Sie sind auf Seite 1von 72

Lectures in Process Fluid Mechanics

P Sunthar

March 8, 2010
Contents

1 Basic Fluid Concepts 5


1.1 Introduction to the Course . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 What Chemical Engineers Do ? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.2 How do Chemical Engineers Carry out the Process ? . . . . . . . . . . . . . . . 5
1.1.3 Product and Service oriented disciplines . . . . . . . . . . . . . . . . . . . . . . 6
1.1.4 Unit Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Dimensions and Units, and Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Measurement errors and propagation of errors . . . . . . . . . . . . . . . . . . . 9
1.4 Index Notation of Vectors and Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.1 Algebra with Index notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4.2 Physical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.3 Comparisons with Bold face notation . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5 Velocity and Stress Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.1 Origin of surface forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.2 Stress Tensor Convention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5.3 Tutorial on Obtaining Local Stress Direction . . . . . . . . . . . . . . . . . . . . 22
1.6 Dimensional Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.6.1 Steps for obtaining dimensionless groups . . . . . . . . . . . . . . . . . . . . . . 26
1.6.2 Example: Drag on a sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.7 Fluid Statics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.7.1 Horizontal and Vertical forces, and Rotation . . . . . . . . . . . . . . . . . . . . 28

2 Differential Analysis of Fluid Flow 39


2.1 Types of flows and Methods of Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.1.1 System, Conservation Laws, and Control Volume . . . . . . . . . . . . . . . . . 39
2.2 Useful Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.1 Reynolds Transport Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.2 Divergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3 Fundamental Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3.1 Continuity Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.3.2 Derivation of Navier-Stokes Equation . . . . . . . . . . . . . . . . . . . . . . . . 45
2.3.3 Energy balance or Heat equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.4 Species Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.4 Boundary Layer Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.4.1 Integral Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.4.2 Laminar and Turbulent Boundary Layers . . . . . . . . . . . . . . . . . . . . . . 57
2.4.3 Differential Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.5 Simple Viscous Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.5.1 Poiseuille and Couette Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

c P Sunthar, March 8, 2010


3
4 CONTENTS

2.5.2 Torsional flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80


2.5.3 Free surface flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.5.4 Wire Coating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.5.5 Tube flow of Power law fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.6 Creeping Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.6.1 Flow past a sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.7 Inviscid flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.7.1 Stagnation point flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.7.2 Flow past a cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.7.3 Porous Media Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.8 Turbulent Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.8.1 What They Said . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.8.2 Origin of Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2.8.3 Boussinesq Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
2.8.4 Prandtl Mixing Length Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . 100
2.8.5 Velocity Profiles in Turbulent Flow . . . . . . . . . . . . . . . . . . . . . . . . . . 101

3 Integral Analysis of Fluid Flow 101


3.1 Mass Balances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.1.1 Multiport Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.1.2 Tank Draining: Unsteady Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.2 Momentum Balances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.2.1 Jet Impingement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.3 Energy Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.3.1 Energy Balance for a Flow through Equipment . . . . . . . . . . . . . . . . . . . 105
3.3.2 Pumping Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

4 Flows in Equipments 113


4.1 Pipe Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.1.1 Major loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.1.2 Minor losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.1.3 Piping Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.1.4 Non-Circular Ducts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.2 Flow Past Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.2.1 Terminal (Settling) Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.3 Fixed Bed of Particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.3.1 Pressure Drop Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.4 Fluidised Beds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

Bibliography 127

c P Sunthar, March 8, 2010



Chapter 2

Differential Analysis of Fluid Flow

2.1 Types of flows and Methods of Analysis


2.1.1 System, Conservation Laws, and Control Volume
Lecture 2.1 Control Volume Analysis in fluid flow is like Free Body diagrams in Mechanics
Concepts You Must Know
1. What is a system, how is it different from a control volume ?
2. Is a “control mass” equivalent to a system or a control volume?
3. What are the basic conservation laws that apply to a system?

The system is defined as an identifiable mass of a body. In thermodynamics we have closed and
open systems. Here we use system to identify with the closed system, in which there can be heat or
work transported in and out of the system, but there mass remains the same. It is also sometimes
called as a Control Mass (CM). For most engineering purposes, where there is no conversion of mass
to energy, it is easily possible to make a identification of a system.
The basic conservation laws applies to an identifiable mass, or a system.
• Mass Conservation
dMsystem
=0 (2.1)
dt
• Newtons second Law:
d(M V)system X
= Fα (2.2)
dt α

• Energy Conservation
dEsystem
= 0 (Isolated) (2.3)
dt
dEsystem
= Q̇ − Ẇ (Closed) (2.4)
dt

The conservation laws in the form presented above are applicable only for a isolated or closed
systems. But for most engineering interest involving fluid flow, it is not useful to consider an iso-
lated system, especially when we are interested in calculating the effects of flow on machinery, or
only a particular part of it. Therefore, we introduce models of flow, which consider the transport of
conserved quantities across a restricted volume. We call this volume as control volume (CV).

c P Sunthar, March 8, 2010


39
40 Chapter 2. Differential Analysis of Fluid Flow

Isolated System
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
Conserved Quantities
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
M
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
MV
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
E
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111
0000000000000000000000000000000000000000000000000
1111111111111111111111111111111111111111111111111

Figure 2.1: Conserved quantities in an isolated system

Consider a case of a water tap connected to an overhead tank through a pipe. The control mass
or the system would be the entire identifiable mass of water, i.e., that in the tank, pipe, and the
bucket. This is of little use, if we were to find out the local effect of the force applied by water flow
in the pipe. We can instead consider a small cylindrical volume of fluid in the pipe, this would be
the control volume. But we have a problem here: The conservation equations, the only governing
equations that we know of the material, applies to the whole system. The goal then is to derive a
suitable form of the equations that applies to the control volume.
Control volume is a concept introduced in continuum mechanics. We are dealing with a limited
volume of a continuum of a fluid (of identifiable mass). The boundary of the volume is called the con-
trol surface. Some of these surfaces could be real, identifiable interfaces: such as the interface between
the fluid and the solid wall. Some surfaces are purely imaginary: such as any arbitrary cross-section
of the pipe. Identification of a control volume is an important step in the solution of fluid mechanics
problems. Many a times the choice can result in simpler algebra. Depending on how the volume is
taken we can have four different models of flow (see Figure 2.2). Note that we use the notation V or
v to denote velocities, and V− (with a cross) to denote volume.
In the following lecture we will derive relationships that will relate the derivatives of a system
to equivalent derivative terms that apply to a control volume of a continuum. These relationships,
in combination with the conservation laws are fundamental to fluid mechanics. These equations
can be applied to arbitrary control volumes. When the CV is a finite volume, such as that of an
equipment, the resulting equations are called as integral balances because it applies to the whole of
the equipment, and when the volume is differential elemental volume, we recover the differential bal-
ances of fluid flow, such as the Navier-Stokes equations. Integral balances are used to obtain gross
effects of the flow on the object it interacts with. Differential balances are used to obtain the detailed
continuum variables such as the velocity and stress profile.

c P Sunthar, March 8, 2010



Lecture 2.2. Reynolds Transport Theorem 41

Finite Volume Differential Volume

V dV

Fixed in space
Moving with flow

dV
V

Figure 2.2: Models of flow: Finite volume or differential volume.

2.2 Useful Theorems


Lecture 2.2 Reynolds Transport Theorem and Divergence theorem are the two building blocks
for deriving equations governing fluid flow.
Concepts You Must Know
1. What derivatives does Reynolds Transport Theorem (RTT) relate?
2. Is RTT valid for arbitrary control volume ?
3. What is Divergence theorem in one dimension?

We first define some notation. Let P be a scalar conserved extensive quantity (one of mass,
momentum, or energy). We also define a corresponding scalar conserved intensive quantity φ. In
a given control volume (CV) V − , φ could in general vary with space and time φ = φ(x, y, z, t). The
extensive property for the entire CV is related to φ by
Z
PCV = − ρφ
dV (2.5)
CV

where ρ is the local mass density. Note that the quantities appearing in the numerator in the LHS of
Equations (2.1) to (2.4) are each an extensive conserved quantity. In the case of momentum, we have
three conserved quantities u, v, w, in each of the three coordinate directions x, y, z. The complete list is
given in Table 2.1. In order to interpret Equations (2.1) to (2.4) for a CV, we have to evaluate the LHS
of these equations for each of the CVs. Therefore we need an expression for
dPsystem
= ? (in a CV) (2.6)
dt
Here is where the Reynolds Transport Theorem (RTT) comes in.

2.2.1 Reynolds Transport Theorem


Consider a identifiable mass of a fluid (control mass, or system) that is displaced in space and time,
as shown in Figure 2.3. For simplicity we consider a two dimensional figure, but the terms “area”

c P Sunthar, March 8, 2010



42 Chapter 2. Differential Analysis of Fluid Flow

Table 2.1: Conserved quantities in extensive and intensive form, as defined in Equation (2.5)
Equation Physical quantity P φ
(2.1) Mass M 1
(2.2) Momentum MV u = u, v, w
(2.4) Energy E e

n
V
V n

dA III
CV α dV
∆l
II System
t + ∆t
I
dA
CSIII
t
CSI

Figure 2.3: The system (identifiable mass) is displaced in time. The dashed line is the system bound-
ary and the shaded region with solid line boundary is the control volume. The System and Control
Volume coincide as ∆t → 0.

and “volume” will be used to identify the “edge” and “area”. Consider a fixed control volume shown
by the shaded region. If we track all the particles in the volume, then after a period ∆t they can be
identified to be present in a region in the neighbourhood (shown by the dashed curve in the figure).
When ∆t → 0, the system coincides with the control volume. The following identities apply

− CV = V
V −I + V− II (2.7)
− sys = V
V − II + V
− III (2.8)

The total change in the quantity P in the system is

dPsys Psys (t + ∆t) − Psys (t)


= lim (2.9)
dt ∆t→0 ∆t
In terms of the individual region quantities

Psys (t + ∆t) = (PII + PIII )(t+∆t) = (PCV − PI + PIII )(t+∆t) (2.10)


Psys (t) = (PI + PII )(t) = PCV (t) (2.11)

Then Equation (2.9) can be written as:

dPsys PCV (t + ∆t) − PCV (t) PIII (t + ∆t) PI (t + ∆t)


= lim + − (2.12)
dt ∆t→0
| ∆t
{z } | {z ∆t } | {z ∆t }
① ② ③

The first term on the RHS ① can be written as


Z
∂PCV ∂
Term ① = = − ρφ
dV (2.13)
∂t ∂t
CV

c P Sunthar, March 8, 2010



Lecture 2.2. Reynolds Transport Theorem 43

For the second term ②, recognise that the term PIII is the total amount of P present in the volume III.
To find this we consider the dependence of a differential amount dP of the quantity P in the volume

dV

dPIII = φ ρ dV (2.14)
The volume dV− is swept by the system boundary from the boundary of the CV in the time interval ∆t.
This volume depends on the direction of the local velocity of the boundary V as well as its direction
n. From the “volume” of the parallelogram we have

− = dA ∆l cos α
dV

where ∆l = V ∆t is the length swept by the volume. In general we can write

− = (dA · V)∆t = (dA n · V)∆t


dV

Term ② can now be written from Equation (2.14), by integrating over the complete surface CSIII
Z Z Z
Term ② = dPIII /∆t = dA · V ρ φ = dA (n · V) ρ φ (2.15)
CSIII CSIII CSIII

A similar expression can be derived for ③ in CSI . The only difference being, in I the direction of V
and that of n are “opposite”, i.e., the expression for the differential volume will have a negative n to
keep the volume positive:
− = (dA (−n) · V)∆t
dV
Therefore we have,
Z Z Z
Term ③ = dPI /∆t = − dA · V ρ φ = − dA (n · V) ρ φ (2.16)
CSI CSI CSI

Combining all the three terms in Equations (2.13), (2.15), and (2.16) we get two equivalent expressions
for the Reynolds Transport Theorem
Z Z
dPsys ∂
= − ρφ +
dV dA · V ρ φ (2.17)
dt ∂t
CV CS
Z Z

= − ρφ +
dV dA (n · V) ρ φ (2.18)
∂t
CV CS

Here we have used the complete integral over the control surface that bounds the control volume to
be Z Z Z
dA = dA + dA
CS CSI CSIII

Reynolds transport theorem basically is an expression for the rate of change of a conserved quan-
tity in a given system (or controlled mass) in terms of variables defined in a control volume (inside as
well as its surface). It simply says that the total derivative (ie, the system derivative) of a conserved
quantity is equal to the sum of that due to the partial time derivative in the CV and the net influx
in to the CV through the CS. Note that this form of the derivative is applicable to control volumes
that is fixed or moving with constant velocity with respect to a inertial frame of reference. We have
derived the model equation for the finite volume case. When the CV is fixed, the form of equation
is called as conservative form. In order to derive the differential form of the conservative equations,
we need to invoke the divergence theorem.

c P Sunthar, March 8, 2010



44 Chapter 2. Differential Analysis of Fluid Flow

2.2.2 Divergence Theorem


For any vector field F defined in the volume V − , divergence theorem relates the volume integral di-
vergence of the field to the surface integral of the normal component of the field.
Z Z
− (∇ · F) =
dV dA (n · F) (2.19)
CV CS

This is a generalisation of the one dimensional integral of a function.

Zb
dx f (x) = F(b) − F(a)
a

dF
where dx = f (x), i.e.,
Zb
dF
dx = F(b) − F(a)
dx
a

Here the one dimensional integral over a line is reduced to a “sum” over the boundary points, ac-
counted for the direction. In two dimensions the divergence theorem will relate an integral over an
area to an integral over the peripheral edge (boundary). Along with Equation (2.19), the Reynolds
Transport Theorem (RTT) Equation (2.18) can be written as
Z Z
dPsys ∂
= − ρφ +
dV − (∇ · u ρ φ)
dV (2.20)
dt ∂t
CV CV

Note that the divergence includes the scalar terms ρ φ. We have also used u for the velocity in place of
V, for convenience of familiarity of the equations to be derived. These both are equivalent, and may
be used interchangeably. V is typically used when the velocity is evaluated on a surface (integral),
and u is used when it is evaluated in a volume (integral).
In order to derive the differential form of the governing equations, we consider an infinitesimally
small control volume, δV − , where the field variables ρ, φ, u are uniform in space. Applying RTT to this
volume we get
!
dPsys ∂
= −
ρ φ + (∇ · u ρ φ) δV (2.21)
dt ∂t

2.3 Fundamental Equations


Lecture 2.3 The fundamental equations for the conserved quantities can be derived from the
conservation laws for the system and RTT

Concepts You Must Know

1. What are the integral and differential forms of the mass conservation law as applied to a con-
tinuum? What is the differential form called as?

We now have the general relationship between derivatives of a conserved quantity on a system
to that in a control volume, we can express the LHS of the conservation laws Equations (2.1) to (2.4)
on a control volume. This is used to derive the familiar equations of fluid mechanics, the derivation
of which you may be familiar with using a shell balance.

c P Sunthar, March 8, 2010



Lecture 2.4. Navier Stokes Equation 45

2.3.1 Continuity Equation


The continuity equation can be derived by using the mass conservation law and the RTT Equa-
tion (2.21). From Table 2.1, we see for P = M, φ = 1. Therefore we get Integral Form:
Z Z
dMsys ∂
= − ρ+
dV dA (n · V) ρ = 0 (2.22)
dt ∂t
CV CS

− can be simplified to
The differential form written for an elemental volume δV

∂ρ
+ ∇ · (u ρ) = 0
∂t (2.23)
∂t ρ + ∂ j (u j ρ) = 0

Note the interchangeable use of u and V, again.

2.3.2 Derivation of Navier-Stokes Equation


Lecture 2.4 Navier Stokes Equation is a “force balance”
Concepts You Must Know
1. What are the integral and differential forms of the Newton’s second law of motion as applied
to a continuum? What is the differential form called as?

2. Interpret each of the terms in the Navier-Stokes equation with respect to the term obtained
from the Newton’s Law.

3. How is a gradient of a stress tensor related to a net force ?

In Equation (2.2), for P = MV, φ = {u, v, w}. We now have to identify the forces. The forces in
Equation (2.2) acted on the system. What are the forces acting on the control volume? Recall that the
implication of the differential equation is an instantaneous balance at time t, that is the LHS and RHS
of conservation equations are evaluated at time t. But at time t the system coincides with the control
volume. Therefore the forces acting on the system (that appear on the RHS) can also be taken to be
the forces on the control volume
X X

Fα = Fα
α sys α CV

The forces acting on the fluid volume can be of two types:

F = FSurface + FBody

The total surface force can be obtained from Equation (1.55) by integrating over the control surface
(or the control volume, using divergence theorem):
Z Z
Fs = dA (n · σ) = − (∇ · σ)
dV (2.24)
CS CV

The local body force is to be integrated over the volume


Z
FB = − ρg
dV
CV

c P Sunthar, March 8, 2010



46 Chapter 2. Differential Analysis of Fluid Flow

For simplicity we consider only one component of the momentum, viz., in the x-direction. For this,
from Table 2.1 we have P = Mu; φ = u. The integral form for the x-momentum conservation reads
Z Z Z Z
dMsys u ∂
= − ρu +
dV dA (n · V) ρ u = dA (n · σ) x + − ρg x
dV (2.25)
dt ∂t
CV CS CS CV

Note the usage of V, which is the surface velocity and the x-component of the velocity u. The x-
momentum integral depends not only on u, but also on the other components coming through the
dot product n·V. To derive the differential form of the momentum balance, we consider an elemental
control volume δV − inside which the field variables can be assumed to be uniform. We also use the
divergence form for the surface flux and the surface force to convert them to volume integrals. This
− , resulting in the differential form
will result in all the terms in Equation (2.25) being multiplied by δV

∂ρ u
+ ∇ · (V ρ u) = (∇ · σ) x + ρ g x (2.26)
∂t
or a more general form in index notation

∂t (ρ ui ) + ∂ j (u j ρ ui ) = ∂ j σ ji + ρ gi (2.27)

Notice the ease of conversion from a scalar form of u to ui : both denote a scalar component of a
vector, but the equations should be read as a collection of three equations as i = 1, 2, 3. The stress
tensor can be expanded in terms of the thermodynamic and viscous stresses σi j = −pδi j + τi j , to give
the Navier-Stokes equations (NSE).

