Fundamentals of Particle Technology Holdich

Micropore Filtration
Sponsors of the book: Fundamentals of Particle Technology
The production costs of this book have been subsidised

by the generous support of Micropore Filtration,
specialists in particle separation and classification.
Micropore specialise in microporous filtration media

designed to separate and classify particles a few microns
and above in diameter. They also supply media for other
operations in particle technology. For more details see:
http://www.microporefiltration.co.uk
Fundamentals of
Particle Technology
Richard G. Holdich
Department of Chemical Engineering,
Loughborough University,
Leicestershire,LE11 3TU,
U.K.
Midland Information Technology and Publishing

Shepshed, Leicestershire, U.K.
http://www.midlandit.co.uk
Fundamentals of Particle Technology by Richard Holdich,
Chemical Engineering, Loughborough University is licenced
under the Creative Commons Attribution-Non-Commercial 2.0
UK: England & Wales License.
To view a copy of this licence, visit

http://creativecommons.org/licenses/by-nc/2.0/uk/
or send a letter to Creative Commons,
171 Second Street, Suite 300,
San Francisco, California 94105, USA.
This work has been released as an open educational

resource through the Open Engineering Resources
project of the HE Academy Engineering Subject Centre.
The Open Engineering Resources project was funded by
HEFCE and part of the JISC/HE Academy UKOER
programme.
You are free:
to copy, distribute, display, and perform the work.

to make derivative works.
Under the following conditions:
Attribution — You must give the original author credit.

Non-Commercial — You may not use this work for
commercial purposes.
The original book was published in the United Kingdom

by Midland Information Technology and Publishing
32, Bridge Street, Shepshed, Leicestershire,
LE12 9AD, U.K.
http://www.midlandit.co.uk
and had the ISBN 0-9543881-0-0
Whilst all due care has been exercised in the preparation of this book, the
author and publishers do not warrant that the information contained
therein is completely free from errors.
Registered names, trademarks, etc. used in this book, even when not
specifically marked as such, are not to be considered unprotected by law.
Preface
There are many industrial surveys reporting poor performance of processes involving
particles, whereas fluid only processes often reach 90 to 95% of design capacity. One
study reports only 6% of particulate processes examined reporting no major
performance problems (in R.A. Williams, CAPE-21, EUREKA project 2311). Clearly,
the presence of particles increases process uncertainties and this book is aimed at
reducing some of those uncertainties for a large engineering and scientific audience.
In many engineering disciplines Particle Technology is now a subject worthy of
study on its own, rather than a sub-section of fluid mechanics. Over the last twenty
years, or so, there have been several books authored on aspects of the subject. This is
an inevitable consequence of the subject’s breadth and the interests of those authors
within it. So, there is no lack of information for the serious student to turn to.
However, where does he, or she, start off? Fundamentals of Particle Technology was
written as a starting point for students new to the subject. The philosophy behind the
book is to provide a text that all students, hopefully, regard as accessible, both in cost
and style. Also, the nature of learning has changed significantly in recent years and
this book reflects some of these changes. There are numerous line diagrams and
illustrations included, but very few equipment pictures. Instead, web addresses are
provided for equipment manufacturers and interested students should access these
sites to see the equipment offered. This provides ready access to colour pictures of
modern machinery, rather than pictures that date all too quickly. The web sites are
those of major manufacturers that are likely to be available for many years to come
and, although the exact URL might alter, a small amount of intelligent browsing will
soon find the device, or its more modern equivalent. Likewise, the exercises at the
end of a chapter have further details of their solution via the Internet site
www.midlandit.co.uk/particletechnology. Most practical questions require a series of
calculations following from problem decomposition and the student may perform an
error at any stage; the multiple choice style is designed to take the student through
the calculations so that he, or she, cannot progress if an error is made on the way.
Thus, the student should know instantly if he, or she, has made a mistake. The
disadvantage of this style is that it provides limited training in problem
decomposition; so, the support web site has additional problems available for practice
in this.
Experience shows that journal references at the end of each chapter are rarely
consulted; so, where appropriate, references are given within the text and to named
research workers. A simple literature search will show up their work – it is not usual
for an author to publish only a single paper; hence it should be easier to look for a
name and context, rather than a specific paper. There is also a further reading section,
with recommended books that correspond to each chapter within this book.
Finally, I would like to express my gratitude to colleagues at Loughborough
University, both present and past, for helping me develop this style of teaching and
encouraging my interest of this subject.
Richard Holdich,
Loughborough, November, 2002.
Contents
1 Introduction 1
1.1 Prerequisites and objectives 2
1.2 The micron 2
1.3 Sampling 2
2 Particle characterisation 5
2.1 Particle size functions 7
2.2 Algebraic representation of size functions 9
2.3 Specific surface area per unit volume 12
2.4 Distributions: moments and conversions 13
2.5 Means of a distribution 15
2.6 Image analysis and particle shape 15
2.7 Example interpretation of distribution data 16
2.8 Summary 17
2.9 Problems 17
3 Fluid flow through porous media 21

3.1 Definitions 21
3.2 Flow regimes 22
3.3 Darcy’s law and Kozeny-Carman 23
3.4 Friction factor 24
3.5 Carman and Ergun correlations 25
3.6 Concentrations by mass and volume 26
3.7 Summary 26
3.8 Problems 27
4 Filtration of liquids 29
4.1 Deep bed and clarifying filtration 29
4.2 Cake filtration 31
4.3 Specific resistance and dry cake mass
per unit volume filtrate 32
4.4 Compressible cake filtration 33
4.5 Filtration modes of operation 35
4.6 Membrane filtration 38
4.7 Filter media 40
4.8 Filter aids 42
4.9 Summary 42
4.10 Problems 43
5 Dilute systems 45
5.1 Weight, drag and Particle Reynolds number 45
5.2 Other forces on particles 48
5.3 Particle acceleration in streamline flow 49
5.4 Settling basin design (Camp-Hazen) 50
5.5 Laboratory tests 51
5.6 Summary 52
5.7 Problems 52
6 Hindered systems and rheology 55

6.1 Hindered settling and zone theory 55
6.2 Batch settling flux 57
6.3 Thickener design 58
6.4 Kynch analysis 60
6.5 Compressible sediments 60
6.6 Homogeneous systems 61
6.7 non-Newtonian rheology 62
6.8 Summary 63
6.9 Problems 63
7 Fluidisation 67
7.1 Minimum fluidising velocity 68
7.2 Types of fluidisation 69
7.3 Bed design and bubbling behaviour 70
7.4 Gas flow patterns around bubble and stability 70
7.5 Davidson and Harrison model 71
7.6 Discrete element analysis 73
7.7 Summary 74
7.8 Problems 75
8 Centrifugal separation 77
8.1 Sedimenting centrifuges 78
8.2 Hydrocyclones 80
8.3 Filtering centrifuges 83
8.4 Washing and dewatering 85
8.5 Summary 88
8.6 Problems 88
9 Conveying 91
9.1 Heterogeneous flow in liquids 91
9.2 Dilute phase pneumatic conveying 92
9.3 Dense phase pneumatic conveying 94
9.4 Other conveying equipment 94
9.5 Summary 96
9.6 Problems 96
10 Powder flow and storage 99

10.1 Powder properties 99
10.2 Flow patterns and stress in a hopper and silo 101
10.3 Hopper opening and angle 102
10.4 The powder flow function 103
10.5 The hopper flow factor and hopper design 108
10.6 Measurement techniques and conditions 109
10.7 Summary 110
10.8 Problems 111
11 Crushing and classification 113

11.1 Energy utilisation 114
11.2 Crushing laws 115
11.3 Breakage and selection functions 116
11.4 Milling circuit matrix 117
11.5 Population balances 119
11.6 Summary 120
11.7 Problems 121
12 Solid/solid mixing 123

12.1 Binary component mixing 124
12.2 Specification and confidence 127
12.3 Equipment 128
12.4 Cohesive powder mixing 129
12.5 Summary 129
12.6 Problems 130
13 Colloids and agglomeration 131

13.1 Forces on small particles – in liquid medium 131
13.2 DLVO and applications 133
13.3 Coagulation 135
13.4 Flocculation 136
13.5 Forces on particles – gaseous medium 137
13.6 Agglomeration and granulation equipment 139
13.7 Summary 140
13.8 Problems 140
14 Gas cleaning 141

14.1 Target; grade and overall efficiencies 141
14.2 Collection mechanism 142
14.3 Dust collection material balance 145
14.4 Equipment types 146
14.5 Summary 148
14.6 Problems 149
15 Powder hazards 155

15.1 Explosion hazards 155
15.2 Physiological hazards 156
15.3 Summary 158
16 Case study 159
Nomenclature 164
Further reading 166
Appendix – Heywood Tables 168
Index 170
1 Introduction
Particle technology may be described as being the study of materials
dispersed within a continuous fluid. The particles may be solid, but
they can also be oil droplets in water, water droplets in air, etc. So, by
a particle we mean any dispersed material within a fluid. In many
cases deformable particles have a slightly different behaviour to rigid
ones, but the starting point for the description of deformable particles
is that of the rigid, and simpler, case. Hence, particle technology
includes the understanding of raindrops, oil emulsions, powders,
slurries, etc., and just about every industrial process uses the subject
at some stage. For example, in petrol production the catalytic
cracking of petroleum is achieved in fluidised beds of catalyst
particles (Chapter 7). An understanding of fluidisation relies upon
knowledge of particle characterisation and fluid flow through porous
media (Chapters 2 and 3). The petroleum processing is performed in
the vapour phase, not the liquid, hence the fluidised beds require
appropriate gas cleaning equipment for recycling and retention of the
catalyst particles (Chapter 14). The catalyst is stored and conveyed
into the system (Chapters 10 and 9) and, of course, due care must be
exercised over powder hazards (Chapter 15). So, even in the case of
an obviously liquid product, petroleum spirit, we encounter a
significant proportion of material covered in this book.
An even greater reliance on particle technology is provided by the
increasing trend towards high value batch processing in the chemical The Reynolds number
and pharmaceutical industries. A prime example is the production of Is a measure of the amount
a tablet. In many cases a reactant is provided in a solid form and of turbulence within a
product recovery involves nucleation and then crystallization of the system. It is numerically
the ratio of the inertial to
product. These two processes are not covered here, but the interested
viscous forces. Flow
reader is directed towards other works [J.W. Mullen, 1997,
Reynolds number for a
Crystallization, Butterworth-Heinemann, 3rd edition; R.J. Davey and fluid through a circular
J. Garside, 2001, From molecules to crystallizers, Oxford Chemistry pipe is:
Primers, No. 86]. Most of the remaining aspects of product recovery duρ
are covered in this work. The crystals may be settled, to increase the Re =
µ
slurry concentration going on to a filter, or filtering centrifuge; the
see the Nomenclature for
resulting cake will need washing free of reaction products and
definitions. Values above
unreacted feed material and mechanically dried, to minimise the 2000 are usually taken to
amount of thermal energy required to complete the drying. After indicate very significant
thermal drying (not covered here), there is likely to be a need for turbulence.
product storage, crushing and classification, solid/solid mixing,
conveying and agglomeration for the purpose of forming the tablet.
Any one of these processes may be the cause of a process bottleneck,
or throughput limitation, and the intention of this book is to provide
a sound understanding of the underlying principles behind these
operations to enable reliable operation and appropriate decisions to
be drawn.
2 Introduction
1.1 Prerequisites and objectives

In common with most engineering subjects particle technology
requires a basic competence in calculus and algebraic manipulation.
A first course in fluid mechanics would also be helpful, as it will be
assumed that the reader is familiar with concepts such as Reynolds
number (see box), Bernoulli’s equation, Hagen–Poiseuille, Newton’s
All sizes in microns
law (F=ma), friction factor and flow regimes. Knowledge of very basic
fine sand: 20 to 200 statistics would be useful for a complete understanding of solid/solid
hair diameter: 100 mixing, but the essential elements are included in Chapter 12.
clouds/fog: 30 Likewise, the concepts of mean, median and mode are used when
red blood cells: 8 discussing particle size distributions.
silt: 2 to 20 The primary objective of this work is to provide a sound basis on
clays: <2 which the more interested reader can build. Particle technology is a
tobacco smoke: <1 very broad subject and there have been many very good books
bacteria: 0.2 to 40
describing specialised aspects of the subject published over the last
viruses <0.5
twenty years.
1.2 The micron

Most industrial processing is performed on small particles with a
diameter typically in the range of 10−6 to 10−3 m; i.e. 1 to 1000 microns
or µm. The box provides some sizes, i.e. particle diameters, of
commonly encountered materials. You will become very familiar
with working in a length scale of microns in particle technology.
However, there are often times when the centimetre, gram, second
(cgs) system of units may be easier to apply.
1.3 Sampling
Processing is usually performed on small particles contained in bags,
drums, 1 tonne Intermediate Bulk Containers (IBC) or trucks.
Fig. 1.1 A grab sample Analysis for particle size, and other characteristics, usually requires
only a few grams. The most convenient method of obtaining a sample
of a few grams is to take a grab sample from the bulk, often scooping
the few grams off the top of the pile, or sack, see Figure 1.1. However,
segregation during particle motion is very common; examples include
the tendency for larger particles to rise to the top in breakfast cereals
such as muesli – and finer particles to fall between the gaps to the
bottom of the packet. Hence, a simple grab sample is rarely adequate
for a reliable analysis. The absence of a truly representative sample
invalidates any further work on the particles, be it laboratory tests for
some process (i.e. unit operation) or for particle size analysis and
characterisation.
When sampling from a suspension the viscous nature of the fluid
may assist in representative sampling; e.g. particles settle slower so
that stirring is often effective in suspending the particles for an
adequate time to obtain a good sample. However, at low
Fig. 1.2 A chute riffler concentrations the hindering (Chapter 6) effect of other particles is
not strong and vigorous stirring can lead to centrifugal separation
Fundamentals of Particle Technology 3
(Chapter 8). In air, or any gaseous medium, motion can help the
segregation of particles by differences in size, density, moment of
rotational inertia, coefficient of restitution, etc. Hence, sampling of
free flowing particles should be performed according to accepted
standards (BS 3406) and using well-known techniques such as:
sample cutters, cone and quartering (25 kg solids and above), chute
riffles (50 kg solids down to 25 g), see Figure 1.2, spinning riffler (1
kg down to 25 g), see Figure 1.3. Ultimately, a scoop, or grab, sample
from the suitably reduced mass of powder will be required for use in
an instrument if, as is usual, 25 g is still too much powder to be used.
When sampling a dilute stream of powder entrained in a gas stream,
isokinetic sampling is often recommended; i.e. equal fluid velocities in
the bulk flow and within the sample probe. Isokinetic sampling is
discussed in greater detail below. Fig. 1.3 Spinning riffler with
A sample cutter is used to take a sample from a flowing stream of vibratory feeder from hopper
powder, such as on a conveyer belt. At a transfer station, from one belt
to another, the powder will be dropped from the first belt on to the
second, and the sample cutter will intermittently move into the
falling powder to remove a sample. It is important that the cutter
removes a sample across the entire width of the falling powder, or
width of the belt, because segregation during transit may result in
particles of a certain size preferentially positioned on the belt.
Cone and quartering is a technique used extensively within the
minerals industries for coarse sized material up to several centimetres
in diameter, to obtain a more tractable sample size, and is often
applied before riffling. The entire contents of the container, or bag, to
be sampled are discharged on to a clean flat surface, which could be a
swept floor. The material is then made into a cone by shovelling
material from the sides to the centre to form a cone. The cone is
usually then flattened to minimise the likelihood of pushing material
at the top in any one direction. It is common for the larger particles to
roll to the periphery of the cone and the finer ones to stay near the
centre. However, so long as the cone is made evenly there should be
an equal distribution of larger particles all around the periphery of
the cone, with no bias towards any position. Thus, quartering the
heap by dividing it into four roughly equal portions should provide
samples representative of the original material. It is usual to then take
two quarters at opposite sides of the heap and remove them from the
sample. The remaining two form the sample and the process can be
repeated to reduce further the overall sample size. A similar
technique can be applied to smaller amounts of finer particles, subject
to adequate safety – health and environment considerations, coned
and quartered on a bench, but the process is operator dependent and
it is unlikely that a sample would be as representative as one
provided by the other techniques described above.
When sampling airborne particles from flow within ducts it is
important that the flow entering the sample filter is the same as that Fig. 1.4 Isokinetic sampling
in the duct, or pipe, see Figure 1.4. This is to prevent particles from a duct or pipe
4 Introduction
inertially separating at the entrance to the sample pipe, see Chapter

14 for details on inertial separation. If particles do inertially separate
the concentration in the sample tube will not be the same as within
the duct and any further analysis will be in error. Another important
factor evident on the illustration of isokinetic sampling is the lack of
sharp changes in direction between the sample entrance and the
sample filter. If a sharp direction change occurs the particles may
again separate from the gas flow by their inertia, leading to
deposition within the tube and a lower than true concentration of
dust found on the sample filter. It is easier to avoid sharp direction
changes in a three dimensional structure than can be represented on a
two dimensional illustration, such as Figure 1.4. Thus, it need not be
necessary to position a sample tapping just up-stream of a bend.
However, it is important that a sample stream should not be placed
just downstream of a bend. Most dust sample systems have a pump
and valve to control the flow entering the sample tube to ensure
isokinetic conditions and it may be possible to use a timer; whereby
the sampling is active for a fixed time, which corresponds to a fixed
volume of gas at a given flow rate. The particles deposited on the
sample filter may be weighed and they may be washed off the filter
and counted. This provides a concentration of dust particles within
the duct and a size distribution. However, it is important to
adequately disperse the dust sample after it has been washed off the
sample filter. This may require some practice with differing types of
chemical dispersing agents. Alternatively, microscope counting of the
test filter may be adequate, in the dry state, if the concentration of
particles is very low.
In some instances, two filters may be used in the sample holder:
one directly after the other. Thus, provided no particles are present
on the second filter and the masses of each filter were carefully
recorded before use, it is possible to weigh both filters and to deduce
the overall concentration of dust by the difference in mass between
the filter which acted on the sampled dust and the second filter which
should remain clean – after having corrected for any initial difference
in mass of the two filters. By placing both filters in-line any difference
in mass due to humidity, or similar, effects should be cancelled out.
Isokinetic sampling is important in gas streams, but not so
relevant to liquid streams because of the greater viscosity of the fluid
medium and the lower density difference between the particle and
fluid. Hence, particles are not so likely to separate inertially in liquid
systems. However, considerable care still needs to be exercised in
deciding where to withdraw samples from liquid systems, to ensure
that a representative sample is taken for analysis. Without a
representative sample the results from even the most expensive
particle characterisation techniques will not represent the particle
dispersion properties adequately.
2 Particle characterisation
An obvious question to ask is, ‘what is the particle diameter of my
powder?’ However, the answer is not so simple. Firstly, most
materials are highly irregular in shape, as can be seen in Figure 2.1 –
where should one make the measurement? Also, if we turn a particle
on its side it is likely that the measurement would be different. When
we have to verbally provide this information to someone, who cannot
see the particle, it becomes almost impossible to describe the particle
simply. In engineering, we wish to perform calculations using the
diameter; so, we need some simple basis for describing the
irregularly shaped particle that can be used in communication and
calculations. This is the origin of the concept of the equivalent spherical
diameter, in which some physical property of the particle is related to
a sphere that would have the same property, e.g. the same volume. Equations for spheres
Volume is easily measured. If the particle is big enough, water
displacement would work and the particle volume can be equated to circumference is πx
the volume of a sphere.
Note that we shall use surface area is πx 2
diameter rather than
radius and the symbol x π 2
rather than d. Also, it is projected1 area is x
4
common practice to talk π 3
about particle size, which volume is x
6
really means particle
6
diameter. specific surface is
A sphere is a readily x
understood geometric
shape and characterised
by a single dimension: its
diameter. If you have Fig. 2.1 Talc particles – as in talcum powder exercise 2.1
completed exercise 2.1 Calculate the equivalent
you will appreciate that, for the same particle, the equivalent spherical diameter of a 10 µm
spherical diameter depends upon the property selected for the cube, using equivalence by:
equivalence. Unless, of course, the particle is spherical in shape. perimeter,
Hence, it is a sphere that all particles are related to and not some projected area,
surface area,
other simple geometric shape, e.g. a cube.
volume,
Even though we can relate a measured property of our particle to
specific surface, and
that of a sphere we should still consider particle shape, as it can have
mesh size (i.e. sieve opening
an important influence on processing requirements. One simple way
size)
to quantify shape is using Wadell's sphericity (Ψ) where:
3 π 3
e.g. for volume 10 = xv
6
surface area of sphere of equal volume to the particle
Ψ= (2.1)
surface area of the particle
1projected area – what is observed when looking at a particle using a microscope

This uses the property that a sphere has the smallest surface area per
unit volume of any shape. Hence, the value of sphericity will be
fractional, or unity in the case of a sphere. There are a variety of
accepted shape descriptors and some of these are provided in Table
2.1.
Table 2.1 Common particle shape descriptions
Descriptor Wadell’s sphericity Example

spherical 1.000 glass beads, calibration latex
rounded 0.82 water worn solids, atomised drops
cubic 0.806 sugar, calcite
angular 0.66 crushed minerals
flaky 0.54 gypsum, talc
platelet 0.22 clays, kaolin, mica, graphite
Particle size analysis equipment is of fundamental importance in

particle technology, as it provides the values used in the calculations.
However, there are many different types of equipment and they
typically provide equivalent spherical diameters based on: volume,
projected area, chord length, area related to the light scattering
properties of the particle, etc. A table of selected devices, the
equivalent spherical diameter measured and links to appropriate
images and descriptions are included in Table 2.2.
Table 2.2 Commonly used particle size analysis equipment
Name Equivalent Example

spherical URL (www)
diameter
microscope projected area
sieve mesh size www.atmcorporation.com
Lasentec™ length www.lasentec.com
(particle chord length
FBRM)
Malvern™ area – light www.malvern.co.uk
(Fraunhofer diffraction) scattering
properties
Fig. 2.2 Example size Coulter Counter™ volume www.beckman.com
distribution – Malvern (electric zone sensing) see Multisizer
Sedigraph™ and sedimentation www.micromeritics.com
Note, for descriptions of Andreasen pipette see Sedigraph
principles of operation –
see the recommended In most cases instrumental equipment, such as the Lasentec, Malvern
web sites. and Coulter Counter, are first calibrated against near monosized
polystyrene latex particles of known diameter. Adjustments are made
within the operating software so that the equipment provides the
calibration material diameter. The equipment provides a full size
analysis, an example is provided in Figure 2.2 and Section 2.7. In
order to simplify subsequent description and calculations it is usual

to attempt to represent the full particle size distribution by a single
diameter. Unfortunately, there are many possibilities and the median,
or x50, is commonly (and usually mistakenly) applied. This diameter
is easily read off the cumulative distribution curve (see right).
At this stage, the reader may well reflect on the best way to
represent the particle size data. There are a number of options to
consider: which equivalent spherical diameter, which size analytical
equipment and which statistical diameter (mean, median, mode)? The
most appropriate technique for the end use of the data should be used. For
example, if the data is required to design a settling basin for effluent
removal a sedimentation diameter is best. If the data is to be used in
reaction engineering, or mass transfer calculations, then the
equivalent spherical diameter by surface area per unit volume (or
mass) is the most appropriate. In the latter case, it is the surface area
that dictates the reactivity and is used in the equations for these
subjects. However, it should be noted that the surface area exhibited
to fluid flow is not, necessarily, the same as that exhibited to reactants
in the gaseous phase. Catalyst particles often have a structure that
provides internal surfaces for reaction. This will be discussed again in
Chapter 7.
2.1 Particle size functions

One of the harder concepts to follow is that the particle size function
of the same powder depends upon the basis on which it is reported.
This is for similar reasons to the different values found in exercise 2.1.
An illustration of this is given in Figures 2.3 and 2.4, and the data
used to construct the curves is included. Firstly, it is a trivial matter to
take the total number of particles and convert to a fractional amount
in each size range (or grade). Hence, it is possible to plot the
cumulative number of particles undersize against particle size.
Clearly all the particles are less than 51 µm; hence, 100%, or 1.00 as a
fraction, are less than this size. At the next size increment boundary,
41 µm, there are 0.975 particles less than this size; so, the cumulative
number less than 41 µm is 97.5%, etc. Finally, there are no particles
less than 1 µm; hence, the cumulative number undersize is zero at
this point.
The above example covers a consideration of the number fraction
of particles within size ranges, or grades (for example: 31 to 41 µm). It
is quite common to want to consider the size distribution based upon
mass rather than number. This can be achieved by conversion from Fig. 2.3 Example distribution
the number distribution as follows. The mass of a single particle is
kv x3 ρ s
where kv is the volume shape coefficient. For spheres, the volume shape
coefficient is π/6; i.e. it is the factor that the diameter cubed must be
multiplied by in order to give the volume of the shape. Hence, the
mass of many particles within a size range will be
kv x3 ρs f
where f is the number of particles in that size range. The mass

fraction, within a size range and compared to the total mass of the
distribution (mi), will be the mass in the size range in question
divided by the mass of the entire distribution; i.e.
k x 3ρ f x3f
mi = v i 3 s i = i 3 i (2.2)
∑ k v xi ρ s f i ∑ x i f i
Equation (2.2) assumes that the volume shape coefficient and the
solid density remain constant throughout all the size grades and,
therefore, can be cancelled. Also, in order to apply the equation we
need to define a size to use for xi; this is usually the mid-point of the
grade. Thus, in the size grade 31 to 41 µm, the mid-point is 36 µm,
and it would be multiplied by the number of particles within the
grade to give a value proportional to the mass in that grade. The
constant of proportionality being the solid density multiplied by the
volume shape factor. On consideration of equation (2.2) it will
become apparent that a mass based distribution will provide greater
emphasis on the coarser particles than a number based one. This
follows from the simple fact that one large particle has much more
mass than one small particle: both represent a number distribution of
one, but the distribution of mass is much greater in the larger particle
grade. Problem 2, page 18, covers the conversion of data from a
number to a mass distribution. On consideration of equation (2.2) it
should be apparent that a mass distribution is exactly the same as a
volume distribution: to convert from volume to mass in each grade
the true solids density is used, but to convert from mass in each grade
to mass fraction the summation of the mass in each grade is divided
into the component grade masses and the density will, in fact, cancel.
Hence, the terms volume and mass distribution are used
interchangeably.
Mass and number distributions are frequently met within
industry, but distributions by length and area also exist. In the latter
case it may be by projected area (like looking down a microscope) or
surface area; see page 5 for the equations of these for a sphere. The
analogue expression for projected area to equation (2.2) is
k p xi 2 f i xi 2 f i
2
= 2
(2.3)
∑ k p xi f i ∑ xi f i
where kp is the projected area shape factor, π/4 for spheres.
In the construction of Figure 2.3 the cumulative number undersize
curve was considered. It is possible to convert the data to mass
fractions, using equation (2.2), and to plot the cumulative mass
undersize. Likewise, length and area undersize curves can be plotted
and these are illustrated in Figure 2.4. So, for the same powder we have
illustrated four different functions to represent the size data. Each
Fig. 2.4 Sizes for the same function will have a different mean, median, etc. So, it should be
powder – different functions becoming apparent why representing, or designing, systems with
particles in them is difficult! In correspondence it is most important
to specify whether the distribution is by number, area, mass, etc.
Considering the data used to construct Figure 2.3, it would be

possible to plot the number of particles within a size grade as a
histogram, rather than the cumulative data considered so far.
Further, if the mid-points of the size grades are used, it is possible to
plot a smooth curve to represent the distribution of particle sizes on
an actual number or number fraction basis. This is illustrated in
Figure 2.5. In summary, we have two types of curves to represent the
particle size data: the cumulative curve that starts at 0 and goes up to
1 and the distribution, or frequency, curve that starts at 0 goes
through a maximum and then returns to zero. The latter curve is
often illustrated as a bell shaped curve, whereas the cumulative
curve is usually an ‘S’ shaped curve. Both of these curves are
obtained from the same original set of data, so it is logical to deduce
that there is a simple mathematical relation between them. It should
be easy to spot that the number fraction distribution, or frequency, Fig. 2.5 Histogram and
curve in Figure 2.5 can be used to construct the cumulative curve distribution curve for data
illustrated in Figure 2.3: for a given particle size the number fractions used in Figure 2.3
of particles below that size are simply added together. At the top size
(i.e. xmax) all the number fractions added together must equal unity
because all the distribution is less than that size.
Distribution curves similar to Figure 2.5 are useful for comparison
purposes and for modelling particle processes, as we will see in later exercise 2.2
chapters. In Figure 2.5 it is easy to see which grade contains the mode What is the approximate area
and it would be possible to compare this figure with a similar under the curve in Figure 2.5?
distribution curve to compare size distributions. The distribution
curve in Figure 2.5 has been normalised to take into account the
micron range; i.e. the total area under the curve is unity. There are
two main reasons for this procedure: one reason is illustrated by
Figure 2.6, the other is discussed later under Figure 2.7. The same
overall data was used to plot Figures 2.6 and 2.5, but the interval
sizes considered were changed: the lowest two intervals were added
together and the 21 to 31 µm class was divided equally in two.
Clearly, this is the same powder, but we can make the distributions
look very different by our choice of interval, or class, sizes. The data
in Figure 2.6 hasn’t been divided by the micron range. Thus, one
reason for normalisation is to produce a fair basis for comparison
between the distributions as, after normalisation, the curves would be
similar.
2.2 Algebraic representation of size functions

In the previous section the cumulative and distribution functions to
represent particle size data were introduced. Also, conversion
between distributions by number, length, area and volume were
considered. For these conversions from a number distribution, we Fig. 2.6 Same data as Figure
must raise the particle diameter to the power 0, 1, 2 and 3 in order to 2.5 but different class sizes
convert to a number, length, area and volume distribution
respectively. This gives us a convenient shorthand by which to
represent the distribution data and a starting point for algebraic
manipulation; for the frequency curves:
n0(x) represents a number distribution

n1(x) represents a length distribution
n2(x) represents an area distribution
n3(x) represents a volume or mass distribution
Similarly, the cumulative size functions are represented by:
N0(x) represents cumulative number
N1(x) represents cumulative length
N2(x) represents cumulative area
N3(x) represents cumulative volume or mass
In fact, the distribution functions n*(x) represents the fractional
number/etc. of particles per micron range. Hence, integrating n0(x)
with respect to particle diameter from xmin to xmax will provide the
total fractional number of particles below size xmax; which is, of
course, unity. Thus, the mathematical relation between n0(x) and N0(x)
is formally written as
dN 0 ( x )
n 0 ( x) = (2.4)
dx
i.e. the cumulative curve is the integral form of the distribution curve
and the distribution curve is obtained by differentiating the
cumulative curve. This has some logic, as the cumulative S shaped
curve starts at low sizes with a small positive gradient, goes through
a period of high rate of change with size, then ends with a period of a
small positive gradient again. Hence, the distribution curve has the
characteristic bell shape: low values – high values – low values.
The above size functions and distribution curves are generic: they
do not imply a specific functional form for the equations. However,
numerous functions have been applied to particle size distributions
over the years. We will consider only the most common ones here,
but before providing the equations it is worth mentioning that the
functions have good practical use when trying to understand what is
Fig. 2.7 Illustration of the taking place within a process. For example, during crushing it may be
relation between the possible to inspect the size distribution of the mill product and fit
cumulative and two, or more, well-known distribution functions, suitably weighted
distribution curves – after in terms of x% of one and y% of the other. This would probably give
normalisation rise to a bi-modal distribution curve and this implies that more than
one crushing mechanism is taking place in the mill. Alterations of the
By normalising the mill conditions could increase the bimodal nature, or remove it
distribution curve altogether. A careful track, and modelling, of the size distribution
(dividing by the micron data enables the significance of the change in mill conditions to be
range) it is possible to evaluated. Hence, process decisions can be arrived at from the
easily relate the two distribution data. This principle is illustrated in Chapter 11.
types of curves: the
cumulative is the integral Rosin-Rammler Bennett – cumulative function
of the distribution curve, n
  x 
up to any size x, and the N 3 ( x ) = 1 − exp −   
distribution curve is the   x 63.2  
   (2.5)
differential of the
cumulative curve. where x63.2 represents the particle size at which 63.2% of the
distribution lies below and n is a constant called the uniformity index.
Gaudin-Schuhmann – cumulative function

n
 x 
N 3 ( x ) =  
 (2.6)
 x100 
where x100 represents the top particle size of the distribution and n is
again a uniformity constant.
Broadbent-Callcott – cumulative function

1 − exp(− x / x100 )
N 3 ( x) = (2.7)
1 − exp(−1)
This is a single parameter model: only the top-size is required. All
previous models need two parameters.
Gaudin-Meloy – cumulative function

n
 x 
N 3 ( x ) = 1 − 1 − 
 (2.8)
 x100 
Normal Distribution function (Figure 2.8)
1  − x − x 50 
n 3 ( x) = exp 2
 (2.9)
σ 2π  2σ 
In a normal distribution the median, mode and mean are all the same.
This is still a two parameter model: x50 (median size) and x84 is used to
determine the standard deviation from
Fig. 2.8 Illustration of the
σ = x 84 − x 50 (2.10)
Normal distribution
this uses the property that 68% of the distribution lies one standard
deviation away from the mean and the distribution is symmetrical.
Log-normal Distribution function (Figure 2.9)

x 84
σg = (2.11); x m = cx 50 (2.12)
x 50
This distribution is biased towards the coarser end, again x50 and x84
define the required parameters, but a geometric standard deviation
and mean are calculated. In practice, the distribution has to be Fig. 2.9 Illustration of the
truncated at some maximum and minimum sizes. log-Normal distribution
1  − 1 / 4b 
b= 2
c = exp(−1 / 2b) Z = (b / π )0.5 exp 
2 ln σ g  xm 
n 3 ( x) = Z exp[−b ln 2 ( x / x m )] (2.13)
2.3 Specific surface area per unit volume

The surface area of a single sphere is πx 2 and the volume of a sphere
is πx 3 / 6 ; so, the specific surface area per unit volume (Sv) for a single
spherical particle is the former divided by the latter, or
6
Sv = (2.14)
x
The specific surface area per unit mass (Sw) for a single spherical
particle is: Sv/ρs. Hence, the SI units are m−1 and m2 kg−1, respectively.
It has become conventional to abbreviate the full name to simply
specific surface and, throughout the rest of this book, when specific
surface is referred to it implies by volume and not mass. It is a very
important parameter because of its relevance to fluid flow problems
and reactivity. In the former case it is the friction of fluid flowing
over the surface of the particles that gives rise to an energy loss, or
pressure drop. In the case of reactivity, a high specific surface is
useful for the provision of many reaction sites in catalysts. For the
same volume of solids small particles have a much higher surface
area than larger ones – see the simple illustration on the left.
The specific surface can be extended to a full size distribution by
considering the total surface area divided by the total volume of the
distribution, for spheres
π ∑ xi 2 f i 6∑ x i 2 f i
Sv = = (2.15)
π 3 ∑ x 3
f
x
∑ i i f i i
6
The fractional number of particles (ni) could be used instead of the
actual number within the increment because
f i = ni ∑ f i (2.16)
and, after substitution, the total number of particles would appear on
both the top and bottom of the equation and cancel. In many cases
the distribution data is in mass, rather than number, terms. Equation
(2.15) can be simply converted to mass terms by first considering the
definition of mass fraction within an increment, equation (2.2), and
rearranging to provide
m
xi 2 f i = i ∑ xi 3 f i
xi
which can be substituted into equation (2.15). The summation of the
entire distribution volume is now on both the top and bottom of the
resulting equation and cancels, to leave the simple equation
m
S v = 6∑ i (2.17)
xi
So, unexpectedly, it is a simpler calculation to determine the specific
surface from a mass, rather than a number, distribution. A similar
derivation can be written using the continuous functions introduced
in Section 2.2. The analogues to equations (2.15), (2.2) and (2.17) are
6 ∫ x 2 n 0 ( x )d x
Sv = 3
(2.18)
∫ x n 0 ( x )dx
x 3n 0 ( x)
n 3 ( x) = (2.19)
∫ x 3 n 0 ( x )d x
n ( x)
S v = 6 ∫ 3 dx (2.20)
x
We can now define another equivalent spherical diameter that may
be used to represent the entire size distribution: the spherical particle
diameter that has the same specific surface as that of the distribution.
Rearranging equation (2.14) provides this value
6
xSv = (2.21)
Sv
This is called the Sauter Mean Diameter, and is sometimes represented
as x(3,2) in some distribution data. It is of great importance in particle
technology because it is the most appropriate equivalent spherical
diameter to represent the size distribution in fluid flow calculations.
An example of this is given in Section 3.4 in the next chapter.
2.4 Distributions: moments and conversions

In statistical terms the moment (Mn) of a probability function P(x) Recommendation
Leave Sections 2.4 and
taken about a point a is defined by
2.5 until after you have
M n = ∫ ( x − a) n P( x)dx (2.22) completed the Problems.
hence if a is equal to the mean and the first moment is considered
(n=1) then the result of equation (2.22) will be zero. The second
moment (n=2) will result in the variance (standard deviation
squared).
In particle technology we often need to integrate distribution
functions, see equation (2.19) for example. If, in equation (2.22), the
point a is taken to be zero then the moment is
M n = ∫ x n P( x)dx
but the probability function could be by number, mass, etc. Hence,
we need to include this in our specification as follows
xmax
M k,r = ∫ x k n r ( x )dx (2.23)
xmin
In the case of k=0, then we have the simple result

xmax
∫ n r ( x )d x = 1 (2.24)
xmin
which is valid regardless of the basis of the distribution data

(number, mass, etc.) because the probability function is based on
fractional number (mass, etc.) per micron range. Hence, an
integration, or summation, of all the fractional numbers per micron
range, from the minimum size to the maximum size, will result in the
fractions adding up to unity. Graphically, this is the same as saying
that the area under the distribution curve must be equal to unity, see
Figure 2.7.
In the non-trivial case (k≠0), it is possible to encounter functions
such as
xmax
π
ρ s ∫ x 3 n 3 ( x )d x (2.25)
6 xmin
which represents the mass weighted mean particle mass for a
distribution, that is of some importance in solid/solid mixing. The
weighting is towards the mass distribution because of the term n3(x)
and it is the mean particle size of the distribution multiplied by the
volume shape factor and solid density – thus giving the mean particle
mass.
Conversion between distributions was first introduced in equation
(2.2) and its analogue (2.19). In general, we have
x r n 0 ( x) x r n 0 ( x)
n r ( x) = xmax
= (2.26)
r M r,0
∫ x n 0 ( x ) dx
xmin
Substituting equation (2.26) into (2.23) results in
k r
∫ x x n 0 ( x )d x
M k + r,0
M k, r = = (2.27)
M r ,0 M r ,0
Using equation (2.27) it is easily possible to show that
M 2 ,0
M −1,3 =
M 3, 0
which is consistent with equations (2.18) and (2.20) combined.
Another illustration of conversion between distributions is to
change a mass distribution into a number one, considering both
discrete data and continuous functions as follows. If we substitute
equation (2.25) into (2.2) and use mid-points to represent the
increment we arrive at the following equation for mass fraction
k v xi 3 ρ s f i x i 3 ni
mi = 3
=3
(2.28)
∑ k v xi ρ s f i ∑ x i ni
where ni represents the fraction by number of the distribution within
the increment considered. We must now rearrange for the number
distribution
m
ni = 3i ∑ x i 3 ni (2.29)
xi
At first sight, equation (2.29) is without solution because the number
fraction appears on both sides, but the summation will result in a
constant which is multiplied by all the incremental values of mass
fraction divided by mid-points. Hence, simply dividing the mass
fraction in the increment by the mid-point cubed, summing all these
values up and then dividing through by this summed value will
provide the number fraction for each increment, as the constant value
(i.e. the summation) will cancel. Rearranging equation (2.26) provides
the analogue equation for the continuous functions
n ( x)
n 0 ( x ) = 3 3 M 3, 0
x
2.5 Means of a distribution

The simplest way to calculate the mean of a distribution, based on
discrete data, is to use the fractional amount by number, mass, etc.
thus
x 0 = ∑ x i ni (2.30)
x 3 = ∑ xi mi (2.31)
where equation (2.30) is mean particle size by number and (2.31)
mean by mass. Further consideration of the means by this approach
are given in Problem 3, at the end of the chapter. However, a more
complete approach considers that the distribution could be weighted
by number, mass, etc. and that we require a single particle diameter Fig 2.10 Graticule used to
that is representative of the entire distribution based on number, compare particle sizes on a
mass, etc.; i.e. microscope – BS 3260
xmax
x k,r k = ∫ x k n r ( x )dx = M k,r (2.32)
xmin
Hence, the general equation for mean particle size is

xk,r = k M k,r (2.33)
2.6 Image analysis and particle shape

Microscope counting of particles is an important characterisation
technique (BS 3260), but when performed manually it is slow and
prone to operator variation. The technique compares the size of the
particle to circles of standard areas, see Figure 2.10, and to count the
number of particles between each size range. Modern techniques
have removed the tedium of assessing particle size against standard
areas by eye and use computer image recognition for the size
analysis. However, they still usually rely on there being no
overlapping particles present. Other forms of electronic image
analysis are possible, including shape assessment. Hence, the
microscope (with computer) is an important tool in particle
characterisation. A microscope is essential for the quick check of
material to assess, for example, if the particles may be aggregating
during a water based particle size analysis by one of the instrumental
techniques mentioned in Table 2.2. An optical microscope is generally
used for particles down to 1 micron in diameter, but the theoretical
limit is slightly below this. Electron microscope images provide high
resolution much this size, see Figure 2.1 for an example.
In automated microscope analysis three diameters are often used:
Feret, Martin and Image Sheared. These diameters are illustrated in
Figure 2.11, together with the equivalent spherical diameter by
projected area. The Feret diameter is the perpendicular distance
between two parallel tangents on opposite sides of the particle, where
the tangents are drawn perpendicular to a fixed direction during the
travel of the microscope stage. The Martin diameter is the length of a Fig. 2.11 Statistical diameters
line which bisects the image of the particle. The line may be drawn in used in image analysis: Feret;
any direction but, once chosen, the direction must remain constant Martin and Image Sheared
for all measurements of the distribution. The Image Shear diameter is
computer generated by taking the image of the particle and moving it

until it is just touching the original image at an edge. The
displacement distance to achieve this is the image sheared diameter.
Particle shape is a very complex property to measure. In practice,
the shape descriptors and approach commonly used was covered in
equation (2.1) and Table 2.1. Shape can considerably influence
particle processing; an example is in the sedimentation of particles
with a low sphericity – these tend to descend like a feather oscillating
from side to side, whereas isotropic particles of the same size would
fall in a single dimension. An alternative approach to the description
of particle shape has considered using ratios based on elongation,
particle length to width, and flakiness, particle width to thickness.
Particle shape has also been described by fractal analysis, as
illustrated in Figure 2.12. However, there has been only very limited
success in quantitatively relating particle shape, by any of these
means, with particle processing requirements or products.
2.7 Example interpretation of distribution data

Figure 2.13 shows the tabular results from a Malvern Mastersizer size
analysis. The ‘Results Statistics’ are reproduced in Table 2.3 for
further discussion. The full data can be seen as a jpg file and the
associated spreadsheet calculation for this section can be found from
http://www.midlandit.co.uk/particletechnology/chapter2.htm
Table 2.3 Illustration of Malvern Results Statistics
Density = 1.000 g/cub.cm Specific S.A. = 0.7046 sq m / g

Mean diameters: D(v,0.1)=3.25 um D(v,0.5)=29.8 um D(v,0.9)=107.1 um
D(4,3)=44.11 um D(3,2)=8.52 um
Malvern use D to represent particle size, against the use of x in this

book. The term D(v,0.1) represents the particle diameter at which 10%
of the distribution is below; and similarly for the other percentiles on
Fig. 2.12 Fractal analysis of
the same row of the table. Thus, the median size is 29.76 µm. The
a particle: perimeter
mean diameter by volume (i.e. same as mass) is 44.11 µm, calculated
increases with smaller step
by equation (2.31), and the mean diameter by surface area to volume
sizes (λ) –rate of increase in
(i.e. Sauter mean) is 8.52 µm, calculated by equations (2.17) and (2.21).
perimeter against
The specific surface area comes from equation (2.17), and the value of
decreasing step size gives
0.7046 square metres per gram is reported. The equation provides a
the fractal dimension.
value in µm−1 but, for the Malvern calculation, the density has been
Maximum particle
given the value of 1 gram per cubic metre so that the reported specific
dimension normalises step
surface is, in fact, in µm−1. Using the size increment data from the
size and perimeter.
analysis and the equations in a spreadsheet provides the values of
44.2 and 8.5 µm for the mean by mass and Sauter mean diameters,
which is within rounding errors to that reported here by the Malvern
software.
2.8 Summary
When using particle size distribution data the most important
considerations to be aware of are: how the data was obtained, i.e. the
basis of the analysis by number, mass, etc., and how the data will be
used. For routine quality control, or assurance, the first principle is
not so important as simply a change in results will be enough to
signal a change in quality. However, these principles are important
for equipment design, or understanding. For design, the equivalent
spherical diameter, or distribution data, appropriate to the end use
should be used; e.g. settling diameter for sedimentation basin design,
Sauter mean for pressure drop calculations.
From a practical point, the repeated testing of samples of the same
powder in a size analysis laboratory can result in significantly
different particle size distribution data. This is a consequence of
variations in both sampling as well as size analysis. Modern
computer based electronics has resulted in size analysis equipment
providing many channels, 256 and 1024, and the use of spreadsheets
has made manipulation of the data relatively easy to accomplish. Fig 2.13 Example tabular data
Hence, the previous practice of fitting mathematical functions, or output from Malvern – the
undertaking graphical analysis of the data, in order to obtain values original is available for
viewing at:
such as the specific surface, has become less important. The level of
www.midlandit.co.uk/particl
effort required to obtain a graphical solution is rarely justified when
etechnology
the distribution data changes from one analysis to another.
The Results Statistics section
Invariably, the requirement is for a single value to be used for design
is reproduced in Table 2.3
(such as an average particle diameter), or comparison purposes. This
value can be adequately obtained from a spreadsheet analysis.
However, given the possible lack of repeatability of the size analysis,
and sampling, it may be wise to obtain a range of values for the
average diameter; as a small change in its value might significantly
influence a design variable, see Chapter 3 for an illustration of this in
the context of fluid flow.
For a more thorough modelling of particle processes, the particle
size distribution functions covered in Section 2.2 have considerable
utility, as they can be used to investigate, or model, what happens
within a process to the different sizes of particles.
2.9 Problems
1. The following size distribution was obtained for glass spheres by
microscope examination. Complete the table rows showing the
relative number of particles, the cumulative number undersize and
the relative (i.e. fractional) number per micron range.
Size range (µm): <5 5-7 7-10 10-15 15-20 20-30 30-40 40-50 >50
Number in range: 0 50 150 200 55 45 20 5 0
Relative number: 0 0.0952
Cumulative no. 0 0.0952 1.000
undersize:
Relative no. per µm: 0 0.0476
2.
The mass of a single
Using the above number distribution data, complete the table:
particle is:
π 3
x ρs Size range (µm): <5 5-7 7-10 10-15 15-20 20-30 30-40 40-50 >50
6
(mid point)3.frequency 10800
where x is particle
Relative mass: 0 0.0039
diameter and ρ s is solid
Cumulative mass 0 0.0039 1.000
density. Thus it is possible
undersize:
to convert a number
distribution to a relative
3.
mass distribution by
The mean particle size by number is best calculated by determining
multiplying the number of
the fractional contribution to the mean from each size range, or grade.
particles by the mass of a
single particle and
This is the product of the relative number and the mid-point size in
dividing through by the the grade. Summing all the contributions together gives the mean
total mass of the system. size. Note that this is mathematically equivalent to the more
The range, or grade, mid- conventional:
point is used for the ∑ f i xi ∑ f i
particle diameter. Note
that the volume shape where fi is the frequency (or number) of occasions that xi occurs, e.g.
factor and density cancel you may have used this formula before to calculate an average mark.
out during this operation.
The relative number can select the correct answer from the following
also be used in preference Thus the mean size by number is (µm):
to the actual number for a a: 12.4 b: 13.5 c: 25.0 d: 28.2
similar reason.
4.
The mean size by mass, calculated in a similar way, is (µm):
a: 12.4 b: 13.5 c: 25.0 d: 28.2
mi 5.
S v = 6∑ The complete distribution specific surface area per unit volume (Sv) is
xi
the total surface area divided by the total volume. It may be
where mi is the calculated by several routes, one is shown left. For this distribution Sv
relative mass - see
is (µm−1):
Section 2.3
a: 0.213 b: 0.240 c: 0.264 d: 0.444
Now convert your answer into one with SI units.

