You are on page 1of 16

ME 313: Fluid Mechanics

Day 23

Some notes on dimensionless groups

Reminders about “dimensionless groups”: the “groups”I’m referring to are groupings of physical
variables. Let me enunciate the principle underlying the use of dimensionless groups:

“Any equation which satisfies a scientific law must be satisfied in all possible systems of units”

In other words, if an equation which represents a scientific law is to work equally well in all systems of
units, the dimensions on both sides of the equations must be the same.

An equivalent way of saying this is that all equations which represent scientific laws must be expressible
in such a way that both sides are dimensionless. Hence we find that it is both natural and necessary to
group physical variables in such a way that the groups are dimensionless. Later on you will understand
just how powerful this principle is.

Probably the most familiar example for you is Mach number [cf. Ernst Mach 1838-1916]. Most of you
have already heard of Mach number; it is just a velocity divided by the speed of sound. When a jet is
traveling at Mach 2, it is traveling at twice the speed of sound. Hence we write

v
M≡
c
where c is the speed of sound. Loosely speaking, the Mach number tells us when compressibility effects
associated with high-velocity fluid motion are important. As is the case with many dimensionless groups,
we can identify a ratio of familiar quantities with the group or its square or some power. In this case

v2 v2 2 12 v 2
M2 = = =
c2 γ RT γ− 1 c p T
γ− 1 2 12 v 2 kinetic energy
⇒ M = =
2 c p T thermal energy

The next important dimensionless group is the Reynolds number Re [cf. Osborne Reynolds 1842-1912].
Re is the ratio of flow momentum to viscous dissipation.

ρVL
Re ≡
µ

This requires some explanation. Rho is the density, V is the flow velocity, L is some characteristic
dimension, and mu is the absolute viscosity. ρV is the momentum density and µ L is proportional to
the shear stress per unit velocity – in other words, the frictional impulse per unit volume. We will find that
the Reynolds number tells us when a flow field will make a transition from laminar to turbulent flow. Also,
things like friction factors in pipes, lift and drag of airfoils, drag coefficients of bluff bodies, all are
expressed as functions of the Reynolds number. I can’t emphasize enough the central role that Re plays
in fluid dynamics.

A dimensionless group that comes up in open channel flow is the Froude number [cf. William Froude
1810-1871]
V
Fr ≡
gL

where V is the flow velocity, g is the acceleration of gravity, and L is a characteristic length. The square
of the Froude number is proportional to the ratio of kinetic energy to gravitational energy.

The Euler number is related to a number of different kinds of pressure coefficients like lift and drag
coefficients [cf. Leonhard Euler 1707-1783]. The Euler number is defined by

V
Eu =
∆p
ρ

and again its square is proportional to the ratio of inertial and pressure forces.

The Weber number is related to the ratio of inertia to surface tension [cf. Moritz Weber 1871-1951]:

V
We =
σ
ρL

This completes our brief survey of dimensionless groups. There are many other groups in common
engineering usage, e.g. the Prandtl number, Nusselt number, and Grashof numbers used in heat
transfer. However the groups I have just discussed are the most important ones for fluid dynamics.
More about Dimensionless Groups and Similarity

We began with a brief discussion of the use of dimensionless parameters such as the Reynolds number
and Mach number. Now we will discuss a powerful tool, called the “Buckingham Pi Theorem”, that
facilitates the formulation of dimensionless parameters. Before we do so, we have to establish a little
terminology.

The premise of dimensional analysis is the notion that two flowfields can be similar.

I think you must have already in your minds a concept that two flowfields can be identical, that is, that the
flow around two identical airplanes flying in exactly the same atmospheric conditions at exactly the same
speed and altitude will be identical. A far more subtle question is the conditions under which the flow
around a small-scale model in a wind tunnel mimics the flow around a full-scale airplane flying in the
atmosphere. This question is the central preoccupation of dimensional analysis (applied, of course, to
many other fields besides aerospace… ).

Implicit in this discussion is that we want to be able to perform scientific tests under controlled conditions
on small scale models and apply what we learn to the design of full-scale models. This activity is central
to all large engineering projects, whether we are talking about finding the optimum wing design for a
supersonic fighter, or finding the dynamic forces on a hydroelectric dam, or predicting the roll
characteristics of an oil tanker under heavy seas. In each case it is simply too expensive or risky to test
full-scale models of every candidate design, and scale-model testing is indispensable.