∂t (ρ ui ) + ∂ j (u j ρ ui ) = −∂i p + ∂ j τ ji + ρ gi (2.28)

This LHS of this equation can be further simplified along with Equation (2.23).

∂t (ρ ui ) + ∂ j (u j ρ ui ) = ρ∂t ui + ui ∂t ρ + ρu j ∂ j ui + ui ∂ j (ρu j )
h : i 0

= ρ∂t ui + ρu j ∂ j ui + ∂tρ + ∂
j (ρu j ) ui

This gives another familiar form of the Navier-Stokes equations
D ui
ρ∂t ui + ρu j ∂ j ui ≡ ρ = −∂i p + ∂ j τ ji + ρ gi (2.29)
Dt
We will spend some time recalling how this equation was derived. The LHS came from the “mass
times acceleration” of Newtons second law (converted to control volume variables by RTT). The LHS
is therefore a term related to the acceleration of the fluid packet δV − , and is called as the inertial term.
The RHS is from the “sum of forces” acting on a elemental volume. ∂i p, or for simplicity ∂ x p is the
normal pressure difference (in x-direction) across a fluid packet. If the pressure gradient is non-zero,
then there will be a net normal force acting on the fluid packet. Similarly if there is an appropriate
gradient in the viscous stress, it leads to a net force. A simple illustration would be to consider
∂y τyx , 0, (stress acting on the horizontal surface) which will lead to a net force in the x-direction.
The last term is simply due to the gravitational force.
In the case of Newtonian fluids, the viscous stress is related to the local velocity gradients by
!
  2
τi j = µ ∂ j ui + ∂i u j + κ − µ ∂k uk δi j (2.30)
3

where µ is the Newtonian (shear) viscosity and κ is called as dilatational viscosity, bulk viscosity, vol-
ume viscosity, or second viscosity. The bulk viscosity is identically zero for dilute mono-atomic
gases.

c P Sunthar, March 8, 2010



Lecture 2.5. Forms of energy 47

2.3.3 Energy balance or Heat equation


Lecture 2.5 The forms of energy can be simply stated to be potential and kinetic, each applied to
a microscopic and macroscopic systems.
In order to derive the continuum version of the energy equation, we need to identify the various
forms of energy associated with matter. Because energy is transferable among these various forms,
and only the sum total is conserved. There are various physical forms of energy: kinetic, potential,
thermal, gravitational, light, elastic, electromagnetic. However, all these can be simply categorised
broadly under two heads: Kinetic (energy due to motion) and Potential (energy due to interaction
with a force field). Remember the known force fields are just four: Gravitational, Electromagnetic,
Strong, and Weak nuclear forces. All the example energy forms cited above can be classified under
one of these heads.
The energy of the system Esys in Equation (2.4) therefore consists of
• Internal Energy, U (microscopic Kinetic and Potential)
• Macroscopic Kinetic Energy, M V 2 /2
• Macroscopic Potential Energy, M g · x
where the microscopic forms of energy have been clubbed in to one head called as the internal en-
ergy, U. In a macroscopic or a continuum system, the microscopic forms cannot be distinguished as
separate. The total energy is therefore:

MV 2
Esys = U + + Mg·x (2.31)
2
In terms of the specific local energy density e
Z
Esys = − ρe
dV (2.32)
CV

we associate specific (intensive) quantities

u2
e = Û + +g·x (2.33)
2
Here Û represents the internal energy per unit mass. This completes the identification of forms of
energy on the LHS of Equation (2.4). We now have to identify the constituent terms on the RHS.
The heat input to the system Q̇ can be due to two forms
• Surface transport (Molecular conduction), Q̇s
• Bulk (Radiation)
The surface term can be written as an integral over the system boundary (or the control volume
boundary, since they both coincide at time t).
Z Z
Q̇s = − dA n · q = − dA ni qi (2.34)
CS CS

qi is the local heat flux vector. Note that the -ve sign is so that the integral is positive when heat is
added to the system, ie, n (outward normal) and q are in opposite directions.
The work done by the system is due to the forms of work done by the system on its surroundings.
Some typical work are

c P Sunthar, March 8, 2010



48 Chapter 2. Differential Analysis of Fluid Flow

• Shaft work (due to impellers or other drives), Ẇshaft


• Due to Surface forces (normal and tangential) (Molecular origin), Ẇs
• Other forces: Electrical, Electromagnetic, Ẇother ,
We will ignore the other work for simple fluids (that do not carry any charge). The shaft work cannot
be quantified using the system variables, but only by the properties of the surroundings on which it
acts, so we retain the term as Ẇshaft . The surface work is very important. Remember that the system
is closed, but could be acted upon by surface forces. The system can change its shape against these
forces, therefore it does a work against these forces. The local rate of work done by an elemental area
is dF · V = dA Σi ui . Expanding the stress vector in terms of the local stress tensor and integrating over
the control surface, we get
Z Z
Ẇs = − dA n · σ · V = − dA n j σ ji ui
CS CS
Z (2.35)
=− dA (n j τ ji ui − ni ui p)
CS

The RTT for energy conservation can be written from Table 2.1 and Equation (2.17) as
Z Z
dEsys ∂
= dV− ρ e + dA · V ρ e (2.36)
dt ∂t
CV CS

Inserting this and expressions for the work done from Equation (2.35) and heat input from Equa-
tion (2.34) into Equation (2.4) we get
Z Z Z
∂  

dV ρ e + dA · V ρ e = dA −ni qi + ni τi j u j − p ni ui − Ẇshaft − Ẇother (2.37)
∂t
CV CS CS

which in complete bold face (vector) notation is


Z Z Z

− ρe +
dV dA n · V ρ e = dA n · (−q + τ · u − p u) − Ẇshaft − Ẇother (2.38)
∂t
CV CS CS

Simplifying using divergence theorem, Equation (2.19)


Z Z Z

− ρe +
dV − ∇ · Vρe =
dV − ∇ · (−q + τ · u − p u) − Ẇshaft − Ẇother
dV (2.39)
∂t
CV CV CV

Here we have ignored bulk heat input due to radiation. For further simpler forms of the equation
we will ignore the shaft and other forms of work. It is often convenient to take the pressure term to
the LHS, and writing the density ρ as reciprocal of specific volume V−̂ and simplifying we get:
Z Z i Z
∂ ∂ h  ∂  
− ρ e + dV
dV − −̂ = dV
ui ρ e + p V − −qi + τi j u j (2.40)
∂t ∂xi ∂xi
CV CV CV

When written for an elemental control volume δV − , as for the Navier-Stokes and continuity equations,
we get the differential form of the energy conservation equation for the continuum:
h  i  
−̂ = ∂i −qi + τi j u j
∂t (ρ e) + ∂i ui ρ e + p V (2.41)

c P Sunthar, March 8, 2010



Lecture 2.6. Species balance 49

Special forms of Energy Equation The form of equation given above is not suitable for most of the
heat transfer applications involving fluid flow. They can however be simplified. One of the most
common form is that expressed in terms of the temperature T . To obtain this, the mechanical energy
equation is subtracted from the above equation. The mechanical energy is not a conserved quantity,
but can be obtained from the momentum balance by multiplying each term by ui , and simplifying it
with the help of the continuity equation. The resulting equation, after subtraction, is the equation of
change for the internal energy: Internal energy equation (Total − Mechanical):

D Û
∂t ρÛ + ∂i ui ρÛ = ρ = −p∂i ui − ∂i qi + τi j ∂i u j (2.42)
Dt
D
Here we have expressed the LHS in terms of the substantial derivative, or the material derivative Dt.
For any scalar φ we can show using continuity equation that

h i h : i 0 Dφ
∂t (ρφ) + ∂ j (u j ρφ) = ρ ∂t φ + u j ∂ j φ + φ ∂ 
tρ + ∂
j (ρu j ) ≡ ρ (2.43)
 Dt
We can make further simplifications to Equation (2.42), in terms of the thermodynamic function
enthalpy (per unit mass) Ĥ. In terms of Enthalpy,

D Û D Ĥ D p
ρ =ρ − (2.44)
Dt  Dt Dt 

 ∂V−̂ 
dĤ = c p dT + V−̂ − T  dp (2.45)
∂t p 

The most useful form of the energy equation is written in tems of the temperature, which is ob-
tained by substituting Equation (2.45) in Equation (2.44) and further in Equation (2.42). In terms of
Temperature:
DT ∂ ln ρ D p
ρ cp = −∂i qi + τi j ∂i u j − (2.46)
Dt ∂ ln T p D t

Simplified forms of Temperature Equation The temperature equation Equation (2.46) can be sim-
plified further for various special cases.

1. Ideal gas (p/ρ = R T/Mw , giving ∂∂ lnln Tρ = −1; qi = −k∂i T )
p

DT Dp
ρ cp = k∂2i T + µ∂i u j ∂i u j + (2.47)
Dt Dt
Dp ∂ρ
2. Constant Pressure System, Dt = 0 or Constant density system ∂T = 0 and Newtonian viscosity
DT
ρ cp = k∂2i T + µ∂i u j ∂i u j (2.48)
Dt

2.3.4 Species Balance


Lecture 2.6 Species balance is similar to continuity equation but written only for one particular
species
In the case of multicomponent systems, the chemical species is conserved as well. Chemical reactions
can take a species from one compound to another. In this case we can write a “mass” balance for a
species α alone.
Z
dMα
= ṙα where, Mα = dV − ρα (2.49)
dt sys
CV

c P Sunthar, March 8, 2010



50 Chapter 2. Differential Analysis of Fluid Flow

The left hand side can be expanded by the Reynolds transport theorem, Equation (2.20), just as it was
done for the continuity equation, except that the local velocity of the control surface will be uα of that
P
of the species α, and not the mass averaged velocity u = α ρα uα /ρ .
Z Z
dMα ∂
= − ρα +
dV dA (n · uα ) ρα (2.50)
dt ∂t
CV CS

Adding and subtracting a term proportional to the mass average velocity u, we get for the species
balance an integral form
Z Z Z
dMα ∂
= − ρα +
dV dA [n · u] ρα + dA [n · (uα − u)] ρα = ṙα (2.51)
dt ∂t
CV CS CS

Differential Form of Species Balance The differential form is obtained, as before, by considering
an elemental volume δV− where the properties are uniform.

∂t ρα + ∂i ui ρα = −∂i ρα (uα i − ui ) + ṙα (2.52)


| {z }

In terms of mass fraction ωα = ρα /ρ, and using Equation (2.43)


D ωα
∂t (ρωα ) + ∂i (ui ρ ωα ) = ρ = ∂i (ρ Dα ∂i ωα ) + ṙα (2.53)
Dt
where, Ficks law for diffusion flux has been used

jα ,i = −ρDα ∂i ωα (2.54)

Lecture 2.7 Problems in Process Fluid Mechanics can be taken to be two broad types: Macro-
scopic and Microscopic.
Problems in process fluid mechanics can be broadly classified under two categories: Macroscopic or
simply Macro-, and Microscopic or Micro-. Macro-problems deal with gross properties of the flow,
such as total drag, work done, energy requirement, equipment sizes, etc, without requiring to find
out the variation in the velocity and stress fields. Macro-problems are usually simple to solve math-
ematically, but may involve several unknowns, which need to be determined either by experiments
or by solving the microscopic problem. Micro-problems involve finding the detailed velocity and
stress fields and using them to calculate other properties, which might be used in macro-problems.
Micro-problems are often difficult to solve mathematically to obtain a closed form analytical solution.
In such cases, a numerical (or computational) solution of the microscopic equations are obtained.
Navier-Stokes equations accurately describes models the behaviour at the normal densities of New-
tonian fluids (only in very low densities, such as that of a plasma, or sudden variation density, as in
a shock wave, does it break down). In such cases, with the availability of computational power, the
computational route of solving the equation is fast emerging as tool to obtain the behaviour of the
fluids. This is called as computational fluid dynamics (CFD), where experimental conditions (geometry
and boundary conditions) are simulated on a computer, and the microscopic balance equations are
solved numerically.
Some macroscopic problems include
• Determination of a pump specification (RPM and size), and energy required to transport a fluid
across a height at a given flow rate.
• Determination of the pressure drop across a bed packed with solids through which a fluid has
to pass at a given flow rate.

c P Sunthar, March 8, 2010



Lecture 2.8. Introduction Boundary Layer 51

• Determination of the energy required to stir (agitate) a tank containing a liquid which is to be
mixed with other components, which may be a bubbling gas, another liquid, or solids (slurry).
• Determination of the rate at which a filter used to separate solids from a liquid can produce the
filtrate.
Microscopic problems are not “microscopic” in the strict sense of the word. We are still dealing
with continuum here. However, we need to find out how the velocity and stress fields vary or
develop inside (or outside) a solid object (equipment).
• Determination of the velocity profiles around a rough sphere (a cricket ball?) as it moves
steadily in a fluid.
• Determination of local heat transfer coefficient, which is dependent on the local velocity varia-
tions. A convective current towards the wall increases the heat transfer as well.
• Determination of net rate of reaction on a catalyst surface which is suspended (fluidised) in air
by its upward flow.
The equations derived in the last section can be applied to macro- as well as micro-problems. The
integral form (over the control volume) is used to set up a macro-problem and the differential form
is used to solve the velocity, stress, temperature and concentration fields. In the next lecture we will
consider an important concept in fluid mechanics, namely the boundary layer, in which both the
macroscopic and the microscopic methods will be applied.

2.4 Boundary Layer Theory


Lecture 2.8 Boundary layers occur in High Reynolds number flows, close to a solid boundary
Concepts You Must Know
1. Boundary layers is a relatively small zone close to a solid surface, in which the velocities vary
rapidly in comparison with a slow variation just outside.
2. Boundary layers occur in high Reynolds number flows, i.e. when the viscous terms are negli-
gible in comparison with the inertial terms.
3. A laminar boundary layer is one in which the velocities are in streamlines.
4. A Turbulent boundary layer has velocities in all directions without any uniformity.
5. The requirement of a Boundary layer is the condition of no slip imposed on the solid surface.
6. The thickness of a Boundary layer (from the solid wall) can vary along the length of the flow,
often very slowly, and therefore a thickness is usually adopted by some convention.
7. The boundary layer influences the transport from the solid to the bulk and therefore is crucial
in determination of drag, heat transfer, and mass transfer coefficients.

It is convenient to start with what is known as a flat-plate boundary layer in which a fluid moving
at a uniform velocity (therefore laminar) approaches a flat plate whose normal is perpendicular to
the flow, as shown in Figure 2.4. At the up-stream end (leading-edge) of the plate the velocities are
uniform, and in the down stream, very far away from the plate the velocities is again uniform. Only
close to the plate there is a layer where the influence of the plate is felt. The plate exerts a drag on the
fluid (no-slip at the walls), which consequently influences (drags) other layers away from the plate.
The laminar boundary layer develops along the flow, and its thickness increases. The flow remains
viscous and laminar. Beyond a critical thickness, turbulence sets in: due to some disturbances in the

c P Sunthar, March 8, 2010



52 Chapter 2. Differential Analysis of Fluid Flow

Turbulent Boundary Layer


y
Buffer Layer
x δ Laminar Boundary Layer
Laminar sub−layer

Flat Plate

Figure 2.4: Development of boundary layer over a flat plate

y a b

x δ

d c
Flat Plate
Figure 2.5: Compute net flow across control surface ab.

flow or roughness in the solid surface. The boundary layer, called as the turbulent boundary layer is
then dominated by turbulent patterns (3D flows, fluctuations, non-uniformity, eddies).
Even in the turbulent BL, a still smaller layer persists where the flow is still laminar. This layer is
called as the laminar sub-layer. This is important to heat and mass transfer because this is the region
of dominant resistance: There is no cross flow, and all the transport must occur through molecular
transport: diffusion and conduction. In between the viscous layer and the turbulent layer is a layer
known as the buffer layer, where the viscous and turbulent transport are equally important.

2.4.1 Integral Analysis


Lecture 2.9 Computing the net mass flow in a boundary layer problem.

Tutorial Example A flat plate of width 3 mm is approached by air (ρ = 1kg/m3 ) having an uniform
velocity of 2 m/s, as shown in Figure 2.5. Calculate the net mass flow rate across the surface
ab. Assume that the velocity inside the boundary layer is given by u/U = 2(y/δ) − (y/δ)2 , and
the boundary layer thickness at c is δ = 5 mm.

Given Data

ρ = 1 kg/m3
w = 3 mm = 3E−3 m
δ = 5 mm = 5E−3 m
U = 2 m/s

Governing Equations :

c P Sunthar, March 8, 2010



Lecture 2.9. Integral Analysis of BL 53

Mass balance Equation (2.22)


Z >
 Z

−ρ + dA (n · V) ρ = 0
dV

∂t 
 CV CS

Assumptions :

① Steady flow
② Laminar flow
③ FM convention for normal (outward)
④ Incompressible flow (constant ρ).

