Proceedings of the Control Conference Africa, 2017

Proceedings of the Control Conference Africa, 2017

Johannesburg,
Proceedings South Africa, December Africa,
7-8, 2017
Proceedings of
Johannesburg, the
the Control
of South
Control Conference
Africa, December
Conference 2017
7-8, 2017
Available
Africa, online at www.sciencedirect.com
2017
Johannesburg,
Johannesburg, South Africa, December 7-8,
South Africa, December 7-8, 2017
2017

ScienceDirect
IFAC PapersOnLine 50-2 (2017) 19–24
On
On the
the current
current state
state of
of flotation
flotation modelling
modelling for
for process
process control
control
On
On the current state, of flotation modelling for process
the current state of flotation modelling for process control
control
D.J.
D.J. Oosthuizen
Oosthuizen **,,**, **, I.K.
I.K. Craig
Craig **,**, S-L. Jämsä-Jounela ***,
S-L. Jämsä-Jounela ***, B.
B. Sun
Sun ******
D.J. Oosthuizen * ,**, I.K. Craig **, S-L. Jämsä-Jounela ***, B. Sun ***
D.J. Oosthuizen * **, I.K. Craig **, S-L. Jämsä-Jounela ***, B. Sun ***
*MINTEK,
*MINTEK, Randburg, 
Randburg, South Africa
South Africa
(Tel: 011 709
*MINTEK,
(Tel: 011*MINTEK, 4704; e-mail:
Randburg,
709 4704;Randburg, kobuso@mintek.co.za).
South Africa
South Africa
e-mail: kobuso@mintek.co.za).
**
** University
(Tel:
(Tel: 011
011 709 of
of Pretoria,
709 4704;
University 4704; e-mail: Pretoria,
e-mail:
Pretoria, South
South Africa
kobuso@mintek.co.za).
kobuso@mintek.co.za).
Pretoria, Africa
** University
** University (e-mail:ian.craig@up.ac.za)
of
of Pretoria, Pretoria, South
Pretoria, Pretoria,
(e-mail:ian.craig@up.ac.za) South Africa
Africa
***
*** Aalto
Aalto University,
University, Espoo,
Espoo, Finland
Finland (e-mail:
(e-mail: sirkka-liisa.jamsa-jounela@aalto.fi)
(e-mail:ian.craig@up.ac.za)
(e-mail:ian.craig@up.ac.za) sirkka-liisa.jamsa-jounela@aalto.fi)
***
*** Aalto
Aalto University,
University, Espoo,
Espoo, Finland
Finland (e-mail:
(e-mail: sirkka-liisa.jamsa-jounela@aalto.fi)
sirkka-liisa.jamsa-jounela@aalto.fi)
Abstract: Despite significant effort in modelling and simulating flotation
Abstract: Despite significant effort in modelling and simulating flotation circuits,
circuits, comprehensive
comprehensive model model
based
Abstract:
Abstract: control
based control and
Despite
Despite optimisation
significant
and significant
optimisation implementations
effort
effort in modelling
in modelling and
implementations on
and industrial
simulating circuits
simulatingcircuits
on industrial remain
flotation
flotation scarce.
circuits,
circuits,
remain In this
this paper,
comprehensive
comprehensive
scarce. In paper, the
model
model
the
factors
based preventing
control
based control
factors and more
and more
preventing widespread
optimisation
optimisation
widespread implementation
implementations
implementations
implementation on of model-based
industrial
onofindustrial
model-based control
circuits
circuits and
remain
remain
control optimisation
scarce.
and scarce. In applications
this paper,
In thisapplications
optimisation the
paper, the
are investigated
investigated
factors
factors
are preventing
preventing by more
by focussing
more on three
widespread
widespread
focussing on threeimplementation
aspects. Firstly,
Firstly,of
implementation
aspects. of themodel-based
the critical variables
model-based
critical variables required
control
control and
required in aa simplified
simplified
and optimisation
optimisation
in flotation
applications
applications
flotation
model
are are identified.
investigated
are investigated
model by Models
focussing
by focussing
are identified. Modelson that
on are
three
three
that currently
aspects.
are aspects. used
Firstly,
currentlyFirstly,
used in inthe control,
critical
thecontrol, optimisation
variables
criticaloptimisation and
required
variables required supervisory
in a applications
simplified
in a simplified
and supervisory flotation
flotation
applications
are thereafter
model
model
are thereafter
are analysed
are identified.