6.
You will need the equations relating specific surface (Sv) to number
frequency and also to mass fraction (both per micron). See Section 2.3 Number distribution:
∫ x n ( x )d x
for the derivation of the last relation starting from the former. For 2
π 0
spherical particles the two equations are given in the box on the right. Sv =
π /6
∫ x n ( x )d x
3
These questions are based on an idealised cumulative size 0
distribution that can be represented as a single straight line on the Mass distribution:
upper figure given to the right. π n3 ( x )dx
Sv =
π /6 ∫ x
(i). The cumulative size distribution of a powdered material may be Relation between two
represented as a straight line on a % number undersize versus types of distribution
particle diameter (x) graph passing through the points 0% by number curves (given for a
at 1 µm and 100% by number at 101 µm. The equation for N0(x) is: number distribution):
N0(x)=... dN 0 ( x )
a: x-1 b: x/100-1 c: x/100-0.01 d: x-100 = n0 ( x )
dx
Note, work in fractional terms and NOT percentages.
The relation between the
mass distributions is
(ii). The equation for n0(x) is: n0(x)=...
similar.
a: 1 b: 1/100 c: 100 d: x/100-1
(iii). What is the shape of the n0(x) graph given by your answer to
part (ii)?
(iv). The best equation to use for calculation of specific surface is:
101 2 101 2
∫1 x (1 / 100)dx 6 ∫1 x dx
a: S v = 6 101 3
b: S v = 101
c: S v = 6 101 3
∫1 x (1 / 100)dx ∫1 xdx ∫1 x dx
(v). The specific surface area per unit volume is (µm−1):

a: 0.013 b: 0.045 c: 0.059 d: 0.079
(vi). Now convert your answer into one with SI units:
(vii). If the sphericity of the material is really 0.9 the specific surface
is (µm−1):
a: 0.071 b: 0.088 c: 0.066 d: 0.053
7.
(i). The specific surface of another size distribution having the same
limits as that given in question 6 but on a mass distribution N3(x)
basis is (µm−1):
a: 0.013 b: 0.138 c: 0.277 d: 0.554
(ii). The Sauter mean diameter of the distribution in part (i) is (µm):
a: 50.0 b: 43.5 c: 21.7 d: 10.8
8.
The attenuation of a parallel beam of light is used to monitor
visibility on an airport approach during fog conditions. It is found,
experimentally, that the droplet size distribution of the fog may be
represented by the equations
for: 50≥ x ≥15 µm N 0 ( x ) = 0.83 + 0.043 ln( x )
15≥ x ≥4 µm N 0 ( x ) = 0.30 + 0.239 ln( x )
4≥ x ≥0.5 µm N 0 ( x ) = 0.21 + 0.304 ln( x )
where N0(x) is the cumulative fraction by number less than the
particle size in µm.
(i). Write down below the equations for n0(x)
for: 50≥ x ≥15 µm n0(x)=
15≥ x ≥4 µm n0(x)=
4≥ x ≥0.5 µm n0(x)=
(ii). The specific surface area per unit volume (Sv) of the fog is the
total surface area divided by the total volume, in number distribution
terms (assuming spherical droplets):
4 15 50
2 2 2
∫ x n 0 ( x )dx + ∫ x n0 ( x )dx + ∫ x n 0 ( x )dx
S v = 6 04.5 4
15
15
50
3 3 3
∫ x n0 ( x )dx + ∫ x n 0 ( x )dx + ∫ x n0 ( x )dx
0.5 4 15
Just considering the size grade 50≥ x ≥15 µm the area integral in the
∫ x n ( x)dx ) is (µm ):
2 2
above equation (i.e 0
a: 48.91 b: 12.35 c: 293.5 d: 1.286

N.B. For Part (vi)
The attenuation of a (iii). The volume integral for the same grade is (µm3):
light beam is given by a: 5895 b: 48.91 c: 1743 d: 102.1
the Beer-Lambert law:
I  1  (iv).The area and volume integrals for the other size ranges are as
= exp− S v CL
Io  4  follows:
(for convex particles)
for: 15≥ x ≥4 µm area is 24.98 µm2 volume is 263.8 µm3
where C is the volume 4≥ x ≥0.5 µm area is 2.39 µm2 volume is 6.473 µm3
concentration of the The specific surface area per unit volume of the fog is (µm−1):
fog, L is the path length, a: 0.0379 b: 0.227 c: 2.96 d: 0.493
and I and Io are the
intensity of the (v). Now convert your answer into one with SI units:
transmitted light and
the intensity of the light
with no fog present (vi). If the path length is 50 m, the concentration of fog that will
respectively. reduce the intensity of the beam by a factor of 100 is (-):
a: 1.6x10-6 b: 1.6x10-5 c: 1.6x10-4 d: 1.6x10-3
3 Fluid flow in porous media
In Chapter 2 we considered how to represent a particle size
distribution by, where possible, a single term that is representative of
all the particle sizes. This term may then be used for modelling,
design or simply to understand a process within Particle Technology.
One such example is in the fluid flow through a porous medium, or
porous media (plural). There are a number of practical applications of
fluid flow, including filtration, flow in a packed column, permeation
of water, or oil, within the matrix of a porous rock, etc. Before
discussing the consequences of our choice of a single value to
represent the distribution, and the appropriate modelling equations,
we must define the commonly used terms.
3.1 Definitions
By definition, a porous medium consists of pores between some
particulate phase, contained within a vessel, or some control volume,
as illustrated in Figure 3.1. The fluid flow rate through the bed is Q
(m3 s−1) and the bed cross sectional area is A (m2). Thus the superficial
(or empty tube) velocity U0 is the total flow rate divided by the cross
sectional area. The existence of the particles within the bed will
reduce the area available for fluid flow; i.e. to preserve fluid
continuity with the entering superficial flow the fluid will have to
squeeze through a smaller area; hence the velocity within the bed (U –
interstitial velocity) will be greater than the superficial. In Particle
Technology calculations it is the volume fraction that is most
important, and not the mass fraction. The volume fraction of solids
present (i.e. volume solids in bed divided by total bed volume) is
usually referred to simply as the volume concentration, or solids
fraction, and the remaining fraction is that of the voids. The void
fraction is also called the voidage and the bed porosity. It is important
to realise that, in liquid systems, the voids are usually filled with
liquid and not to assume that the bed consists of just solids and air.
The porosity is usually an isotropic property (i.e. the same in all
directions); hence, the interstitial velocity is simply related to the
superficial velocity by the following expression, which comes from a Fig. 3.1 Illustration of
consideration of fluid continuity. fluid flow through a
porous medium and
U
U= o (3.1) consideration of the
ε volume fractions present
Clearly, the resistance to fluid flow through the porous medium is
related to the amount of particles present, or volume concentration,
but it is conventional to work in terms of bed porosity. At one
extreme, when the bed is full of solids (porosity is zero – possible exercise 3.1
with cubic particles placed carefully within the bed) the resistance is Using continuity:
infinite. At the other, when no solids are present and the porosity is i.e. Q = constant,
unity, the interstitial velocity will be the same as the superficial deduce equation (3.1)
velocity. The resistance to fluid flow gives rise to a pressure drop in
the fluid (∆P). Pressure is not a vector quantity, but a pressure

Measuring porosity gradient with respect to distance (∆P/L) is. The pressure decreases in
For a dry powder in air: the direction of the fluid velocity, hence the pressure gradient should
fill a weighed
be negative: (-∆P/L). However, for the sake of brevity, we will adopt
measuring cylinder to a
graduation, gently
the term pressure difference, which is a scalar quantity, hence the
vibrate and reweigh. negative symbol will not be used in the following text.
The bed mass over The porosity of a packed bed of material depends strongly upon
volume will give the the nature of the particles and how the bed has been treated.
bulk density (ρb) of the Incompressible glass beads, including marbles, pack to a porosity of
powder: about 45%, but beds of alumina particles (not catalyst pellets) often
ρ b = (1 − ε ) ρ s + ερ reach porosities of 75%. Biological material, such as yeast cells, may
As the fluid is a gas, its form a packed bed with porosities of 90%, or higher. Therefore, it is
contribution to the bulk dangerous to make the common assumption that packed beds are
density is minimal and 50% solids and 50% voids. It is a simple parameter to measure, so
ε = 1 − ρb / ρs long as the solid and fluid densities are known; see the box on the
left.
3.2 Flow regimes

On page 1 the flow Reynolds number was stated. In a porous
medium, as with all fluid flow problems, we need to consider energy
losses from the fluid due to viscous and form drags. The former can
be simply referred to as laminar flow (low Re) whereas turbulent
flow has additional drag due to eddies in the fluid within the porous
medium. The Modified Reynolds number is used to determine the flow
regime of the fluid within the porous medium. Modification to the
fluid velocity term (u) and the characteristic linear dimension (d) are
required. When considering flow within the bed the appropriate
velocity is the interstitial, hence u = U, which can be related to the
superficial velocity by equation (3.1). The characteristic linear
dimension was deduced by Kozeny and is the volume open to the
fluid flow divided by the surface area over which it must flow (i.e.
product of volume of solids and specific surface area per unit
volume)
ALε ε
d= = (3.2)
AL(1 − ε ) S v (1 − ε ) S v
Thus, equations (3.1) and (3.2) used in the Reynolds number
expression give the Modified Reynolds number (Re1)
ε Uo ρ ρU o
Re 1 = = (3.3)
(1 − ε ) S v ε µ (1 − ε ) S v µ
Conceptually, the number still represents the ratio of inertial to
viscous forces in the fluid and provides a means to assess when the
inertial effects become significant. The conventionally applied
threshold to indicate significant turbulence is 2, whereas for the flow
Reynolds number (page 1) the conventional threshold is about 2000.
It is important to note that the density term in equation (3.3) is the
density of the fluid: the turbulences described are that of the fluid,
the particles do not move in a packed bed.
3.3 Darcy’s law and the Kozeny-Carman equation

Darcy’s law, and the Kozeny-Carman equation, are valid for laminar
flow (Re1<2); i.e. for viscous drag by the fluid on the surface of the
particles within the bed. The analogy with electrical flow is shown in
Figure 3.2. A cell, or pump, provides the driving potential and the
flow, in either system, depends on the resistances in the circuit. For
two equal resistances, the potential, or pressure, is equally divided
between the two. In fluid flow, high resistance is provided by high
fluid viscosity (treacle is more difficult to pump than air) and by low
permeability (k) of the bed. A value of zero permeability would give
rise to infinite resistance – both for electrical and fluid flow. Darcy’s
law1 is
∆P µ dV 1
= (3.4)
L k dt A
where V is the volume of fluid flowing in time t. For a given bed
length the pressure drop will rise linearly with volume flow rate, or
fluid velocity, noting that
Q dV 1
Uo = = (3.5)
A dt A
as illustrated in Figure 3.3. The permeability of a packed bed can be
measured in this way, using the gradient from such a plot. The Fig. 3.2 Analogy between
fluid and electrical flows
permeability is often assumed to be a constant in a packed bed,
The flow rate (current or
provided the particle packing is also uniform within the bed. It
fluid) is proportional to
should, therefore, be an intrinsic property of a material. However, in the driving potential:
order to use equation (3.4) for design, e.g. to specify a pump required voltage in Ohms law and
to pass liquid through a bed at a desired flow rate, we need a method pressure gradient in a
for predicting the permeability of the bed. This is provided by the fluid. The constant of
Kozeny-Carman equation. proportionality is
The Kozeny-Carman equation was derived from the Hagen- resistance (R) – which for
Poiseuille equation for laminar flow of a fluid in a circular channel a fluid is viscosity
divided by bed
∆P 32µ permeability (k).
= 2 u (3.6)
L d
where d is the channel diameter. The derivation assumed that flow in
a porous medium can be represented as flow through many parallel
channels and equations (3.1) and (3.2) were used to represent the
equivalent channel diameter and to convert between the fluid
velocity within the channel and the superficial. Hence, substituting
these equations in (3.6) and collecting the constants together in a
single term called The Kozeny constant (K), which includes a factor
relating the tortuous flow channel length to the measured bed depth
gives the Kozeny-Carman equation
∆P  K (1 − ε ) 2 S v 2   K (1 − ε ) 2 S v 2  dV 1
= µ  U = µ   (3.7) Fig. 3.3 Graphical
o
L  ε3   ε3  dt A representation of Darcy’s
law for a bed of fixed
1
overall length
Darcy observed the law during the 1850’s by monitoring pressure drop over
sand filters at Dijon, France.
Comparison of equations (3.4) and (3.7), results in the conclusion that

the Kozeny-Carman equation is simply a subset of Darcy’s law, with
Permeability
Comparing equations an analytical expression for permeability. There are many alternative
(3.4) and (3.7) the expressions for permeability, but the Kozeny approach is the most
permeability is frequently encountered. In many instances the Kozeny constant has a
ε3 value close to 5, but this is not universally true. Inspection of
k= equation (3.7) shows that the permeability, or inverse resistance to
K (1 − ε ) 2 S v 2 fluid flow, is dependent upon the bed porosity (or solids
hence the SI unit is m2. concentration) and the specific surface area per unit volume of the
The Kozeny constant is particles within the bed. This is logical because the higher the bed
often 5, but there is surface area the greater the viscous drag of the fluid on the particles.
much experimental
Equation (2.14) can be substituted into (3.7) to provide a version of
evidence to suggest that
the equation in terms of particle size
K = f (ε )
i.e. the Kozeny ∆P  36 K (1 − ε ) 2 
coefficient is a function
= µ 3 2
U o (3.8)
L  ε xSv 
of porosity.
where xSv is the Sauter mean diameter for the particle distribution. On
page 6 and 7, the question ‘which particle diameter to use to
represent the distribution?’ was asked and Figure 2.2 provided an
example distribution. The x50 and xSv for this distribution are 9.2 and
6.4 µm, respectively. The consequence of using an inappropriate
equivalent spherical diameter can be illustrated by using both of
these diameters in equation (3.8) and comparing the results. Under
identical flow conditions, the ratio of pressure drops calculated by the
Sauter mean to that by the median particle size is 2:1; i.e. there is
100% difference between the pressure drops calculated by these
diameters; yet both are equivalent spherical diameters for the same
particle size distribution. Clearly, the Sauter mean is the most
appropriate diameter to use, as indicated in equation (3.8), but it is a
calculated value and the median is quick to read off the cumulative
distribution curve. Hence, there is great temptation to use the
median, but it would predict only 50% of the likely pressure drop.
3.4 Friction factor

When turbulences within the fluid flowing through the
porous medium become significant, i.e. Modified
Reynolds numbers greater than 2, additional drag terms
to the viscous ones quantified in the last section become
important. In fluid flow through pipes and channels a
friction factor was deduced to represent this region and
Carman extended the analogy with pipe flow to cover
both flow regions in porous media. The porous media
friction factor is illustrated in Figure 3.4 and the method
used to relate the shear stress at the surface of the
solids, to the pressure drop, follows (the same
Fig. 3.4 The friction factor plot for fluid approach for flow in pipes is included in the box
flow through porous media overleaf for comparison).
R
(3.9)
ρU 2
Equation (3.9) is the friction factor and R is the shear stress, or drag
force per unit area, on the particle surface. A force balance at the Force balance on a pipe
particle surface can be constructed as follows. wall
surface area of particles= S (1 − ε ) LA (m2) (for pure fluid – no
v particles)
drag force = R. particle surface area (N)
and
pressure drop on fluid= ∆P (N m−2)
force by fluid= ∆PAε (N)
Equating the two forces and rearranging gives
force on wall:
ε ∆P
R= (3.10) Rπd .δx
S v (1 − ε ) L force on fluid:
Note that equation (3.10) is the analogue of that provided for pipe πd 2
flow in the box. Finally, expanding into a friction factor, equation δP
4
(3.9), together with equation (3.1) gives combine and integrate:
R ε3 ∆P 1 ∆Pd
= (3.11) RL = or
ρU 2 S v (1 − ε ) L ρU 2 4
o d ∆P
Alternatively, the pressure drop per unit length is R=
4 L
2
∆P  R  S v (1 − ε ) ρU o c.f. equation (3.10)
= (3.12)
L  ρU 2  ε 3
 
where the bracketed term is the friction factor. So, given a flow rate, Laminar flow
hence superficial velocity, it is possible to calculate the Modified Carman correlation
without the turbulent
Reynolds number from equation (3.3) and the friction factor from
correction is
Figure 3.4. This can then be used in equation (3.12) to provide the
R 5
pressure drop, or gradient, under conditions of laminar or turbulent =
flow through the porous medium. ρU 2 Re1
Using equation (3.10) and
3.5 Carman and Ergun correlations (3.3) provides
The friction factor plot, with Reynolds number, for fluid flow through ε3 ∆P 1
= ...
porous media is a smoother function than that found in pipe flow. S v (1 − ε ) L ρU 2
This is due to the smooth increase in turbulences within the bed as o
flow rate increases. Thus the friction factor plot can be represented by K (1 − ε ) S v µ
... =
just one, or two, empirical curves. The Carman correlation is ρU o
generally used for solid objects forming a bed So,
R
=
5
+
0 .4
(3.13) ∆P  K (1 − ε ) 2 S v 2 
2 Re 0 .1 = µ U o
ρU 1 Re1 L  ε3 
The Ergun correlation is for hollow objects, such as packing rings
R 4.17 i.e. equation (3.7)
= + 0.29 (3.14)
ρU 2 Re1
The form of both equations is similar: with a correction term added to
the laminar flow term to account for resistance due to turbulences.
This is illustrated in the box, which shows that the Carman

correlation reduces to the Kozeny-Carman equation, with K=5, when
the turbulent correction term is dropped.
In practice, when performing flow calculations with Modified
Reynolds numbers greater than 2, equations (3.13) or (3.14) are used
to determine the friction factor – rather than Figure 3.4, and equation
(3.12) is used to calculate the pressure drop.
3.6 Concentrations by mass and volume

Packing arrangements Solid concentration by volume fraction (C) was illustrated under
Uniform spheres, packed
Figure 3.1 and is simply the volume of solids present divided by the
together in a regular
pattern, vary from a co- total bed volume. It is numerically equal to unity minus bed porosity.
ordination number of 6, In most of the following chapters, it is more convenient to work in
for simple cubic packing, terms of solid concentration rather than porosity. However,
to 12 for hexagonal close- laboratory analyses often provide solid concentration by mass. For
packed; which is the example, taking a sample of a filter cake containing water: weighing,
closest possible packing drying and then weighing the dried cake will provide the
for uniform spheres. The concentration by mass – if the last mass is divided by the first. Hence,
solids concentration for
conversion between the two different types of concentrations is
these arrangements varies
frequently required.
from 0.524 to 0.740,
respectively (porosities of Consideration of what the solid concentration by volume fraction
0.476 to 0.260). Randomly means leads to the following expression
packed spheres have a volume solids
solid concentration of 0.50 C=
volume solids + volume fluid
to 0.60. In theory, with a
size distributed solids the If a sample has resulted in a concentration by mass (Cw), and the total
finer particles could fit sample mass is M, then the volumes present can be deduced if the
inside the gaps between densities of the solid and fluid are known by
the larger particles Cw M / ρ s
providing even higher C= (3.15)
C w M / ρ s + (1 − C w ) M / ρ
solid packing. In practice,
solid concentrations much Dividing through by the volume of solids gives
lower than the regular 1
C= (3.16)
packed arrays are found. (1 − C w ) ρ s
1+
It is always safest to Cw ρ
measure concentration, or
A similar argument can be applied to convert from concentration by
porosity, as described on
page 22. volume fraction to mass fraction.
3.7 Summary
In this chapter we have seen the importance of specific surface: it
defines the surface area that is present within a porous medium; and
it is the friction of fluid flowing over that area that causes a pressure
drop. Finer particles provide a higher surface area per unit volume
than coarser ones, therefore, a higher flow resistance. Turbulences
within the fluid inside the bed, at higher flow rates, cause an
additional flow resistance, or pressure drop. Thus, a calculated
pressure drop by a laminar flow equation, Darcy’s law or Kozeny-
Carman, will always underestimate the true pressure drop if
significant turbulence is present. Often the temptation to use Kozeny-
Carman, rather than the procedure described in Sections 3.4 and 3.5,
is too great even when the Modified Reynolds number is high. This
will lead to an under-design, or specification, for equipment such as
pumps and fans. Providing another example of the failure in design
discussed in the Forward.
Even under conditions of laminar flow, the use of the median size
to respresent the size distributed solids rather than the Sauter mean
diameter can cause significant errors, as shown in Section 3.3. Finally,
a brief consideration of particle packing arrangements has been
included, but in most processes the packing arrangement is random
and not structured. Hence, the safest procedure for the analysis of a
flow through porous media problem is to conduct experiments to
deduce characteristics such as the permeability and if it varies with
Equation summary
flow; a possibility if the finer particles become transported within the Under laminar flow:
bed or if suspended solids within the fluid deposit inside the bed.
∆P  K (1 − ε ) 2 S v 2 
The latter is depth filtration, which is covered in the next chapter. = µ U o
L  ε3 
3.8 Problems
where K is the Kozeny
1. constant At high values of
(i) A powder is contained in a vessel to form a cylindrical plug 0.8 the 'Modified Reynolds
cm in diameter and 3 cm long. The powder density is 2.5 g cm−3 and Number' (Re1>2):
2.20 grams of powder was used to form the plug. The porosity inside Uoρ
Re1 =
the plug of powder is (-): µ (1 − ε ) S v
a: 0.58 b: 0.42 c: 0.75 d: 0.25 turbulent conditions
pertain. A pressure drop
(ii). Air was drawn through the plug at a rate of 6.6 cm3 per minute. given a flow rate can still
A mercury manometer was used to measure the pressure drop be deduced, but first the
during this process: a pressure drop of 60 mm Hg was recorded. The Modified Reynolds number
specific gravity of mercury is 13.6, thus the pressure drop across the is required, then the
friction factor using say the
plug was (Pa):
Carman correlation:
a: 80 b: 800 c: 8000 d: 80000
R 5 0 .4
(iii). The superficial gas velocity in (ii) was (m s−1): = +
a: 0.0022 b: 3.65x10−5 c: 2.2x10−6 d: 0.00365 ρU 2 Re1 Re 0.1
1
A force balance on the
−5
(iv).The viscosity of the air was 1.8x10 Pa s, using the Kozeny- surface of the solids and on
the fluid gives:
Carman equation, the specific surface area per unit volume of the
RS v LA(1 − ε ) = ∆PAε
powder was (m-1):
where L is the bed height
a: 2310 b: 3.0x1011 c: 5.5x105 d: 1.2x106
or depth. Given a value for
the shear stress on the
(v). The Sauter mean diameter of the powder was (µm): solids (R) calculated from
a: 2600 b: 22 c: 11 d: 5.0 the Carman correlation
then the pressure drop (∆P)
(vi). The air density was 1.2 kg m−3, the Modified Reynolds Number can be calculated from the
of the system was (-): force balance.
a: 0.10 b: 8.4x10−10 c: 4.6x10−4 d: 0.00021
(vii). Comment on whether your use of the Kozeny-Carman equation

was valid or not:
2.
(i). A cylindrical ion exchange bed composed of spherical particles 2
mm in diameter packed at a bed voidage of 0.45 is to be used to de-
ionise a liquid of density and viscosity 1100 kg m−3 and 0.0075 Pa s
respectively. The design flow rate is 5 m3 hour−1 and the bed height
and diameter are 2 and 0.2 m respectively, using the Kozeny-Carman
equation the pressure drop is (Pa):
a: 99000 b: 1.32x107 c: 4400 d: 44000
(ii). The Modified Reynolds Number is:

a: 3.91 b: 4.78 c: 1290 d: 0.478
(iii). Comment on your use of the Kozeny-Carman equation:
(iv). The interstitial liquid velocity inside the bed is (m s−1):

a: 0.020 b: 0.098 c: 0.08 d: 0.044
(v). Using the Carman correlation the shear stress on the ion
exchange beads is (Pa):
a: 17.2 b: 3.5 c: 13.5 d: 27.0
(vi). Hence the dynamic pressure drop over the bed is (kPa):
a: 84 b: 130 c: 99 d: 150
(vii). Why is the answer to (vi) different to that in (i)?
(viii). If the liquid has a datum height equal to the position at the base
of the ion exchange vessel and, therefore, needs raising to the top of
the column before it enters the ion exchange bed the additional
pressure drop to effect this, i.e. the static pressure drop over the bed,
is (kPa):
a: 2.16 b: 21.6 c: 216 d: 0.22
4 Filtration of liquids
Filtration is the removal of suspended particles from a fluid,
performed by a filter medium, septum, cloth or bed of solids. In this
chapter we will discuss only liquid filtration, the removal of particles
from gases is covered in Chapter 14. Filtration is commonly
encountered in chemistry laboratories on a Buchner funnel and
within the kitchen during the making of filter coffee. It is a very
important industrial process as it is often a key stage in product
recovery: following reaction, precipitation and crystallisation stages,
but preceding thermal drying and packaging (e.g. in pharmaceutical
production). It is more economic to remove moisture from particles
by mechanical means, including filtration, than by thermal means. Fig. 4.1 In a filter increasing
resistance causes liquid
Thus, domestic washing machines provide higher and higher spin
flow to decrease with time
speeds prior to thermal, or evaporative, drying. – at constant pressure drop
There is a vast range of filtration types; depending upon whether
the objective is to produce a clean liquid, as in drinking water
production, or solids retained in a filter cake, as in product recovery.
The former process is called clarification, or clarifying filtration and is
often performed in equipment containing packed beds, which were
introduced in the previous chapter.
4.1 Deep bed and clarifying filtration

Uses for bed filters
Deep bed filtration uses packed beds of particles between 300 and Drinking (potable) water,
5000 µm in diameter and the bed height is usually between 0.5 to 3 pretreatment for high
metres. Deposition of suspended solids takes place within the bed, by purity water, e.g.
a variety of particle adhesion and collection mechanisms. Some are electronics industry,
discussed in Chapters 13 and 14. The most pertinent are: effluent treatment,
beer and wine
sedimentation (on to the bed grains), inertia, van der Waals,
clarification,
diffusion, electrostatic attraction and repulsion. Within the bed the
sea water filtration
fluid flow condition is predominantly laminar – to minimise the before injecting into oil
scouring effect of turbulences. However, the packed bed will reservoirs.
eventually contain a large amount of deposited solids and need
regenerating; at this stage a backflush is often employed, which may
involve inducing fluid turbulences. The frequency of regeneration
will depend upon the concentration of solids in suspension entering
the bed. This is usually low: less than 0.5 grams per litre. The feed
flow rate is usually less than 8 m3 per m2 of bed area per hour and it
is possible for the bed filter to reduce the outlet (effluent) solids
concentration to below 0.1 milligrams per litre. However, when
processing very small particles, such as bacteria and viruses, the feed
rate may need to be low: 0.1 m3 m−2 h−1, in order to provide the
effluent quality required; i.e. particle removal.
As particles are deposited within the packed bed, the resistance to
fluid flow increases. If the bed is running under a constant pressure Fig. 4.2 A deep bed filter
difference the resulting flow rate will diminish, as illustrated in
Figure 4.1. When the filtrate rate drops to an unacceptably low value,
History
the filter is backflushed at a rate of up to 36 m3 m−2 h−1 and may
In the UK, Dr John Snow
include an air scour and some degree of fluidisation (Chapter 7). So,
first traced a water borne
cholera epidemic to a for continuous production a second filter is required; so that whilst
public water pump in one is filtering the other can be cleaning and put on stand-by. In
Broad Street, London, in general, backflushing uses up to 5% of the previously filtered liquid.
1854. Surrounding areas In some applications the backflush cycle is timed and, for constant
not using the pump were rate filtration – where the pressure drop increases to maintain the
unaffected. The rate, the backflush is initiated by a pressure threshold. Figure 4.2
connection between water shows a schematic illustration of a bed filter with the associated flow
and disease led to mass
lines for operation.
deployment of sand filters
Various media may be used as the bed, including: sand, gravel,
(300 µm grains and 0.75 m
high), which were anthracite, activated carbon and garnet. The carbon based materials
previously just provides removal of some dissolved species as well as filtration of
considered to improve the suspended matter from the liquid. In many instances a mixed bed is
water look and taste. A used: coarse particles of one type on top of fine particles of another.
public house, called the One advantage of this arrangement is that, if the coarse particles treat
John Snow, now stands the down-flowing suspension first, they will remove the coarser
on the site of the old suspended solids – the finer material will pass between the coarse
Broad Street pump.
grains, but will become captured on the finer grains in the lower bed.
This segregation of filter media prevents the finer suspended material
from clogging up the gaps at the top of the filter bed. However,
cleaning by backflushing would normally result in the finer filter
grains transferring to the top of the bed, but this can be overcome by
designing the bed to contain dense fine grains (of say sand) and
coarse light grains (of say anthracite). An illustration of a mixed
media bed is given in Figure 4.3. Filtration performance is also
improved by the use of flocculants and coagulants (Chapter 13), prior
to passage through the bed, to increase the size of the suspended
particles to be captured within the media.
The basic design, or modelling, of deep bed filters can be achieved
using a simple equation
dN
= −λN (4.1)
dz
where λ is a filtration constant (m−1). Unfortunately, the filtration
constant will change with time, as material becomes deposited within
the bed; hence, equation (4.1) can only be treated as an ordinary
differential equation at the start of the filtration and integrated to give
N = N exp(−λ L) (4.2)
o o
where λo is the initial filtration constant, L is the depth within the
filter and N is the concentration of solids in the influent (on any
consistent basis: mass per volume, volumetric, etc.). Thus, the
concentration of solids deposited from the influent suspension
Fig. 4.3 Illustration of exponentially decays within the bed. It is possible to model the
mixed media deep bed performance of a bed filter by allowing the filtration constant to
with cut-away showing become a variable dependent upon the following
particle deposition (4.3)
λ = f (ε ε σλ )
s o o
where εs is the surface deposit porosity (of the material deposited on

the grains), εo is the packed bed porosity of the filter grains and σ is
the specific deposit: the ratio of the volume of solids deposited within
a layer to the volume of the filter layer.
Apart from the filtration performance, the pressure drop during
filtration is important. This can be deduced from the Kozeny-Carman
equation, but it is usual to combine it with the hydrostatic head
equation derived from Bernoulli’s equation, thus
 K (1 − ε ) 2 S 2 
dh v U 1 (4.4)
= µ
dz  ε 3  o ρg
 
where h is pressure in terms of head of liquid. Strictly speaking, the
porosity is the porosity remaining open to flow; which will be that Fig. 4.4 Candle filtration: the
formed from the initial bed voids and what is left after deposition of high surface area provided by
solids. However, the latter effect is normally limited so the initial bed the filters in parallel helps
porosity is used. However, the pressure drop, or head, will increase provide good filtrate rates.
during filtration as the flow channels become smaller, or clogged. Filter cleaning is required on a
Hence, equation (4.4) is only safely applied towards the beginning of periodic basis. Filtration is
a bed filtration. The equation can be integrated over the full bed usually performed under
constant rate conditions.
height to give the overall head requirement.
A more recent development from the batch bed filter design is the
continuous sand filter. The sand bed is continually moving
downwards within the vessel because the bed at the bottom enters an
pictures
air lift pump during which deposited solids are scoured off the sand For illustrations of
grains. At the top of the air lift the sand quickly sediments to rejoin equipment described here
the top of the bed, but the deposits on the sand grains are elutriated see:
(i.e. washed) out through a separate discharge pipe. Thus, there is no www.solidliquid-
need for backflushing and the consequent taking the filter off-line. separation.com
Deep bed filtration is just one example of clarifying filtration,
where the objective is to remove suspended solids at a low
concentration from a liquid. Other filter types that can be used for
this include: membranes, pressure leaf filters, candle filters and pre-
coat filtration, see Section 4.8. Excepting pre-coat filtration, the
intention is to provide a filter medium capable of retaining the
particles in suspension without clogging, or providing high flow
resistance. A set of candle filters are illustrated in Figure 4.4 and,
typically, filtration would take place at up to 6 bar gauge pressure.
4.2 Cake filtration

When the solid concentration in the suspension is sufficiently high, Fig. 4.5 Bridging of
solids will bridge over the entrance of the filter medium pores. This is particles over the surface
cake filtration and it is illustrated in Figure 4.5. The minimum slurry openings of a filter medium
– thus new cake is formed
concentration required for cake filtration depends on the nature of
on top of old layers of cake
the solids and filter medium, but is usually about 0.5% by volume.
and it is possible for the
The modelling of cake filtration assumes that Darcy’s law can be medium openings to be
applied to both the filter medium and the forming filter cake. The much bigger than the
assumption is that the two resistances to fluid flow, or their resulting particles
pressure drops, can be added together similarly to Figure 3.2, thus
∆P = ∆Pc + ∆Pm (4.5)

where ∆P is the total applied pressure (i.e. over both the cloth and
cake), whereas ∆Pc is pressure drop over the cake and ∆Pm is pressure
drop over the filter medium (normally a cloth). Applying Darcy’s law
to the pressure drops, equation (3.4), gives
µ L dV µ Lm dV
∆P = + (4.6)
A k dt A k m dt
where the subscript ‘m’ denotes the medium and L represents the
cake height at any instance in time. The medium height and
Fig. 4.6 The filter cake permeability will be essentially constant during the filtration and can,
concentration step change therefore, be replaced by a constant medium resistance term (Rm).
from the feed concentration Likewise, the cake resistance (Rc) can be defined as follows but,
Cf to the cake concentration unlike the medium resistance, it will increase as the cake height
C increases
L L
Rm = m and Rc =
km k
Substituting the resistance terms into equation (4.6) and rearranging
dt µ
= (R c + R m ) (4.7)
dV A ∆P
Figure 4.6 illustrates the occurrence of a new layer of cake: the
concentration jumps from Cf to that of the cake C; the liquid displaced
by the solids concentration increase must be sucked, or pushed,
through the layers of previously formed cake. Hence, it is appropriate
Fig. 4.7 The variation of to apply Darcy’s law for pressure the drops due to: flow through the
cake resistance with mass layers of previously formed cake and through the filter medium. In
of cake deposited –
equation (4.7) the medium resistance term is a constant, but the cake
definition of specific cake
resistance
resistance term increases with deposit thickness.
4.3 Specific cake resistance and dry cake mass per

filtrate volume
Equation (4.7) has too many variables to be solved: time, volume, Rc
and, under certain circumstances, pressure drop. The cake resistance
must be related to one of the other variables. This is possible because
it is the increasing cake depth that causes Rc to vary and, for
incompressible cake filtration, there will be a constant volume of cake
deposited per volume of filtrate. Hence, L can be related to V,
enabling the substitution to be made. It is formally achieved as
follows. Figure 4.7 illustrates that the cake resistance increases
proportionally with increasing cake mass, or mass per unit filter area.
Fig. 4.8 The variation of cake The constant of proportionality is called the specific resistance to
mass with filtrate volume – filtration (α) and it is similar to an equivalent length used for fittings
definition of dry cake mass in pipe flow calculations; e.g. an extra length of pipe in resistance
per unit filtrate volume terms to account for the extra pressure drop due to the fittings. Thus,
the equation is
Rc = αW / A (4.8)
where W is the total mass of cake deposited, which can be related to

the volume of cake (LA) by the volume fraction of solids in the cake
(C) and the solids density. See Figure 4.6 for an illustration of cake
solids volume fraction and cake height. Thus
W = LACρ s (4.9)
and, for an incompressible cake filtration, the cake mass will
uniformly increase with the filtrate volume. This is illustrated in
Figure 4.8. The constant of proportionality is called the dry cake mass
per filtrate volume and has the symbol c. Thus
LACρ s
c=
V
which can be rearranged to give
LACρ s = cV hence from equation (4.9) W = cV
which is substituted into equation (4.8) to give Fig. 4.9 Illustration of
Rc = αcV / A (4.10) compressible cake
filtration, increasing
For the filtration of a given material, at a constant feed concentration,
concentration of the cake
both specific resistance and dry cake mass per filtrate volume will be with applied pressure
constants (they are the constants of proportionality in Figures 4.7 and difference – compare with
4.8); hence, cake resistance varies proportionally with filtrate volume. Figure 4.6, where C is
Substituting equation (4.10) into (4.7) provides independent of cake
dt µαc µRm forming pressure.
= 2 V+ (4.11)
dV A ∆P A∆P
which is the general filtration equation and is valid for all types of
incompressible cake filtrations. It has three variables: time, volume
and pressure drop. All the remaining terms are constant for the
filtration of a specific material at constant concentration of the feed
slurry. The specific resistance is a property of the particles in
suspension, but not the feed concentration; whereas the dry cake
mass per filtrate volume is both a property of the particles in
suspension and the feed concentration employed. Equation (4.11) is
solved under specific conditions: constant pressure, constant rate or
both varying – as described in Section 4.5
4.4 Compressible cake filtration

Figure 4.9 illustrates what is meant by compressible cake filtration:
the cake concentration, and hence the permeability and specific Recommendation
resistance, are a function of the pressure drop applied during the Leave the compressible
cake filtration section
filtration. It is logical that an increase in concentration increases the
until the problems at
resistance to fluid flow within the cake and it is possible to show that
the end of this chapter
the specific resistance and cake permeability are simply related. have been completed.
Starting with the original definition of cake resistance and
substituting equation (4.8)
L αW
Rc = =
k A
Using equation (4.9) and rearranging for specific resistance
AL 1
α= = (4.12)
kLACρ s kCρ s
Filtration modelling and the permeability can be provided by the expression given on
Permeability can be page 24, which demonstrates that it is a function of porosity (or cake
deduced from a particle concentration by volume fraction), as well as specific surface of the
size distribution –
particles. Likewise, the dry cake mass per filtrate volume will also
hence specific
resistance from (4.12), depend on the applied pressure, again this is because the cake
so long as the cake concentration increases with applied pressure. The relation between
concentration can be dry cake mass per filtrate volume is derived from a mass balance on
estimated. Likewise, solids and liquid entering the filter as follows.
equation (4.14) Total mass slurry filtered =M (kg)
provides a value for c. Mass fraction of solids in slurry =s (-)
Both can then be used Mass of solids in slurry = sM (kg)
in (4.11) for filtration Total mass of liquid in slurry = (1-s)M (kg)
modelling. However,
Mass liquid retained in filter cake = (1-C)ALρ (kg)
laboratory filtrations
used to scale-up So, provided that all the solids are retained in the filter cake, the dry
filtration data are much cake mass per filtrate volume (N.B. volume liquid in the filtrate and
more reliable. not total liquid volume) is
sM
c= (4.13)
[(1 − s) M − (1 − C ) ALρ ] / ρ
The solid mass can be represented in terms of both the solids retained
in the cake and the original solids in the slurry
CALρ s = sM (kg)
Filtration constants
The empirical constants which can be rearranged to give
tell us about the nature sM
AL =
of the material to be Cρ s
filtered:
and substituted in to equation (4.13)
vacuum filtration
sMρ 1
region c= = (4.14)
n < 0.3 and α < 1x1011 [(1 − s) M − (1 − C ) sMρ ] / Cρ s (1 − s) /( sρ ) − (1 − C ) /(Cρ s )
m < 0.1 So, with compressible cake filtration, both the specific resistance and
dry cake mass per filtrate volume are functions of the filtering
pressure filtration pressure because it influences the cake concentration. There are
region
several expressions relating the average specific resistance and
0.3 < n < 0.6
average cake concentration (by volume fraction) to pressure
1x1011< α < 2x1012
difference; the easiest to apply are
0.05 < m < 0.3
α = α o (1 − n)∆Pc n (4.15)
See: and
www.filtration-and-
separation.com for a list C = C o (1 − m )∆Pc m (4.16)
of measured values for where n, m, αo and Co are empirical constants. The coefficient ‘n’ is
common materials. sometimes referred to as the compressibility coefficient and the
greater the value the greater the compressibility. A value of zero
indicates incompressible solids, calcite is about 0.2, kaolin clay about
0.6 and fine sludges approach unity. Unfortunately, equation (4.15) is
unusable when the compressibility coefficient approaches unity.
The average dry cake mass per filtrate volume can be deduced
from the average cake concentration by equation (4.14).
1
c= (4.17)
(1 − s ) /( sρ ) − (1 − C ) /(C ρ s )
In equations (4.15) and (4.16) the pressure forming the filter cake
should be used; this will be less than the total applied pressure drop
because some of the applied pressure will be used to force the fluid
through the filter medium. However, the use of a higher pressure
difference in equation (4.15), than really exists, will result in a higher
specific resistance to filtration which is a safer design option than
under-estimating filtration resistance. The actual difference between
total applied pressure and that over the cake is normally very small.
4.5 Filtration modes of operation

Equation (4.11) contains three variables: time, volume and pressure Fig. 4.10 Data analysis by
difference. It can be solved by holding one of these variables constant. equation (4.19)
Considering constant pressure filtration, the equation can now be
integrated from time zero to some time t, during which the volume
passing through the cake (and medium) changes from zero to V –
which is the total filtrate volume at time t. Integrating and
rearranging provides
 µαc  2 µR m
 2 V + V −t = 0 (4.18)
 2 A ∆P  A∆P
which is a quadratic equation in terms of filtrate volume. Thus, if all
the filtration constants are known and the filter cake formation time
(t), it is possible to solve equation (4.18) for the filtrate volume. In
laboratory tests it is usual to operate at constant pressure, employing
a constant vacuum filter such as a Buchner type, and equation (4.18)
can be rearranged to assist data analysis to give
t µαc µRm
= V+ (4.19)
V 2 A 2 ∆P A ∆P
which is the equation of a straight line, as illustrated in Figure 4.10.
The gradient can be used to deduce the specific resistance and the
filter medium resistance contributes towards the intercept; the other
constants include: fluid viscosity, filter area and total pressure
difference. However, the dry cake mass per filtrate volume must be
independently determined, as this is also a contributor towards the
gradient. This is normally achieved by taking a cake sample and
weighing it, followed by drying then reweighing when dry. This
provides the moisture ratio
mass of wet cake
mR = (4.20)
mass of dry cake
Extending the material balance just before equation (4.13):
Total mass of liquid in slurry = (1-s)M (kg) Fig. 4.11 Schematic
Mass of dry cake = sM (kg) diagram of a Rotary
Mass of wet cake = sMmR (kg) Vacuum Filter (RVF)
Mass of liquid retained in the cake = sMmR- sM (kg)
= (mR-1)sM (kg)
So, provided that all the solids are retained in the filter cake, the dry
cake mass per filtrate volume (i.e. volume liquid in the filtrate and
not total liquid volume) is
sM sMρ
c= =
[(1 − s ) M − (m R − 1) sM ] / ρ (1 − s) M − (m R − 1) sM
which simplifies to
sρ
c= (4.21)
1 − mR s
Hence, a knowledge of the moisture ratio, slurry solids concentration
by mass fraction and liquid density provides the dry cake mass per
filtrate volume and the gradient from a plot similar to Figure 4.10 will
provide the experimentally determined value for specific resistance
to filtration. If a series of tests are conducted, at different total
filtration applied pressures, then it is possible to assess if the
resistance varies with pressure: a log plot of these is suggested by
equation (4.15), with the gradient providing the exponent in the
equation and the intercept can be used to deduce αo. A similar plot is
Fig. 4.12 Schematic used for concentration variation with pressure, in accordance with
diagram of a plate and equation (4.16).
frame filter When all the constants in equation (4.18) are know, including the
specific resistance and dry cake mass per filtrate volume, possibly
coming from laboratory test data, then filtration simulation can be
performed:
filtrate volume in time t =V (m3)
dry cake mass in time t = cV (kg)
wet cake mass in time t = mRcV (kg)
mass slurry filtered in time t = mRcV+ρV (kg)
The height of the filter cake can be obtained from a volume balance
conducted on the filter cake:
volume solids in cake + volume liquid in cake = cake volume
cV ( m R − 1)cV
+ = LA
ρs ρ
cV  1 ( m R − 1) 
L=  
A ρ + ρ  (4.22)
 s 
Industrial filtration equipment operating under constant pressure
includes most vacuum filters, such as rotary vacuum (or drum) filters
and rotary disc filters. One of these is illustrated in Figure 4.11.
Equation (4.11) can be solved under the condition of constant rate
filtration. Under these conditions, the filtration pressure must increase
Fig. 4.13 Schematic in order to overcome the ever-increasing resistance to filtration due to
diagram of a simple cake growth. Equation (4.11) can be rearranged to give
pressure Nutsche filter µRm dV
 µαc dV 
∆P =  2 V + (4.23)
 A d t  A∆P dt
which is the equation of a straight line on a plot of filtration pressure

against filtrate volume. This is a result of the filtrate rate (dV/dt) being
a constant – by definition. Hence, the gradient is provided by the
terms within the brackets in equation (4.23) and the intercept is the
furthest term on the right hand side of the equation.
Industrially, constant rate filtration occurs when a filter is fed by a
positive displacement pump, that provides the same displacement of
fluid with time; such as a piston pump, or progressing cavity pump.
Thus a filter press fed by one of these pumps will operate under For more details on
constant rate, dictated by the pump stroke or performance. However, filtration equipment see
it is usual for the mode of operation to switch from constant rate to www.midlandit.co.uk/p
constant pressure, when a set operating pressure has been achieved. articletechnology.
This is a safety requirement to prevent over-pressurisation of seals,
etc. An example of a filter press is provided in Figure 4.12 and a
pressure Nutsche filter in Figure 4.13.
Feeding a filter press with a centrifugal pump will result in
variable rate and variable pressure filtration. The pump will provide high
flow at low back-pressure and this will fall to low flow at high
Variable pressure and
pressure. Equation (4.11) has to be graphically, or numerically, solved rate filtration
rather than the analytical solutions represented by equations (4.18) The numerical solution
and (4.23), see the box on the right. Firstly, using Q for the volume is obtained by taking a
flow rate of slurry, and neglecting the volume of solids and liquid pressure drop from the
retained in the filter cake compared to the slurry volume, pump characteristic and
rearrangement of equation (4.11) provides the corresponding
∆P µcα µRm volume flow rate. These
= 2 V+ (4.24)
Q A A are used in equation
The pump characteristic provides a relation between the volume flow (4.24) to calculate the
rate and the pressure drop and it is possible to rearrange and filtrate volume, which is
integrate the flow rate expression as follows then plotted against the
V
inverse volume flow rate
dV
t=∫ (4.25) as suggested by
0 Q equation (4.25). Another
In this chapter we have considered bed filtration of suspensions with pressure difference is
concentrations below 0.5 grams per litre and cake filtration at then selected and the
concentrations above 0.5% by volume. The question arises over what procedure repeated until
happens when concentrations less than 0.5% are used with a suitable plot of inverse
conventional filtration media? The solids can block the surface pore flow rate against filtrate
openings, become lodged within the pore – blocking the flow, or can volume is produced. The
deposit within the pore thereby restricting the flow, but not blocking area under the curve, up
it. A general set of equations have been deduced to describe all of this to a selected filtrate
behaviour volume, provides the
n' time taken to filter the
d 2t  dt 
= K'  (4.26) required volume of
dV 2  dV 
filtrate, in accordance
where K’ and n’ are constants. If a value of zero is used for n’, and the with equation (4.25).
equation is integrated, then a similar equation to (4.18) results. Hence,
this equation is a general one that includes cake filtration as one sub-
set. Other values for n’ are: 1, 1.5 and 2 and they represent
intermediate blocking, standard blocking and complete blocking

respectively. Standard blocking is used to describe the reduction in
pore flow channel width due to particle deposition, whereas
complete blocking is the closing off of the flow channel due to
plugging.
4.6 Membrane filtration

In the context of particle technology, the most appropriate membrane
filtration process is microfiltration (MF), but it is quite common to
filter very fine particles using an ultrafiltration (UF) membrane. Also,
Fig. 4.14 Polymer UF mathematical models are sometimes applied to MF applications.
microfiltration membrane So, both techniques will be briefly described here. A general
– note scale bar definition of a MF membrane is one that possesses filtering pores
between 10 and 0.1 µm. A UF membrane has pores below 0.1 µm, but
is usually rated in terms of molecular weight cut-off, and is employed
to retain macro-molecular material. Membranes usually possess a
thin filtering surface, which may be supported by a much thicker
structure. An example photograph, taken under a scanning electron
microscope, is provided in Figure 4.14.
On the figure a variable pore structure is evident and the
manufacturer’s pore rating of the membrane is 0.2 µm, but surface
openings up to 3 µm are evident. The filter is effective for the removal
of all bacteria (sizes down to 0.2 µm), but it is obvious that these small
particles will not be retained on the filter surface – at least some will
penetrate the membrane matrix. The amount of penetration will
Fig. 4.15 Illustration of depend upon the concentration of solids as, at high concentration, a
crossflow filtration – shear cake may form. Thus, if all the fine particles are retained at all
at the membrane surface concentrations deposition must take place inside the filter and the
helps reduce particle microfilter is relying on depth filtration mechanisms similar to those
deposition. described in Section 4.1. The consequence of internal deposition of
particles within the membrane is that the filtration flux rate (or
permeate rate) will decrease, if a constant pressure differential is being
used. The permeate rate will also decrease as solids become
deposited on the surface of the membrane, for similar reasons as
described during cake filtration. In order to limit the occurrence of a
cake crossflow filtration is often employed. This is illustrated in Figure
4.15.
The process selection of either crossflow operation or filtration in
dead-end mode using, for example, a cartridge filter depends upon
the process conditions. Cartridge membrane filters are used
extensively in the electronics and medical industries to ensure
sterility and particle free fluids (gases as well as liquids). The
industry is worth well over $1 billion per year. In most of these
applications the concentration of suspended material is very small
Fig. 4.16 Illustration of and the filter operates in dead-end and it is discarded when the
crossflow filtration pressure drop required to maintain an acceptable flow rate becomes
process operation too great. However, these are high value industries that can afford to
discard these products. A duplex system is often used: a second
cartridge filter is used whilst the first is taken off-line for filter
element changing. However, at higher concentrations of suspended
solids the frequency of element changing would be too great and
crossflow filtration may be appropriate. In this instance a clean liquid
stream is still supplied, but a second liquid stream containing a
higher concentration of suspended material is recycled. This stream is
the retentate and the process is illustrated in Figure 4.16.
Despite the application of crossflow filtration the permeate flux
rate usually decreases due to the deposition of some solids on the
surface of the filter, as well as the deposition within filters when
using the design illustrated in Figure 4.14. A typical flux rate curve is
illustrated in Figure 4.17. The conventional units of membrane flux Fig. 4.17 Permeate flux rate
are litres of permeate per square metre of filter per hour (l m−2 h−1). It behaviour with backflushing
to restore the rate
may be possible to momentarily backflush some permeate through the
membrane to dislodge the solids retained on the surface of the
membrane, which will then be swept away in the crossflow, thus
restoring some of the original flux. However, it is much more difficult
to remove the solids deposited within the membrane filter matrix and
this may lead to an ever-declining flux rate, even with backflushing
applied.
During crossflow filtration two stages of modelling may be
considered: the period in which the flux declines, and a model to
predict the equilibrium value of the filtration flux. Flux decline
models are often based on pore blocking, or diminishing size,
concepts discussed in Section 4.5 and numerically the data is fitted to
versions of equation (4.26). From a practical perspective, models of
the equilibrium flux rate have greater application because these
enable the calculation of flow and process conditions required to
perform an industrial separation over an extended period in time.
Two general models are often used: cake filtration and film theory.
These are illustrated in Figure 4.18. Film theory is normally applied
to very small particles, less than 5 µm, and is founded in classical
mass transfer analysis. Cake filtration is usually applied to larger
particles and some of the resistance terms are described by the
equations discussed in Section 4.5, with additional forces due to
colloidal interactions as described in Chapter 13.
In MF the concentration of suspended material passing through
the membrane is usually negligible, cp=0, hence the retention (Rt) is
100%
 cp 
R t = 1 − 100% (4.27)
 cb 
In the cake filtration model the permeate flux rate (J) is described by Fig. 4.18 Cake and film
a form of the cake filtration equation (4.7) models of membrane
filtration
∆PTMP
J = (4.28)
µ ( R m + R c − Rs )
where the flux is
dV 1
J= (4.29)
dt A
and the filtration is run under the conditions of constant pressure.
The filtration resistance terms are: membrane, cake and shear (for
crossflow). The latter term is negative because the crossflow shear
causes a reduction in the resistance. In many cases it is possible to
relate the reduction in resistance due to shear with the flow
condition, such as surface shear stress, flow rate, degree of
turbulence, etc.
In the film model, the convective transport of solids towards the
membrane is balanced, at equilibrium, by the diffusional transport
monofilament away from the membrane. This gives rise to a smooth concentration
gradient approaching the membrane, rather than the step change
seen in cake filtration. This gradient in concentration of material is
called concentration polarisation and, in the film model, it is here that
the resistance to filtration is assumed to be greatest. Thus, a film
equation is appropriate
 D   cm − cp 
J =  e  ln   (4.30)
 δ   c b − c p


where cm is the concentration at the membrane surface, De is the
effective diffusivity of the filtering species and δ is the distance over
which the diffusion takes place. The last two terms can be replaced by
multifilament
a mass transfer coefficient, as is conventional in film modelling, and
that coefficient can be related to flow conditions using film transfer
correlations. This enables the prediction of permeate flux under
different conditions of shear, crossflow velocity, etc. The
concentration cm is assumed to be a constant, once sufficient material
has arrived at the membrane to form it, so that an increase in filtering
pressure will not give rise to an increase in cm. Hence, once the value
of cm is achieved at the wall, the flux rate determined by equation
(4.30) will not be influenced by further increases in pressure because
the mass transfer resistance is within the diffusional region
represented by the concentration polarisation. Thus, permeate flux
needle felt
will be transmembrane pressure independent, after the threshold of
reaching cm at the membrane. Increasing the pressure difference will
Fig. 4.19 Examples of increase the thickness of the cm deposit, but will not influence the
filter cloths
concentration gradient at the membrane wall, equation (4.30).
4.7 Filter media

In the last section we met particle deposition within the filtration
medium, resulting in long-term flux decline; i.e. increasing filtration
resistance. In the cake filtration section we discussed the need for
bridging over the pores within the filter medium, see Figure 4.5. The
filtration medium used for conventional cake filtration has a much
more open structure than a membrane and is more mechanically
robust. Usually, woven cloths are employed, or pressed needle felts;
these are illustrated in Figure 4.19. Pores within the cloth are much
bigger than the particles and this helps minimise the long-term
increase in medium resistance. However, the situation described by
Figure 4.5 is too simplistic, a more realistic representation is provided
in Figure 4.20: it is difficult to define where the cake finishes and the
filter medium starts and there is some penetration of the medium by
the particles. One important consequence of this is that it is not
possible for filter cloth manufacturers to define a value for Rm for
their cloth. The value must be determined in-situ, or during Fig. 4.20 More realistic
laboratory tests on the slurry, as it is a function of both the cloth and representation of
the material being filtered. Typically, the in-situ medium resistance is particles on a filter cloth
one or two orders of magnitude greater than the clean value. In many – c.f. Figure 4.5
cases this is also true during membrane as well as cake filtration.
One of the main distinctions between filtration media for cake
filtration is the use of monofilament or multifilament fibres for the
weaved cloth. Monofilaments give better cake discharge properties,
lower resistance due to internal solids deposition and lower flow
resistance in general; however, they are less effective at particle
retention and can be weaker cloths. Plugging of a monofilament is
also possible; whereas flow is possible between the fibres constituting
the multifilament cloth, so complete plugging is unlikely. When
filtering fine particles both types of cloths may not provide a good
clarity filtrate from the start and these first runnings can be recycled
back into the feed stream until the clarity of the filtrate improves.
Factors to be considered when selecting a filter medium, apart
from the ability to retain particles or form a cake, include: good cake
release, chemical compatibility with the liquid and solid to be filtered,
mechanical ability to withstand the filtration cycle (often including
stretching and rolling), cost, economic filtration time, adequate cloth
lifetime and resistance to particle blinding and biological growth. In
operation, it is common to include a cloth washing stage between
cake discharge and cloth reuse, to enhance cloth performance.
Laboratory tests to determine the cake resistance, as mentioned in
Section 4.5, are also used to evaluate different types of filter cloth and
their resistances, for possible use on a production filter. However, as
the interaction between the cloth and solids is significant, see Figure
4.20, it is important that the filtration test is performed following the
same orientation that the production filter will use. A rotary vacuum
filter has a filtering surface facing downwards; thus coarser and
easier to filter particles will settle away from the surface, so the test
should also use a downward facing filter leaf. An example filter leaf
is illustrated in Figure 4.21. Fig. 4.21 Laboratory test
When filtering finely divided material that may blind a cloth, or filter leaf – usually at
filter membrane, it is possible to deposit a sacrificial initial layer of constant pressure and the
solids of an additional material. The filtration takes place on the filter orientation should be
deposited solids and, when the filtration performance becomes similar to how the filter
unacceptable, the layer can be removed and replenished with a fresh will be used
one. This is called pre-coat filtration and secondary membrane
filtration in classic and membrane filtrations, respectively.
4.8 Filter aids