We will find that dimensional analysis allows us to reduce the amount of data that we must gather to
explore some design space, and conversely, to summarize a vast amount of test data in an economical
fashion. [It is for this reason that we have to cover dimensional analysis before we discuss, say, the flow
of viscous fluids in pipes.]

It should be obvious that the stakes involved in such testing can be huge. Failure to anticipate and
design for all the forces on a dam can be disastrous. Failure to find the optimum wing design for a fighter
plane can add tens of millions of dollars to life cycle costs. You may have seen the famous newsreel
footage of the Tacoma Narrows Bridge collapse. A modern engineer would identify faulty scale-model
testing as the culprit. A less familiar but more modern example is the John Hancock building in
downtown Boston, a lovely glass and steel skyscraper. The building was erected in the early ‘70’s and
immediately began shedding glass window panes under moderate wind conditions. It happens that the
standard design practice of that era was to do wind-tunnel testing of new skyscraper designs to
determine wind loads, but from the four compass-point directions only. Nature provided a surprise in the
form of some non-linearities in the superposition of wind forces. Today the standard practice includes
360 degree testing.
So here is our paradigm: we consider a system that we want to model. The full-scale system we call the
prototype.

pro-to-type (proh'tuh tiep ) n., v. <-typed, -typ-ing>


n.
1. the original or model on which something
is based or formed; pattern… .
[1595-1605; < NL prototypon < Gk protótypon, n.
use of neut. of protótypos original. See PROTO -,
TYPE]

The small-scale system we call the model. We want to understand the circumstances under which the
flow around the model is similar to the flow around the prototype.

MODEL PROTOTYPE

Immediately we are confronted with additional distinctions. There are different kinds of similarity, of
increasing specificity.

First of all, there is simple geometric similarity, the kind that you hopefully learned about in high-school
geometry. Two objects are similar if the ratio of their linear dimensions is exactly the same for all
corresponding dimensions. The ratio of prototype to model linear dimensions is called the scale ratio or
simply the scale.

Next, there is kinematic similarity, which is to say, similarity of motion. In essence, two flow fields are
kinematically similar if the velocities at corresponding points are in a fixed ratio. Fundamentally we mean
that characteristic times (for example, the time that it takes a fluid particle to traverse the chord of an
airplane wing) are proportional.

It is simply a matter of convention that we isolate the time as a parameter rather than the velocity.

Kinematic similarity necessarily means that the streamlines of the model and prototype flow fields will be
related by a fixed scale ratio. Since the streamlines follow all material boundaries, this means that
kinematic similarity implies geometric similarity (though not the reverse).

If two flow fields are kinematically similar, then the fluid velocities, accelerations, volumetric flow rates,
and kinematic viscosities are necessarily related by fixed scale factors which are in turn related to the
length and time scale factors.

The third kind of similarity is dynamical similarity. Essentially two flow fields are dynamically similar if all
fluid forces at corresponding points are in fixed ratios. Formally we define this notion in terms of a mass
ratio, though again this selection is arbitrary, and in fact it is often more convenient to specify a density
ratio:

Dynamic similarity implies kinematic and geometric similarity.


Similarity: (model and prototype are in proportion)

Kinematic
Similarity: Geometric
Dynamic L,T ratio Similarity:
Similarity: L ratio
L,T,M ratio
For any given physical situation, it turns out that there is a calculable number of dimensionless groups
that are required to describe it. The Buckingham Pi Theorem tells us this number, and it gives a
prescription for finding them.

è If there are…

• n physical variables that affect a given physical phenomenon, and


• r base dimensions to measure those variables (r=3 for {M,L,T}), then
• j = n – r dimensionless groups describe the phenomenon.

Examples abound. Here are some simple examples, which later we will attack with the Pi Theorem.

