CHAPTER 8

Dimensional Analysis and Scaling

8.1 INTRODUCTION
Process development, process support, and scaling go hand in hand:
we cannot do process development without considering upscaling since
moving the process to the commercial scale is the purpose of the effort.
Likewise, we cannot do process support without considering downscal-
ing. To support a commercial process, we have to model the troubled
portion of the production plant in the laboratory or in a pilot plant.
We cannot solve the production problem in the laboratory or in the
pilot plant without considering upscaling the solution into the commer-
cial plant. In other words, process development, process support, and
scaling are, essentially, one and the same activity.

8.2 MODELING
Upscaling and downscaling involve modeling. We build models for the
same reasons we use Dimensional Analysis during process development
or process support: to reduce the time from idea to commercialization
and to reduce the cost of the effort. The major cost savings of modeling
come from not building an inoperable, full-scale commercial plant.
There are four types of models. They are

1. true models;
2. adequate models;
3. distorted models;
4. dissimilar models.
True models involve building all significant process features to
scale. Thus, the model is an exact replica of the commercial plant,
which we call the “prototype.” We build true models in some safety
investigations in order to determine definitely the cause of a grievous,
horrific event. While true models may provide highly accurate infor-
mation, they are capital intensive, expensive to operate, and require
extended time periods to build.
138 Dimensional Analysis

Adequate models predict one characteristic of the prototype

accurately. If the sizes of the model and the prototype are signifi-
cantly different, then it is unlikely that we can achieve complete sim-
ilarity. And, for complex processes, a complete model is actually a
full-scale prototype, i.e., a true model.1 If the modeled characteristic
is the dominant, controlling factor in the process, then an adequate
model may be sufficient. For example, porous solid catalyzed pro-
cesses are generally stagnant film diffusion rate limited or pore dif-
fusion rate limited. If a process is so limited, then we only have to
ensure the same controlling regime in our laboratory or pilot plant
reactors. If we do not ensure equivalent controlling regimes in the
laboratory or pilot plant reactors, then any process development or
process support will be wasted effort. If we do not consider whether
the commercial process is stagnant film diffusion rate limited, pore
diffusion rate limited, or reaction rate limited, then we will finish
our effort with an expensive scattergram of the experimental results.
This situation actually occurs more often than we like to admit.
Many porous solid catalyzed commercial processes are pore diffu-
sion rate limited due to high interstitial fluid velocity through the
reactor. Such high fluid velocity minimizes the boundary layer sur-
rounding each catalyst pellet, thereby making the process pore diffu-
sion rate limited. Unfortunately, most porous solid catalyzed pilot
plant processes, i.e., models, are operated at low interstitial fluid
velocities in order to minimize feed and product volumes at the
research site. In such situations, stagnant film diffusion rate is the
controlling regime. The result of a multiyear, multicatalyst testing
effort will be an expensive scattergram around the average value for
the film diffusion rate constant. On the other hand, considerable
effort can be made at the laboratory scale to ensure that catalyst
testing occurs in the reaction rate limited regime. Plots with impres-
sive correlations result from these types of experimental programs.
Unfortunately, when the best catalyst is tested in the prototype, it
displays the same efficiency and productivity as the current catalyst.
In such cases, the prototype is either stagnant film or pore diffusion
rate limited. It does not matter how reactive the catalyst is in the
laboratory; in the prototype, the process is incapable of keeping the
catalytic site saturated with reactant. In conclusion, the controlling
regime of the model must be identical to the controlling regime of
the prototype. With regard to the process, adequate models behave
 Dimensional Analysis and Scaling 139

similarly to their prototypes, even though they may be many times

smaller than their prototypes.

In distorted models, we violate design conditions intentionally for one

reason or another. Such distortion affects the prediction equation. In
other words, we have to correct data from the model in order to simulate
the prototype. Hydrologic river basin models are the most common dis-
torted models. In these models, the horizontal and vertical lengths do not
have the same ratios or “scaling factors.” In a geometrically similar
model, the horizontal and vertical ratios are equal; for example

P LH P LV
5κ and 5κ
M LH M LV

where P LH is the prototype horizontal length of interest; M LH is the

model horizontal length equivalent to P LH ; P LV is the prototype verti-
cal length of interest; and, M LV is the model vertical length equivalent
to P LV . κ is a constant, the “scaling factor.” For a distorted model

P LH P LV
5κ and 5λ
M LH M LV

where κ ¼6 λ. It is “legal” to use distorted models, just so long as we know

we are doing it and we understand why we are doing it. With regard to the
process, distorted models behave in a manner similar to their prototypes;
however, one dimension of the model will not be scaled equivalently to the
other dimensions. Thus, a distorted model may look squat or tall or broad,
depending on the distortion, when compared to its prototype.2
Dissimilar models comprise the fourth and last model type. Such
models have no apparent resemblance to the prototype. Dissimilar
models have, as their name states, no similarity to their prototypes.
These models provide information about the prototype through
suitable analogies.