Pre-coat filtration uses a filter aid to protect the filter medium and the
filter aid might also be added to the suspension to be filtered in what
is called a body feed. Hence, it may be possible to filter very finely
divided solids using the filter aid to capture the particles by similar
mechanisms to bed filtration described earlier, but employing
conventional cake filtration equipment. This is only economic for the
removal of low concentrations of finely suspended solids because the
filter aid is rarely regenerated.
Two commonly encountered filter aids are diatomaceous earth
(kieselguhr) and perlite. The former is the skeletal remains of
diatoms, a type of algae, that died thousands of years ago and
compacted into sedimentary deposits that can be mined. The latter is
a glassy volcanic mineral that expands on heating and has usually
been crushed. Both types provide highly porous particle beds due to
their very irregular particle shapes. The dendritic, or spiky, nature of
the kieselguhr prevents two particles approaching closely and
provides an open flow path between the spikes. However, capture of
fine particles on the spikes is still possible.
Recently, the trend has been away from conventional filter aids
because it may be difficult to dispose of the silica based ones. When
the cake dries out there is a possibility of the finely divided silica
particles posing a health problem. This has increased disposal costs
and led to the adoption of alternatives such as fuel ash, sawdust and
other waste solid products as filter aids.
4.9 Summary
In this chapter, we have extended the principles covered in Chapter 3
Equipment selection to cover the effects of depositing solids, within a packed bed and on
Attempts to specify
top of a filter medium. In most cases we are concerned with the
equipment based simply
pressure drop required for a given flow rate of filtrate, or permeate.
on particle size are not
successful. However, in
This is required for equipment design, or simply to understand the
general finer particles do process. Numerous filter types are available, these are not described
require higher operating in detail, but the hydraulic operating principles are. Filter cake
pressures for filtration. A washing and air blowing, to displace moisture from the cake prior to
free expert system for thermal drying, are not covered but will be considered in Section 8.5.
selection is available at: The reader should appreciate the many similarities between
www.filtration-and- conventional and membrane filtrations, including the tendency for
separation.com the filter medium resistance to be a function of what is being filtered
and not just the medium itself. For most filter mediums it is not
possible to define a true pore size, or diameter, but an equivalent one
provided by a flow test based on hydraulic resistance, or the ability to
displace one fluid. This is met again in Section 13.5 and is called the
bubble point test.
Selection of equipment for solid/liquid separation depends on
many factors, both equipment and process related. Expert systems
exist for this purpose and the reader is referred to one that is freely
available on the left. However, some existing operating data, or
laboratory tests, are essential for reliable equipment specification.
4.10 Problems
1. Deep bed filtration
A 3 layer filter is used to clarify an effluent containing 60 mg litre−1 The following
(ppm) of solids. Complete the following table: differential equation
describes the
distribution of solids
Depth Bed Initial filtration Inlet Outlet
within the filter bed:
medium constant concentration concentration
to section from section  ∂N 
  = − λN
 ∂z  t
(m) −1 (ppm) (ppm)
(m )
0.45 anthracite 2 60 where N is solid
0.28 sand 8 concentration in any
0.22 alumina 15 consistent set of units
and λ is the filtration
The final treated effluent concentration discharged from the filter will constant. By considering
be (ppm):........................ only the start of the
filtration the above
partial differential
2. Five litres of a 10% w/w slurry of chalk ( ρ s of 2670 kg m−3) in equation may be treated
water (µ of 0.001 Pa s) was filtered under constant pressure on a filter as an ordinary
leaf of 0.0314 m2 area. The specific surface of the chalk is 3x106 m−1 differential equation,
and the cake can be assumed to have had a porosity of 50%. and integrated with
respect to position (z)
i). The volume of solids in the slurry was (cm3):
a: 100 b: 500 c: 200 d: 250
ii). The volume of water retained in the filter cake was (cm3):
a: 400 b: 200 c: 100 d: 0 Cake filtration
3 Resistances in series are
iii). The filtrate volume was (cm ):
a: 4600 b: 4750 c: 4500 d: 4800 additive
iv). The mass of dry solids in the cake was (kg); ∆Ptotal = ∆Pcake + ∆Pmedium
a: 0.533 b: 0.500 c: 1.335 d: 0.668 Hence
v). The dry mass of solids per unit volume of filtrate (i.e. the same as µαc dV µRm dV
∆P = 2 V +
'c') was (kg m−3): A dt A dt
a: 112 b: 100 c: 116 d: 2670 where c is the dry mass of
vi). Using the Kozeny-Carman equation, the permeability of the cake cake deposited per unit
volume of filtrate:
was (m2):
sρ
a: 1.1x10−14 b: 9x1013 c: 3.3x10−8 d: 3x107 c=
−1
(1 − m R s )
vii). The specific resistance of the cake was (m kg ):
where s is slurry
a: 8.3x10−18 b: 22500 c: 2.5x10−11 d: 6.7x1010 concentration as a weight
viii). The time taken to filter the suspension, if the medium resistance fraction and mR is mass wet
can be neglected and the differential pressure forming the cake was cake sample/dried sample
36 cm of mercury ( ρ Hg =13600 kg m−3), was (s): (i.e. the moisture ratio). The
connection between
a: 55 b: 998 c: 1572 d: 1747 specific resistance (α) and
permeability (k) is
3. The same quantity of a similar suspension to problem 2 was 1
filtered on the same filter at constant rate until 2.3 litres of filtrate had α=
k (1 − ε ) ρ s
been recovered. The following empirical equation for pressure was
observed
∆P = 168Pa/second + 6670Pa
Assume an incompressible filtration: i.e. α, c as well as A and µ were

unchanged from above.
i). The pressure drop over the medium during the filtration was
(Pa):
a: need Rm b: 168t+6670 c: 6670 d: need dV/dt
ii). The constant filtrate rate was (m3 s−1) - hint compare cake filtration
equation and the empirical equation for pressure rise:
a: 3.4x10−6 b: 3.7x10−7 c: 9.7x10−5 d: 4.6x10−6
iii). The time taken to achieve the filtration to this stage was (s):
a: 678 b: 6287 c: 24 d: 500
iv). The medium resistance was (m−1) - hint compare the equations
again – consider the two terms separately, this term is static:
a: 6.2x1010 b: 5.7x1011 c: 2.2x109 d: 4.5x1010
4. Following on from problem 3, after the 2.3 litres of filtrate was

Effective medium resistance
A previously formed cake recovered the pressure was then kept constant until all the
can be treated as being part suspension was filtered. Assume an incompressible filtration: i.e. α, c,
of an effective medium A and µ were the same as in Q.1 and see the box on effective medium
resistance (Rm'), and the resistance.
equation for the next i). The effective medium resistance was (m−1):
filtration can then be applied a: 4.5x1010 b: 5.7x1011 c: 6.2x1011 d: 1.2x1012
to a second half of the
ii). The volume of filtrate filtered under constant pressure was (m3):
filtration. The effective
medium resistance has two a: 0 b: 0.0023 c: 0.0046 d: 0.0025
components: the true iii). The total filtration pressure was (Pa):
medium resistance and that a: 90600 b: 10650 c: 1065000 d: 121000
due to the earlier cake iv). The time filtering at constant pressure was (s):
formation a: 519 b: 674 c: 731 d: 2470
V v). The total filtration time was (s):
R m ' = R m + cα
A a: 1230 b: 1350 c: 2490 d: 6810
5. A pilot experiment using a filter of 0.1 m2 area at a constant

Filt’n Filtrate Time pressure of 68.5x104 Pa produced the results on the left. The filtrate
time volume over viscosity was 0.0015 Pa s, slurry concentration was 3% w/w, cake
volume concentration was 52% w/w and the liquid density was 1000 kg m−3.
−
(mins) (litres) (s m 3): i). The moisture ratio was - hint consider 100 kg of wet cake:
10 88 a: 0.52 b: 1.92 c: 1 d: 0.30
20 125 ii). The dry solids per unit volume of filtrate was (kg m−3):
40 180 a: 30.9 b: 1080 c: 17.3 d: 31.8
60 220
iii). Draw a graph and calculate: the specific resistance of the cake
and the resistance of the medium
iv). On a 10 m2 filter what would be the filtrate volume after 2 hours?
v). If the solid density is 2500 kg m−3, what would be the cake
thickness in part (iv)?
5 Dilute systems
This chapter considers the behaviour of a single particle suspended in
Archimedes’ principle
a fluid. In practice, the equations and principles described are used to
States that when a body
understand how a number of particles behave, provided that the is wholly or partially
concentration is sufficiently low enough to ensure that the behaviour immersed in a fluid it
of the particle under consideration is not significantly interfered with experiences an upthrust
by the presence of other particles. Applications of the principles equal to the weight of the
covered include particle size analysis by sedimentation methods; fluid displaced
where the settling rate is related to the size of the particle, and the
industrial process of clarification by sedimentation: in which particles
are removed from a fluid stream by allowing sufficient time for the
particles to settle.
5.1 Weight, drag and Particle Reynolds number

All forces must reduce to Newton's basic equation
F = ma
Forces either cause particle motion in a fluid, or resist it. A force
balance can be written using all the forces described, or some of these.
The easiest force to appreciate is the particle weight, but this is just
one example of a field force. The particle weight is the product of its
mass and the gravitational acceleration. Particles are usually too
small to weigh; hence the particle diameter is used to calculate the
volume, which is then multiplied by the density to give the mass.
Thus, for a spherical particle, the particle weight is (in Newtons)
πx 3 Fig. 5.1 Flow streamlines in
ρs g (5.1)
6 fluid around a sphere
However, the particle will experience an upward force, in accordance
Archimedes’ principle, which numerically is
πx 3
ρg (5.2)
6
Hence, combining equations (5.1) and (5.2) provides the buoyed
particle weight
πx 3
(ρ s − ρ )g (5.3)
6
Before considering other field forces it is illustrative to conduct a
simple force balance to see the application of this approach. In a fluid,
particle weight will cause an acceleration that will be resisted by fluid
drag. When the fluid drag force is equal to the particle weight the
motion will be uniform; i.e. no longer accelerating and the particle Fig. 5.2 Streamlines and
will attain its terminal settling velocity. Fluid drag force comes from a turbulences in a fluid around
suitable solution to the Navier-Stokes equation. However, this has a sphere – at higher Re’
only been achieved analytically under conditions of no turbulence
within the fluid; i.e. streamlines of fluid flowing past the particle, as
46 Dilute systems
illustrated in Figure 5.1. Under these conditions Stokes’ drag

expression is valid
FD = −3πUxµ (5.4)
Galileo (1564-1642) which can be combined with equation (5.3) to provide an expression
Galileo is credited with for terminal settling velocity (Ut), called Stokes’ law
dropping different
x 2 (ρs − ρ)g
sized balls from the top Ut = (5.5)
of the leaning tower of 18µ
Pisa to show that they In equation (5.4) the drag force is related to the particle velocity (U),
fell at the same rate.
for all values of velocity, whereas in equation (5.5) we are referring to
This ignores air drag
the final (terminal) settling velocity of the particle in a static fluid
and Galileo knew better
than this, he had after the period of acceleration (Ut). Clearly, the settling rate of a
worked on wind particle is a function of its size, solid density and physical properties
friction. Note that of the suspending fluid. Equation (5.5) is only valid when the degree
equation (5.5) is not of turbulences within the fluid is negligible, see Figure 5.2. This is
valid for balls in air, but measured by The Particle Reynolds Number
(5.11) is generally valid.
xUρ
Re' = (5.6)
µ
where the threshold for streamline flow past the particle is believed
exercise 5.1 to be about 0.2. The Particle Reynolds Number measures the ratio of
Determine the maximum
inertial to viscous forces within the fluid; hence it is the fluid
particle size at which
properties that should be used in it: fluid density and viscosity. At
Stokes’ law should be
applied for the data: Particle Reynolds Numbers greater than 0.2 the degree of turbulence
solid density: 2500 kg m−3 becomes more significant leading to an additional fluid drag force
liquid density: 1000 kg m−3
due to form drag. Hence, the terminal settling velocity will be lower
viscosity: 0.001 Pa s than that predicted by Stokes’ law, equation (5.5), which considers
i.e. use Re’=0.2 and only viscous drag around the particle.
substitute equation (5.5) in In common with fluid flow in pipes, it is possible to correlate a
to (5.6). friction factor with Reynolds number for the case of fluid flow past a
single spherical particle. This is illustrated in Figure 5.3. The friction
factor is the shear stress in a plane at right angles to the direction of
motion at the particle surface (R) divided by the fluid density and
relative velocity between the particle and fluid squared. The drag
force is the product of the shear stress and the
particle area, which is the projected area to the
fluid flow (Ap). In particle settling it is usual to use a
drag coefficient (Cd), rather than friction factor,
these are related as follows
f R
Cd = = (5.7)
2 ρU 2
where the fluid drag (FD) is
FD = RAp (5.8)
The projected area for a sphere is
π
Fig. 5.3 The drag coefficient or friction Ap = x 2 (5.9)
factor plot for single spherical particles 4
Combining equations (5.7) to (5.9) provides

π
FD = C d ρU 2 x 2 (5.10)
4
Equation (5.10) can be equated with (5.3) to provide a generally valid
equation for the terminal settling velocity
2( ρ s − ρ )gx
Ut = (5.11)
3 ρC d
However, equation (5.11) can only be used to predict terminal settling
velocity if a value of the drag coefficient is known, see Figure 5.3. In
the streamline flow region we know that
12
Cd = (5.12)
Re'
which can be substituted into equation (5.11) together with (5.6) to
provide equation (5.5). Hence, the drag coefficient and Stokes’ law
approach to particle settling are compatible. However, for Particle
Reynolds numbers greater than 0.2 no single and simple analytical
function, equivalent to equation (5.12), can be used. Many
correlations have been suggested; all of them are valid over a
restricted range of Particle Reynolds numbers. An alternative
approach comes from considering the drag coefficient further
2 (ρ s − ρ )g
Cd = x (5.13)
3 ρU t 2
Equation (5.13) contains both settling velocity and particle size and
cannot be used to give the drag coefficient from a diameter because
the settling velocity is also required. However, multiplying by the
Particle Reynolds number squared results in
2 ( ρ s − ρ )g x U t ρ  2 ( ρ s − ρ )gρ  3
2 2 2
C d Re' 2 = x =  x (5.14)
3 ρU t 2 µ2  3 µ2 
The term in the square brackets contains neither particle diameter,
nor settling velocity. Likewise, dividing Particle Reynolds number by
the drag coefficient gives
Re' xU t ρ 3 ρU t
2
3 ρ2  3
= = U t (5.15)
Cd µ 2 ( ρ s − ρ )gx  2 (ρ s − ρ )gµ 
Equations (5.14) and (5.15) can be written as
3
Re' U t
C d Re' 2 = PH 3 x 3 and =
C d QH 3
where both PH and QH are not dependent upon particle size or settling
velocity. The friction factor correlation, Figure 5.3, can then be
Fig. 5.4 Modified drag and
redrafted in these terms to give Figure 5.4. So, for the purposes of Reynolds number plot
determining the settling velocity from a given particle diameter,
equation (5.14) provides a value for CdRe’2, Figure 5.4 is then used to
find Re’/Cd and equation (5.15) to provide the settling velocity.
48 Dilute systems
In practice Figure 5.4 is not very easy to use: it is a logarithmic

plot and the resolution reduces as a decade is approached. Thus it
may be easy to read off an unambiguous value at 11, but difficult to
read off a value at 90. To overcome this problem a set of tables was
produced by Heywood, correlating log10(PHx) against log10(Ut /QH)
and vice-versa, see the Appendix. Thus, in order to determine the
settling velocity from a particle diameter, log10(PHx) is first calculated
and used to determine log10(Ut /QH) from The Heywood Tables. This
value is then anti-logged and multiplied by QH to give the velocity.
Clearly, this procedure is only worth the extra computational effort
when the settling is at Particle Reynolds numbers greater than 0.2.
See the box on page 50 for an example of how to use the tables.
The main advantage of the Heywood Tables approach, over
empirical correlations between the Particle Reynolds number and a
derived function of the drag coefficient, is that it is valid for all
Particle Reynolds numbers. It is also possible to implement on a
computer and is available via the Internet at:
www.filtration-and-separation.com/settling
An alternative popular correlation, using the Particle Reynolds
number and The Archimedes (Ar) number, is
( )
2
Re' =  14.42 + 1.827 Ar 0.5 − 3.798
0. 5
(5.16)
 
valid for 2<Re’<20 000, where The Archimedes number is closely
related to equation (5.14) and is
 ( ρ − ρ )  gρ 2 x 3
Ar =  s  (5.17)
 ρ  µ
2
Thus, to determine a settling velocity the Archimedes number is

deduced from equation (5.17), followed by the Particle Reynolds
number by equation (5.16) and hence the velocity from (5.6).
5.2 Other forces on particles

Field forces other than the gravitational include
πx 3
centrifugal: F = ( ρ s − ρ ) rω 2 (5.18)
6
electrical, thermophoretic (due to a temperature gradient), thermal
creep (due to greater loss of molecules from the hotter side of a
particle), and photophoretic (due to a light intensity gradient). The
centrifugal field force is considered further in Chapter 8, electrical
and thermophoretic forces in 14 and colloidal forces in 13. The inertial
force (Fi) is the rate of change of momentum
π 3 dU
Fi = x ρs (5.19)
6 dt
note that the product of volume and particle density has been used
for mass, assuming a spherical particle.
The fluid drag force may be subject to the mean free path correction,
which is required when the particle size is comparable to the mean
free path of the fluid. This is required because the particles can slip
between the fluid molecules - effectively reducing the viscous drag. It
is more prevalent when the fluid is a gas. The correction to the drag
Brownian motion
coefficient is
When small particles
C d = C d(STOKES) [1 + 1.7λ / x ] −1 (5.20) are suspended in
where λ is the mean free path length of the gas. For air λ=0.1 µm liquids, they are
(approximately) so the error in assuming the Stokes drag term is 17% subject to molecular
for a 1 µm particle and 170% for a 0.1 µm particle settling in air. bombardment giving
When particles come to rest on each other, or a surface, there is a rise to Brownian
solids stress gradient, or reaction force. This force can be rationalised motion. Hence, finely
by considering when a particle is at rest at the base of a vessel, as it divided particles may
experiences no drag or inertia, but still possesses a weight (field) not settle. In practice,
force. This force must be balanced as the particle does not accelerate. particles smaller than
The reaction force is due to a pressure, or stress gradient, exerted 2 µm suspended in
from the vessel base. water will settle
slower than predicted
5.3 Particle acceleration in streamline flow by Stokes’ law and
particles less than 1
In the derivation of equations (5.5) and (5.11) the drag force was
µm might not settle at
equated to the gravitational field force, to determine the terminal
all.
settling velocity of the particle. This simple force balance is only valid
if the particle inertial forces can be neglected. Therefore, the time
taken to reach the terminal settling velocity, or 99% of it, is a useful
check on the validity of the simple force balance used to derive these
equations. A force balance of the apparent mass (buoyed mass), drag
and inertia for a spherical particle is
πx 3 dU
( ρ s − ρ ) g - 3πµxU − m p =0 (5.21)
6 dt
where mp is the actual particle mass, not buoyed mass. Equation (5.21)
can be rearranged to give
dU π ( ρ s − ρ ) x g / 6 − 3πµxU
3
= (5.22)
dt mp
but
π
( ρ s − ρ ) g = 3πµxU t (5.23)
6
i.e. the equation for the terminal settling velocity.
Therefore,
dU 3πµxU t
= (1 − U / U t ) (5.24)
dt mp
We need to integrate equation (5.24) to find time taken to reach a
given velocity, or fraction of terminal settling velocity. The equation
can be rearranged to give
−1 / U t 3πµx
∫ dU = − ∫ dt
1−U / U t mp
50 Dilute systems
i.e. a mathematical relation of the form:

f ' ( x)
∫ = ln[ f ( x )]
f ( x)
Therefore,
t =t
 − 3πµx 
[ln(1 − U / U t )]UU ==U0
= t
t
 m  t =0
the constant of integration is zero. Hence,
mp
t=− ln(1 − U / U t ) (5.25)
3πµx
where the actual mass of particle is mp:
π
mp = x 3 ρs
6
On considering equation (5.25) it should be apparent that the particle
will never reach its terminal settling velocity: it asymptotes to this
value. However, most small particles that are encountered within
Particle Technology will reach 99.9% of their terminal settling
velocity within a very short acceleration time. See Figure 5.5 for an
illustration of this.
5.4 Settling basin design (Camp-Hazen)

Fig. 5.5 Time taken to reach Figure 5.6 illustrates the principle behind continuous settling basin
99.9% of terminal settling design, using a rectangular clarifier. The feed flow enters the vessel
velocity – for the conditions on the left and plug flow conditions are assumed, with treated
illustrated effluent leaving on the right of the vessel. Whilst inside the vessel
particles sediment and if they reach the base of the vessel, before
being removed in the effluent, then the particles are assumed to stick
to the base and be removed from the liquid. Hence, the vessel design
requirement is to allow sufficient residence time within the vessel to
provide adequate particle removal. The design is based on the critical
trajectory model: where a particle size is selected and a balance
undertaken equating the time taken for the particle to settle the full
basin height, and the residence time within the basin assuming plug
flow. The resulting simple vector analysis of the trajectory is a
straight diagonal line: starting at the top left of the vessel and
finishing at the bottom right. All particles of this size will be
Fig. 5.6 Critical trajectory collected; as those starting their trajectory from further down than the
model for continuous
full vessel height (H) will follow a parallel trajectory to the critical
settling basin design
and, therefore, reach the vessel base before the full vessel length (L).
The time taken to settle will be
H
ts = (5.26)
Ut
and the residence time, assuming plug flow, is
HWL
tr = (5.27)
Q
where W is the vessel width (i.e. residence time is vessel volume

divided by volumetric flow rate). Equating these two times,
cancelling and rearranging provides
Q
= LW (5.28)
Ut
The product of the vessel length and width is the plan area. Hence, for
complete removal of particles of a given size, the volume flow rate
divided by the corresponding terminal settling velocity is equal to the
plan area for the complete removal. If the plan area is too small then
not all the particles of the selected size will be removed. In settling it
is always the plan area that is the important design parameter and
not the vessel cross-sectional area. This important fact will be met
again and in all cases the provision of a too small plan area will result
in incomplete settling, or particle removal.
Further consideration of equation (5.28) and (5.5) provides an
indication of the efficiency of removal of particles smaller than the
critical size. All particles larger than the critical will sediment out in
the available time, and the fraction of smaller ones removed will be
directly proportional to the settling velocity – assuming that the feed
flow is in fact uniformly distributed over the height of the vessel and
not all entering the vessel at the top. Hence, particles half the critical
size will only be collected with 25% efficiency because the settling
velocity is proportional to the particle diameter squared.
5.5 Laboratory tests

Practical laboratory tests to deduce settling parameters for the design
of industrial clarifiers involve the short tube and long tube tests. Figure Fig. 5.7 The long tube test
5.7 illustrates the long tube test, where the suspension is allowed to
settle within the tube for a set time and the contents above a sample
point are drained off. The concentration above the sample point is
determined by weighing and drying. For each height it is possible to
plot the concentration remaining in suspension against inverse time,
as illustrated in Figure 5.8. It may be possible to extrapolate this plot
to an inverse time value of zero; which will represent the
concentration of unsettlable solids. These are fine particles that
remain suspended due to molecular bombardment, or colloidal
repulsion forces. Assuming a plot similar to Figure 5.8 provides a
suspended solids concentration that is acceptable for an effluent from
a continuous clarifier, then the required settling time and height of
the sample point (measured downwards from the suspension top),
are used in the following equation for vessel plan area
Fig. 5.8 Results from long
Q 1 tube test
A= (5.29)
H / t EA
where EA is an area efficiency to take into account turbulences, poor
flow distribution, etc. within the vessel.
52 Dilute systems
5.6 Summary
The settling velocity of small particles may be reliably obtained from
Stokes’ law. Larger particles, however, do not obey Stokes’ law.
Alternative correlations between drag coefficient and Particle
Reynolds number do exist – but the settling velocity is a constituent
of the Particle Reynolds number; hence the answer needs to be
known before the appropriate equation to use can be identified! To
overcome this problem Heywood published a set of tables that can be
used over a wide range of settling velocities and particle sizes.
The single particle settling discussed in this chapter is widely used in
engineering calculations. For example, within a spray drier trajectory
analysis is often performed using the drag coefficient and the
difference in velocity between the particle and the gas is of use in
mass transfer calculations. Single particle settling also forms the basis
for understanding the behaviour of more concentrated dispersions,
which is the subject of the next chapter.
Heywood Tables
5.7 Problems
(see Appendix) 1.
 4 ( ρ s − ρ ) ρg 
1/ 3 i). A solid and liquid has a specific gravities of 2.8 and 1.0,
PH =   respectively and the liquid viscosity is 0.001 Pa s, the value of the
 3µ 2  function PH is 2.87x104 m−1, the value for QH is (SI units):
and a: 2.87x10−2 b: 2.87x104 c: 2.87 d: 0.490
1/ 3
 4( ρ s − ρ ) µg 
QH =   ii). The SI units of the function QH are:
 3ρ 2 
a: m−1 b: m s−1 c: m s−2 d: s m−1
Both functions are size
and velocity independent.
When calculating the iii).Use the Heywood Tables to complete the following:
settling velocity given a Particle diameter (µm): 1 10 50 100 1000
particle diameter (x) the log(PHx): ------ ------ 0.
value of log(PHx) is first log(Ut/QH): ------ ------
calculated. Then the first
two significant figures of Settling velocity* (m s−1) ------ ------
−1
log(PHx) are given by the Stokes’ settling velocity (m s ):
first column of the table, * settling velocity using the Heywood Tables
the second and third
come from the scale given iv). Why should the Stokes’ settling velocities of the larger particles
at the top of the table (the always be greater than those found in practice (and given by the
first row). The
Heywood Tables)?
corresponding value of
log(Ut /QH) is then read or
estimated from the table
and converted into a xU t ρ
v). The Particle Reynolds is defined as Re ′ =
value for Ut using the µ
calculated function QH.
which should be below some threshold for Stokes’ law to be
applicable. The maximum particle size at which Stokes’ law is
applicable for the above system is (µm):
a: 59 b: 5900 c: 127 d: 1271
2.
i). See the continuous settling basin on the right. What will be the
trajectory of particles the same size as the critical particle size, but
which start their descent from a height less than H ?
ii). The solid and liquid densities are 2900 and 1000 kg m−3, the
viscosity is 0.001 Pa s and the critical particle diameter is 50 µm, the
terminal settling velocity (Ut) is (m s−1):
a: 2.6x10−3 b: 2.6x10−5 c: 2.6x10−1 d: 0.52
iii). The Particle Reynolds number is:

a: 0.26 b: 129 c: 0.129 d: 0.00013
In the Camp-Hazen
iv). An expression for the critical particle residence time vertically (tv ) settling basin model the
is (s): feed to a basin is
a: t v = L / U b: t v = H / U c: t v = H / v d: t v = L / U assumed to enter well
mixed and distributed
evenly over the full
v). An expression for the critical particle residence time horizontally depth of the vessel. The
(th ) is (s): critical trajectory is given
a: t h = LWH / Q b: t h = H / Q c: t h = L / U d: t h = Q / LWH by a particle of a certain
(critical) size that enters
vi). An expression for LW is (SI units): the basin at the top left
and has just settled by
a: LW = U / Q b: LW = Q / U c: LW = HQ / U d: LW = Q / HU
the end of the basin
length - bottom right.
vii). What are the units of LW, and what does it represent? The critical trajectory will
be a straight line.
viii). If the volume flow rate into the basin is 10 m3 min−1 the Particles smaller than the
critical size will also have
minimum settling area required to remove all particles of the
a straight line trajectory
diameter given in Part (ii), and bigger, is (m2): but one that does not
a: 6.4 b: 64 c: 640 d: 6400 intercept with the base
by the time the fluid
3. element has reached the
i). An effluent containing a mineral in suspension with solid and end of the basin.
water densities of 2600 and 1000 kg m−3, respectively, is pumped into
a batch vessel 5 m high and left for 30 minutes prior to discharge into
a river. The viscosity of water is 0.001 Pa s. The maximum particle
diameter that will be in the discharge is (µm):
a: 3180 b: 56.4 c: 90 d: 66.4
ii). The effluent solid has the following particle size distribution:
Cumulative mass undersize (%): 100 92 80 62 48 31 18 8 4 0
Particle diameter (µm): 90 80 70 60 50 40 30 20 10 0
If the initial concentration of the effluent before settling was 60 mg l−1,

the concentration of solids below the size calculated in your answer
in Part (ii) (the critical size) is (mg l−1):
a: 60 b: 43.2 c: 34.2 d: 25.8
54 Dilute systems
This represents a 'worst case' estimate of the concentration in the

effluent discharge after settling, as it assumes that no solids smaller
than the critical size settle in the allowed time.
iii). Of the concentration of solids below the critical size a

considerable fraction will also have settled out. The amount settled
out at each particle diameter is proportional to the ratio of its settling
velocity compared to the velocity of the critical particle. For example,
a particle with a settling velocity half that of the critical particle will
travel 2.5 m in the allowed 30 minutes and, if we assume the
suspension was homogeneous before settling, half of the solids at that
diameter will settle out. Complete the following table:
Diameter (µm): 70 60 56.4 50 40 30 20 10 0

Fraction settled at
size: 1.00
Fraction undersize: 0.80 0.62 0.57
iv). Now, to estimate the amount of material settled below the

critical size a plot of fraction of particles settling in allowed
time against fraction of material undersize is made and the
area under the curve is calculated by graphical means. Plot
these on the left.
The area under the curve is:

a: 0.133 b: 0.265 c: 0.53
This represents the fraction of the total distribution below the critical
size but which still settles because the particles still reach the base of
the vessel in 30 minutes. Add this fraction to the fraction of material
in the size distribution above the critical size (which has all settled
out). See your answer to Part (ii) to help you find this. NB the next
question does not want the fraction settled – it asks for the effluent
concentration going to discharge.
v). Hence, the concentration in the effluent discharge is (mg l−1):

a: 9.1 b: 43.2 c: 34.2 d: 18.3
vi). If the effluent discharge consent limit is 20 mg l−1, will you be able
to discharge this suspension?
6 Hindered systems and rheology
On increasing the particle concentration from the low values
considered in the previous chapter, the system properties will change
considerably from those of the continuous phase, usually water, and
the individual particles. In settling, the presence of a large number of
fine particles will hinder the fall of the larger particles and the very
small particles will be dragged down more quickly than under free
settling. There is an important question to be considered in our
treatment of high concentration suspensions: are we interested in the
behaviour of the particles compared to the continuous phase, such as
during hindered settling, or the behaviour of the suspension as a new
homogeneous phase, such as during pumping of a mixture? The
latter is the concern of buoyancy, viscosity and rheology and is
considered in the later sections. Initially, the settling of particles
within a continuous phase will be considered. Another fundamental
concern is the particle concentration at which hindered systems are
appropriate, rather than the models discussed in the previous
chapter. In general, hindered settling is appropriate when the particle
concentration is greater than about 1% by mass. To describe
concentrated suspensions we will need to use some of the definitions
Fig. 6.1 Illustration of
from Chapter 3. An illustration of porosity and solid concentration by porous medium –
volume fraction is reprinted in Figure 6.1. including settling
The industrial equipment in which hindered settling is conducted suspension
is simply tanks, which may be operated batch-wise or continuously.
Settling is a cheap method of concentrating solids, the driving
potential for it is free (gravity), but it provides only a limited final
solid concentration and the process is slow. However, it is very often
used in thickening a suspension before a more capital intensive
operation, such as filtration. The design of industrial thickeners is
covered in Section 6.3.
6.1 Hindered settling and zone theory

One method to ascertain if a suspension is settling in the hindered
settling regime is to mix the suspension thoroughly and to watch it
settle in a laboratory measuring cylinder. If an interface between the
settling suspension and clearer residual liquid is apparent, then the
settling is within the hindered settling regime. It may be possible to
observe the rate of descent of the interface with time, as illustrated in
Figure 6.2. Below the settling interface exists a porous medium,
similar to that illustrated in Figure 6.1. If it were possible to turn the
measuring cylinder upside-down, without everything falling out, Fig. 6.2 Hindered settling in
then the liquid velocity upwards, required to keep the settling cylinders – interface fall
interface stationary, would be the same as the superficial velocity with time
illustrated in Figure 6.1. Hence, the settling velocity is the same as the
superficial velocity and we can justifiably use the symbol Uo for both.
The technical term for the residual liquid above the interface is the
supernatant and, for the duration of this chapter, it will be assumed
that it is entirely free of solids. It will also be assumed that there are
no wall support effects for the settling suspension, a vessel of
diameter equal to, or greater than, 150 mm is often recommended for
this purpose, but tests in different diameter vessels can show if this
effect is important.
The hindered settling velocity is a strong function of the particle
concentration and, for a given material, nothing else; i.e. it is
independent of vessel diameter, shape, etc. The dependency on
concentration is logical: at the highest possible concentration no
Fig. 6.3 Variation of settling can take place, at low concentration the particle will settle
settling velocity with under free settling conditions. Thus, in hindered settling the settling
starting concentration velocity will be between these two limits and a mathematical
expression for hindered settling velocity could be considered to be a
correction term to the free settling velocity (Ut). The most famous
empirical relation between settling velocity (Uo) and solid
concentration by volume fraction (C) is the Richardson and Zaki
equation
U o = U t (1 − C ) n (6.1)
where n is a variable constant that depends on the Particle Reynolds
number and may be dependent on vessel diameter (d):
Particle Reynolds number n for small tubes n for large tubes

< 0.2 4.65 + 19.5 x/d 4.65
0.2 < Re’ <1 (4.35 + 17.5 x/d) Re’
−0.03 −
4.35 Re’ 0.03
1 < Re’ < 200 (4.45 + 18 x/d) Re’
−0.1
4.45 Re’−0.1
200 < Re’ < 500 4.45 Re’
−0.1
4.45 Re’−0.1
Re’ > 500 2.39 2.39
In Figure 6.3 all the settling interface plots are straight lines, followed
by curves. This can be explained by considering what takes place
within the settling suspension. At the start, the concentration is Cf and
uniformly distributed within the vessel. At a time δt the concentration
at the base of the vessel is Cf +δC. At the next instance in time, 2δt, the
concentration at the base is Cf+2δC. If we were to track the
concentration Cf+δC we would find that it is now slightly higher up
the vessel than the base. Thus layers of constant concentration appear
to propagate upwards. Of course, all the solids are settling; no solids
are propagating upwards – we are merely looking for a region of a
given concentration of solids. It takes some time before the solid
concentration below the settling interface increases from that of the
Fig. 6.4 Solid concentration original and, according to equation (6.1), settling velocity is a unique
increases starting at the base
function of concentration; hence, the settling velocity must remain
and apparently rising
upwards in the vessel
constant, giving rise to the initial straight lines in Figure 6.3. The
concentration increases from the base are illustrated in Figure 6.4.
In our settling suspension we can see that there are zones, or regions,
of: supernatant liquid, the original concentration (just below the
settling interface) and it is logical that there will be another zone
consisting of fully settled sediment at the base of the vessel. There is a
fourth zone: that of variable (increasing) concentration lying between
the sediment and initial concentration zones. This is illustrated on
Figure 6.5 and the mathematical description of the concentrations in
this zone was first published by Kynch in the 1951, and is covered in
Section 6.4. At this stage, it is reasonable to suggest that because the
settling velocity is a unique function of concentration the velocity at
which the concentrations propagate upwards will be also, giving rise
to lines emanating from the origin called concentration characteristics.
6.2 Batch settling flux

When a characteristic reaches the settling interface the concentration
at the interface will become the value of the characteristic and, in
accordance with equation (6.1), the settling velocity will become the
value given by that concentration. This leads to an increasing
Fig. 6.5 Zones within a settling
concentration at the interface, after the end of the constant settling
suspension at a selected time t
rate period, hence the settling rate decreases with time.
The height of interface against time plot, as illustrated in Figure
6.3, is very simple to obtain from experimental data and is very
powerful in the information that can be deduced from it. Design
information is concerned with the ability to pass a required mass of
solids per unit area and time; i.e. kg m−2 s−1, or mass of solids per unit
time kg s−1, this is the solids flux. Mathematically, the solids flux
(kg s−1) due to settling in a batch vessel is
G = U o CAρ s (6.2)
However, in most cases the vessel area and the solids density are
constant; hence, the solids flux is usually abbreviated to (G’)
G' = U oC (6.3)
which has the units m s−1. A batch flux curve, Figure 6.6, can be
Fig. 6.6 The batch flux curve
plotted by conducting several sedimentations at different
concentrations and measuring the initial settling velocities at the
concentrations used, as illustrated in Figure 6.3. The batch flux curve
possesses a maximum and this can be explained by consideration of
the two limits: at zero concentration the flux must be zero, in
accordance with equation (6.3), and at the highest possible
concentration the flux will again be zero because the settling velocity
term will be zero. Between these two extremes the flux will have
finite values; hence, the batch flux curve must possess a maximum, as
illustrated in Figure 6.6.
Fig. 6.7 Construction to
The knowledge that concentration characteristics reach the settling
transform a settling curve
interface, and then the interface settles at the velocity for that
concentration, can be used to determine the settling velocities for
concentrations greater than that used in the original suspension. The

justification comes from a material balance as no solids are lost
C f H o Aρ s = C1 H 1 Aρ s (6.4)
where Ho represents the full height of the suspension before settling
commences. Thus it is possible to mix the suspension and obtain a
settling velocity at concentration Cf; then after some time to remove
some supernatant liquid and fully mix the suspension again and let it
settle. This will give a slower settling velocity and the solid
concentration of the fully mixed suspension can be deduced from
equation (6.4) by rearrangement to obtain the new initial
concentration
H
C1 = o C f (6.5)
H1
Clearly, it is physically possible to perform the experiment to obtain
the settling rate at the new concentration, but it is also possible to
perform the transformation of the data graphically, without the need
of the experiment, as the two settling curves are identical after the
point at which the concentration characteristic reaches the interface.
This is illustrated in Figure 6.7.
6.3 Thickener design

A continuous thickener is a vessel with a feed, at low solid
concentration, and two output streams: an overflow of clean liquid
Fig. 6.8 Fluxes in a thickener and an underflow suspension of much greater concentration than the
feed. The vessel is normally circular, with a conical bottom that is
raked to bring the solids into the discharge well. The design
requirement is to deduce the plan area required for a given flow rate
of solids entering and to achieve the desired degree of thickening. If
insufficient area is provided then the concentration of solids within
the vessel increases and will eventually leave in the overflow. The
fluxes within a thickener include the batch flux, described above, but
there is an additional flux due to the continual removal of material
from the base; i.e. underflow. The fluxes are illustrated in Figure 6.8,
which includes a schematic diagram of the thickener.
The thickener feed flux is illustrated in Figure 6.9 and this is the
mass feed rate entering the system. Equating all the flux terms
provides, where F is the thickener feed rate (m3 s−1)
G = A(U o + T )Cρ s = FC f ρ s
i.e. the feed flux must be equal to the batch and underflow
withdrawal flux within the thickener, thus
FC f
A= (6.6)
−1
Fig. 6.9 Flux (kg s of solids) (U o + T )C
fed to the thickener The flux at an arbitrary height within the thickener will be equal to
the flux at the underflow (TCu), which can be substituted into
equation (6.6)
FC f
A= (6.7)
(U o + T )C u
but the batch flux at the underflow will be negligible compared to the
underflow withdrawal flux; hence we can write
FC f
A= (6.8)
TC u
Equation (6.8) is the design equation and to use it we must determine
the flux at the underflow concentration.
The underflow withdrawal induces a downward velocity within
the thickener which is constant for all the concentrations present from
Cf to Cu. Hence, the underflow withdrawal flux is a straight line on a
graph of flux against concentration, see Figure 6.10. Adding the batch
and underflow withdrawal flux together gives the composite flux
curve; which has a minimum at a critical concentration between Cf and
Cu. It is this concentration that has the minimum solids flux, or
handling, ability. Hence, if too much solids are added to the thickener Fig. 6.10 Underflow
the critical concentration (Cc) will build up within the device and withdrawal flux
eventually overflow. The batch, underflow and composite flux curves
are illustrated in Figure 6.11.
In Figure 6.12 the construction required to determine TCu, for use
in equation (6.8), is illustrated. The minimum composite flux occurs
at Cc and is numerically equal to the underflow flux at Cu; i.e. TCu.
However, the composite flux at the critical concentration does have
two components: that due to batch settling and underflow
withdrawal, which are marked on the figure. Clearly, it is possible for
the thickener to be operated under conditions that require a batch
flux less than the value at the batch flux curve, at Cc, but batch fluxes
greater are not possible. Hence, the limit of operation is where Cc
meets the batch flux curve. So, for a required underflow Fig. 6.11 All the flux
concentration (Cu) a line drawn through the concentration axis at Cu curves
and tangential to the batch flux curve will meet the flux curve at the
value of TCu. This value can then be used in equation (6.8) to
determine the thickener area.
Thickener height is not determined by flux theory and, in general,
thickeners are short vessels with diameters up to 50 metres.
Minimum heights are allocated for the raked zone (0.5 m), solids
storage zone (0.5 m) and clarification (0.5 m). Thus, a thickener is
usually 1.5 to 4 m in height, unless solids compression is important.
An alternative design method to the use of the batch flux curve
construction described above was originally described by Coe and
Clevenger. A flux balance between the feed and underflow provides
FC f Fig. 6.12 Construction to
G = ATC u ρ s = FC f ρ s hence T =
AC u determine the critical flux –
equation (6.8)
which can be substituted in to equation (6.7) and rearranged to give
FC f 1 1 
A=  −  (6.9)
Uo  C C u 
Equation (6.9) is solved by selecting concentrations between Cf and

Recommendation Cu, where Uo is required for the value of C selected, and using the
Leave Sections 6.4
greatest area for the design. Equation (6.9) is easier to apply than the
and 6.5 until after
graphical technique described earlier, but equation (6.8) and the
completing the
problems. graphical construction has the advantage that it can be used to
predict the underflow concentration from an existing thickener under
different operating loads; i.e. FCf.
6.4 Kynch analysis

Figure 6.13 includes a slice through a settling suspension and
illustrates a batch settling curve. A mass balance on the solid slice
gives, in terms of kg s−1 of solids:
input CU o Aρ s
 d (CU o ) 
output CU o + dz δz  Aρ s
 
dC
accumulation Aρ s δz
dt
where z is the vertical co-ordinate in the batch settling vessel. Using
input - output gives accumulation, taking to an infinitesimal distance,
cancelling and rearranging gives
dz d (CU o )
=− (6.10)
Fig. 6.13 A lamina layer dt dC
within a settling suspension i.e. the rate of increase in height (dz/dt) of a concentration C is the
and the batch settling curve differential of the product of the concentration and the settling
velocity, which could be calculated from equation (6.1). Equation
(6.10) shows that the propagation velocity will be a unique function
of solid concentration, a consequence of the settling velocity, equation
(6.1), being a unique function. Thus, equation (6.10) shows
mathematically that the propagation velocity for a given
concentration will be a constant; hence, the straight lines from the
origin to the point where they meet the settling interface curve.
Equations (6.10) and (6.3) show that the propagation rate of a
concentration characteristic can be obtained from the tangent, or
differential, to the batch flux curve drawn at that concentration value.
6.5 Compressible sediments

The various types of settling are illustrated in Figure 6.14. It is called
Fitch's paragenesis diagram. Clarification was described in Chapter 5
and the earlier part of this chapter discussed zone settling, which
occurs when each particle behaves in an incompressible manner and
is free to move within the suspension, subject to undergoing
Fig. 6.14 Diagram showing hindered settling. Flocculent sedimentation is when the particles are
types of sedimentation stuck together, either by a synthetic flocculent or by natural
aggregation. This leads to compressible sediments. Compression

during sedimentation is sometimes called self weight filtration, and the
compressible equations (4.15) and (4.16) can be applied. During
compression a solids stress gradient exists (dPs/dz), this gradient is not
present in zone settling. A force balance, neglecting inertia, on a
lamina settling layer gives (stress = solids weight – liquid drag)
dPs µ
= Cg ( ρ s − ρ ) − U o (6.11)
dz k
where k is the permeability. For a more detailed discussion on this
topic the interested reader is directed to more advanced texts – see
the further reading section.
6.6 Homogeneous systems

The earlier sections considered particles sedimenting in a continuous
phase, such as water. The remaining sections cover the treatment of
homogeneous systems, where the particles do not separate from the
continuous phase, but their existence leads to modification of the
properties of the continuous phase such as buoyancy and viscosity.
These considerations are, therefore, relevant to situations when the
suspension needs to flow within a pipe, etc.
The buoyancy force exerted by a fluid is the upthrust exhibited by
the weight of the fluid displaced, see Archimedes’ principle on page
Fig. 6.15 Comparison of
45. However, if a large object is suspended in a stable suspension,
Newtonian viscosity
then it will experience an upthrust due to the liquid and surrounding equations correcting for solids
solids combined. The effective density of the surrounding continuous presence and experimental
phase is average density of the fluid and solids combined data
ρ m = Cρ s + (1 − C ) ρ (6.12)
The buoyancy correction for a large object within a suspension is
ρs − ρm
Hence, it is possible to quickly measure the solid concentration of a
suspension using a hydrometer to obtain the mixture density and
equation (6.12) rearranged to give
ρm − ρ
C= (6.13)
ρs − ρ
However, equation (6.12) is only valid for objects that experience the
surrounding suspension as a continuous phase and not just the fluid
component of it; i.e. objects larger than the particles in suspension.
The viscosity of a suspension depends upon the solid
concentration and the nature of the solids. For suspensions exhibiting
Newtonian flow behaviour a well-known correction for the presence
of solids is Krieger’s equation
µ e = µ (1 − K ' C ) −η '/ K ' (6.14)
where µe is the effective Newtonian viscosity, K’ is a crowding factor
which is equal to 1/Cmax (which is 1.56 for spheres) and η ' is the
intrinsic viscosity, which is 2.5 for spheres. A comparison of the

Krieger equation and an alternative is illustrated in Figure 6.15,
together with experimental data obtained with mono-sized latex
spheres in water.
When pumping suspensions, the pressure drop may be calculated
using Newtonian flow equations, with equations (6.12) and (6.14)
employed to correct for the presence of solids. However, this is only
true for suspensions exhibiting Newtonian flow behaviour and most
suspensions, at high concentration, exhibit non-Newtonian rheology.
6.7 non-Newtonian rheology

Figure 6.16 illustrates the common, time-independent, rheograms.
For Newtonian flow the simple relation between shear stress (R) and
rate is
du
R = −µ (6.15)
dz
where µ is the coefficient of dynamic viscosity. Equation (6.15) is a
single parameter model (i.e. just viscosity) to relate shear stress to
rate. The next most complicated rheological model is a two parameter
one
Fig. 6.16 The common time
m
independent rheograms  du 
R = −K   (6.16)
 dz 
where K is the consistency coefficient, m is the flow index and u is the
fluid, or suspension, velocity. Equation (6.16) is known as the power
law model and it mathematically represents both pseudoplastic and
dilitant flow on Figure 6.16; corresponding to m values less than, or
greater than, unity respectively.
Under laminar flow conditions it is possible to combine equation
(6.16) with a force balance at a pipe wall
a∆P
R= (6.17)
2L
where a is the pipe radius, and derive an analytical equation for
pressure drop with suspension velocity, or flow rate, in a similar way
to the derivation of the Hagen-Poiseuille equation for Newtonian
fluids. The resulting equation is
1/ m
mπa 3  a∆P 
Q=   (6.18)
3m + 1  2 LK 
As with single phase systems, on increasing the velocity energy
losses due to turbulence within the flowing suspension will become
more significant. Much of the work on power law fluids was
published by Dodge and Metzner in the 1950 and 60’s, and they
Fig. 6.17 Lamella separator
derived the Generalised Reynolds number to distinguish between
or thickener – has lots of
sedimentation channels in
laminar and turbulent flow. The threshold for the onset of sufficient
parallel turbulence to invalidate the use of equation (6.18) is about 2000 and
the Generalised Reynolds (Re*) number is
ρ m u 2−m d m
Re * = m
(6.19)
K  6m + 2 
 
8 m 
For Newtonian fluids, in the turbulent flow regime, the following
equation correlates the friction factor and flow Reynolds number
−1 / 2   f 1 / 2 
f 
  = 2.5 ln Re   + 0.3 (6.20)
2  2 
 
The analogous correlation for power law fluids, and suspensions, is
( f )−1 / 2 =
m
4
0.75
(
ln Re* f (1− m / 2 )
) − m0.4
1. 2
(6.21)
where the friction factor is related to the wall shear stress by

f R
= (6.22)
2 ρu 2
6.8 Summary
If a continuous thickener is designed correctly sufficient area will be
present to allow the solids to settle into the underflow, whilst
supernatant liquid reports to the overflow. If the total solids flux is in
excess of the design value, the thickener will have insufficient
capacity for the throughput required. Suspension, at the critical
concentration, will build up inside the thickener and eventually
overflow. Above the hindered settling region there may be some free
settling (clarification) or simply a region of clear liquid. Some special
designs exist that enhance the throughput, such as the lamella
separator illustrated in Figure 6.17, but the principles of hindered
settling are similar to those described in the earlier sections – the
designs provide an enhanced effective plan area for the separation.
Equation (6.11) suggests that it is possible to use permeability,
covered in Chapter 3, to deduce the settling rate (Uo) given that for
incompressible (zone settling) the solids stress gradient will be zero.
This is true, and it is possible to rearrange equation (6.11) for settling
velocity and combine with equation (6.3), for flux, or equation (6.9)
for plan area of a thickener. If the Kozeny-Carman model of
permeability is used, however, it is usual to reduce the Kozeny
constant from the usual value of 5 to 3.3, as this is believed to give a
better fit to the data. However, there is a wealth of evidence to
support this so-called constant being a function of solids
concentration.
6.9 Problems
1.
i). A 3% v/v suspension is to be settled in a batch settling tank prior
to water reuse within a process. A sample of the suspension settled
with a clear interface when placed in a 1000 ml measuring cylinder.