Ideal pendulum

The small-angle pendulum (a point mass swinging at the end of a massless string fixed at the top end) is
a simple harmonic oscillator. What are the parameters that describe a pendulum? Arguably the list of
physical variables (and their dimensions) is

• mass m of the bob, m = M []


• length l of the string, []
l =L
• gravitational acceleration g , []
g =L T 2

• angular frequency of oscillation ω , []


−1
ω =T

Thus n=4. Also, there are 3 base dimensions {M , L, T }, so r=3. The Pi Theorem states that since
j = n − r = 1 , there is just one dimensionless group. It is easy to verify that

1 g
Π1 ≡
ω l

is dimensionless. (Later we will learn how to construct these groups). An important corollary to the Pi
Theorem states that if there is exactly one dimensionless group describing a physical
phenomenon, that group must assume a constant value (since there are no other groups for it to be
a function of!). The Pi Theorem does not provide any way to determine this constant – you have to do
some kind of analysis to find it. In this case, the constant happens to be 1, so we find the familiar formula

g
ω= .
l


, instead of ω in our list of physical variables, we
Note it we had chosen the period of oscillation, τ =
ω
g l
would have found Π 1 ≡ τ and our subsequent analysis would show Π 1 = 2 π → τ = 2 π . Note
l g
also this dimensional analysis assumed that the pendulum angle was small and therefore had no effect
on the relation of frequency to the other variables – in hindsight, this is a key assumption, because the
frequency of a pendulum at large angles does depend on the amplitude of the oscillations. Finally, note
that dimensional analysis is all that is required to see why the mass of the bob does not affect the
frequency of oscillation.
Ideal spring-mass system

Along these lines, consider an ideal mass-spring system. The physical parameters and their dimensions:

• mass m of the bob, m = M[]


• length l of the spring, []
l =L
• spring constant k, []k =F L=M T2
• angular frequency of oscillation ω , []
ω = T−1

In this case again n = 4 , r = 3 , → j = n − r = 1 , so there is one dimensionless group. However, in this


case, the dimensionless group is

1 k
Π1 ≡ .
ω m

Dimensional analysis shows why the spring length has no effect on the frequency of oscillation. Again, a
k
detailed analysis applying Newton’s laws reveals Π 1 = 1 → ω = .
m
Pythagorean theorem

The Pi Theorem can also be used to prove results that are purely mathematical. Consider a right
triangle.

L=length of hypotenuse

Area=A
α

Using dimensional analysis, we argue that the area of the triangle is a function of the length of the
hypotenuse and the smaller of the two acute angles. We could find that function using trigonometric
functions, but dimensional analysis by itself tells us that

A = L2 f (α ).

Now split the triangle into similar smaller right triangles:

A2
L2
α
A1
α
L1

Note the hypotenuses of A1 and A2 are also the side lengths of the larger triangle.

Clearly A1 + A2 = A . Now substitute our equation from dimensional analysis, and note that α is the
same angle for all three triangles! Thus

L21 f (α )+ L22 f (α ) = L2 f (α )→ L21 + L22 = L2

This is just the Pythagorean theorem, which we have just proved using dimensional arguments alone.
The Pi Theorem in gory detail (constructing the groups):

The Buckingham Pi Theorem tells us how many dimensionless groups it takes to describe a given flow
problem. And it gives a prescription for finding them:

If there are

• n physical variables that affect a given flow phenomenon, and


• r base dimensions to measure those variables (r=3 for {M,L,T}), then
• j = n – r dimensionless groups describe the flowfield

Here’s the prescription:

1. List any pertinent parameters (n of them)


2. Pick r=3 “repeating variables”
• geometric: L
• kinematic: V
• dynamic: rho
3. Sequentially combine (multiplicatively) the repeating variables with the other j = n – r variables to
form the groupings – called π groups. Each π group has 4 variables, each with its own exponent.
4. For each of the j π groups:
• Express each variable in terms of the fundamental dimensions M,L,T.
• Add exponents for MLT separately and equate to zero, giving three equations in four
unknowns for each group.
• Relate any three of the exponents to the fourth.
5. Eliminate superfluous or negligible groups (like surface tension in oil tanker design?)
6. Remaining groups must be related to each other in some functional form.
7. Conduct experiments to determine functional forms and undetermined coefficients.

Note that the condition r=3 {MLT} is strictly an artifact of the kind of flow problems we are doing. Were
we to consider problems involving fluid temperature, a fourth dimension would be involved; electric
current, a fifth. (The SI system identifies seven independent base dimensions).

