Scale-Up/Scale-Down: Lecture Notes

LECTURE NOTES
on
SCALE-UP/SCALE-DOWN
K.Ch.A.M. Luyben
J.J. Heijnen
R.G.J.M. van der Lans
J.J.M. Potters
2006
Delft University of Technology

Faculty of Applied Sciences
Department of Biotechnology
These Lecture Notes are based on the original Dutch text of the course “Schaalvergroten” by NW
(Nico) Kossen and PG (Peter) Krouwel. The course dates back to at least 1978, when it was initiated
by prof. Kossen of the subdepartment Biochemical Reactors, later the department BioProcess
Technology of the Faculty Chemical Technology. In 1983 prof. Karel Luyben succeeded prof. Nico
Kossen as head of the department and lecturer of this course, together with Jacques Potters. In 1998 an
English text was prepared for an international 3-day course in Lek, Slovenia. Since 2000, prof. Sef
Heijnen, now head of the subdepartment (section) BioProcess Technology, Karel Luyben, now dean of
the Faculty of Applied Sciences, and Rob van der Lans are involved in this course. Several guest
speakers are invited each year to present relevant material or industrial cases.
The present text provides only the basics of tools and methods taught in this course. Other, essential,
material such as examples and exercises are provided for during the course.
April 2002, Rob van der Lans, editor.
Several, relatively minor, improvements were made to the text. Dr. Peter Hamersma, educational
director of the study programme (Bio)Chemical Engineering (Molecular Science and Technology)
joined the group of teachers in 2003.
April 2006, Rob van der Lans
R.G.J.M.vanderLans@tudelft.nl
Scale-up/scale-down Introduction
0. Introduction
In this course on scale-up we will focus on the problems that arise with the
enlargement of equipment and installations as is often the case in the process industry.
Scale-up is not an isolated subject; it basically combines any different engineering
subjects. Insight in what really happens inside the process equipment and being able
to distinguish in the multitude of subprocesses the (speed) controlling step is always
essential.
The theoretical background of the scale-up subject is strictly formal, and should be
applied with the necessary ‘physical feeling’ to be successful.
The reasons for scale-up in the process industry are besides the growth of markets
also the introduction of new processes. In a broader sense the problems linked to
scale-up in the process industry can also be found in other areas like transport of
goods and people or the industries producing units like television sets or cars.
TIME AND CHANGE
There is no absolute way to measure change. In the awesome complexity of the universe, even within
any given society, a virtually infinite number of streams of change occur simultaneously. All ‘things’ –
from the tiniest virus to the greatest galaxy – are, in reality, not things at all, but processes. There is no
static point, no nirvana-like un-change, against which to measure change. Change is, therefore,
necessarily relative.
It is also uneven. If all processes occurred at the same speed, or even if they accelerated or
decelerated in unison, it would be impossible to observe change. The future, however, invades the
present at differing speeds. Thus it becomes possible to compare the speed of different processes as
they unfold. We know, for example, that compared with the biological evolution of the species,
cultural and social evolution is extremely rapid. We know that some societies transform themselves
technologically or economically more rapidly than others. We also know that different sectors within
the same society exhibit different rates of change – the disparity that William Ogburn labelled ‘cultural
lag’. It is precisely the unevenness of change that makes it measurable.
We need, however, a yardstick that makes it possible to compare highly diverse processes, and
this yardstick is time. Without time, change has no meaning. And without change, time would stop.
Time can be conceived as the intervals during which events occur. Just as money permits us to place a
value on both apples and oranges, time permits us to compare unlike processes. When we say that it
takes three years to build a dam, we are really saying it takes three times as long as it takes the earth to
circle the sun or 31 000 000 times as long as it takes to sharpen a pencil. Time is the currency of
exchange that makes it possible to compare the rates at which very different processes play themselves
out.
Given the unevenness of change and armed with this yardstick, we still face exhausting
difficulties in measuring change. When we speak of the rate of change, we refer to the number of
events crowded into an arbitrarily fixed interval of time. Thus we need to define the ‘events’. We need
to select our intervals with precision. We need to be careful about the conclusions we draw from the
differences we observe. Moreover, in the measurement of change, we are today far more advanced
with respect to physical processes than social processes. We know far better, for example, how to
measure the rate at which blood flows through the body than the rate at which a rumour flows through
society.
Even with all these qualifications, however, there is widespread agreement, reaching from historians
and archaeologists all across the spectrum to scientists, sociologists, economists and psychologists, that
many social processes are speeding up – strikingly, even spectacularly.
From: FUTURE SHOCK by Alvin Toffler; Pan Books, London, 1970
There are, of course, limits to the growth although they are difficult if not impossible
to determine or predict exactly. An example of how difficult can be seen if one reads
1 1
the predictions of the ‘Club of Rome’ from the seventies. Nevertheless also nature
teaches us that there are limits.
Size limits of animals
The possible range of size of animals is determined by such factors as the ability of the animal to
support its own weight, the adequacy of blood circulation, metabolic rate, and whether the animal
is homothermic (maintains constant body temperature) or poikilothermic (has a changeable body
temperature, depending on the surroundings).
In the evolution of invertebrates, the development of an adequate circulatory system may have
been the first requirement for growth beyond a certain size, except where the body is filamentous
or greatly flattened. Circulation, together with branchial (relating to gills) respiration, permitted
crustaceans and mollusks to grow to considerable size. Insects are severely limited in their
ultimate size, as their respiratory system consists of tiny tubes that carry air from openings in the
body wall to every cell.
Data compiled for many types of animals indicate that the metabolic rate is a major factor limiting
the size of animals. Within a given group of animals, the metabolic rate (per gram of body weight)
invariably increases with diminishing body size; the higher metabolic rates require
correspondingly higher food intake, faster digestion, and increased circulation. Hypothetically, an
animal might be so small that it would have to eat so much so quickly and expend so much energy
looking for food that it actually could not exist. For example, the smallest mammals, shrews, may
weigh as little as 4 g and must consume nearly their own body weight in food each day; if
deprived of food, a shrew’s high metabolic rate consumes body reserves so fast that it starves to
death in a day or less.
Land animals cannot increase to unlimited size, because their muscles and bones could not support
the weight. As size increases, weight increases more rapidly than strength. The largest living land
animal, the elephant weighs about 3600 kg and the largest dinosaur weighed some 30 000 kg,
whereas the largest of animals, the blue whale, weighing up to 130 000 kg, is aquatic and fully
supported by the surrounding water.
Aspects of Scale-up
There are many different aspects related to scale-up, some of them technical and
others not. Let us shortly look at some of these (in an arbitrary order).
Economical aspect
Financially seen it looks more attractive to build one big apparatus rather than several
parallel ones. The investment costs (Inv) go up with the production capacity (Cap)
and for this it holds roughly the 0.6-rule for process equipment:
Inv ~ Cap 0.6
This can be simply seen by realising that Inv~A (surface area of equipment)~L2 and
Cap~V (volume of equipment)~L3. From these two it follows Inv~Cap2/3.
The exponent 2/3 is called the degression coefficient and changes from case to case.
As long as the degression coefficient is smaller than unity there is an economic
incentive to increase the size of equipment rather than to multiply the equipment. The
degression coefficient depends, of course, on the type of equipment. Also the effect of
‘standard’ sizes is important here. If, for example, a valve is made in different
standard sizes than it could be cheaper to buy two smaller ones than to have a large
one especially made.
1 2
General aspects of scale-up

• Economical aspects:
I ~ C0.6
I ~ A (area)
C ~ V (volume)
• Energy, raw materials, and environment (Bruggink)
• Vulnerability of the production:
- Maintenance
- Sabotage
• Logistics:
- Transportation of parts
- Supply of raw materials
- Discharge of products
• Social aspects:
- Employability, labour conditions (ARBO)
• Mechanical and other technical restrictions
• Start-up, management, maintenance of a plant
• Safety
Scale-up considerations
by number by size enlargement
• fast design procedure: • slower design procedure:
gain of time loss of time
• limited chance on • equipment limitations due to
practical constraints practical constraints
• easier construction: • construction problems:
small mechanical problems mechanical constraints
• good possibility of • limited possibility for
sub-maximal production sub-optimal production
• easier repair/cleaning; • difficult repair;
cheaper spare parts limitation in spare parts
• logistical problems in • logistical problems in
terms of distribution terms of large quantities
• extra risk problems • risk problems
due to interference due to bulk quantities
first size enlargement until constraints, then scale-up by number
1 3
Scale-up/scale-down Scaling factors
1. Scaling Factors
1.1. Introduction
When scaling, two different scales are considered: model scale (M-scale) and
prototype scale (P-scale).
For each physical quantity X playing a role in the process a scaling factor can be
defined:
XP
kX =
XM
For instance for a cylindrical reaction vessel the dimensions height (H) and diameter
(D) are of importance; their scaling factors become:
HP DP
kH = kD =
HM DM
For totally different physical quantities also scaling factors can be defined.
For example for the temperature T in the reaction vessel:
TP
kT =
TM
or for the density ρ of the reaction mixture:
ρP
kρ =
ρM
Three kinds of physical quantities require special attention:
(1) dimensions (lengths): one method of scaling is geometrical similar scaling: all
dimensions change with the same factor. For the cylindrical vessel this means
kH = kD. The scaling factor used for all dimensions (lengths!) is called k: the
scaling factor. Then for all lengths (L) holds kL = k, for all areas (A) kA = k2, and
for all volumes (V) kV = k3 (regardless of form, try to prove this).
(2) material properties, such as density ρ, specific heat capacity cp and heat
conduction coefficient. At both model and prototype scale we prefer to use the
same materials: material identity. At material identity the scaling factor for a
material property is equal to one:
k ρ = 1 k c p = 1 k λ = 1 etc.
Note: sometimes model materials (at M-scale) are used. Then there is no material
identity.
Note: sometimes, physical quantities that are not material properties are treated as
such. For instance in a catalytic reactor the catalyst particles - optimised with
1 1
much effort - are seldom scaled; dimensions and other properties of the catalyst
particles are considered as ‘pseudo’ material properties.
(3) physical constants have by definition a scaling factor equal to one; for us only the
gravity field strength (gravitational acceleration) g and the gas constant R are
possibly of importance:
kg = kR = 1
1.2. Calculations with scaling factors
By introduction of scaling factors scale-up problems become easier to calculate.
Example: flow of a liquid through a smooth, horizontal tube
Process description: As the result of a pressure drop -∆p = pstart - pend a liquid
(density ρ, viscosity η, specific heat capacity cp) flows with a (mean) velocity v
through a smooth, horizontal tube (round profile, length L, diameter D).
-∆p = pstart - pend
pstart pend
ρ, η, cp
D v
L
Aim: topics of interest are the pressure drop required for mass flow Φm, the
dissipated power P and the resulting temperature rise ∆T.
Formulas: in scaling factors:
Φ m = 14 πD 2 ⋅ v ⋅ρ (1) k Φ m = k 2D ⋅ k v ⋅ k ρ
L k
− ∆p = 4 f ⋅ ⋅ 12 ρv 2 (2) k ∆p = k f ⋅ L ⋅ k ρ ⋅ k 2v
D kD
− 14 −1
4 f = 0.316 Re (3) k f = k Re4
ρvD kρk vk D
Re = (4) k Re =
η kη
⎛ −∆p ⎞ k ∆p
P=⎜ ⎟ Φm (5) kP = kΦm
⎝ ρ ⎠ kρ
∆T =
( −∆p ) (6) k ∆T =
k ∆p
ρc p k ρ ⋅ k cp
There are 12 scaling factors:
k Φ m , kv, kρ, kD, k∆p, kf, kL, k cp , kRe, kP, k∆T, kη
and 6 equations.
1 2
For six so-called input variables the scaling factors can be chosen freely; three so-
called output variables are of interest. The remaining three variables are called
auxiliary variables.
Input: kL, kD kρ, k cp , kη kΦm

Output: k∆p, kP, k∆T
Auxiliary: kf, kRe, kv
By taking the logarithm at both sides the six equations in scaling factors can be
converted into six linear equations. These six linear equations with 12 unknowns
can (partially) be solved.
(1) 2 1 -1 1
(2) -1 1 1 1 2 -1
(3) -¼ -1
(4) 1 1 -1 -1 1
(5) -1 1 1 -1
(6) -1 -1 1 -1
kD kL kρ k cp kη k Φ m kRe kf kv k∆p kP k∆T
input auxiliary output
(1) -2 -1 1 -1
(2) ¼ ¼ -¼ -1
(3) -1 -1 1 -1
(4) − 19
4 1 -1 ¼ 7
4 -1
(5) − 19
4 1 -2 ¼ 11
4 -1
(6) − 4
19 1 -2 -1 ¼ 7
4 -1
− 19 −1
1 7
So: k ∆p = k D 4 k L kρ k η4 k Φ4 m
− 19 1 11
k P = k D 4 k L kρ−2 k η4 k Φ4 m
− 19 1 7
k ∆T = k D 4 k L k ρ−2 k c−p1 k η4 k Φ4 m
In case of material identity: kη = kρ = k cp = 1 this simplifies to:
− 19 7
k ∆p = k D 4 k L k Φ4 m
− 19 11
k P = k D 4 k L k Φ4 m
k ∆T = k ∆p
1 3
Scale-up/Scale-down Similarity
2. Similarity
2.1. Introduction in similarity
In the process industry all kinds of physical and chemical processes take place which in
principle can be described mathematically by means of momentum, heat, and mass balances.
The resulting description is usually too complex to allow general analytical solutions.
Although these coupled partial differential equations can be solved numerically after some
computer time, the obtained solutions are very specific and hardly suitable for generalisation
and analysis.
In such cases it might be useful to replace the straightforward process description by a
comparative method in which the process at two different scales (sizes) of the process
equipment are compared. From the conditions for process similarity, scale-up criteria with
respect to that equipment can than be derived.
So scale-up is actually designing using “foreknowledge”; this foreknowledge can be the result
of laboratory experiments or experience with a similar industrial process, possibly at a
different scale.
Considering the two scales we use with scale up (the model scale or M-scale and the
prototype scale or P-scale), then at the M-scale information is available that we want to make
use of at the design scale (P-scale). Naturally, this is only possible if there exists a certain
relation (analogy) between the two processes at M- and P-scale. This analogy will be
indicated as similarity.
Next we will consider what similarity actually is and which conditions must be met to achieve
similarity.
Firstly, two kinds of similarity are distinguished:

