The relationship between hydrodynamic variables and ...
Transcript of The relationship between hydrodynamic variables and ...
The relationship between hydrodynamic variables
and particle size distribution in flotation
Thèse
Ali Vazirizadeh
Doctorat en génie des matériaux et de la métallurgie
Philosophiæ Doctor (Ph.D.)
Québec, Canada
© Ali Vazirizadeh, 2015
iii
Résumé
La flottation industrielle est un procédé continu qui se déroule souvent en plusieurs étapes
et dans lequel des particules d'une espèce de minéral donnée (généralement celles d'intérêt),
présentes en différentes tailles, rencontrent une grande quantité de bulles de gaz
(normalement de l'air) pour produire des agrégats bulles-particules minérales, qui sont
extraits du dispositif de flottation (colonne ou cellule) en tant que produit de valeur
(concentré). Le contenu en bulles est décrit par les conditions hydrodynamiques régnant
dans le réacteur de flottation. Celles-ci sont reconnues pour leur influence sur la
performance de la flottation.
Ce projet de recherche porte sur deux sujets majeurs. Le premier est l'analyse de l’impact
des particules solides sur les variables hydrodynamiques et l’effet de ces variables
hydrodynamiques sur la récupération d’eau au concentré. Pour ce faire, l'effet du solide sur
la distribution de la taille des bulles et le taux de rétention de l’air, ainsi que la corrélation
entre la distribution de taille des bulles et le taux de rétention de l’air dans une colonne de
flottation ont été étudiés. L'effet du taux de rétention de l’air, de la dimension des bulles et
du taux surfacique de bulles (Sb) sur la quantité d’eau extraite au concentré a ensuite été
analysé.
Le second sujet traite de l'utilisation des variables hydrodynamiques pour la modélisation
de la cinétique du procédé de flottation selon distribution granulométrique des particules
introduites. La surface inter-faciale de bulle (Ib) est introduite à cet égard comme une
variable hydrodynamique fournissant plus d'informations sur la distribution de taille de
bulle que le taux surfacique de bulles qui est plus couramment utilisé. De plus, la
corrélation entre la constante cinétique, la taille des particules et certaines variables
hydrodynamiques a été analysée en utilisant une projection de structures latentes (PSL).
Les résultats indiquent que l'importance relative des variables hydrodynamiques pour la
modélisation de la cinétique de flottation dépend de la distribution granulométrique des
particules. Finalement, les variables hydrodynamiques suggérées pour chaque classe
granulométrique considérée ont été utilisées pour produire des modèles de régression
mono-variable de la constante cinétique.
v
Abstract
Industrial flotation is a continuous and often multistage process, where particles of a given
mineral species (usually the targeted one), present in different sizes, encounter a large
amount of gas bubbles (normally air) to produce mineral–bubble aggregates, which are
removed from the flotation device (cell or column) as a valuable product (concentrate). The
bubble content inside the cell is characterized by the prevailing hydrodynamic conditions
(known as gas dispersion variables), which in turn are known to influence the flotation
performance.
This research project deals with two major topics. The first one is identifying the effect of
mineral particles on hydrodynamic variables, and the effects of hydrodynamic variables on
the final water recovery. For this purpose, the effect of solid particles on the bubble size
distribution and gas hold-up, as well as the correlation between bubble size distribution and
gas hold-up in column flotation were studied. It is followed by an assessment of the effect
of the gas hold-up, bubble size and bubble surface area flux (Sb) on the amount of water
reporting to the concentrate.
The second topic deals with applying appropriate hydrodynamic variables for flotation
modeling based on a given introduced particle size distribution. The interfacial area of
bubbles (Ib) is introduced to address this issue as a hydrodynamic variable providing more
information about the size distribution of bubbles than the commonly used bubble surface
area flux. The correlation between the flotation rate constant and particle size as well as
given hydrodynamic variables using a Projection to Latent Structures (PLS) has been
analyzed. Results suggest that the relative importance of hydrodynamic variables for
flotation rate modeling depends on the particle size distribution. Finally the suggested
hydrodynamic variables for each of the various particle size-classes considered were used
to produce single variable models for the flotation rate constant.
vii
Table of contents
Résumé .................................................................................................................................. iii
Abstract .................................................................................................................................... v
List of Tables .............................................................................................................................. xi
List of Figures ........................................................................................................................... xiii
Nomenclature ........................................................................................................................... xvii
Acknowledgments .................................................................................................................... xxi
Foreword .............................................................................................................................. xxiii
Chapter 1 Introduction ................................................................................................................ 1
1.1 Principles of flotation ....................................................................................................... 1
1.2 Hydrophobic and hydrophilic particles ............................................................................ 1
1.3 Use of reagents in flotation practice ................................................................................ 2
1.3.1 Collectors .......................................................................................................................... 2
1.3.2 Modifiers ........................................................................................................................... 3
1.3.3 Frothers ............................................................................................................................. 3
1.4 Bubble-particle collection ................................................................................................ 4
1.4.1 Bubble-Particle Collision .................................................................................................. 5
1.4.2 Particle attachment ............................................................................................................ 7
1.4.3 Particle detachment ........................................................................................................... 8
1.4.4 Particle entrainment .......................................................................................................... 9
1.5 Flotation devices ............................................................................................................ 10
1.5.1 Mechanical cells ............................................................................................................. 10
1.5.2 Flotation columns ........................................................................................................... 11
1.6 Flotation modeling ......................................................................................................... 15
1.6.1 Performance justification ................................................................................................ 15
1.6.2 Kinetic constant modeling .............................................................................................. 18
1.6.3 Residence time measurement in a flotation column ....................................................... 21
1.7 Gas dispersion properties ............................................................................................... 22
1.7.1 Gas hold-up (εg) .............................................................................................................. 22
1.7.2 Superficial gas velocity (Jg) ............................................................................................ 22
viii
1.7.3 Bubble surface area flux (Sb) .......................................................................................... 23
1.7.4 Bubble size distribution .................................................................................................. 26
1.7.5 Gas dispersion properties range and effects ................................................................... 27
1.8 Relation between the kinetic constant and gas dispersion properties ............................. 29
1.9 Problems associated to the use of a single Sb value........................................................ 30
1.10 Assumptions made in this experimental work................................................................ 33
1.10.1 Hydrophobicity ............................................................................................................... 33
1.10.2 Solid particle residence time .......................................................................................... 33
1.10.3 Ultimate recovery (R∞) ................................................................................................... 34
1.10.4 Froth depth and its effect on the final recovery ............................................................. 35
1.11 Experimental set-up and results...................................................................................... 35
1.12 Objectives ....................................................................................................................... 36
1.13 Outline of the thesis ........................................................................................................ 37
Chapter 2 Effect of particles on the bubble size distribution and gas hold-up in column
flotation ................................................................................................................................. 41
2.1 Introduction .................................................................................................................... 42
2.2 Experimental set-up ........................................................................................................ 43
2.3 Results and discussion .................................................................................................... 48
2.3.1 Solid particles on the bubble size distribution ............................................................... 48
2.3.2 Effect of solid particles on the gas hold-up .................................................................... 55
2.3.3 Discussion ...................................................................................................................... 59
2.4 Conclusion ...................................................................................................................... 61
Chapter 3 The effect of gas dispersion properties on water recovery in a laboratory
flotation column ...................................................................................................................... 63
3.1 Introduction .................................................................................................................... 64
3.2 Results and discussion .................................................................................................... 66
3.2.1 Effect of gas dispersion properties on the water recovery ............................................. 66
3.2.2 Effect of hydrophobic particle size on the water recovery ............................................. 69
3.2.3 Effect of gas dispersion properties on the carrying capacity ......................................... 70
3.3 Conclusion ...................................................................................................................... 72
ix
Chapter 4 On the relationship between hydrodynamic characteristics and the kinetics of
column flotation. Part I: modeling the gas dispersion ............................................................. 73
4.1 Introduction .................................................................................................................... 74
4.1.1 Hydrodynamic variables and particle size ...................................................................... 75
4.1.2 Bubble surface area flux models ..................................................................................... 76
4.1.3 Bubble size measurement ............................................................................................... 77
4.1.4 Bubble size distribution .................................................................................................. 78
4.2 Equations of gas dispersion............................................................................................ 80
4.2.1 Normal distribution ......................................................................................................... 80
4.2.2 Log-normal distribution .................................................................................................. 82
4.3 Test procedure ................................................................................................................ 84
4.4 Results and discussion ................................................................................................... 85
4.4.1 Modeling the experimental bubble size distributions ..................................................... 85
4.4.2 Correlations between hydrodynamic variables ............................................................... 88
4.5 Conclusion ..................................................................................................................... 92
Chapter 5 On the relationship between hydrodynamic characteristics and the kinetics of
flotation. Part II: model validation ........................................................................................... 95
5.1 Introduction .................................................................................................................... 96
5.2 Test procedure ................................................................................................................ 99
5.3 Results and discussion ................................................................................................. 100
5.3.1 Particle size and kinetic constants ................................................................................ 100
5.3.2 Hydrodynamic variables and rate constant ................................................................... 101
5.3.3 Experimental validation ................................................................................................ 103
5.4 Interfacial area of bubbles and rate constant ................................................................ 107
5.5 Conclusion ................................................................................................................... 111
Chapter 6 Single variable rate constant models ...................................................................... 113
6.1 Introduction .................................................................................................................. 113
6.2 Flotation kinetics of fine particle size-class ................................................................. 113
6.3 Flotation kinetics of large particle size-class ............................................................... 118
6.4 Flotation kinetics of particle size-class spanning a wide range ................................... 120
6.5 Particle size effect on the kinetics of flotation ............................................................. 121
x
6.6 Conclusion .................................................................................................................... 122
Chapter 7 Thesis conclusion .................................................................................................. 123
7.1 Future work .................................................................................................................. 124
References ............................................................................................................................... 127
Appendix A .............................................................................................................................. 137
A.1 Sampling ............................................................................................................................ 137
A.2 RTD measurement ............................................................................................................. 138
A.3 Kinetic constant calculation .............................................................................................. 140
Appendix B .............................................................................................................................. 141
B.1 Selection of the shear water rate ........................................................................................ 141
B.2 Effect of the temperature and duration of the test on the bubble size ............................... 141
B.3 Relationships between hydrodynamic variables in two and three-phases ......................... 142
Appendix C .............................................................................................................................. 147
C.1 Solid Characteristics .......................................................................................................... 147
Appendix D .............................................................................................................................. 151
D.1 Radial gas dispersion analysis (Banisi et al., 1995) .......................................................... 151
Appendix E ............................................................................................................................... 153
Appendix F ............................................................................................................................... 155
F.1 Correlation between hydrodynamic variables .................................................................... 155
F.2 Regression methods ........................................................................................................... 157
F.3 Datasets .............................................................................................................................. 160
Table F.1 Experimental results ................................................................................................. 160
xi
List of Tables
Table 1.1 Yoon’s model parameters for different flow conditions ......................................... 6
Table 1.2 Example of a same Sb originating two different rate constants. ........................... 31
Table 1.3 Talc particle sizes measured by twice wet screening and Malvern 2000 ............ 34
Table 2.1 Summary of the experimental plan ....................................................................... 53
Table 2.2 Generated gas hold-up and d32 for gas-water and gas-slurry systems .................. 59
Table3.1 Experimental plan ................................................................................................. 67
Table 4.1 Experimental conditions for three types of BSD .................................................. 88
Table A.1 Measured RTDs of the tests ............................................................................... 140
Table B.1 Effect of temperature and duration of test on bubble size distribution .............. 142
Table B.2 Values of hydrodynamic variables generated by manipulating Jg and Jsw ....... 143
Table C.1 XRF results of quartz (%) ................................................................................. 147
Table C.2 Mineralogy analysis results ................................................................................ 148
Table C.3 XRF results for quartz particles (%) .................................................................. 149
Table F.1 Experimental results ........................................................................................... 160
Table F.2 Gorain's results ................................................................................................... 161
Table F.3 Massinaei et al. data base ................................................................................... 162
Table F.4 Kracht et al. data base ....................................................................................... 163
Table F.5 Jincai et al. data base .......................................................................................... 164
xiii
List of Figures
Figure1.1 Distribution of radial components of the particles and liquid velocity on the
bubble surface ......................................................................................................................... 7
Figure 1.2 Mechanical cell schematic (Fuerstenau et al., 2007) ........................................... 11
Figure 1.3 Flotation column schematic ................................................................................. 12
Figure 1.4 Three layers of the froth zone .............................................................................. 14
Figure 1.5 Typical grade–recovery curves for froth flotation ............................................. 17
Figure 1.6 Different bubble size distributions all having the same d32 ................................. 26
Figure 1.7 Different pairs of bubble size and gas velocity leading to a unique BSAF ........ 31
Figure 1.8 Two BSDs having same d32 originate different kinetic constant ........................ 32
Figure 1.9 Light beam passing through loaded bubble ......................................................... 35
erugiF 2.1 Schematic of the experimental set-up ................................................................. 44
Figure 2.2 Schematic of the εg sensor (Gomez et al., 2003) ................................................. 45
Figure 2.3 Schematic of a frit-sleeve sparger (Kracht et al., 2008) ...................................... 46
Figure 2.4 Example of an original captured image and detected bubbles .......................... 47
Figure 2.5 Effect of the solid percent on the bubble size distribution .................................. 49
Figure 2.6 Example of bubble images using 25 ppm F150, and for Jg = 1 cm/s .................. 50
Figure 2.7 Effect of the solid percent on the bubbles size histogram ................................... 52
Figure 2.8 Bi-modal distribution due to coalescence ........................................................... 53
Figure 2.9 The effect of the particle size on the bubble d32 at constant gas rate (1 cm/s) .... 54
Figure 2.10 The effect of solids on the parameters of the bubble size distribution .............. 57
Figure 2.11 Bubble size and gas hold-up correlation in two and three-phase system .......... 60
Figure 2.12 Bubble size and gas hold-up correlation in three-phase system ........................ 61
xiv
Figure 3.1 Effect of the mean bubble size (d32) on water recovery for constant as rate ...... 68
Figure 3.2 Effect of the gas hold-up on the water recovery for constant as rate .................. 68
Figure 3.3 Effect of the mean talc particle size on the water recovery ................................ 69
Figure 3.4 Effect of (a) the bubble size, (b) gas holdup, and (c) bubble surface area flux on
the carrying capacity ............................................................................................................ 71
Figure 4.1 Bubble size distribution histogram ..................................................................... 85
Figure 4.2 Example of a bubble size histogram for T1 distribution .................................... 86
Figure 4.3 Example of a cumulative bubble size T2 distribution fitted with multi-shape
density function .................................................................................................................... 87
Figure 4.4. Example of a cumulative bubble size T3 distribution fitted with a lognormal
density function .................................................................................................................... 88
Figure 4.5 Shows the correlation between Ibc and Ibm (a) for T2 distribution (b) for T3
distribution ........................................................................................................................... 89
Figure 4.6 Correlation between : (a) Ibc and Sb for T2 distributions, (b) Ibc and Sb for T3
distributions, (c) Ibm and Sb for T2 distributions, (d) Ibm and Sb for T3 distributions ........... 90
Figure 4.7 Correlation between: (a) Ibc and εg for T2 distributions, (b) Ibc and εg for T3
distributions, (c) Ibm and εg for T2 distributions (d) Ibm and εg for T3 distributions ............. 91
Figure 5.1. Relative importance of particle size (P.S.), εg, Sb and Ib .................................. 101
Figure 5.2. Importance of variables in VIP projection for three sizes of particles ............ 104
Figure 5.3 PLS regression for hydrodynamic variables .................................................... 105
Figure 5.4. PLS regression for hydrodynamic variables .................................................... 106
Figure 5.5 PLS regression for hydrodynamic variables ..................................................... 107
Figure 5.6 PLS regression for hydrodynamic variables ..................................................... 107
Figure 5.7 PLS regression for hydrodynamic variables using mixed-size class particles . 108
Figure 5.8 PLS regression for hydrodynamic variables ..................................................... 109
Figure 5.9 PLS regression for hydrodynamic variables using the datasets combined ...... 110
xv
Figure 6.1 Talc collection rate constant as a function of the bubble size ........................... 114
Figure 6.2 Talc collection rate constant as a function of Sb for two particle size-classes .. 114
Figure 6.3 Kinetic constant as a function of d32 for the collection zone and the overall
process ................................................................................................................................ 115
Figure 6.4 Nonlinear regression for d32 and kinetic constant, Predicted and actual values 117
Figure 6.5 Nonlinear regression for Sb and kinetic constant, Predicted and actual values . 118
Figure 6.6 Flotation rate constant as a function of εg for large particle size-class .............. 119
Figure 6.7 Predicted and actual kinetic constants for the 106/150 µm particle size-class . 120
Figure 6.8 Flotation rate constant as a function of Ib for large particle size-class .............. 120
Figure 6.9 Linear regression for Ib and k for large range particle size-class ...................... 121
Figure A. 1 shows the sampling points from flotation column .... ......................................137
Figure A. 2 schematic of sampler and its position in the column ....................................... 138
Figure A. 3 Example of measured RTD by conductivity cells; Left) tracer impulse in the
feeding point, Right) the detected response of tracer in the tailing point ........................... 139
Figure B. 1 Bubble size variation by shear water to a frit and sleeve sparger ... .................141
Figure B. 2 d32 variations over time .................................................................................. 142
Figure B. 3 a) Variations of d32 with the gas and shearwater rates, b) variations of εg with
the gas and shearwater rates and c) variations of Sb with the gas and shearwater rates .... 145
Figure C.1 Mineral liberation – talc....................................................................................148
xvii
Nomenclature
BSD bubble size distribution f(db) size distribution function
c concentrate assay Ib interfacial area of bubbles
C total concentrate weight Ibc calculated Ib
db bubble size Ibm predicted Ib from d32 and εg
db max maximum bubble size Jg superficial gas rate
db min minimum bubble size k flotation rate constant
d0 bubble size at Jg = 0 µ mean of the BSD
d10 mean diameter of the BSD PC collision probability
d32 Sauter mean diameter of the BSD PD detachment probability
EA attachment efficiency R recovery
Ec collision efficiency Re Reynolds number
Ek collection efficiency Rp particle radius
erf (x) error function Sb bubble surface area flux
εg gas hold-up t assays of the tailing
f assays of the feed tind induction time
F total feed weight tslide sliding time
Fa attachment force σ2 variance of the BSD
xix
“Whatever you look for-will become your core” 1
To my immortal beloved ones: my parents
1 Rumi, (1207-1273) “Rubaiyat”, verse 1815, line 4.
xxi
Acknowledgments
I would like to sincerely thank my supervisor, Professor Rene del Villar, for his guidance,
support, and insightful comments throughout the years. Under his tutelage, I have learned
to question myself and to think more sophisticatedly in order to perform quality research.
As a mentor, his contributions to my professional development are invaluable.
Special thanks are likewise expressed to Professor Jocelyn Bouchard, my Co-supervisor,
for providing perspective that contributed significant to this work. I am grateful for his
diligent efforts to mentor me in many directions throughout my years as a PhD student.
I would like to express my gratitude towards Professor Claude Bazin for their thought-
provoking comments.
I would like to extend my gratitude to Jonathan Roy and Alberto Riquelme Diaz for their
dedicated cooperation at the beginning of this research. Special thanks to Vicky Dodier for
assistance in setting up the apparatus and technical support. I would like to acknowledge
Dr. Massoud Ghasemzadeh Barbara and Professor Carl Duchesne for cooperating on the
statistical analysis of the data.
I acknowledge the financial support received from NSERC (Natural Sciences and
Engineering Research Council of Canada) and Corem, throughout my studies at Laval,
Finally, I am forever thankful for the encouragement and support from my family,
especially my parents, Dr. Damavandi and Dr. Amir Vasebi without whom this endeavor
would not have been possible.
xxiii
Foreword
Chapter 2 is based on the article "Effect of particles on the bubble size distribution and gas
hold-up in column flotation" submitted in October 2014 to the International Journal of
Mineral Processing authored by Ali Vazirizadeh, Jocelyn Bouchard and Yun Chen. This
article presents some explanations for the solid particle effects on bubble size and gas hold-
up through the investigation of micro-phenomena. My role in the preparation of this article
is that of main writer supervised by the co-authors for corrections.
Chapter 3 uses the material of the article "The effect of gas dispersion properties on water
recovery in a laboratory flotation column" accepted and presented by myself at the 2014
IMPC Conference, Santiago, Chile, from 20-24 October 2014. I participated in the writing
as the first author, followed by Jocelyn Bouchard and Rene del Villar who contributed to
the editing of this document. The main objective is to study the effect of the gas rate, gas
hold-up, bubble size and bubble surface area flux on the water recovery to the concentrate.
I performed the laboratory testwork and analysis under the supervision of professors.
Bouchard and del Villar.
Chapter 4 is based on the article "On the relationship between hydrodynamic characteristics
and the kinetics of column flotation. Part I: modeling the gas dispersion". The interfacial
bubble area concept is introduced in this article as a new hydrodynamic variable; its
correlations with the other hydrodynamic variables are presented. This article was accepted
for publication in Minerals Engineering in September (2014). I did all the laboratory, data
analysis and mathematical developments. I participated in the writing as the first author.
My co-authors, Professors Jocelyn Bouchard and del Villar corrected and edited the
manuscript.
Chapter 5 uses the material presented in the article "On the relationship between
hydrodynamic characteristics and the kinetics of flotation. Part II: model validation". The
correlation between the flotation rate constant and particle size as well as some
hydrodynamic variables is investigated in this article. This article was accepted for
publication in Minerals Engineering in November (2014). My contribution as the first
author was performing the testwork, analyzing the data and writing the manuscript. Mr.
Ghasemzadeh Barbara and Prof. Duchesne of Chem. Eng. Dept. contributed to the
statistical analysis using PLS. Professors Bouchard and del Villar supervised the data
analysis, as well as the preparation and edition of the article.
1
Chapter 1 Introduction
1.1 Principles of flotation
Froth flotation is a mineral separation method based on differences in water-repellency
characteristics of the various mineral species contained in aqueous slurry. The hydrophobic
particles attach to gas (generally air) bubbles injected at the bottom of the separation vessel.
As a result of their lower specific gravity, the bubble-particle aggregates ascend to the
vessel slurry surface, whereas the hydrophilic particles, that remain completely wetted, stay
in the liquid phase and therefore are removed from the bottom of the device. Froth flotation
can be practiced for a broad range of mineral separations, as it is possible to use chemical
treatments to selectively modify mineral surface characteristics so that they have the
required hydrophobicity for the separation to occur. Some examples are separation of
sulfide minerals from siliceous gangue, separation of coal from ash-forming minerals and
removing silicates from iron ores. Flotation is also used in chemical engineering fields such
as processing recycled printed papers, where ink (carbon) is separated from paper-pulp; it is
also used in the oil production from tar sands, where oil (carbon derived) is removed from
sand (siliceous gangue). Flotation is particularly useful for processing fine-grained ores that
are not amenable to other physical separation methods such as gravity- or magnetic-based
methods.
1.2 Hydrophobic and hydrophilic particles
In terms of their behaviour in an aerated liquid (water), particles can be classified in two
groups: those that are readily wettable (hydrophilic) and those that are water-repellent
(hydrophobic). If a mixture of hydrophobic and hydrophilic particles is suspended in water
contained in a device where air bubbles are injected, the hydrophobic particles will tend to
attach to the bubbles and float with them to the top of the device (called a flotation cell)
where they accumulate as a persistent froth, heavily loaded with the hydrophobic mineral,
wherefrom it can be removed as a product, usually a valuable mineral concentrate. Since
hydrophilic particles do not attach to the air bubbles, they will remain in the suspension,
flowing down to the bottom exit of the cell, as a refuse (tail).
2
1.3 Use of reagents in flotation practice
With very few exceptions, minerals are rarely suitable for froth flotation, because they are
naturally hydrophilic. Among the exceptions (natural hydrophobic minerals), it can be
mentioned coal and carbon derivatives (like oil, diamonds, graphite), native sulphur,
molybdenite and talc (used in this thesis as the hydrophobic component). Chemical
reagents are thus required to render hydrophobic the target (valuable) hydrophilic mineral
and also to preserve suitable froth characteristics.
1.3.1 Collectors
Collectors control the selective attachment of particles onto the bubble surface. They form a
monolayer on the particle surface, which essentially leads to generate a thin film of
nonpolar hydrophobic hydrocarbons, rendering the mineral surface hydrophobic. Selection
of the correct collector is important for an effective separation by flotation. Collectors are
generally classified depending on their ionic charge. They can be anionic, cationic, or non-
ionic. The anionic and cationic collectors contain a polar part, which selectively attaches to
the mineral surfaces and a non-polar part (organic), which projects out into the solution,
making hydrophobic the particle surface. Non-ionic collectors are hydrophobic products
(hydrocarbon oils, grease), which adhere to the surface forming a physical coat repelling
water. Collectors can either chemically cover the mineral surface (react with) or they can be
bonded on the surface by physical forces.
In the chemical coverage case, ions or molecules from the solution undergo a chemical
reaction with the surface, becoming strongly bonded. This process drastically changes the
nature of the surface. These kinds of collectors are highly selective, as the chemical bonds
are specific to particular atoms. Among them Xanthates are the most commonly used
collectors for sulfide mineral flotation.
In the case of physical coverage, ions or molecules from solution are reversibly linked with
the surface, the attachment is due to electrostatic attraction or Van der Waals bonding.
These collectors can be desorbed from the surface if some conditions change, such as
variations in pH or in the solution composition. The collectors, bonded through physical
coverage, are much less selective than the collectors chemically attached, as they might
3
adsorb on any surface that has the correct electrical charge or degree of natural
hydrophobicity. The collector available for the flotation of oxide minerals (oxyhydryl
collectors) are one of those. Typical examples are the fatty acid salts, e.g. sodium oleate,
the sodium salt of oleic acid.
1.3.2 Modifiers
Modifiers are chemical reagents changing the form the collector attaches to the mineral
surface. They are categorized in: (a) activators, which increase the adsorption of collector
onto a given mineral, (b) depressants, that prevent collector from adsorbing onto a mineral
and (c) pH-regulators.
The latter are probably the simplest type of modifiers, since the surface chemistry of most
minerals is determined by the pH. For example, minerals in general exhibit a negative
charge under alkaline conditions and a positive surface charge under acidic conditions.
Since each mineral change from negative charge to positive charge at some particular pH, it
is possible to control the degree of attraction of collectors to their surfaces simply by pH
adjustments. Acids are used generally to adjust pH in the low range, while alkalis such as
lime (CaO or Ca(OH)2) are used to rise pH. Pyrite can be separated from chalcopyrite by
regulating the suspension pH to an appropriate value.
The possibility of collector attachment increases by means of activators, i.e. activators
facilitate the task of a collector. A classical example of an activator is the use of copper
sulfate as an activator for sphalerite (ZnS) flotation with xanthate collectors. The copper
ions replace the zinc ions on the surface and the xanthate gets fixed more effectively to the
newly formed ”copper” surface, rendering the Cu-coated ZnS particle hydrophobic.
Depressants act by preventing collectors from attaching onto particular mineral surfaces,
thus they show the opposite effect of activators. For example, cyanide ions act as
particularly useful depressant for pyrite (FeS2) during sulfide minerals flotation.
1.3.3 Frothers
Frothers are compounds that help changing the size and stabilizing the generated bubbles,
so that they remain well dispersed in the slurry and form a stable froth layer, which can be
4
removed before bubble bursting. Many types of organic compounds serve this objective.
However, some of them have also collecting properties (e.g. amines, carboxylic acids, etc.).
In such cases, their dosage control might become cumbersome (same reagent having two
different actions). This is why the most commonly used frothers are alcohols, which do not
have collecting properties. The most commonly frother used in plant practice are the
methyl isobutyl carbinol (MIBC) and water-soluble polymers, such as polypropylene
glycols and polyglycols (F150). The polypropylene glycol frothers in particular are very
versatile and they can be manipulated to a wide range of froth properties. Some other
frothers available are natural product such as cresols and pine oils; however most of them
are considered obsolete, because of their cost, and are not as widely used as the synthetic
ones.
1.4 Bubble-particle collection
The basic flotation separation mechanism is hydrophobicity, which is the ability of a given
particle to get attached to an air bubble after a collision. However, such separation
mechanism is not perfect, as some undesired particles (low-grade non-liberated) might
attach to the bubbles or are entrained in the bubble wake entering into the froth zone.
Inversely, some hydrophobic particles might detach from the air bubbles during the
transport of the bubble-particle aggregates to the froth zone or due to bubble coalescence in
it, thus returning back to the pulp phase; also some hydrophobic particles do not attach to
the bubbles because of their size (too small or too large) and the number of available
bubbles (amount of bubble surface).
In a flotation device, a mineral particle is collected by an air bubble through one of two
main mechanisms: 1) particle-bubble attachment due to the particle surface hydrophobicity,
or 2) entrainment into the wake of the bubble and the boundary layer which does not relate
to the particle hydrophobicity.
The particle collection by attachment occurs due to bubble-particle collision followed by
adhesion of the hydrophobic particle on the surface of air bubble. The efficiencies of both
steps determine the final collection efficiency by attachment. However, there is another
micro process to be considered, called detachment. The detachment usually occurs when
5
the particle-bubble is disrupted by flow turbulence or bubble coalescence. Consequently,
the collection efficiency, Ek, is given by:
Ek = Ec EA (1- PD) (1-1)
where Ec is the collision efficiency, EA is the attachment efficiency, and PD is the
detachment probability.
1.4.1 Bubble-Particle Collision
The collision efficiency is the fraction of all hydrophobic particles swept out by the
projected area of bubbles that collide with the bubbles.
Yoon (1993) explained that the collision efficiency is strongly affected by both the particle
size and bubble size as well as by the system turbulence. He introduced a model between
collision efficiency and bubble size and particle size under different hydrodynamic
conditions:
( / )n
c p bP A d d (1-2)
where dp is the particle size diameter, db is the bubble mean size diameter, A and n are
model parameters depending on the flow conditions. Table 1.1 indicates the value of those
parameters for each flow condition, depending on the bubble Reynolds number (Re). He
quantified how increasing the bubble size decreases the probability of collision regardless
of the flow conditions.