Paradigm: Drag on a smooth sphere in a viscous fluid

Consider a sphere moving through a fluid at some velocity V. There will be some drag force due to
friction. Let that force be F. What does F depend on? We can certainly assume that it depends on the
density and viscosity of the fluid. [After all, a golf ball moves much more slowly through water than
through air… why would that be?] We would also expect that the force would depend on the size of the
ball, so let the diameter D be a parameter. So we have

F = f ( D, V , ρ, µ)

Conceivably we might have to make 10,000 experimental runs to map out the drag characteristic of this
very simple system, with 10-per granularity. But let’s do a dimensional analysis.

Step 1. We have n = 5 parameters: F , D,V , ρ, µ


Step 2. There are m = 3 fundamental dimensions. {M,L,T}. Pick a geometric, a kinematic, and a dynamic
variable: D, V ,ρ are typical choices.
Step 3. There must therefore be j = n - m = 2 dimensionless groups. We combine the extra parameters
one-by-one:

π1 = C1 D aV bρc F d
π2 = C2 D aV bρc µ d

Step 4. For each group we eliminate all but one exponent:

GROUP 1:

π1 = C1 D aV bρc F d
b c d
 L   M   ML 
⇒ M LT = L    3  2 
0 0 0 a
T   L   T 
M c=− d
T b = − 2d
L a + b − 3c = − d ⇒ a = − 2 d
F
⇒ π1 = C1
ρV 2 D 2

1
This is a form of Euler number known as a drag coefficient. The quantity q = ρV 2 is called the
2
π 2
dynamic pressure. The frontal area of the sphere is A = D . If we set C1 = 8 π, then we would find
4
F
π1 = = Cd which is the conventional definition of drag coefficient. Note, however, that the actual
qA
determination of the arbitrary constant C1 is not part of the Buckingham procedure.

GROUP 2:

π2 = C2 D aV bρc µ d
b c d
L  M   M 
⇒ M 0 L0T 0 = La    3   
T   L   LT 
M c=− d
T b=− d
L a + b − 3c = d ⇒ a = − d
µ
⇒ π2 = C2
ρVD
−1
Here we observe that we have generated a Reynolds number; with C2 = 1 we would have π2 = Re

So here is the final desired result for the analysis of the drag on a moving sphere:

Cd = f ( Re)

The drag coefficient is a function of the Reynolds number. This is a very nice result, compressing our
four apparent degrees of freedom into a single independent variable.
Example: Ideal Pendulum

Problem: Consider the frequency of oscillation f = ω 2 π of an ideal pendulum. Suppose we don’t recall
the formula for f from freshman physics, but conjecture that the frequency depends on the length L of the
pendulum, the mass M of the pendulum bob, and the acceleration of gravity g. Find the functional form
( )
of f = f M , L, g using dimensional analysis. Compare with the exact result. Solution: There is only
one dimensionless group with n=4. Our repeating parameters are {L, M , g}. The frequency f has the
dimensions of inverse time.

Then
Π 1 = C1 La M b g c f d
c d
L 1 
⇒ M T L = L M  2
0 0 0 a b
 
T  T 
M0 → b = 0
d
T0 → c = −
2
d
L0 → a + c = 0 ⇒ a =
2
We find that the frequency of the pendulum must be independent of the mass of the pendulum bob.

Also, we know that we have only one dimensionless group,


L
Π1 = f
g
which is necessarily constant, which implies
g
f ∝
L
The exact result is
1 g
f =
2π L
We see that the form of the equation that is derived with Newtonian mechanics can be determined
strictly from dimensional analysis alone.
Principle: if in a given physical situation there is only one dimensionless group, that group must be a
constant.

Application: At low velocities the drag on an object is independent of fluid density. In the case of the
sphere this implies

Cd ∝ Re − 1

which is exactly the trend shown here for low Reynolds number. In fact we can repeat the Pi theorem
analysis for the reduced parameter set {F,V,D,mu}, which will give us j = n - r = 1 dimensionless group.

When there is only one dimensionless group, it cannot be a function of anything, so it must be a
constant. This is a important consequence of the Buckingham Pi Theorem and is useful in many
circumstances ranging far beyond fluid mechanics.