(1) apparatus similarity, in which the dimensions (= lengths!) of the apparatus are being
scaled-up with a common scale-up factor; afterwards it is analysed what effect scale-up
has on the processes in the apparatus.
(2) process similarity, in which a certain process variable (production, energy consumption)
is being scaled-up out of process considerations. Process requirements (such as product
quality) result at process similarity in demands for the dimensions of the apparatus. So
now scale-up of the dimensions is secondary.
Process equipment companies that produce relatively simple equipment often prefer apparatus
similarity for production reasons. Also, a series of similar apparatuses (such as pumps)
‘looks’ better. However, often only the original design of such a series of apparatuses may be
optimal and some (much) of that quality has been sacrificed in case of the ‘derived’
apparatuses. This is (one of) the reason(s) why the scale-up factor is never chosen far from
unity and is seldom bridging one decade.
With more complex equipment and for whole production processes (factories) apparatus
similarity is never considered and only process similarity is used. Therefore, from now on, the
word similarity will be used in the meaning of ‘process similarity’.
The concept of process similarity will be introduced with two simple examples.
2 1
Example 1: Introduction to dimensionless variables

Laminar flow of a Newtonian liquid through a tube
p1 p2
-p = p1 – p2
R r 
We will set up a momentum balance over a cylindrical volume element with length L and
radius r (see figure). In steady state this simplifies to a force balance:
0  r 2  (  p)   rx  2 rL
in which (-p) is the pressure drop over the cylinder and rx is the shear stress acting on
the surface of the cylinder:
r  p 
 rx   
2 L 
For a Newtonian liquid (constant viscosity ) holds:
dv x
 rx   
dr
So:
dv x 1  p 
   r
dr 2  L 
Integration with boundary condition: r = R  vx = 0 gives:
1  p  2 2
v x (r)   R  r 
4  L 
Considering this result at two different scales (RP = 2RM or kR = 2), we may get the flow
velocity profiles depicted below.
RP
RM
M
P
2 2
The first step in making these flow velocity profiles congruent is changing to ‘scaled’
dimensionless variables; instead of external units, such as m and m/s for r and vx, we will
use model parameters as ‘internal units’.
An obvious choice for scaling the radius r is the ‘internal unit’ R. Instead of r we now use
r

R
For the velocity vx the choice is less obvious. We choose as ‘internal unit’:
1  p  2
v x,max  v x (0)   R
4  L 
Instead of vx(r) we then use
vx
x    
v x,max
Using the dimensionless variables x and  the following velocity profile applies:
x     1   2
from which the variables  and (-p/L) have totally disappeared! This is an extraordinary
situation!
Conclusion: In case of laminar flow through a tube there is always (dynamic) process
similarity; the only demand is that the flow stays laminar and thus that there is no transition in
regime (to turbulent flow).
Dynamic process similarity means: congruent dimensionless velocity profiles.
Example 2: Introduction to constant characteristic dimensionless groups

First order reaction in an ideal tube reactor
In a first order reaction (r = krc) a dissolved component (concentration c) is reacting away.

The solvent with the component flows with a (high) velocity v0 through the tube.
dx
v0 r = krc A(m2)
x
0 L
L
Over the volume element Adx a mass balance of the reacting component can be
formulated:
2 3
0  v 0 A c x  v 0 A c x dx  k r cAdx
 V   dV
volume flow volume
So:
kr
dc    c  dx
v0
With the boundary condition c = c0 at x = 0 integration then gives:
 k 
c  c0  exp   r x 
 v0 
In order to have congruent profiles with scale-up (in this case concentration profiles for
so-called chemical similarity) we change to dimensionless variables. Scaling, using x = L
and c = c0, replacing x with  = x/L and c with  = c/c0 yields:
 kL 
 ( )  exp   r  
 v0 
(Chemical) process similarity is attained provided that DaI = krL/v0 is constant (equal at
both scales). In that case the dimensionless concentration profiles at both scales are
congruent. It does not matter how we keep DaI constant; kr (model system, temperature)
and/or v0 and/or L can be varied.
DaI is the Damköhler I number. DaI can be interpreted as the ratio between the
characteristic times of two sub-processes:
sub-process characteristic time

convection : the ‘residence time’ tc = L/v0
chemical reaction : the ‘reaction time’ tr = 1/kr
tc
DaI 
tr
In case of process similarity we have congruent dimensionless profiles and zero, one, or
more than one dimensionless number have to be chosen equal at both scales.
Warning
Do not assume that in order to get chemical process similarity (congruent dimensionless
concentration profiles) only DaI has to be kept constant. In the example a number of
additional assumptions have been made that do not have to be generally valid:
 chemical: one component, first order reaction
 thermal: no heat effects, no temperature difference
 dynamic: ideal tube reactor, constant velocity
2 4
2.2. Process similarity levels
After these two examples the concept of similarity will be treated in a more general way.
First a definition:
Similarity between model and prototype means that the deformation, flow,
temperature and concentration profiles of model and prototype are
congruent in dimensionless form.
From this definition we can conclude directly that there exist more levels of process
similarity. Moreover, there is also a certain hierarchy; each underlying form of similarity
assumes the presence of the one above.
We distinguish them as:
g  g: geometric similarity
shape
 mechanical similarity
s d s: static (mechanical) or d: dynamic (mechanical) similarity
deformation profiles velocity profiles
 t: thermal similarity
t temperature profiles
 c: chemical similarity
concentration profiles
c
As process engineers we will concentrate on the right-hand branch (c 
t  d  g); the left-hand branch is more useful for construction engineers (deformation
mechanics, theory of material strength).
To derive the conditions for the different levels of process similarity we use momentum, heat,
and mass balances; these balances will be formulated such that the analogy between these
three types of transport is maximal.
For each extensive physical quantity X one can formulate a balance over a self selected part
of space (volume). In general a balance consists of the following parts:
[accumulation] = [net transport] + [net production]

‘in’ ‘to’ ‘in’
or further specified:
 transport   molecular 
[accumulation]       transfer    production 
 by convection   transport 
In case of a microbalance (a balance over such a small volume element that all physical
properties within that volume can be assumed constant) we can write in general:
 
t
 
 div   v  div   grad   a       G
accum. convective molecular transfer production
transport transport
2 5

where  is the extensive physical quantity X per volume ( = X/V), v is the velocity of the
medium in the volume element and k is the momentum, heat or mass transfer coefficient.
For momentum, heat and mass this is:
momentum heat mass dimension


 v cpT ci [X]/L3
  = / a/(cp) i L2/T
k ½ f <v> h/(cp) ki L/T
G* -grad(p) r·H r [X]/L3T
* more production terms are possible
Filled out in the general microbalance equation the following balance equations are found:

v      
momentum:
t
     
 div v  v  div   grad v   12 f v  a    v  grad(p)  g
 

c p T     h
heat:
t

 div c p T  v  div 
 c p
 grad  c p T   
 c p h  p 
 a   c T  r  H
 
ci 
component:
t
 
 div ci  v  div i  grad ci   k i  a m   ci  r
All physical quantities refer to the medium (gas, liquid, and solid) in the volume element:
 kg / m3 density

v m/s velocity
t s time
 Pa s viscosity
a 2
m /m 3
specific exchange area (area per volume) for 
a, ah, am m2 / m3 specific exchange area for momentum, heat and mass
f - Fanning friction factor
p Pa pressure
g N / kg gravity field strength
cp J / K kg specific heat capacity
T K temperature
 W/Km heat conduction coefficient
h W / K m2 heat transfer coefficient
r mol / m3s reaction rate
H J / mol heat of reaction
3
ci mol / m concentration component i
i 2
m /s diffusion coefficient component i
ki m/s mass transfer coefficient component i
For simplification we assume the material properties , , cp, , i to be constant in the

macro-environment of the volume element as well. Then we find:
2 6
momentum (z-component):
v z   1 p
t
 

 div v z  v   div grad  v z   12 f v  a     v z  
 z
 gz
If the z-direction is assumed along the gravity field then the production term due to gravity is
g.
heat:
T   H
t

 div T  v 
c p

 div grad  T  
h
c p
 a h  T  r 
c p
component i:
ci 
t
 
 div ci  v  i  div grad  ci   k i  a m   ci  r
To derive the requirements for full process similarity, again we first switch over to
dimensionless variables, by scaling using characteristic values of the variables.
From the macro-environment of the process one dimension is chosen as ‘characteristic length’
L0, one velocity is chosen as ‘characteristic velocity’ v0 and a ‘characteristic time’ t0 is
defined as
L0
t0 
v0
Using these ‘characteristic’ variables the momentum balance can be made dimensionless
using the following dimensionless variables:
z vz v0 t
 z  =
L0 v0 L0
Filled out in the momentum balance this gives:
v02 z  v  1 1 p

L0 
v2
L0

L0 

  0  div ' z   20   div grad '  z   v 02 12 f  a  z 
L0  
g
or
z     p  gL


 div ' z   
v 0 L 0
 div grad '  z   12 f a  L 0 z   2   20
  v 0  v 0
Dynamic similarity
Dynamic similarity requires the partial differential equation including boundary conditions in
dimensionless form to be equal at both scales. Therefore, the following dimensionless groups
must be constant:
v 0 L 0 convection
Reynolds: Re  
 shear
2 7
transfer
Stanton: St   12 f 
convective transport
p production
Euler: Eu  2 
v0 convective transport
v02 convection
Froude: Fr  
gL0 production
From geometric similarity it follows that aL0 is already constant. The conditions for Eu and
St (momentum) result from the boundary conditions.
Not all dimensionless groups are equally important in all cases:

 if there is only one phase then there is no momentum transfer and f does not play a role,
 if the flow is perpendicular to the gravity field then g and thus Fr does not play a role.
Thermal similarity 
In the heat balance the velocity v is explicitly present in the convection term; thus for thermal
similarity at least dynamic similarity is needed.
Choosing a ‘characteristic temperature’ T0 and using  = T/T0 as dimensionless temperature
the heat balance expressed in dimensionless variables reads:
v 0   T  H
T0
L0 
 T0 v 0
1
L0
 
div '   20
L 0 c p
div grad '     T0
h
c p
a h   r 
c p
or
   rL H

 
 div '  
c p v 0 L0
div grad '    
h
c p v 0
a h L 0   0
T0 v 0 c p
For thermal similarity even more dimensionless groups have to be kept constant:
 conduction
Péclet (heat): Pe h1  
c p v 0 L0 convection
h heat transfer
Stanton (heat): St h  
c p v 0 convection
L H production
Damköhler (III): Da III  r 0 
T0 v 0 c p convection
Again ahL0 is constant in case of geometric similar heat transfer from the equipment wall.
Chemical similarity
Finally in the component balance the reaction rate r is explicitly dependent on temperature;
therefore chemical similarity demands thermal similarity.
Choosing a ‘characteristic concentration’ c0 and using i = ci/c0 as dimensionless
concentration the component balance in dimensionless variables reads:
2 8
v 0  i 
c0
L0 
 c 0 v 0
1
L0
  c
div '  i   20 idiv grad '   i   c0 k ia m  i  r
L0
or
 i  

  v0 L0
k
v0
rL
 div '  i   i div grad '   i   i a m L 0  i  0
c0 v0
In order to have chemical similarity again even more dimensionless groups have to be kept
constant:
i diffusion
Péclet (mass): Pe m1  
v 0 L0 convection
k mass transfer
Stanton (mass): St m   and nameless: amL0
v0 convection
rL production
Damköhler 1 (I): Da I  0 
c0 v0 convection
In most cases mass transfer occurs over a dynamic interface (of a drop or bubble). In that case
geometric similarity of the equipment does not imply that amL0 is kept constant.
The dimensionless groups found this way were the result of making balance equations
dimensionless by scaling both the place coordinate and the balanced quantity. Time was
scaled by residence time or characteristic convection time L0/v0. When, as a first step, we do
not scale time, the balance equation (e.g. for heat) takes the following form:
  1  H
t
v
L0
 
