ . The solution is obtained as the sum 1 2) ψ ( ) ( r .Notes on the Solution of Stokes’s Equation for Axisymmetric Flow in Spherical Polar Coordinates R. in the form of infinite series. 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 (η ) . and in Abramowitz and Stegun (1965).θ ) + ψ ( and ψ ( 2) ( r .θ ) where ψ (1) satisfies E 2ψ ( ) = 0 1 (1) is the particular solution of E 2ψ ( ) = ψ ( ) 2 1 (2) The solution. φ ) by the method of separation of variables are given in the book “Low Reynolds Number Hydrodynamics” by J. Their properties are discussed in Sampson (1891). and Cn−1/ 2 (η ) are the Gegenbauer Polynomials of order n and degree −1/ 2 . The result is given below. 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 θ = 0. Brenner. The Gegenbauer Polynomials are solutions of the second order ordinary differential equation d 2Φ (1 − η ) dη 2 + n ( n − 1) Φ = 0 2 (4) where n is an integer.
n ( n − 1)( 2n − 1) 2 δ mn (8) 2 .d 2 dΦ (1 − η ) dη + n ( n + 1) Φ = 0 dη (5) Many important properties of the Legendre Polynomials can be found in MacRobert (1967) and in Abramowitz and Stegun (1965). n ≥ 2. Some useful relationships involving the Gegenbauer Polynomials are given below. Cn−1/ 2 (η ) = Pn−2 (η ) − Pn (η ) 2n − 1 η Cn−1/ 2 (η ) = − ∫ Pn−1 ( s ) ds −1 (6) (7) The first few Gegenbauer and Legendre Polynomials are C0−1/ 2 (η ) = 1 C2−1/ 2 (η ) = 1 (1 − η 2 ) C1−1/ 2 (η ) = η C3−1/ 2 (η ) = 1 η (1 − η 2 ) 2 2 C4−1/ 2 (η ) = 1 (1 − η 2 )( 5η 2 − 1) 8 P0 (η ) = 1 P2 (η ) = P (η ) = η 1 1 1 P3 (η ) = η ( 5η 2 − 3) ( 3η 2 − 1) 2 2 1 P4 (η ) = ( 35η 4 − 30η 2 + 3) 8 The Gegenbauer Polynomials satisfy the orthogonality property ∫ +1 Cm−1/ 2 (η ) Cn−1/ 2 (η ) 1 −η 2 −1 dη = for values of m.
G.” PrenticeHall (1965). Brenner.” Pergamon Press (1967).A. Leal.Gegenbauer Polynomials through Qn (η ) = − Cn−+1/ 2 (η ) . 1 The function Qn (η ) used by Leal in the textbook is related to the References M. “Spherical Harmonics. R. Abramowitz and I.A. 3 . “On Stokes’s Current Function.A. republished by Noordhoff International (1973). MacRobert. “Laminar Flow and Convective Transport Processes. A 182. Sampson. 449518 (1891). “Low Reynolds Number Hydrodynamics. and H. “Handbook of Mathematical Functions. T. J. Happel. Royal Soc. Stegun. Trans.” Phil.” ButterworthHeinemann (1992). L.” Dover Publications (1965).M.
Vorticity cannot be generated internally within an incompressible fluid. In addition.Vorticity R. is an important entity in fluid mechanics. Vorticity is generated at fluidsolid interfaces and at fluidfluid interfaces. in a high Reynolds number flow (implying weak viscous effects) past a rigid body. the vorticity will remain zero in that fluid mass. 2 2 1 1 Ω • dx = − dx × ω = ω × 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 ω . just as energy and species are. vorticity also is intensified by the stretching of vortex lines. It can be shown that if a fluid mass begins with zero vorticity. ⎛ 0 ⎜ ⎜ ⎜ 1 ⎛ ∂u ∂u ⎞ Ω=⎜ ⎜ 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 ⎟ ⎟ 0 ⎟ ⎠ 1 ⎛ ∂u x ∂u z − ⎜ ∂x 2 ⎝ ∂z As we already know. This is the reason why. with the vorticity being confined to a boundary layer near the surface of the body. Ωij dx j = − ε ijk dx jω k or in Gibbs notation. One reason for working with the equations of vorticity transport is that pressure is absent as a dependent variable. The vector associated with the vorticity tensor in this manner is 1 1 − ω . a skewsymmetric tensor Aij can be formed from a vector ak by writing Aij = ε ijk ak . It is transported from one place to another in a fluid by convective and molecular means. We can write its components in terms of the components of the velocity gradient in a rectangular Cartesian basis set as follows. A flow in which the vorticity is zero is known as an irrotational flow. 2 1 . Therefore. a mechanism that is not present in the transport of energy and species. The vorticity tensor Ω is a skewsymmetric tensor. most of the flow can be described by using the equations that apply to irrotational flow. and the fluid is inviscid (meaning the viscosity is zero). Shankar Subramanian The vorticity ω = ∇ × u .