6
Table 1.1 Yoon’s model parameters for different flow conditions
Flow condition A n
Stokes 2/3 2
Intermediate 0.723 4
2 15
Re 2
Intermediate 0.56
3 (3 /16)1
2 1 0.249
Re
Re 2
Potential 3 1
Based on experimental observation, Finch and Dobby (1990) indicated that when particle
size increases, collision efficiency also increases, but attachment efficiency decreases. This
is in accordance with Yoon's model (Eq. 1-2) and common knowledge. Dai et al. (2000)
came to the same conclusion as Finch and Dobby (1990), based on reviewing the previous
collision models.
Independently of the size of the particle being floated, large bubbles exhibit lower
collection efficiencies than fine bubbles. However, using very fine bubbles does not
improve the process selectivity because the possibility of particle-bubble collision for fine
and coarse particles are relatively similar (Dobby and Finch, 1987).
Weber and Paddock (1983) studied the particle-bubble collision based on the equation of
motion of spherical particle relative to a spherical bubble rising in an infinite pool of liquid.
According to their study, the collision efficiency is affected by two main steps: first, by
intercepting collisions, occurring for neutrally buoyant particles exactly following the fluid
streams, and second, by gravitational collisions which would occur for particles with
assumed zero dimension and finite settling velocity. The efficiency of both steps increase
by increasing the particle size and decreasing the bubble size.
7
The collection efficiency improves as a result of increasing the collision independently of
the attachment efficiency. Therefore, regarding Weber and Paddock's conclusion, the
collection efficiency would increase by generating finer bubbles.
1.4.2 Particle attachment
A key concept in the estimation of particle–bubble attachment probability is the induction
time, tind, first introduced by Sven-Nilsson (1934). Basically, the induction time is defined
as the sliding time of a particle on the bubble surface required for thinning and draining the
liquid film interposed between the particle and the bubble, until its rupture and actual
contact between particle and air bubble. In the flotation process, the induction time can be
interpreted as a threshold sliding duration (Nguyen and Schulze, 2003). When a particle
meets a bubble, it will first deviate from its initial trajectory, due to fluid forces, and then it
will slide on the bubble surface for a short period of time called tslide. If tslide equals or
exceeds tind, then attachment is expected to occur. The tslide depends on the distribution of
bubble-particle collision angles, the maximum bubble-particle contact angle and the
particle sliding velocity (Finch and Dobby, 1990). The particle collision angle is
determined from the angle made by the fluid streamlines and the bubble in the vicinity of
the bubble surface (Figure 1.1).
The maximum contact angle corresponds to the angle where the radial component of
particle settling velocity (directed toward the bubble surface) is equal to the radial
component of the liquid velocity (oriented away from the bubble surface).
Figure1.1 Distribution of radial components of the particles and liquid velocity on the
bubble surface
θ θmax
Radial component of particle settling velocity
Radial component of the liquid velocity
8
Moreover, Yoon (1993) stated that surface chemistry and system hydrodynamic conditions
have an enormous effect on the induction time, which is mainly determined by the particle
hydrophobicity. According to Yoon, For example, for two same-size mineral particles, the
strongly hydrophobic could have an induction time of around 15 ms, whereas the second
one, weakly hydrophobic, would require 40 ms for induction.
The reduction of particle size, for a given bubble size, may have an effect on flotation
selectivity. Longer sliding times give more chances to any particle to attach to bubbles,
irrespective of their degree of hydrophobicity. In other words, less hydrophobic particles
(containing some gangue at the surface) can have similar chances to attach to the bubble as
highly hydrophobic (richer) particles. Consequently, both sorts of small particles may
attach to the bubble and selectivity is then reduced.
On the other hand, selectivity increases if bubble size diminishes since the sliding time may
be too short compared to the induction time (Weber and Paddock, 1983).
While the collision process is more determined by physical parameters, adhesion is the
result of both physical and chemical factors. Since both the bubble size and particle size
affect collision and adhesion, the effect of these two variables is more pronounced on
collection efficiency.
1.4.3 Particle detachment
A particle will be detached from an air bubble if the detachment force exceeds the
attachment force. The maximum attachment force of a single particle can be calculated
from:
cosa pF =π σ R 1- θ (1-3)
where Fa is the attachment force, Rp is the particle radius, σ the liquid surface tension and θ
the three-phase contact angle (Nutt, 1960).
The detachment force is the sum of gravitational forces, shear forces and external vibratory
forces, the latter depending on the particle mass, the vibration amplitude and frequency
(Cheng and Holtham, 1995). The particle size and flow turbulence in flotation – the latter
mostly determined by the flotation device – affect on each individual detachment force.
9
However, in general, the probability of detachment is less important in a flotation column
than in a mechanical cell, because the flow turbulence in the flotation column is much
lower than in a mechanical cells (Finch and Dobby, 1990). In addition, the probability of
detachment of fine particles (less than 100 μm) is mostly negligible.
1.4.4 Particle entrainment
According to Cilek (2009), the entrainment mechanism not only refers to the recovery of
hydrophilic particles but also to hydrophobic particles recovered without being attached to
air bubbles. The particle entrainment is directly related to the fraction of water reporting to
the concentrate, called ‘water entrainment’. There is agreement in that the wake of the
bubble swarm is responsible for the water entrainment to the froth. The volume of the
bubble wake is a function of the bubble size and its rising velocity, as well as the liquid
viscosity, therefore the water entrainment should be determined by such variables.
The role of frother type and concentration (as the manipulated variable in flotation) on
water entrainment can be illustrated through its effect on the bubble size and velocity
(Ekmekçi et al., 2003). Since frother modifies the bubble size and its velocity, it indirectly
determines the bubble wake and consequently the volume of water reporting to the
concentrate.
In addition, it appears that gas dispersion properties in general (such as the bubble size, gas
hold-up, etc.) play a major role, similar to the frother effect, on the particle entrainment
through their effect on bubble wake (Nelson and Lelinski, 2000; Phan et al., 2003;
Rodrigues et al., 2001; Yoon, 2000).
Yianatos et al. (2009) observed a selective entrainment of fine particles (less than 45 μm) in
a large industrial cell as. On the other hand, the recovery of coarse particles (larger than
150 μm) by entrainment is considered negligible, less than 0.1% (Yianatos et al., 2009;
Zheng et al., 2006). Experimental evidence also confirms that if the liberation size is
smaller, grinding product (cyclone overflow) must be finer which might imply a greater
content of fine gangue particles, easier to be entrained than coarser ones (Guler and
Akdemir, 2012).
10
Regarding to the micro-processes presented in this section (1.4), it seems that the bubble
size and particle size, as well as the flow conditions in the flotation device, play the most
important role on particle recovery when particles have the same hydrophobicity.
1.5 Flotation devices
The flotation process is accomplished in a device having two main roles: keeping the pulp
in suspension and providing air bubbles. The device should also provide a mean to
evacuate the two products: concentrate (loaded froth) and tails (depleted mineral pulp).
There are two main flotation devices well accepted in the industrial practice, flotation
columns (and other similar devices) and mechanical cells. The latter are usually connected
in banks having enough cells to assure the required particle residence time for adequate
recovery. Subsequently, various banks, each having its particular targets of concentrate
grade and recovery, are interconnected to form a flotation circuit, capable of attaining the
desired overall metallurgy (product).
1.5.1 Mechanical cells
A mechanical flotation cell is basically a cylindrical or rectangular vessel or tank fitted with
an impeller. The impeller function is to mix thoroughly the slurry to keep particles in
suspension, and also to disperse the injected air into fine bubbles, providing conditions
promoting bubble-particle collisions. It is worth mentioning that mechanical cells operate
either on self aspiration mechanism (no air feed rate control) or controlled compressed air.
Formed bubble-particle aggregates rise up through the cell by buoyancy and are removed
from it into an inclined drainage box called "concentrate launder". Water sprays will help
breaking the aggregates to make possible the pumping of the concentrate downstream.
Particles that do not attach to the bubbles are discharged out from the bottom of the cell.
A mechanical cell requires the generation of three distinct hydrodynamic regions for
effective flotation. The region close to the impeller encompasses the turbulent zone needed
for solids suspension, dispersion of gas into bubbles, and bubble-particle interaction. Above
the turbulent zone lies a quiescent zone where the bubble-particle aggregates moves up in a
relatively less turbulent area. This zone also helps in sinking the amount of gangue minerals
that may have been entrained mechanically. The third region above the quiescent zone is
11
the froth zone serving as an additional cleaning step as a result of bubble coalescence and
other phenomena (Fuerstenau et al., 2007). Figure 1.2 shows a typical schematic of
mechanical cell.
Figure 1.2 Mechanical cell schematic (Fuerstenau et al., 2007)
1.5.2 Flotation columns
A flotation column is typically a tall vertical cylinder or rectangular with no mobile parts
(agitator), fed with a mineral pulp (top third of column), air bubbles is injected (always
generated by controlled compressed air) at its very bottom. These bubbles rise-up in
counter-current with the descending flow of pulp, so that the contained hydrophobic
particles are able to attach to the air bubbles. The so-formed bubble-particle aggregates are
carried upwards by a buoyancy effect. The zone where this process takes place is called the
collection zone, corresponding to 75% to 90% of the total column height. The hydrophilic
particles and some non-collected hydrophobic particles move downwards throughout the
collection zone, being finally discharged through the tailing outlet at the bottom of the
column. The ascending bubble-particle aggregates accumulate in the upper part of the
12
column before overflowing into a launder as a concentrate. This zone, which stands from
the pulp-froth interface up to the top of the column, is called the froth zone or cleaning. Its
thickness generally varies between 10% and 25% of total column height (Zheng, 2001 ),
although in specific applications it can be as shallow as 40 cm. Figure 1.3 shows a
schematic of a flotation column unit.
Figure 1.3 Flotation column schematic
1.5.2.1 Bubble generating systems
The bubble size is essentially determined by the type of frother and the mechanism used for
bubble generation. It can be modified through the frother concentration, and gas feed rate to
the device. The bubble generation system is another feature which distinguishes flotation
columns from mechanical cells. In the case of flotation columns the most commonly used
for bubble generation method is through internal spargers near to the bottom of the device.
There are two categories of internal spargers: porous spargers and single or multinozzle
spargers (Finch and Dobby, 1990). Whereas the former are used at laboratory scale, the
second method is the most common at industrial scale. A variant has been introduced in the
FeedConc
Tail
Water
Air
Wash
Water
13
so-called reactor-separator columns, in that gas and slurry are brought into intense contact
in an external section (called reactor), with a very short residence time. Then the aeratered
pulp passes to a quiescent zone (separator) where bubble aggregates separate from settling
gangue particles, each stream exiting the separator at a different outlet. The amount of
slurry introduced to the sparger is usually a small fraction of tails flow rate (around 10%)
(Massinaei et al., 2009).
In the mechanical cells, air enters to the device through a concentric pipe surrounding the
impeller shaft. The rotating impeller tips create a high-shear zone where air is broken up
into a dispersion of bubbles, when passing through a static set of bars around the impeller.
The bubbles are deviated outwards from the impeller tips and dispersed throughout the
solid-liquid mixture (slurry) into the next zone (quiescent) of the cell (Fuerstenau et al.,
2007).
1.5.2.2 Froth zone, wash water and bias rate
Another feature distinguishing the flotation column from the mechanical cell is the addition
of wash water above or slightly inside the froth zone. A fraction of this flow of fresh water
travelling down the froth zone, would wash-out the hydrophilic particles entrained into the
froth zone, thus avoiding their recovery to the concentrate. The remainder of the water
injected helps the concentrate overflowing into its launder. To assess the wash water
cleaning performance, the bias rate concept – i.e. the fraction of the wash water flow rate
going downwards through the pulp-froth interface – was introduced by Finch and Dobby
(1990).
14
Figure 1.4 Three layers of the froth zone
The froth zone normally consists of three regions: an expanded bubble bed (next to the
interface), a packed bubble bed above the previous one and a conventional draining froth at
the top (Figure 1.4). As for practical considerations (plugging of water nozzles) wash water
is sometimes sprayed above the froth and therefore the conventional draining froth does not
establish properly (Finch and Dobby, 1990).
Bubbles pass through the collection zone and enter the expanded bubble bed after colliding
with the first layer of bubbles, which defines a very distinct interface. These incoming
bubbles have a relatively homogeneous and small size and remain spherical all the way up.
Bubbles colliding with the interface, generate shock pressure waves promoting collisions
up through the expanded bubble bed. This phenomenon seems to be the main cause of
bubble coalescence, promoting film thinning and finally its rupture.
The packed bubble bed region expands to the wash water inlet level. The fractional liquid
content is lower than in the expanded bubble region and bubbles keep a relatively spherical
shape, but the range moves towards larger bubbles. Bubbles rise upward with close to
plug-flow conditions, thus promoting a good distribution of the wash water.
Expanded bubble bed
Packed bubble bed
Conventional draining froth
15
Above the wash water inlet level lays the conventional draining froth, therefore the net flow
of water is ascending here is negative. The main aim of this region is to convert froth
vertical motion into horizontal motion to collect the froth.
1.6 Flotation modeling
1.6.1 Performance justification
There is no universal way for expressing the effectiveness of a separation, but several
useful indices exist for evaluating the quality of the flotation process. The following are the
most commonly used.
Ratio of concentration: defined as the weight of feed divided by the weight of concentrate,
that is:
ratio of concentration = F/C (1-4)
where F is the total feed weight and C is the total concentrate weight .
The drawback of this performance index is the need of weight values. Although from
laboratory experiments these data can be obtained, in the industrial practice it is unlikely
that the ore is weighed, only assays are available. However, is possible to express the ratio
of concentration in terms of ore assays. From the definition of the ratio of concentration
(F/C) and using the following mass balance equations, it can be calculated:
F= C+T (1-5)
Ff=Cc +Tt (1-6)
where f, c, and t are assays of the feed, concentrate, and tailings, respectively. Rearranging
Eq. (1-5) and Eq. (1-6) gives:
c tF
C f t (1-7)
Percent recovery: The percent recovery is the percent of mineral in the original feed that is
recovered in the concentrate. By means of weights and assays it can be calculated by:
16
100C c
RF f
(1-8)
or by replacing weights by assays (from material balances):
100
f tcR
f c t
(1-9)
where R is the percent recovery.
Enrichment ratio: The enrichment ratio is directly calculated from the assays as c / f.
Mass pull: The mass pull is the inverse of the ratio of concentration, thus:
f tC
F c t (1-10)
Grade-recovery curves
Although the above mentioned indices are useful for comparing flotation performance for
different conditions, a “grade–recovery curve”, constructed from concentrate grade and
recovery values at different operating conditions, is a more useful tool for inferring the
optimal operating conditions. This is a graph of the recovery of the valuable mineral against
the concentrate grade for given operating conditions, and it is particularly practical for
comparing separations where both the grade and the recovery are varying. A set of grade–
recovery curves is shown in Figure 1.5.
17
Figure 1.5 Typical grade–recovery curves for froth flotation
Controlling of grade and recovery in flotation processes has received significant attention
from researchers in the past years. Since more than hundred variables affect the flotation
process (Arbiter, 1962), knowing which variables have more influence, can help controlling
flotation performance variables, i.e. the grade and recovery. On the other hand, finding the
relationship between manipulated variables and controlled variables is imperative to find
out set-points leading to improved metallurgy. Therefore, some models for recovery
parameters have been proposed. The form of such models and their definition are presented
in the next section.
Another important parameter often used for cell/column design is the so-called carrying
capacity. The carrying capacity of the cell is calculated based on the recovered mass of
particles, both hydrophobic and hydrophilic, per unit cell surface per unit of time. This
Improving Performance
Pure
Mineral Assay (%) Feed
100
0
Reco
very
(%)
18
variable is directly related to the bubble surface available for collection, and therefore to the
bubble size.
2
1( )min
dC kgCarrying Capacity
dt A m
(1-11)
where A is the cross-section area of the flotation cell and t is the time unit and C is total
recovered mass.
1.6.2 Kinetic constant modeling
To model the flotation process, two main approaches are possible: the empirical modeling
and the phenomenological modeling.
In empirical models, input and output variables are related through suitable mathematical
equations. Statistical methods are used to define dependent and independent variables and
to estimate the curve fitting parameters. Obtained parameters do not necessarily have any
physical meaning and they are valid just for the limited conditions in which the model was
calibrated.
Phenomenological modeling provides an explanation for causes and effects, which are
related to physical and chemical conditions of the process. Phenomenological models are
divided in three main categories: kinetic, probabilistic and population balance models.
Probabilistic models are based on some sub-processes such as collision, adhesion and
detachment and can be used as a bridge between micro- and macro-models. Kinetic models
use the chemical reactor analogy and consider flotation as a reaction between bubbles and
particles (Polat and Chander, 2000). A good understanding about mixing conditions in the
flotation process is also required to obtain a reliable kinetic model. For instance, mixing
conditions in a laboratory column (5 cm diameter) are close to plug flow pattern, whereas
in industrial columns they are something between plug flow and perfect mixing patterns
(Finch and Dobby, 1990).
There is an expression relating the collection efficiency to the flotation rate constant. For
the situation of gas bubbles rising through a column of water containing hydrophobic
particles at a concentration cp and having collection efficiency Ek, and given a cubic volume
19
of water with side dimension L, the particle collection process is represented by the rate of
particle removal from the slurry by air bubbles. In other words, it can be calculated from
the rate of particle removed per bubble times the number of bubbles.
At a gas rate Qg and a slip velocity Usg (the velocity of gas relative to slurry), this
expression is equivalent to:
23
34 6
p gbsg p k
sgb
dc Qd LL U c E
dt Ud
(1-12)
After cancelling out some terms:
1.5
p g k
p
b
dc J Ec
dt d (1-13)
where 2
g
g
QJ
L .
Eq. (1-13) is equivalent to the expression of first-order rate process where the first order
rate constant kc is given by:
1.5 g k
c
b
J Ek
d (1-14)
It is very well accepted that the flotation process follows a first order kinetics with respect
to the concentration of the floatable particles c as indicated in equation (1-15).
dc
k cdt
(1-15)
By integration and introduction of the recovery definition, it is possible to obtain the
following expressions, respectively for plug flow conditions (Eq.1-16) and perfect mixing
conditions (Eq.1-17):
(1 )
ckR e
(1-16)
1
c
c
kR
k
(1-17)
20
where τ is the particle mean residence time. Equation (1-16) for plug flow and Equation
(1-17) for perfect mixture indicate that increasing the flotation kinetic constant leads to an
increase in the recovery.
In the case of plug flow conditions (or a batch flotation processes), equation (1-16) is
always valid since the kinetic constant is invariant. However in actual flotation conditions,
particle size and bubble size are present as a population distribution, which leads to
consider a kinetic constant distribution for proper flotation modeling. It is also possible
that the industrial flotation device exhibits a mixing behaviour which is neither plug flow,
nor perfect mixing, in which case the residence time distribution, obtained through tracer
tests, must be used.
In this regard, Polat and Chander (2000) applied a generalized form of equation (1-16)
shown in
0 0
(1 ) ( ) ( )ktRe F k E t dk dt
R
(1-18)
where R is the mineral recovery at time t and R∞ represents the ultimate recovery at infinite
time, E(t) is residence time distribution and F(k) is the kinetic-constant distribution function
for a continues process. This formulation is more appropriate since both the kinetic
constant distribution and the residence time distribution are being considered.
It is worth mentioning that the presented recovery models are for the collection zone only;
the froth zone recovery has not been considered so far. In fact, as a result of various events
taking place in the froth zone (e.g. bubble coalescence) some hydrophobic particle may
detach from the bubbles and return to the pulp zone, making the overall recovery lower
than that of the collection zone, calculated with the previous equations. To eliminate the
froth zone effect on the overall flotation recovery, a very shallow froth height (less than 15
cm) has to be used in the experimental determination of the kinetic constant, so that the
overall flotation recovery can be assumed to be equal to that of the collection zone. Since
the evaluation of k always implies a series of restrictions or inaccuracies, such as the model
will require to be compared to a reference value that is inaccurate. Based on, given all these
21
inaccuracies, a different approach would be worth to be explored. This will be tackled later
on in this thesis.
1.6.3 Residence time measurement in a flotation column
One of the most used methods for determining the residence time distribution (RTD) in
liquid or pulp systems, is the injection of a know amount of tracer (liquid or solid) at the
column feed port and to track its concentration with tailing flow as function of time.
Mean residence time can be calculated based on the time variation of tracer concentration
through:
exp
( )
( )
i iierimental
ii
t C t dt t C
CC t dt (1-19)
where τexperimental is the mean residence time (min), C(t) is the tracer concentration at time t.
The expected mean residence time is simply obtained from:
exp
eff
ected
V
Q (1-20)
where Veff (m3) is effective cell volume (collection zone without gas) and Q (m3/h) is feed
flow rate.
Various sorts of tracers have been used for RTD measurement, such as ionic salt tracers
(NaCl and KCl) and radioactive tracers (Br-82) for liquid RTD and MnO2 as a solid particle
tracer. However, liquid tracers are not as accurate as solid tracers for modeling of flotation
because the aim is to track particle behavior in flotation conditions through studying the
flotation rate. However, ionic salt tracers are simpler to use as their concentration can be
detected by conductivity measurements. These tracers are generally used for liquid RTD
measurements as a mean to evaluate the mixing conditions in the flotation device.
Radioactive tracers provide more accurate results since less amount of tracer is required
and the tracer is tracked by a non-invasive sampling system (measuring the radiation in the
output flow).
22
The best tracer for measuring the residence time of solid particles are the solid tracers, as
long as they have the same physical features (density, size and hydrophobicity) as the target
solid particles. There are some reports on the application of solid tracers like MnO2 for
tracking particle behavior and particle RTD in the flotation cells (Finch and Dobby, 1990).
Yianatos and Bergh (1992) have systematically used the radioactive tracer technique for
coding the different solid particles and measuring the RTD of hydrophilic particles in
industrial flotation columns and cells. More recently, Cole et al. (2010) presented a
Positron Emission Particle Tracking (PEPT) method which can be applied to particles in
froth flotation systems to observe the behavior of individual hydrophilic particles in a
mixed particle–liquid–gas system.
1.7 Gas dispersion properties
Gas dispersion properties have proved to be a key feature of the flotation process. Among
them, the most relevant are: the gas hold-up, superficial gas velocity, bubble size and
bubble surface area flux. This latter has received considerable attention and has been
reported to linearly correlate with the flotation rate constant and bubble carrying capacity,
both related to flotation performance (Gorain et al., 1998).
1.7.1 Gas hold-up (εg)
The volumetric fraction of gas in a given volume of the device is called the gas hold-up.
For a given column section, it can be estimated through the following equation:
g
Volume of bubbles
Volume of aerated pulp (1-21)
where volume could refer to the total volume of a given zone, in which case we are talking
of an overall gas hold-up) or part of it thus a local gas hold-up. The εg value strongly
depends on the prevailing gas rate and bubble size values (Gorain et al., 1998).
1.7.2 Superficial gas velocity (Jg)
The volumetric gas flow rate (cm3/s) divided by the cross-sectional area of the device (cm2)
is called the superficial gas velocity (cm/s), i.e.
23
g
g
QJ
A (1-22)
where Qg is the gas volumetric flow rate and A is the cell cross-sectional area. This
definition is quite useful, since is independent of the device size, therefore it is a scalable
value. For flotation devices the recommended Jg value should vary from 1-2 cm/s.
1.7.3 Bubble surface area flux (Sb)
Bubble Surface Area Flux (BSAF or Sb) is defined as follows:
surface area of bubbles generated per unit of time
cross sectional area of the columnbS (1-23)
Assuming that bubble sizes can be presented in m different discrete bubble sizes, then Eq.
(1-23) can be expressed mathematically as:
1
m
i bi
ib
cell
n S
SA
(1-24)
where Acell is the cell cross sectional area, ni is the number of bubbles of size i generated per
unit of time and Sbi is the surface area of a bubble of size i. Similarly, the volumetric gas
flow rate can be defined as follows:
1
m
g i bi
i
Q nV
(1-25)
where Vbi is the volume of an air bubble of size i. Then, Sb can be written as:
1
1
m
i big i
b m
celli bi
i
n SQ
SA
nV
(1-26)
Assuming that all bubbles are spherical, then2
bi iS d and 3 6bi iV d . Since
g g cellJ Q A , Sb can be calculated in terms of the superficial gas velocity Jg and a mean
bubble size, as
24
32
6 g
b
JS
d (1-27)
where d32 is the Sauter mean diameter defined as:
3
132
2
1
m
i i
i
m
i i
i
n d
d
n d
(1-28)
Another way to calculate d32 from a set of bubble sizes is:
3
132
2
1
b
b
N
i
i
N
i
i
d
d
d
(1-29)
where Nb represents the total number of collected bubbles.
1.7.3.1 Experimental models for Sb
Bubble size measurement in flotation devices is difficult enough to induce researchers to
look for simpler alternatives, among which the empirical modeling of Sb has been
considered as the most appropriate approach.
An empirical model was developed by Gorain et al. (1999), using extensive pilot and
industrial tests obtained with different cell types and sizes. It allows predicting the Sb in
mechanical flotation cells by
0.44 0.75 0.10 0.42
80123 ( )b s g sS N J A P (1-30)
where Ns is the impeller peripheral speed, Jg is the superficial air velocity, As is the impeller
aspect ratio and P80 is feed 80% passing size.
This model was shown to be acceptable for 20-150 µm particles in rougher, scavenger and
cleaner mechanical cells, but was not validated for flotation columns (Gorain et al., 1999).
Heiskanen (2000) evaluated Gorain’s model and formulated some criticisms. In particular,
he mentioned that "the measurement and computation of superficial gas velocity, and
partially also the bubble size may be biased in some conditions. This makes the bubble
25
surface area flux behave such that the final outcome is in doubt." Some other criticisms
were:
- the model validation did not address the different particle sizes at all;
- the results from Gorain’s evaluation show that increasing bubble size gives a higher
flotation rate constant and that poor dispersion gives a good flotation response, both
results being in contradiction with earlier finding and industrial experience;
- lastly, he suggested that the bubble surface area flux would need more validation using
different ore types and that the linear relationship between flotation rate constant and Sb
also required more research.
Finch et al. (2000) suggested another empirical relationship, this time to predict the Sb from
gas hold-up measurements:
5.5b gS (1-31)
This model was tested for mechanical cells and flotation columns, both at laboratory and
industrial scale, and it was deemed to be valid for Sb lower than 130 s-1 and εg lower than
25%. The problem of this model is the method used for bubble size estimation (drift flux
analysis), which does not have a good accuracy. This Sb prediction model was proposed
based on the fact that gas hold-up measurement is easier than that of the bubble size.
A third expression can be developed using the empirical model proposed by Nesset et al.
(2006) for estimating the Sauter mean bubble diameter from superficial gas velocity
measurements. They found that the d32 increases when Jg is increased according on:
32 0
n
gd d C J (1-32)
where C and n are the model parameters and d0 is determined by extrapolation of the d32
graph for Jg = 0. By replacing the d32 in equation (1-27), the following expression for Sb can
be obtained:
0
6 g
b n
g
JS
d C J
(1-33)
26
As can be seen, none of the proposed models consider the actual bubble size distribution
even though it can now be measured and consequently better definition of Sb could be a
targeted.
1.7.4 Bubble size distribution
The bubble size distribution has lately received a lot of attention from researchers.
However, very few mention the effect of bubble size distribution to the flotation process
performance. In fact, bubbles generated in flotation cells or columns show a wide range of
sizes i.e. a bubble size distribution, but most if not all available models use a mean value as
‘the bubble size’, which results in overlooking some distribution features.
Figure 1.6 (Maldonado et al., 2008c) illustrates how different bubble size distributions may
lead to a same Sauter diameter.
Figure 1.6 Different bubble size distributions all having the same d32
(Maldonado et al., 2008c).
Wongsuchoto et al. (2003) showed the relationship between the shape of bubble size
distribution and the superficial gas velocity. They reported that for superficial gas velocities
in the range of 20 to 40 cm/s, bubble size distributions correspond to a normal distribution
in their case of study.
d32 (mm) d32 (mm)
d32 (mm) d32 (mm)
27
Grau and Heiskanen (2005) presented a study on bubble size distributions in laboratory
mechanical cells. Their main objective was to evaluate the influence of operating
conditions, such as the pulp density, frother concentration, air flow rate and impeller speed,
on the bubble size distribution. The results were modeled by applying different distribution
functions (Rosin-Rammler, Tuniyama-Tanasawa, log-normal and upper-limit). It was
shown that the upper-limit function better fit the experimental data as compared to the other
functions. However, there was not a unique distribution function which could represent all
the generated bubble size distributions for all the flotation devices tested.
These modeling results for some gas dispersion properties have been used for optimizing
the flotation performance regarding particular types of particles. In the next section, some
previous research results about influence of gas dispersion on flotation are presented.
1.7.5 Gas dispersion properties range and effects
It has been demonstrated that excessive gas rate might lead to a loss of bubbly flow regime,
pulp-froth interface and positive bias in the case of flotation columns (Hernandez-Aguilar
et al., 2005). In fact, when the gas rate is increased, it first affects the bias rate, which might
turn from positive to negative, but then this is followed by the perturbation of the bubbly
flow regime inside the column and finally the pulp-froth interface might be lost, i.e. it
becomes impossible to detect.
Based on experimental evidence, Deglon et al. (2000) reported that the bubble surface area
flux should stand in the range of 50 to70 s-1. From laboratory scale work using two types of
minerals, Hernandez et al. (2005) demonstrated that Sb values up to 50 s-1 could be used in
full-scale mechanical cells before deteriorating the flotation results.
Yianatos and Henriquez (2007) made a survey of typical ranges of bubble size and
superficial gas velocity, from micro-scale to large industrial flotation cells and columns.