Detailed calculations : The integral over the control surface can be evaluated in four parts over the
four surfaces Z Z Z Z Z
= + + +
CS ab bc cd ad

Before evaluating each of the integrals, let us do a qualitative analysis of the problem. Why
should there be a flux across ab ? If at all there is one, in which direction will the net mass flow
be? The total mass flow across ad (uniform velocity) is more than that in bc (reduced from the
uniform). There is no flux of mass across cd (no-slip), therefore there must be a net out flux
across ab. This is only a guess, we could be wrong.
To evaluate the integral we need to know the variation of the local velocity vector V and the
normal n. We first choose the origin and direction of the coordinate system, as shown in Fig-
ure 2.5.
Across ab we do not know the velocity distribution (not given, so do not assume any). All we
can say is that u x = U across ab. We do not know about uy . Across bc, we only know u x , not uy .
Across cd, V = 0 because of no-slip, and finally across ad, V = {U, 0}. With this information and
with the direction of the local normal we can evaluate the four integrals

1. Across ab, n = {0, 1}, points in the y direction, so the only component that will matter is uy ,
but since this is not known, we denote the integral as the total mass flow rate
Z
ṁab ≡ dA uy ρ
ab

If it is positive (or negative) then the specific mass flux V is in the same (or opposite)
direction to the outward normal n, respectively.
2. Across bc, n = {1, 0}, and only u x matters. uy , even though unknown, does not contribute
to the integral. Substituting the given expression for the variation in the velocity and
integrating we get

Z Zδ !
y  y 2 2ρ U w δ
dA u x ρ = w dy U ρ 2 − =
δ δ 3
bc 0

Note that since the outward normal and the velocity are in the same direction, this integral
evaluates to a positive quantity.
3. The integral over cd evaluates to zero because velocity is zero due to the no-slip boundary.

c P Sunthar, March 8, 2010



54 Chapter 2. Differential Analysis of Fluid Flow

b
y
a

δ0 δ
x
d c
Flat Plate
Figure 2.6: Net mass flux across control surface ab.

4. The integral over ad also has contribution only from u x = U, however n = {−1, 0} in the
negative x direction. The part integral is given by

Z Z Zδ
dA (n · V) ρ = − dA U ρ = −w dy U ρ = −ρ U w δ
ad ad 0

Putting all the integrals together


Z Z Z
ṁab + + + =0
bc cd ad

we get,

2
ṁab = ρ U w δ − ρU wδ
3
1
= ρU wδ
3
Since the mass flux across ab evaluates to be positive, it (V) is in the direction of the normal n,
and there is a net out flux of mass from the surface ab, in agreement with our initial guess.
Symbolic Expression for the Unknown :

1
ṁab = ρU wδ
3
Numerical : Substituting the values the net mass flux out of the surface bc is

ṁbc = 10E−6 m3 /s = 10 cc/s

Lecture 2.10 There is a net influx of mass into the boundary layer
In the same situation as considered in the previous problem, consider a slightly different problem of
the mass flux across the boundary layer, as shown in Figure 2.6. In this problem we wish to find out
the net mass flux across the boundary layer. This is different in that when the control surface ab is
along the boundary layer, we get the net flux across the curved BL, and not a horizontal surface as
before.
To solve this problem, we consider two thickness δ0 at x = d and δ at x = c. The imaginary control
volume is enclosed by is abcd. Most of the derivation is same as that for the previous case. We will

c P Sunthar, March 8, 2010



Lecture 2.11. Integral Momentum for Boundary Layer 55

n α
b

V

a α

p(x+dx)

p(x) δ δ +d δ

dx
d c

x x + dx
τw

Figure 2.7: Combination of differential and integral analysis in Boundary Layer: Differential length
along x and finite length along y

only point out the difference here. We are still wish to obtain an expression for the mass flux across
ab, denoted by ṁab . The normal to this surface changes with x, but the following expression is still
correct
Z
ṁab ≡ dA n j u j ρ
ab

The difference is that we cannot write it in terms of known n j as before. The integral across bc and
cd are the same, but across ad, the integral is similar to that of bc but over a different boundary layer
thickness δ0 .
Z
2ρ U w δ0
dA u x ρ = −
3
ad

Combining all the fluxes we get


2
ṁab = ρ U w (δ0 − δ) (2.55)
3
Taking δ > δ0 as an observable fact, we find that ṁab < 0. Recall that mass flow rate is positive along
the direction of the outward normal (mass flux per unit area ρV is in the general direction of the
normal). Therefore the above equation implies ṁab < 0, and there is a net influx into the boundary
layer. This would seem immediately surprising, since we had just derived a relationship where there
was a flux outside surface ab. But on hindsight we can see that in the earlier case we had integrated
a flat velocity profile (upstream) against a deficient velocity profile (downstream). Here however, we
are comparing a more deficient (upstream) against a lesser deficient (downstream).

Lecture 2.11 The mass and momentum balances is done on a differential element along the di-
rection of flow and a finite length along the boundary layer thickness.
We will now carry out an analysis which is differential in one direction and integral in another. This is
particularly useful in interpreting some qualitative (but very important) features of boundary layers.
Consider control volume shown in Figure 2.7, which is a section of the boundary layer between
lengths x and x + dx, with the thickness going from δ to δ + dδ. We list the directions and magnitudes

c P Sunthar, March 8, 2010



56 Chapter 2. Differential Analysis of Fluid Flow

of known vector quantities across the four surfaces. nab = {− sin α, cos α}, nbc = {1, 0}, ncd = {0, −1},
nad = {−1, 0} and Vab = {U, ?}, Vbc = {u, ?}, Vcd = {0, 0}, Vad = {u, ?}.
Applying the continuity equation to this system, and without assuming any specific velocity
profile (unlike the previous examples) the general expression for the mass flux across ab can be shown
to be
Z Z Z Zδ δ+dδ
Z
ṁab ≡ dA (n · V) ρ = − dA (n · V) ρ − dA (n · V) ρ = w dy ρ u − w dy ρ u (2.56)
ab ad bc 0 0

Here u denotes the x-component of the fluid velocity. The second term can be expanded in Taylor’s
series about x
     δ 
Zδ  Zδ  Zδ   Z 
 ∂ w dy ρ u dx = −w dx ∂  
ṁab = w dy ρ u − w dy ρ u +  dy ρ u  (2.57)
 ∂x    ∂x 
0 0 0 0

Note the x-direction is a differential balance and the y direction is an integral equation.
For the integral momentum balance at steady state, we write down each of the terms from Equa-
tion (2.25). The surface integral on the LHS
Z Z Z Z Z
dA (n · V) ρ u = + + + (2.58)
ab bc cd ad
CS

The integral across the two surfaces ad and bc can be written, similar to the continuity equation, as
     δ 
Z Z Zδ  Zδ  Zδ   Z 
 ∂ w dy u ρ u dx = w dx ∂  dy u ρ u
+ = −w dy u ρ u + w dy u ρ u + (2.59)
ad bc  ∂x    ∂x  
0 0 0 0

This is because, across ad and bc the only component in V that is effective is u, along the direction of
the normal. The integral over cd is zero as the velocities are zero. Across ab, we note the following.
The direction of the normal (the angle α) is not known a priori, and the velocity V over the surface has
both the components u and v non-zero in general. Therefore no simplification can be done on n · V.
However the other velocity u, appearing in the integral on its own, is equal to the local free-stream
velocity U on the boundary layer surface (by definition), therefore
Z Z
dA (n · V) ρ u = U dA (n · V) ρ = U ṁab (2.60)
ab ab

where the second equation has been obtained from Equation (2.57). Note that in general U = U(x)
can be a function of x outside the boundary layer. Only for a flat plate is it a constant. Even then,
for a differential element we can assume it to be a constant across ab. This completes the LHS of
Equation (2.25). Next we move on to the forces.
We neglect the body forces in the x-direction. For the surface force in the x-direction
Z Z
dA n j σ jx = dA (−pn x + n j τ jx ) (2.61)
CS CS

This implies p will contribute only in surfaces where n x is non-zero, namely on ab, bc, ad. On ab
n = {− sin α, cos α}, and assuming the pressure at the midpoint acts uniformly on the whole of ab, we
get
Z !
dx 1 ∂p
dA (−pn x ) = −w p(x) + dx (− sin α) (2.62)
cos α 2 ∂x
ab

c P Sunthar, March 8, 2010



Lecture 2.11. Integral Momentum for Boundary Layer 57

This can be further simplified with the local slope tan α


! 0
*
Z
dδ ∂p dδ
dA (−pn x ) = w dx p + O  (2.63)
dx  ∂x dx
ab

Across the other two surfaces the pressure term can be simplified to
Z Z ! !
∂δ ∂p
dA (−pn x ) + dA (−pn x ) = w δ p − w δ + dx p + dx
∂x ∂x
ad bc (2.64)
∂p ∂δ  *
  0
= −w dx δ − w dx p +Odx2
∂x ∂x 
Combining the three integrals we have for the pressure force term
Z
∂p
dA (−pn x ) = −w dx δ (2.65)
∂x
CS

where, higher order derivatives have been neglected as the leading term in the final expression is
O (dx ∂ x p).
The viscous stress term can be expanded to give n x τ xx + ny τyx . For an incompressible fluid τ xx = 0.
So the only component we have to consider is the shear stress τyx , which has no effect on ad and bc
where ny = 0. On ab, ny is not zero, but outside the boundary layer ∂y u = 0, as u saturates to U.
Therefore τyx = µ∂y u = 0 on the surface ab. So the shear stress has contribution only in the bottom
surface cd, where ny = −1.
Z Z
dA n j τ jx = dA ny τyx ≡ −w dx τw (2.66)
CS cd
where τw is the local average wall shear stress over the length dx, defined by the above equation.
Combining the LHS (from Equations (2.58) to (2.60)), the expression for ṁab from Equation (2.57),
the pressure term Equation (2.65), and the viscous term Equation (2.66) we get
 δ   δ 
Z  Z 
∂   ∂  ∂p
w dx  dy u ρ u − U w dx  dy ρ u = −w dx δ − w dx τw
∂x   ∂x   ∂x
0 0

which can be simplified, for incompressible flow, to


 δ   δ 
Z  Z 
∂  ∂ δ ∂p τw
 dy u2  − U(x)
 
 dy

u = − − (2.67)
∂x   ∂x   ρ ∂x ρ
0 0

This is the general momentum balance equation of the boundary layer, where there can be a variation
in the free stream velocity U(x), and the pressure p(x) outside the boundary layer.

Find Out Yourselves


1. When can τ xx have nonzero values?
2. List all the assumptions made when simplifying the momentum balance to the boundary layer
case in arriving at Equation (2.67)

2.4.2 Laminar and Turbulent Boundary Layers

c P Sunthar, March 8, 2010



58 Chapter 2. Differential Analysis of Fluid Flow

Lecture 2.12 Integral Momentum analysis across a boundary layer gives the scaling behaviour
of the boundary layer thickness and the wall drag with length.
Equation (2.67) can be simplified further for a flat plate boundary. In this case, the velocity distribu-
tion outside the boundary layer is uniform U(x) = U = constant. In the absence of velocity gradients,
the NSE Equation (2.29) shows that pressure distribution is uniform as well (∂i p = 0). From Equa-
tion (2.67) we then have
 δ   δ 
Z  Z 
∂  ∂  dy u = − τw
 dy u2  − U (2.68)
∂x   ∂x   ρ
0 0

To solve this integro-differential equation, we need to know the function u = u(x, y), which requires
the solution of the NSE Equation (2.29). However, we will adopt a simplified approach, but will get
some important qualitative predictions. We assume that the function u(x, y) can be written of the
form
!
y
u(x, y) = U f (x, y) = U f
δ(x)
! (2.69)
u y
ϕ≡ = f ≡ f (η)
U δ(x)

where the dependence on x is through δ, and η = y/δ (η is pronounced as /eeta/, and ϕ is a variant
of the normal φ /fai/) . This form is dimensionally consistent. On the LHS we have a dimensionless
number, it has to be a function only of dimensionless parameters, as the resultant has to be a number.
A function of this form is called as a similarity function, because the form of the function remains the
similar for all x. Though it is difficult to justify this assumption now, we take it as a plausible guess.
The momentum balance Equation (2.68) can be simplified as

Z1
dδ µ dϕ
dη ϕ (1 − ϕ) = (2.70)
dx δ U ρ dη η=0
0

where we have used the shear stress at the wall to be



∂u
τw = µ (2.71)
∂y y=0

We now assume a velocity distribution function that satisfies the boundary conditions:

y = 0, u/U = 0
y = δ, u/U = 1
y = δ, ∂y u = 0

A simple choice is a quadratic function

u  y   y 2
ϕ= =2 − = 2η − η2
U δ δ
Substituting in Equation (2.70) we get
dδ 15 µ
=
dx δ U ρ
Solving for δ we get
30 µ
δ2 = x

c P Sunthar, March 8, 2010



Lecture 2.13. Turbulent Boundary Layers 59

where we have specified δ = 0 at x = 0. This equation can be written equivalently as


s
δ 30 µ 5.48
= ≅ √ (2.72)
x xU ρ Re x

This gives an important relation for the behaviour of δ: the boundary layer increases as the square
root of the distance from the leading edge. In other words, the dimensionless boundary layer thick-
ness at any length x from the leading edge, scales as the local Reynolds number power minus half.
What happens if we assume another functional form for ϕ? From Equation (2.70), we note that the
functional form only modifies constants in the equation. In general we will get

dδ µ
=C
dx δU ρ

where C is a constant. The functional form for δ still remains similar to Equation (2.72)
r
δ 2C
= .
x Re x

The dimensionless wall shear stress is also an important quantity for practical problems. In boundary
layer flows one would like to know the extent of viscous drag force exerted by the boundary on the
fluid. Important applications include, the drag on aeroplanes, cars, and ships, where boundary layers
play an important role. The wall shear stress is made dimensionless by the quantity 12 ρ U 2 (which is
the magnitude of the inertial terms in the equation of motion; more of this we will learn shortly), and
is called as the skin friction coefficient (“skin” refers to the drag due to the boundary layer which is
close to the solid surface like a skin)
τw
Cf ≡ 1
2
2 ρU

From Equation (2.71) we have for the quadratic ϕ function



2 µ dϕ 4µ Re x 0.730
Cf = ≅ ≅ √
ρ U δ dη ρ U x 5.48 Re x

Even in the case of coefficient of drag, changing the functional form only changes the constant that
appears in the final equation.

Lecture 2.13 Experimental velocity profiles for turbulent flows yield a scaling of Re−1/5 instead
of Re−1/2 for δ/x and Cf .
For turbulent flows, the velocity profiles cannot be determined as closed form solutions. Instead, ex-
perimental observations are used. The nature of turbulent flow is strongly determined by the rough-
ness of the boundary. Turbulence occurs even in so-called “smooth” pipes, where the local rough
peaks is much smaller than the laminar boundary layer thickness. Here the size of the roughness
does not affect the turbulence. The velocity profiles in a smooth circular pipe can be approximated
by a power-law equation
u  y 1/n
=
U R
where y is the distance from the wall, R is the pipe radius, and n is the power law index. We can
assume that for a flat plate boundary layer too the velocity has a similar form

u  y 1/n
=
U δ

c P Sunthar, March 8, 2010



60 Chapter 2. Differential Analysis of Fluid Flow

A value of n = 7 is commonly used. In order to find the scaling behaviour of the boundary layer
thickness and the skin friction coefficient, we can use this expression.
ϕ = η1/7
This will not be a problem in evaluating the integral terms appearing on the LHS of Equation (2.70),
however, the wall stress term will result in
dϕ 1
= (2.73)
dη 7 η6/7
which diverges for η = 0 (at the wall). Therefore we need to look for alternate expressions for the
wall shear stress. One way to obtain this is to use experimental correlations observed for a smooth
circular cross section pipe and use it for a flat plate. For a smooth pipe the wall stress is given by
the Fanning friction factors correlations, that relates the dimensionless wall stress to the Reynolds
number. This is empirically obtained from the Blasius equation
!1/4
τw −1/4 µ
fF ≡ 1 = 0.079 Re = 0.079 (2.74)
2 D Ū ρ
2 ρ Ū

where Ū is the mean velocity in the pipe. In order to use this expression for a flat plate, some
approximations have to be made. Firstly recognise that there is no mean velocity Ū for a flat plate.
The skin friction coefficient is given in terms of the maximum velocity outside the boundary layer U.
These two are related by
Z ZR 
1 1 r 1/n
Ū = dA u ≈ dr 2 π r 1 −
π R2 π R2 R
0

For a power law velocity profile, assuming the profile to extend from wall to centerline (the viscous
boundary layer is small) the relationship can be shown to be
Ū 2 n2
=
U (n + 1) (2n + 1)
For n = 7 this evaluates to Ū = 0.817 U. The expression for τw can be written down in terms of the
Blasius equation Equation (2.116) as
!1/4
τw µ
1
= 0.0466
2ρ U
2 RU ρ
Assuming that this law applies to the flat plate boundary layer with R = δ, we get
!1/4
τw µ
1
= 0.0466 (2.75)
2ρ U
2 δU ρ
This expression is directly substituted in Equation (2.68), and the integrals on the LHS are evaluated
using the power law profile Equation (2.73) to give:
!1/4
7 dδ µ
= 0.0233
72 dx δU ρ
which can be solved for δ to give
δ
= 0.381 Re−1/5
x
x
The skin friction can be obtained from Equation (2.75)
Cf = 0.0593 Re−1/5
x

The boundary layer thickness for laminar and turbulent flows is compared in Figure 2.8.

c P Sunthar, March 8, 2010



Lecture 2.13. Turbulent Boundary Layers 61

0.04
Laminar
Turbulent
0.035

0.03

0.025

0.02
δ

0.015

0.01

0.005

0
0 0.2 0.4 0.6 0.8 1

x
Figure 2.8: Developing boundary layers as computed from the laminar and turbulent flow cases, for
Re x = 1E5 at x = 1.

c P Sunthar, March 8, 2010



62 Chapter 2. Differential Analysis of Fluid Flow

Deficient
Mass flow rate

y Equal
δ Areas
δ∗
x

Figure 2.9: Illustration of displacement thickness in an inviscid flow in which the mass flow rate
deficiency is equal to that in the boundary layer.

Lecture 2.14 Displacement Thickness is a convenient experimental measure of the local bound-
ary layer thickness when the velocity profiles can be measured.