The batch settling curve was as follows, plot the data on the grid, or
graph paper.
Time 0 20 40 60 80 100 120 140 160 200 240 280

(mins):
Height 28 24.2 20.4 16.5 13.6 11.2 9.4 8.1 6.8 5.4 4.4 4.0
(cm):
ii). It is proposed to operate the settling tank on a three day cycle, in

which 1000 m3 of suspension is run into the tank each day, allowed to
settle and some of the liquid recycled back to the process. At the start
of the third day 85% of the liquid from the first two days has been
recycled. The total solid volume in the tank after addition of
Rheology summary for Q.4
A commonly used suspension on day 3 is (m3):.......................
correlation between shear The total liquid volume in the tank is (m3):.......................
stress (R) and shear rate (γ) is the tank design is now based on what happens on day 3, on day 4 the
the power law expression sludge in the tank will be pumped out and the cycle started again.
m
R = Kγ iii). It may be assumed that the action of adding suspension on day 3
where K is the consistency
completely mixes the tank to give a uniform suspension leading to a
coefficient and m is the flow tank solid concentration of (% by v/v):
index. For laminar flow the a: 6.7 b: 3.0 c: 9.0 d: 7.1
pressure drop is related to iv). It is possible to write a mass balance relating the height (Hf) of a
the flow rate via Wilkinson's uniformly mixed suspension of one concentration (Cf) to the height
equation (H1) and concentration (C1) of the same mass of solids but mixed to an
m
∆P 2 K  (3m + 1)Q  alternative concentration, as follows:
=
L a  πa 3 m  C f H f Aρ s = C1 H 1 Aρ s
where Q is the volumetric where A is vessel area and ρ s is the solid density. Thus, you have
flow rate and a is the pipe been given the settling data for a 3% v/v suspension above which
radius. The generalised needs to be converted into settling data at the concentration
Reynolds number (Re*) is
determined in Part (iii); the height (H1) required in the above
8 ρ m u 2− m d m equation is (cm):
Re * =
K (6 + 2 / m ) m a: 28.0 b: 9.3 c: 12.6 d: 14.0
For turbulent flow the v). Take a ruler and draw a line from the height determined in Part
friction factor is (iv) making a tangent to the settling curve plotted in Part (i). You
−1 / 2   f 1 / 2  have now obtained a settling curve for a uniform suspension of the
 f 
  = 2.5 ln Re   concentration given in Part (iii). note that settling rate is independent of
2  2 
  vessel diameter (but depends strongly on suspension concentration),
hence the settling curve obtained in a 1000 ml measuring cylinder
and, combining (6.17) & will be the same as that obtained in a large process vessel.
(6.22),
vi). As settling rate is independent of vessel diameter one design is to
f ∆Pd
= construct a vessel of height given in Part (iv). The total vessel volume
2 4 Lρu 2 would need to be sufficient to accommodate the volumes given in
Part (ii). This would make the vessel area (m2):
a: 4825 b: 1070 c: 48250 d: 10700
vii). At the end of the settlement period, i.e. on day 4 after pumping
the supernatant but before pumping the sediment out, the volume of
sediment is (m3):
a: 90 b: 527 c: 437 d: 1351
viii). Hence, given the vessel area from Part (vi), the height required
for the sediment is (m):
a: 0.049 b: 129 c: 0.129 d: 0.00013
ix). Draw a line from the origin of your settling graph to meet the
settling curve at the height given in Part (viii). This occurs at the time
(minutes):
a: 80 b: 160 c: 220 d: 280
You have now completed one design for this settling vessel: the
height comes from Part (iv), the area from Part (vi), and the time
required to settle from Part (ix).
x). Comment below on your design.
xi). The line you have just drawn from the origin to the settling curve
represents a solid characteristic at a concentration greater than that
given in Part (iii). It can be used to provide another vessel design.
Assuming that the maximum permissible time for settling is 24 hours,
the height of this characteristic after 24 hours is (cm):
a: 12.6 b: 28.0 c: 32.0 d: 56.0
xii). Assuming that this height again represents the depth of sediment
in the vessel, the new vessel area required to accommodate the total
sediment on day 4 is (m2):
a: 1650 b: 3290 c: 6590 d: 13200
xiii). Under these conditions the new vessel height will be (m):
a: 4.02 b: 2.17 c: 0.82 d: 0.28
xiv). Explain below how other characteristics can be used to provide
alternative designs:
2.
i). A continuous thickener is to be designed to deal with the effluent
from the last question. It will treat the 1000 m3 per day of suspension
fed at 3% v/v solids concentration and is to discharge underflow at
13.8% v/v solids. Use the settling curve from question (1) and a mass
balance to complete the following table.
Concn (v/v) 0.03 0.039 0.045 0.049 0.056 0.067 0.074 0.092
Height on axis (cm) 28 21.5
Velocity (m s−1) 3.2x10-5
Batch flux (m s−1) 9.5x10-7
note batch flux is the product of settling velocity and solid concentration
Mass balance, where Cf
ii). Plot the batch flux curve on the grid provided, or on graph paper. is any concentration in a
iii). Now a flux balance on a thickener provides the following result: batch settling vessel and
A(TC u ) = FC f = YC u Hf is the corresponding
where A is the thickener area, (TCu) is the critical thickener flux which height:
is the intercept of a line drawn as a tangent to the batch flux curve C f H f Aρ s = C1 H 1 Aρ s
and going through the desired underflow concentration, F and Y are
the volume feed and underflow rates respectively, Cf and Cu are the
volume fraction feed and underflow concentrations respectively.
Note that T is, in effect, the velocity of solid movement in the
thickener caused by underflow withdrawal at the solid concentration
Cu. The critical flux in this thickener giving an underflow discharge
concentration of 13.8% v/v solids is (m s−1):
a: 10x10−7 b: 8.5x10−7 c: 7.2x10−7 d: 5.8x10−7
iv). The minimum thickener area for this duty is (m2):
a: 480 b: 29000 c: 16000 d: 960
v). If the thickener is circular in cross-section the minimum thickener
diameter is (m):
a: 25 b: 190 c: 140 d: 35
2 −1
vi). The underflow rate is (m hour ):
a: 1.25 b: 2.4 c: 4.6 d: 9.1
3 −1
vii). The overflow rate is (m hour ):
a: 40.4 b: 39.3 c: 37.1 d: 32.6
3. An existing 5 m diameter thickener is to be used to thicken 2400

tonnes per 24 hours of flocculated slurry containing 10% solids by
mass (0.037 v/v) in water. The solid density is 2900 kg m3. The
following batch sedimentation results were obtained in a test:
Time (mins): 0 2 4 6 8 10 12 20 30
Interface height 45.6 36.5 28.0 21.6 16.8 14.5 13.2 10.6 9.7
(cm):
What will be the underflow concentration? (Ans 19% by mass)
4.
i). Rheological tests on milk of magnesia have provided the following
In laminar flow of
data. Calculate the consistency coefficient and the flow index.
Newtonian fluids
pressure drop and flow
rate are proportional – shear rate (s−1): 7.2 16 64 320 720
see how this suspension shear stress (Pa): 7.0 9.1 14.3 24.2 31.6
shear thins?
ii). The suspension must be pumped 2.11 m down a pipe of radius
6.95 mm. Mean suspension density is 1300 kg m−3. Complete the
table and sketch the graph of pressure drop against flow rate. How
does it differ from what is expected of a Newtonian suspension?
For Laminar flow: For Turbulent flow:

Pressure Flow rate Velocity Re* Velocity Re* f -1/2 RHS of
drop equn(6.17) equn
− − −
(kPa) (m3 s 1) (m s 1) (m s 1) & (6.22) (6.20)
5 -------- -------- -------- -------- --------
15 -------- -------- -------- -------- --------
30 -------- -------- -------- --------
40 4.66
80 -------- 7.64
7 Fluidisation
The fluidisation principle is straightforward: passing a fluid upwards
through a packed bed of solids produces a pressure drop due to fluid
drag. When the fluid drag force is equal to the bed weight the
particles no longer rest on each other; this is the point of fluidisation.
The superficial velocity at this point is known as the ‘minimum
fluidising velocity’ (Umf). If the fluid velocity is increased further the
pressure drop does not significantly increase – it remains equal to the
bed weight per unit area, but the bed may expand; i.e. grow taller as
illustrated in Figure 7.1. Commercial gaseous fluidised beds are
usually operated at flow rates many times that required for minimum
fluidisation, typically 5 to 20 times. Liquid fluidised beds operate at
values closer to Umf. A material balance indicates that, in general
Cf H o
Ci = (7.1)
Hi
On page 53, the hypothetical case was made for turning the vessel
containing hindered settling solids upside down and noting that the
liquid velocity upwards, required to maintain the position of the
interface, is equal to the settling velocity of the solids in an otherwise
stationary liquid. This upward fluid flow, and balance of forces, is the Fig. 7.1 Bed expansion
hydrodynamic condition that exists during fluidisation. Thus, the during particulate
Richardson and Zaki equation, page 54, is also valid for liquid fluidisation – mass of solids
fluidised systems and the minimum fluidising velocity is a superficial is the same in all beds
velocity, as illustrated in Figure 6.1 inverted.
Fluidisation is a popular means of contacting solids and a fluid
because of the high degree of mixing and the resulting high transfer
coefficients (heat and mass). There are numerous examples including:
catalytic conversion of hydrocarbons, drying, combustion, calcination
(application of heat to decompose a solid – for example calcium
carbonate to oxide or gypsum solids to plaster), agglomeration, etc.
Another useful advantage is the uniformity of the bed temperature,
so that heat sensitive materials can be treated in a well controlled
environment. However, the biggest disadvantage of gas beds is the
need for dust control and treatment – which can be more expensive
than the capital and running cost of the fluidised bed itself. An
example of a gaseous fluidised bed is provided in Figure 7.2. In the
figure several important aspects are recorded: the gas fluidised bed is
not uniform, but bubbles of gas within the bed are observed (see
Section 7.2), a distributor plate supports the solids and distributes the
fluidising gas, above the bed the vessel diameter increases thereby
reducing the gas velocity so that entrained particles may drop back
into the bed from the freeboard and a gas cyclone is used for primary Fig. 7.2 Example design of a
particle separation from the gas stream. The cyclone dip tube enters gas fluidised bed
the bed – hence preventing gas from entering the cyclone from the
68 Fluidisation
solid’s outlet. The cyclone inlet is simply open to the freeboard. Solids
feed into a fluidised bed is often a significant challenge. Screw
feeders used on free flowing materials may form a seal to prevent the
gas escaping but, this cannot be guaranteed and rotary valves may
need to be used. Reaction rates within the fluidised bed are usually
rapid, hence the bed material is often inert, or reaction product, the
actual amount of solid reacting within the bed may be very small; e.g.
in fluidised bed combustion of coal a bed of calcium carbonate may
be used and the coal particles may only be between ½ to 2% of the
bed by mass. The calcium carbonate becomes oxide and then reacts to
form calcium sulphate if sulphur dioxide is present. This provides in-
Fig. 7.3 Up to the fluidisation situ sulphur dioxide emission removal. A high pressure drop over
point pressure with
the distributor plate is usually required to ensure adequate
superficial velocity is linear in
accordance with Darcy’s law:
distribution of fluidising gas over the entire area of the bed. Solids
off-take can be by a simple overflow or via the gas outlet, or cyclone.
µ
∆P = LU o
k
On increasing the flow
7.1 Minimum fluidising velocity
particles rearrange before The fundamental equation for fluid flow through porous media,
fluidisation giving rise to the under laminar flow conditions, is Darcy’s law, equation (3.4). In most
maximum – on decreasing instances the Kozeny-Carman equation (3.7) is preferred because it
flow a lower pressure drop has an explicit expression for permeability in terms of bed porosity
may be found, giving a and specific surface. The pressure at the base of a fluid (due to fluid
hysterisis.
weight) comes from the static component of Bernoulli’s equation
∆P = L ρ g
where L is fluid height, g is the acceleration due to gravity and ρ is
the fluid density. For a suspension, a similar equation is valid but, for
the static pressure due to the solid particles, we must take account of
buoyancy and the proportion of particles present. Clearly, if there are
no solids present there would be zero static pressure due to solids –
but rather than use the solid concentration by volume fraction we use
the porosity, see Figure 3.1, thus
∆P = L( ρs − ρ ) g (1 − ε ) (7.2)
−3
where ρs is the true solid density (kg m ). Combining equations (7.1)
and (3.4) – remembering that we stated that fluidisation occurs when
Fig. 7.4 Pressure gradient
the bed weight (per unit area) equals the fluid drag gives
with superficial velocity up to
and after fluidisation – linear k
U mf = ( ρ s − ρ ) g (1 − ε ) (7.3)
as above whilst L is constant, µ
but drops when L increases:
N.B. the minimum fluidising velocity is a superficial velocity (not
∆P µ
= Uo interstitial). It is common to see equation (7.3) written with the
L k permeability term expanded as provided by the Kozeny-Carman
equation, see equation (3.7), and assuming spherical particles,
equation (3.8), gives an alternative equation for minimum fluidising
velocity
( ρ s − ρ ) gε 3 xSv 2
U mf = (7.4)
180(1 − ε ) µ
7.2 Types of fluidisation

On increasing the fluid velocity, up to the point of fluidisation, flow
patterns are usually well described by Darcy’s law. However, after
the fluidisation point two very distinct types of fluid flow are
observed: particulate and aggregative fluidisation. In the former case
the bed behaves in a uniform manner: as the flow rate is increased the
bed height increases; hence, the increasing fluid flow simply goes to
expand the bed, as illustrated in Figure 7.1. The overall pressure drop
remains constant, and equal to the bed weight per unit area, until
entrained particles are elutriated out by the fluid flow. In particulate
fluidisation the fluid superficial velocity and bed porosity may be
related by the Richardson and Zaki expression, see equation (6.1).
In aggregating, or bubbling, fluidisation aggregates (of fluid) may
be observed within the fluidised bed that move rapidly to the surface.
This type of fluidisation is often associated with the fluidisation of
solids using gases. This is probably the most commercially important
type of fluidisation. Hence, an understanding of aggregative, or
bubbling, fluidisation is important. See the pictures of bubble
formation and passage through a bed illustrated in Figure 7.5.
In an aggregative fluidised bed the fluid can pass through the bed
in a similar fashion to particulate fluidisation and bubbles of fluid
form. The bubbles may travel very quickly through the bed – hence in
the case of catalytic cracking of hydrocarbons this provides a way in
which some hydrocarbon vapour can by-pass the catalyst and,
therefore, reaction. So, a bubbling bed has an emulsion phase
surrounding the bubble and a lean phase where the bubble is lean of
solids. These bubbles form spontaneously and, because most
fluidisations of commercial interest take place in bubbling beds, a
large amount of research has gone into characterising and Fig. 7.5 Numerical
understanding bubble behaviour. simulation of aggregative
There are several correlations designed to help answer the simple (bubbling) fluidisation: a
question: will a bed bubble or not? The simplest is based on the gas bubble rising in a
Froude number: fluidised bed. Note the
scale bar on the left:
U mf 2
Fr = (7.5) white represents 100%
xg voidage and black 50%
voidage. The bed is just
For high (>1) values bubbling is more likely. Other correlations fluidised in the bottom
include the use of Reynolds number, density ratio and bed ratio: picture.
xU mf ρ ρ s − ρ Lmf
Re' = ; ;
µ ρ db
where Lmf is the bed height at minimum fluidisation and db is the bed
diameter. The empirical correlations are:
 ρ − ρ  Lmf 
Fr Re'  s   < 100
 particulate fluidisation
 ρ  d b 
 ρ − ρ  Lmf 
Fr Re'  s   > 100
 bubbling fluidisation
 ρ  d b 
70 Fluidisation
A more comprehensive attempt to describe fluidising behaviour, than

the empirical correlations provided above, has been published and is
known as the Geldart Powder Classification Chart (1973, Powder
Technology, 7, p285), see Figure 7.6. The following regions are
identified in Geldart’s chart:
Group A: Aeratable, may not bubble, if it does then bed expands
before bubbling, may have fast moving bubbles less than 100 mm.
Groups B & D: large bubbles, may form slugs. Group D gives slow
bubbles.
Group C: Cohesive, high interparticle forces leads to difficult
Fig. 7.6 Geldart’s Powder fluidisation, may form channels or slugs instead.
Classification Chart for
fluidisation 7.3 Bed design and bubbling behaviour
In practice, many different types of fluidisation may occur and the
vessel geometry can influence this significantly. Figure 7.7 illustrates
other fluidisation bed behaviour due to: slugging, channelling and
spouting. In the latter two cases, a special design is sometimes used
in which the bed is continually in axial motion, and a distributor
plate might not be employed. Also included on Figure 7.7 is the effect
of the bed behaviour on the pressure curve illustrated in Figure 7.3.
In another special case, the calcination of gypsum, a conical
fluidised bed is used. The application here is to convert the gypsum
(calcium sulphate dihydrate) into plaster (calcium sulphate
hemihydrate), for use as plasterboard on walls, fillers, etc. and the
particles are gypsum produced from flue gas desulphurisation (FGD)
at large power stations. The calcinations can be written:
CaSO4.2H2O → CaSO4.½ H2O + 1½H2O
The particle Sauter mean diameter is 40 µm and the density is
approximately 2.4 g cm−3. The powder is well into Geldart’s Group A
– which should provide a reasonable quality fluidisation, but the
FGD process is a wet one and although the particles are dewatered
the product still has sufficient residual moisture for the wetting force
to be very strong – causing particle cohesion. Hence, the particles can
behave cohesively and do not fluidise easily. The actual, and unusual,
fluidised bed design used is illustrated in Figure 7.8. It does not
employ a distributor plate and uses the combustion products from
natural gas and air, as well as the generated one and a half moles of
water vapour per mole of calcium sulphate, as the fluidising media.
Typical throughputs of 40 tonnes per hour for a fluidised bed 2.5 m in
diameter are possible.
Fig. 7.7 Bubbling behaviour 7.4 Gas flow patterns around bubble and stability
A bubble is stable so long as particles that are drawn up into the
bubble from underneath; i.e. are carried into the bubble by the fluid,
can fall back out again. Unstable bubbles fill in with particles from
the bottom. A simple schematic diagram of a gas bubble, within a
surrounding emulsion phase of the fluidised bed, is illustrated in
Figure 7.9. Real bubbles tend to be more mushroom shaped than

spherical but spheres are assumed for easier modelling. Two different
gas flow patterns may be observed around a bubble; depending upon
the speed of the bubble through the bed relative to the velocity of the
gas through the surrounding emulsion phase. These two types are
illustrated in Figure 7.10. When the gas bubble is travelling slower
than the gas within the surrounding fluidised bed, the presence of the
bubble will provide a region that has lower resistance to the gas flow
than the surrounding emulsion phase. Hence, the gas will use the
path of least resistance and there will be a tendency for gas to pass
Fig. 7.8 Conical kettle type
through the bubble in preference to the surrounding emulsion phase. fluidised bed – no
Thus, lines representing the gas flow will bend towards the bubble distributor plate
and, after passing through the bubble, the lines will bend away from
it again. This is similar to lines of electrical flux, or magnetism,
bending towards a region of higher conductivity within an otherwise
uniformly resistant matrix.
When the gas bubble travels faster than the gas in the surrounding
fluidised bed, the gas that leaves the bubble at the top will then be
immediately overtaken by the bubble (as it is travelling faster than
the gas in the bed). Thus, the gas below the bubble will be the gas
that recently left it at its top and will re-enter the bubble. The gas Fig. 7.9 Gas bubble with
flow pattern is one in which gas recirculates around a fast moving solids trapped in wake
bubble. The gas does come into contact with some of the solids within
the bed, but only those within the cloud associated with the bubble.
Hence, the speed of the bubbles through the bed and the restricted
contact of gas with solid particles means that this type of fluidisation
can have poorer fluid-solid contacting than the slower moving
bubbles and particulate fluidisation. One of the first successful
mathematical analyses of this type of fluidisation was by Davidson
and Harrison, in the 1960’s.
7.5 Davidson and Harrison model

The model assumed that the fluidised particles can be treated as an
incompressible fluid, called the particulate phase, which has the same
porosity as the bed when it just became fluidised, i.e. at incipient
fluidisation. The model also assumed that the fluid can occupy the
same space as the particulate phase and can be treated
incompressibly, and that the fluid and particulate phases can be
linked by Darcy’s law. The model is based on gas velocities centred
around a bubble; hence spherical polar coordinates were used, this is
illustrated in Figure 7.11 and Uθ and Ur represent the gas angular
velocity in the plane of the paper and towards (and away from) the
centre of the circle. The resulting model gives the gas velocities as
R 3  Fig. 7.10 Slow and fast
U r =  b3 (U b + 2U ) − (U b − U ) cos θ (7.6) moving bubbles
 r 
72 Fluidisation
R 3 
U θ =  b3 (U b / 2 + U ) + (U b − U ) sin θ (7.7)
 r 
where U is the interstitial gas velocity, Ub is the bubble velocity, Rb is
the radius of the bubble and r is the radial coordinate (i.e. distance
from bubble centre). The gas velocity in the azimuthal direction is
assumed to be zero; i.e. the angular velocity of the gas coming out of
the plane of the paper is zero.
Considering Figure 7.12 representing a fast moving bubble, on the
horizontal plane the gas flow rate upwards though the bubble will be
the same as that going downwards through the cloud because there is
no interchange of gas with the surroundings. Hence, using A for the
Fig. 7.11 Illustration of cloud radius, the gas flow rate through the bubble will be equal to
spherical polar coordinates A
∫ 2πrεU θ dr (7.8)
Rb
At an angle of 90o sinθ is unity, hence the gas flow rate will be
A  Rb 3 
∫ 2πεr  3 (U b / 2 + U ) + (U b − U ) dr which integrates to give
Rb  r 
  1
πε (U b + 2U )R b 3 
1
( )

−  + (U b − U ) A 2 − R b 2  (7.9)
  Rb A  
Fig. 7.12 Fast moving Now considering the periphery of the cloud, where there is no
bubble and cloud interchange of gas with the surroundings, Ur→0 and r=A can be used
in equation (7.6) to provide
Rb 3
=
(U b − U ) (7.10)
A 3 (U b + 2U )
Crossover velocity with rearrangement and substitution into equation (7.9)
At U=Ub gas recirculation 2
 A3  R
in the cloud extends over: gas flow rate = − 2πε (U b − U )1 − 3  b (7.11)
A→∞  R b  2

i.e. crossover between the
two types of flow round a Equation (7.10) can be used to relate cloud and bubble radii, thus
gas flow rate = − πεR b 2 (− U b − 2U ) = 3πεRb 2U
bubble shown in Figure
(7.12)
7.10. So, at crossover:
U mf Using the assumption that the gas in the emulsion phase has the
Ub = same velocity as it did at incipient fluidisation, then the interstitial
ε mf
and Umf may come from velocity will be
equation (7.4). This links U = U mf / ε mf (7.13)
the properties of the
particles, bubble size and
and the porosity of the emulsion phase remains substantially
fluidisation bubbling unaltered from that of εmf thus combining equations (7.12) and (7.13)
type.
gas flow rate = 3πRb 2U mf (7.14)
Equation (7.14) suggests that the gas flow rate going through a
bubble is three times the minimum fluidising velocity flowing over
the bubble cross-sectional area. Another significant result from the
concepts used in this modelling is that the gas in excess of that
needed to fluidise the bed will pass through the bed in bubbles.
Hence, in aggregative fluidisation bed expansion will occur because
of displacement of particulate phase bed by the bubbles, not due to
the homogenous redistribution of solids represented by equation
(6.1); i.e. the Richardson and Zaki equation is not valid for
aggregative fluidisation.
The Davidson and Harrison model has been refined by other
research workers since the 1960’s, but it represents the starting point
for many of these later models. Its advantages include: a relatively
simple model to apply, appears to predict gas flow near a bubble,
shows how gas can by-pass solids in a fluidised bed, explains
stability of a bubble (by predicting gas flowing up through one) and
it predicts pressure variation near a bubble reasonably. However,
there are several deficiencies including: the bubble shape is wrong
(mushroom shape is observed), it doesn’t explain what happens
when particles enter a bubble, the maximum predicted bubble size is
wrong, it assumes incompressible phases and it uses Darcy’s law to
link the phases together. More recently, with the advent of increasing
and readily available computing power an alternative approach to
the analysis of homogeneous phases has been to consider the
behaviour of individual particles themselves. This is known as
Discrete Element Analysis, or Distinct Element Method.
7.6 Discrete element analysis

Discrete Element (DE) analysis has been applied to many different
operations in particle technology, including: flow in hoppers, mixing,
pneumatic conveying, etc. It is based on determining all the forces
acting on a particle and computing the net acceleration to apply over
a time increment. At the next time increment the step is repeated.
Clearly, this is computationally very complex and it is usually
applied to two dimensional simulations. However, the results can be
more realistic than a continuum based modelling approach as
described in the previous section. An example of the level of detail in
results that is possible is given in Figure 7.5, which is based on a DE
analysis conducted at Tsuji Laboratory, Osaka University. Fig. 7.13 Contact force
In fluidisation, it is possible to model the particle movement by between two particles
represented by spring and
DE and a conventional numerical solution for the gas flow. Mass and
dashpot
momentum balances are applied within the solution. The particle
motion is described by Newton’s equation of motion, which in the
context of the appropriate forces for fluidisation modelling can be
stated as
d2z
= (FC + FD + FW ) / m p (7.15)
dt 2
where mp is the particle mass and the forces are due to respectively:
collision, fluid drag and gravity. Hence, under appropriate conditions
equation (7.15) is solved to give
74 Fluidisation
dz dz d2z
= + 2 ∆t (7.16)
dt t + ∆t dt t dt
and
 dz dz 
z t + ∆t = z t +  + 0.5∆t
 (7.17)
 dt t dt t + ∆t 
The drag force used in the analysis usually comes from either the
correlations described in Chapter 3, for the dense phase – such as
Ergun’s correlation, or the drag coefficient described in Chapter 5, for
a region dilute in particles.
The contact force consists of normal and tangential forces, which
can be modelled in terms of a spring, dashpot and a friction slider as
shown in Figure 7.13. The normal and tangential component of forces
are expressed as the sum of the forces due to the springs and dash
pots. The stiffness of the springs and the viscous dissipation
coefficient of the dash pot determine the force with which the
particles interact. A very soft spring would cause the particles to
embed into each other, causing a large overlapping region between
the particles. By allowing some particle overlapping in space it is
possible to overcome the problem of defining how the particles
deform on contact and the mathematical solution is simpler. In the
model, two particles move towards each other until the contact force
changes its sign; i.e. from attraction to repulsion. Contact between a
particle and a wall is modelled in a similar way, except that the wall
is stationary.
Modelling particle interaction with its neighbours and the wall is
very computationally intensive and is restricted to simulations using
several thousand particles, at best. This may be sufficient for a
reasonable indication of the physics of what is taking place, but
Specific surface actual processes normally contain many millions of particles.
For gas phase reaction Nevertheless, DE modelling of processes within particle technology
modelling the specific will become more reliable and extensively used as computing power
surface is often much increases. They also have the advantage that additional forces can be
greater than that added into the solutions, e.g. colloidal interaction, and that a range of
provided by equations particle sizes can be considered.
covered in Chapter 2.
This is due to internal
cracks and pores in
7.7 Summary
catalyst particles. This In a packed bed of static solids the weight of the particles is
area may be determined transmitted to the base by a network of contacts between the
by gas phase adsorption particles. The two forces present are the solids stress gradient
tests. However, for (upthrust reaction force) and the particle weight. During upflow of
calculations of pressure fluid through the bed the equations for flow and drag, Darcy's law or
drop, e.g. equation (7.4),
Carman correlation, are valid and at some point the fluid drag will
the particle envelope
specific surface should
equal the bed weight. This is the fluidising point. After the point of
be used. This is the same fluidisation the bed may expand further, the gaps between particles
value as that covered in getting bigger. During fluidisation, the pressure loss of the fluid is
Chapter 2. equal to the bed weight and this will remain constant irrespective of
the bed expansion. Hence, the pressure drop remains constant.
When fluidised, the bed adopts all the properties of a usual fluid
including: lighter objects float, surface stays horizontal, solids would
pour out of a hole in the vessel side, the level of two connected beds
would be equal (like a U tube) and the static pressure within the bed
is given by depth x density. These properties assist in some of the
advantages of a fluidised bed, which include: easy to control and
automate, rapid mixing (excellent temperature control), large and
small beds are usable, very high heat and mass transfer coefficients.
Some of the disadvantages include: little scope for temperature
gradients giving thermal shock, a good distributor plate is usually
needed, expensive dust cleaning equipment may be necessary,
particle breakup, erosion and stirred tank behaviour gives non-
uniform residence times.
Fluidised beds are frequently used for reactors, such as the vapour
phase catalytic cracking of hydrocarbons. In this operation multiple Basic equations
beds are used; cracking taking place in one bed under one set of The Kozeny-Carman
equation for fluid flow
reaction conditions and regeneration of catalyst in a separate bed.
through porous media is
Transfer of catalyst is easily performed because of the fluidised
∆P  5(1 − ε ) 2 S v 2 
behaviour of the solids being similar to that of a fluid. In liquid = µ U o
fluidisation, segregation between particles based on size and density L  ε3 
difference is possible and this is used in the regeneration of bed filters where ∆P is the pressure
using mixed beds of sand and carbon. Segregation is not so common drop over the bed, L is the
in gas fluidised beds because solid/solid mixing is encouraged by bed depth, µ is the
entrainment of particles within the wake of the gas bubbles. viscosity, ε is the bed
porosity, Sv is the specific
7.8 Problems surface area per unit
1. Minimum fluidising velocity (Umf) is a superficial velocity. Use volume of the particles and
Uo is the fluid superficial
the Basic equations box to derive an equation for the minimum
velocity.
fluidising velocity where Umf=f[g, ρ s , ρ , µ , ε and diameter (x)]:
During fluidisation the
U mf = pressure drop over a
fluidised bed is constant
2. and equal to the bed
i). A packed bed consisting of 1.96 kg of solids of density 2.8 g cm−3 weight, i.e.
is contained in a cylindrical vessel of 10 cm internal diameter, and the ∆P = (1 − ε ) g ( ρ s − ρ ) L
bed height is 20 cm. The volume of the vessel occupied by the bed is where g is the acceleration
(ml): due to gravity and ρ and
a: 157 b: 1570 c: 3140 d: 6280 ρ s are the fluid and solid
ii). The volume of the solids in the vessel is (ml): densities respectively.
a: 700 b: 1430 c: 70 d: 5490 The modified Reynolds
iii). The porosity of the bed is (-): number (Re1) is:
a: 0.554 b: 0.773 c: 0.446 d: 0.227 ρU o
Re 1 =
iv). The particle size is 500 µm and the liquid density and viscosity S v (1 − ε ) µ
are 1000 kg m−3 and 0.001 Pa s, minimum fluidising velocity is (m s−1): The Froude number (Fr) is:
a: 9.4x10−6 b: 0.021 c: 0.017 d: 0.0094 U mf 2
v). Was the use of the Kozeny-Carman equation justified? Fr =
gx
vi). What type of fluidisation is likely to occur?
76 Fluidisation
3.
Richardson and Zaki i). Another fluidised bed contains 1.8 kg of solids of density 1.3 g
Uo = U tε n cm−3 in a cylindrical 30 cm internal diameter vessel, with a bed height
where Ut is the terminal
of 20 cm. The bed porosity is (-):
a: 0.853 b: 0.976 c: 0.098 d: 0.902
settling velocity (modified for
wall effects if necessary) and ii). The particles used above have a diameter of 200 µm and the fluid
n is an exponent defined as density and viscosity are 1.29 kg m−3 and 2x10−5 Pa s, respectively.
follows The terminal settling velocity of a particle is 0.783 m s−1 (from the
n = 4.6 ; at Re’<0.2 Heywood Tables - check this), the Particle Reynolds number is, see
n = 4.4 Re −0.03 ; at 0.2<Re’<1 Chapter 5:
n = 4.4 Re −0.1 ; at 1<Re’<500 a: 0.101 b: 10.1 c: 10200 d: 1x107
n = 2 .4 ; at Re’>500 iii). For this system, the value of the exponent n in the Richardson
and Zaki equation should be:
a: 3.49 b: 4.60 c: 4.11 d: 2.40
iv). The superficial velocity required to fluidise these particles to
achieve the above conditions is (m s−1):
a: 1.8x10−5 b: 0.71 c: 0.53 d: 0.55
v). The gas flow rate required to fluidise these particles to achieve the
above conditions is (m3 s−1):
a: 0.15 b: 380 c: 0.039 d: 3.9
4.
Hint Q.4 Estimate the bed expansion and bypass fraction for a fluidised bed
See comments at the foot
with incipient fluidisation velocity of 0.1 m s−1 if it is operated at a
of page 72. A volume
balance is needed for the superficial velocity of 0.15 m s−1 and has 2 cm diameter bubbles rising
bed expansion. at 0.5 m s−1. At incipient fluidisation the porosity is 0.4.
5.
A gas bubble in a fluidised bed reactor was found to be typically 100
mm in diameter and to be rising with a velocity of 0.5 m s−1. Using
Hint Q.5 the Davidson and Harrison model, investigate the particle size at
See the box on crossover which the transition in the gas flow pattern occurs: from bubbles
velocity on page 72.
rising slower than the interstitial gas velocity to bubbles with gas
circulation between the bubble and its cloud. Assume typical values
of particle physical properties; take the viscosity of air to be 2x10−5 Pa
s. Finally, draw an estimated graph of the thickness of the cloud
surrounding a bubble of 100 mm diameter as a function of the size of
the fluidised particles.
6.
Using the Davidson and Harrison two phase model, derive an
equation for the rate of flow of gas into and out of a gas bubble in
terms of bubble radius, interstitial velocity and bubble velocity.
Discuss how the model may help in the design of catalytic fluidised
bed reactors.
8 Centrifugal separation
The basic equations for most centrifugal modelling were introduced
in Chapter 5. The liquid drag force was given in equation (5.4), under
streamline flow, and the centrifugal field force was provided in
equation (5.18). It is a simple matter to equate these to arrive at an
analogue equation to the terminal settling velocity, equation (5.5), but
with one significant difference
2 2
dr x ( ρ s − ρ ) rω
= (8.1)
dt 18µ
the distance with time differential is not constant. In a centrifugal
Fig. 8.1 Particle in rotation
field the particle moves radially, see Figure 8.1 and equation (5.18), and definitions
and the radial position is part of the field force – hence the particle
accelerates during its travel in the radial direction. Thus, to determine
the particle position as a function of time integration is required.
It is well known that from a strict physical definition of forces on a
particle, in circular motion, the centripetal force and not the
centrifugal force should be considered. An unrestrained particle
would leave its orbit tangentially if the centripetal force was
suddenly removed. This is what happens with particles in cyclone
separation from gases and this is discussed further in Chapter 14.
However, this chapter is concerned with separation of particles in Buoyancy
rotating flow within a viscous medium, usually water. The particle If a particle floats, rather
will not travel tangentially to one orbit, but to lots of orbits, giving than sinks, then it will
the impression of radial movement outwards (provided the particle is move inwards in a
centrifugal field. Particles
denser than the surrounding continuous phase). Mathematically, we
denser than the fluid will
can use the well-known expressions, such as equations (5.18) and
move outwards. The
(8.1), to describe this travel. centrifugal field acts like
As illustrated in Figure 8.1, the centrifugal acceleration is simply an enhanced
the product of the radial position (r) and the square of the angular gravitational field in
velocity (ω). The SI units of angular velocity are s−1, but calculated by equation (5.3) and it is
converting from revs per minute (rpm) into radians per second – then usual to speak in terms
ignoring the dimensionless radian term. In solid body rotation, such of the equivalent ‘g’
force: i.e. centrifugal
as a centrifuge, this is easily calculated from the rotational speed,
acceleration / 9.81 m s−2.
usually provided in rpm. Thus, 1 rpm is 2π s−1 as an angular velocity.
In free body rotation, such as the hydrocyclone, the angular velocity
is calculated from the tangential velocity (uθ) by
uθ
ω= (8.2)
r
this is also illustrated on Figure 8.1. In the hydrocyclone the principle
known as the conservation of angular momentum is used; in which
knowledge of the tangential velocity at any radial position can be
used to calculate the tangential velocity at another because
u θ1r1 = u θ 2 r2 = constant (8.3)
or to take account of frictional losses within the hydrocyclone

u θ1r1 n" = u θ 2 r2 n" = constant (8.4)
where n” is an empirical constant, usually between 0.6 and 1.
In filtration within a centrifugal field the body force acts on the
liquid, which can pass through the filter medium, or septum, similar
to a washing machine or spin drier. The rotation acts in a similar way
as increasing the pressure effecting the filtration and it is possible to
deduce what this equivalent pressure difference is, using an equation
analogous to that given by the static component of Bernoulli’s
equation (depth x density x acceleration)
∆PCH = ρω 2 ( ro 2 − rL 2 ) / 2 (8.5)
where ro is the radius of the centrifuge and rL is the inner liquid radial
position and ∆PCH is sometimes called the centrifugal head.
From all of the above, it should be apparent that modification of
the equations discussed in Chapters 4 and 5, for an enhanced body
force due to rotation is simply required.
8.1 Sedimenting centrifuges

The analogue continuous gravitational equipment design, to
centrifuges, was covered in Section 5.4. Applying a similar logic to
the critical trajectory model illustrated in Figure 5.6, the critical
particle enters the centrifuge at radial position rL and leaves at radial
position ro – assuming that the particle is denser than the liquid, see
Figure 8.2. Equating the times taken for the particle to move radially
and for it to progress axially provides the following expressions,
based on equations (8.1) and (5.4)
 18µ  1
t (axial) =  2  2 ln (ro / rL ) (8.6a)
 x ( ρ s − ρ )  ω
Fig. 8.2 Critical particle t ( radial) =

( )
π ro 2 − rL 2 L
(8.6b)
2Q
trajectory in a centrifuge
where the equation (8.6b) is the volume of the machine divided by
the volume flow rate; i.e. analogue to equation (5.4). Combining these
equations and multiplying through by the acceleration due to gravity
( )
π ro 2 − rL 2 L 
= 2
18µ  g ln (ro / rL )

2Q  x ( ρ s − ρ ) g  ω2
The term in the square brackets is the terminal settling velocity under
gravity, equation (5.5), making this substitution and rearranging
gives
=
2
( 2
Q π ro − rL Lω 2 ) (8.7)
Ut g ln (ro / rL )
It is notable that the left hand side of equation (8.7) is identical to the
left hand side of equation (5.28). Thus, equation (8.7) represents a
centrifuge that has the same settling capacity as the plan area of a
gravity settling basin. This is illustrated further as follows, the

parameters that are defined by the process are called sigma-process
Q
∑ process = (8.8)
Ut
i.e. for a given particle size there will be a certain flow rate (Q) at
which all particles of this size are removed. If this flow rate is
exceeded then particles of this size start to appear in the effluent.
Thus, the process variables (size and flow rate) defines the sigma
value. If the centrifuge is 100% efficient, the sigma process will be
equal to the machine based parameters, called sigma-machine
∑ machine =
( )
π ro 2 − rL 2 Lω 2
(8.9)
g ln (ro / rL )
Both equations (8.8) and (8.9) have the SI units of area and both
represent the theoretical plan area of a gravity settling basin that
would perform the same separation duty on the solids. Introducing
an efficiency factor (EA) to take account of poor flow distribution
within the machine and other factors reducing the separation
capacity gives
Q
= A o
(
E π r 2 − rL 2 Lω 2) (8.10)
Ut g ln (ro / rL )
It is worth noting that, under gravity, particles less than 2 µm in size
might not settle because of molecular bombardment from the liquid
and colloidal forces. However, in a centrifugal field the body force is
much stronger and these particles have a greater chance of settling.
Hence, a separation that might not be possible under gravity might
be possible in a centrifuge. Obviously, a separation that is possible
under gravity will be much quicker in a centrifuge, due to the
enhanced g force. However, if there is little density difference
between the particle and fluid the separation under gravity and in a
sedimenting centrifuge will be slow, or impossible.
In a continuous sedimenting centrifuge there is always the
problem of how to remove the deposited solids continuously and
Fig. 8.3 Some sedimenting
how to enhance the separation. Various designs are used including:
centrifuge designs
scroll discharge decanter, time activated nozzle discharge disc stack
and the continuous tubular centrifuge. These are illustrated in Figure
8.3. In the scroll discharge machine the Archimedean scroll rotates
very slightly slower than the centrifuge, to convey the solids up the
beach and out of the machine. The disc stack centrifuge provides a
lot of parallel settling chambers, similar to the lamella separator in
Figure 6.16. However, solids discharge is usually intermittent from
this machine, limiting its application to low concentration slurries. In
the tubular bowl centrifuge there are no internal structures, so it is
possible to run the machine at very high rotational speeds, up to
Fig. 8.4 The hydrocyclone
30000 rpm. Continuous discharge relies upon flushing material out of and flow patterns inside
the machine using displacement by incoming material and these
devices are usually used on liquid/liquid separations or emulsions.
The machines with internal structures have slightly modified forms

of the sigma expression to account for the differences in geometry.
However, the sigma values are still related to the equivalent settling
basin plan area.
tangential:
8.2 Hydrocyclones
Figure 8.4 illustrates a hydrocyclone, including the flow patterns
found within the device. The tangential inlet causes the fluid to
rotate, rather than mechanically rotating the wall, hence these devices
are described as having no moving parts. Of course, a pump or other
prime mover for the suspension is required. The flow pattern within
the hydrocyclone is complex and there are three velocities that need
to be considered. The tangential velocity gives rise to particles subject
to the centrifugal field force and is, therefore, critical to the operation
of the hydrocyclone. Tangential velocities of the liquid, and solids,
may be up to 20 m s−1. The radial velocity is much lower, usually less
than 0.1 m s−1. However, within the device there is a net flow of
liquid towards the centre and a net flow of solids away from it.
radial: Hence, it is important to distinguish between the radial liquid, or
solid, flow. The third velocity is in the axial direction. This velocity
has to be considered carefully because the hydrocyclone has two
outlets, continuously splitting the feed in to two separate streams.
The overflow contains a suspension that is more dilute than the feed
and has a finer particle size distribution. By contrast, the underflow is
a suspension more concentrated than the feed and has a coarser
particle size distribution. Thus, the hydrocyclone acts as both a
thickener (i.e. concentrates a suspension) and a classifier (i.e. selects
particles of a specific size). The axial velocity must take material to
the two outlets and the suspension near to the wall of the
hydrocyclone flows axially to the underflow. The material near to the
centre of the hydrocyclone flows axially to the overflow. Hence, there
is axial flow downwards, and upwards, within the hydrocyclone as
axial: illustrated in Figure 8.5.
A further understanding of these axial flows can be obtained by
considering the hydrocyclone primary and secondary vortices. The
primary vortex spirals down towards the underflow taking the larger
particles with it: these are centrifugally encouraged towards the wall
of the hydrocyclone. However, the geometry of the hydrocyclone
causes flow reversal towards the central axis of the device, giving rise
to the secondary vortex that spirals upwards to the overflow. It is the
finer particles that are caught within the secondary vortex. A vortex
finder is used on the overflow to minimise short circuiting of solids
from the feed directly to the overflow.
Referring to Figure 8.5, the argument can be made that because
there is axial flow upwards into the overflow and downwards into
Fig. 8.5 Velocities within underflow there must be a shear plane within the hydrocyclone
the hydrocyclone where there is no net velocity in the axial direction. In fact, the shear
plane is a surface, or locus, because the device is three dimensional;
hence, this has become known as the Locus of Zero Vertical Velocity
(LZVV). However, it is a misnomer because the hydrocyclone could
operate equally efficiently in any orientation and the term should
really be locus of zero axial velocity. The concept of this LZVV is very
important and it can be used to explain some of the observed
behaviour in the hydrocyclone. Within the device it is possible to set
up a force balance between the centrifugal force and the liquid drag
force. The latter pulls the particles inwards, as liquid must flow
inwards in order to enter the overflow. Hence, particles may adopt an
orbit; where the drag force is balanced by the centrifugal force.
Particles that orbit at a radial distance greater than the LZVV will be
in the primary vortex and will tend to report to the underflow.
Particles orbiting at radial distances less than the LZVV will be in the
secondary vortex and will be carried into the overflow. Thus, the Fig. 8.6 Equilibrium orbit at
particle size that orbits at the LZVV will have no preference for the the LZVV with liquid drag
overflow or underflow; i.e. it will have an equal chance of entering and centrifugal forces
either exit. This is defined as being the cut size (x50) of the balanced
hydrocyclone, see Figure 8.7. Again, the term is misleading as it could
be assumed that no particles bigger than the cut size enters the
overflow; which is very significantly different from the true meaning
of equal chance of entering either flows. For a particle to be radially
stationary on the LZVV the force due to liquid drag inwards must be
balanced by the particle centrifugal field force outwards, Figure 8.6
πx 3
3πµxu r = ( ρ s − ρ ) rω 2
6
18µu r
x 50 = (8.11)
( ρ s − ρ ) Rω 2
where R is the radius of the LZVV. Hence, it is possible to predict the
cut size of the hydrocyclone provided the terms on the RHS of
equation (8.11) are known. This approach is known as the equilibrium
Fig. 8.7 Size distributions
orbit theory. There are three variables that must be deduced before the
of feed, overflow and
cut size can be estimated: the radial liquid velocity, the radial position underflow showing the
of the LZVV and the angular velocity at the LZVV. The remaining cut size. Note that this is
physical constants should be straightforward to obtain. an idealised plot
The radial position of the LZVV is deduced by assuming that the assuming equal volume
LZVV is a shape that is identical to that of the overall hydrocyclone, flow split between the
but at a smaller radius, and that the volumetric flow split of overflow underflow and overflow
rate compared to feed rate is equal to the volume ratio within the – hence the two outlet
hydrocyclone; i.e. curves should equal the
feed curve.
volume inside LZVV volume overflow rate
=
volume inside hydrocyclone volume feed rate
see Problem 8 for a worked example of this.
Once the radial position of the LZVV has been deduced it is
possible to calculate the radial liquid velocity if it is assumed that the
liquid flow entering the overflow is uniformly distributed over the
entire surface of the LZVV. Hence, after calculating the equilibrium

orbit radius (R) the surface area of the LZVV can be deduced from the
equations of surface area of a cylinder and cone, see the Problems for
these. The radial liquid velocity inwards is then the volume flow rate
in the overflow divided by this surface area. The angular velocity at R
is deduced using equations (8.2) and (8.3). Firstly, the tangential
velocity at the inlet to the hydrocyclone is determined by dividing the
feed rate by the cross-sectional area of the feed pipe. This gives the
tangential velocity at the wall of the hydrocyclone, equation (8.3), or
(8.4), can be used to convert this to tangential velocity at the LZVV
Fig. 8.8 Reduced grade
efficiency curve and and equation (8.2) is used to deduce the angular velocity at this point.
ideal classifier showing Experimental observations performed using glass and transparent
a vertical line at the cut plastic hydrocyclones have shown that there exists a mantle region
size. within the hydrocyclone, in which there is little radial flow over the
LZVV. The experimenters injected a pulse of dye into the
hydrocyclone and it remained at the LZVV corresponding to the
region where the hydrocyclone is cylindrical, but not the conical
section; i.e. the dye did not accumulate in the conical section. Hence,
in order to use equation (8.11) the liquid velocity inwards can be
modified by neglecting the surface area of the cylindrical section of
the LZVV; i.e. the overflow rate divided by the surface area of the
conical section of the LZVV gives the liquid velocity corrected for no-
Grade efficiency flow over the mantle region.
In particle classification it Equation (8.11) can be used to deduce the cut size, but it does not
is usual to have a single
provide any information on the proportion of particles finer than the
feed stream and two
outlet streams – one with cut size that will enter the underflow (other than the knowledge that
particles finer than the this will be less than 50%), nor on the proportion of particles larger
other, and it is usual to than the cut size that will enter the overflow. This information is
define the grade provided in the grade efficiency curve, which is illustrated by Figures
efficiency as being the 8.7 and 8.8. The formal definition of grade efficiency (E1) is given
fraction by mass of below
particles reporting to the
fine cut of the classifier. Mass flow in size grade in underflow (kg s -1 )
E1 = (8.12)
Grade efficiency in Mass flow in size grade in feed (kg s -1 )
hydrocyclones is defined
as fraction by mass However, there is a problem with this simple definition of grade
entering the underflow, efficiency: if the feed was simply taken into a box and split into two
i.e. the coarser cut. It is separate outlets, with equal flows, the grade efficiency according to
also sometimes called the equation (8.12) would be 50%, but the system has not achieved any
recovery curve. With the degree of particle classification (sorting by size). To overcome this
term grade efficiency
shortcoming the reduced grade efficiency is used, which is the grade
care is always needed to
determine if the efficiency by equation (8.12) minus the volumetric flow split entering
definition is recovery to the underflow. The volumetric flow split is know as the recovery (Rf).
the fine, or coarse, In practice, using this definition for reduced grade efficiency will
fraction from the result in values approaching zero for small particle sizes, but at larger
classifier. particle sizes the values will be 100% minus the recovery. Whereas
the value at larger particle sizes should be 100%. Hence, a modified
reduced grade efficiency is usually used, with the correct limits of
zero at low particle size and 100% at high sizes
E1 − Rf
Reduced grade efficiency = E 2 = (8.13)
1 − Rf
and a sharpness of separation can be deduced from
x
Sharpness of separation = 25% (8.14)
x 75%
A residence time model for hydrocyclones, similar to that derived for
centrifuges before equation (8.7), provides the following result
x50 2 ( ρ s − ρ ) L∆P
= 3 .5 (8.15)
µρQ Fig. 8.9 Definition of
terms within a filtering
The derivation was first published by Rietema and his work was later centrifuge
extended by Svarovsky. Consideration of equation (8.15) leads to the
conclusion that there should be an optimum design: minimising cut
size and pressure drop and maximising flow rate. Equation (8.15) is
useful as it relates cut size to pressure drop over the hydrocyclone.
8.3 Filtering centrifuges

A schematic illustration of a filtering centrifuge is shown in Figure
8.9. As with sedimenting centrifuges, a major design consideration is Cake removal
the removal of solids in order to permit continuous operation of the An oscillating plate that
industrial equipment. In filtering centrifuges, a more coherent solid pushes the cake out of the
structure is obtained due to the two forms of dewatering that apply: machine in a pusher
centrifugation and filtration. However, mechanical devices are still centrifuge, a knife that
required to remove the cake in a semi-continuous manner. This can periodically enters the
be achieved by the methods described in the box. cake and scrapes it away
In most of the above filtering centrifuge types there is an in a peeler centrifuge and a
identifiable cycle which is illustrated in the following table. rotating filter bag that is
periodically pulled inside
Table 8.1 Centrifuge full cycle out; thus discharging its
contents in the inverting
Function Time (s) Cycle time (%) bag centrifuge, the latter is a
Accelerate 50 to 500 rpm 40 5 popular centrifuge type
Load & filter at 500 rpm 277 32 within the pharmaceutical
Accelerate to 1050 rpm 90 10 industry. It is also possible
Spin dry at 1050 rpm 119 14 to stop a centrifuge and
Wash at 1050 rpm 10 1 manually dig the cake
Spin dry at 1050 rpm 236 27 away.
Slow down to 50 rpm 90 10
Discharge at 50 rpm 15 2
Total cycle time 877 100%

Basket load per cycle of solids 140 kg
Productivity 575 kg h−1
The operations in italics in Table 8.1 are considered in the rest of this
chapter, starting with the modification of the filtration theory already
covered in Chapter 4, to account for its application in a centrifugal