Their work showed that for typical gas velocities (Jg = 1.2 cm/s), the optimal bubble size in
the collection zone should be in the range of 1 to 1.5 mm to optimize the bubble surface
area flux, that is Sb should then be in the range 50 to 100 s-1. They concluded that to have
an acceptable flotation performance, bubbles should be generated in this particular size
range.
28
If the bubble size stands in the range of 0.5 to 1 mm, three main effects in flotation were
observed, namely
- the loss of the pulp-froth interface;
- lower bubble surface area flux;
- lower mineral carrying capacity.
On the other hand, increasing bubble size up to 1.5 to 2 mm led to
- lower bubble surface area flux;
- lower mineral carrying capacity;
- greater disturbances at the interface level (“boiling” effect);
- more mineral entrainment into the froth.
Increasing the Sb to reach higher flotation rates might encounter some limitations with
respect to increasing the bubble size and the gas rate. As a result, attention should be placed
on the most adequate range of bubble size and gas rate.
Beside these considerations about bubble size range, it is worth considering some studies
on the effect of other variables on the bubble size. The following paragraphs present some
of these works.
Gorain et al. (1999) reported that the particle size has an important influence on the bubble
size, and thus on Sb. Therefore, they considered the particle size as an independent variable
in their Sb model. The effect of other physical and chemical variables on the bubble size
was also reported. For instance, an increase in the pulp viscosity, pulp density and air flow
rate or a decrease in temperature is reported to produce larger bubbles. Bubble size could
decrease if frother concentration increases; on the contrary collector dosage does not have
any major effect on the bubble size. However it’s not clear whether the reported effect of
the particle size is due to changes in the pulp density (solids concentration) or pulp
viscosity (related to the presence of fine particles and therefore more particles for a given
tonnage) or due to the size of the particles itself. From the analysis of their study, it can be
concluded that the proposed Sb model needs more research to clearly ascertain the effect of
these variables.
29
On the same subject, O’Connor et al. (1990) experimentally demonstrated that using
particles smaller than 38 μm is more efficient for generating smaller bubbles than using
particle larger than 166 μm. However, this result might need more investigation because the
degree of control over other influential variables during the study leading to this conclusion
seems doubtful. It was also observed that the bubble size increases with increasing pulp
density through the addition of quartz particles, was also reported by Grau and Heiskanen
(2005). It is not clear though, whether the effect of the particle size occurs during or
following the bubble generation process.
1.8 Relation between the kinetic constant and gas dispersion properties
Gorain et al. (1998) proposed a linear relationship between the flotation rate constant and
superficial area flux for different particle sizes in a pilot mechanical cell. They found that
the slope of the obtained straight line was different for different size fractions: for smaller
particle sizes (less than 38 µm), the slope increases and vice-versa. This observation was
validated in mechanical cells in the pilot scale and the obtained model was used for
mechanical cell characterization.
They also reported a strong relationship between k and Sb at different froth depths. The
relationship was found to be linear for shallow froths (7 cm). For intermediate (30 cm) and
deep froths (45 cm), the relationship was found to be non-linear, therefore the froth depth
plays an important role on the overall kinetics.
The existence of a linear relationship between the flotation rate constant and froth depth
was also reported by Vera et al. (1999) and a linear model for this relationship was
presented. They also claimed that the collection-zone rate constant of the two evaluated
minerals (chalcopyrite and pyrite) increases as air flow rate increases, at the expenses of a
reduction in froth-zone recovery. This was associated to a detachment of particles from the
bubbles in the froth-zone. The impeller speed also increased the collection-zone rate
constant and decreased the froth-zone recovery for both minerals. Finally, they showed that
the solid concentration and the collector dosage have an important effect on the collection-
zone rate constant, both being related to the mass transfer rate from the collection-zone to
the froth-zone. They appeared to have an indirect influence on the froth-zone behavior by
increasing froth stability.
30
Heiskanen et al. (2000) performed some experiments proving that the rate constant starts
decreasing for coarse particles at high superficial gas rates (more than 2 cm/s) and that fine
particles (20-25 µm) require much higher superficial gas rate for effective flotation.
Before them, Yoon (1993) reported that decreasing the bubble size is a more effective mean
than increasing the gas rate to reach faster kinetics.
On the subject of the relation between the particle and bubble size, Diaz-Penafiel and
Dobby (1994) demonstrated that the kinetic constant for a laboratory flotation process
using small-bubbles (mean bubble size smaller than 1.2 mm, estimated using the drift flux
analysis) varies for different particle sizes (silica). Their work indicates that, for the same
bubble size, the flotation kinetic constant increases when particle size increases up to a
maximum value and then decreases.
Moreover, the flotation kinetic constant obtained using larger bubbles (mean bubble size
larger than 2 mm) exhibits the same trend than that with small bubbles up to a maximum
value, wherefrom further increase of particle size does not affect the kinetic constant value.
Their study has shown that small bubbles are not effective for very fine particles (smaller
than 5 µm) and the influence of the bubble size on flotation is greater as particles become
larger than this size.
1.9 Problems associated to the use of a single Sb value
So far, the Sb has always been considered as a global variable developed using an average
value of the BSD (d32), a value that can be originated from quite different distribution
shapes (Figure 1.6). This practice disregards other distribution features contained in the
measured data (variance, multimodes, etc.).
The application of a unique value of Sb for the determination of the flotation rate constant
values certainly leads to errors. For instance, a given Sb value can be obtained from
different combinations of mean bubble size and superficial gas velocity as demonstrated by
Figure 1.7. It is doubtful that all different pairs of bubble size and superficial gas velocity
leading to a single Sb value will transpose into a same kinetic constant.
31
Figure 1.7 Different pairs of bubble size and gas velocity leading to a unique BSAF value
This hypothesis has been proven in some of the tests completed in this PhD work (full
results will be presented in Chapter 5 and 6). In the two tests (flotation of talc as a
hydrophobic mineral) run under the same operating conditions, by changing the distribution
shape rather than the d32, two different values of the kinetic constant were estimated as
shown in Table 1.2. The only difference between them is the shape of bubble size
distribution as depicted in Figure 1.8.
Table 1.2 Example of a same Sb originating two different rate constants.
Particle size Jg (mm/s) d32(mm) Sb (1/s) τ (s) R∞ k (1/min) Recovery %
+75/-106 10 1.34 44.82 300 1 0.26 73
+75/-106 10 1.35 44.57 300 1 0.24 70
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2
10
12
14
16
18
20
d32
(mm)
Jg (
mm
/s)
Sb=120 (1/s)
Sb=100 (1/s)
Sb=80 (1/s)
Sb=60 (1/s)
32
Figure 1.8 Two BSDs having same d32 originate different kinetic constant
Although the variation in flotation rate value is rather small, it affects the flotation recovery
by 3% (in this example).
From this review of previous works, it seems that the influence of the bubble size (and
particularly the BSD) and particle size on the flotation rate constant has not been
comprehensibly studied, nor the modeling of the BSD.
Bridging the gap between micro and macro flotation models should also be addressed for a
better understand of the flotation process. Consequently, determining the relation between
collision, attachment and detachment efficiencies and the prevailing cell hydrodynamic
conditions having an effect on the kinetics, can be useful.
The effect of solid particles on the bubble size distribution and gas hold-up, as well as the
correlation between the bubble size distribution and gas hold-up should be considered as
well. Developing a better characterization of prevailing flotation conditions will help
understanding how hydrodynamic variables influence the particle collection process.
Finally, optimizing the collection efficiency (improving collision and attachment) does not
seem to improve the selectivity of the flotation process. The effect of hydrodynamic
variables on the flotation selectivity should thus be studied as well. These effects are partly
reflected on the amount of water reported to the concentrate (water recovery).
10 20 30 40 50 600
0.01
0.02
0.03
0.04
0.05
0.06
Bubble size (mm)
De
ns
ity
k= 0.24 (1/min)
k= 0.26 (1/min)
0.5 1.51 2 2.5 3
33
Based on the presented problems the objectives of this study have been defined and
presented hereafter. However some assumptions made should first be presented and
explained.
1.10 Assumptions made in this experimental work
1.10.1 Hydrophobicity
Talc is a natural hydrophobic mineral. As such, it was selected in order to disregard the
effect of variable hydrophobicity and collector efficiency in the evaluation of the
experimental rate constant. This way, the pulp could be reused continuously without the
above-mentioned limitations.
1.10.2 Solid particle residence time
The measured residence time of water is not the same as that of solid particles. The particle
residence time in flotation columns depends on the particle settling time, which is a
function of particle size, particle density, hydrophobicity, pulp percent solids, pulp
viscosity and the mixing conditions of the column. In the present study only the particle
size was modified, all remaining factors were kept constant during the experiments, thereby
particle residence time is only depended on particle size.
The particle size can be determined by means of screening, or measuring the settling time.
These two methods however do not address to the same size. Table 1.3 indicates the
differences between the measured values via either method.
Based on the concept of residence time measurement the measured particle size by settling
should be used however owing to hydrophobicity of talc, this residence time cannot be
accurate (attachment and detachment occur in the cell). It is worth mentioning that in this
research the particle size are defined as fine, middle, coarse and mixed as indicated on
Table 1.3.
34
Table 1.3 Talc particle sizes measured by twice wet screening and Malvern 2000
Particle
size
Measured particle size
by screening Measured particle size by settling
below screen size/ top
screen size (μm)
d80=80% of particles
passing this size (μm) below size/ top size (μm)
Fine 53/75 86 0.5/208
Middle 75/106 113 0.6/275
Coarse 106/150 137 0.6/416
Mixed 53/150 112 0.5/416
As results of obvious restrictions at Université Laval facilities, applying the most accurate
methods for residence time measurement, such as radioactive tracers, is not possible. The
only possible method is through injection of ionic solution tracers (Appendix A) and
measuring their concentration at the tailing port through electrical conductivity. Obviously,
this implies assuming that the particle residence time is equal to the water residence time.
1.10.3 Ultimate recovery (R∞)
Since all hydrophobic particles used in these experiments are fully liberated (pure talc) and
there is enough bubble surface for particle attachment, we can assume that R∞ is always
equal to 1. Figure C.1 (Appendix C) shows that all used (sizes) talc particles are free.
Moreover, Figure 1.9 shows a white spot in the center of loaded bubbles, which proves that
the light beam passes through a transparent part of bubble surface, indicating that bubbles
have enough surface area available for particle attachment. It should be noticed that
Figure 1.9 corresponds to a condition where a minimum bubble surface was generated.
Therefore, in cases where more available bubble surface exists, this assumption should hold
as well.
35
Figure 1.9 Light beam passing through loaded bubble
1.10.4 Froth depth and its effect on the final recovery
A fraction of the collected particles usually drains back from the froth zone. Therefore, the
overall particle recovery is not the same as the collection zone recovery. The amount of
drop-back depends on the bias rate and froth depth. If a very shallow froth zone is set in the
process control system and wash water is not introduced to the flotation column, we can
assume that the overall recovery is approximately the same as the collection zone recovery.
Therefore, a pulp sample taken from the concentrate launder can be considered as
representative of the collection zone concentrate.
In this research, samples were taken from the concentrate launder and at 10 cm below the
froth interface (this position was chosen to eliminate the effect of interface turbulence).
However for the purpose of flotation rate calculation, only overall concentrate values were
retained for data reconciliation using mass balance conservation equations.
1.11 Experimental set-up and results
The experimental work was conducted on a flotation column made of 5.6 cm internal
diameter acrylic tubes for a total height of 650 cm, continuously fed with a mineral pulp
from a 360 liter stirred plastic reservoir, where both column products were recycled to
assure a continuous operation. Air bubbles were provided to the column through a frit-and-
sleeve sparger placed at its bottom. This device, along with a given frother dosage, allowed
the modification of bubble size. The produced bubbles were sized by an adapted version of
McGill bubble viewer, for on-line measuring purposes. The used image processing method
36
allowed the detection of close to 100% of the bubbles present in the pictures taken, a major
improvement with respect to the presently used method. It is detailed in section 2.2. Froth
and pulp conductivities were measured at various heights of the upper part of the column
for detecting and controlling the pulp-froth interface control, and at a lower position in the
collection zone, for gas hold-up evaluation. More details about the set-up and instruments
are presented in section 2.2. Before starting the three-phase test phase, the column
hydrodynamic conditions were assessed using a two-phase system, to determine the most
appropriate range of flotation conditions for the used mineral particle. The determined
hydrodynamic conditions are presented in Appendix B. Various mixtures of almost pure
talc (used as natural hydrophobic mineral) and quartz (as a hydrophilic mineral) was used
in all tests to emulate real flotation conditions. More details about the solid characteristics
are presented in Appendix C. During the flotation tests samples of the feed, concentrate (at
the column upper lauder) and tailing (at the bottom of the column) stream were taken,
processed and analyzed for the evaluation of the flotation kinetic constant. The sampling
points and experimental practice are detailed in Appendix A. The obtained raw data and the
reconciled values (BILMAT), as well as the calculated flotation constants, can be found in
Appendix E. The proper estimation of the kinetic constant requires the knowledge of the
prevailing particle residence time distribution (RTD) in the column. The RTD measurement
method and its results are also explained in the Appendix A. Finally, it must be mentioned
that the experimental set-up included a process control system ensuring the data acquisition
and the automatic control of some operating variables, e.g. froth-pulp interface.
1.12 Objectives
As shown in the above literature review, most previous research work dealing with bubble
size characterization have not focused on the relationship between the hydrodynamic
variables and particle size. Although most researchers and practitioners agree that such a
relationship exists, the lack of an adequate method for on-line measuring the average
bubble diameter has prevented such studies. The on-line version of the McGill bubble
viewer implemented at Université Laval has opened a door for a study of the relationship
between hydrodynamic variables and the particle size. Consequently, the main objectives of
this PhD thesis are the following:
37
to understand the interactions between solid particles and hydrodynamic variables,
and the effect of hydrodynamic variables on the water recovery and carrying
capacity;
to evaluate other hydrodynamic variables potentially providing more information
about the size distribution than the bubble surface area flux (Sb);
to investigate the correlations between flotation rate constant and particle size as
well as hydrodynamic variables.
1.13 Outline of the thesis
This thesis was prepared using the material from four papers that have been either
published, accepted for publication or under journal review. A brief summary of each one is
presented below:
Chapter 2: Effect of particles on the bubble size distribution and gas hold-up in column
flotation
This is a study on the effect of solid particles on the bubble size distribution and gas hold-
up, as well as the correlation between the bubble size distribution and gas hold-up in
column flotation. The experimental results are compared with previously published
correlations and models. Different mechanisms introduced from the literature are discussed
and used to explain the working mechanisms.
Chapter 3: The effect of gas dispersion properties on water recovery in a laboratory
flotation column
This chapter addresses three objectives. The first one is to study the effect of the gas rate,
gas hold-up, bubble size and bubble surface area flux on the water recovery. The second
one is to investigate the effect of hydrophobic particle size on the water recovery. The last
one consists in characterizing the effect of gas dispersion properties on the column carrying
capacity.
Chapter 4: On the relationship between hydrodynamic characteristics and the kinetics of
column flotation. Part I: Modeling the gas dispersion. The interfacial area of bubbles (Ib) is
introduced in this chapter as a hydrodynamic variable providing more information about
38
the size distribution than the bubble surface area flux (Sb). Experimental evidence shows
that the bubble size distribution can exhibit lognormal shape, or even multi-modal and bi-
modal shapes. Later on, the correlation between the hydrodynamic variables – i.e. the
bubble size, bubble surface area flux and hold-up – and the interfacial bubble area is
studied.
Chapter 5: On the relationship between hydrodynamic characteristics and the kinetics of
column flotation. Part II: Model validation
In this chapter, the correlation between the flotation rate constant and particle size and
some hydrodynamic variables is investigated. This correlation exercise was conducted
using a statistical modeling method called Partial Least Squares (PLS), a very powerful tool
to demonstrate the relative importance of the independent variables being considered.
Finally, based on the obtained results, the relationship between the flotation micro-
processes and the relevant hydrodynamic variables is presented.
Chapter 6: Single variable rate constant models
This chapter proposes single variable models for the flotation rate constant based on the
hydrodynamic variables for each particle size-classes specifically suggested in Chapter 5.
Observed trends for each model are then discussed.
Appendix A: Kinetic constant calculation
Firstly, the procedure for sampling input/output streams is presented as well as bubble
sampling technique. Then, the method used for RTD determination and an example are
provided. Finally the kinetic constant calculation is detailed through two examples.
Appendix B: Preliminary tests
Before undertaking three-phase tests (liquid, solid and gas) some tests for the identification
of the hydrodynamic conditions within the experimental column were conducted. Among
them, it can be mentioned the determination of manipulated variable responses and ranges,
temperature and activation time and their effect on the frother efficiency, selection of levels
of shear water rate to the sparger. Finally, the method for determining the residence time
distribution and the obtained results are presented.
39
Appendix C: Solid characterization
This appendix presents the various analysis done on the solid particles used in the tests,
such as XRF, liberation degree and mineralogy.
Appendix D: Radial gas dispersion
The analysis of radial gas hold-up in a section of a column proposed by Banisi et al. (1998)
is presented in this appendix.
Appendix E: Gas dispersion properties
This appendix contains a table showing the effect of the particle size, frother concentration
and shear water rate on the measured hydrodynamic variables such as gas hold-up, Sauter
mean bubble diameter and mean of the bubble size distribution (d10).
Appendix F: Statistical analysis
This appendix is divided in three main parts. In the first one, the existing correlations
among various gas dispersion properties are analyzed. The second one briefly introduces
the Projection to Latent Structures (PLS) method. The last part provides the datasets used
for the validation of the model developed in Chapter 5.
41
Chapter 2 Effect of particles on the bubble size distribution
and gas hold-up in column flotation2*
Abstract
This chapter studies the effect of solid particles on the bubble size distribution and gas
hold-up, as well as the correlation between the bubble size distribution and gas hold-up in
column flotation. Experiments were conducted in two and three-phase systems using a
laboratory flotation column (5.6 cm internal diameter for a total height of 650 cm), and
mixtures of quartz (hydrophilic gangue) and talc (naturally hydrophobic mineral), classified
in four different size fractions. Results are compared with literature correlations and models
to reveal that hydrophobic particles affect the gas hold-up through three different
mechanisms modifying the Sauter mean diameter and rise velocity, namely (1) surface
interactions, and the joint antagonistic effect of (2) bubble loading and (3) coalescence.
Résumé
Ce chapitre étudie l'effet des particules solides sur la distribution de taille de bulle et sur le
taux de rétention de l’air, ainsi que la corrélation entre la distribution de taille de bulle et le
taux de rétention de l’air pour la flottation en colonne. Des expériences ont été menées dans
des systèmes à deux et à trois phases avec une colonne de flottation de laboratoire
(diamètre interne de 5,6 cm pour une hauteur totale de 650 cm), et des mélanges de quartz
(gangue hydrophile) et du talc (minéral naturellement hydrophobe), séparés en quatre
classes granulométriques. Les résultats sont comparés avec des corrélations et des modèles
de la littérature pour révéler que les particules hydrophobes affectent le taux de rétention de
l’air par trois mécanismes différents modifiant le diamètre moyen de Sauter et la vitesse
ascensionnelle des bulles, à savoir (1) les interactions de surface, et l'effet antagoniste du
(2) chargement des bulles et (3) le phénomène de coalescence.
2*Ali Vazirizadeh, Jocelyn Bouchard and Yun Chen "Effect of particles on the bubble size distribution and
gas hold-up in column flotation". International Journal of Mineral Processing, (2014) SUBMITTED
42
2.1 Introduction
In flotation systems, gas dispersion plays a critical role for particle collection (recovery)
and froth mass transport (selectivity). To evaluate this effect at the industrial scale, the
bubble surface area flux (Sb) is typically estimated from the plant superficial gas rate (Jg)
and bubble size distribution (BSD) measurements, where the complete BSD is compressed
into a single value, i.e. the Sauter mean diameter (d32). However, it is now criticized that (1)
a given Sb value can be obtained from different combinations of Jg and d32 (Vinnett et al.,
2012), and (2) a given d32 value can be obtained from different BSDs (Maldonado et al.,
2008c). A better approach consists in adequately parameterizing the overall BSD as
presented by Vazirizadeh et al. (2014).
The gas hold-up is another important parameter used to characterize the hydrodynamic
conditions of bubble column reactors (Luo et al., 1999). It is useful because it combines the
influence of both the bubble size and gas rate. It provides a holistic indication of the
hydrodynamic conditions because it is dependent on various factors such as frother type
and concentration, cell dimensions, operating temperature and pressure, as well as solid
phase properties and concentration.
Banisi et al. (1995) indicated that the presence of solid particles reduced gas hold-up in a
column operated under conditions relevant to flotation. This effect increased with solids
concentration over the range 0 – 15% v/v. Hydrophilic (silica and calcite) and hydrophobic
(coal) particles produced similar reductions in the gas hold-up. Banisi et al. (1995) explored
four possible mechanisms responsible for the gas hold-up reduction in the presence of
solids: (1) coalescence, (2) a shift in the density and viscosity of the pulp, (3) a change in
the radial gas hold-up and flow profiles, and (4) bubble wake effects. They concluded that
changes in gas hold-up due to the addition of solids were due to a combination of the two
latter mechanisms, i.e. a change in radial gas hold-up and flow profiles, and bubble wake
effects.
Gandhi et al. (1999) investigated hydrodynamic behavior of slurry bubble column at high
solids concentrations, and found that the axial distribution of slurry concentration followed
the classical sedimentation–dispersion model. They also showed that the effect of the gas
43
velocity on axial solids distribution were minimal over the range of gas velocities
investigated.
To date, most of the laboratory-based research has been conducted on gas-water (two-
phase) systems. Since one cannot conclude that results in two-phase systems necessarily
apply for slurries, ‘surrogate’ solids such as talc, silica and coal have been used. Although
such synthetic feed material are idealized/simplified compared to natural ores, they provide
important insights about the nature of three-phase systems in flotation (Kuan and Finch,
2010). Despite extensive research on the effect of solids on gas-liquid systems, results are
not conclusive and sometimes contradictory. Little is known about the physical
mechanisms underlying the observed macroscopic effects. The majority of literature on the
topic points to a decrease in gas hold-up with increasing solid concentration (Banisi et al.,
1995; Kuan and Finch, 2010; Reese et al., 1996). In other words, the presence of solids
tends to increase the rise velocity of a bubble swarm.
Most of the literature focuses on the effect of solid particles on a single hydrodynamic
parameter, either the gas hold-up (Banisi et al., 1995; Kuan and Finch, 2010) or bubble size
distribution (Gandhi et al., 1999; Grau and Heiskanen, 2005; O’Connor et al., 1990). In this
chapter, the effect of solid particles is studied not only on the bubble size distribution and
gas hold-up, but also on the relationship between these two hydrodynamic characteristics
jointly. Different mechanisms introduced from the literatures are discussed in light of
experimental results.
2.2 Experimental set-up
Experimental data were generated using a fully automated laboratory column flotation set-
up. The column was made of 5.6 cm internal diameter acrylic tubes for a total height of 650
cm is shown in Figure 2.1.
44
2.1 Schematic of the experimental set-up
A frit-and-sleeve sparger, detailed below, was located at the bottom of the column to
provide controlled air flow rates through local PID controllers. Temperature and absolute
pressure sensors, were mounted at the bottom of the column, to compensate for standard
condition flow rate measurements (referred to a given temperature and pressure). A
conductivity-based sensor (Gomez et al., 2003), installed in the middle of the column (350
cm from the tailings port), allowed measuring the gas holdup using the conductivity of both
the aerated and de-aerated slurry. Maxwell's equation relates the conductivity to the
concentration of a dispersed non-conducting phase (i.e. bubbles) in a continuous liquid
phase (pulp in this case) as:
(2-1)
1
2
3
4
5
14
6
7
8
9
10
11
12
13
1 Flotation column2 Level measurement cells3 Feeding pump4 Conditioning tank5 Aerated conductivity cell6 De-aerated conductivity cell7 Tailing pump8 Shear water pump9 Frit-sleeve sparger10 Image analyzer11 Column control system12 Slurry filter13 Filter pump14 Bubble viewer
(1 )
(1 0.5 )
d p
g
d p
k k
k k
45
where kd is the conductivity of the dispersed phase with air, and kp, the conductivity of air-
free dispersion (de-aerated slurry). Figure 2.2 depicts the schematic of the gas holdup
sensor.
Figure 2.2 Schematic of the εg sensor (Gomez et al., 2003)
In order to measure the bubble size, the McGill bubble viewer was adapted for on-line
operation. The column was operated under automatic control of froth-depth, gas rate, shear-
water and feed flow rate. The froth zone level was measured and controlled using a set of
eleven conductivity cells sensor as described by Maldonado, et al., (2008a). No wash water
was used in the experiments. The air flow rate is measured and controlled using a mass
flow controller, whose readings are converted to volumetric flow rates at reference
conditions of gas flow rate, which then are corrected for actual (test) temperature and
pressure conditions as
(2-3)
where P is the measured absolute pressure (cm H2O) at the bottom of column, T is the
temperature (°C) and is the superficial gas rate (cm/s) at 21°C and 1033.23 cm H2O.
The frit-and-sleeve sparger consisted of a porous stainless steel ring concentrically installed
within a cylindrical sleeve (Kracht et al., 2005) , as shown in Figure 2.3. The porous ring
ref
gJ
46
was 4 cm long and 2.5 cm in diameter and had a nominal pore diameter of 10 µm. The
cylindrical gap (annulus) between the frit and the sleeve was 1 mm. This design allowed an
external flow of water to circulate through the gap providing the necessary shear to reduce
the size of the bubbles generated in the device. This sparger thus provides an additional
degree-of-freedom to control the bubble size with the water velocity, i.e. Jsl, defined such
as:
(2-4)
Figure 2.3 Schematic of a frit-sleeve sparger (Kracht et al., 2008)
where Qsl is the volumetric water flow rate in the sparger, and A is the cross sectional area
of the column. The water flow rate to the sparger is manipulated by the speed of a gear
pump, using a PID controller implemented on a Moore 353.
In order to continuously monitor the bubble size, the McGill bubble viewer (Hernandez-
Aguilar et al., 2004) was adapted to the experimental column at 100 cm above feed port.
The bubbles are collected via a sampling tube and directed into a viewing chamber where
they are exposed under lighting conditions to be photographed. Parameters of the bubble
size distribution – such as the Sauter mean diameter (d32), mean and variance at actual
conditions (i.e. not corrected to reference conditions) – are then determined by image
analysis. With this regard, the Circular Hough Transform (CHT)-based algorithm presented
47
by Riquelme et al. (Riquelme, 2013a, b) was implemented for accurate bubble detection
(including large ones and clusters). Figure 2.4 shows an example of an original picture
taken which contains 800×600 pixels (left hand side), and the bubbles detected by the
CHT-based algorithm, circled in blue and marked by a red cross in the middle (right hand
side). Three frames per seconds have been taken to ensure every picture presented an
entirely ‘new’ sample of bubbles, and avoid multiple counts of the same bubble.
Figure 2.4 Example of an original captured image (left hand side), and detected bubbles
using the CHT-based algorithm, circled in blue and marked by a red cross in the middle
(right hand side)
In the upper section of the column a series of eleven 10-cm spaced stainless-steel
conductivity electrodes, i.e. 5.1 cm external diameter by 0.5 cm height flush-mounted rings
inside the column, allow the measurement of the pulp-froth interface with a precision of 1
cm (Maldonado et al., 2008b).
Quartz was used as a hydrophilic gangue and talc as a naturally hydrophobic mineral,
ground and classified into four size classes. The synthetic pulp fed to the column contained
4% solids (i.e. 40 % talc and 60% quartz), which is a relatively low percent solids, but
exhibiting high proportion of hydrophobic (talc) mineral.
For its constant efficiency over a rather long period of time, a polyglycol frother (F150)
was selected. The gas rate and feed rate were set at 1 cm/s for all tests. Talc, as a natural
hydrophobic mineral consumes a fraction of F150, which is absorbed on the particle
surface (Kuan, 2009). To avoid the results being affected by this interaction, samples
48
started being taken after one hour of column operation time, thus ensuring constant frother
efficiency over time. The operating principle consists in directing a sample of bubbles into
a viewing chamber. Each test allowed sampling populations approximately composed of
10,000 bubbles, thus ensuring representative distribution.
2.3 Results and discussion
2.3.1 Solid particles on the bubble size distribution
Figure 2.2 provides the generated bubble size cumulative distribution functions (CDF), and
Figure 2.3 shows the result of bubble images, respectively, as a function of solid addition in
the presence of a polyglycol frother (F150). The data at 0 % solids (w/w) (i.e. the gas–water
system) shows decreasing bubble sizes (Figure 2.5), and more uniform distributions
(Figure 2.6) with increasing shearwater to the sparger.
The most significant impact of solids is to increase the bubble size. This becomes apparent
when introducing 4 % (w/w) (40% talc and 60% quartz) in the presence of 25 ppm F150
(i.e. above the critical coalescence concentration (CCC)): d32 increased from 0.72 mm to
1.35 mm while introducing 0.2 cm/s shearwater (Jsw), and from 0.64 mm to 1.17 mm with
0.8 cm/s shearwater.