From the analysis we have seen so far, we said that the boundary layer thickness is a distance from
the wall when the local velocity becomes equal to the free-stream velocity. In reality, the local velocity
u only asymptotically becomes equal to U, that is the variation is smooth and u → U as y → ∞. The
location of the boundary layer is dependent on the accuracy of measurement of the velocity u. As
the resolution of measurement increases, so does the specified thickness of the boundary layer! To
reduce this discrepancy in reporting, conventions such as u/U = 0.9 and u/U = 0.99, are adopted
as the edge of the boundary layer. Note that at smaller u/U, the discrepancy in δ arising due to
measurement accuracy in u is also less. There are alternate convenient measures of the boundary
layer that are not so much sensitive to single point values, and are a representation of several values,
such as an integral over the boundary layer: The displacement thickness δ∗ and momentum thickness
θ. In the channel flow case, since the boundary layer is developing, the displacement thickness also
changes along the direction of flow. The displacement thickness can be measured by measuring the
free stream velocity and the pressure (see Assignment problem).
Consider a flow past a flat plate shown in Figure 2.9 as an example. We studied in Lecture 2.9
that there is a net mass flow outside section ab, because there is a deficiency in the horizontal velocity
profile. A similar interpretation can be done for a local velocity profile, instead of a section. At any
section x, if the velocity profile is compared with a uniform flow, without any viscous (boundary
layer) effects, then there is a deficiency in the velocity profile. This means the total horizontal mass
flow rate across the local cross section (with depth w) is smaller than that present when there is no
boundary layer
Z∞ Z∞
dy w ρ u < dy w ρ U
0 0

The deficiency in the boundary layer case is given as

Z∞
dy w ρ (U − u) (2.76)
0

This reduction in the horizontal mass flow rate can be thought of as a reduction in a inviscid flow
when the boundary is itself displaced by a distance δ∗ . The deficiency in the inviscid flow is

Z∞ Zδ∗
dy w ρ (U − U ∗ ) = dy w ρ U = ρ U δ∗ w (2.77)
0 0

c P Sunthar, March 8, 2010



Lecture 2.14. Displacement Thickness 63

Fluid Flow in a Channel

Cross Section
Velocity
Profiles

δ∗

Figure 2.10: Illustration of displacement thickness in a channel flow.

where U ∗ is defined as U ∗ = 0 for 0 ≤ y ≤ δ∗ , and U ∗ = U for y ≥ δ∗ . Equating the two deficiencies


(Equations (2.76) and (2.77)) we get an expression for δ∗ in terms of the local velocity profile
Z∞ 
∗ u
δ = dy 1 −
U
0

δ∗ is a measure of the boundary layer thickness, which depends on the local velocity profile u and not
a single point value (such as u/U = 0.99). For example for a quadratic velocity profile we get δ∗ = δ/3.
The illustration of the displacement thickness is better illustrated in a fluid flow in a channel.
Consider flow in a channel as shown in Figure 2.10; for simplicity we take the channel to have only
top and bottom walls–and is infinitely wide down the depth. The actual velocity profile, with two
boundary layers on the top and bottom of the channel, is shown in blue colour. The red coloured
curve has the same amount of mass flow rate deficiency at each of the walls as does the blue curve.
Note that the red and blue curves have the same free stream velocity. The displacement thickness δ∗
is defined such a way that the loss horizontal of mass flow rate in the boundary layer is equal to the
same loss that happens in an ideal fluid with the same free stream velocity but with the boundary
displaced by δ∗ .

c P Sunthar, March 8, 2010



64 Chapter 2. Differential Analysis of Fluid Flow

2.4.3 Differential Analysis


Lecture 2.15 Ordering analysis is an method to obtain approximate solutions to algebraic and
differential equations.
So far we were only concerned with the general behaviour of the boundary layer thickness. We
assumed some general profile for the velocity in the boundary layer. In this section we seek the
exact solution of the velocity profile. For this we have to turn to the differential balance of mass and
momentum.
The governing differential equations of fluid mechanics, the Navier-Stokes equation, Equation (2.29)
is a non-linear equation in the velocity, because of the terms u j ∂ j ui . In some situations, this term is
identically zero, as in the case of unidirectional laminar flow and can be neglected in low Reynolds
number flows. But in most other situations this term is the origin of the difficulties in obtaining a so-
lution to the problem. Particular examples are boundary layer, diverging, converging, and turbulent
flows. In these situations there are two ways to obtain the solution: one is a numerical method and
the other is to use approximate methods. Here, we will introduce one such approximation technique,
which was initially applied to the viscous boundary layer problem. Similar techniques used in other
physical phenomena are also called as boundary layer solutions.

2.4.3.1 Asymptotics and Scaling Analysis


The basic idea is to obtain a simplified set of equations, for which exact solutions can be computed.
First the governing equations are simplified by neglecting terms. This renders the equations tractable
(meaning a solution can be obtained). The underlying assumption is that an exact solution to the
approximate equations is also an approximate solution to the exact equations. This is mostly true, and
is valid only in the parameter space where the approximations are valid, but there are interesting
counter examples where it fails (See Denn, 1980, Chapter 11, for example).

Example 1: Algebraic Equation We will illustrate the concept by a simple algebraic equation. We
consider determining the roots of a cubic equation

x3 − 4.001x + 0.002 = 0 (2.78)

This solution to equation is equation is “formidable”, unless you can recall the formula for the cubic
roots, but you will see that an approximate solution can be obtained quickly and at the back of an
envelope! Though we know the solution to the roots of the equation (x = 2, −2.0005, 0.0005), we
will adopt an approximate scheme to simplify the equation and then obtain an exact solution to it
(approximated equation).
Firstly, we observe that there are some decimal numbers with some kind of insignificance. For
example, 4.001 ≈ 4, 0.002 ≈ 0 are some initial guesses. But such approximations need to be carried out
carefully. One way to do it is to try and parametrise the equation. In this case only one parameter is
required, let us call it as ǫ (pronounced /epsilon/). ǫ is a small parameter, we rewrite Equation (2.78)
as
x3 − (4 + ǫ) x + 2ǫ = 0 (2.79)
and in this case ǫ = 0.001. Some points about the expected solution are in order

• The solution to x is a function of the parameter ǫ: x = f (ǫ).

• ǫ is a small parameter in comparison to 1, denoted as ǫ ≪ 1.

• Since the complete problem is “difficult” to solve, we seek an approximate solution in the limit
ǫ → 0.

In the asymptotic approximation scheme, we

c P Sunthar, March 8, 2010



Lecture 2.15. Scaling and Ordering Analysis 65

1. Make a guess on the behaviour of x or equivalently of f (ǫ) as ǫ → 0.


2. Carry out simplifications of the equation
3. Seek an exact solution
4. Check if the solution is consistent with the initial guess.
We will use the notation O (.) (read as “order of”) to denote the asymptotic behaviour of a function,
for example:
1. x = f (ǫ) ∼ O (1) as ǫ → 0, meaning x approaches a constant of order unity.
2. x = f (ǫ) ∼ O (ǫ) as ǫ → 0, meaning x approaches zero in a similar rate as does ǫ. If we Taylor
expand the solution about zero, then the first term would be proportional to ǫ. Eg: sin ǫ ∼ O (ǫ)
but cos ǫ ∼ O (1).
 
3. x = f (ǫ) ∼ O 1ǫ as ǫ → 0, meaning x diverges to ∞ in a similar rate as does 1/ǫ.

4. In general we can write x = f (ǫ) ∼ O (ǫ n ), where ∼ means asymptotically equal to, if

f (ǫ)
lim = Constant
ǫ→0 ǫn
If f (ǫ) is written as a Taylor series in ǫ, then it would be

f (ǫ) = ǫ n fn + ǫ n+1 fn+1 + . . .

where fn and fn+1 are all constants. fn is called as the leading term or the dominant term in the
series. A pictorial illustration of the asymptotic behaviour of a function is shown in Figure 2.11.
 
5. If x ∼ O (ǫ), then x2 ∼ O ǫ 2 : it is the leading term in a product of two Taylor expansions.
6. So far we used O () for the asymptotic behaviour of dimensionless numbers x = f (ǫ) in terms
of some other small dimensionless number ǫ. For a dimensional physical quantity, O (.) is also
used to indicate the physical order of magnitude. When we say the velocity u is u ∼ O (1) m/s,
we imply that u is within one order of magnitude from 1 m/s, i.e., 0.3 m/s . u . 3 m/s. In a log
scale one order of magnitude represents one linear unit, and 3 ( ≈ 100.5 ) is approximately half
way between 1 and 10 orders. In linear scale just as we round off any 0.5 ≤ x < 1.5 as x ≈ 1, in
log scale or in physical order of magnitude analysis we say for any 3 ≤ u ≤ 30 as u ∼ O (10).
Let us carry out an asymptotic analysis of the Equation (2.79). In asymptotic analysis we have to
make a series of guesses if required (if we have not found all solutions, or if our guess is inconsistent)
1. Let x = f (ǫ) = O (1) as ǫ → 0, then the terms in Equation (2.79) have orders as under.

Terms: x3 −4x −4ǫ x +2ǫ


O (.) : 1 1 ǫ ǫ

Since ǫ ≪ 1, we can neglect the third and forth term in comparison with the first two and we
get

x−3 − 4x = 0
⇒ x (x2 − 4) = 0

which can be easily solved. x = 0 and x = ±2. The first solution is not acceptable according
to our assumption. as x approaches only a constant O (1). But the other two solutions are
consistent.

c P Sunthar, March 8, 2010



66 Chapter 2. Differential Analysis of Fluid Flow

1
Function
Asymptotic
0.8

0.6
f (ǫ)

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1

ǫ
2
Figure 2.11: Asymptotic behaviour of a function. Here the function is f (ǫ) = 1−e−ǫ and its asymptotic
behaviour is ǫ 2 . Close to zero, the asymptotic closely matches with the function.

2. Let x ∼ O (ǫ), then


Terms: x3 −4x −4ǫ x +2ǫ
O (.) : ǫ3 ǫ ǫ2 ǫ
Neglecting the first and third terms in comparison with the second and forth since (ǫ ≪ 1 ⇒
ǫ 3 ≪ ǫ 2 ) we get
ǫ
−4x + 2ǫ = 0 ⇒ x =
2
This is consistent with our initial assumption. We have therefore found out all the roots of the
equation.
The asymptotic roots of equation are: x = 2, −2, ǫ/2, with ǫ = 0.01. Compare this with the exact roots:
x = 2, −2.0005, 0.0005. The first root and the third roots are exact; the second root is also correct within
the order of the approximation used. When we assumed a root to be O (1), we had ignored terms of
O (ǫ), and 0.0005 = ǫ/2 is O (ǫ).

Lecture 2.16 Singular perturbations are asymptotic problems, where by the very nature of the
assumption of smallness, one of the solutions “lost”
Example 2: Differential Equation Now we will take up the case of a differential equation. We are
not considering a simple equation here but one that is called as a singular perturbation problem. In
this the small parameter multiplies the highest derivative, therefore the as ǫ → 0, if that term has to
be neglected then all the boundary conditions cannot be satisfied.
Consider the differential equation for u(x)
ǫ u′′ + u′ = 1 subject to u(0) = 0, u(1) = 2 (2.80)
′ du
where u denotes dx .
We are seeking solutions of the form u = u(x; ǫ) as ǫ → 0. For functions, u ∼ O (1)
implies at constant x, u = u0 (x) as ǫ → 0, without any leading dependence on ǫ. This means the
function u(x; ǫ) can be written as
u(x; ǫ) = u0 (x) + ǫ u1 (x) + ǫ 2 u2 (x) + . . . (2.81)
Similarly, u ∼ O (ǫ) implies we can expand the function u(x; ǫ) for all x as
u(x; ǫ) = ǫ u1 (x) + ǫ 2 u2 (x) + ǫ 3 u3 (x) + . . .

c P Sunthar, March 8, 2010



Lecture 2.16. Singular Perturbation 67

and
u(x; ǫ)
lim = u1 (x)
ǫ→0 ǫ
With this brief introduction to the asymptotic behaviour of functions, we begin the guess work for
the solution of the differential equation.

1. Guess u ∼ O (1). This implies u′ , u′′ ∼ O (1). This can be easily seen by taking the derivatives of
Equation (2.81). The orders of the terms at constant x in Equation (2.80) read:

Terms: ǫ u′′ +u′ = 1


O (.) : ǫ 1 1

Neglecting terms of O (ǫ) in comparison with 1, we see u′ = 1, or u = x + c. Notice immediately


that the second order differential equation has been reduced to a first order, and there is only
one boundary condition to be satisfied. The two possible solutions are

uouter = x (2.82)
uouter = x + 1 (2.83)
(2.84)

For the second boundary condition to be satisfied, the second order term must be retained in
the equation, and this cannot happen if u ∼ O (1). Let us try other possibilities.

2. Guess u ∼ O (ǫ), this gives u′ , u′′ ∼ O (ǫ). The order of each of the terms give.

Terms: ǫ u′′ +u′ = 1


O (.) : ǫ2 ǫ 1

Neglecting ǫ ≪ 1, this leads to a inconsistent equation 0 = 1. Similar inconsistencies occur for


u ∼ O (ǫ n ), for n > 1.

3. For u ∼ O (1/ǫ), we get u′ = 0, which again leaves out the highest order derivative, and leads to
a solution u ∼ O (1), inconsistent with the initial assumption.

So far we have been seeking solutions in a “parameter separated” form: ǫ and u(x) were occurring as
products in the solution. Another possible guess that is usually made is to assume a form in which x
and ǫ occur as products, and the solution is a function of this product. We seek solution of the form

u(x; ǫ) = u(X; ǫ)

where X = x ǫ n is a new independent variable.

1. We first guess X = x/ǫ. The new variable is X ∼ O (1), this implies the old variable is in the
range {0, ǫ}. The new variable is the “stretched” independent coordinate, as though we are
zooming in a small region ǫ close to zero (Compare Figure 2.12 and Figure 2.13). Next, we
guess u(X; ǫ) ∼ O (1) implying

u(X; ǫ) = U0 (X) + ǫ U1 (X) + ǫ 2 U2 (X) + . . .


 x  x (2.85)
= U0 + ǫU1 + ...
ǫ ǫ
so that  x
u(x; ǫ) x
lim = U0 at constant
ǫ→0 ǫ ǫ ǫ

c P Sunthar, March 8, 2010



68 Chapter 2. Differential Analysis of Fluid Flow

Note that x/ǫ is kept constant, whereas earlier it was only x that was kept constant as ǫ → 0.
With this we have for the higher order derivatives of u from Equation (2.85) as
!
′ 1 ′  x  
′ x 1
u (x) = U0 + U1 + H.O.T ⇒∼ O
ǫ ǫ ǫ ǫ
    !
′′ 1 ′′ x 1 ′′ x 1
u (x) = 2 U0 + U1 + H.O.T ⇒∼ O 2
ǫ ǫ ǫ ǫ ǫ
Carrying out a balance of terms of equal magnitude in Equation (2.80)

Terms: ǫ u′′ +u′ = 1


1 1
O (.) : ǫ ǫ 1
we get
u′′ + u′ = 0
This solution to this equation is sought in the stretched independent coordinate (because the
equation is valid only under that assumption)
x
X≡
ǫ
which gives a solution
uinner = −A e−X + C
This solution is consistent with our initial assumption u(X; ǫ) ∼ O (1). This solution is called
the “inner solution” because the new the variable x = ǫ X goes from 0 to ǫ. Satisfying the inner
boundary condition u(0) = 0 we get

uinner = A (1 − e−X ) = A (1 − e−x/ǫ ) (2.86)

Since the boundary condition at zero is satisfied by this solution, we take the other bound-
ary condition to be satisfied by the solution found earlier in Equation (2.83) (discarding Equa-
tion (2.82)). Note that the constant A for in inner solution cannot be determined by the outer
boundary condition u(1) = 2, as the solution is itself valid only in a small region close to the
boundary, x ∼ O (ǫ).
We have thus got two solutions: outer in Equation (2.83) and inner in Equation (2.86), with one
arbitrary constant A. Note that
• the outer solution is valid for ǫ → 0 with fixed x, and
• the inner solution is valid for ǫ → 0 with fixed x/ǫ.
One way to determine this constant is to do a matching at a boundary. Here, we simply set the
far field value of the inner solution to the zero field value of the outer solution. The outer solution
approaches 1 as x → 0. For the inner solution the far-field X → ∞ gives uinner (X → ∞) = A. Matching
these two we get A = 1 and the matched solutions are

uinner = 1 − e−x/ǫ
uouter = x + 1

This matching of two asymptotic solutions are called as, you guessed right, Matched Asymptotics.
Figure 2.12 and Figure 2.13 show the extent to which the matched solutions agree with the exact
solution. The exact solution to this case is analytically solvable and is given by

1 − e−x/ǫ
uexact = x +
1 − e−1/ǫ

c P Sunthar, March 8, 2010



Lecture 2.16. Singular Perturbation 69

1.5

1
u

0.5
Outer
inner A=0.5
inner A=1.0
Exact
0
0 0.2 0.4 0.6 0.8 1

x
Figure 2.12: The inner, outer, and the exact solution to Example 2 differential equation. Here ǫ = 0.01.

1.2

0.8

0.6
u

0.4

0.2 Outer
inner A=0.5
inner A=1.0
Exact
0
0 0.02 0.04 0.06 0.08 0.1

x
Figure 2.13: The inner, outer, and the exact solution to Example 2 shown zoomed in a region close to
the boundary x = 0. Here ǫ = 0.01

c P Sunthar, March 8, 2010



70 Chapter 2. Differential Analysis of Fluid Flow

U
y

Figure 2.14: A general boundary layer flow past a solid object.