  0 div '   2
L 0 c p
div grad '    
h
c p
a h   r 
c p T0
Each term in the equation can be considered as a “scaled” rate with dimension 1/time.
Because the place coordinate and balanced quantity were made dimensionless by
characteristic quantities, the parameter groups in each term may be considered to be
characteristic “scaled” rates [s-1] or inverse characteristic times [s].
We recognise:
- characteristic convection rate (dilution rate) v0/L0 or residence time L0/v0.
 / c p L20
- characteristic thermal diffusion rate or time
L20 a
ha c
- characteristic heat transfer rate h or time p
c p ha h
r  H c T
- characteristic heat production rate or time p 0
c p T0 r  H
1
The several Damköhler groups are used when describing chemical reactions.
Da II = Da I × Pem; Da IV = Da III × Peh; Da V = Re.
2 9
It now becomes clear that the dimensionless groups that were found are the ratio of any
characteristic “scaled” rate or time and the characteristic convection “scaled” rate (v0/L0) or
time (L0/v0). E.g.:
1  / c p ( / c p ) / L0 2 rconduction L /v 
Péclet (heat): Pe     2 0 0  convection
rconvection L0 /( / c p ) conduction
h
v 0 L0 v 0 / L0
k L0 / v0 
Stanton (mass): St m    convection
v 0 1/ ka m mass transfer
This gives a clear physical insight in the importance of these groups, their meaning within
balance equations and the potential value of these times (or rates) to characterise
subprocesses.
General conclusions about similarity are:
 At each level of process similarity it is possible to indicate the conditions: equal

dimensionless numbers at M- and P-scale
 Soon the number of conditions is that large that these conditions can only be satisfied with
a scale-up factor equal to one
 Fortunately not all dimensionless numbers are important at all scales
 A method is needed to determine which dimensionless groups are really important and
which are allowed to vary (hopefully at a modest scale). Regime analysis provides such a
method.
Finally: only seldom full process similarity is possible (or desirable). If full similarity is not
possible then partial process similarity must be accepted. The dimensionless groups, which
have to be kept constant at both scales, form together the scale-up criterion.
2 10
Scale-up/Scale-down Dimensional analysis
3. Dimensional analysis
When a process is too complex to be able to write down all the relevant balance
equations, it is still possible to find the relevant dimensionless groups from a list of all
relevant quantities. Dimensional analysis uses the dimensions of these quantities to
determine all relevant dimensionless groups. Knowing these groups allows us to design
proper experiments on model scale and use the results to write down scale criteria.
3.1. Buckingham Π-theorem
Dimensional analysis is used to derive how many and which dimensionless groups
(dimensionless numbers) are required to describe a given process.
Starting-point is a list of all relevant physical quantities; the result is a list of
dimensionless groups (numbers) * .
We start with how many: How many dimensionless groups can be derived from a given
list of relevant physical quantities?
Basically this is given by a theorem known as the Buckingham Π-theorem:
Every relation between n physical quantities can be reduced to a

relation between m = n - r mutual independent dimensionless groups, in
which r is the rank of the dimension matrix as can be formulated for the
given physical quantities.
A more informal expression of the Π-theorem being sufficient in most cases is:
A physical system described by n physical quantities of which the

dimensions are set by r primary or basic dimensions can be described
by (n-r) dimensionless (Π -)groups.
To clarify the concepts of the Π-theorem the following example is used.
Example 1
Ffr ρ,η
Consider the friction force F experienced
by a sphere with diameter D, while being
D v
moved with a velocity v through a
(stagnant) liquid with density ρ and
viscosity η. Ffr
Starting-point: a list with all relevant physical quantities with their dimensions
(L = length, M = mass, T = time)
*
Generally referred to as (characteristic) number. However, dimensionless groups are quantities or
variables with dimension [1], not numbers.
3 1
a) geometry
D diameter of the sphere L
b) material properties
ρ density of the liquid ML-3
η viscosity of the liquid ML-1T-1
c) process variables
F friction force MLT-2
v velocity of the sphere LT-1
The friction force F is the target variable, or target quantity: F = F(D; ρ, η; v).
The physical relation comprises five physical quantities: n = 5; the number of basic
dimensions required is three (length, mass, and time): r = 3. So there are m = n - r = 2
independent dimensionless groups.
The dimensions of physical quantities can be given in the following formal notation:
[F] = M1L1T-2
[ρ] = M1L-3T0
[v] = M0L1T-1
[η] = M1L-1T-1
[D] = M0L1T0
The first line reads as ‘the dimension of (friction) force is mass to the power of 1, length
to the power of 1, time to the power of -2’.
All information is obviously contained in the exponents 1,1, and -2. It can be proven that
together these exponents are presented by vectors in a vector space. It is convenient to
summarise the dimensions of the physical quantities in a so-called dimension matrix: at
the columns the physical quantities; at the rows the basic dimensions. This results in a
3x5-dimension matrix.
F ρ v η D
M = mass 1 1 0 1 0
L = length 1 -3 1 -1 1
T = time -2 0 -1 -1 0
The rank (r) of this matrix is the order of the largest sub-determinant not equal to zero we
can derive from this matrix by removing columns and/or rows. Of course the rank r is at
most equal to the smallest size of the matrix.
In our example we remove the F- and ρ-column; the resulting square matrix has the
following determinant
det = 0⋅[(-1)⋅0 - 1⋅1] - 1⋅[1⋅0 - 1⋅(-1)] + 0⋅[1⋅(-1) - (-1)⋅(-1)]

= -1 (≠ 0) ⇒r=3
Usually the rank r is equal to the number of chosen and used basic dimensions. If we add
a row θ (= temperature) with only zeros in our example, the rank r does not change.
To expose the background of the Π-theorem the following formal reasoning is used in our
example:
3 2
[F] = M1L1T-2
[ρ] = M1L-3T0
[v] = M0L1T-1 (system 1)
[η] = M1L-1T-1
[D] = M0L1T0
To stress the importance that we deal with exponents and that vectors in a linear vector
space represent these exponents we formally write the following system completely
equivalent to system 1:
log [F] = 1 log M + 1 log L - 2 log T

log [ρ] = 1 log M - 3 log L + 0 log T
log [v] = 0 log M + 1 log L - 1 log T (system 2)
log [η] = 1 log M - 1 log L - 1 log T
log [D] = 0 log M + 1 log L + 0 log T
Note: system 2 contains exactly the same information as system 1 does.
In matrix notation:
⎛ log[F] ⎞
T
[F] [ρ] [v] [η] [D]
⎜ log[ρ] ⎟ T
⎜ ⎟ ⎛ log M ⎞ ⎛ 1 1 0 1 0⎞
⎜ log[v] ⎟ = ⎜ log L ⎟ ⎜ 1 −3 1 − 1 1 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
log[ η] ⎜ log T ⎟ ⎜ − 2 0 −1 − 1 0 ⎟
⎜ ⎟ ⎝ ⎠ ⎝ ⎠
⎜ log[D] ⎟
⎝ ⎠
Here the dimension matrix appears with all relevant information.
The next step is partitioning; we divide the dimension matrix into a square core matrix
and a rest matrix:
Core matrix:
T T
⎛ log[F] ⎞ ⎛ log M ⎞ ⎛ 1 1 0⎞
⎜ log[ρ] ⎟ = ⎜ log L ⎟ ⎜ 1 −3 1 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ log[v] ⎟ ⎜ log T ⎟ ⎜ −2 0 −1 ⎟
⎝ ⎠ ⎝ ⎠ ⎝ ⎠
Rest matrix:
T
T ⎛ log M ⎞ ⎛ 1 0⎞
⎛ log[η] ⎞ ⎜ ⎟ ⎜ ⎟
⎜ log[D] ⎟ = ⎜ log L ⎟ ⎜ −1 1 ⎟
⎝ ⎠ ⎜ log T ⎟ ⎜ −1 0 ⎟
⎝ ⎠ ⎝ ⎠
In the core matrix (determinant ≠ 0) the dimensions of three physical quantities are
expressed in three basic dimensions. By inversion of the core matrix (solving the system
of three equations in three unknowns log M, log L, and log T) we find the dimensions of
the basic quantities expressed in the dimensions of the chosen physical quantities.
3 3
Inversion gives:
T T −1
⎛ log M ⎞ ⎛ log[F] ⎞ ⎛ 1 1 0⎞
⎜ log L ⎟ = ⎜ log[ρ] ⎟ ⎜ 1 −3 1 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ log T ⎟ ⎜ log[v] ⎟ ⎜ −2 0 −1 ⎟
⎝ ⎠ ⎝ ⎠ ⎝ ⎠
T
⎛ log[F] ⎞ ⎛ 23 1
2 2⎞
1
= ⎜⎜ log[ρ] ⎟⎟ ⎜− 1 − 1 − 1 ⎟
⎜ 2 2 2⎟
⎜ log[v] ⎟ ⎜ −3 −1 − 2 ⎟
⎝ ⎠ ⎝ ⎠
Using this into the rest matrix gives:
T T
T ⎛ log M ⎞ ⎛ 1 0 ⎞ ⎛ log[F] ⎞ ⎛ 23 1
2 2 ⎞⎛ 1 0⎞
1
⎛ log[η] ⎞ ⎜ ⎟ ⎜ −1 1 ⎟ = ⎜ log[ρ] ⎟ ⎜ 1 1 ⎟⎜ ⎟
⎜ log[D] ⎟ = ⎜ log L ⎟ ⎜ − 2 − 2 − 2 ⎟ ⎜ −1 1 ⎟
1
⎜ ⎟ ⎜ ⎟
⎝ ⎠ ⎜ ⎟ ⎜ −1 0 ⎟ ⎜ log[v] ⎟ ⎜ ⎟⎜ ⎟
⎝ log T ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ − 3 − 1 −2 ⎠ ⎝ − 1 0 ⎠
T
⎛ log[F] ⎞ ⎛ 12 2⎞
1
= ⎜ log[ρ] ⎟ ⎜1 −1⎟
⎜ ⎟ ⎜2 2⎟
⎜ log[v] ⎟ ⎜ 0 −1 ⎟
⎝ ⎠ ⎝ ⎠
The dimensions of the physical quantities from the rest matrix have now been expressed
in the dimensions of the physical quantities from the core matrix.
[η] = [ F] [ρ] [ v ]0
1 1
2 2
[ D] = [ F] [ρ]− [ v ]−1
1 1
2 2
In this way we find two dimensionless numbers (= dimensionless groups):

1 1 1
F 2 ⋅ρ 2 F2
and 1
η ρ2 ⋅ D ⋅ v
Since squared dimensionless groups are still dimensionless,
F ⋅ρ F
Π1 = Π2 =
η2 ρ ⋅ D2 ⋅ v2
Buckingham named his dimensionless groups ‘Π-numbers’; hence the name Π-theorem
of Buckingham.
Some dimensionless groups are characteristic and have been named. These are listed in all
sorts of handbooks. In Π2 we can thus recognise the Newton number (abbreviated: Ne).
Eliminating the target variable F from Π1 so that it is present in only one dimensionless
group, we find:
ρ⋅v⋅D
Π1' = ( Π1 / Π 2 ) 2 =
1
3 4
In Π1' we recognise the Reynolds group (abbreviated: Re).
Result
Instead of a five-dimensional relation F = F(D; ρ, η; v) we now have a two dimensional
relation
F ⎛ ρvD ⎞
Ne = f (Re) or =f⎜ ⎟
ρD v
2 2
⎝ η ⎠
in dimensionless form.
3.2. Independence of dimensionless groups
In the former we derived a new dimensionless group Π1' from the dimensionless groups
Π1 and Π2.
In general it can be stated that:
If A and B are dimensionless groups (dimensionless numbers), then

AnBm for every n,m ∈ Z are also dimensionless groups.
Of course not all (infinitely many) dimensionless groups you can derive this way are
independent.
Two dimensionless groups A and B are independent if from

AaBb ≡ 1 it follows that a = b= 0.
In our example:
a b
⎛ F ⎞ ⎛ ρvD ⎞
⎜ 2 2⎟ ⎜ ⎟ ≡1
⎝ ρD v ⎠ ⎝ η ⎠
(Note: ‘≡’ means for all values of F, ρ, D, v, and η)
Working this out gives:
Faρ− a + b v −2a + b D−2a + bη− b ≡ 1
From the factors Fa and η-b it follows directly that a = b = 0 since the relation must be
valid for all values of F and η.
Dimensionless groups containing a unique physical quantity are always independent from
the other dimensionless groups. So Ne and Re are independent because only Ne contains
the physical quantity F and only Re contains the physical quantity η.
3 5
Other examples:
Example 2
Dimensionless groups:
vd v2 gd 3
Re = Fr = Ga = 2
ν gd ν
Dimension matrix:
v d ν g v d ν g
Re 1 1 −1 0 Re 1 1 −1 0
~ ~
Fr 2 −1 0 −1 Fr Re −2 0 −3 2 −1
Ga 0 3 −2 1 Ga 0 3 −2 1
v d ν g v g d ν
Re 1 1 −1 0 Re 1 0 1 −1
~
Ga 0 3 −2 1 Ga 0 1 3 −2
FrGa Re −2 0 0 0 0 FrGa Re −2 0 0 0 0
Re, Fr, and Ga are dependent because
Fr1Ga1Re-2 ≡ 1 ⇒ FrGa = Re2
Example 3
The dimensionless groups
Δp vd d
Eu = 2 , Re = and
ρv ν L
are independent because each contains an unique physical quantity (respectively Δp,
ν, and L).
Dimension matrix:
Δp ν L ρ v d
Eu 1 0 0 −1 −2 0
Re −1 0 1 0 0 −1 −1
−1
⎛d⎞
⎜ ⎟ 0 0 1 0 0 −1
⎝L⎠
3 6
Since now we know how many dimensionless groups there are for a given situation
(number of quantities), the next step is to examine how they can be determined.
There are three methods to obtain the dimensionless groups (dimensionless numbers)
from the (list of) physical quantities:
(1) the reduction method of Gauss-Jordan

(2) the method of Gukhman
(3) the method of Rayleigh
Of these methods only the first two will be discussed.
List of relevant physical quantities
In order to work according to method (1), (2), or (3) we only need to have a complete list
of physical quantities describing the process to be studied and to make a choice which
basic dimensions will be used.
The list of physical quantities must be complete, otherwise we might miss a relevant
dimensionless group:
v2
without g (gravity field strength) no Fr = .
gd
To structure the search for relevant physical quantities we distinguish:

(1) geometric quantities
sizes, radials, angles, space angles, etc.
(2) material properties
density, viscosity, specific heat, thermal conduction coefficient, etc.
(3) process variables
concentration, pressure drop, rotation frequency, flow rate, etc.
(4) physical constants
gas constant, gravity field strength, etc.
As basic dimensions usually the basic dimensions of the SI-system are chosen:
M = mass L= length T = time θ = temperature
Sometimes also Q = heat is chosen as a basic dimension, for example in a heat problem in
which the conversion of mechanical energy into heat is negligible.
(1) Reduction method of Gauss-Jordan
This method resembles the theory treated in the previous section on the Π-theorem of
Buckingham. It consists of five steps:
(1) formulation of the dimension matrix

(2) removal of simplexes
(3) ordering of the columns in an appropriate way
(4) splitting up and solving
(5) formation of known dimensionless numbers
These steps will be explained on the basis of an example
3 7
Pressure drop over a tube.