identified.
analysed to determine
Models
Models
to determine
that
that are to
are to what extent
currently
currently
what extent
used
used in thecontrol,
in
the required
control,
required variables are
optimisation
optimisation
variables are
and modelled.
supervisory
andmodelled.
supervisory Finally, online
applications
applications
Finally, online
instrumentation
are thereafter available
analysed
are thereafter analysed
instrumentation to to
to to
available support
determine
determine these
support theseto models
what
to what extent
extent
models are the
arethe investigated,
required also including
variables
required variables
investigated, are instrumentation
modelled.
are modelled.
also including Finally, that is
online
Finally, that
instrumentation online
is
still under
instrumentation
instrumentation
still development and
available
available
under development andto not
to not commonly
support
support
commonlythese available
models
these models arein commercial
investigated,
availableareininvestigated, applications.
also including
also including
commercial applications. Although models
instrumentation
instrumentation
Although used
models usedthat in
is
that in
is
control
still
still
control applications
under
under development
development
applications tendand
tend to focus
and
to focus
not on subsections
subsections
not commonly
commonly
on of the
available
available
of thein flotation
commercial
inflotation
commercialprocess, there seem
seem
applications.
applications.
process, there to be
be aa good
Although
Although
to good agreement
models
models used
used in
agreement in
betweenapplications
control
control
between the required
requiredtend
applications
the andto
tend
and modelled
focus
focus on
tomodelled variables. Model
on subsections
subsections
variables. Model
of
of the fitting however
the flotation
flotation
fitting however
process, often
there
process,often relies
seem
seem to
there relies onbe
to
on extensive
be aa good sampling
good agreement
extensive agreement
sampling
campaigns
between the
between thethat
campaigns that will
required
required need
and to be
modelled
and modelled
will need repeated
variables.
variables.
to be repeated regularly
Model to
Modeltofitting
regularly maintain
fitting however
however
maintain model
model accuracy.
often
often relies onA number
extensive
relies onAextensive
accuracy. number of of online
sampling
sampling
online
measurements
campaigns of
that
campaigns thatofwill
measurements sufficient
will need accuracy
to be
need accuracy
sufficient to be repeatedare
repeatedstill not available
regularly
are stillregularly
not available to to support
maintain
to maintain
to support these
model
model models,
accuracy.
these accuracy. compromising
A number the
of
A number oftheonline
models, compromising long
online
long
term reliable
measurements
measurements
term use
of of
of of
reliable use models
sufficient
sufficient in online
accuracy
modelsaccuracy
in onlineare applications.
are still not The
available
still not available
applications. The fact fact
to that flotation
support
to that
support these processes
models,
theseprocesses
flotation are in many
compromising
models, compromising instances
the long
the long
are in many instances
not extensively
term
term
not extensively
reliable
reliable use instrumented,
of
of models
models in
useinstrumented, constrains
online
inconstrains online maintenance
online applications. maintenance
applications.
online The
The fact and adaption
fact that
that
and adaption
flotation
flotation ofof model based
processes
processes
model based
are
are in solutions
many further.
many instances
insolutions instances
further.
not extensively instrumented, constrains online maintenance and adaption of model based solutions further.
© 2017,
not IFAC (International
extensively
Keywords: instrumented,Federation
constrains ofonline
Automatic Control) and
maintenance Hosting by Elsevier
adaption Ltd.based
of model All rights reserved.
solutions further.
Keywords: Flotation
Flotation modelling
modelling andand identification,
identification, process
process control
control andand optimisation,
optimisation, model
model based
based control
control
Keywords: Flotation
Keywords: Flotation modelling
modelling andand identification,
identification, process
process control
control andand optimisation,
optimisation, model
model based
based control
control


 empirical models fitted over a limited operating range has
1. empirical models fitted over a limited operating range has
1. INTRODUCTION
INTRODUCTION similar
empiricallimitations,
models
models fittedand require
over frequent
over aafrequent
limited recalibration
operating
operating rangewhen has the
1. INTRODUCTION empirical
similar limitations, fitted
and require limitedrecalibration when has
range the
The origins
origins ofof froth 1. INTRODUCTION
froth flotation
flotation can be
be traced
traced back
back toto the
the 1880s
1880s operating
similar points
limitations, shift
and (Bouchard,
require et
frequental., 2009). In
recalibrationthe absence
when the
The can similar limitations,
operating points shift and(Bouchard,
require frequent et al., recalibration
2009). In thewhen absencethe
(Gaudin,
The origins1957)
of while
froth column
flotation flotation
can be traced was patented
back to the in the
1880s of online
operating
operating measurements
points shift to
points shift (Bouchard, recalibrate
(Bouchard, et these
al.,
et al., models,
2009).