8.3 SIMILARITY
We base our models on similarity. Four similarities are important to
chemical engineers. They are
1. geometrical;
2. mechanical;
140 Dimensional Analysis

3. thermal;
4. chemical.

In general, geometric similarity means that given two objects of dif-

ferent size, if there is a point within the smaller object, which we iden-
tify as the model, with coordinates xM, yM, and zM, and a similar
point within the larger object, i.e., the prototype, with coordinates xP,
yP, and zP, then the two objects are similar at that given point if
xP yP zP
5 5 5L
xM yM zM
The two objects are geometrically similar if the above condition
holds for all corresponding points within the two objects.

Mechanical similarity comprises three subsimilarities, which are

static similarity, kinematic similarity, and dynamic similarity. Static
similarity demands that two geometrically similar objects have relative
deformation for a constant applied stress. This similarity is of interest
to civil and structural engineers.
Kinematic similarity means the constituent parts of a model and
prototype mechanism or process in translation follow similar paths or
streamlines if the model and prototype are geometrically similar. Thus
vP
5V
vM
where vM is the velocity of the translating model part or particle and vP is
the translating prototype part or particle. V is the velocity scaling factor.

Dynamic similarity demands the ratio of the forces inducing accel-

eration be equal at corresponding locations in geometrically similar
mechanisms or processes. In other words, the ratio
FP
5F
FM
where FM is the force at location xM, yM, and zM in the model and FP
is the force at location xP, yP, and zP in the prototype, holds true at
every corresponding location in the two mechanisms or processes.

Thermal similarity occurs when the ratio of the temperature differ-

ence at corresponding locations of a geometrically similar mechanism
or process is equal. If translation, i.e., movement, occurs, then the
 Dimensional Analysis and Scaling 141

process must also demonstrate kinematic similarity for thermal similar-

ity to occur. Thus, thermal similarity requires geometric similarity and
kinematic similarity.

As chemical engineers, our major concern is the reactions occurring

in a process. We want our model to reflect what occurs in our proto-
type. To ensure that outcome, our model must be chemically similar to
our prototype. Chemical similarity demands the ratio of concentration
differences at all corresponding locations in the model and in the pro-
totype be equal. Therefore, our model and prototype must also be geo-
metrically, mechanically, and thermally similar.

Consider two mechanical processes involving the NavierStokes

equation. Let one process be large and the other process be small. Our
question: is the larger process similar to the smaller process? The best
way to answer that question is to convert the NavierStokes equation
into a dimensionless form. Let us define a characteristic length L and
velocity V, then form the dimensionless variables x and v , which are
xS vx;S
xS 5 and vx;S 5
L V
where the subscript S identifies the small process; x, the length in the
x-direction; L, the characteristic length; xS , the dimensionless length in
the x-direction; vx,S, the fluid velocity in the x-direction in the small
process; V, the characteristic velocity; and vx;S , the dimensionless veloc-
ity in the x-direction. We define dimensionless pressure as
p
p 5
ρV 2
and we define dimensionless time as
Vt
t 5
L
The NavierStokes equation in one-dimension for the small process is

@vx;S @vx;S 1 @pS μ @2 vx;S

1 vx;S 52 1 gx 1
@t @xS ρ @xS ρ @ðxS Þ2
where pS is the pressure of the small process; gx is the acceleration due
to gravity; μ is the fluid dynamic viscosity; and ρ is the fluid density.
@vx;S @vx;S
@t represents the local acceleration of the fluid particle; vx;S @xS is the
142 Dimensional Analysis

1 @pS
convective acceleration of the fluid particle; ρ @xS represents the pressure
μ @2 vx;S
acceleration due to pumping action; and ρ @ðxS Þ2 is the viscous decelera-
3
tion generated by objects in the fluid’s flow path. Converting the
dimensional equation to a dimensionless equation yields

2 
2
2 
V @vx;S V 
@vx;S V @pS μV @ vx;S
2
1 v 5 2 1 g x 1
L @t L x;S @xS L @xS ρL2 @ðxS Þ2