What pressure drop (-Δp) is required to press a liquid with density ρ and viscosity η
through a circular tube with length L and diameter D at an average velocity v?
List of quantities
Δp = f (D, L ; ρ, η ; v )
Step 1: formulation of the dimension matrix
Δp v D L ρ η
M 1 0 0 0 1 1
L −1 1 1 1 −3 − 1
T −2 − 1 0 0 0 − 1
Note: the choice of the order of columns is still free.
Step 2: removal of simplexes

Identical columns in the dimension matrix point out physical quantities with the same
dimensions. In our example those are the D- and L-column. From L we form the
simplex
L
Π1 =
D
D is maintained in the dimension matrix (L is removed). Now D plays the role of

characteristic length.
Step 3: ordering of the columns in an appropriate way

The dimension matrix is ordered in such a way that the core matrix contains mostly
ones in the diagonal and mostly zeros outside the diagonal.
Physical quantities in the core matrix can end up in more than one dimensional group.
The physical quantities in the rest matrix automatically end up in only one
dimensional group. In case of the target variable (here Δp) that is a requirement.
ρ D v Δp η
M 1 0 0 1 1
L −3 1 1 − 1 − 1
T 0 0 − 1 −2 −1
Step 4: splitting up and solving

Solving is successively replacing ‘external’ dimensions by ‘internal’ dimensions
through elimination.
ρ D v Δp η
M 1 0 0 M 1 1 M
[Δp]=
L −3 1 1 L −1 −1 LT 2
T 0 0 −1 T −2 −1
3 8
The first column can be used to eliminate M:

[ρ]L3
[ρ] = ML-3 ⇒ M = [ρ]L3 [Δp]=
LT 2
ρL3 D v Δp η Δp η
3
M 1 0 0 ρL 1 1 ρ 1 1 [ρ]L2
⇒ [Δp]=
L 0 1 1 L −1 − 1 L 2 2 T2
T 0 0 −1 T −2 − 1 T −2 −1
The second column can be used to eliminate L:

[ρ][D 2 ]
[D] = L ⇒ L = [D] [Δp]=
T2
ρL3 D v Δp η
M 1 0 0 ρ 1 1
⇒ idem
L 0 1 1 D 2 2
T 0 0 −1 T −2 −1
Finally the third column can be used to eliminate T:

-1 -1 [ρ][D2 ]
[v] = LT ⇒ T = [D][v] [Δp]= 2
[D ][v]−2
D
ρL3 D Δp η
v Δp η
M 1 0 0 ρ 1 1 ⇒ ρ 1 1 [ρ][v 2 ]
[Δp]=
D 0 1 [1]
L 0 1 0 D 2 2
T 0 0 1 D / v −2 −1 v 2 1
Results:
Δp
[Δp] = [ρv2] ⇒ Eu = = Π2
ρv 2
ρvD
[η] = [ρvD] ⇒ Re = = Π3
η
So: Eu = f(Re; L/D)
Step 5: formation of known dimensionless groups

In this example this is not needed as we already found known dimensionless groups.
If you do, be careful not to introduce dependencies.
3 9
Once you know how the method works, it can of course be done faster:
ρ D v Δp η
M 1 0 0 1 1
(sweep with row 1: exchange M with ρ)
L −3 1 1 −1 −1
T 0 0 − 1 −2 − 1
x D v Δp η
ρ 1 0 0 11
(sweep with row 2: exchange L with D)
L 0 1 1 2 2
T 0 0 −1 −2 −1
x x v Δp η
ρ 1 0 0 1 1
(sweep with row 3: exchange T with v)
D 0 1 1 2 2
T 0 0 −1 − 2 − 1
x x x Δp η
ρ 1 . . 1 1
D . 1 . 0 1
v . . 1 2 1
(2) The method of Gukhman
Again an example is used to illustrate the method.
Stirring power
What power input P is required to have a stirrer with diameter d rotating with a
frequency n in a liquid with density ρ and viscosity η? Also gravity is of importance.
P = f(n; g; η, ρ; d).
First simplexes are removed (already done: d/D, H/D etc.)
Now for each basic dimension an auxiliary quantity with that dimension is introduced:
mass M0 length L0 time T0
Using these auxiliary quantities the dimensionless groups can directly be constructed
from the physical quantities:
2 -3 T03
[P] = ML T thus P ⋅ is dimensionless.
M 0 L20
3 10
Likewise:
1 L30 L 0 T0 T03 T02
d⋅ ρ⋅ η⋅ n·T0 P⋅ g⋅
L0 M0 M0 M 0 L20 L0
(1) (2) (3) (4) (5) (6)

1
Now (1): d ⋅ is used to eliminate L0 in all:
L0
1 T P T03 g 2
ρd 3 ⋅ ηd ⋅ 0 n·T0 ⋅ ⋅ T0
M0 M0 d2 M0 d
(2) (3) (4) (5) (6)
1
Now (2) ρd 3 ⋅ is used to eliminate M0 in all:
M0
η P 3 g 2
⋅ T0 n·T0 5 ⋅ T0
⋅ T0
ρd 2 ρd d
(3) (4) (5) (6)
Finally (the last step) a variable (4): nT0 is used to eliminate T0 in all:
η P g
2 ⋅1 ⋅1 ⋅1
ρnd ρn 3 d 5 n2d
P ⎛ ρnd 2 nd 2 ⎞
So: =f⎜ , ,... ⎟
ρn 3d 5 ⎝ η g ⎠
Po = f(Re; Fr; …)
It appears to be convenient to use the division in type of quantities as a guideline for the
order to use in eliminating the auxiliary quantities. The quantities can be ordered as
follows:
- geometry (dimensions)
- material properties
- process variables
- target variable
The elimination starts from the top, using in each group of variables the convenient
quantities first. Normally, that is geometry (dimension length) first. Then mass and finally
time. In the last step a convenient variable is used to ensure that a variable is present in
the final groups. Note that the last step may also be seen as determining ratios of times. In
a table it looks like:
3 11
1
d⋅
L0 *
L30 1
ρ⋅ ρd 3 ⋅
M0 M0 *
L T T η η
η⋅ 0 0 ηd ⋅ 0 ⋅ T0 ⋅1 Π −21
M0 M0 ρd 2 ρnd 2
n·T0 n·T0 n·T0 *
T03 P T03 P 3 P
P⋅ ⋅ 5 ⋅ T0 ⋅1 Π1
M0L0 2 2
d M0 ρd ρn 3 d 5
2
T g 2 g 2 g
g⋅ 0 ⋅ T0 ⋅ T0 ⋅1 Π 3−1
L0 d d n2d
* stands for ‘used for elimination’
So: Po = f(Re; Fr; …)
… stands for all simplexes describing the geometry of stirrer and vessel: D/d, H/d, …
3 12
Scale-up/Scale-down Regime analysis
4. Regime analysis
4.1. Scale-up criteria
Attempts to scale a process fully similar often fail due to the large number of
dimensionless groups that have to be kept constant.
Reasons to partly give up full process similarity can be of
(1) mathematical,
(2) technological or
(3) economical
nature.
In the first case the conditions, which constrain the system by keeping the
dimensionless groups constant, are mathematically contradictory.
In the second case certain technological boundaries are exceeded (gas velocities near
the speed of sound, vibrations near the characteristic frequencies, etc.).
In the third case the design is becoming too expensive (special design, extreme costs
of transportation, high costs for foundations or safety).
In all cases we have to give up full process similarity and have to be satisfied with
partial similarity: not all dimensionless groups required for full similarity will show
in the scale-up criterion, but some will and some do not. Clearly we have a problem of
choice, entering the field of regime analysis.
Example: mixed vessels
In the process industry mixed vessels are used for all sorts of physical and
chemical processes in which a homogeneous reactor composition is essential.
These processes run at a certain rate; the system (vessel plus stirrer) is chosen
such that the stirring process is considerably faster than the reaction rate to ensure
a homogeneous reactor content.
Process description
To describe the process, use is made of a drawing and a list of relevant physical
quantities.
P
(a) geometrical:
D, H, d, h, … L n g
(b) material properties:
ρ density ML-3
η viscosity ML-1T-1
H ρ,η
(c) process variables:
n stirrer rotation frequency T-1
P power input ML2T-3 h
(d) physical constants: d
-2
g gravity field strength LT D
The stirrer diameter d is the characteristic length; the power input P is the target
variable.
4 1
Dimensional analysis yields the following result:
⎛ D H h ⎞
Po = f ⎜ Re, Fr; , , ,... ⎟
⎝ d d d ⎠
P
where Po is the ‘power number’: Po = ,
ρn 3 d 5
ρnd 2
Re is the Reynolds number: Re = , and
η
n2d
Fr is the Froude number: Fr = .
g
For scaling with full process similarity the following should hold:
(1) geometric similarity: for all dimensions (lengths) one scale factor applies:
kD = kH = kd = kh = … = k
(2) material identity: the same components at model and at prototype scale:
kρ = kη = 1
(3) dynamic similarity: the numbers of Froude and Reynolds are constant:
kRe = 1,
kFr = 1
Because the gravity field strength can not be varied, inevitably kg is also equal to
one. Working this out gives us:
kρ
k Re = 1 ⇒ k n k d2 = 1 ⇒ k n k d2 = 1 (a)
kη
1
k Fr = 1 ⇒ k 2n k d = 1 ⇒ k 2n k d = 1 (b)
kg
The only solution of the system (a) and (b) is kn = kd = 1. So it is not possible to
keep Re and Fr simultaneously constant with scaling.
If we still want to scale-up (and we do want to), then we will have to restrict our
scale-up criterion, let go of one of the two dimensionless numbers (or both), and be
content with subprocess similarity.
In this example there are three possibilities, because keeping bóth Re ánd Fr constant
excludes scaling.
Fr Situation a: Fr constant
Re yes no This scale-up criterion is often used with surface aeration of
yes b wastewater, where a stirrer at the air-water surface is used to aerate
the liquid.
no a c
kFr =1 ⇒ kdkn2 = 1 (kd = k) ⇒ kn = k-1/2
What will be the consequences for some other important process variables?
4 2
(1) The mixing time tm: it can be derived that mixing time is inversely proportional to
stirrer rotation frequency:
1 1
tm ~ ⇒ k t m = k n−1 = k 2
n
Some statements: ‘mixing at large scale is always more difficult than at small
scale’ or ‘at laboratory scale there is seldom a mixing problem’
(2) The tip speed vtip: vtip = πnd
1
k vtip = k n k d ⇒ k vtip = k 2
In case of shear sensitive suspensions the tip speed can not increase too much: e.g.
animal cells can be damaged by stirrers.
(3) The power input, often expressed as power input per reactor volume: P/V.
For sufficiently fast rotating stirrers (full turbulence!) it holds that Po = constant.
So:
P = Po ⋅ρn 3d5 ; V = 14 πD 2 H
P n 3d 5 k 3n k 5d
~ ⇒ kP / V = with k d = k H = k D = k
V HD 2 k H k 2D
k P / V = k 3n k 2 = k 2
1
Obviously the ratio P/V corresponds to a cost aspect.
Situation b: Re constant
This scale-up criterion can be considered in case of a fully immersed stirrer and the
use of baffles in order to keep the liquid surface more or less level so that Fr does not
play an important role.
The effects of kRe = 1 are:
kRe = 1 ⇒ kd2kn = 1 ⇒ kn = k-2
and furthermore: k t m = kn-1 ⇒ k t m = k2

k v tip = knk ⇒ k v tip = k-1
kP/V = kn3k2 ⇒ kP/V = k-4
With scaling one hopes that the scaling factors do not become too large or too small.
Ideally this means that − 1 < n < +1 for k x = k n
The scale-up criterion ‘Re constant’ does by no means comply with this ideal.
Situation c: neither Fr nor Re constant

In this situation many combinations are possible, all to be written as FraReb = constant
for different values of a and b.
4 3
c1: a = 2 b = −1 ⇒ k n = 1
k Re = k 2 k Fr = k k vtip = k k P / V = k 2 k P = k 5
k t m = 1 (!): scale-up criterion is ‘tm (mixing time) constant’.
Note the effect of this criterion on the power input.
c2: a = 1 b =1 ⇒ k n = k −1
k Re = k k Fr = k -1 k t m = k k P/V = k -1 k P = k 2
k v tip = 1 (!): scale-up criterion is ‘vtip (tip speed) constant’.
− 23
c3: a = 4 b =1 ⇒ kn = k
4
− 13 2 1
k Re = k 3 k Fr = k k tm = k 3 k vtip = k 3 kP = k3
kP/V = 1 (!): scaling criterion is ‘P/V constant’
This scaling criterion is very close to the scaling criterion ‘Fr constant’ and is very
popular in scaling.
Of course more values of a and b are possible (using a progressive scale with a + b =
1):
Fr 2 Re −1
1 1 4 1
Fr 0 Re1 Fr 2 Re 2 Fr 5 Re 5 Fr 1 Re 0
Re vtip P/V Fr tm
Overview:
Scale kd kn kRe k v tip kP/V kFr k tm
criterion
FraReb (a,b) 1 1 1 1 1 1 1
Re (0,1) k k-2 1 k-1 k-4 k-3 k2
vtip (½,½) k k-1 k 1 k-1 k-1 k
P/V (4/5,1/5) k k-2/3 k4/3 k1/3 1 k-1/3 k2/3
Fr (1,0) k k-1/2 k3/2 k1/2 k1/2 1 k1/2
tm (2,-1) k 1 k2 k k2 k 1
This table shows why the scale-up criterion ‘P/V constant’ is so popular in the process
industry: what is not constant is nevertheless not changing too much, except for Re.
The fact that the ratio P/V is directly coupled to the energy consumption (energy
costs) has probably also contributed to the popularity of this scale-up criterion. In a
limited questionnaire among process industry 30% indicated the use of this criterion.
A value of 2 kW/m3 is often used as upper boundary.
The row with ‘Fr constant’ is (of course) even looking better; unfortunately Fr is
seldom an important dimensionless group in mixing processes.
4.2. Regimes
For a process as used in the process industry, we can distinguish a large number of
subprocesses that take place simultaneously, parallel or in series. Of all possible
subprocesses, several are ‘invisible’ because of two very distinct reasons:
4 4
(1) the subprocess takes place much too slow; the subprocess is ‘frozen’; e.g. in the
production (growth) of baker’s yeast evolution (usually) does not play a role
(2) the subprocess takes place much too fast; the subprocess is in ‘pseudo steady
state’; e.g. at laboratory scale mixing is mostly no problem, because mixing is
much faster than all other processes we want to realise in the mixed vessel.
Subprocesses which are neither ‘frozen’ nor in ‘pseudo steady state’ determine the
regime of the process. Together these subprocesses determine how fast the (total)
process runs. The search for the regime of a process is called regime analysis.
Example
At a chemical laboratory the following chemical reaction is performed:
A(aq) + B(aq) Æ Pi(aq) (desired)
Component A is added dropwise into the aqueous solution of component B. The

mixture is firmly stirred. Furthermore the reaction between A and B is exothermic
(-ΔHr > 0), resulting in a considerable heat production. The mixture is well cooled
e.g. via the reactor wall, in order to keep the temperature low and to suppress the
following undesired side reaction as much as possible.
2A(aq) Æ Qi(aq) (undesired)
The subprocesses or mechanisms are:

A
(1) the desired reaction (Pi)

(2) the undesired reaction (Qi)
(3) the supply of component A
(4) the mixing of the mixture
(5) the transfer of reaction heat
(6) …. (what we did not think of)
B
(7) ….
At laboratory scale mixing and heat transfer usually form no problem; a fast
rotating magnet stirrer and cooling via the reactor wall suffice. These mechanisms
are then at ‘pseudo steady state’.
The undesired reaction we try to keep ‘frozen’; we do this by controlling the
temperature well, by stirring firmly so that the concentration of A is homogeneous
(as low as possible) in the reactor, and by adding component A not too fast. Then
we deal with a pure chemical regime, as is usually the case in laboratory
experiments.
The chemical reaction rate determines how fast the process runs (can run); the
addition of component A is adjusted to that. However, at large scale usually the
transport processes determine the rate of the process. In the example of the ideal
tube reactor we also dealt with a pure chemical regime; we only had to keep the
Da I group constant, the ratio between convection time and reaction time.
4 5
In scale-up we distinguish three regime states:
- a pure regime: at both small and large scale the same unique regime prevails;
while scaling only one dimensional group has to be kept constant: ‘heaven’ for
scaling people.
- a mixed regime: at both small and large scale the same regime prevails, but it is
not unique; more than one subprocess is rate-determining.
- regime transition: at small and at large scale different regimes prevail. All
laborious measurements of kinetics done at laboratory scale are of no use, except
for determining that they are of no use: ‘hell’ for scaling people.
The prevailing regime can be determined experimentally as well as theoretically:
(a) experimentally: if at laboratory scale increasing the rotation frequency of the

stirrer and increasing the cooling capacity have no effect on the conversion we can
conclude that there is neither a mixing regime nor a cooling regime. So variation
of process conditions gives information about the prevailing regime.
(b) theoretically: at large scale, experiments such as in (a) are expensive and time
consuming; therefore, a theoretical regime analysis is preferred. Different
theoretical methods exist to that aim.
4.3. Characteristic times
In Delft (dept. BT) the use of characteristic times is preferred for theoretical regime
analysis. The most important argument in this is the general applicability: each
subprocess or mechanism takes place in time and can therefore be described with a
relevant physical quantity containing the dimension T (= time). This quantity can be
used to construct the characteristic time of the subprocess:
L2
diffusion :D (L2/T) : tD ~ 0
D
1
reaction : kr (1/T) : t r ~
kr
h ρc p L 0
heat transfer : (L/T) : t ht ~
ρc p h
etc.
(L0 is a still to be determined characteristic length)
In the concept of characteristic times emphasis is on time; these characteristic times

are not necessarily constant. Characteristic times also have a reciprocal character:
small times relate to high rates and visa versa (in this respect ‘rate’ defined as a
reciprocal characteristic time may perhaps deserve preference).
Characteristic dimensionless groups can also be interpreted as the ratio of (at least)
two characteristic times:
L
mass transfer : k (L/T) : t mt = 0
k
2
L
diffusion :D (L2/T) : tD = 0
D
t kL
Sh = D = 0
t mt D
4 6
Characteristic times are relatively easy to derive, because they refer to one isolated
subprocess or mechanism. There are - roughly - three methods to derive characteristic
times:
Method 1: from the ratio ‘amount’/’rate’
Example: mechanism = convection
Convection: ΦV (L3/T)
V
‘rate’ = ΦV; ‘amount’ = V ⇒ tc =
φV
Convection: Φm (M/T)
m
‘rate’ = Φm; ‘amount’ = m ⇒ tc =
φm
Convection: v (L/T)
L
‘rate’ = v; ‘amount’ = L ⇒ tc =
v
In case of flow of an incompressible liquid through a tube all three

characteristic times are equal:
L A ⋅ L V ρV m
tc = = = = =
v A ⋅ v φV ρφV φm
Method 2: from the balance equation
Example: mechanism = first order chemical reaction
dc
component balance: V = −k r c ⋅ V
dt
Division by the volume V and using c0 = c(0) makes the differential

equation partial dimensionless:
dγ
= − k r γ with dimension T-1
dt
1
Then the characteristic time is t r =
kr
We can also take the Laplace transform of the balance equation:
V ( sc(s) − c(0) ) = − k r ⋅ V ⋅ c(s)

c(0)
c(s) =
s + kr
c (0) / kr 1
In standard form c(s) = and t r =
(1/ k r ) s + 1 kr
4 7
Method 3: from the solution of the balance equation
Example: mechanism = mass transfer
Most microorganisms need oxygen e.g. for their growth; oxygen

dissolves poorly in water and most other media. Oxygen transfer from
gas bubbles to the aqueous media can thus rapidly become limiting.
Mass balance (oxygen; L-phase)

VL L = k L A LG ( c*L − c L )
dc
dt
L ε 1−ε
G
cL: oxygen concentration (L-phase)

VL: volume L-phase
kL: mass transfer coefficient
ALG: exchanging area between L and G
cL*: saturation concentration
ε: volume fraction L-phase
solving:
= k L LG ( c*L − c L )
dc L A
dt VL
A LG VL + VG *
= kL
VL + VG VL
(cL − cL )
= k L ⋅ a V ⋅ ( c*L − cL )
1
ε
aV: specific exchanging area (interfacial area per dispersion volume) *
⋅ ( cL − cL )
dcL k L a V *
=
dt ε
Solving gives:
⎡ ⎛ k a ⎞⎤
c L ( t ) = c*L ⎢1 − exp ⎜ − L V t ⎟ ⎥
⎣ ⎝ ε ⎠⎦
⎡ ⎛ t ⎞⎤
= c*L ⎢1 − exp ⎜ − ⎟ ⎥
⎣ ⎝ t mt ⎠ ⎦
ε
So: t mt =
k La V
aV
*
Note: Often the interfacial area per liquid volume is used: a L =
ε
4 8
For linear or first order processes all three methods yield the same characteristic
times, the time constant; for non-linear processes usually only one method is
applicable (mostly method 1) and then the characteristic time is not necessarily
constant at all times.
In the table in the appendix a number of characteristic times is listed. Note that
characteristic times belonging to transport processes contain lengths and thus are
scale dependent.
Regime analysis with characteristic times procedure
• choose those subprocesses (mechanisms) from the total process you want to
consider in the regime analysis
• determine the characteristic times of these subprocesses (mechanisms)
• take the subprocess that is assumed to prevail in the total process:
the mixing process for a mixer
the conversion for a reactor
the heat transfer for a heat exchanger etc.;
its characteristic time will be indicated with to
• compare the characteristic times of the other subprocesses with that characteristic
time to:
ti
i = 1,2,...
to
• subprocesses with characteristic times in the range of (about)
1 ti
< <5
5 to
determine (also) the regime
4.4. Serial versus parallel subprocesses
To draw conclusions from a regime analysis it is important to consider the position of

the subprocess or mechanism in the total process.
Two situations are especially of importance:
- serial subprocesses
- parallel subprocesses
A. Serial subprocesses
For a model of a number of mixers in series to describe incomplete mixing the serial
character is evident; the same holds for two or more reaction steps in series. But also
different kinds of subprocesses can be in series. For instance, in heterogeneous
catalysis the reactant must first be transported from the bulk to the catalyst surface:
tmt; after that it can react in the catalyst: tr.
For serial subprocesses the characteristic times must be added up and the mechanism
with the longest time is (for large differences) rate-determining.
4 9
B. Parallel subprocesses
An example of parallel subprocesses is a main reaction with one or more side

reactions. In the example of heterogeneous catalysis after mass transfer, diffusion (tD)
and reaction (tr) are taking place parallel in the catalyst particle.
In this case the reciprocals of the characteristic times have to be added up and the
mechanism with shortest time is (for large differences) rate-determining.
result rate-determining (t1 >> t2)

Serial t = t1 + t 2 t1
1 1 1
Parallel = + t2
t t1 t 2
For serial subprocesses the mechanism with time t1 is rate-determining because the
other subprocess must wait for the slower subprocess. For parallel subprocesses the
mechanism with time t2 is rate-determining because the other subprocess hardly takes
place.
At the end an example for illustration.
Example: shake flask experiment
To grow an inoculum for a bioreactor a shake flask sealed with a cotton plug is
used. The shake flask is an Erlenmeyer flask of 2000 ml which is filled with 800
ml medium to a height of 5 cm. The medium contains sufficient components for
prosperous growth of the aerobic microorganisms, except one: the required
oxygen must be taken up from the air.
Question: What is the regime? d1

h1
D1 = εDGo2 = 4 × 10−6 m 2 s
D2 = DGo2 = 2 × 10−5 m 2 s h2
−9
D3 = D = 2 × 10 m s
L 2 d2
o2
h3 V3
h1 = 0.04 m d1 = 0.05 m
h2 = 0.18 m d2 = 0.125 m d3
h3 = 0.05 m d3 = 0.16 m
The oxygen consumption rate ro increases from 8×10-7 kg/m3s (at the start of the
experiment) to 8×10-5 kg/m3s (at the end of the experiment when there are
sufficient cells.)
The process is partly serial: the oxygen diffuses through the cotton plug into the
flask, diffuses through the gas phase in the flask, and is transferred to the medium.
Thereafter the oxygen is mixed throughout the medium and consumed in parallel.
For the oxygen concentration (co) in the medium the maximum value of 9×10-3
kg/m3 is used. For kL a value of 9×10-5 m/s is used, not really dependent of
process conditions.
4 10
Characteristic times
4 h12
Diffusion through the plug : tD1 = 2 = 1.6 × 10+2 s
π D1
4 h 22
Diffusion through the gas phase : t D2 = 2 = 6.6 × 10+2 s
π D2
4 h 32
Diffusion through the media : tD3 = = 5.1 × 10+5 s
π D3
2
VL
Mass transfer : t kL = = 7.2 × 10+2 s
k L 4 πd 2
1 2
c
Oxygen consumption (maximal) : t o = o = 1.1 × 10+4 s (start)
ro
= 1.1 × 10+2 s (end)
For the serial process (diffusion Æ diffusion Æ mass transfer) mass transfer is
rate limiting. The diffusion through the plug, which might easily be considered to
be the limiting process, appears to be the fastest process and thus far from rate
determining.
Without measures (shaking) the mixing of oxygen throughout the media by
diffusion would be totally insufficient; only the top layer (app. 2 mm) would be
sufficiently aerated.
At the start of the experiment all transport is fast enough; at the end of the
experiment mass transfer and diffusion through the gas phase have become
limiting, resulting in an oxygen free culture. Remember that the consumption
rates are maximum values!
4 11
Appendix A: List of characteristic times
mechanism parameter characteristic time

V
convection (liquid) ΦV (L3/T)
ΦV
m
convection (gas) Φm (M/T)
Φm
L
convection (tube) v (L/T)
v
chemical reaction 1
kr (1/T)
(first order) kr
chemical reaction c
r (A/L3T) 1
(general) r
α
mixing by stirrer n (1/T) (12 < α < 120)
n
h ρc p V ΔTchange
heat transfer (L/T) ⋅ ⋅
ρc p h A ΔTdriving
mass transfer ε 1
kL (L/T) =
(liquid side) k La V k La L
mass transfer m (1 − ε ) m 1− ε
kL (L/T) =
(gas side) k La V k La L ε
diffusion* 1 δ2
D (L2/T) π = 3.14159…
(plate, two-sided) π2 D
diffusion* 4 δ2
D (L2/T)
(plate, one-sided) π2 D
diffusion* 1 δ2
D (L2/T) μ1 = 2.40483…
(cylinder) 4μ12 D
diffusion* 1 δ2
D (L2/T)
(sphere) 4π2 D
* for conduction (thermal diffusion) these characteristic times are still valid if the
diffusion coefficient is replaced by the thermal diffusion coefficient:
λ
a=
ρc p
1
“A” stands for the dimension “amount of substance”
4 12
Scale-up/scale-down Scale-down
5. Scale-down of biotechnological processes

K.Ch.A.M. Luyben
5.1 Introduction
The heart of any bioconversion Bioconversion process

process is the biocatalyst (see figure
1). Without a biocatalyst you have no macro-environment
bioconversion and there is no need to
micro-environment
talk about a bioconversion process.
This means that it is of prime DNA
importance to obtain the best substrate product
biocatalyst possible for the conversion biocatalyst
cell
at hand. If you want to make a certain ‘bioreactor’
product using a particular substrate try
to optimise your biocatalyst first. But
once you have done that, once you
have had the genetic engineering and
all the other tools applied to that Figure 1. Schematic of a bioconversion process
biocatalyst, there is no other choice
than to optimise the microenvironment of that biocatalyst. This should improve the
conversion further to achieve an optimal conversion from substrate to product. There
are no little knobs on the cells that you can tune during the fermentation process, so
the only way to influence the biocatalyst is through the microenvironment. And that
microenvironment can be influenced by something we generally call 'control'. Any
form of control can be meant here, for example the amount of substrate you chose at
the beginning of your process or the temperature or pH you choose. But also very
fancy control systems based on measurement information from the process. Knowing
that the biocatalyst experiences varying conditions (dynamic situation) at the large
scale one can conclude that optimising a bioconversion is not a simple task and needs
a certain strategy.
Before we look at the strategies used in order to do scale-up, it is best to have a look
EXISTING TO BE DEVELOPED first at what we mean by
scale-up. The field of
suited for structured scale-up can be divided
NEW
into several types of