49
Figure 2.5 Effect of the solid percent on the bubble size distribution (CDF)
at different shearwater rate (a) 0 wt%, (b) 4 wt%
0.5 1 1.5 2 2.50
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Qu
an
tity
Bubble size (mm)
Jsw
=0.2 cm/s
Jsw
=0.8 cm/s
(a)
0.5 1 1.5 2 2.50
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Bubble size (mm)
Qu
an
tity
Jsw
=0.2 cm/s
Jsw
=0.8 cm/s
(b)
50
Figure 2.6 Example of bubble images using 25 ppm F150, and for Jg = 1 cm/s
(a) Jsw =0.2 cm/s, 0% wt, (b) Jsw =0.8 cm/s, 0% wt, (c) Jsw =0.2 cm/s, 4% wt (d) Jsw =0.8
cm/s, 4% wt
Kuan and Finch (2010) suggested that bubble coalescence in the presence of talc was the
explanation for this phenomenon. This was supported by bubble size distribution
experimental measurements. For instance, Figure 2.7 shows how the bubble size
distribution evolves from (a) a narrow mono-modal range in a water-only system to (b) a
widespread one following the introduction of talc particles (4 % (w/w) slurry). This result is
typical of coalescence: the coexistence of fine and coarse bubbles would come from a
coalescence-induced break-up phenomenon (Tse et al., 2003). Quinn and Finch (2012) also
reported bi-modal bubble size distributions in the absence of coalescence inhibiting agents,
probably resulting from bubble-bubble interactions. Figure 2.8 depicts an example of
coalescence-induced break-up leading to a bi-modal distribution.
Dippenar (1982) reported that hydrophobic solids can promote coalescence through a
bridging effect of hydrophobic particles between bubbles. In addition, coalescence could
occur because the frother concentration decreases due to frother adsorption by talc (as a
(a) (b)
(c) (d)
1mm
51
hydrophobic solid). With this regard, Kuan and Finch (2010) proposed the following
hypothesis about the adsorption of F150 by talc particles based on the structure of the
frother molecule. Talc particles, via their uncharged surfaces providing natural
hydrophobicity, tend to attach to the bubbles. On the other hand, the F150 molecule has
two end OH groups (hydrophilic) lying flat on the surface of the bubble to accommodate
the two hydrophilic groups (OH and H(C3H6O)). In this orientation, F150 is possibly
exposing its hydrocarbon backbone with talc attached to bubbles, thus inducing uptake by
talc surfaces through hydrophobic interactions.
According to the above-mentioned hypothesis, talc surfaces exposed to bubble surfaces
would play an important role on the adsorption of frother. Frother up-taken by particle is
obviously not a desirable effect in flotation. However, it must be taken into account in cases
where finer grind size is considered to improve the degree of liberation, as it will lead to a
larger total surface of particles in the slurry. Similarly, increasing the solids percent in a
flotation circuit, for instance to increase the production rate, could also lead to higher
frother consumption.
52
Figure 2.7 Effect of the solid percent on the bubbles size histogram (25 ppm F150, Jg= 1
cm/s and Jsw= 0.8 cm/s) (a) 0 % wt, (b) 4 % wt
To check the variation of frother uptake by increasing particle total available surface, a set
of tests have been conducted. Four talc particle size-classes (53/75 µm, 75/106 µm,
106/150 µm and 53/150 µm), each one combined with the same quartz particle size class,
were used. The amount of both the hydrophobic and hydrophilic added solids remained
constant. For each particle class, three frother concentrations (5, 15 and 25 ppm) were
applied in conjunction with three levels of shear water rate (0.2, 0.4 and 0.8 cm/s). The
shearwater rate provided an extra degree-of-freedom for modifying the mean bubble size,
0.5 1 1.5 2 2.50
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Bubble size (mm)
Qu
an
tity
(a)
0.5 1 1.5 2 2.50
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Bubble size (mm)
Qu
an
tity
(b)
53
particularly for tests conducted at a constant frother concentration. Table 2.1 summarizes
the experimental plan for each particle size class.
Figure 2.8 Bi-modal distribution due to coalescence (5 ppm F150, Jg= 1 cm/s, 4 % wt and
Jsw= 0.2 cm/s)
Table 2.1 Summary of the experimental plan
Test # frother concentration (ppm) Shearwater rate (cm/s)
1 5 0.2
2 5 0.4
3 5 0.8
4 15 0.2
5 15 0.4
6 15 0.8
7 25 0.2
8 25 0.4
9 25 0.8
Figure 2.9 presents the effect of the talc particle size on the d32 at a constant gas rate of
1 cm/s. Coalescence leads to an increase of both the d32 and standard deviation of bubble
0.5 1 1.5 2 2.50
0.5
1
1.5
2
2.5
Bubble size (mm)
Den
sity
54
size distribution. As can be seen, the particle size does not show any effect on d32, and
therefore, increasing the available surface of hydrophobic mineral surprisingly did promote
coalescence.
Figure 2.9 The effect of the particle size on the bubble d32 at constant gas rate (1 cm/s)
Tests 4 to 9 in Table 2.1 were conducted using 15-25 ppm frother, which is higher than the
CCC of F150. Results could not reveal any significant effect of the particle size on
coalescence, perhaps because of the concentration of frother in the solution which remained
over the CCC. Even for the first three tests, conducted using 5 ppm frother concentration,
which is below the CCC of F150 (Finch et al., 2008), the particle size effect could still not
be observed. This suggests that other mechanisms could play an important role.
The bubble-particle collision efficiency is a possible explanation. Yoon (1993) introduced a
model linking the probability of collision (Pc) with the bubble (db) and particle (dp) size in
different hydrodynamic conditions as
( / )n
c p bP A d d (2-4)
where A and n are model parameters depending on the flow conditions. According to
equation 2-4, fine particles are less likely to collide with a given set of bubbles than coarse
particles. The effect of increasing the specific particle surface with decreasing particle sizes
1 2 3 4 5 6 7 8 9
1.2
1.4
1.6
1.8
2
Test number
d3
2 (
mm
)
53/75 m
75/106 m
106/150 m
53/150 m
(mm
)
55
would thus be compensated by a lower rate of collection. This assumption however would
remain to be further investigated experimentally using the percentage of solids as a factor,
instead of the particle size distribution to modify the total available hydrophobic mineral
surface.
2.3.2 Effect of solid particles on the gas hold-up
Different hypotheses have been postulated in the literature to explain the influence of
hydrophobic particles on the gas hold-up. The effects would be to:
promote bubble coalescence leading to a reduction of the gas hold-up (Banisi et al.,
1995; Kuan and Finch, 2010);
lead to bubble break-up, which in turn increases the gas hold-up (Yang et al., 2007);
increase the apparent viscosity, and thus the gas hold-up (Banisi et al., 1995; Finch
and Dobby, 1990);
increase bubble wake entrainment effects which decrease the gas hold-up (Banisi et
al., 1995);
change the radial gas hold-up and flow profiles, thus leading to a reduction of the
global gas hold-up (Banisi et al., 1995);
increase the weight of bubbles once attached, and therefore increase the gas hold-up
(Tsutsumi et al., 1991; Uribe-Salas et al., 2003).
Bubble coalescence and break-up
Bubble coalescence occurs due to the bridging effect of particles (Dippenaar, 1982) and/or
adsorption of frother on the hydrophobic particle surface (Kuan and Finch, 2010). Faster
rising larger bubbles thus appear in the population, leading to a reduction of the gas hold-
up. The bubble coalescence induced by solid particles could occur during the generation
process or progressively along the column.
In the present study, a frit-sleeve sparger depicted in Figure 2.3 was used. The mechanism
of frit-sleeve sparger does not permit particles to induce bubble coalescence during the
generation stage since the water injected to the sparger does not contain solid particles.
Therefore, the observed coalescence occurs along the collection zone.
56
Coalescence-induced break-up occurs as one form of coalescence mechanism in which very
fine bubbles are generated along with very large ones (Quinn and Finch, 2012; Tse et al.,
2003). This mechanism decreases the gas hold-up, as larger / faster rising bubbles are
created.
On the other hand, it was suggested that in a three-phase fluidization system, ‘sharp’
particle edges can penetrate and break some bubble surfaces. This would cause bubble to
break-up into smaller ones, thus leading to an increase in the gas hold-up. Yang et al.
(2007) reported this phenomenon for large particles (larger than 1 cm). The exact effect of
solids on bubble break-up is still not clear as Gandhi et al. (1999) had previously explained
the reduction in gas hold-up caused by solids addition leading to a reduction in the bubble
break-up rate. It must be emphasized though that their experiments were conducted at high
slurry concentration (40 % vol.), and using very high gas rate (26 cm/s) that are outside the
typical range of operation.
The coexistence of both large and small bubbles was demonstrated above in Figure 2.8.
Figure 2.10 now illustrates the effect of solid particles on bubble coalescence-induced
break-up. Results show that adding solids (4 % (w/w)) to the gas-water system and keeping
the gas rate and frother concentration constant increased both the d32 and standard deviation
of distribution (σ). As mentioned above, the addition of hydrophobic particles thus
promoted coalescence (larger d32) and resulted in a more widespread distribution, thus
supporting the coalescence-induced break-up phenomenon.
57
Figure 2.10 The effect of solids on the parameters of the bubble size distribution
(Jg = 1 cm/s, 25 ppm F150). σ is standard deviation, d10 is mean bubble diameter and d32 is
Sauter mean diameter
Viscosity and bubble wake
The presence of solid particle increases apparent viscosity, leading to the opposite effect on
the bubble velocity, and thus on the gas hold-up (Finch and Dobby, 1990).
Based on the equation of Roscoe (1952), the slurry viscosity (µsl) is calculated as
2.5(1 )sl l (2-5)
where µl is the liquid viscosity, and α is the volume fraction of solids in the slurry.
Increased slurry viscosity through the addition of solid particles can stabilize the bubble
wake. This, in turn leads to an increase of the bubble velocity, and thus a reduction of the
gas hold-up (Banisi et al., 1995).
In the present work though, µsl / µl =1.038. As can be seen, the value is small enough so the
effect of viscosity on the gas hold-up can be neglected.
Radial gas hold-up and flow profiles
Using a laboratory flotation column (4.47 m height and 10.18 cm in diameter), Banisi et al.
(1995) observed that the flat radial gas hold-up in the water-gas system changes to non-
0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.50.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
d32
Two-phase
Three-phase
(mm)
58
uniform profiles in the slurry-gas system (10% v/v calcite). They reported a significant
difference between the distribution parameters of the gas hold-up in the gas-water and gas-
slurry systems due to non-uniform profile using drift flux analysis. Appendix D details the
derivation of the distribution parameters. This mechanism could explain the lower gas hold-
up in the slurry system.
The radial gas hold-up was not evaluated in the current study. However, more recently
Rabha et al. (2013) obtained contradictory experimental results showing, using a similar
set-up (7 cm diameter and 150 cm height), that the radial gas hold-up remained flat
following the addition of 5% solids, but decreased to a lower value. The material used was
spherical glass particle (100 µm and ρp =2500 kg/m3).
It must be emphasized that Banisi et al. (1995) applied the general drift-flux analysis
method to study the radial hold-up, which is not very accurate compared to the ultrafast
electron beam X-ray tomography used by Rabha et al. (2013) .
Bubble loading
The added weight following hydrophobic particle attachment is known to reduce the bubble
velocity and consequently increase the gas hold-up (Uribe-Salas et al., 2003). In the present
case, talc, as a hydrophobic fraction of added solid, attached on the bubble surface and was
collected at the top of the column. The increase in gas hold-up reported in the present work
supports this mechanism. Table 2.2 provides test results in gas-water and gas-slurry
systems. The addition of the fine particle class did not have a significant effect on the gas
hold-up. However, the d32 almost doubled. The same trend is also observed for the other
particle size fractions. This observation supports the role of bubble loading on the gas
hold-up.
59
Table 2.2 Generated gas hold-up and d32 for gas-water and gas-slurry systems
(Jg = 1 cm/s, Jsw = 0.8 cm/s, 25 ppm F150)
Two-phase Three-phase
εg (%) d32 (mm) Class (µm) εg (%) d32 (mm)
12.95 0.63
53/75 12.48 1.17
75/106 10.85 1.30
106/150 10.46 1.18
53/150 9.34 1.31
2.3.3 Discussion
Among the presented mechanisms, coalescence (and coalescence/induced break-up) and
bubble loading can be used to explain the variations of gas hold-up with the bubble size for
two and three-phase systems as depicted in Figure 2.11. The first mechanism accounts for
decreasing gas hold-up and the second one, for the opposite effect.
By manipulating the shearwater rate, different bubble size distributions can be generated
while the gas rate and frother concentration remain constant (25 ppm F150, Jg =1 cm/s). As
can be observed in Figure 2.11, different combinations of gas hold-up and bubble sizes
were thus produced. Without great surprise, increasing the bubble size reduced the gas
hold-up in the two-phase operation.
Adding solids produced coarser and more spread out bubble size distributions (see
Figure 2.8) due to coalescence. In most of the cases, this led to a reduction of the gas hold-
up that can be explained by the dominant effect of the bubble size on the rise velocity. It is
worth mentioning that shifting the d32 and increasing the standard deviation of distribution
do not change the general shape of bubble size histogram, which can still be represented by
lognormal probability density function in this case.
On the other hand, comparing the obtained results of the gas hold-up in three-phase and
two-phase systems also revealed that in a few cases, notwithstanding coarser bubble size
distributions, the addition of solids led to higher gas hold-up values. The bubble–particle
60
attachment mechanism explains this behavior, i.e. the bubble rise velocity decreases as
bubble loading increases.
Figure 2.11 Bubble size and gas hold-up correlation in two and three-phase system
Results from the test conducted at different frother concentrations in the presence of solids
lead to the same conclusion: coalescence and bubble–particle attachment are the main
mechanisms explaining how the presence of solid affect the ties between the bubble size
and gas hold-up. Reducing the frother concentration below the CCC value revealed the
effect of having larger bubbles in the swarm. At the same time, decreasing the number of
bubbles due to coalescence, results in bubble overloading and decreases the rise velocity.
Figure 2.12 depicts how hydrophobic particles affect the correlation between the gas
hold-up and bubble size for three different frother concentrations. It is worth mentioning
that very similar bubble size distributions – and gas hold-up – were generated at 15 and 25
ppm as in both cases the concentration is the above the CCC.
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.54
6
8
10
12
14
16
d32
(mm)
g (
%)
Two-phase
Three-phase
0 10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
bubble size
Qu
an
tity
0 10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
Bubble size
Nu
mb
er
of
bu
bb
les
0 10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
Bubble size
Nu
mb
er
of
bu
bb
les
0 10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
Bubble size
Nu
mb
er
of
bu
bb
les
d32 (mm)
ε g(%
)
Medium bubble loading & medium bubble coalescence in three-phase
Minimum bubble coalescence in two-phase
Maximum bubble loading & minimum bubble coalescence in three-phase
61
Figure 2.12 Bubble size and gas hold-up correlation in three-phase system
for three frother concentrations
2.4 Conclusion
The experimental results presented in this chapter allowed studying the effect of solid
particles on the bubble size distribution and gas hold-up, but also on the relationship
between these two hydrodynamic characteristics jointly. Different mechanisms introduced
from the literature were analyzed leading to the following conclusions:
1. surface interactions: the effect of solid particles on the bubble size distribution can
be explained by coalescence through (1) the bridging effect of solid particles and/or
(2) the absorption of frother by hydrophobic surfaces;
2. bubble weight and volume: the effect of hydrophobic particles on the gas hold-up is
determined by a combination of two antagonistic mechanisms: (1) bubble loading,
which reduces the rise velocity and (2) bubble coalescence, which increases it;
0 10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
bubble size
Qu
an
tity
0 10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
bubble szie
Qu
an
tity
1 1.2 1.4 1.6 1.8 2 2.2 2.42
4
6
8
10
12
14
d32
(mm)
g (
%)
25 ppm
15 ppm
5 ppm
0 10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
Bubble size
Nu
mb
er
of
bu
bb
les
10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
Bubble size
Nu
mb
er
of
bu
bb
les
0 10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
Bubble size
Nu
mb
er
of
bu
bb
les
0 10 20 30 40 50 60 70
0
0. 05
0. 1
0. 15
0. 2
0. 25
Bubble size
Nu
mb
er
of
bu
bb
les
d32 (mm)
ε g(%
)
Medium bubble loading & medium bubble coalescence
in three-phase
Minimum bubble loading & maximum bubble coalescence in three-phase
Maximum bubble loading & minimum bubble coalescence
in three-phase
Medium bubble loading & maximum bubble coalescence in three-phase
62
3. resulting effect: increasing the bubble size will not lead to the reduction of the gas
hold-up if the effect of the bubble loading outweighs that of the bubble size on the
rise velocity.
Characterizing the role of each one of these two mechanisms in the determination of the
gas hold-up remains to be investigated.
63
Chapter 3 The effect of gas dispersion properties on water
recovery in a laboratory flotation column3*
Abstract
This chapter analyzes the effect of gas dispersion properties on water recovery in a
laboratory flotation column. The experimental work was carried out under a constant gas
flow rate to eliminate the gas volume effect, while other gas dispersion properties have
been varied by modifying the frother concentration and the shearing water introduced to the
frit-and-sleeve sparger. A constant froth depth and no wash water were used, in order to be
able to evaluate the exclusive correlation between the water recovery and the bubble
surface area flux and/or the bubble size. It is worth mentioning, that the particle size of the
hydrophobic particles did not show a significant effect on the amount of water going to the
concentrate. Not surprisingly, experimental results demonstrate that water recovery
increases with the bubble surface area flux and with the gas hold-up.
Résumé
Ce chapitre traite de l’influence de la dispersion des gaz sur la récupération de l'eau dans
une colonne de flottation de laboratoire. Dans cette étude, le travail expérimental a été
effectué sous un débit de gaz constant pour éliminer l'influence de la quantité volumique de
gaz injecté dans la colonne, tandis que d'autres propriétés de dispersion de gaz ont été
modifiées par un changement de la concentration de moussant et de l'eau de cisaillement
introduite dans l'aérateur fritté à chemise. Afin de pouvoir évaluer la corrélation entre la
récupération en eau et la densité de flux d'air et/ou la taille des bulles, la profondeur
d’écume a été gardée constante et aucune eau de lavage n'a été ajoutée dans le système. Il
est à noter que la taille des particules hydrophobes n'a pas montré d'effet significatif sur la
quantité d'eau allant au concentré. Sans surprise, les résultats expérimentaux montrent que
3*Ali Vazirizadeh, J. Bouchard and R. del Villar. "The effect of gas dispersion properties on water recovery
in a laboratory flotation column". IMPC Conference, Santiago, Chile, October 20-24, 2014.
64
la quantité d’eau récupérée dans le haut de la colonne dépend de la densité de flux d'air et
du taux de rétention de l’air.
3.1 Introduction
The overall recovery of particles in a flotation system is the result of the so-called ‘true
flotation process’ and the hydraulic entrainment. The first mechanism would lead to the
collection of hydrophobic particles only, whereas the entrainment mechanism not only
implies the recovery of hydrophilic particles but also increases the amount of hydrophobic
particles transferred to the froth (Cilek, 2009). Furthermore, Tao et al.(2000) and Zheng et
al. (2006) reported a direct relationship between the overall recovery and the entrainment
recovery.
Size by size particle entrainment is a complex phenomenon, which ultimately depends on
process operating conditions, cell design, and froth characteristics. This notwithstanding, it
appears to correlate well with the water recovery to the concentrate (Martínez-Carrillo and
Uribe-Salas, 2008; Yianatos and Contreras, 2010). Entrainment is a two-step process:
(1) transfer of the suspended solids from the pulp region to the pulp–froth interface, and
(2) transfer of the entrained particles in the froth phase to the concentrate. In both cases,
particle entrainment increases with the amount of water reporting to the concentrate, which
in turn increases with the gas rate (Martínez-Carrillo and Uribe-Salas, 2008; Zheng et al.,
2006). In fact, the wake of the bubble swarm is responsible for the water entrainment in the
pulp. Its volume is a function of the bubble velocity and size, as well as the liquid viscosity.
The role of the frother type and concentration can be illustrated through its effect on the
bubble size and velocity, and ultimately on water recovery. For instance, it has been
observed that frother SF-6008 reported more water to the froth than DOW-200 and TEB
(triethoxybutane)(Ekmekçi et al., 2003). They also reported a direct correlation between
water recovery and the amount of frother introduced in the cell.
It appears that the system hydrodynamic conditions play a major role on the entrainment
(Nelson and Lelinski, 2000; Phan et al., 2003; Rodrigues et al., 2001; Yoon, 2000). These
conditions are determined by the prevailing values of gas dispersion properties, which
include gas hold-up, bubble size and bubble surface area flux (Sb). This latter is linked to
particle recovery through the flotation kinetic constant (Finch and Dobby, 1990; Gorain et
65
al., 1995). The correlation between gas dispersion properties and hydrophobic particle
recovery has been considered only via the true flotation mechanism, as a result of the
flotation kinetic. Quantifying the relationship between particle entrainment and gas
dispersion properties thus remains to be addressed. Therefore, the correlation between the
water recovery and gas dispersion properties is a key element to answer the above
statement. With this regard, analyzing the link between the carrying capacity of a column –
which is expressed as the mass of solids to the concentrate per unit of column cross
sectional area – and gas dispersion properties is an interesting avenue. It would lead to the
same information, as would the investigation of the relationship between the recovered
water and gas dispersion properties.
Among the various variables having an effect on the water recovery to the froth zone, the
wash water flow rate plays a key role. Experimental evidence shows that increasing the
wash water flow rate has a substantial effect to reduce the water recovery down to a
minimum, i.e. a saturation point beyond which additional wash water is not beneficial (Tao
et al., 2000). Interestingly, the froth depth only has a marginal effect on water recovery
(Tao et al., 2000; Zheng et al., 2006). It is worth noting that for a constant froth depth, the
water transferred to the concentrate was also used to analyze the froth zone stability in the
presence of different reagents such as collectors, activators and depressants (Wiese et al.,
2011).
Yianatos et al. (2009) observed a selective entrainment of fine particles (size less than 45
μm) in a large cell. On the other hand, the recovery of coarse particles (larger than 150 μm)
by entrainment was considered negligible (less than 0.1%) (Yianatos et al., 2009; Zheng et
al., 2006). Experimental evidence also confirms that if the liberation size diminishes, the
gangue entrainment increases (Guler and Akdemir, 2012). It is also worth noting in this
work, that the hydrophilic particle size also affects the water recovery and the entrainment.
Lastly, it must be emphasized that the water recovery in the industry is estimated based on
steady state mass balances, i.e. for most cases considering the feed, tailings and concentrate
mass flow rates, the mass solid fraction, and the mineral assays (Yianatos et al., 2009).
This chapter investigates (1) the effect of gas dispersion properties on water recovery, and
therefore on particle entrainment, and (2) the role of hydrophobic particle size on the water
66
recovery. The correlation between the carrying capacity and the water recovery is also
studied, to confirm previous results. It is worth mentioning that all conducted tests were
done at constant froth depth and without using wash water, in order to allow isolating the
correlation between water recovery and bubble surface area flux and bubble size.
3.2 Results and discussion
3.2.1 Effect of gas dispersion properties on the water recovery
Table 3.1 summarizes the experimental plan to do such study. The effect of the bubble size
on the water recovery is illustrated in Figure 3.1, showing that the water recovery linearly
decreases with increasing bubble diameter.
Since the total volume of air passing through the column section remained constant,
decreasing the bubble size led to an increased number of bubbles which in turn intensified
entrainment in the wake and boundary layer surrounding bubbles, as described by
Martínez-Carrillo and Uribe-Salas (2008). At constant gas flow rate, the bubble surface
area flux is determined only by the bubble sizes. Thus for the same bubble volume,
increasing the total bubble surface translates into a larger total boundary layer volume –
which contains water and entrained particles – and therefore, to additional water recovered
in the concentrate.
67
Table 3.1 Experimental plan
Hydrophobic particle class Frother concentration (10-6 V) Shear water rate (cm/s)
1 15 0.2
1 15 0.4
1 15 0.8
1 25 0.4
1 25 0.8
2 15 0.2
2 15 0.4
2 15 0.8
2 25 0.4
2 25 0.8
3 25 0.2
3 25 0.4
3 25 0.8
4 15 0.2
4 15 0.4
4 15 0.8
4 25 0.2
4 25 0.4
4 25 0.8
The number of bubbles, their size and velocity determine the gas hold-up. On the other
hand, the water recovery is also related to the number of bubbles and their size, that is to
the gas hold-up. Experimental results, shown in Figure 3.2, illustrate that the gas hold-up
also exhibits a similar relationship with the water recovery. It is worth noting that, since the
gas hold-up can be monitored on-line in industrial cells, it could eventually be used as an
indication of both the water recovery to the concentrate and thus the particle entrainment.
68
Figure 3.1 Effect of the mean bubble size (d32) on the water recovery
for constant gas rate (1 cm/s) and different particle size-classes
Figure 3.2 Effect of the gas hold-up on the water recovery
for constant gas rate (1 cm/s) and different particle size-classes
y = -11.75x + 18.68R² = 0.84
0
1
2
3
4
5
6
1 1,2 1,4 1,6
Wat
er
reco
very
(%
)
d32 (mm)
y = 0.41x - 0.04R² = 0.71
0
1
2
3
4
5
6
0 5 10 15
Wat
er
reco
very
(%
)
Gas holdup (%)
69
3.2.2 Effect of hydrophobic particle size on the water recovery
The experimental plan considered four different particle classes of talc to investigate the
effect of hydrophobic particle size on the water recovery. Figure 3.3 reports the water
recovery as a function of the bubble size for these four different particle size classes. All of
them exhibit the same behavior, hence suggesting that the size class of the hydrophobic
mineral does not influence the water recovery. This also suggests that the particle
entrainment would mainly correspond to the size fraction under 53 µm. Accordingly, using
particle size fractions coarser than 53 µm does not indicate any dependency on entrainment
or water recovery as they are recovered by means of true flotation.
Figure 3.3 Effect of the mean talc particle size on the water recovery
0
1
2
3
4
5
6
1 1,2 1,4 1,6
Wat
er
reco
vey
(%)
d32 (mm)
75/106 Microns
53/75 Microns
106/150 Microns
53/106 Microns
70
3.2.3 Effect of gas dispersion properties on the carrying capacity
Since the carrying capacity (the recovered mass of particles, both hydrophobic and
hydrophilic, per unit cell surface per unit of time) for a column encompasses the overall
recovery of flotation via true flotation and entrainment, its correlation with gas dispersion
properties indirectly determines the role of gas dispersion properties on entrainment. It
must be emphasized that the carrying capacity of the column is different from the bubble
carrying capacity (bubble load), the latter being related to the maximum particle load on the
surface of the bubbles. In this research, the carrying capacity of the column is calculated
based on the recovered mass of particles containing both hydrophobic and hydrophilic
particles. Figure 3.4 shows the effect of gas dispersion properties on the carrying capacity,
for three size classes of talc. Both the gas holdup and the bubble surface area flux have a
positive effect on the recovered mass of particles, as a result of both entrainment and true
flotation whereas the effect is negative for d32.
As mentioned previously, the coarse hydrophobic particles used in this work, were
recovered by true flotation. Thus, the amount of hydrophobic particles transferred to the
concentrate increases with both the bubble surface, and gas holdup. On the other hand, the
total boundary layer volume increases with the total available bubble surface. This results
in additional water, and fine entrained particles (quartz) in the concentrate. In other words,
both the entrainment and true flotation mechanisms follow the same trend with respect to
gas dispersion properties.
71
(a)
(b) (c)
Figure 3.4 Effect of (a) the bubble size, (b) gas holdup, and (c) bubble surface area flux on
the carrying capacity
1
2
3
0,9 1,1 1,3 1,5 1,7 1,9 2,1 2,3
Car
ryin
g C
apac
ity
(gr/
min
cm
2)
d32 (mm)
+53/-75 microns +75/-106 microns +106/-150 microns
1
2
3
0 5 10 15
Car
ry c
apac
ity
(gr/
min
cm
2)
εg (%)
1
2
3
25 45
Car
ryin
g C
apac
ity
(gr/
min
cm
2)
Bubble surface area (Sb)
72
3.3 Conclusion
In this chapter, the relevance of gas dispersion properties and water recovery was studied.
For a constant froth depth and no wash-water addition, increasing the gas hold-up and the
bubble surface area flux linearly increased the water recovery to the concentrate, whereas it
decreases for larger bubbles. These correlations can be explained from the variation of the
total boundary layer volume surrounding the bubbles and their number.
The hydrophobic particle size classes did not show any effect on the water recovery and
consequently on the particle entrainment, since relatively coarse particle size classes were
used. Lastly, the carrying capacity – as defined in this chapter – showed a good correlation
with all gas dispersion properties, the Sauter mean diameter, the gas hold-up and the bubble
surface area flux. This allowed concluding that both the entrainment and the true flotation
mechanisms follow the same trend with respect to gas dispersion properties.
73
Chapter 4 On the relationship between hydrodynamic
characteristics and the kinetics of column flotation. Part I:
modeling the gas dispersion4*
Abstract
Modeling of flotation has been the subject of many investigations aiming at better
understanding the process behavior per-se, and as well for process design, control and
optimization purposes. With this regard, the importance of hydrodynamic characteristics,
either as manipulated or measured variables, is paramount. The interfacial area of bubbles
(Ib) is introduced in this chapter as a hydrodynamic variable providing more information
about the size distribution than the commonly used bubble surface area flux (Sb).
Experimental evidence shows that the bubble size distribution can exhibit normal,
lognormal, and even multi-modal shape. Unlike the Sauter mean diameter (d32) and Sb, the
interfacial area of bubbles is derived from the complete bubble size distribution, and takes
into account these specific characteristics. Fundamental expressions are proposed
characterize Ib using the population mean and standard deviation. Experimental results
indicate that for lognormal bubble size distributions, Ib correlates well with the gas hold-up
and d32. Chapter 5 analyzes the correlation of gas dispersion characteristics with flotation
rate constant.
Résumé
La modélisation de flottation a fait l'objet de nombreuses recherches visant à mieux
comprendre le comportement du processus en soi, et à des fins de conception, de contrôle et
d'optimisation. A cet égard, l'importance des caractéristiques hydrodynamiques, soit en tant
que variables manipulables ou mesurées, est primordiale. La surface interfaciale de bulles
(Ib) est introduite dans ce chapitre comme une variable hydrodynamique fournissant plus
d'informations sur la distribution que le taux surfacique de bulles (Sb) couramment utilisé.