Multiplying the above equation by L=V 2 and simplifying yields

!
@vx;S 
@vx;S @pS gx L μ @2 vx;S
1 vx;S  5 2  1 2 1
@t @xS @xS V ρLV @ðxS Þ2

Now consider the one-dimensional NavierStokes equation for the

larger process, identified by the subscript L: it is

@vx;L @vx;L 1 @pL μ @2 vx;L

1 vx;L 52 1 gx 1
@t @xL ρ @xL ρ @ðxL Þ2
We can convert this NavierStokes equation into a dimensionless
equation just as before. Doing so give us
!
@vx;L @v
x;L @p 
g L μ @ 2 
v x;L
1 vx;L  5 2 L 1 x 2 1
@t @xL @xL V ρLV @ðxL Þ2

Note both dimensionless equations have the same dimensionless

groups, namely
gx L μ
and
V 2 ρLV
which are the inverse Froude number and the inverse Reynolds num-
ber. The Froude number is the ratio of the inertial forces to gravita-
tional forces and the Reynolds number is the ratio of the inertial forces
to viscous forces. Thus, if



gx L gx L
5
V2 S V2 L
and



μ μ
5
ρLV S ρLV L

then the two processes are mechanically equivalent.

 Dimensional Analysis and Scaling 143

However, the two processes must be geometrically similar for them

to be mechanically similar. For each process, we defined
xS xL
xS 5 and xL 5
L L
Thus
xS xL
L5 and L 5
xS xL
Equating the above equations, then rearranging give us
xS xL
5
xS xL
xS xS
5
xL xL

Therefore, the two processes are geometrically similar.

In summary, two processes are similar if their dimensionless geo-

metric ratios are equal and if their dimensionless process parameters
are equal. In other words, each process will generate a set of dimen-
sionless Π parameters. When corresponding parameters are equal, then
the comparator processes are similar. Symbolically

ΠGeometric
1 5 ΠGeometric
2

ΠStatic
1 5 ΠStatic
2

ΠKinematic
1 5 ΠKinematic
2

ΠDynamic
1 5 ΠDynamic
2

ΠThermal
1 5 ΠThermal
2

ΠChemical
1 5 ΠChemical
2

Thus, similarity depends upon Dimensional Analysis.

8.4 THEORY OF MODELS

The most general equation we can write for a prototype is

ΠP1 5 f ðΠP2 ; ΠP3 ; . . .; ΠPn Þ

where the subscript numeral identifies a dimensionless parameter and
superscript P indicates prototype. This equation applies to all
144 Dimensional Analysis

mechanisms or processes that are comprised of the same dimensional

variables. Thus, it applies to any model of the same mechanism or pro-
cess, which means we can write a similar equation for that model

ΠM
1 5 f ðΠ2 ; Π3 ; . . .; Πn Þ
M M M

Dividing the prototype equation by the model equation gives us

ΠP1 f ðΠP2 ; ΠP3 ; . . .; ΠPn Þ

5
ΠM
1 2 ; Π3 ; . . .; Πn Þ
f ðΠM M M

Note that if ΠP2 5 ΠM

2 and Π3 5 Π3 and so on, then
P M

ΠP1
51
ΠM
1

Thus, ΠP1 5 ΠM
1 , which is the condition for predicting prototype
behavior from model behavior. The conditions

ΠP2 5 ΠM
2

ΠP3 5 ΠM
3

ΠPn 5 ΠM
n

constitute the design specifications for the prototype from the model or
the model from the prototype, depending whether we are upscaling or
downscaling. If all these conditions are met, then we have a true
model. If the above conditions hold for the controlling regime of the
model and the prototype, then we have an adequate model. If most of
the above conditions hold, then we have a distorted model that
requires a correlation to relate ΠP1 and ΠM1 ; in other words, we need an
additional function such that

ΠP1 5 f ðcorrelationÞΠM
1

If none of the above conditions holds true, then we have an analo-

gous model.

We generally do not build true models in the chemical processing

industry because the processes are so complex. A true model of a
chemical process implies building a commercial-sized plant, which is
 Dimensional Analysis and Scaling 145

far too costly and time consuming for an organization to do. Most
models in the chemical processing industry are adequate or distorted
models. Of these two types, adequate models are the better since they
model the controlling regime of the process. Distorted models are
more difficult to use because we have to determine the correlation
between the distorted model and the prototype. Developing that corre-
lation takes time and costs money—two commodities in short supply
in our global economy.

8.5 SUMMARY
This chapter demonstrated the dependence of scaling and model theory
upon Dimensional Analysis. It also discussed the types of models
available to chemical engineers.

REFERENCES
1. V. Skoglund, Similitude: Theory and Applications, International Textbook Company,
Scranton, PA, 1967, pp. 7475.
2. R. Johnstone, M. Thring, Pilot Plants, Models, and Scale-Up Methods in Chemical
Engineering, McGraw-Hill, New York, NY, 1957, Chapter 3.
3. R. Granger, Fluid Mechanics, Dover Publications, New York, NY, 1995, p. 204.