NEW
scale-down step-wise
OPERATION
problems. Looking at the

approach scale-up
aim of your operation, so
at the aim of your
KNOWN
optimisation scale-up conversion, at the one

KNOWN
of existing using the hand, and looking at the

operation analogy approach installation, the hardware
you use for that, on the
EXISTING TO BE DEVELOPED other hand, scale-up
problems can be divided
INSTALLATION according to the matrix
Division of scale-up problems
given in figure 2.
Figure 2.
5 1
In this matrix 'new' versus 'known' operations and 'existing' installations versus those
'to be developed' are depicted. In the lower left corner you talk about direct
optimisation of an existing operation; you have a known process and you have
existing equipment.
At the other extreme you talk about the area which a lot of people think about if we
use the word scale-up; a new process (operation) in a new installation. However, in
industry this is not the most common problem encountered with respect to scale-up.
To my knowledge the most common problem in industry is where you have an
existing installation in which you have to do a slightly different or completely
different conversion. You either have another biocatalyst or a modified biocatalyst or
you have another substrate. For example, the people buying your substrate found
something cheaper and you'll have to live with that. This is a much more common
situation than developing a completely new process; new hardware for a new
conversion. And it is especially that more common situation that is suited for what we
call the scale-down approach. Going from a small to a large scale usually you will
find a decrease in productivity.
However, also the opposite sometimes is found where the productivity increases with
scale.
5.2 Scale-up strategies
You can ask yourself what is the best technique to use for scale-up, and the answer
will probably be that there is no single best technique to use for all cases. There are
several techniques you can choose from in order to do the scale-up (Johnstone et al.,
1957). Scale-up, from small to large, and optimisation of existing large scale
processes, are governed less and less by 'trial and error', 'rules of thumb' or just
experience. More and more scale-up is performed according to a well-chosen
methodology. A methodology, that has proven to be successful in various application
areas.
An example of a scale-up problem in the biotechnological field is that of a reduced
yield obtained at large scale compared to the small-scale results. As a consequence of
the scale-up mixing, mass transfer and heat transfer will be more problematic and
therefore the microorganism will 'see' a different microenvironment. Only in
exceptional cases will these 'problems' exist at an experimental scale, for example in
cases of shear sensitive biomass, very viscous mycelia or very high specific
consumption rates.
Basically the problem of is a problem of transfer (mass, momentum or heat) and the
lack of knowledge with respect to the interaction of hydrodynamics and other relevant
subprocesses in a large-scale process. Two approaches can be taken to tackle this
problem:
• an exact approach
• an approximate approach
The first approach involves a complete description of the hydrodynamics and the
interaction with other relevant mechanisms. For this fundamental method all the
microbalances for momentum-, mass-, and heat-transfer in the system, need to be
solved. It will be clear that this is not possible for the scale-up of a biotechnological
5 2
process. However, for some scale-up problems, particularly when the flow is well
defined (or absent), solving the microbalances can be very useful. Or using a semi-
fundamental method based on simplified flow models (for example plug-flow with
dispersion), to avoid the use of too complex balance equations. These (semi-)
fundamental solutions are, however, extremely difficult, slow and expensive.
The second (approximate) approach aims at developing an adequate scale-up
methodology, based on the results obtained at experimental scale. The change of scale
is often based on a dimensional analysis or a regime analysis. This paper deals with
the scale-down method based on such a regime analysis.
Furthermore, there are of course the methods known as 'rules of thumb', 'trial and
error', and 'multiplication of elements'.
'Rules of thumb' (Sweere et al., 1987) can be often adequately used. In practice one is
often well aware of the specific characteristics of a particular process. Based on
experience it is for example known that if one is working with a very shear-sensitive
microorganism the tip speed is an important parameter for scale-up.
Or it is known that the oxygen mass transfer is extremely important and that one
should try to keep that constant. In the past trial and error was the technique to use.
The disadvantage of that is of course that you do the trial at the large scale and you
make the errors at the large scale. This can be very expensive. The most simple
technique for scale-up is of course 'multiplication of elements', which I would only
recommend for the last steps in the total scale-up scheme. But if you are really in a
hurry and were not able to scale-up properly, it can be a fast and very reliable
technique. If you can do a fermentation at 20 litres you can do it at 20 000 litres too;
you just use 1000 fermentors in a row. Where you find the people to run them is of
course another question. Scale-up techniques can thus be divided into three categories
of scale-up methodologies, given in the following table.
scale-up methods in three classes
fundamental method (solving all micro-balances)

semi-fundamental method (using simplified flow models)
dimensional analysis (based on dimensionless groups)

scale-down method (based on regime analysis)
rules of thumb (based on experience and estimations)

‘trial and error’ (based on common sense)
multiplication of elements (based on small scale
experience)
Some of the above-mentioned approaches are more suited than other approaches. It
also depends strongly on the situation at hand, which of these approaches would be
the best to use. None of these techniques is adequate for all possible situations. A
combination of the semi-fundamental approach with rules of thumb is mostly applied.
However, almost always the results are based on an extrapolation from a smaller scale
and this has an inherent risk.
5 3
The scale-down approach of rate-limiting mechanisms (a systematic extrapolation

based on common sense) is a relative new tool to solve scale-up problems. It is this
technique that we will discuss more in detail here.
5.3 Scale-down
Basically scale-down is doing at the small scale what you can do (or intend to do) at
the large scale. By this statement I would like to emphasise the fact that the reference
for this procedure, the framework you are dealing with, is the large scale.
The four step scale-down procedure (Oosterhuis, 1984) looks as follows:
(1) regime analysis of the (proposed) process at production scale;
(2) simulation of the rate-limiting
mechanisms at laboratory scale;
(3) optimisation and modelling of
the process at laboratory scale; REGIME
IMPLEMENTATION
(4) implementation of the process ANALYSIS
at production scale by
transformation of the optimised
laboratory conditions.
See figure 3 for schematic SIMULATION OPTIMIZATION
OPTIMISATION
representation of the scale-down
method.
The most important requirement Figure 3. Scale-down procedure (Oosterhuis, 1985)
for experiments at laboratory scale
is that they have to be representative for the conditions prevailing at the large scale.
This determines the possibilities and limits of small-scale experiments.
Optimisation of the process at laboratory scale and modelling the predominant

phenomena form the third step in the scale-down approach. During optimisation one
has to keep in mind that the optimised situation has to be translated (back) to the
production scale. Consequently, not all possible optimisation results can be used.
With regard to the influence of fluctuations it is important to notice that fluctuations
tend to increase during scale-up (decreased mixing), resulting in transient conditions
for the microorganisms. Much is known about modelling of balanced growth (Barford
et al., 1982) and product formation. However, little is known about growth and
product formation under transient conditions. A distinction here must be made
between empirical and mechanistic models. Most of the models described in literature
are empirical, and thus incapable of predicting. These models add little to the
fundamental understanding of microbial dynamics.
Only few mechanistic models can be found in literature. This is due to the fact that the
biochemical and physiological information, forming the basis of these mechanistic
models, is insufficient. As a result of this representative (of the full scale)
experimental procedures are extremely important.
In the fourth step, the optimised laboratory conditions are implemented at production
scale. Models formed in the previous step can be used for this purpose. Rules used to
scale-down the process conditions can now be used in scaling-up these conditions
again. The success of the scale-up strongly depends on whether one has been
successful in designing representative scale-down experiments.
5 4
5.4 Regime analysis
The first step in the scale-down method (see above) involves a regime analysis. In
order to treat the basis of this analysis, an explanation of the concepts 'time-constant'
and 'characteristic times' is essential.
In biotechnology, as in many other engineering disciplines (or perhaps even more so),
the systems encountered are complex in nature. Rigorous description of the behaviour
of such systems leads to large sets of mathematical expressions describing the time
dependence of a large number of variables (Roels, 1983). The model obtained would
be as complex as the reality itself. Moreover, these equations contain a massive
number of parameters, which in many cases cannot be determined experimentally. To
simplify the complex reality to a model that still can describe the relevant
mechanisms accurately, a consistent methodology is wanted. Regime analysis is such
a methodology based on the analysis of the sub-
processes in a system. In this methodology the
sub-processes are characterised by 'characteristic
times', also called 'relaxation times’ or 'time
constants' (in case of first-order processes). In this
paper, a basis is given for the use of characteristic
times. Practical examples of the use of Φ Cso
characteristic times will be given.
V
5.4.1 What are time constants?
Time constants are defined, in a strict

mathematical sense, only for first order linear Φ Csi
processes. As an example, consider a well-stirred
tank reactor in which a first-order chemical Figure 4. Well-stirred tank reactor with
reaction takes place (figure 4). A substrate chemical reaction
solution of concentration Cso is fed to the system
at a rate Φ). The reactor volume, V, is constant, and therefore the flow out is also
equal to Φ. The concentration of substrate in the flow leaving the system is assumed
to be equal to the substrate concentration in the system, Csi. If equation (1) represents
the rate of chemical reaction, the balance equation (2) describes the dynamics of the
substrate concentration in the system:
r = kCsi (1)
d(VCsi )
= Φ (Cso − Csi ) − k Csi V (2)
dt
At steady state, the left-hand side is equal to zero and it follows that:
Cso
Csi = (3)
V
1+ k
Φ
5 5
The internal substrate concentration can thus be expressed in measurable external

quantities. Now suppose that at a given time the substrate concentration in the inlet is
changed from Cso to Cse. The differential equation can then be solved to give, for Csi:
Φ
1 −( + k) t
Csi = [Cse + (Cso − Cce ) e V
] (4)
V
(1 + k )
Φ
The result is plotted in figure

5. The time constant, tr , is Csi
defined as the time which
elapses before the difference Csi = Cse
between the internal
concentration and the initial
steady state reaches a fraction Csi 1 – e-1( = t / t r )
of (1 - e-1), or about 63 % of
the total difference between
the initial and final value of
Csi. In this case it follows that: time constant
1
tr = (5) tr time
Φ 0
+k
V
Figure 5. Response of the system to a change in the
inlet concentration
The time constant for this
process consists of two contributions, namely the mean residence time in the reactor,
tres = V/Φ, and the reaction time of the first order reaction, trea = 1 / k. The inverse of
the time constant for the process, tr , is the sum of the inverse of the residence time
and the inverse of the reaction time:
1 1 1
= + (6)
t r t res t rea
VARIOUS STEP RESPONSES However, if you do not have a

‘definition of characteristic times’
1.5 first order linear process, you
+ 20% cannot strictly define the time
constant in this way.
relative response
1.0 Nevertheless, we would like to

compare subprocesses in a total
- 20%
0.5
system, which we sometimes
might be able to subdivide into
let's say lower order processes,
0.0 but for example would not want
0 1 2t 3 4 5
t1 t2 3 time to subdivide them.
Figure 6. Step responses of different processes
5 6
We would also like to be able to deal with non-linear processes. In figure 6 a number
of step responses are given for different processes. One of them is the same as for the
first order linear process shown before, but the output to a stepwise input change
could also be something with some delay in it or with an overshoot or many other
things. If you plot them in such a way that you make the response relative to the final
value they will all start at the lower left corner and approach the final value
asymptotically. And if you then scale them with respect to their characteristic time,
which I will define next, they will all look somewhat similar. Now what is the
characteristic time? The characteristic time can be defined as the time it takes to reach
a certain band around the final value of that particular output, a signal relative to the
final value of that output signal. In this example I have taken plus or minus 20% (80%
- 120% of final value), so if the output signal stays within this band, we define this as
being the characteristic time for that particular subprocess. If I had chosen a slightly
smaller band here, then the subprocess with overshoot would have come out of the
band again and would have the highest characteristic time. So you see the
characteristic time is dependent on the basis you choose, defined by the band around
the asymptotic value. This means that if you start to compare characteristic times with
each other, you must try as much as possible to base them on the same deviation from
the final value. You are not always able to do that because some of the characteristic
times you have might come from calculations, others from pure estimations, and again
others come from the ratio of capacity and flow. In that case you should realise that
you can only compare
characteristic times in a microorganism and characteristic times with
in its environment
respect to their order of
-5
10 -4
10 -3
10 -2
10 -1
10 0 1
10 2
10 3
10 4
10 5
10 10 s
mass action enzyme induction

magnitude. If they differ by
a factor of 2 you are not
allosteric controls selection
really sure whether these
mRNA
microorganism subprocesses run with a
environment mixing significantly different rate.
di t
batch, fed--batch, cc--transient
Figure 7 shows
characteristic times of
continuous
subprocesses inside
Figure 7. Characteristic times of subprocesses microorganisms and some
inside microorganisms and in their environment examples of subprocesses
occurring in the
environment of the microorganism, which might lead to interactions with the
microorganism. As you can see there is a large span of characteristic times for the
different subprocesses either in the organism or outside the organism. These
subprocesses are of course a selection, a choice, of a number of subprocesses. You
can think of many more, and that is the trick of doing the regime analysis, which
subprocesses to include, where to get them from, and how to deal with them. There
are no strict rules, there is no strict description on how to do that, one should use its
intuition, and experience, one should listen to experienced people, read the literature
etc. Automatically this analysis comes down to a combination of many sources of
information.
5.4.2 Characteristic times: examples
In general, there are a number of methods for obtaining the characteristic times of a
process. The first method, which has been described in the previous section, consists
5 7
of an analysis of the mathematical solution of a process model. For linear, first order
differential equations, this is relatively simple. Other examples of this procedure are:
heat transfer in a stirred tank:
Tw − T UA Vρcp
= exp ( − t) → t ht = (7)
Tw − To Vρ c p UA
mass transfer in a stirred tank:
C∗ − C 1
∗
= exp ( − k La ⋅ t) → t mt = (8)
C − Co k La
However, in some cases this procedure is not valid, for example, in problems with no
available solution to the differential equations, or in the case of a problem for which
no general differential equation can be defined. In the first case, characteristic times
can be obtained from the differential equation itself. For instance, in the
dimensionless differential equation of a first order process one can often recognise the
characteristic times in the coefficients of the equation. Also literature correlations and
rules of thumb can be used, for instance, in the case of liquid mixing:
t mixing = 4 ⋅ t circulation (9)