To prove this. we can write the average linear velocity along the circle as ∫ u • t ds and 2π a C therefore the average angular velocity as 1 2π a 2 C ∫ u • t ds . we can define a vortex line as a curve to which the vorticity is tangent everywhere. Let the unit normal vector to the surface be n . page 82) that the angular momentum of a spherical element of fluid is equal to onehalf the vorticity times the moment of inertia of the fluid. If the components of the vorticity ω = ∇ × u are (ω x . then we can write the equations of the space curves that are vortex lines as 2 . we see ⎛1 ⎞ that this is equal to the average value of ⎜ ∇ × u ⎟ • n over the surface of the circle. We also can show (see Batchelor. You can prove this by using the fact that Ωij = − Ω ji . Because u • t is the component of the velocity along the periphery of the 1 circle. in the limit as the radius of the circle approaches zero. we should interpret the above statement as implying that the average angular velocity of a fluid element located at a point is onehalf the vorticity vector at that point. From Stokes’s theorem. ⎝2 ⎠ Thus. consider a surface formed by an infinitesimal circle of radius a located at a point x . and the unit tangent vector to the circle at any point be t . ω z ) in a basis set obtained from rectangular Cartesian coordinates. dS is an area element on the surface of the circle S and ds is a line element along the circle C.Because a fluid does not usually rotate as a rigid body in the manner that a solid does. t n a Apply Stokes’s theorem to the velocity field in this circle. Another useful result is that dxi Ω ij dx j = 0 . ∫ ( ∇ × u ) • n dS = ∫ u • t ds S C Here. ω y . Vortex Lines and Tubes Just as a streamline is a curve to which the velocity vector is tangent everywhere. we find that the average angular velocity around the circle approaches the value of onehalf the component of the vorticity vector in a direction perpendicular to the surface of the circle. just as it is for a rigid body.
” Cambridge University Press.G. P. 4. “Mathematical Principles of Classical Fluid Dynamics. Flugge. 3 . Batchelor. Saffman. If we construct an open surface S bounded by this closed curve C. and ds is a line element. 3.” Cambridge University Press. G. S By using Stokes’s C ∫ u • t ds where C is any closed curve around the vortex tube.dx ωx = dy ωy = dz ωz .” Indiana University Press. we can see that this is the circulation ∫ dS • ω . Truesdell. “An Introduction to Fluid Dynamics. The surface that is formed by all the vortex lines passing through a closed reducible curve is known as a vortex tube. 1967. S. Serrin. 1. “Kinematics of Vorticity. we can define the strength of the vortex tube as theorem.” Handbuch der Physik VIII/1. 1959. “Vortex Dynamics. 1992. J. 1954. t is a unit tangent vector to the curve at any point. Some good sources for further study are listed below. 2. C. Ed. Chapter 5.K.