4* Ali Vazirizadeh, Jocelyn Bouchard and René del Villar. "On the relationship between hydrodynamic
characteristics and the kinetics of column flotation. Part I: modeling the gas dispersion". Minerals
Engineering, (2014) IN PRESS
74
Des expériences ont démontré que la forme de la courbe de distribution de la taille de bulle
peut présenter une forme normale, log-normale, et même multimodale. Contrairement au
diamètre moyen de Sauter (d32) et à l’actuelle définition de la Sb, la surface interfaciale de
bulles (Ib) provient de la distribution complète de la taille des bulles, et tient compte de ces
spécificités. Des modèles fondamentaux sont proposés pour permettre la caractérisation Ib
utilisant la population moyenne et l’écart type. Les résultats expérimentaux indiquent que
pour une distribution log-normale de la taille des bulles, Ib est corrélé avec le taux de
rétention de l’air et d32. Le chapitre 5 analyse la corrélation des caractéristiques de
dispersion du gaz avec la constante cinétique de flottation.
4.1 Introduction
Hydrophobic particles present in a mineral pulp are collected in a flotation column by
injecting fine air bubbles through a sparger located at the bottom of the reactor. The
resulting bubble-hydrophobic particle aggregates rise through the collection zone of the
column, allowing their separation from the hydrophilic particles, which settle and flow-out
through the tailings port. The ascending bubble - particle agglomerates finally reach the
froth zone, where typically a counter current wash water stream helps to remove entrained
fine -hydrophilic gangue particles. Bubble coalescence occurs in the froth zone leading to
the loss of some hydrophilic particles but providing improved mineral selectivity. The
overall flotation process can then be considered composed of a reaction process (collection)
followed by a separation process (froth selectivity).
As any other reaction process, the rate of flotation can be characterized by a kinetic
constant, relating the outcome, recovery of valuable species (primary variable) to the
process inputs, bubble and particle concentration and characteristics (secondary variables).
Assuming sufficient air-surface availability, a first order kinetics with respect to the
valuable-particle concentration, is usually accepted for flotation columns. If plug-flow
conditions prevail (as in columns), the relationship between valuable-particle recovery and
flotation time in the collection (reaction) zone is given by
(1 )kR R e
(4-1)
75
where R represents the recovery of floatable particle, R∞ is the ‘ultimate’ recovery
(resulting from the existence of impossible-to-float valuable-mineral particles),τ is the
mean particle residence time, and k is the kinetic constant of such particles. A refinement of
this expression was proposed by Gorain et al. (1997), where the kinetic constant is
decomposed into two elements: (1) the inherent floatability, associated to the
hydrophobicity of the particles, and (2) the bubble surface area flux, representing the
availability of air surface for particle collection.
4.1.1 Hydrodynamic variables and particle size
Hydrodynamic variables have a significant role on the performance of flotation cells since
they affect the reaction rate and the mass transport (water and particles) by increasing the
specific area of the dispersed phase. For instance, it has been experimentally proven that the
rate constant decreases for coarse particles at high superficial gas rates and that fine
particles require much higher superficial gas rates for effective flotation (Heiskanen, 2000).
However, Newell and Grano (2006) reported that increasing the superficial gas velocity
leads to a linearly increased overall flotation rate constant, whereas Yoon (1993) stated that
decreasing the bubble size is more effective than increasing the gas rate to reach higher
recoveries.
Many studies have been conducted to determine the effect of the bubble surface area flux
(Sb) on the collection recovery. For instance, Gorain et al. (1997) obtained a linear
relationship between the flotation rate constant and the bubble surface area flux, the slope
increasing with decreasing particle sizes. This investigation allowed characterizing the
performance of mechanical cells, by using this model. Later on, the same authors reported
that the k – Sb direct relationship was a function of the froth depth. Only for shallow froths,
a linear relationship was found (Gorain et al., 1998) . It can then be concluded that froth
depth plays an important role on the overall kinetics.
The existence of a linear relationship between the flotation rate constant and the froth depth
was also confirmed by Vera and coworkers (1999), The authors also claimed that the
collection zone rate constant of the two evaluated minerals (chalcopyrite and pyrite)
increases with the air flow rate, observing though a reduction in froth-zone recovery. This
76
was associated to a detachment of particles from the bubbles in the froth-zone. This
observation led to the use of a shallow froth in the present work.
4.1.2 Bubble surface area flux models
The bubble surface area flux is defined as the surface of a given number of rising bubbles
per unit time and unit of cell cross-sectional area. Assuming same-size spherical bubbles, a
relation for Sb can be derived as a function of the superficial gas velocity Jg and the bubble
diameter db.
6 g
b
b
JS
d (4-2)
Since bubbles do not have the same size, it is customary to replace db by the Sauter mean
diameter d32
32
6 g
b
JS
d (4-3)
The d32 is a scalar value obtained from a bubble size density function or directly calculated
from observed data. However, important information related to the shape of the bubble size
distribution (BSD), such as multi-modal and tails behavior, is lost in this compression
exercise (Maldonado et al., 2008c). The general Sb definition suggests that the bubble size
distribution play an important role in the metallurgical performance of any flotation
process.
For uniform bubbles, an empirical relationship was presented by Finch & Dobby (1990) to
predict the bubble size from the superficial gas velocity. Finch and Dobby's model was
further developed by Nesset and co-authors (2006), who proposed an empirical model for
estimating the d32 based on do (bubble size at Jg = 0) and two parameters depending on the
bubble generation system, the used chemical reagents, and eventually, the slurry properties.
Based on the collected data from commercial flotation cells, they also reported that the d32
values are sensitive to the relative percentage of large bubbles (larger than 1.5 mm),
whereas d10 values are more sensitive to the amount of small bubbles (smaller than 1 mm).
Instead of using direct values of d32 from histograms describing the size distribution, other
77
authors have used the bubble size density function to evaluate the d32 (Grau and Heiskanen,
2005; Vinnett et al., 2012).
Gorain and coauthors (1999) presented a model to predict the Sb in mechanical cells as a
function of the impeller peripheral speed, the air flow rate per unit cell cross-sectional, the
impeller aspect ratio and the 80% passing size of the feed. This model seems adequate for
cells with forced air feed mechanisms. However, it is known to predict inaccurate values
for self aspirating cells (Gorain et al., 1999).
Heiskanen (2000) criticized Gorain’s model, claiming that “the measurement and
computation of superficial gas velocity, as well as the bubble size in some cases, could be
biased under some conditions”. Moreover, the model validation did not address the real
behavior of various particle sizes, as larger bubble sizes show a higher flotation rate
constant, and poor dispersion conditions give a good flotation response, both results
contradicting earlier research findings and industrial experience. Heiskanen concluded that
the bubble surface area flux needed a broader validation using different types of ore, and
that the linear k – Sb relationship required being further investigated.
4.1.3 Bubble size measurement
Cape Town University (CTU) presented the first attempts to measure bubble size using an
optical system. The experimental set-up uses a belled tube to collect ascending bubbles,
wherefrom they were driven to an inclined viewing chamber. The system combined with a
high intensity lighting and a camera placed at either sides of the viewing chamber to allow
taking pictures (O’Connor et al., 1990; Tucker et al., 1994).
The HUT bubble size sampler developed at Helsinki University of Technology has also
been applied by Grau and Heiskanen (2005) to measure the bubble size distribution in
mechanical cells . They had previously compared the HUT and CTU systems to measure
the bubble size and they found some limitations with the CTU method, “when the gas was
not efficiently dispersed” (Grau and Heiskanen, 2002). The bubble sampling system in the
CTU method also caused bubbles larger than 1 mm to break-up, hence leading to finer
bubble size distribution than the actual ones (Grau and Heiskanen, 2002).
78
The McGill bubble size analyzer is another visual technique which was introduced by
McGill University researchers (Vinnett et al., 2012). Its operational principle is based on
conveying a sample of bubbles (with some pulp) into a viewing chamber where they are
exposed to an appropriate light source and are photographed with a digital camera. An
automated image analysis procedure allows measuring the size of the collected bubbles.
The bubble viewer system consists of a sampling tube attached to the bottom of a sealed
viewing chamber (Hernandez-Aguilar et al., 2004). It is worth mentioning that the viewing
chamber must be periodically purged of the accumulated air, and periodically
cleaned/refilled with fresh water because of progressive slurry buildup, thus making online
measurement very challenging.
Until now, the use of the bubble viewer has been reported basically in offline process audits
(gas dispersion conditions in cells or columns). The only known application for bubble size
control (i.e. continuous db measurement) is the two-phase research work at Université
Laval (Maldonado et al., 2008c).
For process control purposes, Maldonado et al. (2008c) used the commercial software
program Image-J to analyze the bubble images. Based on circular shape detection, the
method however fails at recognizing clustered, large or elliptical bubbles. Vinnett et al.
(2012) proposed a semi-automatic methodology using the USM-IMA in-house software
program to address this issue. The method however is inadequate for continuous
monitoring because of the offline pre-processing step. The introduction of a novel approach
using the Circular Hough Transform (CHT) (Riquelme, 2013a) finally enabled both online
applications (e.g. process control) and the detection of all bubbles, regardless of their shape,
dimension or interaction (clustered, super-imposed, etc.).
4.1.4 Bubble size distribution
Although many bubble size distribution measurements have been reported lately in the
literature (Liu et al., 2013; Xu et al., 2012; Zhu et al., 2014), only a few mention their effect
on flotation performance (Grau and Heiskanen, 2005; Heiskanen, 2000; Vinnett et al.,
2012; Wongsuchoto et al., 2003). Bubble size distributions are now frequently off-line
measured in various flotation cell (mechanical and columns) audits, where they are first
transformed into a scalar, the mean bubble size, thus hiding some important distribution
79
features. Despite their availability, these BSDs have not been used as such in mathematical
models or practical applications.
Experimental data for typical operating conditions in mechanical cells suggests that the
bubble sizes follow a log normal distribution (Vinnett et al., 2012). In some instances, a
modification of the log normal distribution leading to an upper limit distribution is more
suitable to represent bubble size distributions (Grau and Heiskanen, 2005; Heiskanen,
2000). Wongsuchoto et al., (2003) showed that for higher superficial gas velocities, i.e. 20
cm/s to 40 cm/s, which is out of the typical range for flotation conditions, the bubble size
distribution exhibits a normal distribution.
Nesset et al.(2006) studied the relationship between the percentage of bubbles smaller than
1 mm and the d10. They found a unique consistent relationship for different data sets. They
also concluded that increasing frother concentration reduces both the mean and standard
deviation of bubble size distribution. Tucker et al. (1994) reported the same behaviour
when using three different types of frother: one pure and two commercial blends. Vinnett et
al. (2012) also observed a strong relationship between the BSD mean and standard
deviation, and noted that BSDs with a large mean value showed greater dispersion, thus
confirming the existence of a pattern in the shape of distribution. Recent work by Quinn
and Finch (2012) revealed that in absence of frother, bimodal distributions are obtained, i.e.
bubbles finer than 1 mm and bubbles larger than 3 mm.
Considering the abovementioned results, observations and analysis, it seems clear that the
whole bubble size distribution should be considered for modeling the flotation rate
constant. Ideally, the distribution would be narrow, but this is not always the case in
industrial practice (Hernandez-Aguilar et al., 2004; Yianatos et al., 2001; Yianatos et al.,
2012). The interfacial area of bubbles is a solution to characterize normal or lognormal
populations, especially when the bubble size distribution is represented by a multi-model
density function. This chapter proposes fundamental equations to characterize normal and
lognormal bubble size distributions, and also analyses the correlations between the different
hydrodynamic variables.
80
4.2 Equations of gas dispersion
Bubble populations in flotation devices are characterized by the total number of individuals
per unit of volume n, and their size distribution function f(db), between a minimum and a
maximum bubble size. Both the gas hold-up (εg) and interfacial area of bubbles (Ib) can be
calculated from the bubble size distribution.
While Sb indicates the total bubble surface area passing through a horizontal section, the
gas hold-up represents the gas volumetric fraction. The interfacial area of bubbles refers to
the overall surface area of bubble inside the device. It has lately been the subject of several
studies in bubble columns (García-Salas et al., 2008; Gómez-Díaz et al., 2008; Maceiras et
al., 2010). However, it has not been considered so far in flotation modeling.
The bubble interfacial area (m2/m3) and the hold-up of a given set of bubbles are
respectively defined as
max
min
2 ( )db
b
d
b b b bd
I n d f d d (4-4)
and
max
min
3 ( )d6
b
b
d
g b b bd
n d f d d
(4-5)
Depending on the type of distribution used, these equations can be further developed as
presented hereafter.
4.2.1 Normal distribution
If the f(db) is a normal distribution, equations (4-4) and (4-5) become
2
max 2
min
( )
32
2
1d
6 2
bb
b
dd
g b bd
n e d d
(4-6)
and
2
max 2
min
( )
22
2
1d
2
bb
b
dd
b b bd
I n e d d
(4-7)
81
Combining Equations (4-6) and (4-7) allows representing the interfacial area as a function
of the gas hold-up:
2
max 2
min
2
max 2
min
( )
32
2
( )
22
2
1d
2
16 d
2
bb
b
bb
b
dd
b bd
g
ddb
b bd
e d d
Ie d d
(4-8)
or simply
max
min
max
min
2 3
2 2
, , d
6 , , d
b
b
b
b
d
b b bdg
d
bb b b
d
f d d d
I f d d d
(4-9)
Solving for db leads to
max2 2
2 2
2 2
2 2
min
2
2 2 2 2 2
2
2 2
2 ( ) (2 3 ) ( )1
6(2 ) ( ) 2 ( )
bb b
b b
b
dd d
bb b
g
d db b
bd
de d d e erf
I de erf e d
(4-10)
In a normal distribution, 95% of the bubbles are in the domain - 2 and 2. Thus
assuming
min 2bd (4-11)
and
max 2bd (4-12)
Equation (4-10) becomes
32 2
4
32 2
4
1 6(2) (2 3 )
1 2
1 26(2) (2 )
2
g
b
erfe
Ierf
e
(4-13)
or simply
82
2
2 2
1 92.28
6 92.28 46.14
g
bI
(4-14)
Finally, Ib can be expressed as a function of εg and the parameters of bubble size
distribution (μ and σ) as
2
2 2
692.28
92.28 46.14
g
bI
(4-15)
4.2.2 Log-normal distribution
If the BSD is represented by a log-normal distribution, then Equations (4-3) and (4-4)
become:
2
max 2
min
(ln )
22
2
1d
6 2
bb
b
dd
g b bd
n e d d
(4-16)
2
max 2
min
(ln )
2
2
1d
2
bb
b
dd
b b bd
I n e d d
(4-17)
Combining again the expressions for Ib and εg leads to
2
max 2
min
2
max 2
min
(ln )
22
2
(ln )
2
2
1d
2
16 d
2
bb
b
bb
b
dd
b bd
g
ddb
b bd
e d d
Ie d d
(4-18)
Solving for db leads to
max2
2
min
9 23
2
22( )
3 ln( )
21
6 2 ln( )
2
b
b
d
b
g
b b
d
de erf
I de erf
(4-19)
83
Since
2
2
3 ln( )
21
2 ln( )
2
b
b
derf
derf
(4-20)
Equation (4-19) can be approximated as
292
2
6
g
b
e
I
(4-21)
hence
29
22
6g
bI
e
(4-22)
As result of the complexity of manipulating bubble size distribution functions, Bordel et al.
(2006) proposed the following simplified model, based on McGinnis and Little’s approach
(2002), to predict the interfacial area of bubbles (Ibm) from the bubble Sauter mean diameter
and gas hold-up
32
6g
bmId
(4-23)
However, Vinnett et al. (2012) reported that the d32 can be calculated using the second and
third momentum of the lognormal distribution as
22.5
32d e (4-24)
which demonstrates that the model proposed by Bordel and coworkers is not an accurate
simplification of equation (4-22).
Finch et al. (2000) developed a simple approximation linking the bubble surface area flux
and the gas hold-up
5.5b gS (4-25)
84
This relation was validated using various data sets corresponding to different flotation
devices were analyzed and the correlation seems appropriate for Sb lower than 130 s-1 and
εg lower than 25% (typical industrial flotation conditions). To estimate the mean bubble
size, used for calculating Sb, the drift flux analysis method was used. It must be emphasized
that this method exhibits a limited precision, and the correlation is not adequate in some
cases. With this regard, Finch and coworkers (2000) reported that the model did not fit
Gorain’s data base well.
According to Equations (4-23) and (4-25), the bubble surface area flux can be equal to the
specific interfacial area of bubbles under certain condition (see section 4.4.2). Regardless of
their units, Ib and Sb are supposed to be interchangeable variables particularly in their
relation with the flotation rate constant.
4.3 Test procedure
As a result of its constant efficiency over a rather long period of time, a polyglycol frother
(F150) was selected (Appendix B) for the experimental work. The gas rate and feed rate
were set at 1 cm/s for all tests. Talc, as a natural hydrophobic mineral consumes a fraction
of F150, which is absorbed on the particle surface (Kuan, 2009). To avoid the results being
affected by this interaction, samples started being taken after one hour of column operation
time, thus ensuring constant frother efficiency over time. Bubble size distribution was
measured using an on-line version of the McGill bubble size analyzer. The operating
principle consists in directing a sample of bubbles into a viewing chamber, where they are
exposed to an appropriate lighting source, and photographed with a digital camera. Each
test allowed sampling populations approximately composed of 10,000 bubbles, thus
ensuring representative distribution. Figure 4.1 shows a typical bubble size distribution
(histogram).
Bubble detection was performed with the CHT-based algorithm (Riquelme, 2013a, b),
using three frames per second to ensure that every picture presented an entirely ‘new’
sample of bubbles, thus avoiding multiple counts of the same bubble.
85
Figure 4.1 Bubble size distribution histogram
4.4 Results and discussion
4.4.1 Modeling the experimental bubble size distributions
In the current study, bubble size distributions have been generated through manipulating
the shear water rate and frother concentration. Table E.1 in Appendix E provides the
detailed experimental results. The generated bubble size distributions can be categorized in
three main types.
The first type (T1) is obtained using low frother concentration (5 ppm F150) and low
shear-water rate (0.2 cm/s), and exhibits a peculiar shape as shown in Figure 4.2. Both
very fine and large bubbles are observed as a result of the lack of frother. Thus, the bubble
size distribution cannot fit a single or a combination of probability density functions
(PDF), and Ib can only be calculated through the Bordel et al's model (using the d32 and
gas hold-up).
0 10 20 30 40 50 60 700
200
400
600
800
1000
1200
Bubble size (mm)
Qu
an
tity
Pixels
0.820.41 1.25 1.66 2.08 2.5 2.9
86
Figure 4.2 Example of a bubble size histogram for T1 distribution
Increasing the shear water flow rate, and maintaining the frother concentration at the
minimum value changed the bubble size distribution shape and generated the second type
of bubble size distribution (T2), exhibiting a multi-modal bubble size density function.
Mixing two or more PDFs allows fitting multi-shape bubble size density function. In this
study, four normal PDFs have been applied to cover the second type of bubble size
histograms using the maximum likelihood concept. Equation (4-15) thus requires being
adapted as follows
1 1 2 2 3 3 4 4
1 13b g
P P Q QI
(4-26)
where P and Q are the distribution weights, and
2
2 2
92.28
92.28 46.14
i ii
i i
(4-27)
Figure 4.3 illustrates a cumulative multi- shape bubble distribution with fitted model.
0 10 20 30 40 50 60 700
200
400
600
800
1000
1200
1400
1600
Bubble size (mm)
Qu
an
tity
Pixels
2.5 2.90.82 1.25 1.66 2.080.41
87
Figure 4.3 Example of a cumulative bubble size T2 distribution fitted with multi-shape
density function
The third category (T3) of distribution was obtained using operating conditions ensuring
the generation of small bubbles and little coalescence, i.e. with both higher shear- water
flow and frother concentration. In this case, the bubble size histogram can be represented
by a lognormal density function, thus allowing Ib to be calculated using Equation (4-22).
Figure 4.4 shows an example where the cumulative bubble size distribution can be properly
fitted to a lognormal density function.
0 10 20 30 40 50 60 700
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Bubble size (mm)
Cu
mu
lati
ve
Measured
Model
0.41 1.25 1.66 2.08 2.5 2.90.82
Cu
mu
lati
ve f
req
uen
cy
88
Figure 4.4. Example of a cumulative bubble size T3 distribution fitted with a lognormal
density function
Table 4.1 summarizes the experimental conditions used to generate three main types of
BSD. These hydrodynamic conditions have been tested on three particle size classes.
Table 4.1 Experimental conditions for three types of BSD
BSD
Type
Frother
concentration
(ppm)
Shear rate
(cm/s)
Gas rate
(cm/s)
T1 5 0.2 1
T2 5 0.4, 0.8 1
T3 15, 25 0.2, 0.4, 0.8 1
4.4.2 Correlations between hydrodynamic variables
The calculated interfacial area of bubbles using Equation (4-26), denoted by Ibc, for T2
multi-shape bubble size density function does not correlated with Ibm (Equation (4-23)) as
10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Bubble size (mm)
Cu
mu
lati
ve
Measured
Model
0.82 1.66 2.08 2.50.410 1.25
Cu
mu
lati
ve f
req
uen
cy
89
shown in Figure 4.5(left). This could be a result of compacting the whole distribution into a
single value, d32. When d32 does not properly represent the distribution, Ibm does not relate
to Ibc.
The other possibility could be that the PDF model does not fit the bubble size histogram
well. However, in the current study, the PDF completely lies down on the histogram,
therefore the distribution compaction on a single value must be the cause. This is consistent
with the observation for T3 distributions (lognormal), and Ibc, calculated with Equation
(4-22), correlates very well with Ibm (Figure 4.5(right)).
Figure 4.5 Shows the correlation between Ibc and Ibm
(left) for T2 distribution (right) for T3 distribution
According to Figure 4.5 though, Ibm and Ibc are clearly not interchangeable and
dissimilarities (slope ≠ 1) are caused by the inaccurate approximation used for the
denominator of equation (4-23) (see section 4.2.2).
Figure 4.6(a) reveals some correlation between Ibc and Sb, but with an important scattering.
As observed for T2 distributions, Ibc correlates with Sb for T3 distribution, but results also
present some scattering (Figure 4.6(b)). Unlike the correlation between Ibc and Sb for T2
systems, no relationship between Ibm and Sb can be reported here (Figure 4.6(c)). This may
be explained once more by the compaction of the distribution into a single value (d32).
0 20 40 60 800
10
20
30
40
50
60
70
80
Ibm
I bc
y=0.57x
R2=0.98
0 5 10 150
5
10
15
Ibm
I bc
y=0.0006x+0.95
R2=0.00034
90
In addition, based on the good correlation of Ibm and Ibc for T3 distributions, a similar
relationship between Ibc and Sb is expected for Ibm and Sb; this can be seen in Figure 4.6(d).
(a) (b)
(c) (d)
Figure 4.6 Correlation between : (a) Ibc and Sb for T2 distributions, (b) Ibc and Sb for T3
distributions, (c) Ibm and Sb for T2 distributions, (d) Ibm and Sb for T3 distributions
Figure 4.7 shows the correlations between the gas hold-up and both Ibc and Ibm for T2
(multi shape modal) and T3 (lognormal) bubble size distributions.
31.5 32 32.5 33 33.5
0.75
0.8
0.85
0.9
0.95
1
1.05
1.1
Sb
I bc
y=0.11x-2.45
R2=0.69
42 44 46 48 50
10
15
20
25
30
35
Sb
I bc
y=2.94x-89
R2=0.77
31.5 32 32.5 33 33.5
4
6
8
10
12
14
Sb
I bm
y=0.95x-24.3
R2 =0.06
42 44 46 48 50
20
30
40
50
60
Sb
I bm
y=3.76x-136.1
R2= 0.76
91
(a) (b)
(c) (d)
Figure 4.7 Correlation between: (a) Ibc and εg for T2 distributions, (b) Ibc and εg for T3
distributions, (c) Ibm and εg for T2 distributions (d) Ibm and εg for T3 distributions
Obviously, the very good correlation between Ibm and the measured gas hold-up arises
from the definition of Ibm itself, i.e. as a depended function of εg. As seen in Figures 4.7(c)
and 4.7(d), the value of the slopes are different, this being the result of the fact that bubble
sizes in T2 distribution are larger than those in the T3 distribution.
3.5 4 4.5 5 5.5
0.75
0.8
0.85
0.9
0.95
1
1.05
1.1
g %
I bc
y=0.12x+0.4
R2 =0.72
6 8 10 12
10
15
20
25
30
35
g %
I bc
y=3.42x-6.4
R2= 0.97
3.5 4 4.5 5 5.5
10
12
14
16
18
g %
I bm
y=3.25x
R2 = 0.98
6 8 10 12
20
25
30
35
40
45
50
55
60
g %
I bm
y=4.68x
R2= 0.96
92
On the other hand, Ibm fits the measured gas hold-up very well, but the model coefficient is
not exactly identical to that reported by Finch et al. (2000), i.e. 5.5 compared to 3.25 and
4.68 (see Equation (4-25)). It must be emphasized that the bubble surface area flux
calculated using Equation (4-26) is similar to interfacial area of bubbles for d32 values
~1.09 mm. In the present investigation, the average values of d32 are 1.86 mm and
1.28 mm for T2 and T3 distributions, respectively.
Finding the correlation between Ib and the presently used hydrodynamic variables, i.e. Sb,
gas hold-up and d32, is the first step for applying Ib as a hydrodynamic variable in the
flotation process. The next step consists in determining the relationship between Ib and the
flotation kinetics. Since the hydrodynamic variables relate to the flotation rate and Ib, the
flotation rate is expected to have a correlation with Ib. Ib also encompasses both the effect
of gas hold-up and bubble size, and also integrates the whole underlying bubble size
density function. As mentioned earlier, all this information is lost when applying a single
d32 value.
4.5 Conclusion
The interfacial area of bubbles can be considered as a solution to address the issue of
compacting the entire bubble size distribution into a single value (d32) for the purpose of
calculating the flotation rate constant. For normal or lognormal distributions, fundamental
expressions allow characterising Ib using the population mean and standard deviation,
which may lead to a more adequate estimate of the kinetic constant.
In fact, the fundamental equations of gas dispersion have been first presented as a possible
solution for some cases, where multi-shape bubble size density function and lognormal
density function have been well fitted to histograms. Three main classes of bubble size
histograms have been distinguished. The first class (T1) could not be represented by a
single or multi PDF's and has been only determined through Bordel et al.’s model. The
second type (T2) of histogram has been fitted to multi-shape density function and the third
one (T3) follows a lognormal density function. For the two last classes, histogram of the
inter-correlations between gas dispersion properties has been studied. A strong correlation
between gas dispersion properties – namely the gas hold-up and bubble surface area flux –
has been detected for bubble size distributions represented by lognormal density functions.
93
Results were not as convincing for distributions modeled using multi-shape density
function.
Chapter 5 will analyze the correlation of gas dispersion characteristics and the rate constant
of flotation.
95
Chapter 5 On the relationship between hydrodynamic
characteristics and the kinetics of flotation. Part II: model
validation5*
Abstract
The estimation of the flotation rate constant is generally considered difficult because of the
number of variables – not always measurable –required to determine its value. Among
them, the particle size distribution and hydrodynamic characteristics of the device are
considered key elements.
Chapter 4 introduced the interfacial area of bubbles (Ib) as a hydrodynamic variable
providing more information about the size distribution than the bubble surface area flux
(Sb). Fundamental expressions were proposed to characterize Ib using the population mean
and standard deviation. Experimental results indicated that for lognormal bubble size
distributions, Ib correlates very well with the gas hold-up and d32.
Chapter 5 investigates the correlation between the flotation rate constant and the particle
size as well as some hydrodynamic variables using a Projection to Latent Structures (PLS)
analysis. The tests were conducted under 'ideal' conditions (i.e. shallow froth, low mineral
concentration and pure mineral particles). Results suggest that for the fine particle sizes, the
bubble surface area flux (Sb) should be considered for the kinetic constant modeling. For
coarser particle, the gas hold-up (εg) is the determining parameter. In practice though, the
particle size distribution often lies between these two extreme cases, and can either span a
very large range or contains intermediate size particles. In such cases, the interfacial area
of bubbles (Ib) better correlates with the flotation kinetic constant.
5 *Ali Vazirizadeh, Jocelyn Bouchard, René del Villar, Massoud Ghasemzadeh Barvarz and Carl Duchesne.
"On the relationship between hydrodynamic characteristics and the kinetics of flotation. Part II: model
validation". Minerals Engineering, (2014) IN PRESS.
96
Résumé
L'estimation de la constante cinétique de flottation est généralement considérée comme
difficile à cause du nombre de variables qui ne sont pas toujours mesurables. Parmi ceux-ci,
la distribution granulométrique des particules et les caractéristiques hydrodynamiques son
considérées comme des éléments clés.
Le chapitre 4 a introduit la surface interfaciale de bulles (Ib) comme une variable
hydrodynamique fournissant plus d'informations sur la distribution de la taille que la
densité de flux d'air (Sb). Des modèles fondamentaux ont été proposés pour caractériser Ib
en utilisant la moyenne de la distribution de dimension et sont écart-type. Les résultats
expérimentaux indiquent que pour une distribution logarithmique de la taille de bulle, Ib est
corrélé avec le taux de rétention de l’air et le d32.
Le chapitre 5 étudie la corrélation entre la constante cinétique de flottation et la taille des
particules ainsi que des variables hydrodynamiques données en utilisant une projection de
structures latentes (PSL). Les tests ont été effectués dans des conditions idéales (écume
superficielle, une fraction solide faible et des particules minérales pures). Les résultats
suggèrent que pour les particules fines, la densité de flux d'air (Sb) doit être envisagée dans
la modélisation des constantes cinétiques. Pour les grosses particules, le taux de rétention
de l’air (εg) est le paramètre déterminant. Dans la pratique cependant, la distribution
granulométrique des particules se situe souvent entre ces deux cas extrêmes, elle peut soit
couvrir une très large gamme ou contenir des particules de taille intermédiaire. Dans ce cas,
la surface interfaciale de bulles (Ib) a une meilleure corrélation avec la cinétique de
flottation.