V
t circulation = (10)
2 ⋅ H stirrer πDstirrer ⋅ NπDstirrer
So far, only characteristic times referring to physical processes determining the

environment of microorganisms were treated. Let us consider the time constants of
microorganisms more closely. A characteristic time can be chosen for any metabolic
variable of the system, because the specific rates of the changes of the variables can
be different. The system would be even more difficult to describe if the specific rates,
and therefore the characteristic times, are not even constant at all.
Because of the steadily increasing complexity, the description of such a system will
have to be limited to a few subprocesses. To do so, it is useful to compare
characteristic times of microorganisms with characteristic times of environmental
changes. In general, a microorganism can be assumed to have a large capacity to
react to environmental changes, described by a wide range of characteristic times, as
was shown in figure 7.
The time constants of the environment narrow our interest to a 'window'. Processes
occurring within the organism with time constants outside this 'window' can be
considered 'frozen' or in 'pseudo steady state' and thus can be left out of the
description (figure 8). In the following example the meaning of the terms 'frozen' and
'pseudo steady state' will be demonstrated. Morjanoff (1979) carried out continuous
culture experiments with Bacillus coagulens, grown on starch. The response of the
culture to changes in feed rate and composition were investigated (figure 9). It was
concluded that the culture was at steady state after 15 to 20 generations, as judged by
metabolic variables such as dry mass and specific rates of oxygen uptake and carbon
dioxide production. The production of extracellular amylase, however, took much
longer to achieve steady state (in excess of 100 generations). It was suggested that
5 8
amylase was subject to a long-term regulation mechanism such as feedback control

with a time lag.
I: window of 10 characteristic times of a system

generations; amylase
synthesis 'frozen' “pseudo
“frozen”
steady-state” “window of observation”
approximation
II: window of 100 approximation
generations; growth at
interaction with environment
'pseudo steady state'
Thus, when the observations not dynamic dynamic overshoot, not dynamic
always adapted oscillations, always in the
'window' is narrowed to the to environment lag phenomena initial state
first 10 generations, the
predominant feature is the tss < tmin tmin < tss < tmax tmax < tss
adaptation of the culture with
respect to biomass production
and gaseous metabolism. In Figure 8. Window of observation
this case, the production of
extracellular amylase can be considered 'frozen'. If the 'window' is expanded to
include more than 100 generations, the primary metabolic feature becomes the
regulation of amylase synthesis. In this case, the other cellular processes can be
assumed to be at 'pseudo steady state'.
On the basis of the
characteristic times for the
cellular growth
various mechanisms in a
specific rate
process one can perform a so-

called regime analysis. This
regime analysis should answer
questions such as “which
mechanisms are rate limiting or sis
e s ynt he
ruling the regime of the as
amyl
process?” Is the regime ruled by
one mechanism (pure regime) or
more mechanisms (mixed 10 time (generations) 100
regime)? Does a change of
Figure 9. Response of a continuous culture of
regime occur, going from one Bacillus coagulens grown on starch to a
scale to another? shift in feed rate (Morjanoff, 1979)
To answer these questions, the analysis has to allow for changes of scale and/or
changes in process parameters during the course of the process. The determining
factor in the performance of microorganisms is their microenvironment. Are there in
the existing regime any substrate or nutrient limitations due to fluctuations in the
concentrations in the microenvironment or in the pH? Are there significant changes in
the temperature or shear rate? All these can effect the yield of biomass or products.
Comparison of characteristic times and their change as a function of scale can give an
answer to the above questions.
However, it should not be forgotten that other forms of information on the process
should be brought in as well. Rules of thumb, literature data, experimental
correlations and experience can play an important role in this part of the analysis.
5 9
Also characteristic times are not the only parameters to characterise processes. In
some cases, stresses, heights (e.g. HETP, the height equivalent of a theoretical plate),
or dimensionless numbers may give a better insight into the problem (Pawlowski,
1971).
The characteristic time concept has, however, two advantages:
• the effects of transient conditions on large scale are very poorly known; therefore
it is difficult to find the right (dimensionless) characteristic parameters and it is
relatively easy to work with characteristic (relaxation) times.
• dimensionless numbers always give the ratio between two (characteristic) times
and not the separate values; therefore they contain less information.
In general we can conclude, especially under the dynamic conditions found in a large
fermentor, that various mechanisms inside the microorganism may have characteristic
times of the order of magnitude as those of the changes in the micro-environment of
the microorganisms (like mixing, mass transfer etc.). Therefore, those subprocesses
will interact and change the 'behaviour' of the microorganism. These subprocesses fall
in the so-called "window of observation".
5.4.3 Example: Gluconic acid production
Gluconic acid is produced in an aerobic batch process at a scale of about 25 m3. The
fermentor is equipped with two Rushton-type impellers, a star-shaped sparger just
below the lowest impeller and four baffles. Table 1 gives the relative dimensions of
the production-scale fermentor. To get an impression of the rate-limiting steps, regime
analysis was carried out (Kossen & Oosterhuis, 1985).
Table 1: Dimensions of production-scale fermentor for gluconic acid

production
Impeller/vessel diameter 0.32

Number of impellers 2
Impeller blade width/impeller diameter 0.2
Impeller speed 1.3 or 2.6 s-1
Baffle diameter/vessel diameter 0.09
Liquid height/vessel diameter up to 1.8
Gas flow rate per reactor volume up to 0.5 vvm (m3/min per m3)
Liquid volume up to 25 m3
Table 2: Characteristic times (in seconds) of the mechanisms that are

important in gluconic acid production
Transport phenomena
Oxygen transfer 5.5 (non-coal.) - 11.2
(coalescing)
Circulation of the liquid 12.3
Gas residence 20.6
Transfer of oxygen from a gas bubble 290 (non-coal.) - 593
(coalescing)
Heat transfer 330 - 650
5 10
Conversion
Oxygen consumption, zero order 16
first order 0.7
Substrate consumption 5.5 × 104
Growth 1.2 × 104
Heat production 350
Table 2 gives the characteristic times of the different mechanisms that play a role in
the process. The following conclusions can be drawn from the table.
Substrate consumption and growth have no influence on the performance of the
process. The times for oxygen consumption and oxygen transfer to the liquid phase
are of the same order of magnitude. Therefore, oxygen limitations may occur. Also,
the liquid circulation time is of the same order of magnitude, so the conclusion can be
drawn that oxygen gradients are likely to occur.
From a comparison of the gas residence time and the time for oxygen transfer from
the gas phase it is clear that no substantial exhaustion of the gas phase will occur.
Heat transfer will balance heat production. From a comparison heat transfer time with
the liquid circulation time it can be concluded that no temperature gradients will be
present in the fermentor.
coupled reactor system
It has to be remembered that
the correlations used give only
a rough estimation of the reactor 1
characteristic times, so only

pO
the order of magnitude can be 2
considered and compared.

This regime analysis has nitrogen
reactor 2
formed the basis of small-
scale experiments, looking at pO2
mass transfer and the

influence of fluctuating air
oxygen concentrations on Figure 10. Two-compartment system
gluconic acid production.
Figure 10 shows the experimental set-up for the small-scale experiments. With this
configuration the gluconic acid-producing bacteria can be ‘offered’ a similar
microenvironment as that
500
experienced in the large-scale gluconic acid
mM
fermentor. Figure 11 illustrates concentration
400
the influence of this changing
environment. In this figure exp. 1 exp. 2
300
three curves are shown for the
gluconic acid production as a
200
function of time. Experiment 1
refers to the situation with one exp. 3 (two compartment system)
100
fermentor that is well aerated
and well stirred (reference
0
case). 0 10 20 30 time 40 h
Experiment 2 was performed Figure 11. Experimental results gluconic acid production
with the two-compartment
5 11
system depicted in figure 10. However, both the fermentors were well aerated and
well stirred. This experiment was performed to test if the circulation of the broth over
the two fermentors as such would have any effect on the gluconic acid production by
the bacteria. As can be seen this is not the case and the production is the same (within
the experimental error). Experiment 3 shows clearly the influence of oxygen
limitation on the productivity of gluconic acid. This behaviour resembles that
observed in the production fermentor.
5.5 Conclusions
Scale-up often has to do with process modification. In other words, it is often not a
new process in a new piece of equipment that you are dealing with if you talk about
scale-up in practice. Scale-down can be used for the cases where you are talking about
this modification of the existing process in the existing equipment.
Regime analysis is the first and most critical step in the scale-down methodology.
Optimisation of the scale-down process is based on regime analysis. It is essential to
keep in the back of your mind that the possibilities you are creating by optimisation of
the small-scale or the alternatives you are suggesting can be implemented at the large
scale.
List of symbols
symbol property units
A heat transfer area m2

C concentration kg/m3
cp heat capacity J/kg
D diameter m
Φ feed rate m3/s
H height m
k reaction rate constant s-1
kLa mass transfer coefficient s-1
N stirring speed s-1
r reaction rate kg/(m3s)
T absolute temperature K
t time constant s
t time s
U heat transfer coefficient J/(m2s)
V volume m3
x co-ordinate m
subscript
se substrate at equilibrium
si substrate in tank
so substrate in inlet
ss subsystem
x biomass
w wall
* saturated
5 12
References
Barford, J.P.,Pamment, N.B., Hall, R.J., 'Lag phases and transients',

Microb.Pop.Dynamics, Bazin, M.J. (Ed.), CRC Press, Boca Raton (1982), 55-89.
Johnstone, R.E., Thring, M.W., 'Pilot plants, models and scale-up methods in
chemical engineering', McGraw-Hill, New York (1957).
Kossen, N.W.F., Oosterhuis, N.M.G., in 'Biotechnology' (Rehm, H.J., Reed, G., eds.),
VCH Verlagsgesellschaft, Weinheim (1985), 2, 572-605.
Levenspiel, O., 'Chemical Reaction Engineering', Wiley New York (1972).
Morjanoff, P.J.C., 'Kinetic studies on the growth of microorganisms on cereal starch',
PhD Thesis, Sydney (1979).
Oosterhuis, N.M.G., PhD Thesis, Delft (1984).
Pawlowski, J., 'Die Aehnlichkeitstheorie in der physikalisch-technischen
Forschung', Springer-Verlag, Berlin (1971).
Roels, J.A.,'Energetics and kinetics in biotechnology', Elsevier Biomedical Press,
Amsterdam (1983), 205-222.
Sweere, A.P.J., Luyben, K.Ch.A.M., Kossen, N.W.F., 'Regime analysis and scale-
down, tools to investigate the performance of bioreactors', Enzyme Microb.Technol. 9
(1987), 7, 385-398.
5 13
Scale-up/scale-down Pilot plant studies
6. Pilot plant studies
6.1 Introduction
In previous chapters attention was paid in a rather formal way to scaling up of a

model to a prototype. Of course, a pilot plant is a sort of model and a commercial
plant is the prototype. Among the several different definitions of a pilot plant a simple
one is [Ogorzaly, 1958]:
A pilot plant is an assembly of equipment devoted to studying the

critical features of a process operation.
Pilot plants form a link between research and the actual application of a process.
Actually a pilot plant offers the opportunity to do research on a larger scale.
A more general definition of a pilot plant is given by [Paluzzi, 1992]:
A pilot plant is a tool intended to allow investigation of a process

or process problem on a manageable scale in a realistic manner
and in a timely fashion.
This definition clearly shows that a pilot plant is not an end in itself, but a way to
achieve a certain goal, which is a commercial, full-scale plant. A manageable scale
implies a limited commitment of time, money, and other resources. A pilot plant
should be large enough to include all effects that might be of importance and that are
also present in a full-scale plant. Finally, research in a pilot plant is generally
temporary in nature.
A number of general characteristics of pilot plants are:

• critical process properties are being studied.
• the size is in between laboratory scale (bench scale) and commercial scale.
• the most important problems studied are of technical and economical nature,
because the scientific basis has been looked at earlier.
• in most of the cases several pieces of equipment (unit operations) are involved.
Of course, the separate unit operations can be studied (first), but generally more unit
operations are processed together to simulate the entire process.
6.2 Goals of a pilot plant
There are a number of strategies for scaling up. The two extremes are:
• Fundamental: Each process development consists of small-scale studies
immediately followed by a successful design of a commercial plant.
• Empirical: The desired process is simulated and the required answers for a
successful process development are obtained by empirical trial and
error adjustments in the set-up and operation of a plant.
In practice there is not such a large distinction but a mixture of both methods, namely
an integral method in which both theoretical analysis and fundamental experiments
(fundamental) and process simulation on an adequate scale (empirical) are used.
6 1
Which method is more important depends on a large number of factors including the
characteristics (complexity and scale) of the regarded process, the amount of
experience, and the available scaling up technology. Because there is not a clear
distinction between the two scale-up techniques, a multitude of possibilities for
scaling is available.
There is a wide variety in scale and complexity of pilot plants. Independent of the size
and nature of the equipment, the fundamental goal of a pilot plant is to reduce risks
and errors in the market evaluation, design, start-up, and processing of a full-scale
production plant. The chance of total failure is minimised.
A pilot plant can be used for existing as well as new processes. In the case of a new
process (in new or existing equipment) the understanding of chemistry, physics, and
technology for the considered process is not adequate to extrapolate form bench scale
to full scale. The pilot plant then is an intermediary stage in the development of those
processes en equipment. Moreover, useful economical data can be obtained and often
tests on pilot plant scale have influence on the decision whether to proceed with a
project. In the case of adjustment of an existing process (in new equipment) it often is
not practical to carry out tests on full scale because of the high degree of complexity,
the too high costs, or disturbance of production.
A pilot can be used to for several purposes. A number of them are:

• to test product quality.
• to produce small amounts of product for marketing.
• to set-up or improve process and product formulation.
• to evaluate different feedstock and auxiliary materials
• to consider time effects such as accumulation, fouling and corrosion on process,
materials and chemicals.
• to study dynamic behaviour.
• to test construction (materials).
• to enhance development of high-technological processes.
• to form reliable scale-up criteria due to a better insight in specific scale-up effects.
• to demonstrate convincingly that the process is indeed (technically) feasible.
• to carry out an economical analysis with attention for use of raw materials and
auxiliary materials among other aspects.
• to achieve good process improvements for a next plant. Especially if the time for
development of a process is long (for instance in the case of a patent).
• to limit the chance of failure or a less efficient process or even a process that is
not paying off. Especially if the process has a large scale and requires very large
investments.
Advantages of pilot plants:

• more accurate design possible.
• reduction in investment costs.
• training.
• primary production (for marketing).
• demonstration.
Disadvantages of pilot plants:

• The application of a process on a commercial scale is slowed down.
• Pilot plants are expensive.
6 2
• Possible alternatives are less ‘visible’.