2009). In or
the at least
absence
In theorabsence
The origins of froth flotation can be
(Gaudin, 1957) while column flotation was patented intraced back to the 1880s
the of online measurements to recalibrate these models, at least
alert
of the operator
of online measurements of the need
to for recalibration,
to recalibrate these these or
these models, strategies
early
(Gaudin,
(Gaudin,
early 1960s
1960s (Finch
1957)
1957)
(Finch &
while
while Dobby,
column 1990).
column1990).
& Dobby, Froth
flotation
flotation
Froth flotation
was
was patented
patented
flotation hasin been
the
hasinbeen
the online
alert measurements
the operator of the need recalibrate
for recalibration, these or
models, at
at least
least
strategies
modelled
early 1960sextensively
(Finch & to include
Dobby, factors
1990). Frothsuch as the
flotation chemical
has been are bound
alert the
alertbound to
the operatorfail.
operator of the need for recalibration, these strategies
early 1960s
modelled (Finch & to
extensively Dobby,
include 1990). Froth
factors flotation
such has been
as the chemical are to fail. of the need for recalibration, these strategies
reactions
modelled (King, 1982),
extensively to how different
include factors size classes
such as participate are bound to fail.
modelled (King,
reactions extensively
1982), to how
include factorssize
different such as the
classes chemical
theparticipate
chemical are
Thebound
aim oftothisfail.paper is hence to identify the key variables
in the
reactions process
(King, (Lynch,
1982), et al.,
how 1981) and
different the
size physical
classes processes
participate The aim of this paper is hence to identify the key variables
reactions
in (King,
the process 1982),ethow
(Lynch, different
al., 1981) andsize classes participate
the physical processes required
The in flotation control applications, determine ifvariables
existing
such
in as particle-bubble collision (Finch & Dobby, 1990). The aim
aim in
required of this
this paper
offlotationpaper is
is hence
control hence to
to identify
identify
applications, the
the key
determinekeyif variables
existing
in the
such process
the as
process (Lynch,
(Lynch, et
particle-bubble et al.,
al., 1981)
1981) and
collision and
(Finchthe
the physical
physical
& Dobby, processes
processes
1990). models
required take
in these
flotation variables
control into account,
applications, and consider
determine if to what
Collaboration
such as between
particle-bubble industry
collision and
(Finchacademia
& Dobby, as part
1990).of requiredtake
models in flotation control into
these variables applications,
account, and determine
considerif existing
existing
to what
such as particle-bubble
Collaboration collision and
between industry (Finch & Dobby,
academia 1990).
as part of extent
models online
take measurements
these variables intorequired
account, byandthese models
consider to are
programmes
Collaboration such
between as AMIRA
industry P9
and focussed
academia efforts,
as part and
of models online
extent take these variables intorequired
measurements account,by andthese
consider
modelsto what
what
are
Collaboration such
programmes between industry P9
as AMIRA and focussed
academiaefforts,as partand of commonly
extent online available on
measurements froth flotation
required byplants,
these or are
models being
are
significant
programmes progress
such has
as been
AMIRA madeP9 during the pastefforts,
focussed 25 yearsand in extent online
commonly measurements
available on frothrequiredflotation by theseormodels
plants, are being are
programmes such as AMIRA P9 focussed
significant progress has been made during the past 25 years in efforts, and developed
commonly and can
available potentially
on froth be made
flotation available
plants, on
or industrial
are being
understanding
significant the
progress intricacies
has been of the
made froth
during flotation
the past process.
25 years As
in commonly and
developed available on froth be
can potentially flotation plants, or
made available on are being
industrial
significant progress
understanding has been of
the intricacies made duringflotation
the froth the pastprocess.