Equipment field and filtering on the inside of a cylindrical surface.
See:
It is possible to combine the basic filtration equations (4.7) and
www.midlandit.co.uk
(4.10) and rearrange to give
/particletechnology
for links to equipment  αcV Q
∆P = µ  + Rm  (8.16)
suppliers with pictures  A A
of the equipment However, when filtering on the inside of a cylinder the effective
described here.
filtering area will be continually decreasing because of the shrinking
of the radius on which the new cake forms. Hence, a consideration of
the two areas contributing towards the equations (4.7) and (4.10),
provides the alternative form to equation (8.16)
 αcV R 
∆P = µ  2 + m Q
 (8.17)
 A Ao 
where Ao is the area before any solids deposition occurs and is
Ao = 2πro h (8.18)
See Figure 8.9 for an illustration of these terms. Integration of
equation (8.17), under constant pressure conditions, provides
∆P
Q= (8.19)
 µαcV µR m 
 + 
 Alm Aav Ao 
where Alm and Aav are the log-mean and average areas respectively.
They can be calculated from
2πh( ro − rc )
Alm = (8.21)
ln(ro / rc )
2πh( ro + rc )
Aav = (8.22)
2
For filtering centrifuges the constant pressure, to be used in equation
(8.19), can be calculated by equation (8.5). For data analysis, equation
(8.19) can be rearranged into the form
t  µαc  V µRm
=   + (8.23)
V  Alm Aav ∆P  2 Ao ∆P
a similar relation to equation (4.19) which was used for constant
pressure cake filtration analysis and illustrated in Figure 4.10.
Equation (8.23) can be used for data analysis provided that the areas
do not change significantly; this is true at the start of the filtration.
Hence, under these conditions, a plot of t/V against V will be a
straight line that can be used to deduce the filter cake specific
resistance from the gradient and the cloth resistance from the
intercept. Under these conditions Ao may need to be used for the log
mean and average areas.
Fig. 8.10 Filtration on the More detailed information on the areas can be obtained by
inside of a cylinder – terms conducting a material balance on the solids inside a batch centrifuge.
used in material balance
Imagine a batch centrifugation has resulted in two products: a filtrate
of volume V and a cake of volume hA. The balance provides:
total volume slurry in the centrifuge was (V+hA) m3

volume of solids at the start is Cf(V+hA) m3
volume solids filtered and in cake is ChA m3
where C is the cake concentration and Cf the feed slurry
concentration of solids, both by volume fraction. Hence, the mass
balance on the solids gives
Cf
ChA = C f (V + hA) which rearranges to hA = V
C − Cf
and, see Figure 8.10, from the geometry the cake volume is
hA = hπ ( ro 2 − rc 2 )
Hence, combining these equations and rearranging gives
1/ 2
 Cf V 
rc =  ro 2 −  (8.24)
 C − C f πh 
Equation (8.24) provides the cake radius as a function of filtrate
volume, which can be used in equations (8.21) and (8.22) to give the
filtering areas. This enables a more detailed analysis of the filtration
data than that described under equation (8.23) and it can be used as
part of the simulation, or modelling, of centrifugal filtration as
described in Figure 8.11. Combining equations (8.19) to (8.22) and
(8.5) for total pressure, noting that c is the dry cake mass per unit
volume of filtrate, but forming inside a cylinder between ro and rc, i.e.
C ( ro − rc ) Aav ρ s
c= (8.25)
V
provides the following simulation equation which is valid for
filtration and for cake washing Fig. 8.11 Centrifugal
filtration modelling as
ρ mω 2 ( ro 2 − rL 2 ) / 2 could be applied within a
Q= (8.26)
 µαCρ s  ro  µRm  computer spreadsheet
 ln  +  N.B. Rm must be finite to
 2πh  rc  2πro h  start solution off, its
effect can be minimised
where C is the cake concentration by volume fraction and ρm is the
by reducing ∆t and
mean suspension density, see equation (6.12). During cake washing,
comparing results as
the cake radius (rc) will remain constant and the mean suspension
Rm → 0
density will be the density of the washing liquid, not including any
solids. In many instances the liquid radius (rL) is also a constant as
the feed is controlled by an overflow weir.
8.4 Washing and dewatering

After filtration the cake will contain pores between the solids with
retained solution present. In many cases, e.g. in some pharmaceutical
formulations, it is not desirable to leave solutes from the initial
reaction present in the final solid product. Hence, fresh solvent is
used to wash the solutes from the cake, or at least to reduce their
concentration to a low value. After washing to remove the solutes the
cake is often dewatered to reduce the amount of retained solvent in the

pores between the solids forming the cake. The term dewatered is
generally used regardless of whether the solvent is water, or
otherwise. As the cake radius remains constant during washing, it is
possible to integrate equation (8.26) directly for the time taken to
wash the cake
µ   ro  Rm 
t= αC ρ  +
2 2  s ln  r  r V w (8.27)
2πhρω 2 ( ro − rL ) / 2   c 
o 
where Vw is the volume of wash water passed in the time t. The

values to use in equations (8.26) and (8.27) for specific resistance and
cake concentration could be obtained from conventional cake
Fig. 8.12 Comparison of filtration tests and empirical relations such as equations (4.15) and
washing performance with (4.16), where cake forming pressure is obtained from equation (8.5).
dewatered and flooded cakes Prior to washing, and as indicated in Table 8.1, it is usual to
dewater the cake slightly. This will reduce the amount of solution
that needs to be displaced from the cake and helps to reduce any
inhomogeneity within the filter cake. Figure 8.12 compares the
performance of cake washing with, and without, a dewatering stage
before washing. The effectiveness of washing is usually judged by
comparing the solute concentration in the filtrate with the value of
solute in the solution. The wash ratio is the volume of wash liquid
divided by the volume of the pores within the filter cake. Hence,
under plug flow conditions, a wash volume of 1 would give a solute
concentration of zero in the filtrate. In filter cakes the flow conditions
Fig. 8.13 Example washing can never be adequately described by plug flow, but a dewatered
performance at 50 and 70% cake helps to reduce the volume of wash water required to obtain a
efficiencies as provided by given residual solute concentration.
equation (8.28) For high removal of solute from the cake it may be necessary to
reslurry wash the cake; i.e. take the dewatered filter cake and form
another well dispersed slurry with it, then filter again. This is because
cake washing on any filter is a mixture of displacement and
diffusional flow of the solute and the latter process is very slow. The
mass fraction of solute remaining in the washed cake (WS) can be
related to the wash ratio (WR) and the washing efficiency (EW) by the
following equation
WR
 E 
WS = 1 − W  (8.28)
 100 
Typical washing efficiencies are between 40 and 80%. An example
wash ratio curve is illustrated in Figure 8.13.
After washing it is common to spin the cake as dry as possible,
before discharge. There are two aspects to consider in this part of the
dewatering process: the equilibrium residual moisture and the kinetic
approach to that equilibrium. In the former case, for a given degree of
force removing liquid from a cake (provided by a centrifugal field in
this case) there will be a given amount of residual moisture retained
by the capillary pressure force. This force is considered further in
Chapter 13. A finite time is required for the cake to drain, hence the
rate of approach to the equilibrium saturation level of the filter cake
is also important. The equilibrium saturation is known as the
irreducible saturation and saturation is usually reported in terms of a
ratio based on the initial value: i.e. after cake formation a fully
saturated cake has a value of 1, but during dewatering the relative
saturation approaches the irreducible saturation if sufficient time is
allowed. This ratio of saturation values is known as the relative
saturation. There are two well-known models for saturation
modelling, one developed by Wakeman and based on particle
Fig. 8.14 Comparison of rotation
properties and another based on the work of Zeitsch, which is based speeds and resulting drainage
on a Boltzman distribution of pore diameters. However, in curves.
application the Zeitsch model applies the data obtained from the Operating data:
filtration stage and will be discussed further here. filter diameter 0.3 m
The relation between filter cake permeability and cake specific solvent density 980 kg m−3
resistance was provided in equation (4.12); permeability is used in solids density 1400 kg m−3
Zeitsch’s dimensionless Drainage (DN) number solvent viscosity 0.001 Pa s
cake concn. 0.3 v/v
kro 2ω 4 ρ 2 ( ro − rc ) 2
DN = (8.29) specific resistance 6x1010 m kg−1
(1 − C )σ 2 cos 2 θ rotational speed 3400 rpm
where σ is the surface tension and θ is the contact angle of the liquid surface tension 0.07 N m−1
contact angle 10 Degrees
on the solid. Zeitsch deduced the irreducible saturation ( S ∞ ) to be
cake thickness 50 mm
 −1  πD N   1 
S ∞ = 1 − exp + 1 − erf   (8.30)
D  2D   D 
 N N
  N 
The error function (erf) is built in to most modern spreadsheets,
hence evaluation of operating parameters, such as rotational speed,
by equations (8.29) and (8.30) on a spreadsheet is straightforward.
For a kinetic evaluation of drainage, Zeitsch defined a drainage
rate constant (φ), which has the SI units of s−1, as
σ 2 cos 2 θ
φ= (8.31)
2 µ ( ro − rc ) 3 ρroω 2
and, for convenience, it is possible to combine the drainage rate
constant and drainage number into a single dimensionless term (B)
1
B= + φt (8.32)
DN
where t is the time during drainage. According to the Zeitsch model,
the relative saturation with respect to time is
S t = S∞ +
1 1
 exp(− B ) −
[ ]
1 − erf ( B ) π 
 (8.33)
2
DN B  B 2 B  Fig. 8.15 Effect of cake
thickness on saturation at a
Figure 8.14 provides an illustration of the above theory used to rotational speed of 3400 rpm
compare rotational speeds on relative saturation and drainage times. – all other variables are as
On Figure 8.14, it is noticeable that there is little saturation given below Figure 8.14
reduction by spinning at 1400 rpm; the equilibrium value (irreducible
saturation) is 96%. At 3400 rpm the irreducible saturation is 24%.

Hence, the beneficial effect of dewatering at higher rotational speeds
is obvious. However, a relatively slow filtration rotational speed may
be required to reduce the tendency for particles to penetrate the filter
cloth. If this occurs the medium resistance term will increase and the
cloth may even blind. Thus, gentle filtration conditions are often
employed during the start of any filtration, building up to more
severe dewatering conditions as time progresses. The data on Figure
8.14 demonstrates that there is a threshold pressure that needs to be
overcome before significant dewatering occurs. The data on Figure
8.15 shows that dewatering is favoured by thicker cakes rather than
thin ones. This is evident from a consideration of equations (8.29) and
(8.30): the higher the drainage number the greater the dewatering will
be. The rotational velocity has a significant effect on the dewatering
as it is raised to the power 4 in the expression and the cake thickness
(ro-rc) is raised to the power 2. Clearly, the benefit from increasing the
cake thickness, on the reduction in cake moisture, becomes less
significant at higher depths. Also, productivity calculations often
suggest that thin cakes provide greatest throughputs; so, it is likely
that there is an optimum cake thickness for any operation and that it
is a balance between filtration, washing and dewatering
requirements. The analysis provided by the above equations can be
used to deduce that optimum cake height and, therefore, optimum
throughput provided the numerical input variables are known, or
can be measured.
8.5 Summary
A centrifugal field force can be used to speed up sedimentation and
filtration. A sedimenting centrifuge, or hydrocyclone, may be used
instead of gravity settlement and a filtering centrifuge enhances the
rate at which liquid passes through the filter cake and cloth. In
practice, in filtering centrifuges centrifugal sedimentation should be
expected as well as filtration. The models of gravity sedimentation
and pressure filtration were adapted to a centrifugal field force in this
chapter.
8.6 Problems
1.
i). The equation for the rate at which a particle will settle in a
gravitational field, neglecting the acceleration of the particle, is Ut=...
x3 (ρs − ρ )g x 2 (ρs − ρ )g x 2 (ρs − ρ )g
a: U t = b: U t = c: U t =
18µ 9µ 18µ
ii). The equation for the rate at which a particle will move in a
centrifugal field, neglecting the acceleration of the particle, is dr /dt =...
d r x 2 ( ρ s − ρ ) rω 2 d r x 3 ( ρ s − ρ ) rω 2 dr x 2 ( ρ s − ρ ) g
a: = b: = c: =
dt 18µ dt 18µ dt 18µ
iii). Why is the rate of motion radially in a centrifugal field not a

constant, unlike settling under gravity?
iv). The diagram on the right illustrates a homogeneous oil emulsion

contained in a circular channel rotating at 1000 rpm. The diameter of
the outer circle is 30 cm, and the distance between the walls, i.e. the
channel width is 2.5 cm. The angular velocity of the channel is (s−1):
a: 1000 b: 6280 c: 105 d: 3140
v). If the oil is less dense than the surrounding water an oil droplet Questions 2 and 5
will travel inwards on the application of a centrifugal field force, just Consider the axial and
as oil floats in a gravitational field. In the integration of the radial flow in a tube
centrifugal rate expression with respect to time and radial distance, centrifuge. If the start
the upper or top limit of the integration of radial position is (cm): radius is, in fact, the
a: 30 b: 12.5 c: 15 d: 27.5 inner diameter of the
centrifuge, i.e. between r
L
vi). The oil and water densities are 800 and 1000 kg m−3, respectively and the central axis there
and the viscosity of water is 0.001 Pa s, the time taken (neglecting is only air, and we
acceleration) before a sample of emulsion withdrawn at the inner consider a critical
radius contains no particles bigger than 10 µm in diameter is (s): particle trajectory we can
a: 15 b: 2290 c: -15 d: 150 derive the following
equations
Q
2. In a continuous tube type centrifuge 5.4 m3 min−1 of an aqueous ∑ PROCESS =
suspension is being processed and all the particles of diameter 10 µm Ut
or more are being removed. The solid and liquid specific gravities where Ut is the terminal
are 2.8 and 1.0 respectively, and the liquid viscosity is 10−3 Pa s. settling velocity of the
critical particle under
i). The volume flow rate was (m3 s−1): consideration. This is the
a: 0.0015 b: 0.09 c: 324 d: 5.4 area of a settling basin
ii). Stokes' settling velocity of the 10 µm particle was (m s−1): that would perform the
a: 9.8x10−6 b: 9.8x10−7 c: 1.5x10−4 d: 9.8x10−5 same duty under the
iii). The particle Reynolds number was: given flow rate. The
theoretical capacity of
a: 2.7x10−3 b: 275 c: 9.8x10−5 d: 9.8x10−4
the centrifuge is
2
iv). The sigma process value was (m ); ∑ MACHINE = ...
a: 920 b: 150 c: 9200 d: 590
π ( ro 2 − rL 2 ) Lω 2
…
3. The machine used in the above question had a length and g ln(ro / rL )
diameter of 1.5 and 0.75 m respectively. The pool depth (i.e. ro - rL)
was 0.1 m, and the operating speed was 1800 rpm.
i). The rotational speed was (s−1):
a: 190 b: 380 c: 11000 d: 94
3
ii). The volume of the centrifuge pool ONLY was (m ):
a: 0.31 b: 0.20 c: 0.62 d: 0.05
iii). The 'sigma machine' value was (m2):
a: 19 b: 840 c: 3600 d: 2200
iv). Compare the two sigmas for the efficiency of the machine (%):
a: 26 b: 4.3 c: 110 d: 30
4. The sigma values of four machines labelled a, b, c and d are 200,

400, 600 and 1200 m2, respectively. Assuming they cost the same,
which machine would you recommend?
5.
i). Derive an expression for the volume fraction of a centrifuge, based
on an imaginary start radius that exists between rs and ro (i.e. p=).
ii). Using rs instead of rL and equating the two sigma terms (see box
above), derive an expression for rs.
iii). Combine the answers to parts (i) and (ii) to give an expression for
the proportion of particles removed as a function of flow and
material properties, i.e. p or the grade efficiency=...
The equilibrium orbit (R) of
8. The feed and overflow rates of the hydrocyclone illustrated are 36
a particle at the locus of
zero vertical velocity is
and 23 litres per minute, the liquid viscosity is 0.0015 Pa s, the solid
given by and liquid densities are 2000 and 1000 kg m−3, respectively. These
x 2 ( ρ s − ρ ) Vi 2 questions determine the separation size for the hydrocyclone using
R= equilibrium orbit theory.
18µ U
i). If Ro is the internal radius of the hydrocyclone, the fractional
where Vi is the tangential
volume inside the locus of zero vertical velocity is:
velocity of the liquid in the
hydrocyclone at the a: (R/Ro) 3 b: (R/Ro) 2 c: πlR 2 d: lR2/lRo2
equilibrium orbit position ii). By assuming that the fractional volume inside the locus of zero
and U is the inward radial vertical velocity is equal to the proportion of the feed flow going into
velocity of the liquid. The the overflow (O/F), the equilibrium radius was (m):
principle of the
a: 0.0052 b: 0.016 c: 0.032 d: 0.0080
conservation of angular
momentum gives
iii). By considering the locus of zero vertical velocity as a surface over
which the flow entering the O/F is uniformly distributed the liquid
Vi ri = constant
and the surface area and
velocity over the LZVV was (m s−1):
volume of a cone are a: 0.025 b: 0.057 c: 0.0186 d: 0.069
πrl πr 2 l / 3 iv). The inlet velocity, hence tangential velocity at Ro, was (m s−1):
where l is height or length. a: 5.97 b: 11.9 c: 7.32 d: 3.0
v). Using the principle of the conservation of angular momentum,
and assuming that all the inlet flow occurs at Ro, the tangential
velocity at the equilibrium orbit radius was (m s−1);
a: 14.9 b: 45.8 c: 9.54 d: 7.47
vi). The separation or cut size of the hydrocyclone was (µm):
a: 2.0 b: 6.0 c: 9.4 d: 12
vii). The particle Reynolds number at the LZVV was:
a: 0.074 b: 0.12 c: 0.15 d: 0.30
9. Reitema's residence time model for an optimum design is given

x 2 ( ρ s − ρ ) L ∆P left, where L is total length. Using your answer for (8.vi) for the
= 3 .5
µρQ separation size, the hydrocyclone pressure drop was (bar):
a: 0.84 b: 3.4 c: 5.8 d: 7.9
9 Conveying
This chapter covers the transportation of particles by fluids. It
considers hydraulic, pneumatic and mechanical conveying means.
The former uses liquids the second gases and an example of the latter
is a conveyer belt. A basic course in fluid mechanics will equip the
reader with sufficient knowledge to calculate the pressure drop for a
given flow rate of a uniform, or homogeneous, fluid. When
considering particles suspended in flow, however, it is possible for
the particle to slip in the fluid; i.e. the fluid travels faster than the
solid particle. It is also possible for two distinct regions to be visible: a
dilute, or emulsion, region with few particles and a dense region
below it. This is illustrated in Figure 9.1. In the case of homogeneous
flow, where there is no slip and the particles merely add to the
effective liquid viscosity and density the methods of analyses covered
already in Sections 6.6 and 6.7 are appropriate.
An example of a positive pressure pneumatic conveying flowsheet Fig. 9.1 Heterogeneous flow
is provided in Figure 9.2. The process is straightforward and requires showing only the particles
a blower to provide the motive force, feed silo for solids, a means of within the dense region
controllably passing solids into the conveying line and a destination
vessel with particle-gas separation equipment. If the blower was to be
positioned after the gas cleaning equipment and the particles sucked
through the system it would be negative pressure pneumatic
conveying. Systems with mixed positive and negative pressures are
also possible, such as might occur when there are two conveying
lines: one operating under suction on the blower inlet, the other from
the blower outlet.
9.1 Heterogeneous flow in liquids

Most of the work in this subject was due to the requirement to move
coal, and other minerals, over significant distances as slurries. A
resurgence of interest occurred when clean-up of nuclear solids
deposited in ponds started. There are two key mathematical analyses
required for the flow of solids in a liquid filled pipe: identification of
the flow velocity which is insufficient to prevent solids depositing on
the bottom of the pipe and the calculation of the pressure drop, or
gradient, during the heterogeneous flow. Homogeneous flow is dealt
with elsewhere. If the slurry velocity is reduced from a high value
which entrains all the solids a point will be reached when solids will
be observed to become stationary on the pipe surface. This is the limit
deposit-velocity (uLDV) of the heterogeneous suspension. In order to
hydraulically convey material efficiently this velocity should be
exceeded. An analysis based on boundary layer theory suggests that
a correlation of the following form is appropriate
92 Conveying
0. 8 K2
ρ   x85 
u LDV = K1 gx85  s − 1   (9.1)
 ρ   D 
where x85 is the particle diameter below which there is 85% by mass
of the distribution, K1 and K2 are constants. The value of K1 is 650,
when viscous forces predominate, and 0.19 in the turbulent region.
The value of the constant K2 is 0, when viscous forces predominate,
and -2 in the turbulent region.
During heterogeneous flow there will be two identifiable regions,
as illustrated on Figure 9.1, a homogeneous region flowing above a
heterogeneous one. It is assumed that the overall pressure drop, or
gradient, for the two phase flow is the sum of the pressure drops in
both of these regions. Using the empirical correlation proposed by
Durand (Wasp, E.J., Kenny, J.P. and Gandhi, R.L., 1979, Solid-liquid
flow slurry pipeline transportation, Gulf Publishing, Houston, Texas.)
  gD ( ρ − ρ )  
1.5

∆P = ∆Pw 1 + 85C  s   (9.2)
  u2ρ C  
  d  

where ∆Pw is the calculated pressure drop for the homogeneous
phase, u is the mean suspension velocity and Cd is the particle drag
coefficient. The homogeneous phase pressure drop may be calculated
using Newtonian flow equations with equation (6.14) for the
viscosity. The drag coefficient comes from equation (5.13), with the
terminal settling velocity in that equation provided by Stokes’ law or
The Heywood Tables.
9.2 Dilute phase pneumatic conveying

Particle slip occurs when solids are conveyed by gases but, at low
solid concentration, there will be a uniform density of particles
throughout the conveying pipe. This is dilute phase pneumatic
conveying and is another form of homogeneous flow. At the saltation
point a significantly greater concentration of solids is found at the
base of the pipe. These solids are not stationary, they will still be
dragged through the pipe by the gas, but they will have an additional
frictional energy loss due to close contact with the pipe wall. Thus,
they will have a slower velocity than the particles contained within
the emulsion phase above the bottom of the pipe. A phase diagram
illustrating flow conditions in dilute, and other, pneumatic conveying
Fig. 9.2 Positive pressure regions is illustrated in Figure 9.3.
pneumatic conveying system When designing a dilute phase pneumatic conveying system a key
parameter to assess is the gas velocity at which saltation takes place.
Operating at gas velocities above this will be in the dilute phase
region. An empirical correlation for saltation velocity is due to Rizk
1 /( b +1)
 4 M s 10 a g b / 2 D (b / 2− 2) 
u salt =  (9.3)
 ρπ 
where
a = 1440 x + 1.96 (9.4)
b = 1100 x + 2.5 (9.5)
where Ms is the mass flow rate and SI units must be used throughout.
The saltation velocity is a superficial velocity and the required gas
velocity for use in the friction correlations is the interstitial. Also, to
ensure that the flow is in the dilute phase region an excess gas
velocity of 50% may be employed. Thus, with 50% excess the gas
velocity is
u salt
u g = 1 .5 (9.6)
ε
where ε is the porosity of the gas phase and is, in general, close to
Fig. 9.3 Types of flow found
unity. Having established the superficial gas velocity (Uo) the solids in pneumatic conveying
velocity (us) can be calculated using an equation for slip
u s = U o (1 − 0.0638 x 0.3 ρ s 0.5 ) (9.7)
where SI units must again be used for dimensional consistency.
Clearly, the greater the particle size and solid density the greater the
degree of slippage between the gas and particle velocities. The
interstitial gas velocity is used to calculate the friction of the gas in
the pipe, comes from standard methods for single phase pressure
drop calculations. It is assumed that the overall pneumatic conveying
pressure drop is the sum of all the pressure drops due to the following,
which also indicates where the equation originates:
acceleration of the gas (from Bernoulli’s equation)
1
ρεu g (9.8)
2
acceleration of the solids (from Bernoulli)
1
ρ s (1 − ε )u s (9.9)
2
friction of the gas on the pipe wall (from friction factor)
L
2 f g ρu g 2 ε (9.10)
D
friction of the solids on the pipe wall (from friction factor)
L
2 f s ρ sus 2 (1 − ε )
D
which, using continuity, is the same as
L
2 f s Gs u s (9.11)
D
where Gs is the mass flux of solids,
the static head of the gas is (from Bernoulli)
ρεgL sin θ (9.12)
94 Conveying
static head of the solids (from Bernoulli)

ρ s (1 − ε ) gL sin θ (9.13)
where θ is the angle from the horizontal, thus sin θ = 0 for horizontal
conveying and sin θ = 1 for vertical conveying. It is usually assumed
that the pressure drop due to a 90o bend is 7.5 times that of a vertical
pipe, i.e. an equivalent length. The solid particle friction factor for use
in equation (9.11) depends upon the pipe alignment. For horizontal
conveying it has a reminiscent form
2
6 ρ D  u g − us 
fs = C d  
 (9.14)
8 ρs x  us 
but for vertical conveying it must be modified to
0.057 D g
fs = (9.15)
2 us D
The practical application of these equations is shown in Problem 1,
which is a worked solution to be completed for a dilute phase
pneumatic conveying system design.
9.3 Dense phase pneumatic conveying

When the gas velocity is lower than that required to produce
saltating flow dense phase conveying exists. It is possible to observe
dunes, similar to sand dunes, flowing through a conveying pipe. At
very high solids loadings, or low velocities, complete plugs of solids
may form in the pipe and still be conveyed. However, there is a
danger of the pipe completely filling up with solids that become
stationary. Hence, the design of high concentration dense phase
pneumatic conveying can be critical and it is well known to be an
exceedingly difficult subject to model mathematically. All supplying
companies require careful test work before specifying such systems.
The advantages of dense phase conveying over dilute phase
include: considerably lower product degradation from particle-wall
collisions and much lower energy costs because the air velocities are
much less than during dilute phase flow. However, the pressure in a
dense phase system operating in plug flow is often higher than a
dilute phase system. So, compared to a dilute phase system, the
blower might need to produce lower gas velocities, but provide
higher pressures. The lower velocities found in dense phase systems
lead to lower maintenance requirements for such systems.
9.4 Other conveying equipment

Considering the transportation of mainly dry materials, one of the
main advantages of pneumatic conveying is the complete enclosure
of the product. Hence, the material is protected from the surrounding
environment and vice-versa. Also, pipes can be easily altered to
Fig. 9.4 Screw conveyor change the flow route, few moving parts mean low maintenance,
easy control and the ability to handle a range of products. However,

the main disadvantages are: high power, limited distance and limited
throughput. When transporting materials in thousands of tonnes per
hour a pneumatic conveying system would be prohibitively
expensive.
Alternative popular mechanical conveying methods include: belt
conveyors, chain conveyors, screw conveyors, bucket elevators and
vibratory conveyors. In most cases the equipment are true plug flow
devices and design is straightforward given the mass of material to
be conveyed and the manufacturer’s data sheets on motor power and
capacity. However, screw and vibratory conveyors can have a small
amount of axial mixing, but this is not significant for throughput Fig. 9.5 Vibratory conveyor
calculations. The reader will undoubtedly be familiar with a
conveyor belt. Illustrations of a screw conveyor and vibratory
conveyor are provided in Figures 9.4 and 9.5, respectively. Drives for
vibratory conveyors can be mechanical, eccentric wheels,
electromagnetic and hydraulic. These drives give varying degrees of
amplitude and frequency. Conveying speed is generally less than 0.4
m s−1 with a vibratory conveyor. The mass flow rate (Ms) for a screw
conveyor can be calculated from the following formula
M s = A' u s ρ b E f (9.16)
where A’ is the cross-sectional area available for flow (i.e. not

occupied by the screw and shaft) and Ef is the filling efficiency. This
efficiency does not relate to the introduction of solids into the
conveyor, it is the fill level within the device; i.e. the area occupied by
the bed of solids compared to the cross-sectional area. Thus, it takes
in to account any space above the bed of solids. The solids velocity
inside a screw conveyor (us) is the product of the blade pitch and the
revolutions per second, assuming no slippage of particles.
Bucket elevators are designed to raise material and the loading
and discharging of such elevators is illustrated in Figure 9.6. Chain
conveyors, sometimes called en-masse conveyors, provide an
arrangements of flights suitably spaced along a chain and the product
to be conveyed is dragged by the flights through a trough that is
usually completely enclosed. Product is normally introduced into the
conveyor through a hole in the top of the trough and discharged
through a hole at the bottom. The slip between the particles and the
flights determines the efficiency of the device; ideally this should be
zero. Chain velocity is usually less than 0.5 m s−1 and widths of up to
0.5 m are possible.
Particle properties that are important for all of these conveying
methods include: moisture and stickiness of the particles, hazard to Fig. 9.6 Bucket elevator
the environment (and vice-versa), resistance to particle breakage and loading and discharge
product temperature. Process requirements that need to be
considered include: required angle of inclination, mass throughput,
conveying length, automation, sampling, loading and discharge.
Careful consideration of the conveyor type and the above list should
96 Conveying
result in the identification of the most suitable conveying methods for

a given system. For example, particularly sticky material might not
discharge from a belt conveyor, but a chain or vibratory conveyor
could be appropriate. For high throughputs of low value minerals a
belt conveyor is appropriate and, if contamination from the
environment is not important, including rainwater, the conveyor
need not be covered. A search of the Internet will provide many
conveyor suppliers and example throughputs for their machinery.
9.5 Summary
The transportation of particulate solids is an important consideration
of most processes involving solids. Often equipment selection is
based on process considerations such as the need to avoid
contamination from the environment of the valuable solid such as a
pharmaceutical product. If required throughput is high, normally
associated with low value products such as minerals, then a
mechanical device is appropriate such as a belt conveyor. If the
material is sticky, due to moisture or temperature softening, then a
chain conveyor may be required. For most chemical and food
processes pneumatic conveying is preferred because it is enclosed
and controllable. Using a rotary valve to meter the solid input into
the system the solid mass flow rate can be easily relayed to the
control room, and changed by it, if the valve is controlled by an
inverter.
9.6 Problems
Design a positive pressure dilute phase pneumatic conveying system
to transport 800 kg h−1 of sand of density 2500 kg m−3 and mean
particle diameter 80 µm between two points in a plant separated by
30 and 15 m in the vertical and horizontal directions, respectively.
Use ambient air ( µ = 1.84x10 −5 Pa s; ρ = 1.2 kg m -3 ) and assume that
eight 90o bends are required in the circuit. Start your worksheet
solution based on a pipe with an internal diameter (D) of 40 mm, then
try the pressure drop using pipe diameters of 50, 63 and 78 mm.
Which one would you recommend? Provide answers to the following
calculation stages.
1. The saltation velocity (usalt) as given by the Rizk correlation.
2. The design superficial gas velocity is U o = 1.5u salt .
3. The solid velocity (us) is u s = U o (1 − 0.0638 x 0.3 ρ s 0.5 ) - this

equation allows for slip.
4. The solid flux (Gs) is Gs = M s / A .

Gs
5. The porosity (ε) is ε = 1 − .
ρ s us
6. The interstitial gas velocity is u g = U o / ε .
7. The slip velocity is u g − u s .
(u g − u s ) ρx
8. The particle Reynolds number is Re' = .
µ
9. The drag coefficient is C d =

12
Re'
( )
1 + 0.15 Re' 0.687 , valid for
0.2<Re'<800.
10. The flow Reynolds number is.
11. The gas friction factor is f g = 2(0.0396 Re −0.25 ) ...Blasius equation.
12. The pressure drop due to acceleration of the gas is = 0.5( ρεu g 2 ) .
13. The pressure drop due to acceleration of the solids is

= 0.5( ρ s (1 − ε )u s 2 ) .
Considering the horizontal lengths (LH):

2
6 ρ D  u g − us 
14. The horizontal solid friction factor is f s = C d   .

8 ρs x  us 
15. The pressure drop due to gas friction is = 2 f g ρεu g 2 LH / D .
16. The pressure drop due to solid friction is = 2 f s ρ s (1 − ε )u s 2 LH / D .
17. The total horizontal pressure drop is.
Considering the vertical lengths (LV):

0.057 D g
18. The vertical solid friction factor is f s ' = .
2 us D
98 Conveying
19. The vertical pressure drop due to solid friction is

= 2 f s ' ρ s (1 − ε )u s 2 LV / D .
20. The vertical pressure drop due to gas friction is

2
= 2 f g ρεu g LV / D .
21. The pressure drop due to the solid's static head is

ρ s (1 − ε ) gLV sin θ .
22. The pressure drop due to the gas static head is ρεgLV sin θ .
23. The total vertical pressure drop is.
24. The pressure drop due to ONE bend is

= 7.5(total vertical pressure over distance).
So, the total pressure drop due to all the bends is.
25. The total pressure drop is.
26. After setting the above solution up on a spreadsheet, try internal

pipe diameters of 50, 63 and 78 mm.
27. Which pipe diameter would you recommend?
28. Now calculate the gas flow rate in m3 s−1 for each pipe.
29. Which pipe diameter would you recommend now?

10 Powder flow and storage
Pumping fluids is simple: you need a pump and some pipework. The
To flow or stick?
higher the viscosity, the higher the pressure drop and, therefore,
Property Free Difficult
pump and energy costs. Much is known about how to characterise flowing flowing
and specify a fluid system. What is the equivalent for dry particles? Size >400 µm <100 µm
Also, tanks can be used to store liquids, but how should we store Range narrow wide
powders to ensure that they can be reliably used within our process? Shape spheres needles
This is a subject that has received considerable attention over many Moisture not too high &
centuries and is still a long way from a complete understanding. This low* none
chapter considers simple characterisation based on solid properties Internal
and a one-dimensional particle mechanics analysis. A more thorough friction low high
description is possible by Discrete Element Analysis, see Chapter 7.
* a certain amount of
It would be a mistake to assume that the problem is just one of moisture may help to
ensuring that the powder flows in a hopper, or down a chute. There lubricate the flow and to
are many recorded instances of process difficulties due to powders prevent any electrostatic
suddenly flowing too easily! If a hopper is discharging into a process attraction between particles
and suddenly the powder surges out, it is likely that it will overflow from stopping the flow.
the process vessels and cause disruption: this is called a powder flood.
An example of a powder flood in nature is an avalanche and it is, of
course, potentially very dangerous. Process operators can be killed in
a powder flood, so these must be avoided at all costs. Powder flow in
a controlled and predictable fashion is desired. Floods are usually
associated with aerated powders, in which gas is mixed with the
powder and it behaves in a fluidised fashion, see Chapter 7.
Fig. 10.1 Angle of repose
and its possible relevance
10.1 Powder properties to hopper design
An understanding of particle behaviour starts with a consideration of
particle properties and some basic techniques to measure them. One
of the simplest measurements is the angle of repose, which is
illustrated in Figure 10.1, and is often assumed to be the angle that
the hopper needs to exceed in order to assure powder flow. This may
be acceptable for free flowing powders, and the angle is typically 30o, Fig. 10.2 Alternative measures
but in most cases this ignores the tendency for particles to form a
cohesive structure depending upon how they have been treated.
Pouring the powder into an upside down funnel and then carefully
removing the funnel to leave the heap in place can be used to
measure the angle. Alternative techniques include measuring the
angle of slide and the angle of rotation, as illustrated in Figure 10.2.
The particle size distribution has a complex effect on the angle of
repose and a graph of the angle plotted, against the percentage of
fines present, usually shows a minimum, see Figure 10.3. The angle of Fig. 10.3 Influence of
repose is a property of a powder that does not exist in a liquid and it fines on angle of repose
is not a very consistent measurement for a powder because most
powders exhibit some degree of cohesiveness.
The bulk density is the combined density of the powder and the
void space (i.e. porosity see Figure 3.1). Hence the bulk density is the
same as the mean suspension density, equation (6.12), but as the fluid
is air its contribution to the mean density is ignored. Thus
Fig 10.4 Two planes needing

ρ b = (1 − ε ) ρ s (10.1)
to dilate (open out) before However, the bulk density is a function of porosity, i.e. how the
moving
powder has been treated. To standardise this the tap density is used
(BS 3483). The powder (and voids) volume is measured in a
Porosity is isotropic: i.e. measuring cylinder type vessel after 0 to 800 taps and the powder
the same in all directions, weighed. This provides a more consistent density than the bulk.
but the assumption here Another property that powders posses, but liquids do not, is that
is that the overall porosity of dilatancy. This arises from particles resting on each other such that
increase is due to a shear plane of particles must rise vertically before it is possible for
expansion in one plane the plane to move horizontally, as illustrated in Figure 10.4. As the
(i.e. height).
planes move apart from each other the porosity increases. The
porosity at which the powder can shear is known as the critical
porosity. Dilatancy is important for powders flowing in chutes; in
Figure 10.5 a box is provided at the point where the powder changes
direction, to allow for the powder to dilate, or expand. Without the
expansion box the powder may be prevented from dilating and,
therefore, cease to flow.
Powders have the ability to sustain shear forces better than fluids.
Thus, it is possible to walk on a bed of powder, such as a sandy
beach. The weight of an object on the powder is transmitted through
a network of contacts within the powder compact to the underlying
base, or to the walls of a container. Hence, Archimedes’ buoyancy
Fig. 10.5 Expansion box
principle does not occur: the 'up-thrust' experienced by a body
in flow chute
submerged, or partially submerged, is not equal to the weight of the
material displaced. The up-thrust may be equal to the entire weight
of the body - just like a solid surface. Unlike fluids, there isn't a linear
increase in pressure with depth of particles. In fact, the pressure
stabilises after a short distance and the rate of discharge from a
hopper will, therefore, be substantially constant. The rate of discharge
(Mp) for free flowing powders has been correlated using the following
empirical equation
π
Mp = ρ b gB 5 2 tan θ H (10.2)
4
where B is the opening diameter and θH is the hopper half angle; i.e.
the angle from the vertical, see Figure 10.6. Note that this equation
does not include powder height. However, powders may form a
Fig. 10.6 Hopper half angle stable arch over the opening and block the powder flow, see later.
Whenever powders flow there is an opportunity for the powder to
segregate by, primarily, size and density difference (and other
secondary factors such as rotational inertia). This will occur in heaps,
hoppers, mixers, conveyors, etc. By contrast, miscible fluids do not
un-mix like this.
10.2 Flow patterns and stress in a hopper and silo

A hopper is the conical, or converging, section of a powder storage
vessel; the bin is the parallel sided section, usually cylindrical or
rectangular, and the word silo is used to cover the entire vessel.
However, these terms are often used interchangeably. There are two
main flow types that describe how powder discharges from a silo and
these are illustrated in Figure 10.7. In mass flow the flow pattern is
often described as: first in, first out and in core, or funnel, flow the
pattern is last in, first out. In true mass flow the powder at the edges
of the hopper has to accelerate and shear over that towards the
centre: as it has a longer distance to travel to the discharge hole. Fig. 10.7 Flow patterns
Dilatancy is required at this point and the stress on the hopper is during hopper discharge
greatest here.
A hopper may acceptably operate under funnel or core flow
conditions so long as piping, or rat holing doesn't occur. These two
terms refer to a silo in which the storage capacity consists
Table 10.1 Flow patterns in a
substantially of stationary powder with just a hole within the silo
silo – italicised text indicates
taking newly deposited material from the top straight to the bottom preferred behaviour
and discharge. Hence, there is no net storage capacity within the silo.
Table 10.1 compares the advantages and disadvantages of the two
types of flow.
Powder stresses inside a hopper may be analysed by Janssen's
method of differential slices. Consider a slice dz at a height z, the
downward force (vertical pressure times applied area) is
πD 2 πD 2
Pv upward force is ( Pv + δPv )
4 4
The resultant solids stress (i.e. equivalent to pressure
in a fluid and equal to force over area) is
πD 2
δPv (10.3)
4
Another force comes from the wall. If the horizontal stress at the wall
is Ph and the coefficient of friction is µw, then the wall support force is
µ w ( Ph πDδz ) (10.4)
The weight (i.e. force) of solids in the differential slice is
πD 2
δzρ b g (10.5)
4
Combining the forces: upthrust + wall friction = weight
πD 2 πD 2
δPv + µ w Ph πDδz = δ zρ b g
4 4
giving
dPv 4 µ w Ph
+ = ρb g (10.6)
dz D
The problem is in relating the horizontal with the vertical stresses, or
pressures. If the material in the hopper is a liquid these two pressures
are the same (Pascal's principle). However, consider a stack of coins,
one on top of another, in air. In this case the horizontal pressure is

zero – there is no force accelerating the coins sideways, nor any
reaction force needed to prevent this. Now, as powders have
properties between that of fluids and continuous solid bodies (such
Janssen’s kJ factor
as a stack of pennies), it would be expected that we could write
Ph = k J Pv Ph = k J Pv
where kJ = 0 for coins, see box. This was Janssen's assumption. Substituting this simple
where kJ = 1 for liquids, relation into equation (10.6) and integrating (using integrating
and factors) results in the following equation
0 < kJ < 1
ρ b gD
for powders. Pv = [1 − exp(−4µ w k J z / D )] + Pvo exp(−4µ w k J z / D) (10.7)
4µ w k J
where Pvo is the pressure at z=0, called the surcharge or uniform stress
applied at the top of the powder. For Pvo=0 and at small values of z
ρ b gD 4 µ w k J
Pv = z as exp(-Az) ≈ 1 - Az for low z (10.8)
4µ w k J D
Thus, Pv = ρ b gz – a similar result to that of liquids but only for small
values of z. At large values of z the exponential term disappears,
hence
ρ gD
Pv = b (10.9)
4µ w k J
i.e. pressure asymptotes to the above uniform value. The results from
these equations are illustrated in Figure 10.8.
Janssen's work was important because it showed that stress is not
transmitted in a similar way to hydraulic head, and wall friction has a
very significant influence on the internal powder stresses. However, the
assumption of a constant coefficient linking the vertical and
Fig. 10.8 Stresses inside horizontal stresses has no theoretical justification. Also, arches can be
a silo formed, suggesting that the surface of interest is not planar and that
the stress in a plane is not uniform. Nevertheless, it provides a useful
semi-theoretical analysis of stress inside a hopper. Practical
measurements have supported the above analysis: powder stress
does build up to a constant value within a parallel sided bin. At the
position of the start of the hopper (converging section) the stress
rapidly increases, see the section on dilatancy, followed by
diminishing values as the hopper diameter then reduces to a value of
zero at the outlet. However, measurements show that the powder
retains some extra stress over what is expected in the hopper section
– this is the memory effect included on Figure 10.8.
10.3 Hopper opening and angle

The correct design, or operation, of a hopper to ensure consistent
(mass flow) discharge is based upon two factors: providing a steep
enough hopper angle and ensuring that the discharge opening is wide
enough. Laboratory tests are performed under conditions of stress
and consolidation that are similar to that expected in the hopper.
Values of the required stress to break a stable arch are deduced and
the hopper is designed to provide these conditions with this powder.
This is illustrated in Figure 10.9. Although most of the following text
considers hopper design, the principles are more generally applicable
to powder flow and characterisation than simply hopper design: i.e.
the Powder Flow Function (PFF) or sometimes called the Material Flow
Function, characterises the ease, or otherwise, of powder transport
and storage.
During the discharge of a mass flow silo there is a possibility that
the powder flow may stop, or become intermittent, due to the
formation of a stable arch within the hopper. The reasoning is that the
particles interlock to form a stable arch over the opening much like
bricks in an arched bridge. In practice, sufficient force should be
provided by the system, if correctly designed, to break this bridge if it
forms. Firstly, a force balance can be used to determine the
magnitude of forces present in this powder bridge. Consider a
hopper with an opening of diameter B and a slice of powder of depth Fig. 10.9 Stable arch formed
∆h. If an arch forms there will be air on one side and powder on the above the hopper outlet
other. The powder within the arch will have a yield stress; given
sufficient stress above the arch it will break. This value of stress is
called the unconfined yield stress (fc): unconfined because the powder is
open to air on one side. An illustrative balance (taking moments from
the wall) on the plane forming the arch provides the following result,
see Figure 10.9
B
f c AreaL = W (10.10)
2
where W is the powder weight, B can be taken to be the hopper
opening diameter and L is the linear distance that the arch can be said
to act over. The force from the weight of powder is
W = Area∆hρ b g (Newtons) (10.11)
Combining these two equations gives the following result
∆h ρ b gB
f c = ρ b gB = (10.12)
2 L H (θ )
where L, and hence H(θ), is a function of the geometry of the opening.
Thus the minimum hopper opening diameter needs to be
f c H (θ )
B= (10.13)
ρb g
The next stage is to identify the unconfined yield stress for a powder
inside a hopper, and to know more about the functional relation H(θ).
10.4 The powder flow function

We can represent the normal (σ) and shear stresses (τ) on a plane by
an equation that describes a circle, this is the Mohr's circle. Adopting a
coordinate system where the normal stress is plotted against the
shear, Figure 10.10, the centre of the circle (Ce) is
Ce = σ y + r and the radius (r) of the circle is

1
r= (σ x − σ y ) (10.14)
2
Thus, by inspection and substitution the centre is also
1
C e = (σ x + σ y ) (10.15)
2
By Pythagoras’ theorem for triangles
Fig. 10.10 Mohr’s circle 2
1 
representing shear stress τ 2 + (σ − C e ) 2 =  (σ x − σ y )  thus, from equation (10.15)
and normal stress on a  2 
plane by a circle 2 2
 1  1 
τ 2 = − σ − (σ x + σ y )  +  (σ x − σ y ) 
 2  2 
and
2 2
1   1 
τ 2 =  (σ x − σ y )  − σ − (σ x + σ y )  (10.16)
2   2 
i.e. in this coordinate system, equation (10.16) provides an equation to
represent the shear and normal stresses, which is the equation of a
circle where the centre is given by equation (10.15) and the radius by
(10.14). The question now becomes: can the normal and shear stresses
on a plane be represented by the above equation?
To answer this question the principal planes are considered, as
illustrated in Figure 10.11. A plane of stress is resolved into two
principal planes: where the shear (τ) and normal (σ) stresses of one
plane are represented by two planes with zero shear acting. The
resulting values are known as the maximum and minimum principal
stresses, with an angle of 90o between the principal planes.
Now, these are planes of say length l and width m, and force is the
product of stress and area (stress is similar to pressure). A force
balance gives:
horizontally
lm sin θσ x + lm cos θτ = lm sin θσ
Fig. 10.11 Plane of interest τ
∴σ x + =σ (10.17)
resolved into two principal tan θ
planes at right angles to and vertically
each other – on the
principal planes no shear lm cosθσ y = lm cosθσ + lm sin θτ
acts.
∴ σ y = σ + τ tan θ (10.18)
Rearranging for tan θ and combining these equations gives
τ 2 = (σ − σ x )(σ y − σ ) = σσ y − σ 2 − σ x σ y + σσ x
hence
2
[ ]  1
2


4
4
1
τ 2 = − σ 2 − (σ x + σ y ) − σ x σ y = − σ − (σ x + σ y )  − σ x σ y + (σ x + σ y ) 2
 4
and
2
 1  1
τ 2 = − σ − (σ x + σ y )  + (σ y 2 − 2σ yσ x + σ x 2 )
 2  4
Thus,
2 2
1   1 
τ 2 =  (σ x − σ y )  − σ − (σ x + σ y )  (10.19)
2   2 
i.e. the equation of a circle in σ and τ coordinates, see equation (10.16)
above. So, the principal planes illustrated above can be represented
on a Mohr's circle. Note that the minor principal plane occurs where
the normal stress axis (σ) is cut at the lower end of the circle (τ=0) and
the major principal plane is at the top end of the circle. We call these
σy and σx respectively - see the Mohr's Circle diagram.
The angles between planes and in the Mohr's Circle are related, as
can be seen from the vertical and horizontal force balances on the
principal planes (rearranged for tan θ)
τ σ y −σ σ y −σ
tan θ tan θ = tan 2 θ = =
σ −σ x τ σ −σ x
and using the general result
2 tan θ
tan 2θ =
1 − tan 2 θ
So,
2τ
σ − σx
tan 2θ =
σ y −σ
1−
σ −σ x
Hence,
2τ τ
tan 2θ = = (10.20)
(
2σ − σ x − σ y σ − 1 σ x + σ y
2
) Fig. 10.12 Mohr’s circle with
angle by inspection
Comparing equations (10.20) and Figure 10.12, shows that
2θ = φ (10.21)
i.e. the principal plane angle is half that given by the Mohr's circle.
We have now seen how it is possible to consider an arbitrary
plane inside a powder compact, with both normal and shear stresses
acting on it, and resolve it into two principal planes. A Mohr's circle
represents shear stress (on the y axis) and normal stress (on the x
axis). Hence, we can draw the shear condition of this arbitrary plane
on these axes and obtain the equivalent minimum and maximum
principal plane values. This is of interest if we know the stress
condition that will cause the powder compact to break, or fracture. It
is argued that the stress of interest is the maximum principal stress
(σx): this represents the maximum stress consolidating the powder,
i.e. giving it strength. Also of interest, for the purpose of breaking the
powder compact, is the unconfined yield stress – again a maximum
principal plane stress. The latter is obtained from the Mohr's circle
that must also go through the origin. Before we see this in practice we
should consider the Mohr's circle a bit further.
If we increase the normal, or shear stress, on a powder the Mohr’s
circle representing it will become bigger, see Figure 10.13. If we know
where the yield locus lies, then the circle can only become bigger
until it touches the yield locus. Note that if we treat powder compacts
as Coulomb solids the yield locus cuts the shear stress axis (i.e. does not
go through the origin) and usually has a small positive gradient. For
a given powder compact the maximum principal stress can be
obtained by drawing a Mohr circle that is tangential to the yield locus
at the upper end. The corresponding unconfined yield stress comes
from drawing a Mohr circle again tangential to the yield locus, but
Fig. 10.13 Mohr’s circle
with the minor principal plane stress going through the origin (i.e.
with yield locus – when a
circle touches the yield
unconfined with zero stress acting).
locus failure occurs However, the Coulomb solid formed by the powder compact will
have properties that depend upon how it has been treated. The
greater the original consolidating load, the larger will be the yield
locus. Hence, a series of yield loci will exist for a given powder,
dependent upon the consolidation conditions used to form the
compact. This series of yield loci can be used to provide a relation
between the strength of the powder to resist breakage and the
consolidating conditions used to form the compact; this is the
material, or powder, flow function.
Fig. 10.14 Cohesion and In a Coulomb solid there is a limit to the range of stresses that will
friction in a Coulomb solid cause no permanent deformation. A stress equal to the limit causes
plastic flow, see Figure 10.14, where Cy is similar to a yield stress and
is called the cohesion
τ = σ tan θ + C y (10.22)
Equation (10.22) provides the shear stress needed to cause failure of

the specimen at a given normal force. A free flowing powder will
have no cohesion, resulting in a line through the origin (Cy=0). For
cohesive powders, a shear cell can be used to determine a yield locus
where the powder is first consolidated to a given bulk density and
Fig. 10.15 Yield loci for state, then sheared under different values of consolidating load or
powder compact normal stress. The tests can be repeated to provide several yield loci
for the same powder, but under conditions of different initial
consolidation, see Figure 10.15.
The top point on each locus was obtained from a powder by
applying a fixed consolidating load before and during the shear test.
The same load was initially applied for all the other tests used, but
lower values were used during the test; the powder retains the
properties of the material formed during the pre-shear consolidation
process. The highest pre-shear consolidation load was used for YL3,
and YL1 had the lowest The powder porosity should decrease with
Fig. 10.16 Effective angle increasing consolidating load so YL3 represents the strongest and
of internal friction least porous powder compact.
The effective angle of internal friction is given by the angle of a line

going through the origin that is tangential to the Mohr's circle drawn
at the end of the yield locus, see Figure 10.16, and compare with
Figure 10.14.
Consideration of the Mohr's circle, provides two key elements that
are used in the characterisation of powders and hopper design: the
unconfined yield stress (fc) and the maximum principal stress (σx). The
unconfined yield stress is the maximum stress on a powder plane
where the other principal plane is under conditions of zero shear and
zero consolidation. This is the condition at the open side of a stable
arch, - see Figure 10.9. So, in order to break a stable arch - or to stop
one forming, we need to ensure that conditions within the hopper are
such that the state of stress is greater than the unconfined yield stress,
as given by the Mohr's circle going through the origin and tangential
to the yield locus. The physical state of the powder is given by the
consolidation conditions; treating the powder as a Coulomb solid the
powder strength will be greater when the consolidation forming the
compact is greater. The powder has been formed by the conditions of
shear stress and consolidation given by τa and σa respectively.
However, under the Mohr's circle development, we may represent
this state of stress as a single (maximum) principal plane stress σx.
Perhaps the best way of considering these parameters is to think of a
maximum consolidating stress (σx) that will give rise to a single value
of the unconfined yield stress (fc). The greater the consolidation then
the stronger the powder: hence the larger the value of fc and the Fig. 10.17 Construction for the
hopper will have to be designed to provide a greater arch breaking required data for the PFF
condition: e.g. steeper angle from the horizontal or larger opening.
These terms are illustrated in Figure 10.17.
This relation between the unconfined yield stress and the
maximum consolidation stress is often a single repeatable function
that characterises how the powder compact behaves; it is known as
the Powder Flow Function (PFF), or Material Flow Function (MFF). For
cohesive powders, it is a more useful and reliable form of powder
characterisation than angle of repose, etc.
The PFF is obtained by the following methodology. A single yield
locus is used to provide the Mohr's circle that can be fitted to the
stress condition at the end of that locus: i.e. a circle is drawn through
the top point of the locus such that it is tangential to the locus. The
maximum consolidating principal stress is read off where the Mohr's
circle cuts the normal stress axis. The unconfined yield stress comes
from another Mohr's circle plotted near the origin: the minimum
normal stress has to be at the origin and the circle must again be
tangential to the yield locus, see Figure 10.17. The unconfined yield
stress is read off where this Mohr's circle cuts the normal stress axis.
The two values are plotted as a single data point on the illustrated Fig. 10.18 The Powder Flow
PFF (or MFF) graph. This procedure provides one data point for the Function
PFF, so other yield loci must be obtained (under different
consolidation conditions) to provide sufficient data to draw the full
curve. The origin on the unconfined yield stress and maximum

consolidation stress figure provides an additional data point. An
illustrative PFF is shown in Figure 10.18.
For the powder we now know: the PFF and the effective angle of
internal friction, from Figure 10.16, which tells us something about
how stress will be transmitted within the powder. Thus, we have
characterised our powder and must consider the state of stress in the
powder caused by the hopper design and see whether it will be
sufficient to overcome the formation of a stable arch.
Fig. 10.19 Jenike design chart for 10.5 The hopper flow factor and hopper design
a powder with an effective To determine the Hopper Flow Factor (HFF) the following procedure is
angle of internal friction of 50o used: the appropriate Jenike design chart is identified using the
effective angle of internal friction. The angle of wall friction is
obtained from tests described later and the hopper slope from the
vertical is either measured, for an existing hopper, or selected for a
new design. The value selected is based on the knowledge that the
design chart, an example is illustrated in Figure 10.19, is split into two
regions. The bottom left region is for a mass flow hopper and the top
right will result in a core flow hopper. Thus, a value of the hopper
slope from the vertical that rests on the dividing line between these
regions is bordering on a mass flow design. For the sake of safety,
Fig. 10.20 Test for flow by
assuming that a mass flow design is required, it is usual to come 3
comparing the PFF and the HFF
degrees back towards the vertical and into the mass flow regime from
this line. On the design chart, the hopper ratio between compacting
stress and applied shear stress (ff) is estimated, and the HFF is plotted
on the same graph as the PFF, as a line with gradient of 1/ff, Figure
10.20.
To test for a stable arch in the hopper the following logic is
applied. Where the PFF lies above the HFF the powder has greater
strength to resist shear and collapse than the hopper/powder system
can provide. A stable arch is, therefore, possible. When the HFF is
above the PFF the system has sufficient stress to break an arch and
reliable flow should exist. The point at which the PFF and MFF meet
gives the critical unconfined yield stress (fc -critical) and this can be used to
determine the minimum hopper opening, from equation (10.13)
f c −critical H (θ)
B= (10.23)
ρb g
Fig. 10.21 H(θ) for types of where H(θ) is a constant dependent upon the geometry of the hopper
hopper opening where L is opening. The factor H(θ) arose in the derivation of the stable arch
one rectangular dimension where the arch was assumed to act over a linear distance L, which
and B the other: if L<3B use acts up to the centre of the hopper. The arch forms above the opening
the square curve but the linear dimension measured, or deduced, is the hopper
opening itself. This is not the same as L, or even 2 L, but bears some
relation to it. Jenike provided another chart, Figure 10.21, for this
relation, dependent upon the opening geometry and the hopper slope
from the vertical.
10.6 Measurement techniques and conditions