Lecture 2.17 Differential Analysis of Boundary Layer Problem results in a Singular Perturbation
problem.
Concepts You Must Know
1. Differential Analysis of Boundary layer gives the scaling√of the boundary layer thickness, with-
out actually fully solving the differential equation: δ ∼ x ν/U.
2. Inside the boundary layer, the pressure variation along the boundary layer (along x) is much
larger than that across the boundary layer (along y).
∂p ∂p
3. ∂x inside the boundary layer equals ∂x outside.
4. The pressure outside the boundary layer is obtained by the solution of the inviscid (Euler)
equations.
We now apply the governing equations of fluid flow to the boundary layer problem. We will
consider a general two-dimensional boundary layer flow, in which the “external” flow field need not
be uniform (This can happen for example for a flow past a sphere, converging/diverging duct, airfoil
etc). In the Figure 2.14 L and D represent the two length scales in the x and y directions respectively.
L = D corresponds to a sphere, L > D corresponds to an air-foil, and L ≫ D corresponds to a flat
plate,
The governing equations at steady state can be written down
∂ x u + ∂y v = 0 (2.87)
1  
u ∂ x u + v ∂y u = − ∂ x p + ν ∂2x u + ∂2y u (2.88)
ρ
1  
u ∂ x v + v ∂y v = − ∂y p + ν ∂2x v + ∂2y v (2.89)
ρ
There are two boundary conditions to be satisfied for each velocity field
u(surface) = 0 inner boundary condition (2.90)
u(far field) = {U, 0} outer boundary condition (2.91)

c P Sunthar, March 8, 2010



Lecture 2.17. Boundary Layer Equations 71

In general the velocity is two-dimensional, and varies in both x and y. Before any further analysis
on the guess work, we non-dimensionalise the equations. In contrast to the earlier example problems,
we are now dealing with a dimensional equation, apart from the usual orders of magnitude, we also
have to deal with the units. We choose certain reference quantities for each of the variables (indepen-
dent and dependent) and re-write the equations in a dimensionless form. The reference quantities
can be chosen arbitrarily, but choosing them with some physical insight simplifies the analysis, and
can give a meaningful physical result.

Choice of Reference quantities The choice for the reference variables (constants) can be two types:
one simply a dimensionally consistent quantity, and another a quantity that represents a characteris-
tic value for the variable. Consider an absurd example. We wish to find out a characteristic velocity
for a fluid flow past a sphere; we can take several reference constants as velocities: velocity of earth’s
surface, speed of sound, velocity of light. Though all these choices will give a dimensionless quantity
each, it will not provide any useful simplification of the important terms in the equation. Recall, in
asymptotic analysis, we try to simplify the balance between several terms to that between a very few
(one or two terms). A better choice would be the far-field velocity. This is the characteristic velocity
of the fluid. We say
u(x) ∼ O (U∞ ) (2.92)

implying, u at a typical location x goes to zero as does U∞ . For this to hold good u must be linearly
and uniquely dependent on U∞ . This also means that the variable u varies between 0 and U∞ ; and
for most of the “region” u is a significant fraction (0.1 or more) of U∞ . Now, we do not know all these
before hand; but by experience with similar problems, we hypothesise and then validate it later.
What is this “region” we are talking about. This also answers the question: What are the reference
quantities for the independent variables x and y? Over what region in space does u change signifi-
cantly (from zero to U∞ ). Using these rough thumb rules, we guess the orders of magnitude of the
variables.

x ∼ O (L) y ∼ O (D)
u ∼ O (U) v ∼ O (U)
p ∼ O (P)

Here P is some pressure of the incoming flow. We have written a relation for the independent vari-
ables to denote that the changes in the dependent variables occur significantly in the lengths scales
of O (L) and O (D), as shown. We can use a physical method to obtain an estimate of the orders of
magnitude of the derivatives. Physically, a derivative represents a change: we ask the question, over
what region does u change significantly, and use this to estimate the derivative:

∂u ∆u U − 0 U 
≈ = ⇒∼ O
∂x ∆x L−0 L
∂v ∆u U 
≈ ∼O
∂y ∆y D

A good starting example is the parabolic flow in a pipe, where u(r) ∼ O (U), r ∼ O (R), and the
derivative ∂r u ∼ O (U/R).
We now use these to estimate the orders of the terms in the x-momentum equation

Terms: u ∂x u +v ∂y u = −∂ x p +ν ∂2x u +∂2y u

U2 U2 P νU νU
O (.) L D ρL L2 D2
P 1 1
simplified O (.) 1 1 ρ U2 Re Re

c P Sunthar, March 8, 2010



72 Chapter 2. Differential Analysis of Fluid Flow

U2
where for getting the third line, we have assumed L ∼ D, and divided by L . The third line shows
the orders of magnitude for the dimensionless form of the equation:
!
L ∗ P 1 L2
u∗ ∂ x∗ u∗ + v ∂y∗ u∗ = − 2
∂ x∗ p∗ + ∂2x∗ u∗ + 2 ∂2y∗ u∗ (2.93)
D ρU Re D

The superscript ∗ on each of the variables denotes the dimensionless variables eg: u∗ = u/U. 1/Re
and P/ρ U 2 are two dimensionless parameters in the problem. P is so far unspecified,
For very large Reynolds numbers, Re ≫ 1, the viscous terms are negligible. We can write ǫ ≡ 1/Re,
and state ǫ → 0 equivalently. This problem is very similar to the singular perturbation problem,
where the small parameter multiplied the highest order derivative. This means the leading order
solution u/U ∼ O (1) cannot satisfy both the boundary conditions.
For the leading order solution (Re → 0) there is still one parameter left: P/ρU 2 . If the pressure
gradient is important then it must be O (1), to balance with the other terms remaining in the equation.
Most of the problems involving inviscid flow this happens to be the case: The leading order solution
is a balance between
 the inertial (LHS) and pressure (RHS) terms. Therefore the pressure scale is
dictated by P ∼ O ρU 2 (for inviscid flow problems). ρ U 2 is also called as the inviscid pressure scale.
The method to obtain the solution to the variables u, v, and p is through irrotational flow theory, which
will be discussed in later lectures.
We will assume that such a solution has been found out. The solution will be of the form

uouter (x) = U f (x) (2.94)


2
pouter (x) = ρ U g(x) (2.95)

The inner boundary condition can be written as one for the normal component of the velocity and
one for the tangential. We can enforce only one condition. Since it is necessary to enforce that the
fluid does not “penetrate” the solid, the zero normal velocity is a stricter condition. For the tangential
velocity, we simply allow a slip at the boundary.
In order to enforce the no-slip boundary condition at the surface, we need to retain the second
derivative. To do this we assume a small layer, called the boundary layer where the viscous terms
cannot be neglected. In this region we have y ∼ O (δ). and

∂u U 
∼O
∂y δ
2 
∂ u U
∼ O
∂y2 δ2

We estimate terms in the continuity equation, by assuming an unknown scale for the velocity v ∼
O (V)
∂ x u +∂y v = 0

U V
Order Of: L δ =0
The continuity equation holds good always; that is there is no parameter which dictates importance
of terms. For non-zero velocities, all terms are equally important here. This gives the vertical velocity
scale  δ
V∼O U (2.96)
L
Using this with the other estimates into the x-momentum balance:
 
u ∂ x u +v ∂y u = − ρ1 ∂ x p +ν ∂2x u + ∂2y u
  (2.97)
U2 U2 U U
Order Of: L L ? ν L2 + δ2

c P Sunthar, March 8, 2010



Lecture 2.17. Boundary Layer Equations 73

Among the viscous terms in the RHS, the first is much small compared to the second by the virtue of
δ ≪ L. In order that the viscous terms be retained in the equation, they must be of the same order as
of the other terms. That is !
νU U2
∼ O (2.98)
δ2 L
for the viscous and the inertial forces to balance. The pressure variation in the boundary layer is not
known. All we know is that if it is important then it must at least be order of the inertial terms, i.e.,

1 U2
∂x p ∼ (2.99)
ρ L

We will show below that the pressure the variation in the x-direction, inside the boundary layer is
the same order as it is outside the boundary layer. There can be situations where the inviscid (outer)
solution has the gradient in pressure to be zero (as in flat plate). This means the viscous terms should
balance the inertial terms in general. This gives a scaling for the boundary layer thickness: from
Equation (2.98) we have r !
µ  p 
δ∼O L ∼ O L/ ReL (2.100)
LU ρ
This is an important achievement. The scaling of the boundary layer has been obtained without even
solving the equations. This is identical to that we obtained using integral analysis of boundary
layer: irrespective of the assumed velocity profile, the dimensionless boundary layer thickness al-
ways scaled as Re−1/2 (see Equation (2.72)).
We now estimate the terms in the y-momentum equation.
 
u ∂ x v +v ∂y v = − ρ1 ∂y p+ ν ∂2x v + ∂2y v
 
U2 δ U2 δ Uδ Uδ
Order Of: L2 L2
? ν L3
+ δ2 L

Here we treat the scaling of the pressure term in the boundary layer as unknown. The first among
the viscous term is still smaller than the second. All other terms other than the pressure term are of
equal magnitude. This means that if the pressure variation in y-direction is important, then it scales
as
1 U2 δ
∂x p ∼ (2.101)
ρ L L
Comparing Equation (2.101) with Equation (2.99), we see that the pressure variation in the y direction
is insignificant in comparison with the x direction. This means inside the boundary layer the pressure
is almost constant along y.
∂p ∂p ∂p
≪ ; ≈ 0 or p = f (x) (2.102)
∂y ∂x ∂y
This implies the pressure gradient inside the boundary layer is only in the x-direction, and the pres-
sure at edge of the boundary layer is impressed upon on the boundary layer. Recall that the pressure
outside the boundary later was obtained by potential flow solution. It is this same pressure that
applies inside the boundary layer, because there is no variation in the vertical direction, inside the
boundary layer. Recipe: Determine the pressure outside using the inviscid solution, use the same
variation inside the boundary layer. Outside the boundary layer we had the pressure variation to
scale as given by the inviscid flow as in Equation (2.95). We can now use this estimate into equation
Equation (2.97) and rewrite it as

u ∂x u +v ∂y u = − ρ1 ∂ x p +ν ∂2y u
(2.103)
U2 U2 U2 U U2
Order Of: L L L ν δ2
∼ L

c P Sunthar, March 8, 2010



74 Chapter 2. Differential Analysis of Fluid Flow

This implies inside the boundary layer, in general, all the above terms are of equal magnitude, and
the pressure term has to be retained in the x-momentum equation.
The final set of equations of motion for the flow inside the boundary layer can then be written as

∂ x u + ∂y v = 0 (2.104)
1
u ∂ x u + v ∂y u = − ∂ x p + ν ∂2y u (2.105)
ρ
0 = ∂y p (2.106)

c P Sunthar, March 8, 2010



Lecture 2.18. Blasius Solution 75

Lecture 2.18 Blasius solution for a Boundary Layer over a flat plate is a numerical solution to an
assumed similarity solution for the velocity profile.
Concepts You Must Know
1. The similarity variable occurs in flat plate Boundary Layers due to the absence of any geomet-
rical length scale.
2. In problems where there are no external length or time scales, the independent variables are
combined with the other parameters to provide the similarity variable.

The boundary layer equations Equations (2.104) to (2.106) can be applied to a simple flat plate
situation where the outside velocity profile is uniform. This implies there is no pressure variation
outside the boundary layer. From Equation (2.106), this also implies there is no pressure variation in
the x direction inside the boundary layer (as the outside pressure is impressed as is). The simplified
set of equations are therefore

∂ x u + ∂y v = 0 (2.107)
u ∂ x u + v ∂y u = ν ∂2y u (2.108)

The solution to this equation is obtained by assuming a similar variation in velocity—similarity so-
lution. The dimensionless similarity variable is taken to be
r
U y2
η≡ (2.109)
νx
which is formed by combining the independent variables y and x. The motivation for this choice is
as follows.
1. For a flat plate boundary, there is no length scale imposed by the geometry. So by dimensional
analysis the only variable that can occur as an independent variable should be a combination
of the independent parameters in the problem, viz. : U, ν, x, y The dependent variable u has a
velocity scale U. Therefore we can expect
!
u U y2
= f (2.110)
U νx

2. Dimensional analysis does not dictate what power of the combination of these variables should
occur. But in our analysis of the boundary layer, we expected a small region of the order of δ in
the y-direction where gradients are important. This implies y ∼ δ. The variable η defined above
can also be seen as a dimensionless y, and δ is given by Equation (2.100), with L replaced by x
in general:
y y
η≡ = √ (2.111)
δ ν x/U
Two dimensional equations of motion can be solved by the method known as stream functions. We
will learn more of this in inviscid flows. We introduce a function known as the stream function ψ
that jointly represents the variation of both the components of velocities.
∂ψ ∂ψ
u= v=− (2.112)
∂y ∂x
It can be easily verified that this choice automatically satisfies the continuity equation. Only the mo-
mentum equation is to be solved, and that can be reduced to a single ordinary differential equation.
We try a separation of variables
ψ = g(x) f (η) (2.113)

c P Sunthar, March 8, 2010



76 Chapter 2. Differential Analysis of Fluid Flow

where the functional form of g is determined from the requirement that the dimensionless velocity
u/U is a function only of the variable η.
r
u 1 ∂ψ 1 ′ U
= = g(x) f (η) , (2.114)
U U ∂y U νx

which gives g(x) = ν U x, and r
v ∂ψ 1 ν
=− = (η f ′ − f ) (2.115)
U ∂x 2 U x
Substituting these expressions in the boundary layer equations, we get an ordinary differential equa-
tion
f f ′′ + 2 f ′′′ = 0 (2.116)
subject to the boundary conditions

f = f ′ = 0 η = 0 no slip at the wall (2.117)


f ′ = 1 η → ∞ outer boundary (2.118)

The Blasius solution was obtained as an approximate solution to this equation. It can also be solved
numerically.

c P Sunthar, March 8, 2010



Lecture 2.19. Boundary Layer Separation 77

Lecture 2.19 The Boundary Layer detaches from the surface when there is an positive pressure
gradient along the direction of flow.
Concepts You Must Know
1.

Now consider the limit of the x-momentum equation in the boundary layer given by Equa-
tion (2.105), as y → 0 or more precisely y ≪ δ. This is very close to the wall inside the boundary
layer. In this limit the derivatives still scale as

∂u U ∂u U
∼ ; ∼ (2.119)
∂x L ∂y δ

but the u velocity scales as


y
u∼U , (2.120)
δ
where, close to the wall the velocity can be assumed to be linear in the variable y/δ. From the conti-
nuity equation we have v velocity scale as
y
v∼U , (2.121)
L
Substituting this in the momentum balance we obtain

u ∂x u +v ∂y u = − ρ1 ∂ x p +ν ∂2y u
(2.122)
y U2 y U2 U2 U U2
Order Of: δ L δ L L ν δ2
∼ L

Note scale estimates in this equation reduces to Equation (2.103) when y ∼ O (δ). But considering a
limit y ≪ δ, which is very close to the wall, we see that the inertial terms in the LHS become small
comparison to the other terms. Therefore in the limit y ≪ δ, the balance is between the pressure
and viscous terms (the inertial terms drop out). This is equivalent to a case in mechanics of a body
moving with constant velocity when there is a force balance, and no acceleration. Later we will study
this to be a case of creeping flows or low Reynolds number flows.
Consider the simplified equation near the wall

∂p ∂2 u
=µ 2 (2.123)
∂x ∂y

The second derivative of the velocity near the wall depends on the pressure gradient in the x direc-
tion, which in turn is determined by the inviscid solution outside the boundary layer. Consider three
simple cases of the pressure gradient
∂p ∂u
1. ∂x = 0: This implies the velocity gradient ∂y = constant all along the wall.

∂p
2. ∂x < 0: Implies the velocity gradient decreases with y (second derivative of u < 0). Since
the velocity at any point reaches the free-stream value at the edge of the boundary layer, this
implies there is a higher slope of the velocity profile near the wall, which tends to zero at the
edge of the boundary layer. Note that in the figure the slope has to be seen along the y direction
which is the independent variable.
∂p
3. ∂x > 0: Implies the gradient increases with y. But since ultimately the gradient has to go to
zero, this leads to a situation of maximum slope at some intermediate height.

c P Sunthar, March 8, 2010



78 Chapter 2. Differential Analysis of Fluid Flow

∂p
4. ∂x is highly positive: The velocity gradient can become zero at the wall or even negative. This
leads to a reversal of direction of flow close to the wall. This pressure gradient is also known
as an adverse pressure gradient as it leads to adverse effects in the flow.

Under the reversal of flow, the boundary layer is no longer “attached” to the wall, but is “separated”.
Visually this is seen as small viscous region which continues in the same direction as the main flow,
and a separated region in which there is mixing and turbulence.
Boundary layer separation leads to the formation of wakes—regions of swirls and mixing. This
leads to loss of pressure (the reason for loss of pressure will be clear when we study macroscopic
balances and viscous losses) and leads to high drag. An example is the flow past a sphere. The drag
on the sphere is mainly due to the boundary layer separation.
The drag can be reduced by delaying the boundary layer separation. Delaying can be done by:

1. Streamlining the object. Aerodynamic shapes, as they are conventionally known, are shapes
which lead to a well behaved streamline inviscid flow in which the pressure gradient is limited
to small positive quantities to avoid boundary layer separation.

2. Another way is to make the surface rough. Such as in the case of golf balls or cricket balls. A
rough surface leads to a turbulent boundary layer. The velocity profile in turbulent boundary
is steeper than in the laminar case. This implies a stronger positive pressure is required in order
to reverse the flow and separate the boundary. Golf balls therefore have lesser drag compared
to smooth balls, and hence can travel farther.

c P Sunthar, March 8, 2010



Lecture 2.20. One Dimensional Flows 79

2.5 Simple Viscous Flows


Lecture 2.20 In one dimensional flows, the inertial term is absent and the problems are usually
amenable to analytical solutions.
In this lecture we will consider simple flow application of Navier-Stokes equation, on which it is pos-
sible to obtain a closed form analytical solution to the velocity fields. Many problems are discussed
in the introductory texts Wilkes (1999); Fox and McDonald (2001); Denn (1980), we will consider only
a few illustrative applications which have more of chemical engineering relevance.