” The viscous dissipation per unit volume is written as τ : ∇u = µΦ v where Φ v for a Newtonian fluid is given below in different coordinate systems. Rectangular Cartesian Coordinates ( x. 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. y. z ) ∂ur 2 1 ∂uθ ur 2 ∂u z 2 Φ v = 2 + + + r ∂z ∂r r ∂θ ∂ u + r θ ∂r r 1 ∂ur 1 ∂u z ∂uθ + + + ∂z r ∂θ r ∂θ 2 2 2 2 2 ∂u ∂u + r + z − (∇ • u) ∂r 3 ∂z 1 . This is known as “viscous dissipation.Viscous Dissipation term When the dot product of each term in the NavierStokes 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. z ) 2 ∂u 2 ∂u y 2 ∂u 2 ∂u ∂u y x x z Φ v = 2 + + + + ∂x ∂x ∂y ∂z ∂y ∂u ∂u ∂u 2 2 ∂u + y + z + z + x − (∇ • u) ∂y ∂z 3 ∂x ∂z 2 2 Cylindrical Polar Coordinates ( r .θ .
θ .Spherical Polar Coordinates ( r . φ ) 2 2 ∂u 2 1 ∂u ur 1 ∂uφ ur uθ θ Φ v = 2 r + + + + + cot θ r r sin θ ∂φ r r ∂r r ∂θ ∂ u + r θ ∂r r 1 ∂uθ 1 ∂ur sin θ ∂ uφ + + + r ∂θ r ∂θ sin θ r sin θ ∂φ 2 2 2 1 ∂ur ∂ uφ 2 2 + + r − (∇ • u) ∂r r 3 r sin θ ∂φ 2 .
v = 0 . and mass transport to a falling liquid film. Here. 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. y b x U The fluid is initially assumed to be at rest. We neglect edge effects in the z − direction so that we can ∂v ∂v set vz = 0 and = 0 . Shankar Subramanian Introduction and Problem Statement We encounter partial differential equations routinely in transport phenomena. We shall assume the flow to be incompressible with a constant density ρ and Newtonian with a constant viscosity µ . and the subscripts denote components. 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. Here. We shall also learn when the method can be used. There is no applied pressure gradient.Solution of Partial Differential Equations Combination of Variables R. Some examples are unsteady flow in a channel. with motion being caused strictly by the movement of the bottom plate. 1 . steady heat transfer to a fluid flowing through a pipe. and assume fully developed flow. implying ∂x ∂z stands for the velocity vector. Motion is initiated by suddenly moving the bottom plate at a constant velocity of magnitude U in the x − direction. we shall learn a method for solving partial differential equations that complements the technique of separation of variables.
given in Equation (4). Outside of this region of influence. Therefore.It can be established from the continuity equation and the kinematic condition at one of the walls that v y = 0 . 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. Physically. at values of time t for which the scaled time ν t / b 2 is small compared to unity. This problem was first considered by Lord Rayleigh. we obtained a solution of these equations that can be written as follows. But. Mathematically. The initial condition is (2) vx ( 0. 0 ) = U vx ( t . Therefore. vx ( t . y ) . we replace the boundary condition at the top plate. y ) 2 y = 1− − U b π ∑ n =1 ∞ νt ⎤ ⎡ exp ⎢ − n 2π 2 2 ⎥ b ⎦ ⎛ nπ y ⎞ ⎣ sin ⎜ ⎟ n ⎝ b ⎠ (5) The infinite series in Equation (5) is uniformly convergent for all values of time t . 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. and ν is the kinematic viscosity. 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. which can be shown to satisfy the following partial differential equation. y ) = 0 and the boundary conditions are and vx ( t . t represents time. b ) = 0 (3) (4) Using separation of variables. one might approximate the system for such small values of time by another in which the top plate is absent. with 2 . the fluid is practically stationary. corresponding to the scaled time ν t / b2 not being too small compared to unity. and therefore is known as Rayleigh’s problem. the only nonzero velocity component is vx ( t . we find that a large number of terms needs to be included to obtain a sufficiently accurate answer. if we attempt to calculate the sum numerically for small values of time (ν t / b 2 small compared with unity) when the exponential factor is not as helpful. Therefore. ∂vx ∂ 2 vx =ν ∂t ∂y 2 (1) In Equation (1).