The simplest way to determine the unconfined yield stress is from a
Uniaxial Compression test, see Figure 10.22. Firstly, the powder is
compacted whilst constrained by the application of a normal
consolidating load. The constraints are then removed and another
compressive load applied. The value of this load at the point of
failure is the unconfined yield stress; i.e no side-walls and no shear
stress; solely a consolidating stress applicable at the point of failure. It
is fairly easy to relate this state of affairs to that of breaking an arch Fig. 10.22 Uniaxial compression
inside a hopper. However, the uniaxial compression test is not a test
reliable method. The most common method employed is the Jenike
shear cell.
During the Jenike test, Figure 10.23, two rings are employed
(upper and lower). The powder fills the rings and has a consolidating
load applied. This load is removed and a lower load applied, together
with a shear stress via the bracket on the side of the top ring. When
the shear stress is sufficient the top ring will slide over the bottom,
and the powder has sheared. This gives one value for shear and
consolidating stress, which may be plotted on a yield locus. A
measurement of wall friction can be achieved simply using the top Fig 10.23 Jenike shear cell
ring from the Jenike shear cell: a given normal force is applied and
the shear force required to slide the ring over the solid surface can be
measured. A series of experiments at different normal forces gives
rise to a graph from which the angle of friction can be deduced.
Using a load cell connected to a chart recorder the trace (i.e. shear
load) should increase steadily until the powder yields. This will be
the proper consolidation curve, or plot, illustrated in Figure 10.24.
However, if the powder only forms a loose aggregate the trace will
follow the under consolidated curve. Conversely, if the powder is so
tightly consolidated that dilitancy needs to be overcome (i.e. the
powder must expand before it can shear) then the over consolidated
curve will be followed. Only the test result from the proper
Fig. 10.24 Test for correct
consolidation should be used.
consolidation
An alternative to the Jenike cell is the Ring Shear Tester, Figure
10.25. Again a normal force is applied together with a shear or
rotational stress, see Powder Handling and Processing, Vol 8, No. 3,
1996, pp 221 - 226, or http://members.aol.SchulzeDie/grdle1.html
for further details.
If the powder is left for some length in time, then the compact
usually becomes stronger due to time consolidation. Hence, a new
yield loci, and powder flow function (usually stronger), will be
formed. The strength of a powder is also greratly influenced by the
prevailing humidity. Hence, it is important to conduct the tests under
Fig. 10.25 Ring shear cell
controlled humidity conditions that will be similar to that used for
powder storage.
A triaxial test cell, consisting of a rubber sleeve around a cylinder
of powder with side stress applied together with compressing stress
at the ends, can be used to investigate compact failure. The
continuum approach in three dimensions is illustrated in Figure

10.26, showing an element, or cube, of material with the
corresponding stress tensor
σ 11 τ 12 τ 13 
τ 
 21 σ 22 τ 23  (10.24)
τ 31 τ 32 σ 33 
However, consideration of only three mutually perpendicular

principal planes (no shear forces acting) results in the simpler matrix
shown in Figure 10.26, which can be drawn as a tetrehedron, or
pyramid, with three principal planes (x,y,z) that represent the
resolved forces from a plane joining the three principal planes. The
corresponding three dimensional Mohr's circle, represented on two
dimensional paper, can be drawn as three circles and is shown in
Fig. 10.26 Three dimensional Figure 10.27.
consideration of stresses with In order to maintain stable discharge from a hopper many bin
principal plane matrix below inserts have been marketted. The formation of a stable arch may be
disrupted by surfaces placed close to where an arch is most likely to
form. See Lyn Bates, The Chemical Engineer, 14 November 1996 and
numerous articles at: www.powderandbulk.com/. Consideration of
the fastener arrangement, and discharge system, within the hopper is
also important. Some examples of these are shown in Figure 10.28.
10.7 Summary
The hopper design procedure is summarised as follows: when
considering a powder for the first time a series of yield loci can be
Fig. 10.27 Three dimensional obtained by laboratory measurements and the Powder Flow Function
representation on Mohr’s circles (PFF) deduced. The effective internal angle of friction is also obtained
and the appropriate Jenike design chart identified. Tests also give the
wall friction. For mass flow conditions (no arch and uniform flow),
the operating region should be in the bottom left section of the Jenike
design chart. For the sake of safety, it is usual to come three degrees
back into the mass flow regime, i.e. reduce the hopper angle from the
vertical axis. This is, therefore, the hopper half angle. The hopper
flow factor is also deduced from this chart. A plot of the hopper flow
factor and PFF can be used to deduce the critical unconfined yield
stress and, therefore, the minimum hopper opening from equation
(10.23). Hence, the hopper is now specified in terms of angle and
opening diameter; storage capacity per silo can be deduced from the
geometry and the maximum discharge rate can be estimated from the
empirical equation (10.2), but this neglects coherency and should only
be used as a rough estimate.
Fig. 10.28 Hopper fastener The powder flow function is useful for hopper design, but is also a
and discharge arrangements
property of the powder and can be used as a characterisation
to avoid arch formation
technique in itself. For example, it may be used to correlate some
behaviour within a process, or from a product, with the PFF.
However, changes with the PFF dependent upon humidity and time
should be considered. See:
Apart from powder storage in silos, other common powder www.jenike.com
containers include an Intermediate Bulk Container (IBC), which is a for more information on
large sack that can hold over 1 tonne of powder. For transportation of hopper design
amounts in excess of IBC values trucks and wagons are needed.
10.8 Problems
Explain and answer the following terms:
Dilatancy Segregation Shear stress
Critical Porosity Normal stress Principal stress
angle of internal friction angle of wall friction
effective yield locus powder flow function
hopper flow factor unconfined yield stress
1. Mohr circle (why is the centre always on the τ=0 line)?

2. Given any value of τ and σ on a Mohr circle, at what angle to a
principal plane will failure occur?
3. Give physical meaning to the points at the opposite end on a
Mohr's circle lying on the axis τ=0.
4. How does increasing the state of stress influence the Mohr circle?
5. What is the yield locus and why is it a locus?
6. Sketch the yield locus of a free flowing powder and a cohesive
powder - is the latter a meaningful question?
7. Why is the maximum consolidation stress important?
8. Please explain the condition required for a stable arch.
9. A, B and C are badly drawn Mohr circles, together with three
data points from a yield locus. Explain what is happening as the
state of stress goes from A, through B and on to C. Do they all
really exist?
10. A Jenike shear test on a powder resulted in the following data:
Normal Shear Normal Shear Normal Shear

stress stress stress stress stress stress
Pa Pa Pa Pa Pa Pa
Maximum normal 2200 1340 1600 1000 1200 740
load
At subsequent 1600 1250 1000 920 900 740
loads 1000 1100 700 840 600 640
400 900 400 700 400 560
Derive the powder flow function for the powder. If the material is
hygroscopic, outline a series of tests to check on the importance of
changes in humidity on the hopper design.
11. Draw a diagram representing the variation of solids stress inside a

silo, and explain its shape. Contrast the solids stress with the
variation of hydrostatic liquid pressure.
The following results were obtained using a Jenike shear cell with
time consolidation of 0, 7 and 14 days. Obtain the theoretical
minimum opening for a cylindrical silo of hopper half angle 30o, H(θ)
of 2.2 and angle of wall friction of 15o. Use a powder bulk density of
1800 kg m−3.
0 Days 7 Days 14 Days

−2 −2 −2 −2 −2 −2
fc kN m σx kN m fc kN m σx kN m fc kN m σx kN m
2.6 3.0 5.5 7.0 9.0 10.0
3.1 5.5 5.7 9.0 9.2 12.0
3.4 8.0 5.9 12.0 9.3 14.0
The actual hopper opening is 1 m in diameter; determine the time

that the powder can be left in the hopper without arching. If this time
is too short what steps can be used to ensure that the material still
flows reliably?
12. Intermittent flow is experienced from a cylindrical silo with a

conical hopper that is several years old. The design criteria have been
checked by remeasuring the powder properties and wall friction. The
powder properties were found to be unchanged but it was
discovered that the hopper wall had become badly scored. Using the
following data and Figures 10.19 and 10.21 determine what minimum
diameter opening is now required. What solution would you offer to
ensure that the problem will not recur?
Powder characteristics:
Bulk density: 1800 kg m−3

Unconfined yield stress kPa 8.5 5.8 3.0
Maximum consolidation stress kPa 16.0 10.0 3.5
Wall friction results

In 1998 Normal stress kPa 8.00 4.00 2.00
In 1998 Shear stress kPa 1.41 0.71 0.35
In 2002 Normal stress kPa 8.00 4.00 2.00
In 2002 Shear stress kPa 3.23 1.62 0.81
Hopper half angle is 15o.
11 Crushing and classification
Estimates vary, but it is generally accepted that of all the energy used Classification is the term for
in the World something like between 1 and 10% is in comminution, i.e. industrially sorting particles
the processes of crushing, grinding, milling, micronising, etc. into different size fractions or
Changing the size of the particles by crushing creates many grades.
important industrial products. An example is sugar, which has three
different grades: granular, castor and icing; chemically they are the
Table 11.1 Moh’s scale
same material the main difference is their particle size. So, grinding
followed by classification is all that is needed to produce the three
different products.
Historically, grinding became a mechanised industrial process
with the advent of water and wind powered mills to process wheat,
barley. animal feed, etc. The mills used a flat stationary stone with a
moving mill wheel revolving on-top. A derivative of this type of mill,
an edge-runner mill, can still be found in use today, albeit electrically
driven. In old mills classification was important to separate the flour
from the husk, this was often achieved by sieving. Modern mills
often combine classification and milling within the same device by
having an up-draught to carry the finer particles away from the
milling section. These are known as air swept devices.
Milling of minerals has been an important part of the recovery of
metals and industrial minerals for many centuries. Often a mineral of
interest is surrounded by rock of a different type, which may be Diamonds are processed by
worthless; i.e. a gangue mineral. The grain boundary between the crushing the rock to liberate
desired mineral and the gangue will be the weakest mechanical point the diamond (hopefully the
and the most likely to break. Thus, grinding to the liberation size will diamond doesn't crush
release the valuable mineral so that it may then be separated from the sometimes it does). One big
gangue. This is a process that is employed for metal ore mining as diamond is worth a lot
well as precious minerals recovery – see the box below Table 11.1. In more than two small ones!
the table the Moh's scale of hardness is shown. It is based on the
mineral lower in the table being able to scratch the mineral above.
'Soft' minerals are 1 to 3, 'medium' are 4 to 6 and 'hard' are 7 to 10.
However, the hardness value is entirely arbitrary, it is merely a
ranking of the described minerals. The table also includes the Bond
Work Index, which is discussed in the next section.
An initial size reduction from material several centimetres in
diameter down to one centimetre, or so, is often termed primary
crushing. This may be followed by further size reduction, secondary
crushing, and then pulverising, or fine grinding. The term micronising
has become popular for the reduction of particle size to this
dimension. Thus, coal being fed into a pulverised fuel burner for
electricity generation will undergo the first three grinding operations.
Primary crushing normally takes place close to the mine head and
relies on equipment with very large throughputs, usually having two Fig. 11.1 An attritor mill
surfaces approaching and retreating from each other. Examples
include Jaw and Cone Crushers. Secondary crushing can be by rotating

surfaces such as swing hammer mills, for brittle materials, and roll
crushers. Finer grinding usually takes place in rotating vessels, such
as ball and rod mills. Very fine grinding, to sizes less than 10 µm,
requires high energy and attrition mills are often used. These have
much lower throughputs than the primary crushers and an example
of an attrition mill is illustrated in Figure 11.1. These are usually
found in the fine chemical industry. Examples of these and other
types of machines can be easily found on The Internet.
Experimental observations suggest that particles may undergo
Fig. 11.2 Plastic particles
elastic cracking and viscoelastic cracking. In the former case the crack
before grinding
propagates around curves like onion peels and high longitudinal
tensile stress is displayed. Viscoelastic is a term used to imply
properties of both a liquid and a solid and such cracking propagates
through a particle like orange slices and the longitudinal strain is
compressive.
The different type of fracture mechanisms may be observed from
the particle size distribution functions of a mill product. Figure 11.2
provides the particle size distribution of a mixture of two plastic
waste products. After grinding, the particle size distribution changed
to that shown in Figure 11.3. Clearly, the two different types of plastic
displayed different behaviour during crushing. Both were crushed:
the coarse end of the distribution shown in Figure 11.2 has been
removed, but the effectiveness of the crushing depended on the
material processed. Hence, two separate distribution curves can be
deduced from the composite mill product distribution curve. The two
separate distribution curves were deduced from the Rosin-Rammler
Bennett (RRB) cumulative function, equation (2.5)
  x n

N 3 ( x ) = 1 − exp −  


  x 63.2  
Fig. 11.3 Plastic particles after  
grinding – two separate where x63.2 represents the particle size at which 63.2% of the
distributions added together to
distribution lies below and n is a constant called the uniformity index.
provide the composite curve
Hence, two separate curves were deduced for the crushed plastics
with RRB model parameters
and they were added together in the mass ratio of 0.27:0.73 from
distributions 2 and 1, respectively. The different crushing behaviour
of mixed plastics is one way in which they may be separated for
recycling. Also, their behaviour could be changed by the temperature
employed during crushing.
11.1 Energy utilisation

If a particle undergoes brittle fracture a crack should propagate
through the particle creating new surface area, see Figure 11.4. There
will be a corresponding increase in the surface energy (γ – measured
Fig. 11.4 Crack propagation in the units of J m-2). The relation between the tensile stress (σB), crack
in brittle fracture length (cL), Young’s modulus (E) and surface energy is
σ B = (2γE / πcL )1 / 2 (11.1)
Efforts to relate the increase in the surface area to the energy required Noise can be used to
for the crushing have met with very limited success. The theoretical control the mill - an
energy required in comminution to create a new surface is often less empty mill 'rings'.
than 1% of the required amount – a lot of energy goes into effects
other than the new surface itself. During crushing, energy is used by
Elastic deformation
the processes of: elastic deformation (not including breakage) of the returns to the
particles, inelastic deformation, elastic deformation of equipment, previous shape but an
friction between particles, friction between particles and machinery, inelastic particles
noise, heat and vibration, and friction losses in equipment drives. bends out of shape or
Together with the energy required to form a new surface. flattens like putty.
11.2 Crushing laws

All the common basic equations relating energy requirement for
comminution can be derived from a single ordinary differential
Exercise 11.1
equation
Using equation (11.2)
dE
= −kLm (11.2) derive the three
dL crushing law equations
where E is power, L is particle dimension (diameter), k is a constant by substituting in n and
and m takes one of three values depending on the particle size: -2, -1.5 integrating.
or -1. The resulting equations are due to: Rittinger, Bond and Kick,
respectively.
Rittinger (1867) postulated that the energy per unit mass required is
proportional to the new surface area produced, i.e. energy = (final
surface area - initial surface area). Using an energy per unit mass
basis, where
surface area = fsa x2
mass = fv x 3
Then,
x 2 x 2 
E ∝  f 3 − i 3  or E = k R ( S vfinal − S vinitial ) (11.3)
 x f xi 
where xf and xi are the final and initial diameters. Kick (1885)
assumed that the energy required is equal to the reduction ratio
based on the ratio of the volume before to the volume after crushing.
Thus, the energy required is directly proportional to the size
reduction, e.g. the same amount of energy is required to go from 2 to
Fig. 11.5 Crushing laws
1 cm as from 1 to 0.5 cm.
x 
E = k K ln f  (11.4)
 xi 
Bond (1952) suggested that the total work required to break a cube of
size x is proportional to x3. On fracture and creation of the first crack
the energy becomes proportional to x2, so it was argued that for
irregular particles the energy must lie between x2 and x3 and a value
of x5/2 is taken. Again on an energy per unit mass basis and the
xf=x80% value
The x80% value is the  x 5/ 2 x 5/ 2 

particle diameter E∝ f 3 − i 3 
where 80% of the  x f x i 
distribution lies below  10 10 
on the cumulative giving E = Wi  −  (11.5)
mass undersize curve.  x f xi 
where Wi is the Bond Work Index, again in energy per unit mass
terms. The index is defined as the energy required to crush from
infinite size down to 100 µm, hence the 10 inside the brackets. Bond's
Work Index values, in kWh per short ton: i.e. 2000 lbs or 907 kg,
roughly follow the Moh's scale of hardness, see Table 11.1.
In Summary, Kick's law is better for larger particles and Rittinger's
for fine grinding, see Figure11.5. Large particles are easier to break
than smaller ones as they have more cracks and faults. Often the laws
are used in conjunction with empirically given constants: it can be
dangerous to use the book values without some test work.
11.3 Breakage and selection functions

The Breakage function, B(y,x), is the fraction by mass of breakage
products from size x that fall below size y, where x≥y. The Selection
function, S(x), is the fraction by mass of particles that are selected and
broken in time t – sometimes referred to as the specific rate of
breakage and related to revolutions of the mill.
Remember that N3(x) is the cumulative mass fraction below size x,
and we can define the mass of the mill charge as W. Then the amount
broken below y in time dt is:
dN 3 ( x ) dN 3 ( x )
comminution mechanisms WB( y , x ) dxdt where = n3 ( x )
dx dx
compression: between two
in general, N3(x) and n3(x) are functions of x and t, call these N3(x,t)
solid surfaces,
and n3(x,t). Concentrating on size y in the mill, noting the definition of
attrition & impact: against a
solid surface and other
breakage function, then at t=0 W=W0 (the original mill charge) and
particles, using F for mass feed rate and R for mass removal rate then:
cutting: of the particles, mass in mill below size y at time t is =
shear: against surrounding original mass + mass added - mass removed + mass created from
fluid, particles and surfaces, sizes above y and =f[S(x) & B(y,x)]
non-mechanical: e.g. laser and Algebraically this is represented as:
t t
plasma ablation.
WN 3 ( y , t ) = W0 ( N 3 ( y ,0)) + ∫ [ N 3 ( y , t )] F Fdt − ∫ [ N 3 ( y , t )] R Rdt...
0 0
∂N 3 ( x, t )
+
∂t ∫∫
S( x )B( y , x )dxdt (11.6)
In practice, it is difficult to distinguish between the breakage and

selection functions, both are measured together. Tests have been tried
with single particles and closely sized mill feeds. For most
calculations it is necessary to assume S(x)=1.
There are many factors that influence the breakage and selection
functions for particles, these include: toughness, abrasiveness, feed
size, cohesiveness and adhesion, particle form and structure,

softening and melting, organic content, particle size distribution, bulk
density and agglomeration. In addition the following factors need to
be considered to ensure safe operation: toxicity, explosion and fire
hazard. Some of these properties may be favourably changed by
milling cryogenically, to encourage brittle fracture, and this is
common for plastics. Milling under an inert atmosphere is common
for explosive materials and ones that may spoil by exposure of fresh
surfaces to air.
11.4 Milling circuit matrix

Broadbent and Callcott introduced a breakage function of the form
exp( y / x ) − 1
B( y , x ) = (11.7)
exp(1) − 1
For example, taking the milling of material less than 1000 µm it is
possible to define five grades: 1000 to 800, 800 to 600, 600 to 400, 400
to 200 and material all less than 200 µm. Hence, the grade boundaries
(y) will be: 800, 600, 400 and 200 µm. We can define the fraction by
mass in these grades before milling as: x1, x2, x3, x4 and x5. The
breakage function for grade 1 (1000 to 800 µm) with y=800 µm will
provide the fraction, by mass, of material starting in this grade and
ending up below 800 µm. In order to apply equation (11.7) a single
value must be selected to represent the grade, which is usually the
mid-point; i.e. 900 µm for grade 1 above. The fraction by mass of
material after milling in grade 2 (800 to 600 µm), and originating from
grade 1, will be the amount calculated above minus the amount
passing through the grade 2 to 3 boundary; i.e. the breakage function
with x=900 µm and y=600 µm. All the resulting fractions by mass of
material starting in grade 1, and now residing in grades 1 to 5, must
be equal to unity; to ensure conservation of mass. This logic is
illustrated in Figure 11.6.
The example provided above only considers the milling of the size
grade 1, i.e. 1000 to 800 µm. Size grade 2 will also be milled, resulting
in breakage products in the grades 600 to 0 µm together with
Fig. 11.6 The breakage
unbroken material remaining in grade 2: 800 to 600 µm. The same
function and conversion into
logic can be applied, using equation (11.7), but with a value of x=700
amounts entering differing
µm to represent this grade. Likewise grades 3 and 4 must be grades after milling
considered this way. However, all the material entering the mill that
is less than 200 µm will leave the mill still less than 200 µm; hence, the
breakage function will be unity for grade 5. It should be evident that,
as this is a milling operation and not an agglomeration, the fraction
by mass starting in a lower sized grade and reporting, after milling,
to a higher sized grade will be zero. If we represent the mill product
by p and the mill feed by f, in mass flow rates, then the material
breaking from the top size and remaining in the top size grade will be
p1 = b11 f1
where b11 was defined in Figure 11.6. Likewise, the mass flows
entering all the other grades will be
p 2 = b 21 f1 ; p 3 = b 31 f1 ; p 4 = b 41 f1 ; p 5 = b 51 f1
The full mass flow rates for the mill product, and a schematic
representation of it, are illustrated in Figure 11.7. It is possible to
represent this material balance in matrix form, as shown below
 p1  b11 0 0 0 0   f1 
 p  b b22 0 0 0   f 2 
 2   21
 p 3  = b31 b32 b33 0 0  f3  (11.8)
    
 p 4  b41 b42 b43 b44 0  f 4 
 p 5  b51 b52 b53 b54 b55   f 5 
  
Fig. 11.7 Mill product and The matrices may be represented as follows
feed material balance
{p} = M{f} (11.9)
where M is the milling matrix. Figure 11.7 illustrates a single pass of
material through the mill and it is highly unusual that such a single
pass would produce the desired mill product quality. It is common to
recycle mill product back to the mill for a second, or more, chance of
breakage. Thus, Figure 11.7 represents what is called open circuit
grinding. In order to return oversize material back to the mill, but
remove undersize material that is our product, a classifier is required.
A classifier is schematically represented in Figure 11.8 and the
Fig. 11.8 Schematic classifier matrix is shown below
representation of a classifier
c11 0 0 0 0 
0 c 22 0 0 0 

C=0 0 c33 0 0  (11.10)
 
0 0 0 c 44 0 
0 0 0 0 c 55 

It is usual to define grade efficiency for a classifier as being the fraction,
by mass, of material entering the fine cut of the classifier. During
classification no alteration of particle size takes place, it simply sorts
the particles within a grade: some entering the coarse cut and some
the fine. Hence, the classifier matrix is a leading diagonal matrix of
Fig. 11.9 Closed circuit fractional values, where the top left element should be a low value
grinding
and the bottom right element should be approaching unity.
The mill and classifier, with recycle, constitute closed circuit
grinding and are illustrated in Figure 11.9. At steady state, the mass
fed to the circuit {f} must be in balance with the circuit product {q}.
The mass flow rate of material being returned to the mill by the
classifier is large and the circulating load is defined as the ratio of the
amount returned {p-q} to the amount fed {p} or {q}. Typical values of
circulating loads are in the 100’s of percent. Clearly, the circuit
product is the classifier matrix operating on the mill product
{q}=C{p} (11.11)
Using the labelling shown in Figure 11.9, the mill product is
{p}=M{k} (11.12)
and the recycle is
{p-q}=(I-C){p} (11.13)
where I is the identity matrix. Considering the mass flow entering the
mill, which is the summation of the feed to the circuit and the recycle
{k}={f} + (I-C){p}={f} + (I-C)M{k}
and rearranging for {k} using matrix algebra
{k}=[I-(I-C)M]-1{f} (11.14)
It is possible to substitute equation (11.14) into equation (11.12), to
provide the mill product and then into equation (11.11) to provide the
circuit product.
The practical possibilities represented by equation (11.14) are
significant. Provided the operating matrices remain constant, i.e. C
and M do not change under different loadings, then using equations
(11.11) to (11.14) provides a means to determine what the effect of a
change in the feed condition will be on all the mass flow rates
throughout the rest of the circuit, including the circuit product. In
order to use equation (11.14) the feed column vector {f} must be
known, but this is likely to be the case. It represents the feed mass
flow rates in the size grades selected: 1000 to 800 µm, etc. in the
example quoted. Hence, simulation of the milling circuit under
different loading conditions can be provided: covering both a simple
increase in the overall feed flow rate into the circuit and the influence
of a change in size distribution of the feed. Also, it is possible to check
the change in energy requirement by the mill using the matrix
modelling: e.g. if Rittinger’s constant in equation (11.3) is known,
then the specific surface of the feed and mill product may be
calculated from the mass flow rates in these streams by dividing the
mass flow in each grade by the total mass flow, to give the mass
fraction, and then using equation (2.17) for specific surface. More
complicated mill circuits can be modelled in a similar way, the
objective is always to relate the column vectors within the circuit to
the feed column vector, using only the operating matrices that are
assumed to remain unchanged under the different operating
conditions.
11.5 Population balances

The milling matrix approach described above is a convenient method
of conducting a material balance within a particle technology unit
operation. It is possible to write a set of simultaneous algebraic
equations to perform the same task, but the matrix format is simpler.
Also, computer spreadsheets provide all the required matrix
functionality required to solve the equations, including matrix
inversion. The generic term for the material balancing described Fig. 11.10 Two particles merged:
above is a population balance. Splitting the distribution into separate volumes are additive but not
grades enables the investigation of each grade separately from its diameters
neighbours to be performed. Thus reactivity, or physical processing

characteristics, that are specific to that grade can be applied. Further
examples of unit operations that can apply this method include:
crystallisation, agglomeration, particle production by atomisation
and attrition in fluidised beds. In many cases both particle break-up
and formation have to be considered simultaneously.
Care has to be exercised over some population balance processes,
as illustrated in Figure 11.10. If two particles are completely merged
Fig. 11.11 Illustration of a ball together it is reasonable to assume that the resulting volume will be
mill with an attached grate the same as the volume of the two particles added together.
discharge to retain balls and However, the resulting particle diameter is not the same as the two
oversize material diameters added. Thus, it has been argued that a more complete
description of the breakage function should be based on volumes,
and not particle diameters. The deaths of particles within a given
volume increment D(V) is represented by
D(V ) = S(V ) no (V ) (11.15)
where S(V) is the selection function and no(V) is the number of
particles per unit volume. All the terms in equation (11.15) are
volume increment dependent; so, (V) is used. The births B(V) are
provided by particles from sizes above breaking and creating
products that lie within the volume range currently considered
∫
B(V ) = b(V1 , V 2 ) D(V1 )dV (11.16)
where the breakage function has been modified for particle volume
rather than diameter. Equation (11.15) can be introduced into (11.16)
to provide an expression for the births in terms of both the particle
breakage and selection functions. Numerical values may be assigned
to the selection function based on the assumption that it is
proportional to the particle volume.
The birth and death rates may be combined with the appropriate
flow particle equations for the system under consideration; e.g. once
through or recycle, in order to conduct a complete material balance
based on the usual input-output=accumulation.
11.6 Summary
The production of particles by crushing and grinding is a very
Grinding media important operation throughout the world. In most cases mechanical
Common grinding media classification is also required, to return the oversize material to the
such as balls in a ball mill, or
mill. Common classification equipment includes sieves, or when used
attrition mill, include:
on an industrial scale these are called screens. In some cases decks of
alumina (ceramic),
screens may be used to fractionate the product: the largest screen size
carborundum,
on the top deck. Slotted screens are often used to minimise screen
silica,
blinding, or blockage. Alternative classifier designs include cyclones
zirconia,
and hydrocyclones for dry and wet milling operations respectively,
nylon, and
steel.
and wet classifiers using screws and rakes to lift the coarser particles
to one end, whereas the finer particles leave with the liquid flow at
the other. In some cases the screen can be bolted onto the mill itself;
for example, on a ball mill the feed enters at one end and the product
may leave over an attached screen, called a trommel, at the other.
Grates, or screens, may be used within the mill to retain the grinding
media and some of the oversize, see Figure 11.11.
No mechanical classification will be perfect, normally the
separation curve, see Chapter 14, is very shallow. Hence, a significant
amount of fine material is usually recycled to the mill and oversize
material may be found in the circuit product. For critical applications
it may be necessary to classify the product for a second, or more,
times to remove oversize. One of the most fundamental decisions to
be made is whether to wet or dry mill a material. In most cases a dry
product is required, but dry milling very fine material (<20 µm) is not
efficient. The dry powder tends to coat the grinding media and
vessel. Wet milling provides a viscous medium that absorbs some of
the grinding energy, but it does help to disperse the fine particles so
that they can receive impact, shear and abrasion. However, the need
for wet milling, or otherwise, is a material dependent property and
should be considered for each case individually.
11.7 Problems
1. Rock is crushed in a cone crusher. The feed is a nearly uniform 50
mm sphere. The screen analysis of the product is given below. The
power required to crush this material is 0.429 MW/tonne, of this
11.18 kW is required to drive the empty mill. By reducing the
clearance between the crushing head and the cone, the screen
analysis of the product becomes finer, i.e. second grind product.
Tyler Mesh size Mesh size in First grind Second grind

µm product (%) product (%)
-4 +6 4699 to 3327 3.1 -
-6 +8 3327 to 10.3 3.3
-8 +10 2362 to 20.0 8.2
-10 +14 1651 to 18.6 11.2
-14 +20 1168 to 15.2 12.3
-20 +28 833 to 12.0 13.0
-28 +35 589 to 9.5 19.5
-35 +48 417 to 6.5 13.5
-48 +65 295 to 4.3 8.5
-65 0.5 -
-65 +100 208 to - 6.2
-100 +150 147 to - 4.2
-150 104 - 0.3
a) Using Rittingers Law, what do you estimate the power Rittingers Law
hint: use energy per unit
requirement for the second grind to be?
mass is proportional to
change in specific surface
b) Using Bond’s method, estimate the power necessary for each
grind, refer to Table 11.1 for a suitable Work Index.
2. A mineral has the following breakage function

exp( y / x ) − 1
B( y, x) =
exp(1) − 1
and is being classified and crushed in the circuit shown left. Where M
is the mill and C is the classifier. The feed to the circuit is:
cumulative mass undersize (%): 100 85 24 10

particle diameter (µm): 1000 750 500 250
and the feed rate into the circuit is 5 tonnes per hour.
i) Construct a mill matrix based on the above diameters using the

given breakage function.
ii) If the classifier can be represented by a leading diagonal matrix of
elements: 0.1, 0.3, 0.5 and 0.7, what will be the size distribution and
total flow rate of the coarse cut from the classifier?
3. i) When operating a mill in closed circuit show how it is possible to

represent the stream entering the mill (k) as
k = af
where f is the column vector representing the feed entering the circuit
and a is an operating matrix.
ii) A mill requires 0.4 MW/tonne when the operating matrix a,
milling matrix and circuit feed column vector are
 1.923 0 0   0 .6 0 0   2 
   
 0.361 1.563 0 ;  0.3 0.9 0 ;  1 
 0.025 0.017 1.111  0.1 0.1 1   0.1
   
where the size ranges corresponding to the rows are 5000 to 2000,
2000 to 1000 and 1000 to 0 µm, and the feed column vector rows are
in tonnes per hour. Calculate the empirical constant in Rittinger's
equation in the units of metres Mega-Watts per tonne, and the
circulating load. If the circuit feed rates in each grade were to change
explain briefly how the milling model and energy equation may be
used to predict circuit performance and mill energy requirement.
Suggest other important factors that would influence the actual mill
power requirement if the feed rate was changed.
4. Granulated sugar is sieved at 430 µm in sieve A. The oversize is

Note, this question expects a milled and the product sieved in an identical sieve B. The oversize
series of operator matrices from this sieve is recycled to the mill. The undersize is combined
and the only column vector with the undersize from sieve A and is further sieved at 250 µm to
should be for the feed. yield two products: a fine cut which is icing sugar and a coarse cut
which is castor sugar. Sketch the circuit and obtain an expression for
each product assuming that the mill is represented by matrix M,
sieves A and B by matrix C and the final sieve by matrix D.
12 Solid/solid mixing
The mixing of solids is a critically important operation in many
industries, especially in pharmaceutical production where the active
ingredient in a formulation may be toxic and be present at only 0.5%
by mass overall. A product with too low an active ingredient will be
ineffective and a product with too high active ingredient may be
lethal. To provide good solid mixing the phenomenon to be avoided,
or overcome, is the particles’ tendency to segregate. Segregation
occurs when a system contains particles with different sizes,
densities, etc. and motion can cause particles to preferentially
accumulate into one area over another; e.g. large particles work their
way to the top of breakfast cereal – fines are found at the bottom of
the packet. In contrast, motion of gases and miscible liquids due to
flow (convection) provides mixing on a large scale and molecular
diffusion is important for completing the process at the micro-scale
of mixing. Thus, removing the top off a perfume bottle in a room will
result in the vapour evenly distributing itself throughout the room. A
particle mixture will never be as homogeneous as that of a fluid as
particles tend to segregate, whereas fluid molecules tend to mix.
In general, for particle mixing the following physical particle
properties should be considered. Monosize particles are easy to mix,
provided they are free flowing, but segregation by size, density and
rotational inertia are possible with free flowing powders possessing
differences by these properties. Fine particles with high surface
forces (diameters <100 µm and very high forces with diameters <10
µm), may need agglomerate breakage requiring high power, but can
give good mixing of cohesive powders. Aeration: e.g. catalyst particles
in gas fluidisation, may undergo diffusional type mixing, which is a
low energy process, but with a risk of powder flooding. Friability: for
delicate particles mixing by shear mechanisms would be
inappropriate. Explosion hazard: an inert gas blanket is needed and
low specific power input (low shear) is required. Physiological hazard:
need to avoid airborne dust formation. Adherence to surfaces: easy to
clean surfaces needed and if a liquid cleaning fluid is used then a
new pollution problem may result.
Three mechanism types are often used to describe mixing
performance: diffusion, but not molecular diffusion – an expanded
bed of free flowing material occurs with particles in random
movements, convection – when volumes, or regions, of the mix are
moved en-masse to different areas, and shear – mixing occurs along
the slip planes between regions of particles. All the mechanisms may
exist in a single mixer, but one or two may predominate. The mixer Fig. 12.1 Stages in mixing dark
type needs to be right for the material mixed, e.g. cohesive powders and light coloured beads to
are more likely to require shear (and convection) hence blades and give a complete random mix
ploughs are more appropriate than tumbling.
The mathematical description of the mixing process starts by

considering the simple case of mixing two components differing only
by colour.
12.1 Binary component mixing

Figure 12.1 illustrates 113 light
coloured chips and 101 dark
coloured ones at various stages
of mixing. The overall
proportion of dark chips is 0.472
(i.e. 101/214) and, at any
instance in time, if we were to
split the mixture into many
different samples and
investigate the proportion of
dark chips within the sample we
would expect to obtain a
proportion of 0.472. However, it
is unrealistic to expect all the
samples to have this proportion
of dark chips: some samples will
have more, some less, but the
Fig. 12.2 The random mix split into 21 samples and the
overall mean average of dark
proportion of dark chips measured in each sample: mean
chips as a proportion must equal
proportion is 0.472 with a standard deviation of 0.195
this value. Figure 12.2 illustrates
such a collection of samples, where there are 21 separate samples: the
lowest proportion of dark chips is 0.1 and the highest is 0.78. The
mean proportion calculated over all the samples is 0.472, the same as
the overall proportion used, and the standard deviation and variance
around this mean is 0.195 and 0.0382, respectively. This illustrates an
important concept: the between sample variance. If these samples were
to be sold as our product, then the variance between the products
would be an important measure of the difference in quality of our
product. Thus an understanding of the expected difference is
important. The statistical terms and the normal distribution are
briefly discussed in, and below, Figure 12.3.
Using our knowledge of the normal distribution, and assuming
that our randomly sampled coloured chips follow it, we would
expect 68.2% of the samples to have a proportion of dark chips equal
to 0.472 plus or minus the standard deviation, i.e. 0.472 ±0.195.
Hence, out of 21 samples 3.3 samples should be less than 0.277 and
3.3 samples more than 0.667. In Figure 12.2 we find 4 samples greater
than 0.667 and 3 samples less than 0.277, in accordance with the
Fig. 12.3 The Normal
distribution. The lowest proportion was 0.1, the highest 0.78; both are
probability distribution –
symmetrical around the mean
within the 95% that we expect: i.e. 2 times the standard deviation
value from the mean. We would need to take 40 samples to find one below
0.082 and one above 0.862.
Figure 12.4 illustrates an ordered dispersion of particles, rather

than a random dispersion. This might be the degree of mixing
desired, but it is very unlikely that it will be achieved by random
mixing; i.e. it could only be achieved by placing the particles in order
and not by a mechanical mixing process.
In powder mixing, pioneering work was by Lacy (1943, Trans.
IChemE, 21, p53 & 1954, J. Appl. Chem., 4, p257), in showing that for Basic definitions
a binary mixture, of identical particles apart from colour, the variance Mean:
is p = ( p1 + p 2 ... + p n ) / no
pq
σR2 = (12.1) Standard deviation:
no
2
∑ ( p − p)
where no is the number of particles in the sample, p and q represent σR =
no − 1
number proportion of the two components (where p+q=1), p and
or divide by no for values
σ R 2 are the mean and variance between the samples. Lacy also greater than 30.
showed that the variance between samples for one component in the
Variance:
binary mixture in its unmixed state (σo2) will be
= σR2
σ o 2 = pq (12.2)
For a normal distribution
This represents the worst case that one would expect. An example of the mean = median=
the use of these equations follows. Assume that a type of children's mode
confectionery is sold in tubes containing 100 coloured sweets, equal
proportions of blue and red. In all other physical characteristics the
sweets are identical. Hence, if the sweets are batch mixed before
filling the tubes we should have the following situation: comparing
the worst-case random variation of red sweets between tubes we
would expect the variance to be
= 0.5x0.5 = 0.25 (i.e. standard deviation σo = 0.5)
and at best the variance will be
= 0.5x0.5/100 = 0.0025 (std. deviation σR = 0.05)
Now, to put these values into context. If we assume a normal
distribution then 95% of the distribution lies ±1.96σ from the mean
value. Hence, when fully mixed, out of 100 tubes of sweets we would
expect to find 95 with a proportion of 0.5 reds ±1.96 x 0.05, i.e. 50 reds
± 9.8 sweets. For the sake of rounding let's call this 2σ, and 50 reds
plus or minus 10 sweets. Five tubes contain sweets outside of this
limit: 2.5 tubes with less than 40 reds and 2.5 tubes with more than 60
reds. Similarly, using the 99.8% confidence limit (3.06σ) we have: 0.5 Fig. 12.4 An ordered
reds ±3.06 x 0.05 = 0.653 hence out of 1000 tubes 1 tube will have mixture – not realistically
more than 65.3 red sweets and 1 tube will have less than 34.7. In possible by random mixing
summary, out of 1000 tubes 25 tubes have less than 40 red sweets and
1 tube has less than 35. Similarly, 25 tubes have more than 60 red
sweets and 1 tube has more than 65. This assumes a normal
distribution and no bias in mixing or filling the tubes. The idealised
perfect mix is all tubes containing 50 red sweets but this cannot be
obtained by random mixing, only by structured mixing (i.e. counting
and selecting particles during filling). Now, if our mixer doesn't

provide adequate random mixing then the number of tubes outside
the above limits will be even greater; hence we have calculated the
best case random mixing based on particle properties alone. Tests may
be required to establish which mixer provides the best random
mixing and how quickly it is achieved.
In industry it is common to weigh materials to be mixed: this is
fast, accurate and convenient. Hence, the work of Lacy was extended
to provide the random variation using masses, first by Stange in 1954
PQ   σ Q2  
 2
σR2 = P wQ  1 +  + Qw  1 + σ P  (12.3)
P 
M   wQ 2  wP 2
    
where P, Q, are mass fractions of two components, M is the mass of

sample taken, wp and wQ are the mean mass by number of particles in
the two components, i.e.
π x - max
wP = ρ s ∫ x 3 n 0 ( x ) dx (12.4)
6 x - min
and σ Q and σ P are the standard deviations of the distribution of

weights. This equation was later (1964) simplified by Poole, Taylor
and Wall, based on mass distributions
PQ
σR2 =
M
(
PWQ + QWP ) (12.5)
where WP and WQ are the mean mass by weight of particles in the two
components. i.e.
π x - max
WP = ρ s ∫ x 3 n 3 ( x ) dx (12.6)
6 x - min
assuming spherical particles and a homogeneous solid density. The

resulting variance and, therefore, standard deviation is that for the
variation around the mean value by mass proportion assuming a
normal distribution. Equation (12.5) may be extended to provide
information on mixtures containing more than two components by
creating a pseudo-two component mixture; i.e. considering one of the
components of interest against all the other components aggregated
together. However, to do this the value of the weight weighted mean
Fig. 12.5 How the variance
particle mass of the other components must be calculated using the
changes during mixing
mean masses of its components combined together in the correct
mass proportions in which they are present.
12.2 Specification and confidence

In the example of sweets we may have needed to meet a certain
specification for our product; let's say that customers are prepared to Specification is what
accept some variation in the product, up to a point, and that is 15 red you want, and
sweets around the mean value. This would be our specification: 50 confidence is the
likelihood of
±15 red sweets in each tube. Our confidence in meeting this
achieving it.
specification is 99.8%; we know that 1 tube will be below this limit,
and 1 above, from 1000 tubes. Unfortunately, in random powder
mixing we can never be 100% confident of meeting a specification.
We can use the basic equations to tell us how we should minimise the
number of off-spec products, and we can adopt different mixing
strategies to help: e.g. add liquids to create a cohesive mixture that is
not so subject to these random variations. However, all these
strategies have disadvantages; e.g. cohesive mixtures take a lot of
effort to mix.
We have seen already that mixture quality can be quantitatively
assessed using Lacy's work on the random mixture variance (σR) and
the unmixed variance (σo). However, we may need to know how
good a mixture is between these limits. It is possible to define a
mixing index M, which is equal to unity at the start of an unmixed
process and approaches zero when perfectly randomly mixed
σ 2 −σ R2
M = (12.7)
σ o2 −σ R2
However, there are at least 30 more definitions of mixing index, so it

is important to understand whichever definition is being used.
Recently the term blender efficiency (BE) has been proposed where
σ N.B. The variance
BE = (12.8)
σo reduces in a logarithmic
way towards the
For a more comprehensive analysis of mixing indexes see Fan, L.T., et random value - but
al, 1979, Powder Technology, 24, p73. segregation may cause
The rate of mixing is very important: equations (12.1), (12.3) and the curve to rise up
(12.5) only provide the mixture quality when fully randomly mixed. again and poor mixer
Equation (12.2) provides the starting mixture quality and the quality selection, or operation,
may mean that the
between these two extremes may vary in accordance with Figure 12.5,
random value is never
if the mixing is mainly by a mechanism that is similar to diffusion; i.e. attained.
random motion of particles at a length scale similar to the particle
size. However, it is perfectly possible for the variance to rise again
during mixing because of segregation.
Another important consideration when investigating mixing is the

size of sample to take, i.e. the scale of scrutiny. Note, this is not the
same as the size of sample required for analysis, sub-division may be
required and multiple analysis of fractions. The correct sample size is
the same mass that the material is being mixed for, examples include:
a pharmaceutical tablet's worth, a packet of cake mixture, a cement
sack one cow dinner – and not the entire sack! Hence, the sampling
provides an indication of how well mixed the solids are, and can be
used to follow the rate of mixing and the analysis may also be used to
predict the variation in the final product quality.
12.3 Equipment
It is possible to obtain a good mixer quality, but a poor product
See: quality as there are other operations between mixing and use, e.g.
www.midlandit.co.uk/part emptying, transportation and use, all of which may induce
icletechnology/chapter12 segregation. In a process analysis the permissible variation in the
for further information on product quality must be known and its relation to quality from the
mixer types and links to
mixer deduced. Batch mixing large quantities (up to 2 tonnes)
suppliers web sites
reduces labour costs but as size goes up so does time to reach desired
quality, filling and emptying times per batch. Continuous mixing
depends on metering rates, capacity of mixer, axial and radial
dispersion performance. Loading two, or more, components together
should reduce batch mixing time, but requires metering or practice. A
brief equipment type description follows.
Rotating shapes: tumbling action that induces particles to roll and fall,
material is elevated beyond its angle of repose and it falls to the free
surface. Good mixing of free flowing materials (tea, seeds, etc.). Easy
cleaning and emptying and power consumption and wear are low to
moderate.
Ribbon blade: an agitator mixes material in a trough. Reasonably
gentle mixing but with shear and impaction. Not suitable for very
cohesive materials, unless a dough is required. Used for addition of
small amounts to larger components, but can be difficult to clean.
Orbiting screw: charge is lifted up through mixer and spread out. Can
be modified for paste use. Moderate power consumption and
reasonable to clean.
Pan mixer: blades or ploughs move through mixture with various
angles of attack. Mainly found in food and pharmaceutical use, easy
to clean but has a high power consumption.
Z-blade: two contra-rotating blades, seldom used for dry materials
Fig. 12.6 Blending by ideal for a dough. Very high power requirement and difficult to
metering correct proportions clean.
of components from storage Muller or edge runner mill: crushes aggregates and mixes well, but not
hoppers used for free flowing or too cohesive mixtures – just the right amount
of friction is needed to resist the rollers.
High speed impeller: material hits impeller and is thrown out towards
the wall. Very high power consumption, but rapid mixing.
Others include fluidised beds and air lifts, blending from hopper
(bunker) discharges and stockpile blending. Continuous mixing of
free flowing powders includes metering the solids out at a
continuous and well defined rate, such as on a conveyor belt, see
Figure 12.6. This avoids the problem of mixing, but requires precise
powder control. This is only sometimes achievable in practice. Also,
this arrangements assumes that it is acceptable to form the product
from two separate streams; e.g. mixing will take place in the bag or
by the consumer. Metering, together with continuous mixing, has
become accepted for cohesive powders, where segregation is less Table 12.1 Comparison of
likely and is sometimes used with free flowing ones. Table 12.1 continuous and batch mixing
compares batch and continuous mixing strategies. of free flowing powders:
italicised description
12.4 Cohesive powder mixing indicates preferred option
When mixing bread a lot of effort is required to stir the cohesive

mixture but it doesn't segregate. However, only small pockets of
mixture receive shear, and mixing, and the energy input is high. In
commercial systems this means a powerful motor is necessary.
Cohesive mixtures tend to be formed by fine particles and systems
with binders present. Smaller particles lead to a larger number of
particles for a given sample mass. As the number of particles goes up
the variance between samples will go down, see equation (12.1). This
is intuitive, as the limit would be to go to molecular mixing (very
small particles) where the variance is zero between samples.
However, with powders, Lacy pointed out that the situation could
arise where smaller particles actually increase the variance between
samples; if a mixture contains aggregates, say of paint pigments blue
and red in colour, then overall the mixture may be good but, if the
sample size is equal to the aggregate size, it is possible that each
aggregate could be selected. This would give a very large value for
the calculated variance. So, although the primary particle size is
small the particles aggregate and the true particle number within the
sample is just one! Clearly, some careful interpretation of the data is
required and consideration of the appropriate scale of scrutiny is
most important.
12.5 Summary
Even when fully mixed we know that there will be a difference
between samples, or products, made from a powder. This variation is
given by Lacy's or Poole, Taylor and Walls' equations. If this variation
is too great for our specification we must either supply components
separately or decrease the size of the particles to give a more intimate
mix. This may lead to powder handling problems and a strategy of
size reduction, followed by mixing, then granulation before further
processing may be required. In general, mixing is easier when
particle size and other physical properties are the same. When a small
amount of one component is mixed with a large amount of another,
we may need to grind finely and coat larger particles.
A common misconception is that to overcome poor mixing an

increase in mixing time may be applied. If segregation occurs an
increase in time will increase segregation, therefore, mixing for longer
produces a poorer dispersion. In many cases process considerations
are important as well as mixing efficiency. For example, in
pharmaceutical production it is important to be able to identify batch
numbers; hence, batch mixing is preferred. To overcome segregation,
deliberately forming a cohesive mixture by using a binder might be
performed, but this will require higher energy costs than dry powder
mixing and may result in more difficult to clean mixing machinery.
12.6 Problems
1. According to a British Pharmacopia standard, the allowable
Consult equations variation in composition of an active ingredient is ±10%. A tablet
(12.1) and (12.5), contains 10% active ingredient and 90% inert diluent. Calculate the
but you will need to minimum number of particles to each tablet so that it may be within
convert between the B.P. specification (assume the standard is based on number):
number and mass a) assume that the particles are all the same size, and
proportions. See
b) assume that the particles are size distributed so that the mean
discussion around
weight of a particle of the active ingredient is 300 µg, and the other
Figure 12.3 for a
discussion on the ingredient is 100 µg.
number out of spec c) Out of a batch of 1 million tablets how many will be outside the
tablets. B.P. specification. How could you improve the situation?
2. Three powders (A, B and C) are to be mixed in the proportion of