A general procedure to solve the Navier-Stokes equation The Navier-Stokes Equations Equa-
tion (2.29) is a set of four coupled non-linear second order partial differential equations. Even if
the boundary conditions are known, obtaining a solution is not a straight forward mathematical
problem. For most engineering applications, at least if the problem can be simplified, it is possible
to reduce the complexity in the equations at the very beginning. This is achieved by making some
assumptions, and idealisations. Good approximations come with experience and practice. But there
is a general sequence of steps that may be followed to simplify the problem.

1. Guess which is the dominant velocity direction, if there could be one.

2. Guess the independent variables that this depends on.

3. Validate the guess by substituting it in the continuity equation. Remember there are no physical
parameters appearing in the continuity equation which could alter the balance. It could be
possible to obtain additional information about the velocity field. This approach was used in
the differential analysis of the boundary layer.

4. Substitute the guesses in the momentum equation and simplify the differentials. For many
problems, the concentrating on pressure terms usually leads to further simplifications.

We will illustrate this procedure in a few problems.

2.5.1 Poiseuille and Couette Flows


Poiseuille flow is a pressure driven motion of a fluid. Couette flows are those in which the pressure
gradients are absent, and the flow is driven by motion of the boundary; it is also called as a drag flow.
You have studied Poiseuille flows in a pipe and steady and unsteady Couette flows, in your earlier
course Transport Phenomena. Here we will consider pressure driven flow in a rectangular channel.
Consider a rectangular channel of length L (along the x direction), and cross-section of W × H
(along z and y directions respectively), through which there is a flow driven by a pressure gradient
acting in the x-direction. We will make certain assumptions and simplify the problem. If we obtain a
solution and it is consistent with the assumptions, then we take that we have found out one solution
to the non-linear differential equations. But we must admit there there could be other possibilities as
well, which can be verified only by experiments, or the complete solution to the equation.
Assumption-1: Steady flow. This immediately drops all the time derivatives, and we have al-
ready restricted our solution space. Assumption-2: The aspect ratio of the channel is very large. This
implies W ≫ H. This seems a restrictive assumption, but again it is only an assumption. We can
take this to an extreme, which we call idealisation, and take W/H → ∞. This implies the wall does
not influence the velocity, or that there is “no wall” in the z direction. This means the velocities do
not change in the z-direction. Assumption-3 The flow is fully-developed. This means that regions of
influence of the entrance and exit are negligible in comparison with the length L. This is also to say
that the velocities are not a function of x. This reduces the problem statement in the velocities to the
following
ui = fi (y) (2.124)

c P Sunthar, March 8, 2010



80 Chapter 2. Differential Analysis of Fluid Flow

Substituting in the continuity equation we get uy = v = constant. From the no-slip condition, we
have that the velocity has to vanish at the boundaries. This implies v = 0 through out. The NSE in
Cartesian coordinates reduce to
∂p d2 u
0=− +µ 2 (2.125)
∂x dy
∂p
0=− (2.126)
∂y
∂p d2 w
0=− +µ 2 (2.127)
∂z dy
Note we have used complete derivative for the velocities, and partial for the pressure. Inspecting the
pressure terms, we find that the pressure is only a function of (x, z). In Equation (2.125), ∂p
∂x is not a
function of y where as the viscous term is, and the viscous term is not a function of x whereas the
pressure term is. This implies each of them must be equal to the same constant. A similar argument
applies for the z-momentum equation. This gives us the solutions (along with the no-slip conditions)
! !2 
H 2 ∆p  2y 
u= − 1 −  (2.128)
8µ L  H 
! !2 
H 2 ∆p  2y 
w= − 1 −  (2.129)
8µ W  H 

The average velocity in the z-direction is


ZH/2 !
1 H2 ∆p
hwi ≡ dy w = − (2.130)
H 12 µ W
−H/2

Since there is no net-flow in the z-direction because of the presence of the walls, this implies the
pressure gradient is zero, or from Equation (2.129), w = 0. Equation (2.128), v = w = 0 is the solution
we obtain. This is consistent with our assumptions.

2.5.2 Torsional flow


Viscometers are used to determine the viscosity of fluids. They are usually employed to measure
viscosities of liquids. The commercial viscometers can typically measure only high viscosity liquids,
about 100 times that of water. A cylinder is rotated in a liquid and the torque required to generate
a constant rpm is measured. This is used to compute the viscosity of the fluid. Rheometers are
more sophisticated and have many more controls, and can measure many more properties of flow
for complex fluids that exhibit both viscous and elastic behaviours.
In reality the velocities around the shaft can be complex, but we make simplifications as before.
We assume that the length of the shaft L ≫ R, the radius of the shaft, thereby neglecting the end
effects between the bottom of the vessel and the shaft. Further we assume that there is no axial or
radial motion, but motion only in the θ-direction. From the continuity equation we have that vθ = f (r)
(we have neglected the variation in the z direction due to the length assumption above). The NSE
reduce to
vtheta ∂p
−ρ =− (2.131)
r ∂r " #
d 1 d(r vθ )
0=µ (2.132)
dr r dr
∂p
0=− (2.133)
∂z

c P Sunthar, March 8, 2010



Lecture 2.20. One Dimensional Flows 81

From Equation (2.132) we get


C2
vθ = C1 r + (2.134)
r
At the outer radius of the vessel the velocity is zero, so for any large r the velocity is finite, therefore
the term proportional to r term must vanish from the equation. No-slip at the shaft surface (vθ = R ω
at r = R,) gives
ω R2
vθ = (2.135)
r
The torque is proportional to the surface force exerted by the fluid on the shaft. We require to estimate
τrθ , which can be obtained from the tables for the velocity gradient in cylindrical coordinates

τ = µ(∇ v + ∇ vT )
!
∂vθ 1 ∂vr vθ (2.136)
τrθ = µ + −
∂r r ∂θ r
The stress at r = R is therefore
τrθ (r = R) = −2 µ ω (2.137)
The differential force on an element swept by dθ across the length is τrθ L R dθ and the corresponding
torque is τrθ L R2 dθ. Integrating over the angle, we get an expression for the total torque

G = −4 π L R2 µ ω

2.5.3 Free surface flows


So far we have considered a flow where there is only a single fluid. In the presence of more than
one fluid, proper interface condition for the force needs to be specified. The simplest condition when
the interface is stationary or moving with a constant velocity. In this case the forces (surface forces)
acting on either side of the interface must balance each other.

σ1 = σ2 or σ(1) (2)
i j = σi j (2.138)

We can also equate the normal and tangential component of the forces on either side. For the simplest
case of a stationary fluids with a curved interface this reduces to the normal stress balance:

p(1) = p(2) + (2.139)
R
at the interface, where γ is the surface tension and R is the radius of curvature. When two fluids flow
past each other, with a flat interface, the tangential force arises due to the viscous stress. A stress
balance gives
∂u1 ∂u2
τ(1) (2)
yx = τyx or µ1 = µ2 (2.140)
∂y ∂y
In the case of air-water interface, we rewrite this as
∂uw µa ∂ua
= (2.141)
∂y µw ∂y
In most cases the air region is unbounded (like in the case of a jet of liquid in air) therefore the velocity
gradient is also negligibly small. Even if the region of air is bounded by walls present in comparable
length scales, the viscosity of water is about 50 times that of air, making the RHS negligibly small for
the velocity gradient in water. This leads to a general condition that there is no shear stress acting
on the surface, or it is a stress-free boundary; the condition is also known as an inviscid boundary
condition. This is a very useful simplification in treating flows of air water interface. The stress free
condition leads to the conclusion that in the cases of jets the velocity profile in the liquid is flat.

c P Sunthar, March 8, 2010



82 Chapter 2. Differential Analysis of Fluid Flow

Die Wire Coating


Radius: Rd Rw Rc
Figure 2.15: Schematic diagram of wire coating.

2.5.4 Wire Coating


Wire coating is the process of coating a polymeric material on, usually, metallic wires. The process
is done in the liquid state of the polymer, and then it is let to dry. A wire is pulled through a die
connected to a reservoir of molten polymer, leaving a coat of constant thickness. The problem is
to determine the coating thickness as a function of rate of pulling and the other parameters of the
system. Also of importance is the force required to pull the wire, at a given velocity.
Consider the schematic shown in Figure 2.15. Inside the die the liquid is flowing in an annular
space created by the void between the wire and the die. At distances far away from the die exit, the
coating is uniform, implying the interface is flat and the free surface condition (stress-free) yields a
flat velocity profile. Performing a integral mass balance across a large control volume, that takes the
shape of the liquid we get at steady state

ZRd
− dr 2 π r ρ vz + Vw π (R2c − R2w ) = 0 (2.142)
Rw

where Vw is the velocity the wire is pulled with. In order to find Rc we need to know the velocity
profile vz (r) inside the die annulus. As before, we consider a fully developed velocity profile, as this
makes the equations simpler (Note that when we positioned the CV boundary in the die, we did
not require it to have a fully developed velocity profile, i.e., the Equation (2.142) is valid even if the
velocity is not fully developed, but the expression is easier to evaluate for fully developed flow). The
NSE reduce to p = constant and " !#
1 ∂ ∂vz
0=µ r (2.143)
r ∂r ∂r
With the no-slip boundary conditions this gives
ln r/Rd
vz (r) = Vw (2.144)
ln Rw /Rd
Substituting in Equation (2.142) we get
 2 1/2
 Rd − R2w 
Rc =   (2.145)
2 ln Rd /Rw

and the force required to pull the die is obtained by integrating the surface force on the wire over the
length of the die
2 π µ Vw L
Fw = (2.146)
ln Rw /Rd
assuming the same the fully developed velocity profile to be dominating the entire length. This also
ignores any radial variations in the velocity from the die exit to the regime of constant wire thickness.

c P Sunthar, March 8, 2010



Lecture 2.20. One Dimensional Flows 83

2.5.5 Tube flow of Power law fluid


Most of the fluid we have considered are Newtonian, that is the shear stress is related by a linear
relationship to the local velocity gradient as given in Equation (2.30). However, there are several
fluids which do not exhibit this behaviour. This is especially true of polymeric fluids, and other
fluids that have a constituent substance that cannot be considered to be simple spherical molecules
of same size. These fluids exhibit behaviours that are vastly different and rich in variety. Any fluid
that does not show the Newtonian behaviour, in the velocity gradients of practical interest, is called a
non-Newtonian fluid (Strictly speaking, there could be very high gradients of velocity where even the
so called Newtonian fluids, such as air and water, exhibit non-linear behaviour; but these are very
rare for practical considerations.)
The local velocity gradient (also called as the local shear rate) can be vastly different for various
processes. Below is a compilation of some of the processes.
Phenomenon Shear rate (s−1 ) Application
−6
Sedimentation of powders 10 Medicines, paints
Extruders 10 Polymers
Mixing 102
Rubbing 103 Application of creams
High speed Coating 105 Printing inks
Lubrication 106 Engines
For non-Newtonian fluids, the viscosity is not a constant. It could be dependent on the local
velocity gradient (or shear rate). It could be time-varying as well. It is convenient to define a viscosity
function
τ
η≡ (2.147)
γ̇
where τ is the local shear stress, in a simple Couette geometry, and γ̇ is the local shear rate (or velocity
gradient). Some of the common non-Newtonian behaviours are
• Shear-thinning materials. The shear stress is no longer linear with the shear rate, but increases
sub-linearly. Or in other words, the viscosity (which is the local slope in the stress vs rate curve)
decreases as the shear rate is increased. Most polymeric substances exhibit this behaviour. The
behaviour is also called as pseudo-plastic.
• Thickening. Slurries and suspensions of particles in a fluid (paint for example) show a increase
in the viscosity with shear rate, and is called as shear-thickening or dialatant materials.
• Yeild Stress materials: Some materials deform elastically upto a certain stress and begin to
flow only beyond a stress known as yeild stress. These materials may exhibit a Newtonian or
non-Newtonian behaviour beyond the yield point. They are also known as Bingham Plastic
• The three types discussed here have a steady viscosity function, so long as the applied shear
rate is kept constant. However some materials show a time-dependent (or history dependent)
viscosity. Thixotropic fluids show a viscosity thinning with time at a constant shear rate, Rheopec-
tic fluids show a viscosity thickening with time.
With that brief introduction to non-Newtonian behaviour, we take up a simple problem of a
Poiseuille flow of a power-law fluid. “Power law” fluids are a class of shear thinning materials that
show a power-law dependence of the viscosity function with the shear rate.

η = K (γ̇)n−1 (2.148)

where K is a empirical constant called as the consistency factor. For a Couette flow the shear stress is
given by
τ xy = η γ̇ = K γ̇n (2.149)

c P Sunthar, March 8, 2010



84 Chapter 2. Differential Analysis of Fluid Flow

A more general expression for the shear rate tensor is given by


 
τ = η(II) ∇v + ∇vT (2.150)
(n−1)/2
η(II) = K |II| (2.151)
 
where II is called as the second invariant of the velocity gradient tensor defined as, II = 21 (tr2D)2 − tr(2D)2 ,
D = 12 (∇v + ∇vT ) is the rate of deformation tensor, and tr denotes the trace of a matrix.
As with the Poiseuille flow we will assume the following (fully-developed, axis-symmetric, and
one-dimensional)
vz = vz (r), vr = vθ = 0 (2.152)

This satisfies the continuity equation. To solve for the velocities, we have to go back to the general
momentum balance, not that for Newtonian fluids. We look up the cylindrical coordinate equations

∂p
0=−
∂r
∂p
0=−
∂θ
∂p 1 d
0=− + (rτrz )
∂z r dr

Solving for τrz gives

∆P
τrz = r (2.153)
2L

Note that even for non-Newtonian fluids the stress is still a linear function of the distance from the
walls. It is only the velocity profiles that will be different. Inserting the expression for the shear stress
n−1
dvz dvz
τrz = K (2.154)
dr dr

and solving for the velocity and using the usual boundary conditions we get
" n+1 !#1/n "  r (n+1)/n #
n R ∆P
vz = − 1− (2.155)
n + 1 2K 2K L R

In terms of the mean velocity hvz i we get, after simplifications,


"  r (n+1)/n #
3n + 1
vz = hvz i 1 − (2.156)
n+1 R

For shear thinning fluids, n < 1 (n is usually in the range 0.3 < n < 1), the profile is flatter towards the
centre in comparison with a parabolic velocity profile for Newtonian fluids.

Physical interpretation of the velocity profile The stress distribution is identical in the cases of
Newtonian and non-Newtonian fluids. For a given pressure drop, the non-Newtonian fluid experi-
ences an lower viscosity, therefore we can expect a higher flow rate, implying the average velocity
hvz i would be higher. This can be seen from Figure 2.16, where the area under the curve for the
non-Newtonian fluid is higher than that for the Newtonian fluid.

c P Sunthar, March 8, 2010



Lecture 2.20. One Dimensional Flows 85

0.8

0.6
vz

0.4

0.2

n = 1, Newtonian
n = 0.3
0
0 0.2 0.4 0.6 0.8 1

z
Figure 2.16: Fully developed velocity profiles for a power law fluid with n = 0.3 compared to that of
a Newtonian fluid.

2.6 Creeping Flows


2.6.1 Flow past a sphere
Concepts You Must Know
1. Simplifications of NSE for low Reynolds Number flows

2. Creeping flow equations are linear differential equations: Unique solution exists

3. Uniform flow past a sphere is easily solved in spherical coordinates

4. The net drag force on the sphere is due to “form friction” (normal stresses) and “skin friction”
(tangential stresses) on its surface.

5. The total drag may also be obtained equating the energy supplied to move the sphere to the
total viscous dissipation in the fluid.

c P Sunthar, March 8, 2010



86 Chapter 2. Differential Analysis of Fluid Flow

2.7 Inviscid flows


Lecture 2.21 Euler’s Equation in Streamline Coordinates provides a method to qualitatively anal-
yse the pressure distribution in an inviscid flow.
Concepts You Must Know
1. Streamline Euler Equations in two dimensions can be derived by assuming local cylindrical
coordinates.
2. The pressure decreases along a streamline when the fluid’s velocity increases along it; or the
fluid accelerates in a direction of decreasing pressure.
3. The pressure is constant across parallel streamlines.
4. For curved streamlines, the pressure increases going outward from the local centre of the radius
of curvature.

Inviscid flows form a class of problems which reduce to linear differential equations of fluid flow.
Creeping flows is the other class of problems in which the equations are linear. This linearity provides
a basis to obtain an unique solution to the flow equations. Inviscid flows and Creeping flows form
two extremes of the Reynolds number parameter, the only parameter appearing in the Navier Stokes
Equation for steady flows. As the name implies, in inviscid flows, the viscosity is not of consequence
(implying it is an ideal fluid) and in creeping flows the viscous term is important but the inertial
terms are negligible.
In dimensionless variables, the Navier-Stokes Equation can be written from Equation (2.29) and
Equation (2.93) in bold face notation as
Du 1  
= −∇p + ∇ · ∇u + ∇uT (2.157)
Dt Re
(Here we have not shown the gravity term explicitly, as it can be absorbed into the pressure term;
see modified pressure in Bird et al. (2007)) In the limit of Re ≫ 1, the viscous terms can be neglected
in comparison with the inertial terms (see the development given in boundary layers, Lecture 2.17).
Though the LHS of the equation which contains the term u j ∂ j ui appears to be non-linear, this set of
equations along with the continuity equation can be reduced to a linear equation in another variable
(potential function) as will be shown later. The resultant equation by ignoring the viscous terms is
called as the Euler’s equations, and which formed the foundations of most of the early differential
analysis of fluids. Since the viscosity is neglected, the fluids are also called as ideal fluids.
We first discern some important qualitative aspects of this equation in its dimensional form
Du
ρ = −∇p − ρ g (2.158)
Dt
This equation is called as the Euler’s equation of fluid flow. Here the LHS represents the acceleration
of the local fluid element, and the RHS is the force acting on it. In this case there is only pressure
force and gravity force because the fluid is ideal.
An important illustration of the features of the flow is two dimensional flows. Here we seek
to understand the pressure distribution by analysing the velocity streamlines. Euler’s equation can
be derived in a streamline coordinate (direction along the streamline and the direction of local the
normal), either from first principles (Fox and McDonald, 2001, see), or by using a local cylindrical
coordinate. We will adopt the later method here, which is simpler and straightforward. Consider
a curved streamline, and place a local cylindrical coordinate such that the origin of the coordinate
coincides with the centre of the local radius of curvature. Then the unit vector along the streamline
is
e s = eθ

c P Sunthar, March 8, 2010



Lecture 2.21. Euler’s Equation along Streamlines 87

and the vector perpendicular to the streamline is

en = er

Euler’s equation in cylindrical coordinates gives

D vr −v2 ∂p
ρ =ρ θ =− (2.159)
Dt r ∂r
D vθ vθ ∂vθ 1 ∂p
ρ =ρ =− (2.160)
Dt r ∂θ r ∂θ
where we have made use of the following conditions.