25 yearsAs in sites in future.
developed and The focus be is made
on long termon industrial
aunderstanding
result, manythe
understanding of the principles
intricacies of thedescribed
froth earlier
flotation have
process. been
As developed
sites and can
in future. canThe potentially
focus be
potentially is made
on long available
termon
available industrial
a result, manythe of intricacies
the principles of thedescribed
froth flotation
earlierprocess.
have been As implementations
sites in future.
sites in future. rather rather
The than
The thanfocus short
focusshort is term
on
is term pilot
long
on pilot plant
term
long plant campaigns.
industrial
termcampaigns.
industrial
aintegrated
a result, into of
result, many a comprehensive simulator earlier
(Schwarz, etbeen
al., implementations
integrated into of
many the
the principles
principles described
a comprehensive simulator earlier
described (Schwarz, have
haveet been
al., implementations
implementations rather rather than
than short
short term
term pilot
pilot plant
plant campaigns.
campaigns.
2006).
integrated
integrated into a comprehensive simulator (Schwarz, et al.,
into a comprehensive simulator (Schwarz, et al.,
2006). 2. MODEL REQUIREMENTS
2006).
2006). 2. MODEL REQUIREMENTS
Despite the rich modelling framework available, the number 2.
2. MODEL
MODEL REQUIREMENTS
REQUIREMENTS
Despite the rich modelling framework available, the number
of successful
Despite the industrial implementations of model based 2.1 Structure
ofDespite the rich
successful modelling
richindustrial framework
framework available,
modellingimplementations available, the
the number
of model number
based 2.1 Structure
control
of and
successful optimization
industrial strategies
implementations(other than
of basic
model level 2.1
2.1 Structure
of successful
control industrial implementations
and optimization strategies (other of basic based
thanmodel based
level Structure
control as
anddescribed
optimizationby Schubert et al.(other
strategies (1995)than) remain
basic scarce
level The motivation to develop many of the existing flotation
control as and optimization strategies (other than
described by Schubert et al. (1995) ) remain scarce basic level The motivation to develop many of the existing flotation
(Shean
control & described
as Cilliers, 2011).by Reasons
Schubert et for (1995)
al. this include
) remain a lack
scarceof models,
The is to model
The motivation
motivation to
to developdifferent
develop many circuit
manycircuitof
of the configurations
existing
existing flotation
theconfigurations and
flotation
control as described by Schubert et al. (1995)
(Shean & Cilliers, 2011). Reasons for this include a lack of ) remain scarce models, is to model different and
instrumentation,
(Shean & Cilliers, lack
2011). of reliable
Reasons fordynamic
this models
include a lack and
of operating
models, practices
is to
models, ispractices model
to model with high accuracy,
different circuit in order to recommend
configurations and
(Shean & Cilliers,lack
instrumentation, 2011). of Reasons
reliablefordynamic
this includemodels a lackandof operating with different
high accuracy, circuit configurations
in order to recommend and
inadequate regulatory
instrumentation, lack control
of (Bergh
reliable & Yianatos,
dynamic 2011).
models The
and changes
operating that
operatingthat would
practices
practices improve
with high operation.
accuracy, in Examples
order to of
recommend such
instrumentation,
inadequate regulatorylackcontrol
of reliable
(Bergh & dynamic
Yianatos,models
2011). The and changes wouldwith high accuracy,
improve operation. in order to recommend
Examples of such
issues of theregulatory
inadequate lack of suitable dynamic models and insufficient activities
changes
changes that are described
would
would improve by Schwarz et al. (2006).
operation. Examples Under of these
such
of theregulatory
inadequate
issues control (Bergh
(Bergh &
controldynamic
lack of suitable & Yianatos,
Yianatos,
models 2011).
2011). The
and insufficient The activities that
are described improve
by Schwarz operation. Examples
et al. (2006). Under of these
such
instrumentation
issues of the lack toof interface
suitable with
dynamic these
modelsmodels
and cannot be
insufficient circumstances
activities are a
activities are described detailed
described by sampling
Schwarz
by Schwarz campaign
et al. (2006). can often
Under be
these
issues of the lacktoof interface
instrumentation suitable dynamic
with thesemodels and insufficient
models cannot be circumstances a detailed sampling etcampaignal. (2006).can Under
often these
be
separated and need
instrumentation to tointerface
be addressedwith simultaneously.
these justified to provide
circumstances a the datasampling
detailed to fit a comprehensive
campaign set of
can model
often be
instrumentation
separated and need to to interface
be addressed these models
models cannot
with simultaneously. cannot be be circumstances a detailed sampling campaign
justified to provide the data to fit a comprehensive set of model can often be
separated and need to be addressed simultaneously. parameters.
justified
justified to
to Wherethe
provide
provide models
the data
data to
to are
fit
fit a
a to be used online,
comprehensive
comprehensive set
set of
ofmanual
model
model
separated and need to be addressed simultaneously.