60:30:10 (A:B:C) by weight. Each powder consists of monosized
spherical particles of density 2600 kg m−3 and particles of diameter
100 µm, 80 µm and 50 µm for A, B and C respectively. In the
evaluation of the mixing process with respect to powder C, the
between sample standard deviations of 100 g samples were obtained
as a function of time at 20, 40 and 60 minutes, the corresponding
standard deviations were: 5.2 x 10−4, 1.4 x 10−4 and 5.6 x 10−5. Estimate
the minimum mixing time for complete randomness. What
confidence would you have in this result? Is further experimentation
necessary?
3. A mix contains two components, A and B, the fractional mass

undersize particle size distributions are (where x is in mm):
0. 5
 x 
A: N 3 ( x) =   for x ≤ 0.5 mm and
 0.5 
2
 x 
B: N 3 ( x) =   for x ≤ 0.25 mm
 0.25 
A and B are in mass proportions 2:3 and have densities 3500 and 2500
kg m−3 respectively. The specification requires that in samples of 4 g
the proportion of A should be within ±5% of the nominal value.
Comment on this.
13 Colloids and agglomeration
The processing of fine particulate materials is becoming increasingly
important and the term nanotechnology is typically used to describe any
system with particle diameters less than 1 micron. In liquid systems,
colloidal forces become significant when particle diameters are less
than 40 microns and very significant below 10 microns. At higher
particle sizes the liquid drag and particle weight forces are normally
dominant. These thresholds are, of course, material dependent and the
equations covered in this chapter should help in determining the
relative importance of the various forces. In gas based systems, some
of the forces appropriate to liquid systems may still be applied, such as
van der Waal’s attraction, but others will not: e.g. electrical double
layer repulsion. Thus, this chapter covers the forces on fine particles
(other than drag and inertia) in both systems separately.
In a gaseous continuous phase agglomeration may be performed,
which is the process of sticking particles together, either relying on
natural adherence forces to bring the particles together or by adding
binding agents. Granulation is the process of forming particle granules
that are usually large enough to flow readily and are easy to process.
Industrial granulation relies upon the formation of strong
agglomerates which are, usually, compressed or heated (or otherwise
treated) to give granules that do not easily subsequently break. Hence,
within a process it may be necessary to grind a powder to a small size,
to intimately mix the small grain sizes with another powder, and then
to agglomerate the resulting mixture into granules for the purpose of
storage and further powder processing. The granules may, eventually,
be made into tablets or a similar compressed compact. Another process
worth mentioning is the formation of engineering components from
Fig. 13.1 Electrical charge
very small particles, which after pressing form strong solid objects.
around a particle in an
Complex shaped objects with tailored physical properties, and electrolyte solution
composites, can be made by this technique. Hence, an understanding
of fine particle processing, and forces, is important to many modern
processing industries.
13.1 Forces on small particles – in liquid medium

Figure 13.1 illustrates a particle in water and the associated electric
field around it. Most mineral particle surfaces are negatively charged,
due to defects in the solid crystal lattice, chemical reactions at the
surface, slightly soluble ions dissolving from the crystal and
adsorption (and exchange) of ions from the surrounding solution.
Some biological particles may be positively charged. The charge on the
particle surface gives rise to ions of counter charge strongly bound to
the surface. The total depth of strongly bound ions to the particle is
called the Stern layer. At some distance further away from the particle
there is the shear plane, which marks the start of the diffuse layer. It is
not possible to shear the ions off the particle that are closer to it than
the shear plane. Thus, if the particle surface potential is measured
using a device involving motion, the potential will be measured at the
shear plane, and not the true potential on the particle. This is called the
Zeta potential. So, a particle in an electrolyte solution contains a
strongly bound layer and a diffuse layer of ions close to its surface.
This is known as the Double layer. It is worth noting that ions are
hydrated, often by several solvent (usually water) molecules, so the
distance of these layers is not that of the ions alone, but is a distance
dependent on the hydrated ion. The degree of hydration depends
upon the ionic strength of the solution (I), which is defined as follows
1 2
I= ∑ ci z i (13.1)
2
where ci is the molar concentration of an ion and zi is the ionic charge,
Dimensions or valency; i.e. the ionic strength is based on the sum of contributions
There is some confusion in for all ions in solution. The electrical potential at the particle surface
colloid chemistry with units (ψo) can be deduced from the Nernst equation, based on the
and dimensions. The concentration (or more correctly solution activity) difference between
recommended system for the particle surface and bulk solution. The potential at the Stern layer
equation (13.3) is: (ψd) will be less, and can even have a different charge value (i.e.
−
ci in number of ions m 3, negative or positive) to the potential at the surface, if the counter ions
−19 bound to the surface are polyvalent. The diffuse layer potential (ψ)
e is 1.6x10 Coulombs
− diminishes exponentially, as first derived by Gouy-Chapman, giving
εo is 8.85x10 12 Farad m-1
εd is typically 81 for water ψ = ψ d exp(− Κ∆z ) (13.2)
− −
kB is 1.38x10 23 J K 1
where ∆z is distance from the surface and Κ is the Debye-Hückel
so
function. For a simple electrolyte consisting of a cation of
ci is 1000 x Molarity x
concentration c1 and anion of concentration c2, this function is
Avagadro’s number
1/ 2
 c z 2e 2 c2 z 2 2e 2 
Κ = 1 1 +  (13.3)
 ε p k BT ε p k BT 
 
where e is the charge on an electron, no is the concentration of ions
Potential and energy present, εp is the electrical permittivity, kB is the Boltzmann’s constant
The potential is defined as and T is the temperature in degrees absolute. However, in SI units it is
work done in moving a unit usual to calculate the permittivity from the dimensionless permittivity
charge of the same sign as
(εd) and the permittivity of a vacuum (εo)
the surface from infinity to
that point; i.e. if a charge of ε p = 4πε d ε o (13.4)
ze is moved from infinity to a
The inverse Debye-Hückel function has the units of length and it is
point within the electric field
convenient to assume that this is the diffuse layer thickness for
surrounding the particle the
practical use. However, the real diffuse layer extends to infinity.
work done is
zeψ Equations (13.2) to (13.4) can be used to predict the electrical potential
with separation distance from a particle surface. Clearly, on bringing
which is the potential energy
two particles together the electric fields will overlap and this could
possessed by the charge at
that position.
cause the particles to repel each other. This is known as electrical
double layer repulsion.
At the same time, and if the particles are in close enough proximity,
the London-van der Waal’s attractive force will tend to pull the
particles together. This is a short range force between particles that

arises because of electrical dipole fluctuation between neighbouring Hamaker coefficient
particles. In particle force terms, the van-der Waal’s force is short It is convenient to use a
ranged: up to 0.1 µm. The potential due to van-der Waal’s attraction Hamaker constant, but
(ΨA) may be deduced from in colloid chemistry it
has been known for
AH  2(1 + H s )  Hs  some time that the
ψA = −  + ln  
 (13.5)
12  H s ( 2 + H s )  2 + Hs  value is a variable
dependent on the
where AH is the Hamaker constant for a given system and Hs is the separation distance.
ratio of the separation distance (∆z) between the particles and the Thus, it is sometimes
particle radius. So, in terms of particle diameter referred to as the
2 ∆z Hamaker coefficient,
Hs = (13.6) which is a function of
x
separation distance. A
The Hamaker constant varies with material, but is about 5x10−20 J for true analysis of this
water. In systems containing two different types of particles and a coefficient requires a
liquid, the constant may be estimated by consideration of
quantum mechanics,
H s:132 = ( H s:10.5 − H s:3 0.5 )( H s:2 0.5 − H s:3 0.5 ) (13.7)
where the liquid Hamaker constant is Hs:3 and the other two
subscripted constants refer to the two particles. Thus, Hs:132 represents
the resulting mean constant value for use in equation (13.5), etc.
Equation (13.5) is just one form of the London-van-der Waal’s
equation; there are many others depending on the assumptions made
in the solution, such as surface geometry. When the separation
distance is very small, in comparison to particle diameter, equation
(13.5) simplifies to
AH
ψA = − (13.8)
12H s
Molecular forces and valence energy are only applicable for
separations between particles in the order of less than 10 Å, which may Atomic Force
Microscopy
be applicable to particles only after the application of very high
Is a device that uses a tip
pressures. In most instances the particle separation will be in excess of
mounted on a cantilever
10 Å due to surface roughness.
to move over a solid
surface. It is possible to
13.2 DLVO and applications keep the tip and surface
There is enormous difficulty in measuring the appropriate Hamaker apart by providing a
constant. In recent years, Atomic Force Microscopy (AFM) has been constant force – hence
used for his purpose, but ensuring pure components with no surface constant distance apart.
adsorption of ions still makes this a difficult task. In many cases the The deflection on the
Hamaker constant has been used as an empirical variable in order to cantilever is a measure of
reconcile measured data and theory. Nevertheless, the equations the surface contours. The
described above can still be used for a very useful semi-quantitative AFM can also be used to
analysis of observed phenomena. The theory first expounded by measures force whilst the
Deryagin and Landau, and separately by Verwey and Overbeek tip approaches the
(DLVO) was to use the potential field around the particles, as surface from a distance
determined by equation (13.2) and calculate the repulsive potential away.
formed when overlapping the potential fields for each particle. This
provides the electrical double layer repulsion between two particles,

adding the van-der Waal’s attraction potential, or force, provides the
total interaction potential (ΨT). At constant potential, the double layer
repulsion potential is
εp x    
ψR =  2ψ Z:1ψ Z:2 ln 1 + exp(− Κ∆z )  + (ψ Z:12 + ψ Z:2 2 ) ln[1 − exp(−2 Κ∆z ]
 
8   1 − exp(− Κ∆z  
(13.9)
and at constant charge the repulsion potential is
εp x    
ψR =  2ψ Z:1ψ Z:2 ln 1 + exp(− Κ∆z )  − (ψ Z:12 + ψ Z:2 2 ) ln[1 − exp(−2 Κ∆z ]
 
8   1 − exp(− Κ∆z  
(13.10)
where ψZ:1 and ψZ:2 are the Zeta potentials on particles one and two
respectively, see Figure 13.1. Simpler forms of equations (13.9) and
(13.10) exist, depending on the assumptions made in their derivation
and application. The total interaction energy, by DLVO theory, is
ψ T = ψ A +ψ R (13.11)
The potential energy curves are illustrated in Figure 13.2. Often, the
total interaction energy is normalised (divided) by the kinetic energy
due to thermal motion: this is called the dimensionless interaction
Fig. 13.2 Potential energy energy and may be represented by
curves for attraction and ψT
repulsion as well as total (13.12)
k BT
interaction energy curve
or more correctly
eψ T
(13.13)
k BT
The potentials described above can be converted into forces on a
particle by differentiation with respect to separation distance
dψ
F= (13.14)
dz
hence a comparison of the forces influencing a particle is possible.
Plots of the dimensionless interaction energy are useful for
assessing the stability of colloid suspensions. Figure 13.3 illustrates
three cases: curve 1 has a primary minimum and a maximum, curve 2
has a primary minimum and maximum together with a secondary
minimum, and curve 3 only has a primary minimum. To destabilise a
colloid, such as in effluent treatment, the requirement is to bring the
particles together so that they form a bigger aggregate that will settle
faster and be easier to remove from suspension. Thus, the particles
Fig. 13.3 Dimensionless
need to closely approach each other for the attractive forces to stabilise
interaction energy curves to
the aggregate. A high repulsion force provides a stable colloidal
illustrate colloidal
behaviour by DLVO theory
suspension. The DLVO theory uses plots like Figure 13.3 to explain
colloid stability as follows. The explanation considers the potential
curve based on one particle surface whilst bringing the second particle
towards that surface, from right to left on Figure 13.3. In the case of
curve 3, the primary well is close to the particle surface and from
equation (13.14) the force is attractive (positive gradient) until the
approaching particle sits in the primary well, which is at a low enough
separation distance for settling as an aggregate to take place. Thus, in
case 3 the colloid in unstable as the two particles will settle as an
aggregate. For curve 2, there is a mildly attractive force until the
particle sits in the secondary minimum. In order to approach further
the particles must have sufficient energy to overcome the maximum –
and to overcome the repulsive force at separation distances less than
this maximum. Once over the secondary maximum there is a strong
force pulling the particle into the primary well. Hence, this situation
could be a stable colloid (because of the repulsive force preventing the Fig. 13.4 Clays in freshwater
approaching particle from reaching the primary minimum), or it may and estuary water salinities
be unstable because the maximum to be overcome is fairly low in and stability – conditions:
height. In curve 1 the maximum is very significant and the force, Zeta potential -30 mV
equation (13.14), is strongly repulsive: this is a very stable colloid. particle size 4 µm
−20
One practical application of this theory is to explain the occurrence Hamaker water 5x10 J
−
of silting within estuaries. Colloidal clay is carried down a river in Hamaker particle 6x10 20 J
stable suspension because rivers are freshwater and have a low ionic ion valancies 1
concentration, the electrical double layer extends to a large distance temperature 298 K
around the particles providing a strong repulsion force. This is dimensionless dielectric 81
illustrated in Figure 13.4, the 0.01 M NaCl curve. However, on meeting also:
the saline seawater the ionic concentration increases, and may reach permittivity of a vacuum
− −
0.3 M, which reduces the distance that the electrical repulsion acts 8.85x10 12 F m 1
over, equation (13.3), and the particles can approach each other. This is Boltzmann constant
− −
illustrated as the 0.3 M NaCl curve on Figure 13.4. The colloidal 1.38x10 23 J K 1
particles, therefore, aggregate together and settle out into the estuary electron charge
−19
causing silting. 1.6x10 Coulombs
The DLVO theory is a useful approach to explaining observed Avagadro’s number 6x1023
phenomena, but many of the equations are difficult to apply and have
different forms depending upon the assumptions made in their
derivation. Also, colloidal stability, or instability, may be due to
adsorption of solutes and ions onto the particle surface. So, the above
description is not a complete model of colloidal behaviour.
13.3 Coagulation
In the silting of estuaries example given above particles were brought
together to form an aggregate that is easier to remove from suspension
than the primary particles. This is an example of the process of
coagulation, and the resulting aggregate is called a coagula which is
formed by adding a coagulant (usually a metal salt) in solution. The
process of coagulation forms coagula that may be broken by the
application of shear, such as during pumping, but they will usually
reform when the shear field is removed.
Consideration of equation (13.3) results in the conclusion that the
distance over which the electrical double layer repulsion acts is
Fig. 13.5 Zeta potential of
strongly dependent on the valency of the electrolytes in solution. Thus, iron oxide colloid particles
a tertiary valent salt is more effective at causing coagulation than a
secondary, or monovalent salt. In water treatment, ferric (iron III)

sulfate, and aluminium salts, are often used for this reason. They are
also inexpensive chemicals to apply. By the addition of a salt solution,
or by the adjustment of the pH, it may be possible to effectively reduce
the electrical potential at the shear plane (Zeta potential) to zero. Thus,
the double layer repulsion term in equations (13.9) and (13.10)
becomes zero and the particles will have no repulsion force on them to
prevent aggregation by van-der Waal’s attraction. This is illustrated in
Figure 13.5, for iron oxide at various pH values. The point at which the
Zeta potential is zero is often referred to as the iso-electric point. Thus,
in the example of Figure 13.5 adjusting the pH to 9.5 should result in a
less stable colloid than is found at other pH values. Conversely, at a
pH of 5 a very stable colloidal suspension would be expected by
DLVO theory.
13.4 Flocculation
Another means by which primary particles are brought together to
form aggregates, that is easier to remove from suspension than
primary particles, is by flocculation, see Figure 13.6. In this process the
aggregation is caused by bridges from one particle to another, usually
formed by high molecular weight polymers in solution. The polymers
may be man-made, or they can be naturally occurring; examples of the
latter are found with several biological compounds. The resulting
Fig. 13.6 A floc is much
aggregate is called a floc which is formed by adding a flocculant
bigger than a particle and is
(usually a polymer) in solution. The process of flocculation forms flocs
easier to remove from
suspension, but a floc
that may be broken by the application of shear, such as during
network settles slowly pumping, and the physical bridges linking the particles do not usually
reform when the shear field is removed. Thus, unlike coagulation, the
process of flocculation is shear sensitive. So, when flocculants are used
in high shear applications, such as to assist dewatering in a scroll
discharge decanter centrifuge (Figure 8.3) a special shear resistant
flocculant is required. These are normally very high molecular weight
polymers. However, the higher the molecular weight the more difficult
it is to dissolve the polymer and it may be more difficult to disperse
the flocculant within the feed suspension to be treated.
Synthetic polymer flocculants (polyelectrolyte) are supplied in three
forms: non-ionic, anionic and cationic. The first has no net charge, the
second is negatively charged in solution and the latter is positively
charged. Hence, when treating mineral suspensions cationic
flocculants are preferred because they can more easily bind on to the
negatively charged mineral surface. For similar reasons anionic
flocculants are applied to biological suspensions. There are many
instances when flocculation follows a primary coagulation stage. The
coagulation is designed to destabilise the suspension and the
flocculation stage to then form aggregates that are even easier to settle,
Fig. 13.7 Floc bed clarifier filter, etc. Hence, anionic polymers may be applied to mineral systems
after coagulation and cationic to biological systems. So, the optimum
flocculant and coagulant dose, and strategy, is one that can only be
determined by experimental testing. The quickest way to achieve this

is by the simple settling jar test, illustrated in Figure 6.2: fast settling
times are desired. However, the optimum dose is usually provided by
the cost of flocculant purchase, rather than any physical consideration,
as these chemicals constitute a significant recurrent cost.
Conventionally coagulants and flocculants are added to the feed
launder to the thickener, however, in the floc blanket clarifier (popular in
the potable water industry) the feed is introduced, with flocculant, into
the flocculated and settling bed of solids, see Figure 13.7. Likewise the
high rate thickener applies the same principle; the clarifier is used for
dilute feeds and the thickener at higher concentrations usually found
in the minerals industry.
Fig. 13.8 Simple aggregate
13.5 Forces on particles – gaseous medium used to compare forces
Forces on particles in a gaseous continuous phase, or medium, are
important in the understanding of particle behaviour during many
operations, such as powder flow during hopper discharge and for the Table 13.1 van-der Waal’s
process of granulation. It is useful to compare the forces on a particle force compared with gravity
in a gaseous medium to that of the weight of the particle; i.e.
gravitational force. Thus, if the particle has a mass of mp the particle
weight (Fw) is
Fw = mp g
and when considering the simple aggregate situation represented by
Figure 13.8, if mpg>F then no permanent bond is formed, whereas
when mpg<F a permanent bond may be formed, where F represents the
force under consideration.
The van-der Waal’s attractive force is applicable to a gaseous
continuous phase as well as in liquids. Table 13.1 compares the ratio of
the van-der Waal force to the particle weight for particles 0.01 to 10
microns in diameter and at separation distances of 0.01 and 0.1
microns between the particles. For a separation of 0.1 µm this ratio is 1
when the particle diameter is 3 µm. In general, the force may be
important for particles less than 10 µm.
In a gaseous continuous phase an electrostatic force exists, but not
the double layer repulsion discussed earlier. The electrostatic force can Fig. 13.9 Capillary rise
be calculated from due to surface tension
q1q 2
Fes = (13.15)
πε p x 2
The moisture force is very
where q represents charge on the particles (1 and 2) and εp is the strong between particles.
electrical permittivity of the fluid medium (normally quoted relative to You may notice this when
a vacuum). This force is difficult to quantify because a particle may riding a bike on a fine
only pick up a fraction of its equilibrium charge. Typical values of gravel path after rain: the
particle diameter, when the ratio of this force to the particle weight is gravel sticks together better
1, are: 0.3 µm at 0.1% of the equilibrium charge attained, and 30 µm for than when dry.
1% of equilibrium charge attained. Hence, this force may be significant
for particles in the sub-sieve range (less than 45 µm).
The surface tension force, or moisture force, arises from the well-
known capillary pressure effect (see Figure 13.9). There is a contact
angle (θ), which is the equilibrium angle of a liquid on a surface in the
presence of another phase such as air (i.e. three phases are required:
two fluids and a solid); e.g. as with the meniscus illustrated here. The
Young-Laplace equation is
2γ cos θ
∆P = (13.16)
R
where γ is the surface tension of the liquid and R is the radius of the
capillary. The height of rise of the fluid (h) to balance this pressure is
2γ cos θ
h= (13.17)
R ρg
From a consideration of the Young-Laplace equation Newitt and
Conway-Jones (1958, Trans. IChemE) deduced the following equation
for the force due to the pendular liquid between particles:
2πγr
Fpend = (13.18)
1 + tan(θ / 2)
where r is particle radius. However, the angle formed by the meniscus
will change as liquid is lost from the agglomerate (it dries out). When
θ→0 the particle diameter with the ratio of force divided by particle
weight of one, is 4200 µm. Hence, this force is more significant than
van-der Waal’s, or electrostatic, forces for bringing particles together in
the agglomeration purposes; i.e. for particle diameters less than 4 mm
the moisture force is greater than the force due to gravity. However,
equation (13.18) is for pendular liquid bridges between the particles,
other bridges are not so strong, so some consideration of the states of
dehydration within a particle assemblage is required, see Figure 13.10.
Initially, as a droplet, the strength of the agglomerate is due only to
the liquid surface tension; this forms a weak agglomerate that may be
Fig. 13.10 Stages during the
deformed or broken easily. As some moisture is lost the capillary
formation of a granule as
agglomerate state is achieved, in which all the interstices are filled, but
liquid is removed the agglomerate strength comes from curvature of the liquid alone. As
more moisture is lost the funicular state is achieved, which still forms
only a weak structure; followed by the pendular state of agglomerate.
In the pendular state the liquid bridges are all independent of each
other and the liquid cannot move between bridges, but draws particles
together by the capillary pressure effect. It is a very strong force.
Finally, it is possible to form a dry agglomerate, possibly bound
together by van-der Waal’s or electrostatic forces but, in the absence of
solid bridges between constituent particles, the agglomerate will be
weak. However, if solid bridges have been formed, caused by
precipitation from solution, or drying of a binder, the agglomerate can
be very strong. Such dry granules may be formed when a soluble
binding agent, such as starch dries out to chemically bind the particles
together, or if the particles have been sintered together: a high
temperature causing some of the particle surface to melt and form the
bridge between particles after cooling. Strong bonds are also formed
from chemically reacting species, crystallising solids or suspended

colloidal material that dries out between the particles. Granules with a
large number of bonds per particle are strongest.
During granulation using a liquid binder it is possible to identify
several stages during the process. Initially, nucleation takes place when
a particle plus liquid meets a second particle and sticks. This is
followed by nuclei growth as other particles stick together on the nuclei.
The nuclei themselves may then stick together during a nuclei
coalescence stage. At this point the aggregate has many dendritic (i.e.
spiky) parts to its shape and these may break off during densification,
which will leave a stronger granule. Finally, layering or snowballing
may occur, during which small fragments and fresh powder now coat
granules to form new strong dense particles.
13.6 Agglomeration and granulation equipment

Agglomeration is the process of bringing particles together to form
agglomerates and it may be performed in a liquid medium as well as
gaseous. Granulation is a term usually applied to the formation of Fig. 13.11 Pan granulator
granules and is, therefore, restricted to particle agglomeration in the
gaseous phase. Granulation is required for many reasons including: to
improve powder flow properties within a process, to avoid dust
explosions and physiological hazards from fine particles, to increase
bulk density in storage, to prevent segregation, to meter a dose (e.g.
tablets) and to achieve specified particle properties such as size and
specific surface. Hence, there are many industries that use granulation
and a variety of different equipment is employed. In the mineral
industries a simple pan granulator, see Figure 13.11, open to the
environment, may be sufficient but this would not be the case for
pharmaceutical production. Enclosed granulation equipment includes
fluidised bed granulators and high shear mixers, fitted with rotating
blades that provide an optimum granule size for a given shear input
with the addition of a binding agent.
In most cases the binding agent is a liquid added to a powder in a
granulating machine (pan, drum, fluidised bed, etc.) and
agglomeration is induced to form the granules. This is the process of Fig. 13.12 Fluidised bed
wet granulation. However, granulation can also be achieved by the granulator
action of heat and pressure on powder compacts due to sintering and
this is called dry granulation. A brief summary of the main types of
equipment, and their characteristics follows. A dish or pan granulator
has been used for many years in metallurgical industries: a titled pan
rotates and fresh powder and solution (if required) are added, it may
be run continuously and gives consistent 5 mm spherical granules.
Fluidised bed granulators have powder injection into a granule
forming bed, together with solution sprayed on the bed. Granules
overflow from the bed. However, there may be problems with
explosion risk, loss of fines and weak granules. High shear mixers,
both vertical and horizontal axes machines, are available with high
power input. A rotating drum (slightly tilted for continuous use with
an outlet weir) may be used for wet and dry granulation; the latter just
Capillary pressure and using the moisture associated with the powder mix.
pore size
Powder compacts are also formed by pressing and tabletting and
Equation (3.16) can be
by extrusion of pastes. The forces considered earlier in this chapter can
rearranged to give a
method to determine the again be used to assist in the understanding of these processes, but in
pore size (d) of a these instances very high pressures and, sometimes, shear fields are
membrane, cloth or porous also employed. The equipment required for tabletting, for example, is
medium: a highly specialised mould and punch system; generating considerable
4γ cos θ compaction pressure to compress and form the tablet. Usually, many
d=
∆P punches are used simultaneously, to provide high throughput.
Hence, immersing a Another important consideration in the pharmaceutical and food
membrane, in a filter related industries is an ability to comprehensively clean the
holder, within a liquid and machinery, possibly using a Cleaning In Place (CIP) strategy. In many
slowly increasing air instances it may be possible to clean the granulator and additional
pressure in the holder until processing equipment at the same time. However, in highly regulated
air flow is observed can be
production environments it is usual for agreed procedures and
used to determine the
biggest pore size from the
protocols to be in place, under the remit of Good Manufacturing Practice
above equation. This is (GMP), which might restrict the alteration of existing operating
known as the bubble point strategies, or adoption of new equipment, or techniques. Thus, when
test. considering granulation equipment for these industries both GMP and
CIP are factors that need to be considered carefully.
13.7 Summary
Forces on particles, in both gases and liquids, exert important
influences on their behaviour in processes and processing equipment.
Particles less than 10 µm in diameter are considerably influenced by
the colloidal forces described in this chapter and, towards the lower
end of this size range, the weight force may become insignificant in
determining the particle’s behaviour. In wastewater and potable water
treatment, coagulation and flocculation are extensively used processes
for the removal of unpalatable and dangerous finely dispersed
materials. The processing of fine particles is becoming increasingly
important, but fine particles tend to stick, arch in hoppers, cause safety
hazards and are generally difficult to process. Thus, it may be
important to create fine particles for a processing stage, such as
solid/solid mixing, but this may need to be followed by granulation in
order to provide acceptable later processing performance.
Advances in laboratory analytical equipment, such as the Atomic
Force Microscope, is leading to better direct experimental
measurement of colloidal forces, which will help to refine the
theoretical models and lead to a better understanding of particle-
particle interaction.
13.8 Problems
1. Using the data supplied with Figure 13.4, write a computer
spreadsheet to investigate the influence of salt concentration and ion
valency on the separation of two particles with the Zeta potentials, and
other physical properties, assumed to remain constant.
14 Gas cleaning
In the earlier chapters unit operations that cause particles to become
entrained in gas stream were covered; examples include fluidisation,
pneumatic conveying and grinding. In the next chapter the effect of
dust in the working environment is described. In certain workplaces
removal of particles down to very low sizes and concentrations is
essential, such as within operating theatres, during fabrication of
electronic equipment and production of pharmaceutical grade
materials. Hence, a critically important subject is the removal of
particles from gas streams. The subject is significantly different from
solid-liquid separation because the fluid medium is much less
viscous than a liquid and this influences the forces that are most
relevant to the trajectory analysis.
In many industrial processes sedimentation is the primary
mechanism for particle removal from a gas stream. Thus, gas
fluidised beds have the characteristic shape illustrated in Figure 7.2:
the increase in the bed diameter leads to a decrease in the gas velocity
above the bed and particles will fall back into the fluidised bed.
However, finer particles may still be carried over with the gas, i.e.
entrained in the gas flow, and additional particle/gas separation
equipment is required. The material presented in Chapter 5 is
appropriate to calculate the sedimentation rates of particles in gases,
this chapter covers other relevant mechanisms.
14.1 Target; grade and overall efficiencies

To remove a particle from a gas stream it must be encouraged to hit a
target and then to stick to it. The gas then passes on leaving the
particle removed from it. Targets can be made from many objects
including: liquid drops, fibres, larger particles, plates and walls.
There is an enormous range of equipment based on one, or more, of
these targets. Clearly, for the particle to be removed the target needs
to have the property of retaining the particle from the flow but, if the
targets are reused it must then be encouraged to release the particle
when required during a cleaning cycle. There is often confusion Fig. 14.1 Particle collected on
between the different types of efficiencies used to describe the a target from a gas streamline
process. The simplest is the single target efficiency, which can be and single target efficiency
represented by Figure 14.1. The single target (e.g. a fibre) efficiency is
(ηs)
r
ηs = c (14.1)
rt
where the particle is collected by inertial separation: the gas can
easily change direction and does so round the fibre, but the particle
has much greater inertia and will continue moving towards the fibre,
142 Gas cleaning
possibly contacting it. If it does so, and sticks, it will be removed from
the gas stream. All particles of a critical size starting from r=0 to r=rc
are captured, particles starting their trajectory from r=rc to r=rt will
not be collected. Clearly, rc=f(x) or more accurately a function of the
inertia provided by the particle and fluid system.
The grade efficiency is the collection efficiency for a given particle
size range and would be calculated by dividing the mass of particles
in a grade retained in the device compared to the mass entering the
dust collection equipment for the size range. In most cases, it is easier
to remove larger particles; hence, the grade efficiency curve for dust
collection is at, or approaching, 100% for large particles and lower for
finer particles. An example grade efficiency curve is provided in
Figure 14.2.
The grade efficiency considers collection of particles in as many
grades as are defined, but the overall collection efficiency is a single
value that represents the total mass retained compared to the total
mass challenging the dust removal equipment. Hence, it may be
deduced using the grade efficiency curve if the total dust mass flow
rate entering the device is known. The total mass flow rate of dust
leaving can be calculated and the overall efficiency is the total mass
Fig. 14.2 Grade efficiency retained divided by the total mass entering.
– note a different
definition from a classifier: 14.2 Collection mechanisms
here the intention is to
remove particles hence a Considering Figure 14.1, the dust collection mechanism was inertial
good grade efficiency separation. The inertia is a property of the system, and not just the
occurs with the larger particle; the significance of the direction change of the gas, viscosity
particles. Efficiency is still of the gas and velocity of the gas entraining the particle all influence
based on a size range – or the likelihood of the particle inertially separating from the gas
grade. streamline and hitting the target. Hence, a measure of the inertia of a
system needs to contain all these parameters and it is called the
Stokes Number (Stk)
xu g ρ s
Stk = (14.2)
9 µD
where it is assumed that the particle starts its trajectory at the same
speed as the gas (ug) and D is a characteristic dimension of the system.
In the case of Figure 14.1 the obvious dimension would be the fibre
diameter. The Stokes number is dimensionless and is not to be
confused with the Stokes’ settling velocity.
Another very common dust collection mechanism is diffusional
collection. This is illustrated on Figure 14.3 and is more relevant to
the capture of very small particles. The dust is subject to
bombardment by molecules of gas and, as the particle is small, the
Fig. 14.3 Particle momentum of the gas can induce particle motion (Brownian motion).
captured by diffusional
Hence, it is possible for a dust particle to be small enough to follow
mechanism on a fibre
the gas streamline around the target, i.e. not collected inertially, but it
may be pushed onto the target after it has passed it because of

molecular bombardment, as illustrated in Figure 14.3.
Inertial collection mechanisms are important with larger particles
and diffusional collection for smaller; hence, there is a minimum on a
plot of target efficiency against particle size, Figure 14.4. Other
mechanisms that are important include: bounce (re-entrainment),
sieving/straining, sedimentation, electrostatic and thermophoretic
forces. Bounce occurs when the particle has sufficient inertia so that
the forces attempting to retain the particle on the target are
insufficient to do so and the particle bounces off.
The Stokes number can be explained by considering a force
balance in the axial direction (z axis or direction), considering just the
forces of inertia and drag. Calling the particle and gas velocities: Up
and ug respectively and using the Stokes drag expression is valid
dU p
F = ma = m = 3πµx (u g − U p ) (14.3) Fig. 14.4 Different
dt
collection mechanisms
Using the product of volume and density for mass, and expanding and particle size range
particle velocity gives
x 2 ρ s d 2 z dz
+ − ug = 0 (14.4)
18µ dt 2 dt
The above equation can be rendered dimensionless as follows:
z ug U
z* = U* = τ= ot
rt Uo rt
where Uo is the mainstream (initial) gas velocity, hence
x 2 ρ sU o d 2 z * dz *
+ −U* = 0 (14.5)
18µrt dτ 2 dτ
which is a dimensionless equation where the Stokes number is
x 2 ρ sU o
= Stk (14.6)
18µrt
The difference between equations (14.2) and (14.6) is in the
characteristic dimension, which is based on a diameter and radius
respectively. Hence, the associated constant is18 for a radius and 9 for
a diameter. The Stokes number is very important in modelling and
scale-up of equipment sizes because, if the collection mechanism is
primarily due to inertia, there will be a unique function between
particle collection efficiency and the Stokes number, see Figure 14.5
for an example. Thus, if this relation is known it is possible to predict
the collection efficiency for one set of particles from a knowledge of
efficiency for another set of conditions. This is used in the scale-up of Fig. 14.5 For inertial collection
only – a unique relation
gas cyclones, where the collection efficiency is transposed between a
between Stokes number and
test cyclone and an industrial unit, see Problem 6. collection efficiency
Another parameter used to characterise the inertia of a particle is
the stop distance, which can be considered to be the distance required
144 Gas cleaning
for a particle to come to a halt when it is injected into a stationary gas

at the velocity Uo. Again the only forces considered here are inertia
and drag. The particle weight is insignificant. Integrating the
dimensional force balance, equation (14.4), provides an expression for
the particle velocity as a function of time (assuming ug=0 and Up=Uo
as a boundary condition). Expanding velocity into distance with time
then permits the integration to be repeated providing an equation for
the stop distance (zs)
x 2 ρ sU o
zs = (14.7)
18µ
i.e. the stop distance is the product of the Stokes number and the
target radius (or characteristic linear dimension).
Regardless of the mechanism bringing about the collection it is
possible to conduct a critical trajectory analysis. For laminar flow
conditions, this procedure is very similar to clarification and is
illustrated in Problem 2. However, gas flow rates are usually very
high and the resulting fluid flow is normally turbulent. A critical
trajectory analysis is still possible, but must be conducted within the
boundary layer outside of the turbulent flow. The trajectory is
illustrated in Figure 14.6. The time taken to travel radially is again
equated to the time taken to travel axially, albeit very small amounts
δy δz
Up = and u g = Thus, equating the times and rearranging
δt δt
Up
δy = δz (14.8)
ug
The proportion of particles removed from a system (-dN/N) is simply
Fig. 14.6 Critical trajectory
within a boundary layer assumed to be equal to the ratio of the volume from which they are
captured, compared to the total volume (where W is channel width)
dN Wδyδz U p δz
− = = (14.9)
N WHδz ug H
The negative sign is required for particle removal. The particle
velocity towards the wall could be due to electrophoresis (in an
electrostatic precipitator), thermophoresis (due to a temperature
gradient such as in a chimney) and simple gravity sedimentation.
Regardless of the mechanism for deposition the mathematical
analysis is identical. Integrating equation (14.9), under the boundary
conditions that N=No at z=0 and N=N at z=L provides
N  UpL 
= exp −  (14.10)
No  ug H 
 
Equation (14.10) provides an expression for the fraction of particles
remaining in gaseous suspension at a distance L down the dust
collection device, with turbulent flow. The expression for particle
collection efficiency (η) is one minus this value
N  UpL 
η =1− = 1 − exp −  (14.11)
No  ug H 
 
Equation (14.11), when applied to dust collection by means of an
electrostatic precipitator, is known as the Deutch Equation.
14.3 Dust collection material balance Fig. 14.7 Fibres in a

differential slice within a
A material balance can be written for particle collection in a system,
HEPA filter
such as capture in a fibrous filter designed for high efficiency
particulate air filtration (HEPA). These are typically thick fibrous pads
of greater than 95% porosity that capture particles by diffusion.
However, the following analysis can be adopted to any system
involving a target, such as a spray tower. The fibrous filter is
illustrated in Figure 14.7 and a material balance is as follows
mass input - mass output = accumulation
Using αf for the packing density of fibres, i.e 1 − ε = α f

the volume of fibres in height dL is: α f AdL
The length of fibres in dL is fibre volume over fibre area, i.e.
α f AdL
(π / 4) d f 2
where df is the diameter of the fibre. The projected area to the gas
flow is the product of the length and diameter of the fibre, thus
4α f AdL
πd f
Now it is generally true that the mass of dust removed in a Fig. 14.8 A spray tower
differential layer per unit time, i.e. the accumulation, is the product of for dust removal
(SI units of kg s−1)
Interstitial x Projected area x Mass concentration x Efficiency of
velocity of target of the dust collection
Ug 4α f AdL
. .Cρ s .η s
1−αf πd f
where η s is the single target (fibre) collection efficiency - NOT the
overall collection efficiency.
Now, from a mass balance:
rate of dust input into layer is Cug ρ s A
 ∂C  Fig. 14.9 Target efficiency

rate of dust output from layer is Cu g + u g ∂L dL  ρ s A
  within a spray tower
hence accumulation is −u g dCρ s A

taking a fixed instant in time so that the partial becomes a full
differential and equate with above expression for accumulation to
give
146 Gas cleaning
dC 4η sα f dL
− =
C πd f (1 − α f )
Integrate from C=Co at L=0 to C=C at L=L to give
C  4η sα f L 
η =1− = 1 − exp −  (14.12)
Co  πd f (1 − α f ) 
note single target efficiency varies according to Figure 14.4 and
usually has a minimum at approx 0.4 µm.
Similarly, for a spray tower (Figure 14.8), a mass balance on dust
entering and leaving a layer of depth dz in a spray tower gives
∂N
− ug A dz = accumulation
∂z
where u g is the superficial gas velocity, A is the cross-sectional area
and N is the dust concentration (kg m−3). Using
mass flow rate x target area x efficiency of capture
per unit area (or target efficiency)
The mass flow rate per unit area is: Nu g /(1 − α s ) , the projected area of
the spray drops is: Aα s dz 3 / 2 x and the target efficiency is: η s , where
α s is the volume fraction of the liquid in the tower and x is the liquid
Fig. 14.10 A gas cyclone
droplet size.
Combining the equations for accumulation and integrating under
appropriate boundary conditions provides the following expression
for the overall efficiency (η), based on the tower height (H)
 3α sη s H 
η = 1 − exp −  (14.13)
 2(1 − α s ) x 
Equation (14.13) shows that the collection efficiency for the tower
depends on the liquid droplet diameter and the target efficiency. The
latter is dependent on the dust particle diameter, but it is generally
found that the optimum liquid drop diameter is approximately the
same for all dust diameters and is about 600 µm. This is illustrated in
Figure 14.9. Rain has a diameter close to this value; thus rainwater is
efficient in removing particles suspended in air.
14.4 Equipment types

The principles of operation of many of these devices were discussed
in the last section. In essence, the dust particles must be made to
travel to a surface in the device and to stick at that surface. The
equipment types are described very briefly in the following.
Settlement chambers often contains multiple plates, may be called
louvre collector. Coarse particles may be collected even in turbulent
flow. Centrifugal collectors such as a gas cyclone (Figure 14.10): inertial
Fig. 14.11 Venturi scrubber separation between the particles and the carrier gas causes particles
(of greater inertia) to be thrown out of the gas stream and collected
on a surface, wall, etc. Scrubbers capture particles on liquid droplets

(spray tower) or liquid surfaces (packed or tray tower). A venturi
scrubber achieves capture in the throat of a specially designed venturi
(Figure 14.11). The particle collection mechanism is primarily inertial
interception. Fibrous and bag filters capture particles on a fibre. The
filter has a very high porosity (>90%), hence a low pressure drop. A
significant depth may be required (often 5 cm). Bag filters have many
(thinner) bags in parallel. Principle collection mechanisms are inertia
and diffusion. Gravel bed filters capture particles on larger ones, which Fig. 14.12 Principle of an
then need to be taken off line and cleaned - similar to a deep bed electrostatic precipitator –
filter. Principle collection mechanisms are inertia and diffusion. the external plate is usually
Electrostatic precipitators use electrostatic forces to induce particle at Earth’s potential to avoid
electric shocks
motion onto a collection surface, see Figure 14.12.
Table 14.1 Industrial Gas Cleaning Devices and Mechanisms
Equipment Collection efficiency (%) at following sizes:

50 µm 5 µm 1 µm High Relative
temperature cost*
Inertial collector 95 16 3 yes 1
Medium efficiency cyclone 94 27 8 yes 3
Low resistance cellular cyclone 98 42 13 yes 2
High-efficiency cyclone 96 73 27 yes 4
Impingement scrubber 98 83 38 no 7
Self-induced spray deduster 100 93 40 no 5
Void spray tower 99 94 55 no 11
Fluidised bed scrubber >99 99 60 no 8
Irrigated target scrubber 100 97 80 no 6
Electrostatic precipitator >99 99 86 yes 9
Irrigated electrostatic precipitator >99 98 92 no 13
Flooded-disc scrubber - low energy 100 99 96 no 10
Flooded-disc scrubber - medium energy 100 >99 97 no 15
Venturi scrubber - medium energy 100 >99 97 no 14
High efficiency electrostatic precipitator 100 >99 98 yes 16
Venturi scrubber - high energy 100 >99 98 no 18
Shaker type fabric filter >99 >99 99 no 12
Reverse jet fabric filter 100 >99 99 no 17
Ceramic filter elements 100 >99 >99 yes 18
*relative cost per 1000 m3 of gas treated - the lower value the better
Figure 14.12 illustrates an electrostatic precipitator, where the voltage

difference between two electrodes is several thousand but, as the air
is non-conducting, the current flow is minimal. Hence, they are
energy efficient devices to induce particle motion towards the
collection plate, which also acts as an electrode. Dust enters the
precipitator and is subject to a high potential gradient and picks up
148 Gas cleaning
an electric charge. It then moves towards the collecting electrode and

deposits. Periodic removal of the dust is important in order to
prevent the charge on the dust reversing; leading to re-entrainment of
it in the gas flow. Their use is restricted to dust that has the right
electrical conduction properties: not too low or a charge will not be
induced and not too high as charge reversal will take place too easily.
They have very large throughputs and low pressure drops and are
very common on power stations. Table 14.1 compares some of the
operating conditions for several devices.
There is considerable interest in the treatment of hot gases,
without having to use techniques that require the temperature to be
reduced. Treatment with a water based system would require
temperature reduction to significantly below 100oC before contact, to
avoid excessive flashing of the water at atmospheric pressure. If a hot
temperature can be maintained the gas plume will rise further from a
chimney, which will help dilute and disperse any discharges made
from the chimney. This is preferable to having any discharge leave
the chimney and immediately sink back to the ground. The same net
amount of pollution will be provided in either scenario, but dilution
and dispersion is preferable to a high local concentration. Hence,
techniques such as electrostatic precipitators, which do not require a
reduction in the gas temperature and can process very large gas
flows, have become commonplace at fossil fuelled power stations.
Another consideration is in the need to avoid secondary pollution.
Thus, spray towers may be adequate for low temperature dust
removal, but it may be argued that they solve a gas/solid pollution
problem and create a liquid/solid pollution problem instead. Hence,
additional equipment is required for solid/liquid separation if spray
towers are used.
14.5 Summary
There are many different types of gas cleaning devices available for
both high throughput industrial use as well as small scale high purity
applications. The latter usually rely on high efficiency particulate air
(HEPA) filters that may be several centimetres thick, but posses a
very high porosity between the fibres. Deposition takes place within
the fibre matrix. For high throughput industrial use, such as mineral
processing and fossil fuel power generation, gas cyclones may be
used as a pre-treatment technique because they are maintenance free
and have a low pressure drop followed by electrostatic precipitators
or special design filters. Ceramic candle filters are becoming popular
in high temperature applications, such as foundries, as they can
withstand corrosive gases, abrasive particles and high temperatures.
It is also worthy of note that the human body provides efficient
particle removal surfaces in both the nose and upper respiratory
tract: nasal hair helps capture particles by inertial impact and
diffusion and further inertial collection takes place in the throat and
whenever the air flow changes direction before entering the lungs.
14.6 Problems Separation span

For the separation size and
1. 1100 kg of powder was fed at a uniform rate to a classifier and 583 sharpness of separation see
kg was recovered in the fine product. The cumulative undersize Figure 14.12 the separation
particle size distributions were determined by sieving, as given in x 25% − x 75%
two columns of the following table. Complete the grade efficiency span is:
x 50%
table, sketch the grade efficiency curve and find the separation size.
Particle Weight % undersize Mass in grade Mass in grade Grade

diameter Feed Fines Efficiency
(µm) (%) (%) (kg) (kg) (%)
850 98.2
600 95.5 100
420 91.8 99.7 40.7 1.749 4.3
300 85.9 98.2 64.9 8.745 13
210 75 92.6 119.9 32.648 27
150 47.7 71
100 24.5 42.5 255.2 166.155 65
75 8.2 15.1 179.3 159.742 89
53 1.8 3.4 70.4 68.211 97
2. See the box and diagram for details on this question.