1. Steady in time, ∂t = 0.

2. vr = 0 since we are considering streamline coordinates, and by definition there is no flow across
the streamline.

3. vz = 0 since we are considering a two dimensional flow.


4. No external body forces.

By using the following transformation of variables to the streamline coordinates with the local veloc-
ity represented as V:
vθ = V; r dθ = ds; dr = dn (2.161)
we get,

V 2 ∂p
ρ= (2.162)
r ∂n
∂V ∂p
ρV =− (2.163)
∂s ∂s
This has an important application in inferring the nature of stream line given pressure and vice-
versa. Equation (2.162) implies that the pressure always increases outward from the centre of the
local radius of curvature. This also implies that when the streamlines are parallel, there is no pressure
variation across the stream lines. From Equation (2.163) we can infer that pressure always decreases
if the velocity increases along a streamline; or the positive acceleration of the fluid happens in a
direction of negative pressure gradient.
As an example consider a stagnation point flow of an inviscid fluid flowing from large y and out-
wards to large |x|. At large y, the streamlines are parallel, implying that there is no pressure variation
in x. Closer to the stagnation point, the curves (streamlines) are bent away from the stagnation point,
which means that the pressure is the highest at the stagnation point. Along the y axis the velocity
goes from a large finite value to zero at the stagnation point, also implying the pressure increases as
the fluid particle is decelerated to zero velocity.
Another example is to calculate the approximate flow rate of air in a bend. Assume the flow is
inviscid, and relate the pressure difference across the channel width to the local velocity through
Equation (2.162). Although there could be viscous effects such as boundary layers, and its sepa-
ration, this estimate of the velocity is a good approximation, since the flow is inviscid. Verify the
inviscid approximation by calculating the Re for the flow. In the case of the problem given in Fox
and McDonald (2001), Re ≈ 0.1 × 30/2E−5 ≈ 1E5.

c P Sunthar, March 8, 2010



88 Chapter 2. Differential Analysis of Fluid Flow

Lecture 2.22 Bernoulii equation is equivalent to Eulers equation along a streamline


Concepts You Must Know
1. Bernoulii’s equation is to be applied only along a streamline, not any arbitrary points in the
flow.

2. Bernoulii’s equation is applicable only for inviscid flows, or flows in which the viscous terms
are negligible.

Including the gravitational acceleration in Eulers equation Equation (2.163) we get

∂V ∂p
ρV =− + ρ gs (2.164)
∂s ∂s
where g s is the component of the gravitational acceleration along the streamline, which is given by

∂z
g s = −g sin θ = −g (2.165)
∂s
where z is the elevation of the streamline from an arbitrary reference. Inserting this in Equation (2.164),
we get all the derivatives along the streamline

∂V ∂p ∂z
ρV + + ρg = 0 (2.166)
∂s ∂s ∂s
or,
!
∂ V2 p
+ + gz = 0 (2.167)
∂s 2 ρ
2 p
Integrating the expression we get the function Φ = V2 + ρ + g z is constant along a streamline, which
is the steady state expression of Bernoulii’s theorem.

Lecture 2.23 Different pressures terminologies are employed depending on the type of measure-
ment.
Concepts You Must Know
1. Static pressure is same as the thermodynamic pressure p, that appears in the derivation of NSE.

2. Stagnation pressure is the pressure developed when a moving fluid particle is brought to rest
by an inviscid process.

3. Dynamic pressure is the stagnation pressure that is in excess of the static pressure.

4. Side or wall taps, with a hole facing perpendicular to streamlines, measures static pressure.
5. A Pitot (/pea toe/) static measures stagnation pressure, and has a hole facing upstream.

6. A Pitot static tube measures both the static and stagnation pressures simultaneously.

7. The flow velocity can be obtained from the difference of the stagnation and static pressures.

Static pressure probes can be placed either on the wall of the pipe with the hole perpendicular to
the flow, or through a probe that is inserted in the flow, but with the walls still perpendicular to the
flow streamlines.

c P Sunthar, March 8, 2010



Lecture 2.24. Rotation of a Fluid Element 89

Lecture 2.24 Vorticity is related to the local rotation of a fluid element is equal to twice the angu-
lar velocity.
Consider a square element with faces a and b each of which rotates independently. Then the angular
velocity of rotation of face a is given by
∆α ∆η/∆x
ωa = lim = lim
∆t→0 ∆t ∆t→0 ∆t
Realise that lim∆t→0 ∆η/∆t = Vy = v(x + ∆x), the local velocity in the vertical direction at x + ∆x. Let the
local vertical velocity at the origin be v0 then a Taylor expansion yields

∂v
v(x + ∆x) = v0 + ∆x
∂x
Assuming, without loss of generality, that the translational velocity v0 of the origin is zero (or equiv-
alently we are moving with the same velocity) we get for the angular velocity
∆η/∆t v(x + ∆x)
ωa = lim =
∆t→0 ∆x ∆x
∂v
=
∂x
Similarly an expression for the angular velocity of the face b can be obtained, this time only with a
negative sign for the direction of u relative to ξ
∆ξ/∆t u(y + ∆y)
ωb = lim =−
∆t→0 ∆y ∆y
∂u
=−
∂y
The average rotation of the fluid element with the faces a and b is
1
ω= (ωa + ωb )
2
The rotation direction is represented by the direction of the normal to the plane which rotates. Con-
ventionally the right hand screw rule is followed, with the thumb pointing out of the plane towards
you, and the other fingers swept from x to y axis. Therefore the above equation corresponds to the
component of the angular velocity in the z direction.
!
∂v ∂u
ωz = − (2.168)
∂x ∂y

Similarly we can derive components in the other two directions as well, by a cyclic rule. It follows
that the components represent the curl of the velocity field
1 1
ω= ∇×u≡ ζ (2.169)
2 2
where ζ is defined and called as the vorticity vector, and whose magnitude is equal to twice the
local angular velocity of the fluid element. The word has origins in the word vortex, in which fluid
particles move along a circular path. A clear distinction between a vortex and vorticity has to be made
here. Whereas vortex is with respect to the bulk the fluid motion with respect to a fixed coordinate
axis, vorticity is the circular motion of the fluid element with respect to a coordinate axis translating
along its centre of mass. Vortices (plural for a vortex) can have either zero or non-zero vorticity, that
is although the fluid motion is in circular paths, the fluid elements may or may not rotate. This is

c P Sunthar, March 8, 2010



90 Chapter 2. Differential Analysis of Fluid Flow

illustrated by placing a small rotating fan or a cork that moves in circular paths. The motion of the
centre of mass of the fan represents a vortex, but the rotation of the fan itself represents vorticity.
The flow pattern generated by stirring a liquid using rotating paddle generates circular motion.
Inside the area swept by the paddle, the motion is a solid body like rotation with vθ = r ω, and is
called as a forced vortex. Here the vorticity is given by
" #
1 ∂r vθ ∂vr
ζz = − = 2ω
r ∂r ∂θ

Outside the swept area the velocity is known to be vθ = c/r, where c is a constant. This circular
motion is called as free vortex, and the vorticity is ζz = 0. Therefore, for a circular motion it is possible
to have no rotation in the fluid elements. Flows with zero vorticity are known as irrotational flows.
They have a special significance in inviscid flows.

Lecture 2.25 Incompressible irrotational flows are inviscid, but vice-versa is not always true
Concepts You Must Know
1. Solution to inviscid equations: Obtain u(x) that satisfies continuity and irrotationality. Integrate
Euler equations to obtain pressure p(x).
2. Inviscid flows can be rotational or irrotational.
3. Irrotational flows solution can be obtained by potential function or stream function (in two-
dimensions and axis symmetric case).
4. Methods of complex variables are used to obtain potential flow solutions.

Here we will seek solution to the inviscid equations. A solution has to satisfy the continuity
equation (we will restrict to incompressible flows)

∇·u=0

and Euler’s equation given earlier in Equation (2.158). There are four equations and four unknowns.
The general procedure to solve inviscid equations is to find a solution to the velocity field u(x) that
satisfies the continuity equation. This is then substituted in the Euler equations and integrated to
obtain the pressure field p(x).
The equations of motion for an incompressible fluid can be written in terms of the velocity vector,
or equivalently in term of the vorticity vector ζ.

∂u 1
+ u · ∇u = − ∇ · p + g + ν∇2 u (2.170)
∂t ρ
∂u 1 1
+ ∇(u · u) − u × ζ = − ∇ · p + g − ν∇ × ζ (2.171)
∂t 2 ρ
We have used the following identities (prove them using index notation)
1
u×ζ = ∇(u · u) − u · ∇u (2.172)
2
∇ × ζ = ∇(∇ · u) − ∇2 u = −∇2 u (for incompressible flows) (2.173)

We will now provide some general relationships between an incompressible flow being potential, ir-
rotational or inviscid.
1. An irrotational flow (ζ = 0) implies that the flow is inviscid (∇2 u = 0). This can be seen from
Equation (2.173).

c P Sunthar, March 8, 2010



Lecture 2.25. Irrotational flows 91

2. The converse is not true. There can be inviscid solutions for which ζ , 0. An example is the
case of solid body rotation of a fluid, in which the viscous stresses (σ) is identically zero. More
generally, from Equation (2.171) we can see that in the absence of viscous forces a solution may
be obtained in which the vorticity is non-zero. For example for steady inviscid flows in the
absence of gravity we get from Equation (2.171)
1 1
u×ζ = ∇ · p + ∇(u · u)
ρ 2
which has to be solved for given boundary conditions. There is nothing known a priori that will
set the LHS to zero. Therefore a non-zero value for ζ is possible for inviscid flows.
3. Irrotationality and the continuity equation form the key conditions to determine inviscid flow
solutions.
4. Potential flows are a special class of flows in which the velocity can be expressed as a gradient
of a potential (φ)
u = ∇φ (2.174)
Here, as the equation implies, φ is a scalar field.
5. Potential flows are Irrotational, and vice versa. By definitions Equation (2.169) and Equa-
tion (2.174) we have
ζ = ∇ × u = ∇ × ∇φ
We can use index notation to prove
∇ × ∇φ = ǫi jk ∂ j ∂k φ
= ǫi jk ∂k ∂ j φ (derivatives switched)
= −ǫik j ∂k ∂ j φ (definition of ǫ)
= −∇ × ∇φ
= 0 (in general)

6. If a potential flow solution can be obtained to a problem, u = ∇φ, inserting this result in Equa-
tion (2.171), we obtain an expression for p(x) with the viscous terms absent (because ζ = 0).
This means any solution u(x) for which ζ = 0, is a solution to the inviscid equations as well.
7. Potential flow solution to an equation can be obtained by substituting Equation (2.174) in the
continuity equation for incompressible flows.
∇ · u = ∇ · ∇φ = ∇2 φ = 0
This is called as the Laplace equation, which is a linear homogeneous second order equation.
8. An alternative solution scheme for two dimensional irrotational flows is the use of stream func-
tions defined as
∂ψ ∂ψ
u= v=−
∂y ∂x
This choice automatically satisfies the continuity equation
∂ x ∂y ψ + ∂y (−∂ x ψ) = 0
We now require that ψ satisfies the irrotationality condition. In two dimensions the vorticity ζz
from Equation (2.169) is given by
!
∂v ∂u
ζz = 2 −
∂x ∂y
!
∂2 ψ ∂2 ψ
=2 − 2 − 2
∂x ∂y

c P Sunthar, March 8, 2010



92 Chapter 2. Differential Analysis of Fluid Flow

Irrotationality (ζz = 0) therefore implies

∇2 ψ = 0

which is again a Laplace equation. Stream function method can also be used for axis-symmetric
flows.
9. Potential function automatically satisfies irrotationality and the continuity equation leads to a
Laplace equation. Stream function in two dimensions automatically satisfies continuity and
irrotationality leads to a Laplace equation.

10. For two dimensional flows, there is a deeper relationship between potential and stream func-
tions. We have the conditions from the potential and stream function definition

∂φ ∂ψ
(u =) =
∂x ∂y
(2.175)
∂φ ∂ψ
(v =) =−
∂y ∂x

These are called as Cauchy-Riemann conditions in calculus of complex variables, and is a well
studied system. These conditions must be satisfied by the real and imaginary parts of an ana-
lytic function (existence of a complex derivative). That is for any function w(z)

w(z) = φ(x, y) + i ψ(x, y) (2.176)

where z = x + i y, to be analytical ( dw
dz exists) its real part φ and imaginary part ψ must obey
Equations (2.175). Differentiating one of Equations (2.175) by ∂ x and other by ∂y and adding
gives two conditions: ∇2 φ = 0 and ∇2 ψ = 0.

11. Conversely, if we find an analytical function w(z), then its real and imaginary parts satisfy
Laplace equation, and therefore are solutions to the incompressible and irrotational (i.e. in-
viscid) flow. w(z) is called as the complex potential, and its derivative (see Arfken and Weber,
1995, for an introduction).
dw
= u − iv
dz
is called as the complex velocity.

12. For two-dimensional flows, lines of constant φ and constant ψ are orthogonal. From vector
calculus we know that the gradient of a scalar ∇φ represents the direction of the normal to lines
of constant φ. This implies the local velocity vector is normal to the lines of constant φ. Lines
of constant ψ on the other hand imply that there is no flow across it (The difference ψ2 − ψ1
between any two streamlines represents the volumetric flow rate, per unit depth.

Zy2 Zy2
Q ∂ψ
= dy u = dy = ψ2 − ψ1 (2.177)
depth ∂y
y1 y1

Therefore, the flow rates between a pair of points separated by the same ∆ψ on constant ψ lines,
are the same, and no fluid escapes through the lines of constant ψ).

Irrotational flow theory therefore forms an important aspect of solution of inviscid flows, mainly
because of its mathematical convenience, since the solution techniques for the Laplace equation is
very well developed. The inviscid solutions simply reduce to solving the second order Laplace equa-
tion, and integrating the pressure from Euler equations.

c P Sunthar, March 8, 2010



Lecture 2.26. Potential flow Problems 93

Lecture 2.26 Example problems in Potential Flow

Concepts You Must Know

1. Stagnation point flow of ideal fluid

2. Pressure distribution in flow past cylinder, and a paradox

3. Superposition principle

4. Porous media flow (low Reynolds number) and potential flows

Methods of obtaining solutions to potential flow problems are beyond the scope of this course. We
will only discuss the application of the solution so found in deriving useful fluid mechanical results.
Even in actual applications, one does not strive to obtain solutions, as these have been thoroughly
researched and probably exhausted by mathematicians for centuries! All one needs to do is to find
out suitable analytical solution, and use it for further analysis.

2.7.1 Stagnation point flow


We have seen the approximate distribution of pressure and velocity in a stagnation point flow. Now
we will quantify them. The potential function for a planar stagnation point flow is given by

A 2 
φ= x − y2 (2.178)
2
giving
u= Ax v = −A y (2.179)

We can verify that it satisfies the boundary conditions at the solid surface, and there is a stagnation
point at the origin. The stream function can be obtained by integration from the Cauchy-Riemann
conditions Equation (2.175)
ψ = Axy (2.180)

The streamlines are therefore hyperbolic functions (of y on x or x on y). Note also a finite velocity at
y = 0 implies, the fluid does not obey no slip at the wall. The pressure distribution can be obtained
by integrating the momentum balance equations.

A2 2
p = p0 − ρ (x + y2 ) (2.181)
2
where p0 is the pressure at the origin.

2.7.2 Flow past a cylinder


Planar flow past a cylinder has solution as the functions
!
R2
φ=U r+ cos θ (2.182)
r
!
R2
ψ=U r− sin θ (2.183)
r
(2.184)

c P Sunthar, March 8, 2010



94 Chapter 2. Differential Analysis of Fluid Flow

The velocity components are given by


!
R2
ur = U 1 − cos θ
r2
!
R2
uθ = −U 1 − 2 sin θ
r

The pressure distribution is given by


 
2 2
!  R 2
ρ U2  1 − R
 2 

p = p0 − + 4 sin θ  (2.185)
2 r2 r

which has a fore-aft symmetry (replace θ by π − θ). This implies that the net force on the cylinder
in the direction of flow is identically zero. That is, the form drag (due to normal stress variation)
in inviscid flows is absent. (In the case of viscous flows the pressure is not fore-aft symmetric, and
real fluids do have a form drag). Since the viscous stress (tangential) is absent in inviscid flows, it
follows that inviscid solutions predicts a frictionless motion of solid bodies in flow. This is known
as the d’Alembert’s paradox: An experimental drag could be measured on flow past objects, but could
not be explained by the theory (that is before the viscous theory of fluids).