5.1 Introduction
In chapter 4 the interfacial area of bubbles (Ib) was introduced as an alternative variable for
flotation kinetics modeling. The interfacial area of bubbles refers to the overall surface area
of bubble inside of the flotation cell. While Sb indicates the total bubble surface area
passing through a section, the gas hold-up (εg) represents the gas volumetric fraction inside
of the device. The bubble interfacial area and hold-up of a given set of bubbles are
respectively defined as
97
(5-1)
and
max
min
3 ( )d6
b
b
d
g b b bd
n d f d d
(5-2)
where n is the total number of bubbles per unit of volume, f(db), their size distribution
function, and db min / db max are the minimum and a maximum bubble size respectively.
The correlation between Ib and other hydrodynamic variables was studied aiming at finding
a relationship with the flotation rate constant. Results demonstrated that Ib is related to both
gas hold-up (εg) and bubble surface area flux (Sb).
Ib could be considered as a compromise between εg and Sb for flotation modeling, since,
according to its definition, it encompasses both the bubble size and gas hold-up effects.
Investigating the ties between the kinetics of flotation and the hydrodynamic variables
requires an adequate model and an appropriate set of tests for calibration purposes.
Moreover, the effect of the particle size distribution must also be factored in, which has
been the case in this investigation.
For the purpose of model-parameter calibration, collection-zone timed-recovery values
must be available. Samples of feed, concentrate and tail flows, must then be obtained over
a given time span, for subsequent processing and analyses for the considered species. In the
case of a flotation column, the device used in this study, normal launder samples were
taken, as sampling of the collection zone was not possible in the small device used for this
study. Therefore, using such samples for the estimation of the overall flotation recovery, to
infer the true collection-zone recovery, can strictly be valid only when the froth zone is
very shallow.
The collection-zone recovery has already been modeled by various researchers. Among
them, Yianatos et al. (2005) used a rectangular distribution function of the rate constant (k)
and a tank in-series model for residence time distribution. Others have demonstrated that
the shape of the distribution function does not have a substantial effect on the average rate
constant (Polat and Chander, 2000). In the present work, a single value of the kinetic
constant was used to represents the rate constant distribution. In fact, the effect of the
max
min
2 ( )db
b
d
b b b bd
I n d f d d
98
particle size distribution and hydrodynamic variables – i.e. the gas rate, bubble surface area
flux, gas hold-up and bubble size – on the ultimate rate constant is the dominant one, and
largely outweighs that of using a rate constant distribution model.
With this regard, it has been reported that the flotation kinetics of single-size particles
varied significantly with the ultimate recovery (R∞) and the mean of the kinetic-constant
distribution of the process (Polat and Chander, 2000). The effect of the particle and bubble
size on micro processes, such as collision and attachment, has also been reported in several
studies(Dai et al., 2000; Nguyen, 1998; Sarrot et al., 2005; Weber and Paddock, 1983). The
presented micro-process models were applied to develop a particle-size dependent
flotation-rate model. For instance, Pyke et al. (2003) reported a characteristic bell-shape
flotation rate constant curve with respect to particle sizes (Ahmed and Jameson, 1985),
exhibiting a maximum at intermediate particle size between 8 µm and 80 µm.
Since an empirical kinetic model would subsequently be derived, an appropriate approach
is required to relate hydrodynamic variables to the kinetic constant value. However, it was
shown in Chapter 4 that εg, Sb and Ib are inter-correlated. Classical linear regression
methods are then not applicable because they assume the predictor variables are
uncorrelated. The multicolinearity in the predictor variables is also demonstrated in
Appendix F.1. The analysis shows how the inter correlation adversely affects the regression
technique performance.
For this reason, a multivariable regression using a Projection to Latent Structures (PLS) is
proposed. The technique is briefly introduced in Appendix F.2 PLS regression takes
advantage of the multicolinearity in the predictor variables, and allows in the present case
comparison of the relevance of εg, Sb and Ib jointly to predict the flotation rate constant.
This comparison accredits a new hypothesis about the effect of particle size on
hydrodynamic variable selection for kinetic modeling, which was corroborated using four
different datasets extracted from the literature. It is worth mentioning that the PLS models
were developed using custom scripts written in MATLAB R2010a (MathWorks)
environment and using the PLS Toolbox (Eigenvector Research).
99
5.2 Test procedure
The experimental set-up used for the experimental work was already described in
Chapter 2. Here, only the experimental design and the test procedure are presented.
Four mixtures of talc (hydrophobic component or valuable mineral) and quartz (hydrophilic
component or gangue), classified with wet screening (two cycles for talc), were used. In all
mixtures, quartz was present with the same complete particle size distribution, which was
combined to a ‘single class’ of talc. Three different single classes were used: +53 / -75 μm,
+75 / -106 μm, and +106 / -150 μm. Nine tests were also conducted using a complete
‘mixed’ size distribution. The relative amount of quartz and talc varied between mixtures
(between 40% and 60%). The talc was very hydrophobic, producing a voluminous froth.
Hence a solid concentration of 4% was used to prevent the froth from overflowing the
launder.
A full factorial design was used with three independent variables: the talc particle size class
(4 levels: fine, medium, coarse, and mixed), frother concentration (3 levels: 5, 15, and 25
ppm) and shear-water superficial velocity to the sparger (3 levels: 0.2, 0.4, and 0.8 cm/s).
Other experimental conditions were set as follows:
gas superficial velocity: Jg = 1 cm/s
type of frother: F150
froth depth: H = 10 cm
pulp conductivity: 1000 – 20,000 µS.
To be in this range small amount of sodium chloride is added to the pulp and no other
reagent is used during the test.
The froth layer was kept very shallow (at 10 cm) in all tests, in order to estimate, with
minimum error, the collection zone recovery from global column recovery values. No
wash-water was used.
After one hour of column operation, the system was deemed to be working under steady-
state conditions, and data collection (bubble pictures and gas holdup measurements) was
launched where samples of tails, feed and concentrate were taken. The pulp residence time
100
was calculated by means of direct measurement of column tailings flow rate, since the
column feed rate and froth height were kept constant throughout the test. One-liter tailing
and concentrate samples were taken simultaneously. This sample volume did not have a
significant effect on the feed composition, compared to the 350 liters in feed tank. .An XRF
analyzer was used to determine the talc content of the samples. Chemical analysis were
then reconciled using Bilmat™. The reconciled data allowed evaluating the recoveries,
which in turn allowed estimating the flotation kinetic constant from
𝑅 = 𝑅∞(1 − 𝑒−𝑘𝜏) (5-3)
assuming a plug flow behavior – acceptable for a small diameter laboratory column – and
where R is the recovery of floatable particle, R∞ represents the ‘ultimate’ recovery, τ is the
mean particle residence time, and k is the talc kinetic constant.
5.3 Results and discussion
Table F.1 of Appendix F present the experimental results used to develop the PLS model.
5.3.1 Particle size and kinetic constants
As presented in Appendix F.2, the Variable Importance on the Projection (VIP) allows
quantifying the weight of the independent variables in the PLS model. The VIP score
thresholds are as follows:
VIP > 1: for the most influential independent variables in the PLS model,
0.8 < VIP < 1: for moderately influential independent variables in the PLS model,
VIP < 0.8: for the least significant independent variables in the PLS model.
As can be seen in Figure 5.1, the particle size has a significantly higher VIP score than any
of the other variables, thus showing that the particle size effect is predominant for the
flotation system investigated.
101
Figure 5.1. Relative importance of particle size (P.S.), εg, Sb and Ib
5.3.2 Hydrodynamic variables and rate constant
Notwithstanding the abovementioned results, the particle size obviously cannot be
manipulated for controlling a flotation cell performance. Therefore, other variables must be
considered. Among other possible manipulated variables, the hydrodynamic characteristics
seem to be good candidates, since they present a strong correlation with the flotation rate.
The rate constant is correlated to the collection efficiency, which in turn depends on two
main phenomena: collision and attachment efficiencies. It has also been reported that the
particle size weights the value of these two antagonistic terms on the collection process
(Diaz-Penafiel and Dobby, 1994; Heiskanen, 2000). Thus, small particle size shows low
collision efficiency, but in return exhibit a high probability of attachment (Dobby and
Finch, 1987). On the other hand, the coarse particles are likely to collide with bubbles, but
detach readily.
5.3.2.1 Fine particles
Several theoretical studies have proven the direct relationship between collision efficiency
and particle size – bubble size ratio (Nguyen, 1998; Sarrot et al., 2005; Weber and
Paddock, 1983; Yoon, 1993). For four different flow regimes, Yoon (1993) determined that
the collision efficiency can be calculated from the particle size – bubble size ratio as
1 2 3 40
0.5
1
1.5
2
2.5
3
Variables
VIP
Sco
res f
or
k
Sb
g
IbP.S.
102
(5-4)
where Pc is the collision probability, dp the particle size, db the bubble size. A and n are
model parameters related to the Reynolds number.
For a constant fine-size particle distribution, Equation (5-4) indicates that the main factor
on the collision efficiency is the bubble size. On the other hand, bubble surface area flux is
directly determined by bubble size when the gas rate is constant. Consequently, under the
assumed conditions, collision efficiency would be a function of the bubble surface area flux
which means that for collecting fine-size particles, the bubble surface area flux should be
considered as the major influencing factor. Therefore, the kinetic constant of fine particles
would exhibit a close correlation with the bubble surface area flux.
5.3.2.2 Coarse particles
As mentioned above, the collection efficiency of coarse particle is mostly determined by
the attachment process. The basic assumption in analyzing the likelihood of attachment is
that it occurs when the film between the bubble and particle breaks and the three-phase
contact is created. The total time required, following the particle-bubble collision, is called
the induction time. A given particle will then attach to the bubble if the particle sliding time
on the bubble surface is longer than the induction time. As presented by Finch and Dobby
(1990), the attachment efficiency depends on particle sliding velocity. Since in actual
flotation, instead of a single-bubble / single-particle phenomenon, a bubble swarm and a
population of particles are implicated, the slip velocity (Usg) can be assumed to be equal to
the sliding velocity on average. The slip velocity is a function of the gas and pulp
superficial velocities, and gas hold-up, i.e.
(5-5)
where Jl is the pulp superficial velocity, and +/- stands for counter-current flow and co-
current flow respectively. Consequently, the attachment efficiency should be related to gas
hold-up when gas and pulp flow rate are kept constant.
n
p
c
b
dP A
d
(1 )
g lsg
g g
J JU
103
5.3.3 Experimental validation
To validate the above phenomenological analysis, a set of tests was completed for each
given single size-class of talc. It was expected that for fine size particles, Sb with k would
exhibits a better correlation than εg and Ib. However, for a coarse size class, the correlation
between the kinetic constant and the εg should be better than with the other hydrodynamic
variables.
The relative importance of the various hydrodynamic variables for their use in the modeling
of the flotation rate, for various single-size particles was studied. For each single-size class,
the VIP scores for εg, Sb and Ib were calculated. Figure 5.2 shows how the VIP scores of εg,
Ib and Sb for (a) fine, (b) mid-, and (c) coarse size classes. Results indicate that Sb has a
stronger effect than εg and Ib, on the kinetic constant, for the flotation of fine particles. As
the particle size increases, the importance of Sb decreases, whereas the importance of εg and
Ib increases, thus corroborating the phenomenological analysis, i.e. the collision efficiency
mainly determine the rate constants for fine particles.
When the particle size increases, the weight of εg in the relationship becomes more
important, but the experimental results do not reveal a clearly dominant effect, because the
‘coarse’ size fraction had a top size of 150 μm, which is far from being extreme for a
flotation feed. The validation of the phenomenological analysis was completed using data
extracted from the literature. With this regard, three datasets were examined.
104
(a) (b)
(c)
Figure 5.2. Importance of variables in VIP projection for three sizes of particles
(a) 53/75 µm, (b) 75/106 µm, (c) 106/150 µm
5.3.3.1 Gorain et al. dataset (1997)
A 2.8 m3 flotation cell designed at the JKMRC was used in this study to evaluate the zinc
cleaner circuit of the Hellyer Concentrator in Tasmania. This mineral exhibited a d80
between 20 and 25 microns. Gas dispersion properties and the flotation kinetic constant
Eg Sb lb0
0.5
1
1.5
2
2.5
Variables
VIP
Sco
res f
or
k
Ib
Sb
g
Eg Sb lb0
0.5
1
1.5
2
2.5
Variables
VIP
Sco
res f
or
k
Sb
Ib
g
1 2 30
0.5
1
1.5
2
2.5
Variables
VIP
Sco
res f
or
k
Ib
g S
b
105
were evaluated. The test results for a Chile-X impeller are shown in Table F.2 of
Appendix F.
The authors concluded that Sb presents an appropriate correlation with the flotation rate
constant, but the correlation with εg was also observed. According to their test conditions,
the particle size is considered as a fine size-class particle, and therefore their results would
confirm that Sb should be the best hydrodynamic variable for flotation rate modeling. This
is in agreement with the PLS analysis presented in Figure 5.3 in which the bubble surface
area flux exhibits the highest VIP score.
Figure 5.3 PLS regression for hydrodynamic variables using the dataset from Gorain et
al.(1997)
5.3.3.2 Massinaei et al. dataset (2009)
For their study, Massinaei et al. used the rougher feed of the flotation column at Miduk
concentrator inIran. The ore had a d80 of 100 microns and contained 54% of chalcopyrite
particles smaller than 30 microns. In other words, most of the hydrophobic particles were in
the fine-size range. Their work presented a set of the tests conducted in an industrial
column (Metso minerals CISA Microcell) of 12 m in height and 4 m in diameter, 150
m3 volume) and a pilot column (Plexiglas, 10 cm diameter, 400 cm height). The data-set is
presented in Table F.3 of Appendix F. The correlation between hydrodynamic variables
1 2 30.75
0.8
0.85
0.9
0.95
1
1.05
1.1
1.15
1.2
1.25
Variables
VIP
Sco
res f
or
k
Ib
g S
b0.1 0.2 0.3 0.4 0.5 0.6
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
kmeasured
kp
red
icte
d
R2 = 0.963
3 Latent Variables
RMSEC = 0.027036
Calibration Bias = -5.5511e-017
Data
1:1
Fit
Estimated Error
106
and the kinetic constant is depicted in Figure 5.4, where the PLS regression results and VIP
scores are presented. Again, Sb exhibits the highest VIP score, thus supporting the
phenomenological analysis for fine particles.
Figure 5.4. PLS regression for hydrodynamic variables using the dataset of Massinaei et al.
(2009).
5.3.3.3 Kracht et al. dataset (2005)
In this case, samples of copper sulphide ore, taken from the rougher flotation feed of the
Andina Division, property of Codelco Chile, were used. The samples assayed 1.29 % Cu,
mainly chalcopyrite (95%) and chalcocite (4%). The particles exhibited a d80 of 182 m,
thus laying in the coarser end of the size range for flotation. The tests were conducted in
two types of batch flotation cell: an Outotec batch cell (3.7 L), and Labtech-Essa cell
(4.9 L). The experimental data are presented in Table F4 of Appendix F. Figures 5.5 and
5.6 provide the results of the PLS regression. The VIP scores for εg are this time
predominant in the regression, thus supporting the phenomenological analysis for coarse
particles.
1 2 30.75
0.8
0.85
0.9
0.95
1
1.05
1.1
1.15
1.2
Variables
VIP
Sco
res f
or
k
Ib
g S
b
0.1 0.15 0.2 0.250.1
0.15
0.2
0.25
kmeasured
kp
red
icte
d
R2 = 0.652
3 Latent Variables
RMSEC = 0.024286
Calibration Bias = -8.3267e-017
Data
1:1
Fit
Estimated Error
107
Figure 5.5 PLS regression for hydrodynamic variables using the dataset from in the Outotec
cell
Figure 5.6 PLS regression for hydrodynamic variables
using the dataset from the Labtech-Essa cell
5.4 Interfacial area of bubbles and rate constant
The interfacial area of bubbles (Ib) contains the main effects of εg and Sb. According to the
previously presented data-analysis, Ib does not offer any significant interest for its use in
modeling the kinetic constant when the particle size distribution is either fine or coarse, in
both cases within a narrow range. To unveil the influence of Ib on the kinetic constant, a
sample with a broader size distribution must be used.
1 2 30.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
Variables
VIP
Sco
res f
or
Variables
k
Ib
Sb
g
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.30.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
kmeasured
kp
red
icte
d
R2 = 0.662
2 Latent Variables
RMSEC = 0.084234
Calibration Bias = -1.1102e-016
Data
1:1
Fit
Estimated Error
1 2 30.5
0.6
0.7
0.8
0.9
1
1.1
1.2
Variables
VIP
Sco
res f
or
k
gI
b Sb
R2 = 0.718
2 Latent Variables
RMSEC = 0.084308
Calibration Bias = 0
1 1.1 1.2 1.3 1.4 1.5 1.6 1.71
1.1
1.2
1.3
1.4
1.5
1.6
k
pre
dic
ted
kmeasured
Data
1:1
Fit
Estimated Error
R2=0.718
2 Latent Variables
RMSEC =0.084308
Calibration Bias = 0
108
Therefore, a mixed-size particle sample, composed of one third of each prepared size talc-
particles was fed to the column. It was expected that the interfacial area of bubbles would
exhibit a larger VIP score than Sb and εg. In the mixed size-class flotation test, the
collection rate of fine particles should be sensitive to collision efficiency, while for coarse
particles, the attachment efficiency should determine their collection rate. Therefore,
neither Sb nor εg should be appropriate for flotation rate constant modeling. Figure 5.7
shows the VIP scores of Sb, Ib and εg for these mixed size-class particles.
The weight of Ib in the relationship becomes more important than previously observed for
single class feeds. The validation of the abovementioned hypothesis was completed using
another dataset extracted from the literature, exhibiting a wide size distribution.
Figure 5.7 PLS regression for hydrodynamic variables using mixed-size class particles
5.4.1.1 Jincai et al. dataset (2013)
Jincai et al. (2013) used a pilot scale flotation column (15 cm diameter and 350 cm height
Plexiglas cylindrical column) for the treatment of oily waste water. In this case oil droplets
show the same behavior as hydrophobic particles and flotation can be applied for separating
water from oil. On the studied sample, oil droplets are distributed in various size fractions
with a d85 below 100 µm.
Eg Sb lb0.8
0.85
0.9
0.95
1
1.05
1.1
VIP
Sco
res f
or
k
Variables
Sb I
b
g
109
Table F.5 of Appendix F presents the dataset. The statistical analysis was done by means of
standard regression and PLS methods. Figure 5.8 shows the PLS regression results showing
again that Ib is the most influential variable in the regression for mixed-size particle
distributions.
Figure 5.8 PLS regression for hydrodynamic variables using the dataset from Jincai et al.
(2013)
A final PLS regression was performed using all the combined datasets, regardless of the
particle size, hydrophobicity, and machine type. Figure 5.9 shows the results of the
regression clearly showing that Ib is the most influential variable in the rate constant model.
1 2 30.9
0.92
0.94
0.96
0.98
1
1.02
1.04
1.06
1.08
1.1
Variables
VIP
Sco
res f
or
k
Sb
g
Ib
0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.20.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
kmeasured
kp
red
icte
d
R2 = 0.958
2 Latent Variables
RMSEC = 0.094246
Calibration Bias = 2.2204e-016
Data
1:1
Fit
Estimated Error
110
Figure 5.9 PLS regression for hydrodynamic variables using the datasets combined
Results thus show that in a typical flotation operation, where floatable particles are found in
all particle size classes, optimizing the value of Ib should be the main focus for optimizing
the recovery.
This notwithstanding, Sb and εg are still variables of interest in very specific applications.
For fine particle size classes (e.g. following a regrind or ultrafine grinding), measuring and
optimizing the mean bubble size and gas rate could be an appropriate tool to obtain the best
recovery. On the other hand, coarse particle flotation (e.g. carbonates, phosphates or
graphite), εg is the main variable influencing the kinetic constant. Gas hold-up is a function
of the gas rate, frother type, frother concentration and pulp chemistry.
It is generally accepted that the bubble size presents an inverse correlation with gas hold-up
but this conclusion is not always true. It was observed that two bubbles having the same
size but generated from two sort of frother may cause dissimilar gas hold-up (Tan et al.,
2013). In other words, it is possible that the same bubble size and gas rate show different
gas hold-up. Optimizing the coarse particle recovery thus should be performed through
controlling the gas hold-up at the proper target, and not necessarily with a reduction of the
bubble size.
1 2 30.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Variables
VIP
Sco
res f
or
k
g
Ib
Sb
0 0.5 1 1.5 2 2.50
0.5
1
1.5
2
2.5
kmeasured
kp
red
icte
d
R2 = 0.641
2 Latent Variables
RMSEC = 0.27642
Calibration Bias = -1.1102e-016
Data
1:1
Fit
Estimated Error
111
5.5 Conclusion
The hydrodynamic variables present a strong inter-correlation. Because of this correlation,
it is difficult; to use multivariable regression techniques to determine which hydrodynamic
variable is the most influential for the determination of the kinetic constant. This problem
can be solved with PLS regression, a method using the covariance structure between
independent variables.
This chapter proposed a phenomenological analysis to determine which hydrodynamic
variable is the most influential for recovering fine and coarse particle, based on results from
tests conducted under ideal conditions and from four datasets extracted from the literature.
Conclusions were that the flotation of fine particles mostly relies on Sb, while coarse
particle flotation is mainly determined by the value of εg. PLS regressions on original
experimental data (talc flotation) and various other data-sets (on oil drops, zinc and copper
sulphide ores) extracted from the literature, confirmed these conclusions.
This chapter also introduces the concept of interfacial area bubbles as a gas dispersion
variable that encompasses the properties of both εg and Sb, and shows good correlation with
the flotation kinetic constant, for general cases where floatable particles f are of any size
class. In such cases, optimizing the value of Ib should be the main focus for optimizing the
recovery.
113
Chapter 6 Single variable rate constant models
6.1 Introduction
The interfacial area of bubbles (Ib) was introduced in chapter 4 as a hydrodynamic variable
suitable for flotation rate modeling; its correlation with the other hydrodynamic variables
such as the gas hold-up and bubble surface area flux was then analyzed.
In Chapter 5, it was demonstrated that the use of hydrodynamic variables for flotation rate
modeling depends on the particle size range. Bubble size and bubble surface area flux
showed better correlation with the flotation rate of fine particle size-classes. On the other
hand, gas hold-up was more adequate to characterize the flotation rate of coarse particles.
This notwithstanding, particle size distributions in flotation systems typically span large
size ranges, and thus do not fall into either of the above mentioned extreme cases.
Experimental evidence showed that the interfacial area of bubbles, which encompasses both
the effect of the gas hold-up and bubble size, is more flexible and could be advantageously
used for flotation rate modeling in general applications.
In this chapter, the most suitable hydrodynamic variable for each particle size-classes is
used to produce single-variable models for the flotation rate constant. Observed trends for
each model are then discussed.
6.2 Flotation kinetics of fine particle size-class
The bubble surface area flux was recommended for modeling the kinetics of fine particle
size-class. Since the gas rate is constant in all the conducted experiments, the bubble size
exhibits the exact inverse effect of Sb on the flotation rate. Figure 6.1 and 6.2 show the
effect of the bubble size, and Sb respectively on the kinetic constant value of two particle
size-classes (53/75 µm and 75/106 µm). Both particle size-classes behave similarly: the
flotation kinetic constant increases when bubble size increases to a peak value and then
decreases with increasing bubble size.
114
Figure 6.1 Talc collection rate constant as a function of the bubble size
for two particle size-classes (Jg= 1cm/s)
Figure 6.2 Talc collection rate constant as a function of Sb for two particle size-classes
(Jg= 1 cm/s)
Interestingly, results also show that the use of smaller bubbles will not always improve the
metallurgical performance. A possible explanation for this result could be that coalescence
1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1
0.1
0.15
0.2
0.25
0.3
d32
(1/m
in)
k
actual
53/75 m
actual
75/106 m
(mm)
30 35 40 45 50
0.05
0.1
0.15
0.2
0.25
Sb
(1/m
in)
actual
actual
53/75 m (model)
75/106 m (model)
(1/S)
k
k
115
occurs in the froth zone and/or might be related to froth stability and particle drop-back. It
has been reported that both hydrophobic particle and bubble size affect froth stability and
consequently final flotation recovery (Aktas et al., 2008). The froth stability may be
determinant although a shallow froth was used to minimize the interference in the results.
The collection zone kinetics has been directly calculated by sampling the ‘concentrate’ just
below the interface. Comparing results for the collection zone kinetics and the overall
flotation kinetics, proves that the effect of froth zone and pulp-froth interface do not change
the observed trend between bubble size or Sb and the overall flotation rate. Figure 6.3
shows the flotation rate constant trend for both collection zone and overall cases.
Figure 6.3 Kinetic constant as a function of d32 for the collection zone and the overall
process (53/75 µm, Jg= 1 cm/s)
The explanation must therefore lie in the antagonistic effect of the bubble size:
(1) for fine bubbles, a lower rise velocities contribute to lower flotation rates (Koh and
Schwarz, 2008), and
(2) the decreasing kinetics for large bubbles comes from a reduction of the collision
efficiency in fine particle with larger bubble sizes.
1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
0.1
0.15
0.2
0.25
(mm)
(1/m
in)
actual
collection zone kinetics
actual
overall flotation kinetics
d32
k
116
A nonlinear least square algorithm was applied to model the effect of the d32 on the
flotation kinetic, leading to
2
1 32 2 32 3
2
32 1 32 2
p d p d pk
d q d q
(6-1)
for the 75/106 µm particle class coefficients (with 95% confidence bounds) are
p1 = 0.1571 +/- 0.02
p2 = -0.4138 +/- 0.07
p3 = 0.2751 +/- 0.04 q1 = -2.67 +/- 0.09
q2 = 1.792 +/- 0.12
and for 53/75 µm particle class are
p1 = 0.09121 +/- 0.04
p2 = -0.2461 +/- 0.11
p3 = 0.167 +/- 0.08
q1 = -2.724 +/- 0.18
q2 = 1.859 +/- 0.25
Similarly, empirical models for the effect of the Sb on the kinetic constant are given by
3 2
1 2 3 4b b bk p S p S p S p (6-2)
for 53/75 µm particle size-class, correlation coefficients (with 95% confidence bound) are
p1 = 0
p2 = -0.0007658 +/- 0.0005
p3 = 0.06372 +/- 0.04
p4 = -1.137 +/- 0.8
and for 75/106 µm particle size-class the correlation coefficients are
p1 = -5.309e-005 +/- 4e-005
p2 = 0.006034 +/- 0.004
117
p3 = -0.2192 +/- 0.1
p4 = 2.744 +/- 2.2
It must be emphasized that the magnitude of the bubble size or Sb effect on the rate constant
was affected by the particle size. For a constant bubble size, the larger particle size-class
exhibits a higher kinetic constant.
Figure 6.4 and Figure 6.5 compare the prediction of the models with the actual kinetic
constant, as calculated from experimental results.
(a) (b)
Figure 6.4 Nonlinear regression for d32 and kinetic constant – Predicted and actual values;
(a) particle size-class 53/75 µm and (b) particle size-class 75/106 µm
0 0.05 0.1 0.15 0.2 0.25 0.30
0.05
0.1
0.15
0.2
0.25
0.3
kactual
km
od
el
(1/min)
SSE: 0.0012
R2: 0.926
RMSE: 0.020
0 0.05 0.1 0.15 0.2 0.25 0.30
0.05
0.1
0.15
0.2
0.25
0.3
kactual
km
od
el
(1/mim)
SSE: 0.0002
R2:0.95
RMSE:0.007
(1/m
in)
(1/m
in)
118
(a) (b)
Figure 6.5 Nonlinear regression for Sb and kinetic constant – Predicted and actual values;
(a) particle size-class 53/75 µm and (b) particle size-class 75/106 µm
6.3 Flotation kinetics of large particle size-class
The gas hold-up is the hydrodynamic variable presenting the best correlation with the
flotation rate of large particle size. Consequently, the gas hold-up was applied for flotation
rate modeling of particles in the 106/150 µm class, and results are presented in Figure 6.6.
As expected, the kinetic constant increases with the gas hold-up, but the fluctuation remains
on a modest scale in the tested range. This seems to indicate that for large particles,
hydrodynamic variables have only a moderate influence on the flotation rate constant.
0 0.05 0.1 0.15 0.2 0.25 0.30
0.05
0.1
0.15
0.2
0.25
0.3
kactual
km
od
el
SSE: 0.0033
R2: 0.79
RMSE: 0.0256
(1/min)0 0.05 0.1 0.15 0.2 0.25 0.3
0
0.05
0.1
0.15
0.2
0.25
0.3
kactual
km
od
el
(1/min)
SSE: 0.0005
R2:0.97
RMSE: 0.0097
(1/m
in)
(1/m
in)
119
Figure 6.6 Flotation rate constant as a function of εg for large particle size-class
(106/150 µm, Jg= 1cm/s)
A linear model relates εg to flotation kinetics and linear least square algorithm is applied for
curve fitting. The obtained model is
1 2gk p p (6-3)
whose coefficients (with 95% confidence bound) are
p1 =0.008583 +/- 0.003
p2 = 0.2022 +/- 0.02
Figure 6.7 compare the prediction of the models with the actual kinetic constant value, as
calculated from the experimental results.
3 4 5 6 7 8 9 10
0.23
0.24
0.25
0.26
0.27
0.28
0.29
g
k
(%)
(1/m
in)
120
Figure 6.7 Predicted and actual kinetic constants for the 106/150 µm particle size-class
6.4 Flotation kinetics of particle size-class spanning a wide range
The effect of Ib on the flotation rate was investigated for a wider particle size distribution
(53/150 µm). As can be seen in Figure 6.8, the flotation rate significantly increased with Ib.