In the figure. We can see how the change in velocity made at the bottom plate at time zero propagates deeper into the fluid with increasing time.vx ( t . if we define a certain combination of the original variables as a 3 . nor a natural time scale. y → ∞ ) → 0 . By implying similarity of shape. A Speculation There is neither a natural length scale in the problem. We can use the reference velocity U as a natural scale for the velocity vx . It is tempting to speculate that these profiles are similar in shape. The solution of Equations (1) to (3) and (6) is qualitatively sketched below for two different values of time. In mathematical language. we must write Equation (6) as vx ( t . but it is convenient to work with the remaining physical variables just as they are. ∞ ) = 0 (6) Note that to be precise. The solid (black) line corresponds to a small value of time. we mean that scaling the distance variable with the thickness of the affected region δ ( t ) should lead to these two curves and others like them collapsing into a single universal curve. the symbol v is used to represent vx . and the equation must be read to imply only such a meaning. and the dashed (red) line to a larger value of time.
y ) = U φ (η ) (7) (8) η= y δ (t ) and δ ( 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.” Solution by Combination of Variables We now proceed to state the above speculation in mathematical form and follow through the consequences. The transformation to η is known as a “similarity transformation” and the variable η is termed a “similarity variable.new variable η = y / δ ( t ) . Therefore.” Assume where vx ( t . y ) to become a function U φ (η ) that depends only on the single new variable? This speculation is shown in the sketch below. we need to insert Equations (7) and (8) into the governing equation and the initial and boundary conditions and see if the 4 . can we expect the velocity field vx ( t . This is the method of “Combination of Variables. but to transform to a single new variable is not always possible.
Recall that we assumed that φ explicitly depends only on the similarity variable η . but will not affect the final solution for vx ( t . φ ′′ + η ⎢ ⎡ δδ ′ ⎤ ′ φ =0 ⎣ν ⎥ ⎦ (12) In writing Equation (12). leads to the following equation after slight rearrangement. If our conjecture proves to be incorrect. writing Equation (12) as 5 . ∂vx ∂η dφ U dφ =U = ∂y ∂y dη δ dη (10) and ∂ 2 vx ∂ ⎛ ∂vx ⎞ ∂ ⎛ U d φ ⎞ U ∂ ⎛ dφ ⎞ = ⎜ ⎟= ⎜ ⎟= ⎜ ⎟ 2 ∂y ∂y ⎝ ∂y ⎠ ∂y ⎝ δ dη ⎠ δ ∂y ⎝ dη ⎠ U ∂η d ⎛ dφ ⎞ U d 2φ = = δ ∂y dη ⎜ dη ⎟ δ 2 dη 2 ⎝ ⎠ (11) Substituting Equations (9) and (11) into the governing differential equation for vx (Equation (1). For this purpose. Later. ∂vx ∂η dφ y d δ dφ η d δ dφ =U = −U 2 =−U ∂t ∂t dη δ dt dη δ dt dη (9) Note that when writing the derivative of φ with respect to η . Therefore. and φ were to depend explicitly on both η and t . We choose δδ ′ = Cν (13) where C is an arbitrary constant. Here is our chance to do so and eliminate the inconsistency at the same time. But.process leads to a consistent mathematical framework. we already have assumed that φ can depend explicitly only on the single variable η and used the ordinary derivative instead of the partial derivative. y ) . we shall use the chain rule of differentiation as needed. We have not yet specified δ ( t ) . for convenience. we shall see that the value of C will affect the result for δ ( t ) . we have used the expedient of designating derivatives with respect to the argument of each function with primes. the above chain rule result will need to be modified to include a partial derivative of φ with respect to time. Let us now obtain expressions for the derivatives with respect to y . we set C = 2 . however. Equation (12) suggests that time also will explicitly appear in the result for φ because of the presence of the timedependent term δδ ′ .