Lecture 2.27 Linearity of the Laplace equation allows superposition of elementary flows to con-
struct solution to flows in complex domains.
Uniform flow, Flow past a cylinder, Stagnation flows are some of the simple flows, and also called
as Elementary flows or building blocks (see Fox and McDonald (2001); Wilkes (1999) for more such
examples). The governing equations in terms of potential or stream functions is the Laplace equation.
This equation is linear, implying if ψ1 and ψ2 are solutions to the equation then ψ3 = ψ1 + ψ2 is also a
solution. This provides interesting application of combination of elementary flows, and comparing
with actual flows.
As an example consider a wide channel of water having a small canal emptying out perpendicular
to it (Example 7.3 in Wilkes (1999)). The elementary flows in this case are the uniform flow

ψ=Ux

and the line sink


Q
ψ=− θ

The line source/sink is a source/sink of mass flow that emits/sucks from/into a line perpendicular
to a plane. ±Q is the volumetric flow rate (per unit depth along the line) of mass, for a source/sink
as the sign may be.
In the upstream far away from the small canal the flow is uniform. Near the canal the flow is
similar to the flow generated by a line sink. In this case the line sink has only one half to suck from
therefore the superposed solution looks like

Q
ψ=Ux− tan−1 (y/x)
π

2.7.3 Porous Media Flow


Another application of potential flow theory is in viscous flows in a porous medium. A porous
medium is a complex connected network of micro-channels through which fluid flows. Because the
size of the channels are small, and the velocities are typically very small (compared to usual speeds),

c P Sunthar, March 8, 2010



Lecture 2.27. Superposition Principle 95

the Reynolds number in these flows is very small (Re ≪ 1), and creeping flow equations apply for
the velocity field in the pores. However, a “superficial” velocity can be defined as

Q
u0 =
A
where Q is the volumetric flow rate and A0 is the overall cross-section area (including that of the
pores): A0 = A pores /ǫ, where ǫ is the void (pore) fraction. Obtaining the detailed flow profiles (inside
the pore given the complex distribution of pores) is a difficult task. But the superficial velocity is
related to the pressure gradient by a simple expression in certain limits
κ
u0 = − ∇p (2.186)
µ

where κ (/kappa/) is called as the permeability of the medium, and µ is the viscosity of the fluid. This
is called as the d’Arcy’s law, though it is not a “law”, it is just an empirical observation (a theoretical
derivation will be provided later.) It states that superficial velocity is expressed as a gradient of a
potential (in this case the pressure itself), therefore the flow in the superficial velocity is irrotational,
and therefore inviscid. Note the microscopic velocities inside the pores are still viscous, and the
viscosity enters the problem through d’Arcy’s law. Problems in porous media can therefore be solved
by methods of potential flow. Examples include: vertical water bores/oil wells (which is a line sink),

c P Sunthar, March 8, 2010



96 Chapter 2. Differential Analysis of Fluid Flow

2.8 Turbulent Flows


2.8.1 What They Said
I am an old man now, and when I die and go to Heaven there are two matters on which I hope for
enlightenment. One is quantum electrodynamics and the other is the turbulent motion of fluids.
And about the former I am rather optimistic.

Sir Horace Lamb (1932)

Turbulence is the most important unsolved problem of classical physics.

Richard Feynman

Turbulence is an irregular motion which in general makes its appearance in fluids, gaseous or
liquid, when they flow past solid surfaces or even when neighboring streams of the same fluid flow
past or over one another.

G I Taylor and von Kármán (1937)

Big whirls have little whorls, which feed on their velocity,


And little whirls have lesser whirls, and so on to viscosity.

Lewis Richardson

Lecture 2.28 The origin of turbulent flows is due to external disturbances, and a typical charac-
teristic is its chaotic fluctuations in velocity.
Concepts You Must Know
1. Characteristics of turbulent flows

2. Governing equations of turbulent motion is the Navier Stokes, and no new fundamental physics
is involved

3. Non-linearity of NSE leads to multiple solutions, turbulent pattern is one of them

When do we know a flow is turbulent, and otherwise? Here are a few characteristic features of
turbulent flows, that are seen in all such flows.

• Highly unsteady (violently fluctuating) fluid motion, appears to be random/chaotic.

• Three dimensional, despite time averaged velocities being two dimensional: u = ū + u′

• Large Vorticity ζ = ∇ × u

• Increases rate of stirring of conserved quantities

– packets of different concentration of a scalar variable φ (such as momentum, Temperature,


concentration) brought into contact
– Actual mixing by molecular diffusion

• Mixing causes increased dissipation into heat

• Wide range of length and time scales

c P Sunthar, March 8, 2010



Lecture 2.28. Turbulent Flow Characteristics 97

• Good: for mixing chemical species and heat transfer


• Bad: for momentum (as more energy is required to move)

• Engineering: understand, predict and control

– Not interested in predicting the actual fluctuating motion, but in the mean motion, and
transport coefficients in turbulent flow.

2.8.2 Origin of Turbulence


It is difficult to give a general theory into the origin of turbulence, but in most cases it follows the
pattern that can be described in a simple pipe flow (one that is also of most consequence to process
engineers).

• Consider a fully developed laminar flow in a smooth pipe. The nature of the flow (velocity
profile) is the same for any small Reynolds number. We have seen earlier that the creeping
flow solution (Re ≪ 1) is also the one dimensional flow solution (any Reynolds number). If we
increase Re gradually, we will not notice any change in the flow till a critical Re.

• However, around Re ≈ 2100, we can observe that the flow assumes a different character. It
starts to show a three dimensional character, not always, but intermittently.

• Beyond about Re & 4000, the flow assumes a fully 3-D pattern. Note that even at this high
Reynolds number the one-dimensional laminar Poiseuille solution still holds good. However,
it is not easily realisable experimentally.

• In carefully controlled experiment, with an ultra smooth pipe, in a laboratory table and sur-
roundings isolated from any external vibrations, it is possible to continue up to Re ≈ 23, 000 in
a laminar one-dimensional flow. But for most practical purposes it is only around 2000.

• This experiment suggests that external disturbances lead to the flow seeking a state, which is
more favourable at high Re. An analogy can be drawn to a ball in a potential well. The ball
likes to stay in a well that is deep (low energy), in comparison with a well that is shallow. The
number of wells correspond each to an allowable solution to the governing equations equa-
tion. The non-linear NSE allows multiple solutions both laminar and turbulent. The external
disturbances are like some kinetic energy that is provided to the ball which can kick it out of a
well.

– For small Re, the “laminar well” is deep, no amount of external disturbance can kick it out
of the well. It is also said that in this state the flow is stable to disturbances.
– For large Re, the opposite happens: the “turbulent well” becomes deeper, and is preferred.
Here, the laminar flow is unstable to disturbances and is not practically realisable.
– In the transition region both the wells are equally deep, and the ball can easily shift from
one state to another due to external disturbances.

c P Sunthar, March 8, 2010



98 Chapter 2. Differential Analysis of Fluid Flow

Steady

Unsteady

t
Figure 2.17: Mean and fluctuating velocities in steady and unsteady flows.

Lecture 2.29 For engineering applications the velocity field can be decomposed into a mean mo-
tion and a fluctuating component.

Consider a laminar flow at small Re, such as the pipe flow discussed earlier. When this motion goes
unstable, the unsteady nature of the flow developed is of two types, one in which the mean flow is
steady, and the other in which the mean flow is itself unsteady. The meaning of mean flow being
unsteady means, if we were to repeat the experiment over several realisations of the setup, and then
take an average over realisations (ensembles) then there is a unsteady behaviour (example in a pipe
flow that is externally driven by a pulsating pressure at a definite frequency). See Figure 2.17.
For steady mean flows, the velocity can be decomposed as

u(x, t) = ū(x) + u′ (x, t)


|{z} | {z }
Mean flow Fluctuation
Zt
1
ū(x) = lim dt′ u(x, t′ )
t→∞ t
0

The second equation represents the mean flow (denoted by an overbar) what is called as a time aver-

c P Sunthar, March 8, 2010



Lecture 2.29. Time Averaged equations 99

aging. In the case of unsteady mean flows we define


u(x, t) = ū(x, t) + u′ (x, t)
N
1 X
ū(x, t) = lim u(x, t)
N→∞ N 0

Here the mean flow is obtained by averaging over several realisations N → ∞, and is called as an
ensemble averaging. For these averages, we can obtain the following relationships.
ū = ū; u′ = 0 by definition of the time average
u v = (ū + u′ ) (v̄ + v′ ) = ū v̄ + u′ v′
Here observe the term u′ v′ , which represents the average correlation between u′ and v′ . This term need
not be zero in general. For example if u′ > 0, v′ can be any thing in general: there is nothing known
a priori that will dictate its behaviour. In cases that v′ takes the same sign as u′ the time averaged
product is always positive, and is negative when it takes the opposite sign. Only in cases where it
takes equally random same and opposite values, is the correlation u′ v′ = 0; then it is said that the
“fluctuations are uncorrelated”.
With these identities, time averages can be carried out for each of the terms in the NSE Equa-
tion (2.28), and we obtain as a result
∂ρūi ∂   ∂ p̄ ∂τ̄i j
+ ρ ū j ūi + ρ u′i u′j = − +
∂t ∂x j ∂xi ∂x j
This equation is rewritten with the correlation of fluctuations taken to the right, and defining it as
τ̃i j ≡ −ρ u′i u′j (2.187)
we get
∂ρūi ∂   ∂ p̄ ∂τ̄i j ∂τ̃i j
+ ρ ū j ūi = − + + (2.188)
∂t ∂x j ∂xi ∂x j ∂x j
The term τ̃ (with a squiggle, tilde above) is called as the Reynolds stress. It represents the turbulent
transport of momentum. Recall (from page 21) that when coarse graining from particulate system
to continuum, we introduced a stress arising from the correlation of fluctuations of the microscopic
particle velocities. Similarly, the fluctuations in the continuum velocities leads to an effective stress,
known as the Reynolds stress. Just as hρ C Ci represents the molecular transport of momentum, ρ u′i u′j
represents turbulent transport of momentum, or τ̃)i j is called as the turbulent momentum flux tensor.
The important achievement of this section is to derive equations governing mean motion. The
equations look identical to Equation (2.29), except for the Reynolds stress term. This again brings us
back to a closure problem, of how to specify the stress terms τ and τ̃ in terms of the other unknowns
to solve the equations. τ can be approximated by Newtons law of viscosity, however, τ̃ does not have
such a generic simple dependence on the gradient of the mean velocity. This is a major stumbling
block in the theory of turbulence. Further progress to obtain a meaningful solution can be obtained
only by some approximations to the Reynolds stress (which is called as turbulence modelling), or from
experimental empiricism (correlations). Also similar to the viscous stress term, is the difficulty that
we have introduced an additional variable for which we do not have a dynamic equation by itself.
Even if we write a dynamic equation of change, that will introduce further unknown variables, and
so on. This is known as a closure problem. That is when we make approximations (here neglecting
fluctuations), we introduce additional variables that are not known from first principles. We need to
invoke a phenomenological model to get closure. Newtons law of viscosity is one such model that
provides the closure. The stress is related to the known variables, velocity gradient. Similarly we
need additional equations for describing the Reynolds stress. It is obtaining these relations is what is
called as Turbulence modelling.

c P Sunthar, March 8, 2010



100 Chapter 2. Differential Analysis of Fluid Flow

Lecture 2.30 Elementary turbulence modelling involves simple algebraic expressions for the Reynolds
stress.
Concepts You Must Know
1. Regions of flow near a wall

2. Boussinesq Approximation

3. Prandtl Mixing length Hypothesis

4. Velocity profile near wall, and Universal velocity profile.

We first provide empirical data on the nature of the mean flow velocity profile in turbulent flow,
in the simplest case, near the wall. The flow near a wall can be roughly classified into four regions

1. Viscous sublayer. In this region the flow is laminar, and the shear viscosity plays an important
role. The momentum is primarily transferred by molecular transport of momentum.

2. Buffer layer. This is a transition layer between the viscous and the next layer.

3. Intertial sublayer. In this layer the momentum is primarily transferred by macroscopic eddies,
and viscous effects are sub-dominant.

4. Turbulent Core. This is the main stream, where the eddies transfer momentum (and other
scalar properties such as temperature and species concentration), and velocity profile is nearly
flat. Viscous effects are unimportant.

This classification is not a strict demarcation, as the third and fourth regions may not be distinguish-
able. Some authors use just three regions. The crucial regions are the viscous sublayer and the inertial
sublayer.

2.8.3 Boussinesq Approximation


In order to solve the equations of motion for the mean velocity, we need a closure expression for
the Reynolds stress. Similar to the Newtonian shear viscosity, Boussinesq proposed an eddy viscosity
relationship of the form
dū
τ̃ xy = µ̃ (2.189)
dy
where µT is the eddy viscosity. Unlike Newtonian viscosity (it has been experimentally observed
that defined this way) µ̃ is not a constant property of the fluid. Therefore such an expression is still
incomplete, and must be supplemented by further modelling.

2.8.4 Prandtl Mixing Length Hypothesis


The analogy of the viscous tensor arising from the neglect of molecular fluctuations to a continuum
fluctuations leading to Reynolds stress, can be carried further, as was done by Prandtl. Just as in the
case of Newtonian fluids the molecular flux of momentum can be written as
r
kB T du
τyx ∼ ρ λ (2.190)
m dy

where kB T/m is an RMS fluctuating velocity, and λ is the mean free path, Prandtl proposed for
Turbulent transport of momentum
dū
τ̃yx ∼ ρ vmix ℓ (2.191)
dy

c P Sunthar, March 8, 2010



Lecture 2.30. Turbulence Modelling 101

where vmix is a RMS velocity of turbulence, and ℓ is the mixing length, the length over which eddies
interact. Prandtl’s hypothesis was actually in the expression for vmix
dū
vmix ∼ ℓ . (2.192)
dy
The expression for the Reynolds stress becomes
dū dū
τ̃yx ∼ ρ ℓ2
dy dy
| {z }
µ̃ (2.193)
!2
dū
∼ ρ ℓ2
dy
Had the mixing length been a constant, or dependent on the properties of the fluid, further anal-
ysis would have been simplified. However, Prandtl, through observations proposed the following
expression
ℓ = ky (2.194)
where k is dependent on the nature of flow! It is different for flows near a wall and for free turbulence,
such as in atmosphere. Any constant which is dependent on the nature of flow, is not of much use in
a general prediction, in which the nature of flow is not known before hand.

2.8.5 Velocity Profiles in Turbulent Flow


The mixing length model can be used to derive a velocity profile in turbulent flow near a wall. Very
close to the wall (i.e. in the intertial sublayer) we make an assumption that the turbulent shear stress
is nearly equal to viscous stress at the wall
τ̃yx ≈ τw
Substituting this in the mixing length model Equations (2.193) and (2.194), we get
!2
dū
τw = ρ k2 y2 (2.195)
dy
which can be integrated to give r
1 τw
ū = ln y + c
k ρ
where c is the integration constant. This equation can be written in a dimensionless form by the
following definitions

u+ = p (2.196)
τw /ρ
p
y τw /ρ
y+ = (2.197)
ν
to give
u+ = A + B ln y+ (2.198)
p
Here, u∗ ≡ τw /ρ is called as the friction velocity. The form Equation (2.198) very accurately describes
the experimentally observed mean turbulent flow near flow near a wall in the inertial sublayer. An
expression for the velocity in the viscous sublayer can be obtained by integrating the viscous stress
dū
τw = µ
dy

c P Sunthar, March 8, 2010



102 Chapter 2. Differential Analysis of Fluid Flow

to give the following equivalent expressions

1
τw y
ū =
µ
ū ρ p (2.199)
p = τw /ρ y
τw /ρ µ
u+ = y+

Putting Equation (2.198) and (2.199) together the following empirical expressions have been obtained
experimentally.

Viscous sublayer: 0 < y+ < 5 u+ = y+ (2.200)


+ + +
Buffer layer: 5 < y < 30 u = −3.05 + 5.0 ln y (2.201)
Intertial layer / Core: 30 < y+ u+ = 5.5 + 2.5 ln y+ (2.202)

A plot of these velocity profiles in log and linear scales is shown in Figure 2.18. These profiles are
also known as universal velocity profiles, because they are valid for turbulent flow of any fluid past a
wall. These are used to calculate the heat and mass transfer coefficients in turbulent flow.

c P Sunthar, March 8, 2010



Lecture 2.30. Turbulence Modelling 103

100
Viscous Sublayer
Inertial Sublayer
Buffer Layer
u+

10

1
1 10 100 1000

y+
30
Viscous Sublayer
Inertial Sublayer
Buffer Layer
25

20
u+

15

10

0
0 100 200 300 400 500

y+
Figure 2.18: Empirical universal mean turbulent velocity profile: in logscale and linear scale.

c P Sunthar, March 8, 2010



Bibliography

Arfken, G. B. and Weber, H. J. (1995) Mathematical Methods for Physicists. Academic Press (Prism
Books, India), fourth edition.

Batchelor, G. K. (1993) An Introduction to Fluid Dynamics. Cambridge University Press, Indian edition.

Bird, B. R., Stewart, W. E. and Lightfoot, E. N. (2007) Transport Phenomena. Wiley India, second
edition.

Denn, M. M. (1980) Process Fluid Mechanics. Printice-Hall.

Foust, A. S., Wenzel, L. A., Clump, C. W. and Anderson, L. B. (1994) Principles of Unit Operations. John
Wiley, second edition.

Fox, R. W. and McDonald, A. T. (2001) Introduction to Fluid Mechanics. John Wiley and Sons, fifth
edition. Wiley Singapore Edition.

McCabe, W., Smith, J. and Harriot, P. (2005) Unit Operations in Chemical Engineering. McGraw Hill,
7th edition.

Richardson, J. F., Harker, J. H. and Backhurst, J. (2001a) Coulson & Richardson’s Chemical Engineering,
volume 1. Pergamon Press, fourth edition.

Richardson, J. F., Harker, J. H. and Backhurst, J. R. (2001b) Coulson & Richardson’s Chemical Engineer-
ing, volume 2. Pergamon Press, fourth edition.
Sunthar, P. (2009) Polymer simulations companion: An introduction to brownian dynamics. In SERC
School on Molecular Simulations. Department of Chemical Engineering, IISc Bangalore, India.

Wilkes, J. O. (1999) Fluid Mechanics for Chemical Engineers. Printice-Hall.

c P Sunthar, March 8, 2010


127

Das könnte Ihnen auch gefallen