Figure 6.8 Flotation rate constant as a function of Ib for large particle size-class
(53/150 µm, Jg= 1cm/s)
0 0.05 0.1 0.15 0.2 0.25 0.3 0.350
0.05
0.1
0.15
0.2
0.25
0.3
kactual
km
od
el
(1/min)
SSE: 0.0008
R2: 0.83
RMSE: 0.0106
10 15 20 25 30 35 40
0.2
0.22
0.24
0.26
0.28
0.3
0.32
Ib
k
(1/m
in)
(1/m
in)
121
Flotation kinetics depend on Ib through a linear model and the curve fitting is done with
linear least square algorithm. By means of
1 2bk p I p (6-4)
whose coefficients (with 95% confidence bound) are
p1 = 0.002931 +/- 0.0012 p2 = 0.1722 +/- 0.025
Flotation kinetics of larger range particle size are linked to interfacial area of bubbles.
Figure 6.9 compares the prediction of the models with the actual kinetic constant, as
calculated from the experimental results.
Figure 6.9 Linear regression for Ib and flotation kinetics for large range particle size-class
6.5 Particle size effect on the kinetics of flotation
Particle size remains the most influential parameter on the flotation rate. As can be seen in
Figures 6.1 and 6.2, the flotation rate significantly increases with the particle size. This is in
agreement with the VIP scores presented in Figure 5.1and previous studies (Diaz-Penafiel
and Dobby, 1994; Karimi et al., 2014; Ofori et al., 2014). However, in real flotation
conditions, the particle size is typically not a manipulated variable for the flotation circuit,
due to the required degree of liberation. The grinding circuit performance and ore
mineralogy determine the final particle size. Consequently the hydrodynamic variables are
0 0.05 0.1 0.15 0.2 0.25 0.3 0.350
0.05
0.1
0.15
0.2
0.25
0.3
kactual
km
od
el
(1/min)
SSE: 0.0031
R2: 0.80
RMSE: 0.0212
(1/m
in)
122
the only variables applicable to improve flotation performance for a given hydrophobicity
degree.
6.6 Conclusion
It was demonstrated that the relationship between the flotation rate constant and d32 (or Sb)
for fine particles, exhibits a maximum. The correlation fits to actual values with R2=0.93
for the 53/75 µm class, whereas for the 75/106 µm class the fit has a R2=0.95.
The kinetic constant for the 106/150 µm class linearly increases with increasing εg. The
same trend can be reported using Ib as the independent variable for a wider range of particle
size-class.
In the current study, the particle size has been considered as an important feature of mineral
particles. However, other features such as the particle hydrophobicity and the degree of
liberation are determinant on the flotation recovery.
The single-variable modeling introduces the appropriate hydrodynamic variable for
modeling and optimization with respect to the particle size. On the other hand, finding the
proper hydrodynamic variable to model the kinetic of flotation can better illustrate the role
of micro-phenomena (collision and attachment efficiencies) on the collection efficiency.
For example, when Sb presents high correlation with the flotation kinetics means that the
attachment efficiency plays a major role, while the effect of collision can be seen through
the relationship with εg.
123
Chapter 7 Thesis conclusion
Specific conclusions have been presented on each chapter, with cross-reference between
them. This Thesis conclusion chapter, summarizes the original contributions of the present
work, and provides some suggestions for future work.
The thesis addressed two main original contributions. The first one, presented in chapter 2
and 3 is related to the relationship between hydrodynamic variables, the water recovery and
carrying capacity in a laboratory flotation column. The second contribution consists in
better defining the relationships between hydrodynamic characteristics and the flotation
kinetics. This is described in chapters 4, 5 and 6.
Chapter 2 contains the effect of solid particles on the bubble size distribution; also gas
hold-up correlation in flotation column has been studied. It was demonstrated that bubble
coalescence affects the bubble size distribution through particle bridging effect and/or by
the frother absorption on talc (hydrophobic) particles. The solid particle effect on gas hold-
up is explained through the bubble loading and bubble coalescence. Consequently, it was
observed that a reduction in bubble size does not always lead to an increase on gas hold-up,
depending on which one of the latter factor is dominant.
In Chapter 3, the relevance of gas dispersion properties – as measured hydrodynamic
variables – on water recovery was studied. Tests were conducted in a laboratory scale
flotation column using a pulp with 4% solids content. They demonstrated a good
correlation between the water recovery and the gas hold-up, bubble size and bubble surface
area flux. It was demonstrated that increasing the gas hold-up and bubble surface area flux,
linearly increased the water recovery to the concentrate. The opposite effect was observed
for increasing bubble diameters. However, the hydrophobic particle size distribution did not
show any effect on the water recovery and consequently on the particle entrainment.
Modeling of gas dispersion properties in flotation was proposed in Chapter 4. The
fundamental equations of gas dispersion have been first developed as a possible solution for
some cases, where multi-shape bubble size density function and lognormal density function
have been well fitted to histograms. In this regard, the interfacial area of bubbles was
124
introduced as a hydrodynamic variable to address the issue of compacting the entire bubble
size distribution into a single value (d32) for the purpose of calculating the flotation rate
constant.
The inter-correlations between gas dispersion properties was also studied. A strong
correlation between gas dispersion properties – namely the gas hold-up and bubble surface
area flux – and interfacial area of bubbles has been detected for bubble size distributions
represented by lognormal density functions.
The difficulty of using multivariable regression techniques – because of strong co-linearity
between hydrodynamic variables – for flotation rate modeling was demonstrated in
Chapter 5. Projection to Latent Structures (PLS) was suggested to solve this problem as a
method using the covariance structure between independent variables.
A phenomenological analysis was presented to determine which hydrodynamic variable is
the most influential for recovering fine and coarse particle. Conclusions were that the
flotation of fine particles mostly relies on the bubble surface area flux, while coarse particle
flotation is mainly determined by the value of gas hold-up. Also, the interfacial area of
bubbles shows good correlation with the flotation kinetic constant for general cases,
floatable particles can be found in any size-class.
It was demonstrated that the flotation rate of fine particle size-class increases with the
bubble surface area flux to a maximum value and then decreases. The same trend was
reported for the relation between flotation rate and bubble size. Furthermore, the flotation
rate of coarse particles linearly increased by increasing the gas hold-up. Similar results
were reported for wider particle size distribution: increasing the interfacial area of bubbles
could directly increase the flotation rate constant. However, regardless of which
hydrodynamic is applied, the particle size shows a large influence on the flotation rate.
7.1 Future work
Future work should target the following issues.
Implementation of Microcell sparger instead of frit-sleeve sparger in the laboratory
and pilot scale flotation column to manipulate the bubble size. This device is used in
125
the industrial scale column and the extracted model for manipulating the bubble size
could then be applied to the industrial cells.
Validation of the model on the relationship between hydrodynamic variables and
flotation rate using original industrial data.
Investigating and modeling the froth zone and wash water effects on the final
recovery.
Modeling the kinetic constant distribution using the PLS technique. Bubble and
particle size distributions would be the input variables and the size-by-size flotation
rate constants would be the output.
Online measuring and controlling of bubble size distribution in slurry-gas systems
via manipulating the gas rate, frother concentration and the sparger shear-water rate.
127
References
Ahmed, N., Jameson, G.J., The effect of bubble size on the rate of flotation of fine
particles. International Journal of Mineral Processing, 1985, 14(3), 195-215.
Aktas, Z., Cilliers, J.J., Banford, A.W., Dynamic froth stability: Particle size, airflow rate
and conditioning time effects. International Journal of Mineral Processing, 2008, 87(1–
2), 65-71.
Arbiter, N., Harris,C.C.,, 1962. flotation kinetics: in froth flotation, ed. D.W.Fuerstenau,
New York.
Banisi, S., Finch, J., Laplante, A., Weber, M., Effect of solid particles on gas holdup in
flotation columns—II. Investigation of mechanisms of gas holdup reduction in presence
of solids. Chemical Engineering Science, 1995, 50(14), 2335-2342.
Bordel, S., Mato, R., Villaverde, S., Modeling of the evolution with length of bubble size
distributions in bubble columns. Chemical Engineering Science, 2006, 61(11), 3663-
3673.
Camacho, J., Picó, J., Ferrer, A., Multi‐phase analysis framework for handling batch
process data. Journal of Chemometrics, 2008, 22(11‐12), 632-643.
Cheng, T.-W., Holtham, P., The particle detachment process in flotation. Minerals
Engineering, 1995, 8(8), 883-891.
Chong, I.-G., Jun, C.-H., Performance of some variable selection methods when
multicollinearity is present. Chemometrics and Intelligent Laboratory Systems, 2005,
78(1), 103-112.
Cilek, E.C., The effect of hydrodynamic conditions on true flotation and entrainment in
flotation of a complex sulphide ore. International Journal of Mineral Processing, 2009,
90(1–4), 35-44.
Cole, K.E., Waters, K.E., Fan, X., Neethling, S.J., Cilliers, J.J., Combining Positron
Emission Particle Tracking and image analysis to interpret particle motion in froths.
Minerals Engineering, 2010, 23(11–13), 1036-1044.
128
Dai, Z.F., Fornasiero, D., Ralston, J., Particle-bubble collision models - a review. Advances
in Colloid and Interface Science, 2000, 85(2-3), 231-256.
Deglon, D.A., Egya-Mensah, D., Franzidis, J.P., Review of hydrodynamics and gas
dispersion in flotation cells on South African Platinum concentrators. Minerals
Engineering, 2000, 13(3), 235-244.
Diaz-Penafiel, P., Dobby, G.S., Kinetic-Studies in Flotation Columns - Bubble-Size Effect.
Minerals Engineering, 1994, 7(4), 465-478.
Dippenaar, A., The destabilization of froth by solids. I. The mechanism of film rupture.
International Journal of Mineral Processing, 1982, 9(1), 1-14.
Dobby, G.S., Finch, J.A., Particle-Size Dependence in Flotation Derived from a
Fundamental Model of the Capture Process. International Journal of Mineral
Processing, 1987, 21(3-4), 241-260.
Ekmekçi, Z., Bradshaw, D.J., Allison, S.A., Harris, P.J., Effects of frother type and froth
height on the flotation behaviour of chromite in UG2 ore. Minerals Engineering, 2003,
16(10), 941-949.
Eriksson, L., Johansson, E., Kettaneh-Wold, N., Wold, S., Multi-and megavariate data
analysis: principles and applications. 2001. Umea, Sweden: Umetrics, AB, 2001.
Finch, J.A., Dobby, G.S., Column flotation. 1990, Pergamon Press, New York.
Finch, J.A., Nesset, J.E., Acuña, C., Role of frother on bubble production and behaviour in
flotation. Minerals Engineering, 2008, 21(12), 949-957.
Finch, J.A., Xiao, J., Hardie, C., Gomez, C.O., Gas dispersion properties: Bubble surface
area flux and gas holdup. Minerals Engineering, 2000, 13(4), 365-372.
Flores-Cerrillo, J., MacGregor, J.F., Within-batch and batch-to-batch inferential-adaptive
control of semibatch reactors: A partial least squares approach. Industrial & engineering
chemistry research, 2003, 42(14), 3334-3345.
Fuerstenau, M.C., Jameson, G.J., Yoon, R.-H., Froth flotation: a century of innovation.
2007, SME.
129
Gandhi, B., Prakash, A., Bergougnou, M.A., Hydrodynamic behavior of slurry bubble
column at high solids concentrations. Powder Technology, 1999, 103(2), 80-94.
García-Salas, S., Rosales Peña Alfaro, M.E., Michael Porter, R., Thalasso, F., Measurement
of local specific interfacial area in bubble columns via a non-isokinetic withdrawal
method coupled to electro-optical detector. Chemical Engineering Science, 2008, 63(4),
1029-1038.
Gómez-Díaz, D., Navaza, J.M., Sanjurjo, B., Interfacial area evaluation in a bubble column
in the presence of a surface-active substance: Comparison of methods. Chemical
Engineering Journal, 2008, 144(3), 379-385.
Gomez, C.O., Cortés-López, F., Finch, J.A., Industrial testing of a gas holdup sensor for
flotation systems. Minerals Engineering, 2003, 16(6), 493-501.
Gorain, B.K., Franzidis, J.P., Manlapig, E.V., Studies on Impeller Type, Impeller Speed
and Air-Flow Rate in an Industrial-Scale Flotation Cell .1. Effect on Bubble-Size
Distribution. Minerals Engineering, 1995, 8(6), 615-635.
Gorain, B.K., Franzidis, J.P., Manlapig, E.V., Studies on impeller type, impeller speed and
air flow rate in an industrial scale flotation cell. Part 4: Effect of bubble surface area
flux on flotation performance. Minerals Engineering, 1997, 10(4), 367-379.
Gorain, B.K., Franzidis, J.P., Manlapig, E.V., The empirical prediction of bubble surface
area flux in mechanical flotation cells from cell design and operating data. Minerals
Engineering, 1999, 12(3), 309-322.
Gorain, B.K., Napier-Munn, T.J., Franzidis, J.P., Manlapig, E.V., Studies on impeller type,
impeller speed and air flow rate in an industrial scale flotation cell. Part 5: validation of
k-Sb relationship and effect of froth depth. Minerals Engineering, 1998, 11(7), 615-626.
Grau, R.A., Heiskanen, K., Visual technique for measuring bubble size in flotation
machines. Minerals Engineering, 2002, 15(7), 507-513.
Grau, R.A., Heiskanen, K., Bubble size distribution in laboratory scale flotation cells.
Minerals Engineering, 2005, 18(12), 1164-1172.
130
Guler, T., Akdemir, Ü., Statistical evaluation of flotation and entrainment behavior of an
artificial ore. Transactions of Nonferrous Metals Society of China, 2012, 22(1), 199-
205.
Heiskanen, K., On the relationships between flotation rate and bubble surface area flux.
Minerals Engineering, 2000, 13(2), 141-149.
Hernandez-Aguilar, J.R., Coleman, R.G., Gomez, C.O., Finch, J.A., A comparison between
capillary and imaging techniques for sizing bubbles in flotation systems. Minerals
Engineering, 2004, 17(1), 53-61.
Hernandez-Aguilar, J.R., Rao, S.R., Finch, J.A., Testing the k-S-b relationship at the
microscale. Minerals Engineering, 2005, 18(6), 591-598.
Höskuldsson, A., PLS regression methods. Journal of Chemometrics, 1988, 2(3), 211-228.
Jaeckle, C.M., MacGregor, J.F., Product design through multivariate statistical analysis of
process data. AIChE Journal, 1998, 44(5), 1105-1118.
Karimi, M., Akdogan, G., Bradshaw, S.M., A CFD-kinetic model for the flotation rate
constant, Part II: Model validation. Minerals Engineering, 2014(0).
Koh, P.T.L., Schwarz, M.P., Modelling attachment rates of multi-sized bubbles with
particles in a flotation cell. Minerals Engineering, 2008, 21(12–14), 989-993.
Kourti, T., Application of latent variable methods to process control and multivariate
statistical process control in industry. International Journal of Adaptive Control and
Signal Processing, 2005, 19(4), 213-246.
Kracht, W., Gomez, C.O., Finch, J.A., Controlling bubble size using a frit and sleeve
sparger. Minerals Engineering, 2008, 21(9), 660-663.
Kracht, W., Vallebuona, G., Casali, A., Rate constant modelling for batch flotation, as a
function of gas dispersion properties. Minerals Engineering, 2005, 18(11), 1067-1076.
Kuan, S.H., 2009. The Effect of Solids on Gas Holdup, Bubble Size and Water Overflow
Rate in Flotation, In Department of Mining, Metals and Materials Engineering. McGill
University, Montreal.
131
Kuan, S.H., Finch, J.A., Impact of talc on pulp and froth properties in F150 and 1-pentanol
frother systems. Minerals Engineering, 2010, 23(11), 1003-1009.
Liu, J., Gui, W., Tang, Z., Yang, C., Zhu, J., Li, J., Recognition of the operational statuses
of reagent addition using dynamic bubble size distribution in copper flotation process.
Minerals Engineering, 2013, 45(0), 128-141.
Luo, X., Lee, D., Lau, R., Yang, G., Fan, L.S., Maximum stable bubble size and gas holdup
in high‐pressure slurry bubble columns. AIChE Journal, 1999, 45(4), 665-680.
Maceiras, R., Álvarez, E., Cancela, M.A., Experimental interfacial area measurements in a
bubble column. Chemical Engineering Journal, 2010, 163(3), 331-336.
Maldonado, M., Desbiens, A., del Villar, R., An update on the estimation of the froth depth
using conductivity measurements. Minerals Engineering, 2008a, 21(12–14), 856-860.
Maldonado, M., Desbiens, A., del Villar, R., An update on the estimation of the froth depth
using conductivity measurements. Minerals Engineering, 2008b, 21(12–14), 856-860.
Maldonado, M., Desbiens, A., Del villar, R., Girgin, E., Gomez, C., 2008c. On-line
estimation of bubble size distributions using Gaussian mixture models, In V
International Mineral Processing Seminar, pp. 389-398.
Martínez-Carrillo, D., Uribe-Salas, A., An experimental study of the recovery of
hydrophilic silica fines in column flotation. Minerals Engineering, 2008, 21(15), 1102-
1108.
Massinaei, M., Kolahdoozan, M., Noaparast, M., Oliazadeh, M., Yianatos, J., Shamsadini,
R., Yarahmadi, M., Hydrodynamic and kinetic characterization of industrial columns in
rougher circuit. Minerals Engineering, 2009, 22(4), 357-365.
McGinnis, D.F., Little, J.C., Predicting diffused-bubble oxygen transfer rate using the
discrete-bubble model. Water Research, 2002, 36(18), 4627-4635.
Nelson, M.G., Lelinski, D., Hydrodynamic design of self-aerating flotation machines.
Minerals Engineering, 2000, 13(10–11), 991-998.
132
Nesset, J.E., Hernandez-Aguilar, J.R., Acuna, C., Gomez, C.O., Finch, J.A., Some gas
dispersion characteristics of mechanical flotation machines. Minerals Engineering,
2006, 19(6-8), 807-815.
Newell, R., Grano, S., Hydrodynamics and scale up in Rushton turbine flotation cells: Part
2. Flotation scale-up for laboratory and pilot cells. International Journal of Mineral
Processing, 2006, 81(2), 65-78.
Nguyen, A., Schulze, H.J., Colloidal science of flotation. 2003, CRC Press.
Nguyen, A.V., Particle–bubble encounter probability with mobile bubble surfaces.
International Journal of Mineral Processing, 1998, 55(2), 73-86.
Nutt, C., Froth flotation: The adhesion of solid particles to flat interfaces and bubbles.
Chemical Engineering Science, 1960, 12(2), 133-141.
O’Connor, C.T., Randall, E.W., Goodall, C.M., Measurement of the Effects of Physical and
Chemical-Variables on Bubbles Size. International Journal of Mineral Processing,
1990, 28(1-2), 139-149.
Ofori, P., O’Brien, G., Hapugoda, P., Firth, B., Distributed flotation kinetics models – A
new implementation approach for coal flotation. Minerals Engineering, 2014, 66–68(0),
77-83.
Phan, C.M., Nguyen, A.V., Miller, J.D., Evans, G.M., Jameson, G.J., Investigations of
bubble–particle interactions. International Journal of Mineral Processing, 2003, 72(1–
4), 239-254.
Polat, M., Chander, S., First-order flotation kinetics models and methods for estimation of
the true distribution of flotation rate constants. International Journal of Mineral
Processing, 2000, 58(1–4), 145-166.
Pyke, B., Fornasiero, D., Ralston, J., Bubble particle heterocoagulation under turbulent
conditions. Journal of Colloid and Interface Science, 2003, 265(1), 141-151.
Quinn, J.J., Finch, J.A., On the origin of bi-modal bubble size distributions in the absence
of frother. Minerals Engineering, 2012, 36–38(0), 237-241.
133
Rabha, S., Schubert, M., Wagner, M., Lucas, D., Hampel, U., Bubble size and radial gas
hold-up distributions in a slurry bubble column using ultrafast electron beam X-ray
tomography. AIChE J., 2013, 59, 1709–1722.
Ran, J., Liu, J., Zhang, C., Wang, D., Li, X., Experimental investigation and modeling of
flotation column for treatment of oily wastewater. International Journal of Mining
Science and Technology, 2013, 23(5), 665-668.
Reese, J., Jiang, P., Fan, L.-S., Bubble characteristics in three-phase systems used for pulp
and paper processing. Chemical Engineering Science, 1996, 51(10), 2501-2510.
Riquelme, A., Bouchard, J., Desbiens, A. and del Villar, R, 2013a. Bubble detection in
flotation columns based on Circular Hough Transform, In 23rd World Mining
Congress, Montreal, Quebec, Canada.
Riquelme, A., Bouchard, J., Desbiens, A. and del Villar, R., 2013b. Parameterization of
Bubble Size Distribution in Flotation Columns, In 15th IFAC symposium on Control,
Optimization and Automation in Mining, Mineral and Metal Processing, San Diego,
California, USA.
Rodrigues, W.J., Leal Filho, L.S., Masini, E.A., Hydrodynamic dimensionless parameters
and their influence on flotation performance of coarse particles. Minerals Engineering,
2001, 14(9), 1047-1054.
Roscoe, R., The viscosity of suspensions of rigid spheres. British Journal of Applied
Physics, 1952, 3(8), 267.
Sarrot, V., Guiraud, P., Legendre, D., Determination of the collision frequency between
bubbles and particles in flotation. Chemical Engineering Science, 2005, 60(22), 6107-
6117.
Sven-Nilsson, I., Effect of contact time between mineral and air bubbles on flotation. Kol
loid, 1934, 230.
Tan, Y.H., Rafiei, A.A., Elmahdy, A., Finch, J.A., Bubble size, gas holdup and bubble
velocity profile of some alcohols and commercial frothers. International Journal of
Mineral Processing, 2013, 119(0), 1-5.
134
Tao, D., Luttrell, G.H., Yoon, R.H., A parametric study of froth stability and its effect on
column flotation of fine particles. International Journal of Mineral Processing, 2000,
59(1), 25-43.
Tse, K., Martin, T., McFarlane, C., Nienow, A., Small bubble formation via a coalescence
dependent break-up mechanism. Chemical Engineering Science, 2003, 58(2), 275-286.
Tsutsumi, A., Nieh, J.-Y., Fan, L.-S., Particle wettability effects on bubble wake dynamics
in gas—liquid—solid fluidization. Chemical Engineering Science, 1991, 46(9), 2381-
2384.
Tucker, J.P., Deglon, D.A., Franzidis, J.P., Harris, M.C., Oconnor, C.T., An Evaluation of a
Direct Method of Bubble-Size Distribution Measurement in a Laboratory Batch
Flotation Cell. Minerals Engineering, 1994, 7(5-6), 667-680.
Uribe-Salas, A., de Lira-Gómez, P., Pérez-Garibay, R., Nava-Alonso, F., Magallanes-
Hernández, L., Lara-Valenzuela, C., Overloading of gas bubbles in column flotation of
coarse particles and effect upon recovery. International Journal of Mineral Processing,
2003, 71(1–4), 167-178.
Vazirizadeh, A., Bouchard, J., Villar, R.d., On the relationship between hydrodynamic
characteristics and the kinetics of column flotation. Part I: modeling the gas dispersion.
Minerals Engineering, 2014, IN PRESS.
Vera, M.A., Franzidis, J.P., Manlapig, E.V., Simultaneous determination of collection zone
rate constant and froth zone recovery in a mechanical flotation environment. Minerals
Engineering, 1999, 12(10), 1163-1176.
Vinnett, L., Contreras, F., Yianatos, J., Gas dispersion pattern in mechanical flotation cells.
Minerals Engineering, 2012, 26, 80-85.
Weber, M.E., Paddock, D., Interceptional and Gravitational Collision Efficiencies for
Single Collectors at Intermediate Reynolds-Numbers. Journal of Colloid and Interface
Science, 1983, 94(2), 328-335.
Wiese, J., Harris, P., Bradshaw, D., The effect of the reagent suite on froth stability in
laboratory scale batch flotation tests. Minerals Engineering, 2011, 24(9), 995-1003.
135
Wold, S., Sjöström, M., Eriksson, L., PLS-regression: a basic tool of chemometrics.
Chemometrics and Intelligent Laboratory Systems, 2001, 58(2), 109-130.
Wongsuchoto, P., Charinpanitkul, T., Pavasant, P., Bubble size distribution and gas-liquid
mass transfer in airlift contactors. Chemical Engineering Journal, 2003, 92(1-3), 81-90.
Xu, C., Gui, W., Yang, C., Zhu, H., Lin, Y., Shi, C., Flotation process fault detection using
output PDF of bubble size distribution. Minerals Engineering, 2012, 26(0), 5-12.
Yacoub, F., MacGregor, J.F., Product optimization and control in the latent variable space
of nonlinear PLS models. Chemometrics and Intelligent Laboratory Systems, 2004,
70(1), 63-74.
Yang, G.Q., Du, B., Fan, L.S., Bubble formation and dynamics in gas–liquid–solid
fluidization—A review. Chemical Engineering Science, 2007, 62(1–2), 2-27.
Yianatos, J., Bergh, L., Condori, P., Aguilera, J., Hydrodynamic and metallurgical
characterization of industrial flotation banks for control purposes. Minerals
Engineering, 2001, 14(9), 1033-1046.
Yianatos, J., Bergh, L., Pino, C., Vinnett, L., Muñoz, C., Yañez, A., Industrial evaluation of
a new flotation mechanism for large flotation cells. Minerals Engineering, 2012, 36–
38(0), 262-271.
Yianatos, J., Bucarey, R., Larenas, J., Henríquez, F., Torres, L., Collection zone kinetic
model for industrial flotation columns. Minerals Engineering, 2005, 18(15), 1373-1377.
Yianatos, J., Contreras, F., Particle entrainment model for industrial flotation cells. Powder
Technology, 2010, 197(3), 260-267.
Yianatos, J., Contreras, F., Díaz, F., Villanueva, A., Direct measurement of entrainment in
large flotation cells. Powder Technology, 2009, 189(1), 42-47.
Yianatos, J.B., Fluid flow and kinetic modelling in flotation related processes: Columns
and mechanically agitated cells - A review. Chemical Engineering Research & Design,
2007, 85(A12), 1591-1603.
136
Yianatos, J.B., Bergh, L.G., RTD studies in an industrial flotation column: use of the
radioactive tracer technique. International Journal of Mineral Processing, 1992, 36(1–
2), 81-91.
Yoon, R.H., Microbubble flotation. Minerals Engineering, 1993, 6(6), 619-630.
Yoon, R.H., The role of hydrodynamic and surface forces in bubble–particle interaction.
International Journal of Mineral Processing, 2000, 58(1–4), 129-143.
Zheng, J., 2001 Une nouvelle approche pour la détermination de la constante cinétique dans
les colonnes de flottation. Université Laval, Quebec.
Zheng, X., Johnson, N.W., Franzidis, J.P., Modelling of entrainment in industrial flotation
cells: Water recovery and degree of entrainment. Minerals Engineering, 2006, 19(11),
1191-1203.
Zhu, J., Gui, W., Yang, C., Xu, H., Wang, X., Probability density function of bubble size
based reagent dosage predictive control for copper roughing flotation. Control
Engineering Practice, 2014, 29(0), 1-12.
137
Appendix A
A.1 Sampling
Kinetic constant is estimated by means of recovery value and RTD. The mineral recovery is
calculated through Eq. (1-9). Representative sample for mineral assaying (from feed, tailing
and concentrate) assists to estimate more accurately the kinetic constant. Also for
estimating the kinetic constant of the collection zone, one sample per test has been taken
from 10 cm below the interface to avoid the effect of disturbance on sampling. It is worth
mentioning that the samples should be taken in the same time while the system is working
in the steady state conditions however due to system restrictions the sampling from feed
flow and collection zone have been done before sampling from the concentrate flow and the
tailing flow (feed sampling causes the system interruption). As seen in Figure A.1 the
sampling points are marked.
Figure A.1 Sampling points of the flotation column
Bubble sampling is also considered for investigation on the relationship between
hydrodynamic variables and the flotation rate constant. There is a region above the feeding
point and below the interface where axial mixing condition can be ignored. This district is
appropriate for bubble sampling. Besides, the sampling method should be considered. The
Collection zone sample
Tailing sample
Feed sample
Concentrate sample
138
bubbles should be extracted from the flotation column and transferred to the bubble viewer.
The angle of sampler, position of the aperture in the cross section of the column, sampler
diameter and sampler length should be considered for taking a representative bubble
sample. Based on that, the port of the bubble sampler was installed 50 cm above the
feeding point in the center of column cross section. The sampler is a hydrophilic tube (to
eliminate bubble attachment on the tube) with 12.5 mm aperture diameter and 60 cm
length. The sampler diameter is selected based on the maximum size of generated bubble
(two times larger). It is worth mentioning that the minimum detected bubble size is 0.1 mm
regarding to image resolution (1mm is equal 24 pixel). Figure A.2 shows the schematic of
bubble sampler. It is worth mentioning that the bubble viewer and bubble sampler have
been filled by water contains the same frother concentration as slurry in the column.
Figure A.2 Schematic of sampler and its position in the column
A.2 RTD measurement
The experimental methodology consists of introducing an impulse of tracer inside the
column at the feed entrance, and to measure the transient response (tracer concentration) at
tailing point. Experiments were done using a solution of NaCl as the liquid tracer. During
45º
50.4 mm
12.5 mm
600 mm
139
0 200 400 600 800 1000 1200 1400 1600
6150
6200
6250
6300
6350
6400
6450
6500
6550
6600
Time (s)
Me
asu
red
co
nd
uctivity (
mS
)
the RTD measuring the tailing flow did not re-circulate to the conditioning tank. The tracer
detector is the pair of conductivity cells in the feeding entrance and tailing point working in
the particular range of conductivity. Figure A.3 shows the tracer impulse in the feed and its
response in the tailing point. Table A.1 presents the measured RTDs were used for kinetic
constant estimation.