becomes φ (∞) = 0 The initial condition. described by Equation (6). given in Equation (2). consider the boundary condition at the bottom surface y = 0 . This transforms in a straightforward manner to φ (0) = 1 (15) The fact that a quiescent condition is approached as y → ∞ . the three conditions on vx ( t .φ ′′ + 2η φ ′ = 0 (14) We now need to transform the initial and boundary conditions. and we have a completely consistent mathematical framework for the problems for φ (η ) and δ ( t ) . transforms to (16) φ⎜ ⎜ ⎛ y ⎞ ⎟=0 δ ( 0) ⎟ ⎝ ⎠ (17) and we see that we have not completely eliminated the original variables from appearing explicitly in the problem statement for φ . First. given in Equation (3). y ) yield two conditions on φ (η ) and one initial condition on δ ( t ) . but only a second order differential equation for φ (η ) . To remove this inconsistency. the general solution of Equation (14) can be written as φ (η ) = a1 + a2 ∫ e −γ d γ 2 η (19) 0 6 . we must set δ ( 0) = 0 (18) The choice in Equation (18) makes Equation (17) collapse into Equation (16). First. therefore. Note that there are three conditions on the velocity field vx ( 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. The specification of the arbitrary constants that arise in the integration of the latter requires only two conditions. y ) . and at the same time select an initial condition for δ ( t ) .
The solution of Equation (13) with the constant C = 2 . y ) = U erfc ⎜ (24) ⎟ ⎝ 2 νt ⎠ If we had made a different choice of value for the constant C that appears in Equation (13).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 φ (η ) = erfc (η ) where “ erfc ” means “complementary error function. the solution for the velocity field can be written as ⎛ y ⎞ vx ( t . δ ( t ) = 2Cν t = C 2 νt 2 (25) φ (η ) = erfc ⎜ ⎜ ⎛ C ⎞ η⎟ ⎟ ⎝ 2 ⎠ (26) 7 . erfc (η ) = 1 − erf (η ) (20) (21) where the “error function” “ erf ” is defined as η erf (η ) = 0 ∞ ∫e ∫e 0 −γ 2 dγ = 2 η −γ 2 dγ π −γ ∫ e dγ 2 (22) 0 You can find out more about the error function and the complementary error function from Abramowitz and Stegun [1]. it would have affected the results as follows.” This function is defined as follows. when specialized using the initial condition given in Equation (18). is δ (t ) = 2 ν t (23) When this result for δ ( t ) is used in Equation (8) in which η is defined.
This is perfectly natural because in a diffusive process. also yielding an initial condition for the thickness of the affected region. This means that there can be no unambiguous definition of a finite thickness for the affected region. whereas the velocity field is uniquely determined. 3.5 % of the value at the surface of the moving plate. obtained an ordinary differential equation for the thickness of the affected region and another ordinary differential equation for the velocity field.678 × 10−3 when its argument is 2 . leading to the same result for the velocity field given in Equation (24).You can see that when the definition of η given in Equation (8) is used in Equation (26). we have learned the method of combination of variables for solving partial differential equations. 5. noted that the thickness of the affected region can only be defined to within a multiplicative arbitrary constant. collapsed the three boundary conditions on the velocity field into two on the velocity field as expressed in the similarity variable. the factor C / 2 cancels out. You may wonder about the uncertainty in the value of δ ( t ) . we identified the governing partial differential equation and boundary conditions for our system. solved these ordinary differential equations to obtain results for the thickness of the affected region and the velocity field. Then we 1. 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. and can be considered negligible for practical purposes. 4. 7. requiring that the original independent variables not be allowed to appear explicitly in the problem posed in the new similarity variable. the influence of a change is felt everywhere in the fluid instantaneously.” caused by the indeterminacy of the value of C . 2. First. the estimate 4 ν t is sometimes used for the thickness of the region influenced by the ( ) sudden movement initiated at the boundary. Because of this. only its scaling can be established uniquely. traced the consequences of this similarity hypothesis mathematically. at a distance y = 4 ν t . Summary In this module. which is the thickness of the “affected region. 8 . along with δ ( t ) from Equation (25). it complements the method of separation of variables. 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. Therefore. 6. the velocity would be less than 0. The complementary error function assumes a value of 4.