Model based implementations often fall into disuse after some parameters. Where models are to be used online, manual
sampling
parameters. campaigns
Where
Where modelsare not are viable on an ongoing basis, and
Model based implementations often fall into disuse after some
time
Model due to the
based models not being
implementations robust
often fall overdisuse
a large range of parameters.
sampling campaigns are not are
models viable to
to onbe
be an used
used online,
online,
ongoing manual
manual
basis, and
Model
time based
due to theimplementations
models not being often fall into
robust overdisuse
into a large after
after some
some
range of online
sampling
sampling
online
measurements
campaigns
campaigns are
measurements
would
are
would not have
viable
nothave
viable to be
on used
an
onused
to be
to
an ongoingmaintain
ongoing basis,
basis,
to maintain
model
and
and
model
operating
time due toconditions
the models (Shean
not &
being Cilliers,
robust 2011).
over a The
large use
range
time due toconditions
operating the models(Shean not being robust over
& Cilliers, a large
2011). Therangeuse ofof integrity.
online A model
measurements
online measurements
integrity.
for
A model forwould
continuous
would have tocontrol
be
have tocontrol
continuous used applications
to maintain
be usedapplications would
model
to maintain would
model
operating
operating conditions (Shean & Cilliers, 2011). The use
conditions (Shean & Cilliers, 2011). The use of integrity.
of integrity. A A model
model for for continuous
continuous control control applications
applications would
would
2405-8963 ©
Copyright © 2017, IFAC (International Federation of Automatic Control)
2017 IFAC 19 Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©under
2017 responsibility
IFAC 19 Control.
of International Federation of Automatic
Copyright
Copyright © 2017 IFAC
© 2017 IFAC
10.1016/j.ifacol.2017.12.004 19
19
CCA 2017
Johannesburg, South Africa
December 7-8, 2017
20 D.J. Oosthuizen et al. / IFAC PapersOnLine 50-2 (2017) 19–24
thus need to be structured in such a way that available online
measurements provide the stimuli to the process and are also
used to maintain model integrity by estimating parameters
where possible.
Basic froth flotation models are typically extended with the
aim of characterising some behaviour that cannot be explained
by existing models. While more detailed modelling is essential
in improving model accuracy and in advancing knowledge
about the process, it does result in a significantly larger set of
parameters to be fitted initially and updated regularly to
maintain model consistency. For control purposes, some
accuracy may be sacrificed in exchange for fewer parameters.
The direction of change and relative magnitude is generally of
more importance than the absolute value of a variable, as
measurement feedback can correct for model inaccuracies.
While the decision on which interactions to ignore are not
trivial, a reliable model would have to be based on a
significantly reduced parameter set, to ensure that model
accuracy is not degraded by the use of estimates based on
parameters that cannot be updated dynamically.
Despite the requirement of minimising the set of parameters,
the model needs to be able to estimate process dynamics
required for control with sufficient accuracy. It must also be
able to model non-linear phenomena such as peak air recovery
(Hadler & Cilliers, 2009) and discrepancies between mass-pull
and recovery (Hadler, et al., 2010) that currently receive
research interest, with sufficient accuracy.
While laboratory and pilot-plant scale applications show some
benefit of deriving empirical models (Bouchard et al., 2009),
the long-term reliability of these models in the presence of
changing operating conditions is a concern.
Phenomenological models should thus take priority. Bouchard
et al. (2009) however also commented that empirical models
should not be dismissed completely, as all models as well as
sensors require calibration.
2.2 Variables
A number of authors listed the key variables required in the
control of froth flotation processes, for example Finch &
Dobby (1990), Lynch et al. (1981), Bergh & Yianatos (2011)
and Laurila et al. (2002). There is to a large extent agreement
on the set of variables required, and a summary follows:
As inputs, or manipulated variables, the following variables
can be used to drive the process in a desired direction:
- Reagent additions
- Pulp level setpoints
- Air flow rate setpoints
- Froth wash water rate (particularly in columns)
Lynch et al. (1981) also included addition points for reagents
and collection points for concentrate, but these form part of
circuit design parameters rather than online control
parameters.