• There is no guarantee for success on a commercial scale, scale effects can still be
present. In this case the pilot plant is too small.
A pilot plant is paying off, in terms of time and money, if it is preferable to other
methods such as literature search, calculations (computer simulations), lab scale
studies, and full-scale plant tests as a source of new information needed for process
design. Pilot plant work (in a narrower sense) is often unattractive due to high costs
involved. Already in 1965 the editorial board of the journal Industrial & Engineering
Chemistry wrote: "Eliminating the intermediate steps in process scale-up from bench
to plant becomes more an economic necessity and less an academic ideal". This is
why the pilot plant stage is often omitted in the present way of scaling up. However,
the pilot plant is often still important during the design stage of a process.
6.3 Important practical aspects of pilot plants
• Size.
On the basis of size, the generally accepted classification of pilot plants is as follows
[Paluzzi, 1992]:
- Laboratory-scale, bench-top pilot plants or micro-units.
These are pilot plants that generally fit on a bench-top or inside a small laboratory
neighbourhood. In general the volumes are smaller than 1 litre, the total ground area
is in the range of 0.5 to 1.0 m2, and the costs are up to € 250 000. Traditionally
totally manual and continuously attended, many new versions are heavily automated
and designed to run continuously, unattended.
- Integrated pilot plants or research-scale pilot plants.
This remains the workhorse of the industry and may vary in size from several frames
or pallets to a unit occupying a small building. In general tanks are used with
volumes up to 4 m3, the total area is in the range of 2 to 15 m2, and the costs are
from € 100 000 to € 1 000 000. They are usually automated and may frequently
designed for unattended operation.
- Demonstration units, semi-works units, or prototype units.
These are units designed to operate at the lower end of plant scale. They are very
large, tanks are 4 m3 to 40 m3, area is typically in the order of 900 m2 or more, and the
costs are € 500 000 to € 25 000 000.
• Costs.
Pilot plants are generally expensive, the equipment cost varies, but the instruments
and process controllers can account for up to 50% of the total cost. This is due to the
fact that in general the same number of instruments is needed for a pilot plant as for a
full-scale plant. Moreover, often extra instruments and process controllers are
necessary to gather data necessary for scaling up. Also with regard to operating costs
a pilot plant is expensive, because often a project team is involved. However, all these
costs can be earned back, for example by simplifying of the start up procedure as a
result of the knowledge gained on pilot plant scale. The costs of a pilot plant can be
reduced by approximately 20 to 40% by using a reduced scale, i.e. a mini-plant that is
roughly 10 times smaller than a pilot plant. An important aspect to keep in mind is
that the stability of processes on a smaller scale (mini-plant) usually is not very large
en also difficult to achieve by means of process control. It is also possible to consider
a part of a process, however then some parts cannot be simulated, for instance recycle
streams which often are a major reason for fouling of a process.
6 3
• Time scheduling.
It’s an important question at what stage of a product development a pilot plant should
be built. Usually there is pressure from the commercial site to get to the production
stage as soon as possible so that there is a large temptation to skip the pilot plant
phase. A possible solution is to start simultaneously with the design and constructing
of the full-scale plant ánd the pilot plant. Since the time required for design and
construction of a pilot plant is much shorter than that of a full-scale plant, it is
possible to do tests in the pilot plant before the start up of the full-scale plant.
However, of course there is a risk that the pilot plant tests show that the wrong choice
has been made for the equipment or even that the desired product cannot be produced
at all, while the full-scale plant has already been built or is being built.
• Hold-up.
The hold-up of the equipment should be small so that short start-up and adjustment
times are possible.
• Flexibility.
The pilot plant must be adequately flexible. There have to be enough possibilities to
take samples frequently and at a lot of places for analysis without disturbing the
process. Also, an undisturbed supply of raw materials and a suitable discharge of
products must be possible. The possibility for variation of independent variables must
be widely available.
• Product research.
A pilot plant can be built with the single purpose to produce small amounts of product
that can be used for product research such as market research, quality research, or
clinical tests.
• Recycle streams.
Recycle streams are applied in processes for increasing the yields and reducing the
amount of waste and costs. Recycling also causes the risk of fouling and other
negative influences on the process. Often certain materials accumulate in recycle
streams, which is an effect not adequately visible on bench scale.
• Safety.
For analysing the safety of a process, especially in the case of risky process steps
and/or dangerous raw materials, precursors or products, it can be advisable to map the
risks of a process.
• Use of computers.
Computers play an essential role in the operation of pilot plants. Computers mostly do
the process control, on-line analysis, and data processing. For instance, feed streams
can be exactly controlled, and by strongly improved and accelerated process control,
processes are often more reproducible, controllable and also safer. Another advantage
of the use of computers is the decrease of costs because of lower expenses on
personnel and also equipment.
Another use of computers is to simulate processes with the help of models. It is often
not necessary to design, construct, and built an entire pilot plant. If certain data
concerning a process are already known, simpler experiments can be sufficient to
obtain the required parameters. A good model provides a quick and simple simulation
of a process without doing a lot of experiments. E.g. (standard) heat exchangers are
6 4
nowadays fully computer designed. Of course there is a risk of premature and over-
enthusiastic use of models for scaling up that can lead to overlooking possible
changes of regime. This last note pleads for a careful application of models for scaling
up. Complex processes such as large-scale aerobic fermentation can still not be scaled
up by computer.
6.4 Rules of thumb for pilot plants
Rules of thumb are frequently used in chemical industry. In the case of pilot plants
they are often very useful during the first design stages and also for troubleshooting,
especially to have a good first estimation for the solution of a problem. A number of
rules of thumb for pilot plant are summed up below [Aerstin et al., 1985]:
General:
Reaction rates may double every 10°C.
Vapour pressures may double every 20°C.
Liquid and vapour handling are easier than solids handling.
Recycle streams usually build up impurities, which can lead to surprising
problems.
A “boiling reactor” or adiabatic reactor is usually easier to scale up than a jacket
cooled reactor.
Small volume production favours multipurpose use of equipment; large-scale
production favours single use of each item.
Use digital (on-off) control in small equipment to save cost.
Back-mixing necessitates longer lengths to get the same results.
Heat transfer:
Heat transfer coefficients in jacketed reactors are 0.23 kW/m2K for metal- and
0.17 kW/m2K for glass-lined equipment.
Conservative heat transfer coefficients for condensers may be 0.54 kW/m2K and
1.1 kW/m2K for thermosiphon reboilers.
Separations:
Countercurrent multistage equals high efficiency in separations.
Column diameters should be at least eight times the packing size to minimise
large voids near the wall.
Column length and throughput per cross-sectional area should be held constant to
get identical results.
First pass guess for boil-up vapour rate in a distillation column may be 1.4 kg/sm2.
Use distillation if the relative volatility is greater than 1.2.
If two liquids are immiscible the infinite dilution activity coefficient is > 8.
Lower temperatures generally give a high relative volatility in distillation except
for alcohols in water.
10 percent salt in water usually doubles the activity coefficient of a dissolved
organic.
1000 ppm of a less dissolving organic is essentially infinite dilution.
The mass transfer unit is a better mathematical model than the theoretical stage
when the stripping factor or adsorption is high.
A stripping factor or adsorption of 5 is about as good as infinity.
Liquid distribution is the problem with a packed tower in 80 percent of the cases.
6 5
First guess at the residence time required in a decanter may be 20 minutes or an

up-flow of 0.09 g/sm2.
Total flow through a liquid-liquid extraction column may be 6.5 (m3/h)/m2 (1.8
mm/s) with a density difference of 0.2 kg m-3.
Solid particles tend to collect at liquid-liquid interfaces.
The relative separation factor in a liquid-liquid extraction column should be at
least 5.
Liquid-liquid extraction column volume can be reduced with mechanical agitation
if the interfacial tension is high.
Dissolving 2 to 20 percent organic solute generally reduces interfacial tension.
Freezing points may be suppressed 1°C for every 1.5 mol% impurity present.
A ratio of impurity concentration between a solid and a liquid phase greater than
0.2 is probably due to solid solution.
High viscosity often leads to supercooling in crystallisation.
Certain (parts of) processes often demand for the use of a pilot plant more often than
other (parts of) processes. This is due to the fact that important process properties
such as specific mass transfer area and velocity profiles are difficult to scale up
directly. The table below gives some scaling up rules of thumb for a number of unit
operations. Some comments are given in the last column [Aerstin et al., 1985]:
Process Pilot plant required? Comments

Distillation Usually not. Sometimes Skipping pilot plant can sometimes
needed to get tray produce surprises, for example,
efficiency or HETP foaming.
data.
Fluid flow Usually not for single Some polymer systems are also
phase. Almost always very difficult to predict.
for two-phase flow.
Reactors Frequently. Scale-up from lab often justified for
homogeneous systems, seldom for
heterogeneous.
Polymerisers Almost always.
Evaporators, Usually not unless
reboilers, heat there is a possibility of
exchangers, coolers, fouling
condensers
Dryers Yes - Almost always. Usually done using vendor
equipment.
Solids handling Almost always. Usually done using vendor
equipment.
Crystallisation Almost always. Many times done with vendor
equipment.
Extraction Almost always. Should pilot in similar type of
equipment proposed for production
plant.
6 6
6.5 Literature
• Pilot Plant Design, Construction, and Operation (1992), Richard P. Palluzi.
• Schaalvergroten, Dictaat TU Delft (1984), N.W. Kossen, P.G. Krouwel.
• Gaining experience in through pilot plants and demonstration units (1985), F.G.
Aerstin, L.A. Robbins, A.J. Vogel.
In: Scaleup of Chemical Processes - Conversion from laboratory scale tests to
successful commercial size design; Attilio Bisio, Robert L. Kabel (eds.).
• The use of pilot plants in scale-up (1958), Henry J. Ogorzaly.

In: Scale-up in Practice; Richard Fleming (ed.).
• The philosophy of pilot plants (1966), J.H. Chesters.

In: Pilot Plants in the Iron and Steel Industry; C.H. Inman (ed.).
• Theoretical aspects of the design and operation of models and pilot plants (1966),
A.V. Bradshaw.
In: Pilot Plants in the Iron and Steel Industry; C.H. Inman (ed.).
6 7

Scale-Up/Scale-Down: Lecture Notes

Uploaded by

Copyright:

Available Formats

Scale-Up/Scale-Down: Lecture Notes

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Scale-Up/Scale-Down: Lecture Notes

Uploaded by

Copyright:

Available Formats

LECTURE NOTES

Delft University of Technology

April 2002, Rob van der Lans, editor.

April 2006, Rob van der Lans

TIME AND CHANGE

From: FUTURE SHOCK by Alvin Toffler; Pan Books, London, 1970

Size limits of animals

Inv ~ Cap 0.6

General aspects of scale-up

first size enlargement until constraints, then scale-up by number

or for the density ρ of the reaction mixture:

Three kinds of physical quantities require special attention:

1.2. Calculations with scaling factors

By introduction of scaling factors scale-up problems become easier to calculate.

Example: flow of a liquid through a smooth, horizontal tube

Formulas: in scaling factors:

Input: kL, kD kρ, k cp , kη kΦm

In case of material identity: kη = kρ = k cp = 1 this simplifies to:

2.1. Introduction in similarity

Firstly, two kinds of similarity are distinguished:

Example 1: Introduction to dimensionless variables

For a Newtonian liquid (constant viscosity ) holds:

Integration with boundary condition: r = R  vx = 0 gives:

Instead of vx(r) we then use

Example 2: Introduction to constant characteristic dimensionless groups

In a first order reaction (r = krc) a dissolved component (concentration c) is reacting away.

With the boundary condition c = c0 at x = 0 integration then gives:

sub-process characteristic time

2.2. Process similarity levels

[accumulation] = [net transport] + [net production]

momentum heat mass dimension

For simplification we assume the material properties , , cp, , i to be constant in the

Filled out in the momentum balance this gives:

Not all dimensionless groups are equally important in all cases:

General conclusions about similarity are:

 At each level of process similarity it is possible to indicate the conditions: equal

3.1. Buckingham Π-theorem

Every relation between n physical quantities can be reduced to a

A physical system described by n physical quantities of which the

To clarify the concepts of the Π-theorem the following example is used.

det = 0⋅[(-1)⋅0 - 1⋅1] - 1⋅[1⋅0 - 1⋅(-1)] + 0⋅[1⋅(-1) - (-1)⋅(-1)]

log [F] = 1 log M + 1 log L - 2 log T

Note: system 2 contains exactly the same information as system 1 does.

Here the dimension matrix appears with all relevant information.

Using this into the rest matrix gives:

In this way we find two dimensionless numbers (= dimensionless groups):

Since squared dimensionless groups are still dimensionless,

In Π1' we recognise the Reynolds group (abbreviated: Re).

3.2. Independence of dimensionless groups

In general it can be stated that:

If A and B are dimensionless groups (dimensionless numbers), then

Two dimensionless groups A and B are independent if from

(Note: ‘≡’ means for all values of F, ρ, D, v, and η)

Working this out gives:

Faρ− a + b v −2a + b D−2a + bη− b ≡ 1

Re, Fr, and Ga are dependent because

Fr1Ga1Re-2 ≡ 1 ⇒ FrGa = Re2

The dimensionless groups

(1) the reduction method of Gauss-Jordan