The assumptions are that there are no sources or sinks. the method is equally applicable to nonlinear problems. and that the slab is very long and very wide so that end effects and edge effects can be neglected. This method has used successfully in solving the NavierStokes equations including inertia (and therefore nonlinear) in forced boundary layer flows. Abramowitz and I. is described by the same governing equations and boundary conditions.The important features of the method are that the domain must be semiinfinite. Handbook of Mathematical Functions. even though the problem we posed is linear. when the temperature at the surface y = 0 is changed to a new value at time zero. New York. M. it can be seen that the same equations also describe unsteady diffusion in a similar situation. Concluding Remarks The problem of unsteady onedimensional heat conduction in a semiinfinite solid slab (or a quiescent liquid layer) in the y − direction. A. All of these cases can be handled by the same solution method. and also in solving problems of natural convection in boundary layers wherein the fluid mechanics and heat transport problems lead to coupled nonlinear governing equations. Dover. combination of variables does not require the system of governing equation and boundary conditions to be linear. 9 . 1965. By analogy. heat transport occurs only by conduction with a constant thermal conductivity. the density and specific heat of the material are constant. Stegun. Note that unlike separation of variables. Reference 1. and the boundary condition at infinity must be the same as the initial condition.
Here only a limited discussion is given. focusing on 1 . Shankar Subramanian Mixing is a common operation in the chemical process industry. You can learn about mixing and mixers from McCabe et al. Shaft Liquid Level Baffle DT Blade DA Typically. can result that would not cause mixing. In addition. a simple rotating flow. mixers typically have baffles that are attached to the wall to break up this vortex flow. such a simple flow would cause a dip in the surface and this can exacerbate the entrainment of air into the liquid. (2001) and Holland and Bragg (1995) as well as Perry’s Chemical Engineers’ Handbook. In the absence of baffles on the wall. 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. (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.An Introduction to Mixing R. a mixer is a hollow cylinder filled with the substances to be mixed to some appropriate level. The objectives of mixing are to achieve one or more of the following. Therefore. sometimes called a vortex flow.
p 2 2 ρ p N p DA ρ m N m DA = µm µp 2 . In designing a scale model. A limited list of the important parameters is given below. m Therefore. in the following. it is important to maintain geometric similarity so that all the aspect ratios and shapes are preserved. we cannot use a scale model. Fr ) Here P NP = 5 ρ N 3 DA 2 ρ NDA Re = µ N 2 DA Fr = g 3 ρ N 2 DA Also. Rem = Re p . Several other geometric parameters are omitted. 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. Let us investigate why. in that case. For dynamic similarity. 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. two liquid phases. respectively.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 can show that the Power number N P will be some function of the Reynolds number Re and the Froude number Fr . We use the subscripts m and p to designate model and prototype. σ is the interfacial tension between the If both the Reynolds and Froude numbers are important. We use dimensional analysis to find the important groups. ρ: µ: 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. the Weber number We = σ can be important in some situations where the objective is to emulsify a mixture. N P = φ ( Re.
we can write the result for the power input to the mixer as 3 . we must simultaneously satisfy 2 N m D Ap = 2 = N p DAm DAp DAm . the data for a given type of mixer can be fitted to N P = c′Re a This can be recast as log N P = c + a log Re where the new constant c = log c′ .Assuming we use the same fluid. This means that the model must be of the same size as the prototype. ρm ρ p = . Power Requirement for a Mixer If we restrict attention only to nonvortex forming systems. µm µ p So. If the flow in the mixer is laminar. Therefore. The implication of this result is that the speed of rotation in the model should be substantially higher than that in the prototype. The only way to do this is to set DAm = DAp . In that case. dynamic similarity requires that Nm = N p 2 D Ap 2 DAm assuming we use the same liquid in both the model and the prototype. Nm = Np D Ap DAm This implies g = 2 N p D Ap g or So. 2 N m D Ap = 2 N p DAm . which defeats the purpose of building a scale model. consider the equality 2 N m DAm Frm = Frp . Baffling the system eliminates vortex formation and removes the Froude number as a significant parameter. for a laminar mixer. it is found that a = −1 . Now.
For a generic baffled mixer. Edward Arnold. 4 . Chapter 9. 2. we can set a ≈ 0 .A.000 . In this case. Fluid Flow for Chemical Engineers. W. 2001. the flow is turbulent. Holland and R. Chapter 5.C. Unit Operations of Chemical Engineering. London. McGrawHill. F.L. Usually. New York. and P. and the reverse in the turbulent flow result. Smith. 1995. J. Can you rationalize this from your knowledge about the mechanisms of dissipation in these two types of flows? References 1. and when the flow is fully turbulent corresponding to Re ≥ 10. McCabe.3 P = c′µ N 2 DA Laminar mixers are not common. the constant a is found to be small. Harriott. 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. Bragg.3ρ N 3 DA Note the significantly larger exponents on the rotation speed and the diameter of the agitator in the case of turbulent flow. the power input is found to be given by 5 P = 6.