Depending on the circuit configuration, feed characteristics are
either considered as disturbances or as manipulated variables,
20
as indicated by Lynch et al. (1981). Bergh & Yianatos (2011)
considered slurry flow as a manipulated variable rather than a
disturbance, which would also be the case in integrated
grinding and flotation control and optimisation applications
(Conradie et al., 2003). The following feed properties can be
classified as manipulated variables or disturbances:
- Pulp density
- Volumetric flowrate
- Fineness of grind
The main outputs of the process relate to its economic
performance, and are grade and recovery. Lynch, et al. (1981)
also included concentrate density and flowrate in the outputs
referred to as “performance variables”, as the total production
also affects profitability.
A number of process states has a direct influence on the
economic outputs of the process, and are typically affected by
the manipulated variables and disturbances. The following
states, also referred to as “intermediate variables” by Lynch, et
al. (1981) are listed:
- Froth depth
- Gas holdup
- Bias superficial velocity (mostly columns)
- Air superficial velocity
- Feed, tailings and concentrate flow rates
- Mineral concentrations in all intermediate streams (grades)
- Densities of all streams
The largest discrepancy between the variables described by the
authors, is in the variables considered disturbances. Bergh &
Yianatos (2011) only included the first 3 disturbances listed
below, while the full list is described by Laurila et al. (2002).
- Feed size distribution
- Feed grade (minerals concentration in feed)
- Feed density
- Feed mineralogy (fineness of crystallisation, minerals)
- Electrochemical potentials (Eh, pH)
- Particle properties (size distribution, shape, degree of
liberation)
- Froth properties (speed, bubble size distribution, stability)
The fact that Bergh & Yianatos (2011) managed to explain
92% of variance with a reduced parameter set model, using
only 6 latent variables obtained through principle component
analysis, indicates that a model with a small parameter set may
still provide sufficient accuracy for control purposes, but the
complexity required is likely to be process dependent (Laurila
et al., 2002). Shean & Cilliers (2011) confirmed that all these
variables are not necessarily required to obtain good control
performance, but that their impact needs to be considered.
Simplifying assumptions, for example that the feed
distribution and density would not vary significantly if the
grinding circuit control is effective have been proposed (Wills
& Napier-Munn, 2006) and can potentially be used to simplify
models without degrading controller performance.
 CCA 2017
Johannesburg, South Africa
December 7-8, 2017
D.J. Oosthuizen et al. / IFAC PapersOnLine 50-2 (2017) 19–24 21
3. MODELS USED FOR CONTROL
The flotation process involves complex physical and chemical
reactions with various contributing factors. Generally,
flotation modelling approaches can be divided into two
categories: Kinetic (first-principle) modelling and data-driven
modelling. The kinetic modelling studies mainly include:
- Flotation rate modelling: The influence of operational
parameters, such as pulp density, chemical reagents, aeration
rate and, froth depth, on the flotation rate;
- Mass balance modelling: Mass and flow conservation in a
flotation cell;
- Probability modelling: The probability of collision /
adhesion / detachment between particles and bubbles, and
the merging / bursting of bubbles.
Data driven modelling mainly involve:
- Performance evaluation: The relationship between flotation
performance and froth features;
- Grade / recovery prediction: Predict the concentrate grade /
recovery using inlet conditions and operational variables;
- Soft sensing: Estimate key process variables, e.g. pH, pulp
level, when the default instrument is not available or out of
order.
Bascur (2005) developed a detailed phenomenological
flotation model. The model linked together the particle /
bubble and water transport mechanisms, as well as the
hydrodynamic characteristic of a flotation cell, and is able to
describe the behaviour of particles with different
mineralogical composition and particle sizes under a wide
range of steady state and dynamic operating conditions. This
model provides detailed understanding of flotation operations
at the expense of increasing complexity.
The simplest models used in froth flotation control have been
derived to stabilise pulp levels in interacting cells in the
presence of variations in feed rate. Despite their simplicity,
flotation level control systems, such as Mintek's FloatStar
(Schubert et al., 1995) can significantly improve the economic
performance of a flotation plant (Craig and Henning, 2000;
Craig and Koch, 2003).
In Jämsä-Jounela (1992), a simplified flotation model based on
both mass and volume balances was developed for the control
of a rougher flotation bank. The model parameters were
obtained through an industrial experimental campaign.