The settling chamber below is
i) The particle settling velocity (m s−1):
to be used to collect particles
a: 0.00303 b: 0.151 c: 1.51x10−6 d: 0.00272
50 µm diameter and 2000 kg
ii) The residence time horizontally is, where n is the number of m-3 density from a stream of
channels, (s): −
10 m3 s 1 of standard air*.
a: 1.5WL / Q b: 1.5WL / nQ c: 1.5WLn / Q d: 1.5WLn / nQ
The chamber is 1.5 m high
iii) The residence time vertically is (assuming settlement over the full and wide. Use Stokes’
height of a channel) (s): settling equation and*:
a: 110000 b: 61.2 c: 1.10 d: 55.1 µ of 1.8x10−5 Pa s, and
iv) Using a critical trajectory model, the length of the chamber for ρ of 1.2 kg m−3.
100% collection efficiency at 50 µm is (m):
a: 20 b: 10 c: 5 d: 2
v) The characteristic linear dimension to use in a flow Reynolds
number calculation is the hydraulic mean diameter which is 4x the
area open to flow divided by the wetted perimeter. The hydraulic
mean diameter of the above settling chamber system is (m):
a: 0.3 b: 1.5 c: 0.60 d: 0.33
vi) The flow Reynolds number is:
a: 440000 b: 180000 c: 99000 d: 89000
vii) The number of trays required to bring the flow into the laminar
regime is:
a: 644000 b: 385 c: 175 d: 36
150 Gas cleaning
viii) Assuming 100% collection of 50 µm particles, what will be the

collection efficiency (%) for 25 µm particles?
a: 100 b: 50 c: 25 d: 0
3. Please refer to Question 2. The flow through the 9 channel settling

chamber is turbulent but we can assume that the particles are
deposited at their terminal velocity within the laminar boundary
layer. This is illustrated by the accompanying figure.
i) Using the critical particle trajectory approach the boundary layer
height ( δy ) is:
Ut ug ug
The figure represents the a: δz b: δz c: d: δz
flow on one of the tray
ug Ut Ut
surfaces. The critical ii) The most appropriate equation for the fractional volume inside the
particle trajectory laminar layer compared to the total volume of the flow channel is -
approach (residence times where W is the channel width is:
horizontally and δyδzW δy u g δz U t δz
vertically are equal) may a: b: c: d:
HδzW H Ut H ug H
still be employed, but the
analysis has to be Note that the answers to parts (i) and (ii) could be written as
restricted to the laminar differentials due to the linear relation between δy and δz inherent in
boundary layer. The the critical particle trajectory model. You will need to integrate the
following symbols are differentials in the following part.
used on the figure: ug is iii) Using N for the number of particles, the fractional number of
the horizontal gas and particles removed in δx is -dN/N, equate this with the fractional
particle velocity, Ut is the volume given in part (ii), and write down the equation for the
terminal particle velocity number of particles left in suspension at a distance L in the channel.
in the boundary layer and
H is the channel height. Ans: N=No...
iv) The fractional number of particles still in suspension at distance L

is N/No, hence the fractional particle REMOVAL or efficiency (η) is:
Ans: η = 1 − ...
where No is the initial number of particles at L=0.

v) The gas velocity along the settling chamber (i.e. axially) is (m s−1):
a: 3.33 b: 4.44 c: 6.67 d: 17.8
vi) The length of settling chamber required to remove 99% of the 50 µ
m particles is (m):
a: 23 b: 90 c: 34 d: 17
vii) The settling chamber length required to remove 100% of the 50 µ
m particles is (m):
a: 46 b: 64 c: ∞ d: 100
viii) The removal efficiency for 25 µm particles using the length from
(vi) is (%):
a: 25 b: 58 c: 68 d: 78
4. Please refer to Question 3, you will need to apply the same logic:
the flow up a chimney is turbulent but we can assume that particles
are deposited at some velocity within the laminar boundary layer.
i) Using the concept of the fraction of particles removed (-dN/N) is
equal to the fraction of the volume occupied by the differential ring dr
within the slice dz, over which the capture takes place, as you did in
the previous tutorial, derive an equation in which N/No=f(ug,r,L,Q),
where Q is the axial air flow rate and L is the chimney height.
N=No...
ii) The thermophoretic velocity of a particle in a temperature gradient
Considering a chimney
may be assumed to be
of cylindrical symmetry
−0.036 dT
Up = the radial (Up) and axial
T dr (ug) velocities are
where Up is velocity in cm s−1, z is distance in cm and T is temperature dr dz
in degrees absolute. The temperature gradient at the wall of a Up = and u g =
dt dt
chimney is 315oC per cm, and the temperature of the bulk gas is
1000oC (assume T=1273 K). The drift velocity towards the wall is
(feet s−1):
a: 0.00037 b: 0.00055 c: 0.00029 d: 0.00069
iii) The chimney is 3 feet in diameter and 50 feet tall, and gas is
flowing at 500 cubic feet per minute (cfm). The inlet dust
concentration is 1 milligram per cubic foot, the outlet concentration is
(milligram per cf) - NB convert cfm to cubic feet per second first:
a: 0.984 b: 0.967 c: 0.970 d: 0.979
iv) The dust mass deposited on the chimney wall over 28 days is (g):
a: 0.66 b: 13.8 c: 330 d: 655
The deposit thickness ½ way up chimney - using above physical data:

v) The dust concentration ½ way up the chimney (milligram per cf):
a: 0.984 b: 0.992 c: 0.985 d: 0.990
vi) The equation for mass of solids deposited per unit time is:
dr dr
a: ρ s 2πrdz (1 − ε ) b: ρ s 2πrdz c: ρ s (1 − ε )dr 3 / dt
dt dt
where ε is the deposit porosity, dr is the deposit thickness and ρ s is
the solid density.
vii) The mass of particles removed from the gas stream AT THE
SAME TIME is:
dz dN dN dN
a: πr 2 ρ s b: 2πr 2 dz c: πr 2 d: πr 2 dz
dt dt dt dt
viii) A mass balance in layer dz within the chimney provides the
following result
dN dN
− ug =
dz dt
and the differential form of your answer to Part (vi) gives another
equation for -dN/dz, combine these two equations and your answers
152 Gas cleaning
to (ii) and (iii) to give an equation for the rate of increase in deposit
thickness with time dr/dt=f[ug, N (at z=0.5L), ρ s , (1- ε )]:
s.g. is specific gravity dr
– or relative density:
= ...
dt
relative to water at ix) The dust s.g. and deposit porosity are 2.6 and 0.5, the deposit
−3
1000 kg m thickness after 28 days is (µm):
a: 2.9 b: 5.8 c: 11 d: 5760
5. Spray tower
i) A spray tower reduces the concentration of a dust emission from
Combining the equations 0.009 to 0.00135 g m-3, the overall collection efficiency is (%):
for accumulation and a: 85.0 b: 82.4 c: 15.0 d: 17.6
integrating under 3
ii) The spray tower uses 1.5 m of water per hour and the residence
appropriate boundary time of a drop in the tower is 30 seconds, the volume of water in the
conditions provides the
tower at any instance is (m3):
following expression for
the overall efficiency ( η )
a: 0.75 b: 1.5 c: 0.0125 d: 0.0083
iii) The tower is 15 m high and 5 m in diameter, the volumetric
 3α sη s H  concentration of the liquid drops is (-):
η = 1 − exp − 
 2(1 − α s ) x  a: 0.0051 b: 0.0025 c: 4.2x10−5 d: 2.8x10−5
iv) The liquid drops are 150 µm in diameter; target efficiency is (%):
a: 1.0 b: 1.1x10−4 c: 30 d: 85
iv) How could the overall efficiency of the tower be improved?
ANS (include a sketch of efficiency against drop diameter if
necessary):
6.
Geometrically similar Air Classified (AC) fine test dust was used in a model 8 inch diameter
cyclones can be gas cyclone and the following collection efficiencies were obtained
compared or scaled when operating with an inlet velocity of 18 m s−1
using the stokes number:
 x 2ug ρs  Particle diameter (µm): 1 3 5 7 9 11 22
Stk =  
Collection efficiency (%): 20 45 60 72 79 84 100
 9 µD 
 
i.e. collection efficiency is The specific gravity of the dust was 2.6 and the viscosity of the gas
a unique function of
was 1.8x10−5 Pa s.
Stokes number. If the
Stokes number is
An industrial cyclone of similar geometry to the model but 5 ft in
maintained constant then
diameter is to be used to remove dust of specific gravity 1.8,
the collection efficiency
will also be a constant. contained in a gas stream at 120oC - viscosity 2.2x10−5 Pa s, with an
inlet velocity of 66 feet per second. The size distribution of the
industrial dust is provided in the figure on the next page.
i) Calling the model cyclone test StkA and the industrial application
StkB fill in the missing terms in the following (using subscript A and
B where appropriate):
Stk A    
= unity =   /  
Stk B    
ii) Now rearrange the above equation to provide the appropriate

TRANSFORM in order to change the particle size in system A into
that in system B which provides the same efficiency, i.e.
( )
 
x B =   x A
 
iii) The transform value, i.e. the constant of proportionality between
xB and xA, used to convert particle diameter is:
a: 0.46 b: 1.90 c: 2.82 d: 3.45
iv) The inlet concentration of the dust is 200 g m−3, estimate the dust
concentration in the emission by completing the following table
xA Efficiency xB size range N 3 ( xB ) n 3 ( xB ) Mass in Mass not

η grade collected
(g m−3) (g m−3)
1 20 3.45 0 to 6.9
3 45 to
5 60 to
7 72 to
9 79 to
11 84 to
Totals:
v) The overall collection efficiency is (%):

a: 60 b: 70 c: 80 d: 90
vi) If the pressure drop in the model cyclone was 4.5 inches of water Most of the pressure drop in
gauge (WG) the pressure drop in the industrial cyclone will be (WG) a cyclone is due to setting up
a: 3.6 b: 4.0 c: 5.0 d: 5.6 the centrifugal head,
2
vii) Increasing the temperature of the inlet stream will alter the gas i.e. ∆P ∝ ρu g
viscosity. If the gas is air the viscosity will: The proportionality constant
a. INCREASE b. DECREASE is the number of velocity heads.
viii) By considering the terms in the Stokes number, write below The relation is valid for
some ways that may be used to improve the collection efficiency of geometrically similar
the cyclone: cyclones, i.e. independent of
1) cyclone diameter.
2)
3)
4)
154 Gas cleaning
7. See the diagram of a tube within an electrostatic precipitator.

i) Using the critical particle trajectory model (residence times in the
differential layer are equal) the distance dr is:
dr =
ii) Now using the concept of the fraction of particles removed (-dN/N)
is equal to the fraction of the volume occupied within dL by the
Modelling is again based
differential layer dr, i.e. the part over which capture takes place (as
on the assumption that
you have done in previous questions) derive an equation in which
the flow is turbulent with
N/No=f(ug,R,L,w), where L is the tube length:
particles deposited at
some velocity within the
N=No...
laminar boundary layer.
Considering a tube the
iii) An electrostatic precipitator consisting of 100 tubes each of length
axial (ug) and radial (w) 5 m and internal diameter 10 cm is operating on a power station flue
velocities are gas. The total gas flow rate is 300 m3 min−1. The axial velocity in the
dL dr precipitator is (m s−1):
ug = w=
dt dt a: 1.6 b: 0.11 c: 3.2 d: 6.4
iv) Tests show that the collection efficiency for 5 µm particles is
99.10%, the radial, or drift (w), velocity of particles of this diameter is
(m s-1):
a: 0.0003 b: 0.15 c: 0.3 d: 0.6
v) The force due to an electrostatic field is:
pEC E P x 2 / 4
where p is a constant dependent on the dielectric properties of the
system, x is the particle diameter, EC and EP are the charging and
precipitating electric field strengths. Ignoring all forces other than
hint Q7.(v)
Use the fluid drag
electrostatic and fluid drag, the drift velocity is:
expression given by pE C E P x 2 pE C E P x 12µπ 12 µπ
a: b: c: d:
Stokes 12 µπ 12 µπ pE C E P x pE C E P x 2
vi) The collection efficiency of particles 2 µm in diameter is (%):

a: 99.1 b: 84.8 c: 77.2 d: 92.5
8. Assuming the axial flow through the above precipitator is laminar,

This question is solved rather than turbulent, how long would the precipitator need to be to
by a similar technique achieve the same collection efficiency for 2 µm particles? Use the
used for centrifuges - physical and flow data given above, and answer in metres:
take r1=0 a: 0.6 b: 1.3 c: 3.3 d: 5.2
15 Powder hazards
Some of the hazards posed by powders should be obvious and several
aspects have been covered in earlier chapters; e.g. powder floods in
Chapter 10. From an operating point of view, the strongest motivation
for the prevention of a powder hazard is the health and safety of the
process operators and those in the surrounding environment. Health
and safety legislation varies from country to country, and changes
with new legislation and directives, so the following references to
standards and procedures should be viewed only as an introduction to
this topic. The main hazards posed by powders have been split into
two: explosion within a process and personal health hazards.
15.1 Explosion hazards

Eighty percent of organic dusts have been found to be explosive, as are
many very fine metal powders, e.g. magnesium and aluminium. Dust
explosions are commonplace and it has been estimated that on average
there is one accidental explosion per week within the UK. The main
distinction between a vapour explosion (detonation) and dust is the
lower flame velocity found in the latter case. Dust explosions are
usually deflagrations; i.e. the flame speed is less than sonic velocity and
usually ranges from a few m s−1 to low 100’s m s−1, sonic velocity is
approximately 330 m s−1. For explosion five conditions must be met:
• dust must be suspended in air or gas supporting combustion,
• must have a particle size capable of propagating a flame,
• dust concentration must be in the explosive range (the lower
threshold ranging from 20 to 50 g m−3 for most dusts),
• must be above minimum ignition temperature – but this may
be achieved in various ways, so use this only to compare
dusts, and
• there must be an ignition source of sufficient energy (which
may locally provide the heat for the last point).
The lower flame speed of a deflagration means that there is a better
chance of pressure venting to control an explosion with powders,
compared to vapour phase detonations.
Particle size is important and, in general, smaller particles are more
likely to be explosive. They are also more likely to become airborne. In
many cases a dust explosion is followed by a secondary explosion that
can have even greater force than the first. This is due to finer particles,
which previously rested on ledges and floors, becoming airborne in
the first deflagration and creating the secondary explosion. Hence,
good housekeeping by minimising particles on ledges and floors is
important. In the UK powders have been classified into two: Group A
and B. Group A is deemed to be the most dangerous and there are
standard tests to determine in which group a powder belongs.
156 Powder hazards
However, even the safer Group B powders may be combustible and

present a fire hazard.
The rate of pressure rise due to a powder explosion can be
estimated by the cube root law, which relates the rate of increase in the
pressure to a vessel volume (Vv) and a material dependent constant
(KST in bar m s−1)
 dP  1/ 3
  V v = K ST (15.1)
 d t  max
Nomographs relating the vessel volume, value of KST and required
vent area, based on the above equation, are available – see The Chemical
Engineer, January, 1989, pp18 – 21 and ‘Guide to dust explosion prevention
and protection’ – Part 1, IChemE, Rugby, 1984.
15.2 Physiological Powder Hazards

High concentrations of dust in air can cause erosion of tooth enamel,
Table15.1 Typical air
abrasion of skin, etc. Powders that can be absorbed into the body and
respiratory requirements
for a human give rise to chemical, or biochemical, reaction are potentially very
dangerous. However, the most significant powder hazards are due to:
• skin – dermatological disease such as dermatitis, which is
possible to prevent with the use of barrier creams and
protective clothing,
• eyes – cause soreness but not usually permanent damage
unless powder dissolves and chemically attacks the eye, and
• lungs – the respiratory system.
Most concern has focussed on the influence of particles on the
respiratory system as this can cause both acute (quick acting) and
chronic (over a long time period) health effects. Table 15.1 provides a
guide to the amount of air required by a person under different
conditions at 20oC in litres per minute.
A typical manual labourer will use 1 to 3 m3 hr−1. Hence, the
concentration of an airborne dust in mg m−3 can be used to estimate the
intake of dust to the lungs. Not all the airborne dust will make it into
the mass transfer region of the lungs (the alveoli), only particles in the
respirable range. This particle size range is illustrated in Figure 15.1,
where the diameter is the aerodynamic diameter; i.e. the particle size that
has the same settling velocity in still air as a particle of relative density
of one (i.e. 1000 kg m−3). The remaining dust is removed by the body’s
natural defences against foreign material.
Particles above this range are removed by: impingement on nasal
hairs, back of the throat and splits in the respiratory tract, see Chapter
14 for a more comprehensive description of the appropriate
mechanisms. Particles in the respirable range may be deposited in the
alveoli by sedimentation in the very slowly moving air, or diffusion
onto the surfaces. Particles below the respirable range may be exhaled
Fig. 15.1 The respirable range
after entering and leaving the lungs. Figure 15.2 provides a large-scale
schematic representation of the respiratory tract with data on air flow
and particle size of particles that may be found deposited in the
regions. The body has several mechanisms for dealing with particles
including:
The cilia, which are very thin hairs up to 4 µm long that line bronchi
and trachea and catch foreign bodies in the respiratory system.
Trapped particles are covered in mucus and passed up into the throat
where they are swallowed, sneezed or spat out; N.B. cilia are
destroyed at an early stage of smoking induced lung cancer, causing
mucous to accumulate in the air passageways and lungs resulting in
smokers cough.
Phagocytes are cells that surround the particle and reduce its irritation.
They can take the particle to the bronchioles, and out by the cilia, as
above, or into the blood stream and eventually excreted, or into the
lymphatic system, and possibly back into blood.
Fig. 15.3 Human head and

dust sampling strategy
Fig. 15.2 Schematic diagram of the human respiratory tract
In the UK, the threshold limit value (TLV) applies to the dust in the
respirable range and were published by the Health and Safety
Executive (HSE), but the Control of Substances Hazardous to Health,
COSHH, act 1988, introduced the Maximum Exposure Limit (MEL) and
158 Powder hazards
the Occupational Exposure Standard (OES). For monitoring and

protection both standards include such phraseology as: is reasonably
practicable, but the former is essentially the TLV and the latter is
applied to the less well-known materials. See the HSE Guidance Notes
EH/40 series, 1989. The old TLV values, defined by different groups of
powders, will be discussed below, as they provide a framework on
which to base decisions on how hazardous particles may be.
Group I – very dangerous, 0 to 50 µg m−3, as they readily give rise to
fibrosis and include: beryllium, silica in the cristobalite form and blue
asbestos (5 fibres/cc less than 5 µm in length). Note that asbestos is not
classified using aerodynamic diameters; this is because of its unusual
fibre like shape.
Group II – dangerous, 50 to 250 µg m−3, including: asbestos (other forms
of and with 5 fibres/cc > 5 µm in length), silica such as quartz, and
mixed dusts containing 20% or more of silica.
Group III – moderate risk 250 - 1000 µg m−3, including: mixed dusts of
less than 20% silica, talc, mica, kaolin, cotton, organic dust, graphite
and coal.
Group IV - >1 mg m−3, the least problematical dusts including: cement,
limestone, glass, barytes, perlite, iron oxide, magnesia and zinc oxide.
The general term for a dust induced lung disease is Pneumoconiosis.
Well-known forms of Pneumoconiosis include: asbestosis and silicosis.
The prevention of a respiratory hazard includes: keeping dust levels
low, switching the process to a powder of lower danger where
possible, using dust extraction hoods and personal protection equipment
such as masks and gloves. It is critically important to monitor the
working air quality frequently by taking samples according to
approved procedures. This normally means sampling the airflow
within ducts isokinetically. When sampling is required in a normally
still environment, proprietary devices designed to only pass material
within the respirable range on to a test filter are available, using an air
pump to ensure that the gas velocity within the sample tube is similar
to what would be experienced by a human. This is illustrated in Figure
15.3. The resulting concentration can be checked against the
appropriate standard.
15.3 Summary
The head and throat regions have very effective means for trapping
dust: nasal hair and splits in the respiratory tract provide targets for
diffusional and inertial collection. Mucus is used to ensure that the
particles do not become re-entrained in the flow. Ultimately, the
particles are discharged from the body or become swallowed. This
protects the lungs from dust ingress. However, the prevention of dusty
environments by good working practices, such as fume hoods and
flow booths is an important part of health and safety provision. The
likelihood of powder deflagrations is also reduced if powders are not
allowed to remain deposited on surfaces within operating
environments.
16 Case study
In this concluding chapter the application of certain aspects of the
preceding knowledge will be illustrated. The case study is entirely
fictional, but it includes many aspects appropriate to a large number
of industries. Recently, process engineering industries have moved
towards high value, relatively low throughput and batch identifiable
compounds and this is the strategy that will be used for this example.
An organic compound is produced in a batch reaction between a
mixture of liquids. The solvent in the reaction is an alcohol. The
organic compound forms a precipitate that needs to be recovered and
cleaned of unreacted species, and the presence of residual solvent
must be minimised, before being stored as a dry powder, or granule.
The solubility of the precipitate in the reactor is temperature
sensitive; dissolving at elevated temperature. Each batch produces
approximately ½ tonne of dry powder product. The product is to be
put into a 2 tonne storage vessel and then mixed with other powder
products to provide a particulate product with total mass of 20
tonnes, per batch. The intention is to sell this product in 50 kg drums,
or kegs, as a chemical intermediate for further processing. The
specifications on the product state that it must flow easily and
reliably, and that it must not provide a serious health risk from dust
during any further processing. Given the nature of the organic
compound produced in the batch reaction, it is desirable to minimise
operator contact with the reactor product and subsequent operations.
Let’s assume that the particle size distribution of the crystallised
product has been analysed and is as shown in Figure 16.1, marked
before recrystallisation. Being a batch system, there is probably the Fig. 16.1 Comparison of two
intention of moving the product out of the reactor as quickly as size distributions provided
possible to allow for another batch, or cleaning, or a different product by initial crystallisation and
to be formed in the reactor. This would make good economic sense as after recrystallisation by
the most expensive capital item could be highly utilised. However, heating and cooling – RRB
downstream processing; i.e. unit operations following on from the parameters are: x63.2% of 8
reactor may provide a bottleneck, if fed with material that is difficult and n of 1.5, and x63.2% of 12
to process. In general, finer particles are more difficult to process than and n of 2.5, respectively, see
bigger ones and one possibility is that the original precipitate could equation (2.5)
be reheated, to dissolve the particles, then gently cooled again to
grow larger crystals. On reheating the finer particles will
preferentially dissolve, as they have a high surface area to volume
ratio. On cooling, the particles remaining in suspension will provide
sites on which the solid material, coming out of solution, can deposit
onto thus forming bigger particles. This is represented by the second
curve shown in Figure 16.1. Appropriate data for the two curves are
included in the caption. The effectiveness of this strategy can be
evaluated by considering the downstream process of filtration.
160 Case study
The permeability of the resulting filter cake from these two

suspensions can be determined from the specific surface area and the
cake porosity, as described in the box on page 24. The specific surface
can be calculated from equation (2.17) and is 1.74 and 0.74 µm-1 for
the before and after recrystallisation size distributions respectively.
The spreadsheet to calculate this is available at:
http://www.midlandit.co.uk/particletechnology
In the absence of testing, the cake porosity will have to be assumed
and a reasonable value for a non-biological product with a regular
particle shape (i.e. not dendritic) is about 60%, or 0.6 as a fraction.
Hence, the cake permeabilities are 8.9x10−14 and 5.0x10−13 m2
respectively. These can be converted into specific resistance to
filtration using equation (4.12), remembering that solid concentration
and porosity must equal one – see Figure 3.1. The organic product
Fig. 16.2 Filter productivity
has a solid density of 1400 kg m−3; hence, the specific resistances to
with two particle sizes –
full spreadsheet is available
filtration are 20.0x109 and 3.59x109 m kg−1 for the two size distributed
at: materials. Using a liquid density and viscosity of 980 kg m−3 and 0.001
http://www.midlandit.co. Pa s respectively, and an assumed constant pressure forming the
uk/particletechnology filter cake of 1 bar and 0.15 w/w feed slurry the productivity (filtrate
volume with time) on a filter 1 m2 in area can be calculated using
equation (4.18). In this calculation the filter medium resistance (Rm) is
assumed to be negligible. The resulting productivity curves are
shown in Figure 16.2.
So, by increasing the size of the particles, with the Sauter mean
diameter changing from 3.5 to 8.1 µm, the productivity is shown to
increase significantly: from about 20 kg in five minutes to 60 kg per
m2 of filter area. Thus, if the filtration capacity is limited it would be
better to use the reactor, or a suitable free vessel, to solubilise and
Fig. 16.3 Productivity at recrystallise the solids before filtration to obtain a threefold increase
different feed slurry in filtration performance.
concentration marked in % Productivity in the filtration stage can be improved further by
by mass feeding a higher concentration of solids to the filter. This is evident
from equation (4.17), where dry cake mass per unit volume of filtrate
(c) is seen to be a function of feed slurry concentration. Using
Filter cake resistance different values of feed slurry concentration and equations (4.17) and
This example uses a (4.18) provides the data illustrated in Figure 16.3 – investigating
calculated cake resistance throughput and feed concentration. The feed concentration can be
from a size distribution. In increased in this way by allowing the solids formed in the reaction, at
many instances the 15% by mass, to settle before passing on to the filtration stage. The
measured cake resistance is
time taken for the settling can be estimated from equation (6.1).
much higher: due to transfer
of fines within the cake to a
Firstly, the terminal settling velocity of the particles is required.
position of high liquid flow Using the Sauter mean diameter in Stokes’ law, equation (5.5),
resistance and cake provides a value of 1.5x10−5 m s−1. In order to use equation (6.1) the
compaction. Ideally, Particle Reynolds number must be first calculated, equation (5.6), to
laboratory tests should be check which exponent to use in the Richardson and Zaki equation.
used to determine the cake The Particle Reynolds number is much less than 0.2; thus 4.65 is used
resistance to be used in in equation (6.1). On investigating solid concentrations between 15%
design studies.
and 35% by mass, 0.11 to 0.27 by volume – see equation (3.16), the
settling velocities are found to vary between 1.34x10−5 and
1.1x10−5 m s−1. Thus, the time taken to settle one metre is between 21
and 25 hours! So, gravity sedimentation is too slow to provide
thickening in this case. Clearly, the low density difference between
the solid and liquid, and small particle size are factors in this
disappointing result.
As there is a wish to minimise operator contact with the product
and the need for rapid processing of the batch, one suitable method
would be to use an automated filtering centrifuge. One popular type
of machine is the inverting bag centrifuge, which can discharge its
solids contents directly into a dryer. Washing of the filter cake and
spinning dry of residual liquid within the cake is possible prior to
cake discharge. These operations will now be considered with the
feed concentration of 15% by mass.
In the operation of a filtering centrifuge with particles less than 10
µm in diameter it would be unrealistic to expect to have no filter
medium resistance. The fine particles are likely to penetrate the cloth
to some extent and cause a significant resistance. Hence, a value of
Fig. 16.4 Time required for
1x1011 m−1 will be assumed. The cake resistance and physical
a wash ratio of 4 on the
properties are as before and other data is as follows: filter basket
filtering centrifuge
height 0.22 m, filter basket diameter 0.45 m, and rotational speed of
1200 rpm. Equations (8.24) and (8.26) must be solved and a critically
important factor is the cake radius: this is limited by the geometry of
the machine and filtration must stop before the cake becomes too
great. Under the stated operating conditions, an analysis shows that
the radius starts at 0.225 m, the full radius of the device, and after 5
minutes the cake is at a radius of 0.13 m; providing a cake thickness
of 0.1 m. In order to wash unreacted chemicals from the product a
wash ratio of 4 has been selected, see Figure 8.13. Clearly, the thicker
the cake the greater the wash time will be. The wash time is provided
by equation (8.27). A reasonable wash time is provided by the 0.1 m
thick cake, which was formed by filtering for 300 seconds. The filter
has 14.2 kg of solids in it and has filtered 65.6 kg of filtrate. An
additional 15.8 kg of liquid is retained in the cake prior to the
Fig 16.5 Zeitsch’s
dewatering stage. The time taken to spin the cake as dry as possible
dewatering model – with
and the residual moisture value are obtained from equations (8.29) to
a surface tension of
(8.33). One advantage of a low cake resistance is that the dewatering
0.08 N m−1 and 10 degrees
stage is quick and very efficient at removing residual solvent within
solvent to solid surface
the cake. Figure 16.5 illustrates the speed with which the cake contact angle and a very
dewaters at a rotational speed of 3400 rpm. All other properties are as low cake resistance
provided before, or stated below the figure.
The cycle time, on a single small inverting bag centrifuge is,
therefore, 300 seconds for filtration and 500 seconds for washing.
Time is also required for speed changes and discharge; thus a total
cycle time of 1200 seconds would be anticipated, to provide 14.2 kg of
product. This equates to 43 kg per hour and 12 hours to process the
entire reactor product using a single machine. This represents best
162 Case study
case productivity, as the cake resistance could be higher than we used

– see the box on the earlier page.
As the powder must have a low solvent content a drier must be
used to reduce the residual saturation of solvent to a very low value
and it is likely that agglomerates will probably form. These may need
to be broken before storage in the two tonne silos. However, an
alternative strategy that fulfils the overall requirement would be to
discharge the moist cake from the inverting bag centrifuge into a
screw conveyor that feeds a combined fluidised bed drier and
granulator. The discharge from the screw conveyor may have to pass
through a mesh to form an extrusion that breaks off into the fluidised
bed and forms granules during drying. A suitable granule size within
the fluidised bed would be about 200 µm.
Under the conditions provided above the required minimum
fluidising velocity can be calculated from equation (7.4) and is 0.038
m s−1, for an assumed voidage of 50% at incipient fluidisation and
using a gas viscosity and density of 2x10−5 Pa s and 1.2 kg m−3,
respectively. It is likely that a significant multiple of the minimum
fluidising velocity would be used, at least 5 times the value. The
settling velocity of the 200 µm granules can be calculated from the
Heywood Tables and is 0.84 m s−1, with a Particle Reynolds number
of 10, so a fluidising velocity of 0.19 m s−1 is safely below the velocity
at which significant entrainment of solids in the gas stream above the
fluidised bed will occur. This is a low throughout operation and a
very small fluidised bed would suffice. Even in a bed of diameter 0.2
m, and height 0.3 m, the average residence time in the bed would be
about 9 minutes. The residence time is greater in larger beds and 9
minutes is likely to be sufficient for drying and agglomerate
formation. A more detailed analysis of this would require mass
transfer data on the particles. An experimental study of this, and the
granulation process would be warranted.
The granules from the fluidised bed are sufficiently large and
should be spherical in shape; so, they should easily flow and simple
gravity discharge from the fluidised bed into the two tonne storage
hopper is possible. In this hopper further batches of the reactor
product will be added. When the equipment is available, the two
Pneumatic conveying tonnes from the hopper need to be mixed with another powder and
conditions stored in a 20 tonne hopper before putting into drums. One
mean size 200 µm possibility is to batch mix the powders directly above the 20 tonne
Gas density: 1.2 kg m−3 storage hopper and to use what is known as bomb-doors discharge
Solid density 1400 kg m−3 into the 20 tonne hopper. The wide opening discharge doors of the
gas viscosity 1.8x10−5 Pa s hopper ensure that the product is discharged with minimal
vertical distance 20 m opportunity to segregate. The storage hopper would need to be
horizontal distance 30 m designed for mass flow discharge so, again, segregation is minimised
number of bends 8
and the drums should contain a well mixed product.
To raise the granules from the 2 tonne intermediate hopper to the
batch mixer a pneumatic conveying pipe may be used, probably with
an inert gas. If the conveying rate is 1000 kg h−1, the theory presented
in Chapter 9 suggests that a pressure drop of 0.28 bar results from the
operating conditions provided in the box, in a 63 mm diameter pipe.
The gas flow rate required for this operation is 0.05 m3 s−1, operating
at 50% excess gas over the saltation velocity. Dilute phase conveying
should help minimise granule breakup.
Assuming that the powder to be mixed with the granules has a
mean size, by mass, of 100 µm the quality of the batch mixture can be
estimated from equation (12.5). Note, this will represent the best
quality we should expect from the powders; it is possible that the
mixture quality will be worse, due to the performance of the Mean particle mass of
equipment. The mass proportion of our granules will be 0.1, i.e. 2 the granules is:
tonnes going to form an overall mixture mass of 20 tonnes, and the π
Wp = ρs x3
mean particle mass of the granules will be 5.9x10−9 kg, see box. 6
The other powder is likely to be a filler and will have a density where the density is
closer to that of an inorganic material, such as 2000 kg m−3. Thus, the 1400 kg m−3 and
mean particle mass for the other powder will be 1.0x10−9 kg. The diameter is 200 µm, but
appropriate sample mass to use in equation (12.5) is the mass of the this assumes no internal
product leaving the process, which is in 50 kg drums. Hence, the voidage of the granules.
The real density would
between drum variance of our active ingredient will be 9.7x10−12,
be slightly less, but the
giving a standard deviation of 3.1x10−6. Now, to put these values into overall mixing result will
context: using the analysis described in Chapter 12, one drum in 1000 not significantly change.
will have a proportion of the granules greater than 0.1 plus 3 times
the standard deviation, and one in 1000 will have a proportion less
than 0.1 minus 3 times standard deviation. This suggests that 998
drums out of 1000 will have masses of the granules varying from
4.9995 to 5.0005 kg. This analysis suggests that a very good mixture
quality will be achieved by batch mixing these dry powders.
However, segregation as the powder falls from the mixer into the
hopper will have to be assessed. Thus, the batch mixing theory
provides an indication of the best possible quality that can be
obtained, operating procedures and mixer type choice may reduce
the mixture quality obtained in practice.
The 20 tonne storage hopper should be designed to provide mass
flow, in accordance with Chapter 10. This will minimise segregation
during the filling of the 50 kg drums. It is not possible to provide a
design here without laboratory data obtained from a shear cell,
similar to that provided on page 112. In practice, it is likely that the
hopper will discharge directly into the drums which will be
positioned on a load cell to weigh the drum contents. An alternative
would be to discharge into a screw conveyer, which then deposits the
mixture into the drum positioned on a load cell. The latter system is
more controllable and less likely to provide a hazard to the process
operators.
The above simplified example provides some indication of how
unit operations in particle technology are related and how some of
the principles covered in the preceding pages may provide an insight,
or even an outline design, of the process.
Nomenclature
Symbols defined and used locally are not included here.
A area, m
2
−2
a acceleration, ms
Ac cloud radius around bubble, m
C solid concentration by volume fraction, -
−3
c dry cake mass per unit volume filtrate, kg m
Cd drag coefficient, -
Cf feed solid concentration by volume fraction, -
Cu underflow concentration by volume fraction, -
D vessel, or pipe, diameter, m
d pipe diameter, m
df fibre diameter, m
E efficiency, -
−2
F force, kg m s
f friction factor, -
fg gas friction factor, -
fi number of particles within an increment, -
fs solids friction factor, -
−2 −1
Gs mass flux solids, kg m s
h pressure head; i.e. equivalent to height of liquid, m
H height of channel, or vessel, m
Hi height of suspension at some time, m
Ho original height of suspension, m
2
k permeability to fluid flow, m
L length of pipe or bed, m
mi mass fraction within a size range, or increment, -
mp mass of particle, kg
−3
N mass concentration of particles, kg m
−3
No original mass concentration of particles, kg m
−1
PH Heywood settling factor, see page 50, m
3 −1
Q volume flow rate, m s
−1
QH Heywood settling factor, see page 50, ms
−1 −2
R shear stress, kg m s
−1
Rm filter medium resistance, m
r radial coordinate, m
rc radial position of cake in filtering centrifuge, m
rcr start radius for critical particle size, m
Rf recovery of flow to underflow of hydrocyclone, -
rL inner radial position of liquid in a centrifuge, m
Ro equilibrium orbit radius, m
ro radius of a centrifuge, m
rt target radius, m
Rb bubble radius, m
2 −3
Sv specific surface area per unit volume, m m
s slurry concentration by mass fraction, -
t time, s
−1
U interstitial velocity, ms
−1
u fluid velocity, ms
−1
Ub bubble velocity, ms
−1
ug gas velocity (can include entrained solids with no slip), ms
−1
Umf minimum fluidising velocity, ms
−1
Uo superficial velocity, ms
−1
Up particle velocity, ms
−1
Ur radial gas velocity, ms
−1
ur radial velocity, ms
−1
us solids velocity with slip (us < ug), ms
−1
Ut terminal settling velocity, ms
−1
Uθ angular gas velocity, ms
−1
uθ tangential velocity, ms
3
V volume of fluid, m
W channel width, m
x particle diameter, m
x mean particle diameter, m
xi mid-particle diameter within a size range, or increment, m
xSv Sauter mean diameter, m
z height, or axial distance, coordinate, m
zs stop distance, m
Greek
−1
α specific resistance of filter cake, m kg
αf packing density of fibres, -
αs volume fraction of liquid in spray tower, -
−1 −2
∆P pressure difference, or drop, kg m s
ε void fraction, voidage or porosity, -
η particle removal efficiency, -
ηs single fibre, or target, particle removal efficiency, -
−1
λ bed filtration constant, m
−1 −1
µ fluid viscosity, kg m s
−3
ρ fluid density, kg m
−3
ρb bulk density, kg m
−3
ρm mean suspension density, kg m
−3
ρs solid density, kg m
−1 −2
σ normal stress, kg m s
−1 −2
τ shear stress, kg m s
−1
ω angular velocity, s
Further Reading
As this is an introductory work, the number of references in each
chapter has been kept to an absolute minimum. Where provided, they
are given in the text. An intelligent search of the Internet will result in
excellent equipment illustrations and pictures and these have also been
kept to a minimum. However, for the purpose of further in-depth
study of the topics introduced here the following books are
recommended. All these books are currently available from well-
known Internet retailers.
Introduction to Particle Technology

M. Rhodes, Paperback, 336 pages (27 August, 1998), John Wiley and
Sons Ltd; ISBN: 0471984833.
Processing of Particulate Solids

J. Seville, U. Tuzun, R. Clift, Hardcover, 384 pages (May 1997), Kluwer
Academic Publishers; ISBN: 0751403768.
Characterization of Powders and Aerosols

Brian H. Kaye, Hardcover, 312 pages (January 1999), John Wiley & Son
Ltd; ISBN: 3527288538;
Particle Size Measurement (Powder Technology Series), 5th Edition

Terence Allen, Hardcover, (December 1996), Kluwer Academic
Publishers,
Hardbound Set of 2 vols., ISBN 0-412-75350-2.
Dynamics of Fluids in Porous Media

Jacob Bear, Paperback, 764 pages (June 1988), Dover Publications;
ISBN: 0486656756
Solid-Liquid Filtration and Separation Technology, 2nd edition

A. Rushton, A.S. Ward, R.G. Holdich, Hardcover, 587 pages (May 15,
2000),
Wiley-VCH; ISBN: 3527296042.
Sedimentation and Thickening: Phenomenological Foundation and

Mathematical Theory
Book Series: mathematical modelling: theory and applications :
Volume 8
María Cristina Bustos, Fernando Concha, Raimund Bürger, Elmer M.
Tory, Hardbound, 304 pages (September 1999), Kluwer Academic
Publishers, Dordrecht, ISBN 0-7923-5960-7
Fluid Flow for Chemical Engineers

F.A. Holland, R. Bragg, Paperback, 373 pages 2nd Ed (17 March, 1995),
Butterworth-Heinemann; ISBN: 0340610581
Fluidization Engineering (Chemical Engineering Series) 2nd edition

Daizo Kunii, Octave Levenspiel, Daizeo Kunii, Hardcover, 491 pages
(March 1991), Butterworth-Heinemann; ISBN: 0409902330.
Industrial Centrifugation Technology

Wallace Woon-Fong Leung, Hardcover, 416 pages (February 1, 1998),
McGraw-Hill Professional; ISBN: 0070371911;
Handbook of Conveying and Handling of Particulate Solids

(Handbook of Powder Technology, V. 10)
A. Levy, H. Kalman (Editors), Hardcover, 872 pages (October 2001),
Elsevier Science Ltd; ISBN: 0444502351
Crushing, Grinding and Classification

Austin, Hardcover, 256 pages (1 March, 1996), Chapman & Hall; ISBN:
0412374005.
Powder Mixing, Particle technology series, Volume 10

Brian H. Kaye, Hardbound, 276 pages, (September 1997), Kluwer
Academic Publishers, Dordrecht; ISBN 0-412-40340-4
Introduction to Colloid and Surface Chemistry 4th Ed

Duncan J. Shaw, Hardcover, 306 pages (24 February, 1992),
Butterworth-Heinemann; ISBN: 0750611820
Gas Cleaning in Demanding Applications

by J. P. K. Seville (Editor), 308 pages (January 15, 1997), Chapman &
Hall; ISBN: 0751403512.
Also, there is the following well-known series of books:

Particle Technology Series
(successor to the Chapman & Hall Powder Technology Series).
Series editor: Brian Scarlett, Kluwer Academic Publishers, Dordrecht.
URL: http://kapis.www.wkap.nl/prod/s/POTS
APPENDIX
Values of log10 Ut/QH in terms of log10 PHx for spherical particles – see page 52
i.e. within the table are values of log10 Ut/QH which correspond to the values of log10 PHx given
by the left hand column, with the second decimal place coming from the scale on the top row,
interpolation may be used for the third decimal place.
log PHx 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
–0.2 –1.780
–0.1 –1.580 –1.600 –1.620 –1.640 –1.660 –1.680 –1.700 –1.720 –1.740 –1.760
–0.0 –1.382 –1.402 –1.422 –1.442 –1.461 –1.481 –1.501 –1.521 –1.541 –1.560
0.0 –1.382 –1.362 –1.343 –1.323 –1.303 –1.283 –1.264 –1.244 –1.225 –1.205
0.1 –1.185 –1.166 –1.146 –1.126 –1.106 –1.087 –1.068 –1.048 –1.029 –1.010
0.2 –0.990 −0.971 −0.952 −0.932 −0.912 −0.893 −0.874 −0.855 −0.836 −0.817
0.3 −0.799 −0.780 −0.762 −0.743 −0.725 −0.707 −0.688 −0.670 −0.652 −0.634
0.4 −0.616 −0.598 −0.580 −0.562 −0.544 −0.527 −0.510 −0.492 −0.475 −0.457
0.5 −0.440 −0.423 −0.406 −0.389 −0.373 −0.357 −0.341 −0.325 −0.308 −0.292
0.6 −0.276 −0.260 −0.245 −0.229 –0.213 −0.198 −0.183 −0.168 −0.153 −0.138
0.7 −0.123 −0.109 −0.095 −0.080 −0.066 −0.052 −0.038 −0.024 −0.011 0.003
0.8 0.017 0.030 0.043 0.057 0.070 0.083 0.096 0.109 0.122 0.135
0.9 0.148 0.161 0.173 0.186 0.199 0.211 0.224 0.236 0.248 0.261
1.0 0.273 0.285 0.297 0.309 0.321 0.333 0.345 0.356 0.368 0.380
1.1 0.391 0.402 0.414 0.425 0.436 0.447 0.458 0.469 0.480 0.491
1.2 0.502 0.513 0.523 0.534 0.545 0.555 0.565 0.576 0.586 0.596
1.3 0.607 0.617 0.627 0.637 0.647 0.657 0.667 0.677 0.686 0.696
1.4 0.706 0.715 0.725 0.734 0.744 0.753 0.762 0.772 0.781 0.790
1.5 0.800 0.809 0.818 0.827 0.836 0.844 0.853 0.862 0.870 0.879
1.6 0.887 0.895 0.904 0.912 0.920 0.928 0.936 0.944 0.951 0.959
1.7 0.967 0.974 0.981 0.989 0.996 1.004 1.011 1.018 1.026 1.031
1.8 1.040 1.048 1.055 1.062 1.069 1.076 1.083 1.090 1.097 1.104
1.9 1.111 1.118 1.125 1.132 1.139 1.146 1.153 1.160 1.167 1.174
2.0 1.180 1.187 1.194 1.200 1.207 1.214 1.220 1.227 1.233 1.240
2.1 1.246 1.253 1.259 1.265 1.272 1.278 1.284 1.290 1.296 1.302
2.2 1.307 1.313 1.319 1.324 1.329 1.335 1.340 1.345 1.350 1.355
2.3 1.360 1.364 1.369 1.374 1.378 1.383 1.388 1.392 1.397 1.401
2.4 1.406 1.411 1.415 1.420 1.424 1.428 1.433 1.437 1.441 1.445
2.5 1.450 1.454 1.458 1.462 1.466 1.470 1.474 1.478 1.482 1.486
2.6 1.490 1.494 1.498 1.502 1.506 1.510 1.514 1.518 1.521 1.525
2.7 1.529 1.533 1.537 1.541 1.545 1.549 1.553 1.557 1.561 1.565
2.8 1.569 1.573 1.578 1.582 1.586 1.590 1.594 1.598 1.603 1.607
2.9 1.611 1.616 1.620 1.624 1.629 1.633 1.637 1.642 1.646 1.651
3.0 1.655 1.660 1.665 1.669 1.674 1.679 1.684 1.689 1.694 1.698
3.1 1.703 1.708 1.713 1.718 1.724 1.729 1.734 1.740 1.746 1.751
3.2 1.757 1.763 1.770 1.776 1.782 1.788 1.795 1.801 1.808 1.814
3.3 1.821 1.828 1.834 1.841 1.848 1.854 1.861 1.868 1.875 1.881
Values of log10 PHx in terms of log10 Utt/QH for spherical particles
i.e. within the table are values of log10 PHx which correspond to the values of log10 Ut/QH given
by the left hand column, with the second decimal place coming from the scale on the top row,
interpolation may be used for the third decimal place.
log Ut/QH 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
−1.7 −0.160
−1.6 −0.110 −0.115 −0.120 −0.125 −0.130 −0.135 −0.140 −0.145 −0.150 −0.155
−1.5 −0.060 −0.065 −0.070 −0.075 −0.080 −0.085 −0.900 −0.095 −0.100 −0.105
−1.4 −0.009 −0.014 −0.019 −0.024 −0.029 −0.034 −0.040 −0.045 −0.050 −0.055
−1.3 0.041 0.036 0.031 0.026 0.021 0.016 0.011 0.006 0.001 −0.004
−1.2 0.093 0.087 0.082 0.077 0.072 0.067 0.062 0.057 0.052 0.046
−1.1 0.143 0.138 0.133 0.128 0.123 0.118 0.113 0.108 0.103 0.098
−1.0 0.195 0.190 0.185 0.179 0.174 0.169 0.164 0.159 0.154 0.148
−0.9 0.246 0.241 0.236 0.231 0.226 0.221 0.216 0.211 0.206 0.200
−0.8 0.299 0.293 0.288 0.283 0.278 0.272 0.267 0.262 0.257 0.252
−0.7 0.354 0.348 0.343 0.337 0.332 0.326 0.321 0.316 0.310 0.305
−0.6 0.409 0.404 0.398 0.392 0.387 0.382 0.376 0.370 0.364 0.359
−0.5 0.465 0.460 0.454 0.448 0.442 0.437 0.432 0.426 0.420 0.414
−0.4 0.524 0.518 0.512 0.506 0.500 0.494 0.488 0.483 0.477 0.471
−0.3 0.585 0.579 0.573 0.567 0.561 0.555 0.548 0.542 0.536 0.530
−0.2 0.649 0.642 0.636 0.629 0.623 0.616 0.610 0.604 0.597 0.591
−0.1 0.716 0.709 0.702 0.695 0.688 0.682 0.675 0.668 0.662 0.656
−0.0 0.781 0.773 0.766 0.759 0.752 0.745 0.738 0.730 0.723
0.0 0.788 0.795 0.802 0.810 0.818 0.825 0.832 0.840 0.848 0.856
0.1 0.863 0.871 0.879 0.886 0.894 0.902 0.910 0.917 0.925 0.933
0.2 0.941 0.949 0.957 0.965 0.973 0.981 0.989 0.997 1.006 1.014
0.3 1.022 1.031 1.039 1.048 1.056 1.064 1.073 1.082 1.090 1.099
0.4 1.108 1.117 1.126 1.135 1.144 1.153 1.162 1.171 1.180 1.189
0.5 1.198 1.208 1.217 1.227 1.236 1.245 1.255 1.265 1.274 1.284
0.6 1.294 1.303 1.313 1.323 1.333 1.343 1.353 1.363 1.373 1.384
0.7 1.394 1.404 1.415 1.425 1.436 1.446 1.457 1.468 1.479 1.490
0.8 1.500 1.511 1.522 1.533 1.545 1.557 1.568 1.580 1.592 1.604
0.9 1.616 1.628 1.640 1.652 1.665 1.678 1.691 1.704 1.718 1.731
1.0 1.745 1.759 1.773 1.786 1.800 1.813 1.827 1.841 1.855 1.870
1.1 1.884 1.899 1.913 1.927 1.941 1.956 1.970 1.985 2.000 2.015
1.2 2.030 2.045 2.060 2.075 2.090 2.106 2.122 2.138 2.154 2.170
1.3 2.187 2.204 2.222 2.241 2.260 2.280 2.300 2.321 2.343 2.365
1.4 2.387 2.409 2.431 2.454 2.477 2.500 2.524 2.549 2.574 2.600
1.5 2.626 2.651 2.677 2.703 2.728 2.753 2.778 2.803 2.827 2.851
1.6 2.874 2.897 2.920 2.943 2.966 2.988 3.010 3.032 3.053 3.073
1.7 3.093 3.113 3.133 3.152 3.170 3.188 3.204 3.220 3.236 3.252
1.8 3.268 3.283 3.298 3.313 3.328 3.343 3.358 3.373 3.388 3.402
Index
aerodynamic diameter 156 colloid stability 134
agglomeration 131-140 comminution 113
Andreasen pipette 6 – mechanisms 116
angle of internal friction 106-108 composite flux 59
angle of repose 99 compressible filter cake 33
angular momentum 77 compressible sediments 60
Archimedes number 48 concentration
atomic force microscopy 133 – by mass and volume 26
attritor mill 113 – characteristics 57
– polarisation 40
backflush 29, 39 cone and quartering 3
ball mill 120 confidence 127
batch flux 57 continuous powder mixing 129
Beer-Lambert law 20 conveying 91-98
between sample variance 124 core flow 101
bin 101 Coulter Counter™ 6
body feed 42 crack propagation 114
breakage function 116-118 critical concentration 59
critical trajectory 50, 78, 144
Broadbent & Callcott 11, 117 crossflow filtration 38
Brownian motion 49 crushing 113
bubble wake 71 crystallization 159
bubbling bed 69
bucket elevator 95 Darcy’s law 23
bulk density 22, 100 Davidson and Harrison 71
buoyancy 45, 61, 77 deflagration 155
densification 139
cake washing 85 dewatering 85
Camp-Hazen 50, 53 diffusional collection 142
capillary rise & pressure 137-140 dilatancy 100
Carman 25 discrete element analysis 73
centrifugal collector, gas 146 DLVO theory 133
centrifugal force 48, 77 Dodge and Metzner 62
centrifugal separation 77-90 double layer repulsion 131-135
centrifuges, sedimenting 78 downstream processing 159
chain conveyor 95 drag coefficient 46
chimney, particle deposit 151 drainage number 87
circulating load 118 drainage rate constant 87
clarification 45 dry cake mass per filtrate 32
clarifier tank 50 volume
clarifying filtration 29 dust sampling 157
classification 113, 118 dust, collection mechanism 142-146
classifier, grade efficiency 118
coagulation 135 efficiency
cohesion 106 – overall 141
cohesive powder mixing 129 – target 141, 145
electrical permittivity 132 friction factor 24, 46, 63

electron microscope 15 Froude number 69
electrostatic force 137
– precipitator 147 gangue mineral 113
emulsion phase 69 gas cleaning 141
energy, during crushing 114 – devices 147
equilibrium flux rate 39 – cyclone 146
equivalent diameter 5 – cyclone transposition 143, 152
Ergun correlation 25 Geldart 70
Generalised Reynolds no. 62
Feret diameter 15 grab sample 2
field force 45 grade efficiency 82, 141
film theory, membranes 39 granulation 131, 139
filter grinding 113
– aids 42 – media 120
– bed 29
– cake resistance 160 Hamaker constant 133
– media 40 HEPA filter 145
– productivity 160 heterogeneous flow 91
filtering centrifuges 83 Heywood tables 48, 52
filtration, liquid 29-44 hindered settling 55
– bridging 31 hopper 101
– cake formation time 35 – angle 102
– candle 31 – flow factor 108
– constant pressure 35 – half angle 100
– constant rate 36 – opening 102
– constants 34 human lungs 157
– equipment selection 42 humidity, effect on flow 109
– first runnings 41 hydrocyclone 79-83
– pre-coat 41 hydrometer 61
– self weight 61
– variable pressure & rate 37 image analysis 15
floc bed clarifier 136 – sheared diameter 15
flocculation 136 inertial force 48-49
Flocculent sedimentation 60 – separation 141-142
fluid flow 21 irreducible saturation 87
fluidisation 67-76 iso-electric point 136
– aggregative/bubbling 69-70 isokinetic 3
– cloud 71
– distributor plate 67 Janssen 101
– particulate 69 jar test 55, 137
fluidised bed 67 Jenike design chart 108
– granulator 139
fractal 16 Kozeny-Carman 22-23
free flowing powder 99 Krieger’s equation 61
freeboard 67 Kynch 57, 60
172
lamella separator 62 particle

Lasentec™ 6 – bounce 143
lean phase 69 – charge around 131
liberation size 113 – entrainment 141
London-van der Waal 132, 137 – shape 6, 15
long tube test 51 – size analysis 6
– size functions 7-11
Malvern™ 6, 16 – slip 91
Martin diameter 15 pendular liquid bridge 138
mass flow 101 permeability 24
material flow function 103-107 permeate flux 39
maximum exposure limit 157 permeation 29
mean 125 plate and frame filter 36
mean free path correction 49 pneumatic conveying 92-94, 162
medium resistance 32 pneumoconiosis 158
– effective 44 polyelectrolyte 136
membrane filtration 38 Poole, Taylor & Wall 126
microfiltration 38 population balance 119
micron 2 porosity 21
microscope 6 porous media 21
milling 113 potential energy curves 134
– matrix 117 powder
mixed media bed 30 – flood 99
mixing mechanisms 123 – flow function 103-107
Modified Reynolds no. 22 – flow/ storage 99-112
Moh’s hardness scale 113 – hazard, explosion 155
Mohr’s circle 103 – hazard, physiological 156
moisture force 137-138 – mixing equipment 128
moisture ratio 35 pressure relief 156
moments 13 primary vortex 80
principal planes 104
nanotechnology 131
non-Newtonian rheology 62 rat holing 101
Normal distribution 11, 124 respirable range 156
normalisation 9 retentate 39
nucleation 139 Reynolds number 1
Nutsche filter 36 – Generalised 62
– Modified 22
occupational exposure 158 – Particle 45
standard rheogram 62
overflow 80 Richardson and Zaki 56
riffler 2
packing arrangement 26 Rizk 92
pan granulator 139 rotary vacuum filter 35
Particle Reynolds no. 45
sampling 2 ultrafiltration 38
saturation 87 unconfined yield stress 103-107
screw conveyor 94 underflow 80
secondary vortex 80 – withdrawal flux 59
Sedigraph™ 6 uniformity index 10
sedimentation 45-66
segregation 2, 123 van der Waal 132, 137
selection function 116 variance 125
separation span 149 – with mixing time 126
separation, liquid-liquid 79 velocity
settling basin, liquid 50 – angular 77
settling chamber, gas 146 – axial 80
shape coefficient 7 – bubble 71
sharpness of separation 142 – interstitial 21
shear cell 109 – locus of zero vertical 81
sieve 6 – limit deposit 91
sigma theory 79 – minimum fluidising 67, 68
silo 101 – saltation 92
– stress 102 – superficial 21
solids flux 57-60 – tangential 80
solids fraction 21 venturi scrubber 146
specific resistance, cake 32 vibratory conveyor 95
specific surface area 12, 74 viscosity, Newtonian 61
specification 127 voidage 21
spherical diameter 5 volume concentration 21
spherical polar coordinates 72 vortex finder 80
spray tower 145
stable arch 103 Wadell’s sphericity 5
standard deviation 125 Wakeman 87
Stern layer 131 wash ratio & volume 86
Stokes number 142
Stokes’ law 46 yield locus 106
stop distance 144 Young-Laplace equation 138
stress gradient 49
supernatant 56 Zeitsch 87, 161
surface tension force 138 Zeta potential 131-135
Svarovsky 83 zone theory 55-57
terminal settling velocity 45

thickener 58
– high rate 137
threshold limit value 157
time consolidation 109

Fundamentals of Particle Technology Holdich

Uploaded by

Document Information

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

Fundamentals of Particle Technology Holdich

Uploaded by

Copyright:

Available Formats

Micropore Filtration

Sponsors of the book: Fundamentals of Particle Technology

The production costs of this book have been subsidised

Micropore specialise in microporous filtration media

Midland Information Technology and Publishing

To view a copy of this licence, visit

This work has been released as an open educational

You are free:

to copy, distribute, display, and perform the work.

Attribution — You must give the original author credit.

The original book was published in the United Kingdom

and had the ISBN 0-9543881-0-0

3 Fluid flow through porous media 21

6 Hindered systems and rheology 55

10 Powder flow and storage 99

11 Crushing and classification 113

12 Solid/solid mixing 123

13 Colloids and agglomeration 131

14 Gas cleaning 141

15 Powder hazards 155

16 Case study 159

1.1 Prerequisites and objectives

1.2 The micron

inertially separating at the entrance to the sample pipe, see Chapter

1projected area – what is observed when looking at a particle using a microscope

Table 2.1 Common particle shape descriptions

Descriptor Wadell’s sphericity Example

Particle size analysis equipment is of fundamental importance in

Table 2.2 Commonly used particle size analysis equipment

Name Equivalent Example

order to simplify subsequent description and calculations it is usual

2.1 Particle size functions

where f is the number of particles in that size range. The mass

Considering the data used to construct Figure 2.3, it would be

2.2 Algebraic representation of size functions

n0(x) represents a number distribution

Gaudin-Schuhmann – cumulative function

Broadbent-Callcott – cumulative function

Gaudin-Meloy – cumulative function

Log-normal Distribution function (Figure 2.9)

2.3 Specific surface area per unit volume

2.4 Distributions: moments and conversions

In the case of k=0, then we have the simple result

which is valid regardless of the basis of the distribution data

2.5 Means of a distribution

Hence, the general equation for mean particle size is

2.6 Image analysis and particle shape

computer generated by taking the image of the particle and moving it

2.7 Example interpretation of distribution data

Table 2.3 Illustration of Malvern Results Statistics

Density = 1.000 g/cub.cm Specific S.A. = 0.7046 sq m / g

Malvern use D to represent particle size, against the use of x in this

Now convert your answer into one with SI units.

(v). The specific surface area per unit volume is (µm−1):

(vi). Now convert your answer into one with SI units:

a: 48.91 b: 12.35 c: 293.5 d: 1.286

the fluid (∆P). Pressure is not a vector quantity, but a pressure

3.2 Flow regimes

3.3 Darcy’s law and the Kozeny-Carman equation

Comparison of equations (3.4) and (3.7), results in the conclusion that

3.4 Friction factor

This is illustrated in the box, which shows that the Carman

3.6 Concentrations by mass and volume

(vii). Comment on whether your use of the Kozeny-Carman equation