Plate Frame Heat Exchanger
Plate Frame Heat Exchanger
1
Exchanger Matrix
2
PIPE FLOW
Re =
DVρ
μ
Laminar, Transition, Turbulent
Entrance Length Le / D = 0.06 Re for Laminar Le / D = 4.4 Re1/6
f =
ΔP D 1 / 2 ρV 2 L
f Fanning =
1 f Darcy 4
2 ⎛ L ⎞⎛ ρV ΔP = f ⎜ ⎟⎜ ⎜ ⎝ D ⎠⎝ 2
⎞ ⎟ ⎟ ⎠
Head Loss
hf =
2 ΔP ⎛ L ⎞⎛ V ⎞ ⎟ = f ⎜ ⎟⎜ ⎜ ⎟ ρg ⎝ D ⎠⎝ 2 g ⎠
Friction Factor – Laminar Flow
Q=
πR 4 ΔP πD 4 ΔP πD 2 = = V 8μL 128μl 4
f =2 ⎛ μ ⎞ ΔP D = 64⎜ ⎟ 2 ⎜ DVρ ⎟ L ρV ⎠ ⎝
f =
64 Re
3
Friction Factor .7 Re f f ⎣ ⎤ ⎥ ⎥ ⎦ 4 .51 = −2.51 = −2.0 log10 ⎢ D + ⎢ 3.0 log10 ⎢ D + ⎢ 3.Turbulent Flow ε ⎞ ⎛ f = φ ⎜ Re. ⎟ D⎠ ⎝ Colebrook 1 ⎡ε 2. ⎟ D⎠ ⎝ 1 ⎡ε 2.7 Re f f ⎣ ⎤ ⎥ ⎥ ⎦ Useful Equations f = ΔP D 1 / 2 ρV 2 L f Fanning = 1 f Darcy 4 2 ⎛ L ⎞⎛ ρV ΔP = f ⎜ ⎟⎜ ⎜ ⎝ D ⎠⎝ 2 ⎞ ⎟ ⎟ ⎠ hf = 2 ΔP ⎛ L ⎞⎛ V ⎞ ⎟ = f ⎜ ⎟⎜ ⎜ ⎟ ρg ⎝ D ⎠⎝ 2 g ⎠ Q= πR 4 ΔP πD 4 ΔP πD 2 = = V 8μL 128μl 4 f =2 ⎛ μ ⎞ ΔP D = 64⎜ ⎜ DVρ ⎟ ⎟ L ρV 2 ⎝ ⎠ f = 64 Re ε ⎞ ⎛ f = φ ⎜ Re.
Description Drawn tubing (brass.03 5 . ft) 0.0004 0.0005 0.001 – 0.003 – 0.00085 0.01 0. lead.00015 0. glass etc) Commercial steel or wrought iron Asphalt cast iron Galvanized iron Cast iron Concrete Riveted steel Surface roughness (ε.000005 0.
000(0.667 x12.001375⎢1 + ⎜ 20.00085 ft ε/D = 0.9 Given: Q = 2000 gpm 8 inch ID Kinematic viscosity 104ft2/s Density = 1.00475 ⎡ ⎛ 10 6 f = 0.0 log10 ⎢ D + ⎢ 3.8 ft / s A DV = 0.0235 6 .8( gal / min/ ft 3 / sec) A= πD 2 4 = π 4 (0.0013 1 ⎡ε 2.500 10 − 4 Re = ν For cast iron ε=0.667 = 0.46 ft 3 / s 448.0002 and compare the value to that obtained by equation: 1 ⎡ ⎤ ⎛ ε 10 6 ⎞ 3 ⎥ ⎟ f = 0.667 ft )2 = 0.001375⎢1 + ⎜ 20.349 ft 2 V = Q = 12.0047 Calculate: Head Loss BHP of the Pump If Efficiency is 0.7 Re f f ⎣ ⎤ ⎥ ⎥ ⎦ f=0. Read the value of friction factor for Re=105 and (ε/D) = 0.00085/0.0002 ) + 5 ⎜ ⎢ ⎝ 10 ⎢ ⎣ 1 ⎤ ⎞3 ⎥ ⎟ ⎟ ⎥ ⎠ ⎥ ⎦ f = 0.000 + ⎜ ⎟ ⎥ ⎢ ⎝ D Re ⎠ ⎢ ⎥ ⎣ ⎦ From the chart Using the equation: f = 0.8 = 85.8 slug/ft3 Q= 2000 gal / min = 4.Q1.51 = −2.