Figure A.3 Example of Measured RTD by conductivity cells; Left) tracer impulse in the
feeding point, Right) the detected response of tracer in the tailing point
200 400 600 800 1000 1200 1400 1600
1
2
3
4
5
6
7
x 104
Time (s)
Me
asu
red
co
nd
uctivity (
mS
)
140
Table A.1 Measured RTDs of the tests.
Particle size class (µm) 53/75 53/75 53/75 53/75 53/75 53/75 53/75 53/75 53/75
Frother (ppm) 5 5 5 15 15 15 25 25 25
Shear water rate (cm/s) 0.2 0.4 0.8 0.2 0.4 0.8 0.2 0.4 0.8
RTD 338 304 257 340 302 254 357 302 247
Particle size class (µm) 75/106 75/106 75/106 75/106 75/106 75/106 75/106 75/106 75/106
Frother concentration (ppm) 5 5 5 15 15 15 25 25 25
Shear water rate (cm/s) 0.2 0.4 0.8 0.2 0.4 0.8 0.2 0.4 0.8
RTD 337 307 257 334 303 254 343 302 255
Particle size class (µm) 106/150 106/150 106/150 106/150 106/150 106/150 106/150 106/150 106/150
Frother concentration (ppm) 5 5 5 15 15 15 25 25 25
Shear water rate (cm/s) 0.2 0.4 0.8 0.2 0.4 0.8 0.2 0.4 0.8
RTD 385 307 256 340 302 254 347 296 250
A.3 Kinetic constant calculation
After mineral assaying and data reconciliation, the flotation recovery is calculated. By
assuming R∞=1 and Eq. (A-1), the kinetic constant is estimated. The following examples
demonstrate the kinetic rate estimation.
ln(1 )100
60
R
k
(A-1)
Particle size (µm) Frother (ppm) Jsw (cm/s) R (%) τ (s) Kinetic constant (1/min)
75/106 15 0.8 66.5 254 0.26
53/75 15 0.4 56.1 302 0.16
141
Appendix B
B.1 Selection of the shear water rate
Determining Jsw values have been done by identifying the most effective Jsw values on
varying the bubble size in two-phase tests. In these tests, 20 ppm of F150 used as frother
and bubble size distributions (BSDs) were built by sizing 12,000 bubbles.
Figure B.1 Bubble size variation by shear water to a frit and sleeve sparger
Regarding to Figure B.1, the maximum bubble size variations occurs in 0.2 cm/s, 0.4 cm/s
and 0.8 cm/s, therefore these levels have been chosen for bubble size-particle size testing.
B.2 Effect of the temperature and duration of the test on the bubble size
The time required each set of three tests is over 8 hours. Therefore, the effect of the frother
should remain constant for this period of time. On the other hand, the frother should not be
sensitive in variation of the pulp temperature resulting from the action of the pumps and
agitator (the slurry is recycled). With this regard, F150 was tested for 8 hours of system
0 0.5 1 1.5 2 2.5 3 3.50
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Jsw
(cm/s)
(mm
)
data1
data2
data3
mean
variance
d32
shear water
&
bubble size
142
operation. By measuring the bubble size over time, the effect of temperature and test
duration was assessed. Table B.1 depicts the variation of the bubble size with time and
temperature. As seen, the effect of F150 remains constant for the duration of the test
Figure B.2 shows that the d32 is relatively stable for the duration of the test.
Table B.1 Effect of the temperature and duration of test on the bubble size distribution
Time (hour) 1 2 3 4 5 6 7 8
Temperature (oC) 19.4 19.6 20.2 20.9 21.4 21.9 22.5 22.8
d32 (mm) 0.89 0.781 0.791 0.782 0.780 0.774 0.785 0.773
mean (mm) 0.579 0.54 0.479 0.491 0.474 0.512 0.51 0.489
variance 0.0845 0.065 0.0718 0.0708 0.0691 0.0686 0.0706 0.0731
The system needed one hour of operation before reaching steady-state conditions thus
sampling have always been executed following at least one hour of test. In addition, it must
be emphasized that the slurry required on hour of conditioning.
Figure B.2 d32 variations over time
B.3 Relationships between hydrodynamic variables in two and three-phases
Studying the response of measured variables to manipulated variable changes allowed
determining the range of operation for hydrodynamic conditions in the experimental set-up.
0 1 2 3 4 5 6 7 80.75
0.8
0.85
0.9
Time (hours)
d32 (
mm
)
143
The shearwater rate to the sparger (Jsw) and the gas flow rate (Jg) are the manipulated
variables. The frother concentration remained constant (20 ppm of F150) in all the tests.
This notwithstanding, manipulating Jsw and Jg allowed controlling the bubble size and gas
hold-up (as measured hydrodynamic variables). Table B.2 shows values of hydrodynamic
variables generated by manipulating Jg and Jsw.
Table B.2 Values of hydrodynamic variables generated by manipulating Jg and Jsw
εg (%)
Jg (cm/s)/Jsw (%) 0.25 0.375 0.5 0.75 1 1.25
0 1.80 2.48 3.11 4.32 5.19 6.23
25 2.43 3.27 3.88 5.19 6.09 6.86
35 2.30 3.38 4.03 5.91 7.76 9.35
45 2.35 3.32 4.23 6.29 8.57 10.44
65 2.627 3.68 5.00 7.53 10.30 10.80
d32 (mm)
0 1.06 1.00 1.13 1.22 1.40 1.38
25 1.12 1.05 1.06 1.22 1.26 1.38
35 0.85 0.93 1.01 1.02 1.07 1.08
45 0.86 0.87 0.91 0.95 0.98 1.05
65 0.78 0.80 0.83 0.85 0.93 0.91
Sb (s-1)
0 14.13 20.83 26.34 36.83 42.69 53.99
25 13.27 21.27 28.19 36.82 47.43 54.18
35 17.59 24.08 29.60 43.99 56.03 69.34
45 17.27 25.57 32.62 46.91 60.65 71.24
65 19.03 27.87 35.95 52.39 64.25 82.27
mean (mm)
0 0.86 0.81 0.86 0.91 0.95 0.91
25 0.96 0.89 0.86 0.95 0.92 0.95
35 0.80 0.83 0.89 0.90 0.92 0.93
45 0.76 0.76 0.78 0.82 0.82 0.86
65 0.70 0.72 0.74 0.76 0.82 0.77
variance
0 0.07 0.06 0.10 0.12 0.18 0.18
25 0.07 0.06 0.07 0.11 0.14 0.18
35 0.03 0.04 0.05 0.06 0.07 0.08
45 0.03 0.04 0.04 0.05 0.06 0.07
65 0.028 0.02 0.03 0.03 0.04 0.05
144
Figure B.3 shows how modifying the gas rate and shearwater rate can generate a large
range of bubble size and gas hold-up values.
(a)
(b)
0 10 20 30 40 50 60
0.8
0.9
1
1.1
1.2
1.3
1.4
Jsw
(%)
d32 (
mm
)
Jg 0.25 cm/s
Jg 0.375 cm/s
Jg 0.5 cm/s
Jg 0.75 cm/s
Jg 1 cm/s
Jg 1.25 cm/s
data8
data9
data10
data11
data12
0 10 20 30 40 50 60
2
3
4
5
6
7
8
9
10
11
Jsw
(%)
g (
%)
Jg 0.25 cm/s
Jg 0.375 cm/s
Jg 0.5 cm/s
Jg 0.75 cm/s
Jg 1 cm/s
Jg 1.25 cm/s
data8
data9
data10
data11
data12
145
(c)
Figure B.3 a) Variations of the d32 with the gas and shearwater rates, b) variations of εg
with the gas and shearwater rates and c) variations of Sb with the gas and shearwater rates
0 10 20 30 40 50 60
20
30
40
50
60
70
80
Jsw
(%)
Sb (
1/s
)
0 10 20 30 40 50 60
2
3
4
5
6
7
8
9
10
11
Jsw
(%)
g (
%)
Jg 0.25 cm/s
Jg 0.375 cm/s
Jg 0.5 cm/s
Jg 0.75 cm/s
Jg 1 cm/s
Jg 1.25 cm/s
data8
data9
data10
data11
data12
147
Appendix C
C.1 Solid Characteristics
The solid used contains quartz as a hydrophilic mineral and raw rock from a talc mine
around Quebec City. Based on the chemical analysis and microscopy results, quartz is 99%
liberate and does not show any significant contamination. Table C.1 shows the chemical
analysis results.
Table C.1 XRF results of quartz (%)
SiO2 Al2O3 Fe2O3 MgO CaO Na2O
98.7 0.2 1.1 0.04 0.01 <0.04
K2O TiO2 MnO P2O5 Cr2O3 LOI
0.01 <0.04 <0.01 0.01 0.02 <0.02
Crushing, grinding and two stages of wet screening have been done to prepare the desired
liberated particle size of talc for three different particle size classes. Table C.2 shows the
mineralogy analysis results of each particle size class. Table C.3 presents the XRF results
of the particle size classes. Figure C.1 depicts the liberated information about each class of
talc particles.
148
Table C.2 Mineralogy analysis results
Mineral 53/75 m 75/106 m 106/150 m
Chrome-Chlorite (Kämmererite) 16.9 17.7 20.4
Mg-Fe carbonate 10.2 14.9 18.3
Talc 66.0 61.4 54.1
Ankerite 4.4 3.8 5.0
Serpentine group 0.1 0.3 0.4
Quartz 0.1 0.3 0.2
Pyrrhotite 1.2 0.8 0.4
Pentlandite 0.8 0.8 0.8
Fe-oxi 0.3 0.1 0.2
Electrum 0.0 0.0 0.0
Zircon 0.0 0.0 0.0
Apatite 0.0 0.0 0.1
Total 100.0 100.0 100.0
Figure C.1 Mineral liberation – talc
149
Table C.3 XRF results for talc particles (%)
SiO2 Al2O3 Fe2O3 MgO CaO Na2O
51.2 1.8 5.8 31.1 0.43 < 0.04
48 1.8 6.2 32.1 0.47 < 0.04
43.3 2 6.8 32.3 0.52 < 0.04
K2O TiO2 MnO P2O5 Cr2O3 LOI
< 0.01 < 0.04 0.05 < 0.01 0.31 11
< 0.01 < 0.04 0.05 < 0.01 0.37 13.2
< 0.01 < 0.04 0.06 0.01 0.42 15.8
151
Appendix D
D.1 Radial gas dispersion analysis (Banisi et al., 1995)
The distribution parameter (C0,) is defined as
0
1( )
1 1( )
g g sl
g g sl
J J dAAC
dA J J dAA A
. (D-1)
where A is the column cross-section and Jsl is superficial slurry velocity. C0 takes into
account the effect of non uniform (i.e. non-flat) flow and profile in the cross section of the
column.
The terminal velocity coefficient (K0) is calculated from
0
1(1 )
1 1(1 )
m
g g
m
g g
dAAK
dA dAA A
(D-2)
where m = 3 for both the gas-slurry and gas-water systems. K0 is a parameter encompassing
both the uniformity of gas hold-up profiles and the hindering effect of the gas bubbles on
bubble rise velocity.
The gas hold-up and superficial velocity profiles are flat in section A if C0 = K0 = 1.
The difference between C0 in the gas-water (C0 = 1.12) and gas-slurry (C0 = 2.21) systems
suggested a change from flat gas hold-up and flow profiles in the water-gas system to non-
uniform profiles in the slurry-gas system.
153
Appendix E
Particle size(µm) F150 (ppm) Shear water (cm/s) εg (%) µ σ2 d32 (mm)
53/75 5 0.2 4.02 1.66 0.48 2.07
53/75 5 0.4 5.17 1.57 0.34 1.92
53/75 5 0.8 5.55 1.40 0.30 1.78
53/75 15 0.2 5.72 1.03 0.18 1.40
53/75 15 0.4 7.54 0.96 0.14 1.30
53/75 15 0.8 8.69 0.89 0.14 1.28
53/75 25 0.2 7.1 0.97 0.17 1.36
53/75 25 0.4 9.8 0.86 0.12 1.21
53/75 25 0.8 12.48 0.85 0.11 1.17
75/106 5 0.2 3.13 0.99 0.89 2.17
75/106 5 0.4 4.25 1.41 0.48 1.92
75/106 5 0.8 5.25 1.44 0.34 1.80
75/106 15 0.2 5.85 1.09 0.19 1.47
75/106 15 0.4 8.35 0.99 0.14 1.32
75/106 15 0.8 9.93 0.94 0.14 1.30
75/106 25 0.2 6.85 0.96 0.16 1.35
75/106 25 0.4 9.68 0.92 0.13 1.26
75/106 25 0.8 10.85 0.92 0.15 1.30
106/150 5 0.2 2.54 1.21 0.82 2.11
106/150 5 0.4 3.59 1.36 0.49 1.91
106/150 5 0.8 5.87 1.32 0.40 1.80
106/150 15 0.2 4.30 0.99 0.17 1.38
106/150 15 0.4 6.54 0.98 0.15 1.33
106/150 15 0.8 8.47 0.99 0.15 1.34
106/150 25 0.2 4.58 0.94 0.17 1.35
106/150 25 0.4 7.93 0.94 0.13 1.24
106/150 25 0.8 10.46 0.91 0.11 1.18
53/150 5 0.2 2.21 1.40 0.77 2.10
53/150 5 0.4 3.14 1.48 0.45 1.92
53/150 5 0.8 4.53 1.53 0.28 1.83
53/150 15 0.2 4.69 1.05 0.17 1.39
53/150 15 0.4 7.25 1.02 0.13 1.31
53/150 15 0.8 8.00 1.00 0.12 1.27
53/150 25 0.2 5.11 1.01 0.21 1.45
53/150 25 0.4 7.85 0.99 0.16 1.33
53/150 25 0.8 9.34 0.96 0.15 1.31
154
Particle size(µm) F150 (ppm) Shear water (cm/s) Feed G. Cons. G. Tailing G. Recovery (%)
53/75 5 0.2 33.61 82.04 24.81 37.55
53/75 5 0.4 36.11 82.03 24.07 47.21
53/75 5 0.8 38.11 81.39 29.41 35.74
53/75 15 0.2 34.85 80.39 17.01 64.92
53/75 15 0.4 37.47 81.01 22.21 56.10
53/75 15 0.8 38.10 82.32 27.84 40.69
53/75 25 0.2 37.16 81.21 15.62 71.78
53/75 25 0.4 38.92 80.39 25.35 50.91
53/75 25 0.8 37.61 81.51 29.10 35.20
75/106 5 0.2 38.41 83.77 20.42 61.93
75/106 5 0.4 39.35 79.95 22.93 58.49
75/106 5 0.8 41.54 81.21 27.10 52.16
75/106 15 0.2 37.16 79.05 14.39 74.91
75/106 15 0.4 39.97 79.88 16.02 74.95
75/106 15 0.8 41.22 81.51 20.80 66.51
75/106 25 0.2 38.41 80.49 13.56 77.80
75/106 25 0.4 39.77 80.69 14.71 77.05
75/106 25 0.8 39.66 79.88 19.28 67.75
106/150 5 0.2 40.99 77.73 17.62 73.73
106/150 5 0.4 40.96 82.43 19.47 68.69
106/150 5 0.8 41.75 79.48 22.60 64.11
106/150 15 0.2 39.13 82.16 15.72 73.99
106/150 15 0.4 41.74 82.70 17.73 73.21
106/150 15 0.8 39.56 76.38 19.71 67.63
106/150 25 0.2 46.70 82.16 21.93 72.34
106/150 25 0.4 44.17 76.07 18.87 76.17
106/150 25 0.8 47.58 76.38 24.93 70.66
53/150 5 0.2 39.26 80.70 18.86 67.81
53/150 5 0.4 39.78 80.20 22.37 60.71
53/150 5 0.8 40.72 79.88 24.35 57.84
53/150 15 0.2 40.07 80.18 18.20 70.62
53/150 15 0.4 41.47 80.31 20.19 68.55
53/150 15 0.8 38.01 79.05 17.43 69.45
53/150 25 0.2 39.46 80.50 18.56 68.84
53/150 25 0.4 39.47 79.37 19.67 66.70
53/150 25 0.8 38.21 79.97 17.04 70.40
155
Appendix F
F.1 Correlation between hydrodynamic variables
The purpose of this section is to assess the capability of classical regression method to
predict k by means of the hydrodynamic variables. It has shown here that, due to
correlation between these parameters, and consequently the sensitivity of the (XTX)-1 to
small variations, the traditional regression methods are not the appropriate tools. Therefore
in this work PLS regression has been implemented.
Fine particle class of test work (53/75 microns)
When X is mean-centered and scaled to unit length, the XTX shows the correlation between
variables:
The following matrices show the sensitivity of (XTX)-1 to an arbitrary ‘small’ change,
Δ=0.01, in the diagonal elements. The regression could be applied only if ‘small’ changes
lead to ‘small’ changes in . Otherwise, the correlation between the input
variables would be too important, and the results would be biased.
In the present case,
and thus,
1.000 0.889 0.996
0.889 1.000 0.9185
0.889 0.9185 1.000
g b b
T
S I
X X
1( )TX X
1
281.995 45.718 322.904
( ) 45.718 13.805 58.223
322.904 58.223 376.142
TX X
1
72.499 11.048 82.368
( ) 11.048 7.693 18.070
82.368 18.070 99.650
TX X
156
As can be seen regression is not suggested for this data base because the changes between
diagonal elements of and is too large.
Coarse particle class of test work (106/150 microns)
The differences between the diagonal elements of and is more
than normal therefore the same condition as fine particle size class occurs in coarse particle
size class.
Gorain et al data base
The same results as test works is achieved therefore regression cannot be a good method for
modeling of kinetic constant.
1( )TX X 1( )TX X
1.000 0.780 0.984
0.780 1.000 0.856
0.984 0.856 1.000
g b b
T
S I
X X
1
55.45 12.660 65.384
( ) 12.660 6.623 18.123
65.384 18.123 80.835
TX X
1
35.040 7.713 41.075
( ) 7.713 5.296 12.120
41.075 12.120 51.782
TX X
1( )TX X 1( )TX X
1.000 0.901 0.727
0.901 1.000 0.888
0.727 0.888 1.000
b g b
T
I S
X X
1
6.143 7.414 2.115
( ) 7.414 13.671 6.747
2.115 6.747 5.452
TX X
1
5.356 6.173 1.585
( ) 6.173 11.625 5.831
1.585 5.831 5.025
TX X
157
F.2 Regression methods
In order to model a response matrix (Y) by means of a predictor matrix (X), the Multiple
Linear Regression (MLR) approach is usually applied, as long as the X-variables are few
and fairly uncorrelated (Wold et al., 2001). In multiple linear regression (MLR), the X links
to Y matrices directly using a linear relation:
𝐘 = 𝐗𝐁 + 𝐄 (F-1)
where B is the matrix regression coefficients and E consists of the regression residuals,
often assumed to be identically and independently normally distributed. A number of
procedures have been developed for estimation of B in linear regression. The simplest and
most commonly method is applying ordinary least squares (OLS) where the regression
parameters is estimated by minimizing the sum of squared residuals (E):
𝐄 = 𝐘 − 𝐗𝐁 (F-2)
�̂� = (𝐗𝐓𝐗)−𝟏𝐗𝐓𝐘 (F-3)
where �̂� is the estimated regression parameters.
However, since the variables describing the hydrodynamic conditions of flotation
(X-variables) are strongly correlated the alternatives can be PCR (principal component
regression) or PLS (Projection to Latent Structures analysis, also known as the Partial Least
Squares) to model (Y) by (X).
In PCR, the X matrix is replaced by the T matrix computed by the PCA (the dimensions
corresponding to the largest eigenvalues are only kept). In the next step, the multi-linear
regression can be expressed as:
𝐘 = 𝐓𝐁 + 𝐄 (F-4)
�̂� = (𝐓𝐓𝐓)−𝟏𝐓𝐓𝐘 (F-5)
Similar to MLR, B and E correspond to the regression parameters and residuals,
respectively. The matrix TTT is diagonal since the columns of T are orthogonal, so the
elements of the inverse are merely the mutual of the diagonal elements. Therefore, PCR
avoids the problems corresponded to colinearity.
158
However, in the PCR approach, the decomposition of X (i.e. projection onto a lower latent
variables subspace) is done without using information from Y, and this dimensional
reduction is not essentially the most predictive of Y. It is worth mentioning that merely
selecting the components with the highest variance (large eigenvalues) may as well be
problematic.
The third method discussed here is the Projection to Latent Structures (PLS). This method
was used throughout this work to build the empirical latent variable models linking the
Hydrodynamic variables to the flotation rate constant. PLS was selected in this work since
the columns of X were highly collinear (proved in the previous section of this appendix).
Due to simplicity and small volume of calculations, the PLS method has been extensively
implemented in different research fields such as product design and process analysis and
optimization (Camacho et al., 2008; Flores-Cerrillo and MacGregor, 2003; Jaeckle and
MacGregor, 1998; Kourti, 2005; Yacoub and MacGregor, 2004). In fact the PLS regression
relates two blocks of variables, X and Y, by maximizing the covariance between them
(Höskuldsson, 1988).
The purpose of the PLS regression is to explain the relationship between X (N×K) and Y
(N×M) matrices, within each data-set, by the following equations (Eriksson et al., 2001):
𝐗 = 𝐓𝐏′ + 𝐄 (F-6)
𝐘 = 𝐓𝐐′ + 𝐅 (F-7)
𝐓 = 𝐗𝐖∗ (F-8)
𝐖∗ = 𝐖(𝐏′𝐖)−1 (F-9)
where the P (K×r) and Q (M×r) are loading matrices that respectively represent the X and
Y spaces, and T (N×r) contains a set of orthogonal components (i.e. latent variables),
defined by linear combination of the X-variables collected in the weight matrix of
W*(K×r). The E (N×K) and F (N×M) matrices respectively represent the model residuals
for X and Y. A cross-validation procedure is often used to select the number of PLS
components or latent variables r (r < K).
159
Another advantage of PLS approach along with solving the colinearity problems is
commonly used metric for judging the importance of the X- variables on the Y. In order to
quantify the importance of the X variables in the PLS model, according to their
contribution to explain the variance of Y by each PLS component, the Variable Importance
on the Projection (VIP) is defined as (Chong and Jun, 2005):
VIPk = √K ∑ ((
wak‖𝐰𝐚‖
)2
(𝐪𝐚𝟐𝐭𝐚
′ 𝐭𝐚))ra=1
∑ (𝐪𝐚𝟐𝐭𝐚
′ 𝐭𝐚)ra=1
(F-10)
where K is the total number of variables, 𝐰𝐚𝐤 is the weight of kth variable in principal
component a, r is the number of principal components, and wa, ta and qa are the ath column
vector of W and T and Q respectively. According to Eriksson et al. (20012001), the
variables with a VIP greater than 1, are most influential in the PLS model, because the
average of the squared VIPs is equal to 1. The variables having a VIP between 0.8 and 1.0
correspond to those having a moderate influence, and VIP less than 0.8 correspond to the
least influential ones.
160
F.3 Datasets
Table F.1 Experimental results
Particle size(µm) F150 (ppm) Shear water (cm/s) εg d32 (mm) Sb Ib k (1/min)
53/75 5 0.2 4.02 2.07 28.91 11.61 0.07
53/75 5 0.4 5.17 1.92 31.30 16.18 0.12
53/75 5 0.8 5.55 1.78 33.75 18.75 0.10
53/75 15 0.2 5.72 1.40 42.16 24.12 0.18
53/75 15 0.4 7.54 1.30 45.787 34.54 0.16
53/75 15 0.8 8.69 1.28 46.87 40.76 0.12
53/75 25 0.2 7.1 1.36 44.23 31.39 0.22
53/75 25 0.4 9.8 1.21 49.71 48.98 0.16
53/75 25 0.8 12.48 1.17 51.27 63.98 0.10
75/106 5 0.2 3.13 2.17 27.68 8.67 0.17
75/106 5 0.4 4.25 1.92 31.28 13.28 0.17
75/106 5 0.8 5.25 1.80 32.75 17.21 0.17
75/106 15 0.2 5.85 1.47 40.71 23.83 0.23
75/106 15 0.4 8.35 1.32 45.478 37.99 0.27
75/106 15 0.8 9.93 1.30 46.10 45.77 0.26
75/106 25 0.2 6.85 1.35 44.32 30.36 0.30
75/106 25 0.4 9.68 1.26 47.66 46.13 0.21
75/106 25 0.8 10.85 1.30 46.06 50.00 0.27
106/150 5 0.2 2.54 2.11 28.38 7.22 0.24
106/150 5 0.4 3.59 1.91 31.44 11.30 0.23
106/150 5 0.8 5.87 1.80 33.38 19.61 0.24
106/150 15 0.2 4.30 1.38 43.56 18.75 0.24
106/150 15 0.4 6.54 1.33 45.25 29.59 0.26
106/150 15 0.8 8.47 1.34 44.82 37.96 0.26
106/150 25 0.2 4.58 1.35 44.57 20.41 0.22
106/150 25 0.4 7.93 1.24 48.25 38.25 0.28
106/150 25 0.8 10.46 1.18 50.93 53.27 0.29
Mixed 5 0.2 2.21 2.10 28.60 6.31 0.20
Mixed 5 0.4 3.14 1.92 31.17 9.79 0.18
Mixed 5 0.8 4.53 1.83 32.78 14.86 0.20
Mixed 15 0.2 4.69 1.39 43.17 20.26 0.22
Mixed 15 0.4 7.25 1.31 45.92 33.31 0.24
161
Mixed 15 0.8 8.00 1.27 47.22 37.80 0.28
Mixed 25 0.2 5.11 1.45 41.23 21.075 0.21
Mixed 25 0.4 7.85 1.33 45.19 35.48 0.22
Mixed 25 0.8 9.34 1.31 45.73 42.69 0.29
Table F.2 Gorain's results for Chile-X impeller k= flotation rate, 1/min; Sb = Bubble surface
area flux (1/sec), Jg = superficial gas velocity (cm/sec), d32 =Sauter mean bubble diameter
(mm), εg = Gas holdup (%), Ib interfacial area of bubbles %
Ib Jg εg d32 Sb k
21.41 0.68 3.64 1.02 40 0.10
28.29 1.94 6.13 1.30 89.534 0.23
45.14 2.77 11.06 1.47 113.06 0.26
54.57 2.34 13.46 1.48 94.86 0.21
34.47 0.65 5.17 0.90 43.33 0.10
55.20 1.42 8.74 0.95 89.68 0.20
58.36 2.99 11.77 1.21 148.26 0.43
78.42 3.07 16.86 1.29 142.79 0.43
50.15 0.64 6.77 0.81 47.41 0.12
55.30 1.38 8.48 0.92 90.00 0.22
73.07 2.52 13.64 1.12 135.00 0.40
85.28 3.03 17.34 1.22 149.01 0.50
49.06 0.8 6.46 0.79 60.76 0.14
71.24 0.98 9.38 0.79 74.43 0.18
92.20 1.68 15.06 0.98 102.86 0.33
108.38 3.06 20.05 1.11 165.40 0.54
162
Table F.3. Massinaei et al. data base
industrial
Ib Jg εg d32 Sb k
35.53 1.1 8.35 1.41 46.81 0.15
32.39 1.1 7.99 1.48 44.59 0.11
27.45 1.1 7.32 1.6 41.25 0.10
36.21 1.6 10.08 1.67 57.48 0.26
37.20 1.6 10.23 1.65 58.18 0.19
35.14 2 11.42 1.95 61.54 0.20
Pilot-
scale
12.41 1.1 16.34 0.79 83.54 0.13
10.23 1.1 14.49 0.85 77.65 0.15
9.06 1.1 13.59 0.9 73.33 0.15
20.24 1.6 26.32 0.78 123.08 0.17
19.44 1.6 25.6 0.79 121.52 0.17
17.94 2 28.1 0.94 127.66 0.19
163
Table F.4 Kracht et al. data base
Ib Jg εg d32 Sb k
Outotec cell
74.22 1.2 12 0.97 74.23 0.60
64.82 1.5 12.1 1.12 80.36 0.58
75.00 1.8 13 1.04 103.85 0.7
57.10 2.1 11.8 1.24 101.61 0.83
87.79 1.2 13.9 0.95 75.79 0.72
81.65 1.5 13.2 0.97 92.78 0.86
105.33 1.8 15.8 0.9 120 0.88
60.00 2.1 13.4 1.34 94.03 0.86
102.76 1.2 14.9 0.87 82.76 0.90
108.71 1.8 18.3 1.01 106.93 0.90
86.28 2.1 17.4 1.21 104.13 0.96
103.14 1.2 15.3 0.89 80.90 0.95
122.89 1.5 17 0.83 108.43 0.99
98.53 1.8 15.6 0.95 113.68 1.04
110.59 2.1 18.8 1.02 123.53 1.13
102.47 1.7 15.2 0.89 114.61 0.85
Labtech-Essa
cell
46.45 0.4 9.6 1.24 19.35 1.07
105.00 0.8 14.7 0.84 57.14 1.21
129.88 1.3 18.4 0.85 91.76 1.33
89.59 0.4 10.9 0.73 32.88 1.03
113.68 0.8 14.4 0.76 63.16 1.24
116.88 1.3 15 0.77 101.30 1.44
82.22 0.4 11.1 0.81 29.63 1.17
77.08 1.3 16.7 1.3 60 1.44
83.70 0.4 11.3 0.81 29.63 1.19
98.00 0.8 14.7 0.9 53.33 1.46
101.87 1.3 16.3 0.96 81.25 1.52
164
.
Table F.5 Jincai et al. data base
Ib Jg εg d32 Sb k
10.09 2.35 5.72 3.40 41.47 0.62
39.63 3.14 8.72 1.32 142.72 1.08
101.50 3.14 12.35 0.73 258.08 1.47
50.55 2.35 8.93 1.06 133.02 1.20
136.38 3.14 11.82 0.52 362.31 2.07
15.63 3.92 6.41 2.46 95.61 0.72
101.91 2.35 11.72 0.69 204.35 1.74
16.19 3.14 6.53 2.42 77.85 0.74
42.57 3.92 9.01 1.27 185.19 1.18