Another important principle involves the special case where we have exactly two dimensionless groups.
In this case the value of the second is fixed whenever the value of the first is known. In other words, if we
know Π 2 at a particular value of Π1 , then we know Π 2 at all combinations of parameters that yield that
particular value of Π1 , even if we are ignorant of the functional form. For example, consider the moving
sphere again. The drag coefficient at a Reynolds number of 1 is 10. Hence we know the drag coefficient
for all the following conditions:

ρ = 1000 kg m3 ,V = 0.001 m s , L = 0.01m, µ = 0.01Pa s


ρ = 500 kg m3 , V = 0.00002 m s , L = 1m, µ = 0.01Pa s
ρ = 50 lbm ft 3 ,V = 0.01 ft s , L = 01
. ft , µ = 0.05 lbm ft s
Example: Flow over a weir.

Width=B
H Q

Assume that the discharge Q, width B, height of weir P, gravitational acceleration g, density rho,
viscosity mu, surface tension sigma, and head H are all pertinent parameters affecting the flow over the
weir. Thus n=8, and j=n-m=5.

First we select three repeating variables:

Geometric: pick one of H, P, B, say H. (One of the fallouts of the method is that P/H and B/H are
dimensionless groups).

Kinematic: Q is the obvious choice.

Dynamic: rho is the obvious choice.

List five groups:

π1 = π1 (ρ, Q, H , B)
π2 = π2 (ρ, Q, H , P )
π3 = π3 (ρ, Q, H , g )
π4 = π4 (ρ, Q, H , µ)
π5 = π5 (ρ, Q, H , σ)

Solve for exponents:

GROUP 1:
π1 = C1ρa Q b H c B d
b
 M  L  c d
a 3
⇒ M LT = 3   LL
0 0 0
 L  T 
M a=0
T b=0
L c=− d
d
B B
⇒ π1 = C1   ⇒ π1 = π1  
H  H 

GROUP 2: exactly the same


π2 = C2ρa Q b H c P d
b
 M  L  c d
3 a

⇒ M LT = 3   LL
0 0 0
 L  T 
M a=0
T b=0
L c=− d
P
⇒ π2 = π2  
H 

GROUP 3: the first interesting case

π3 = C3ρa Q b H c g d
b
 M  L  c L 
a 3 d

⇒ M LT =  3    L  2
0 0 0
 L   T  T 
M a=0
T b = − 2d
L 3b + c + d = 0 ⇒ c = 5d
 Q 
⇒ π3 = C3Q − 2 d H 5 d g d ⇒ π3 = π3  1 2 5 2 
g H 

The term in parentheses is a form of the Froude number.

GROUP 4: getting more complicated… .


π4 = C4ρa Q b H c µ d
b
 M  L   M 
3 a d

⇒ M L T =  3    Lc  
0 0 0
 L   T   LT 
M a=−d
T b=− d
L − 3a + 3b + c − d = 0 ⇒ c = d
 ρQ 
⇒ π4 = C4ρ− d Q − d H d µ d ⇒ π4 = π4  
µH 

[Reynolds number]

GROUP 5: similar in spirit:


π5 = C5ρa Q b H c σd
b
 M  L   M 
a 3 d

⇒ M L T =  3    Lc  2 
0 0 0
 L   T  T 
M a=−d
T b = − 2d
L − 3a + 3b + c = 0 ⇒ c = 3d
 ρQ   Q H2 
⇒ π5 = C5ρ− d Q − 2 d H 3d σd ⇒ π5 = π5   = π  
 σH   σ ρH 
32 5

[Weber number]

Hence

B P Q ρQ Q H 2 

π= π , , , , 
H H gH
52
µH σ ρH 

We want to express the discharge Q as a function of the dimensionless groups; with a little insight that
gravity dominates the problem we can pick the Froude number group to expand the expression as

B P 
Q = π , , Re,We g H 5 2
H H 

and with the reasonable conjecture that the flow rate will be proportional to the width, so that B/H appears
with power unity in the final expression,

P  B P 
Q = π , Re,We g H 5 2 = π , Re,We g BH 3 2
H  H  H 
which is the desired expression.

You might also like