pressure drop and head losses correlate directly with the dynamic pressure.9 HP Pipe Flow Losses occurring in straight pipes – Major losses Losses occurring in pipe systems – Minor losses hL = f KL = ( hL Δp = V / 2 g 1 / 2 ρV 2 2 l V2 D 2g Q hL = Δp hL Δp 2 g Δp = ⇒ = ρg V 2 / 2 g ρg V 2 1 / 2 ρV 2 ) ( ) KL is called the loss coefficient Flow which is dominated by inertia rather than viscous effects.9 /0. Re) However for most practical cases: Δp = K L 1 / 2 ρV 2 V2 2g ( ) hL = K L K L ⇒ Δp = 1 / 2 ρV 2 ( ) KL = ø (geometry) Dynamic pressure 7 .Head Loss hf = 2 ΔP ⎛ L ⎞⎛ V ⎞ ⎟ = f ⎜ ⎟⎜ ⎜ ⎟ ρg ⎝ D ⎠⎝ 2 g ⎠ = 134 ft loss = hf x g Power supplied = loss x mass flow rate = 62. KL = ø (geometry.9 = 69.9 HP (550 ftlb/s = 1HP) BHP = 62.
5 Slightly rounded entrance – KL = 0.04 Vena contracta 1 2 V2>V3 3 Flow separation p1 ρV22/2 Ideal full recovery of kinetic energy ρV32/2 KL ρV32/2 Actual x1 x2 x3 x 8 .2 Well rounded – KL = 0. hL = f leq V 2 D 2g D and f are based on the pipe containing the component.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.8 Sharp edge KL = 0. leq = KLD f Entrance flow conditions – Reentrant KL=0.
Determine the volumetric flow rate m3/s? S 150m (2) 47m 16m (1) Rounded entrance Standard 900 elbows D = 0. Union threaded.4 0.156m 150m Density of water ρ = 103 kg/m3 Viscosity of water = 1 x 103 N s/m2 ε = 5 x 105 KL1 = 0.1 0 0 0.25 rounded entrance KL2 =0.2 0.20 0. Valves Table 6.3 0.9 elbow 9 .15 0.1 0.5 r 0. 180 0 return bends. Tees.25 KL D r/D Minor losses are often a result of the dissipation of kinetic energy Elbows.0.5 page 387 White Water flows from a large reservoir and discharges into the atmosphere.
By continuity V1=V2=V p1 V12 p + + z1 = s + z s ρg 2 g ρg p1 − ps V2 = ( z s − z1 ) − 1 ρg 2g Between sections 1 and 2 p1 V 2 p V2 fL V 2 V2 V2 + + z1 = 2 + + z2 + + K L1 + 2K L2 ρg 2 g ρg 2 g D 2g 2g 2g p1 − p2 fL V 2 V2 V2 = ( z 2 − z1 ) + + K L1 + 2K L2 ρg D 2g 2g 2g (z s − z2 ) = V ⎞ ⎛ fL ⎜ + K L1 + 2 K L 2 + 1⎟ 2g ⎝ D ⎠ 2 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 f V (calculated) m/s 3 115 122 5(10)5 7 7 0.0165 1.8(10) 0.9(10) 0.0150 122 Volumetric flow rate Q = π (D2/4) x 122= π (0.33 m3/s 10 .156)2/4 x 122=2.Write Bernoulli’s equation between the free surface and the entrance.0150 1.