The level (yi) response of a cell can be calculated from the
difference between the slurry inflow from an upstream (Q i-1)
cell and the slurry outflow from the cell (Qi), as shown in (1),
assuming a constant cross section area (Ai) over the cell height.
𝜕𝜕𝑦𝑦𝑖𝑖 𝑄𝑄𝑖𝑖−1 −𝑄𝑄𝑖𝑖
= (1)
𝜕𝜕𝜕𝜕 𝐴𝐴𝑖𝑖
Jämsä-Jounela et al. (2003) improved on this model by taking
the structure of flotation cells and valve sizing into account.
Various control strategies for pulp level control, including PI
control, feed-forward control, decoupling control and
multivariable control, have been tested and compared using
this model (Kämpjärvi & Jämsä-Jounela, 2003).
21
When the froth phase is considered in addition to the slurry
phase as part of a model, a mass balance forms the core of the
model to track mass flows between the feed, tailings and
concentrate streams. Depending on the complexity of the
model and data available for fitting, the mass balance may be
performed on specific components in a stream, for example
PGMs, chromite and gangue (Du Preez et al. 2013) or further
divided into floatability classes per component based on
granulometry (typically size classes) (Putz & Cipriano, 2015).
Maldonado, et al. (2009) implemented a Model Predictive
Control (MPC) strategy to control a flotation column pilot
plant. Gas holdup in the collection zone was measured using a
conductivity probe, while bias rate was calculated from a
difference in conductivities between the wash-water and feed
water streams in a two-phase system. A 2x2 model was
implemented in an MPC controller, with the outputs defined
as gas holdup and bias rate, and the inputs as wash water feed
rate and aeration rate. The models were however empirical
models fitted from operational data. Upper and lower limits
for both gas holdup and bias rate were included in the
controller, based on a desired operating range and expected
constraints that would ensure such operation.
Bergh & Yianatos (2013) developed a simulator for rougher
flotation banks and calibrated it using experimental data. The
simulator was initially used in parallel with the existing control
system to evaluate the effect of changes in operating
conditions. It was later redesigned to act as an expert system,
but this functionality has only been tested in a simulated
environment. Other industrial implementations of expert
systems (Kewe, et al., 2014; Kewe, Moffat and Schaffer, 2014)
and optimisation strategies (Baas, et al., 2007) are often rule
based, with limited detail provided on the underlying models.
Bergh & Yianatos (2013) modelled mass transfer between the
pulp and froth phase by defining a global cell recovery (RG)
based on collection zone recovery (RC) and froth recovery
(RF), as defined in (2). Yianatos et al. (2008) used a similar
approach. Although (2) can be simplified by lumping the
froth- and collection zone recoveries together, Du Preez et al.
(2013) noted that a more detailed model is likely to improve
their results that was based on a single parameter model.
𝑅𝑅 ∙𝑅𝑅
𝑅𝑅𝐺𝐺 = 1−𝑅𝑅 𝐶𝐶(1−𝑅𝑅
𝐹𝐹
(2)
𝐶𝐶 𝐹𝐹 )
The complexity of the equations describing collection zone
and froth recoveries shows some variation between models.
Bergh & Yianatos (2013) modelled RC as a function of pulp
residence time, flotation rate constants and maximum
achievable recovery. RF was modelled using two equations.
The first is a function of bubble load, superficial gas rate,
concentrate mass flow, cell dimensions and concentrate- and
bubble load grade. The other is a function of a froth stability
parameter, froth depth, gas holdup, superficial gas velocity and
maximum froth recovery. For both RC and RF empirical
equations were fitted to model the degradation in flotation rate
distribution and the reduction of froth stability when moving
down a bank of flotation cells. Gangue recovery is modelled
based on water recovery (a function of superficial gas rate,
froth depth and froth stability). The model fit requires an
CCA 2017
Johannesburg, South Africa
December 7-8, 2017
22 D.J. Oosthuizen et al. / IFAC PapersOnLine 50-2 (2017) 19–24
extensive sampling campaign, including grade and mass data
per size class on all streams, density (per stream), and also air
holdup, bubble loading and grade, and aeration rates for each
cell.
The compartment model described by Savassi (2005) include
similar elements as described by Bergh & Yianatos (2013), but
all the factors contributing to overall recovery is included in a
single equation. In (3), collection zone recovery is described
by kcz (flotation rate for the collection zone) and τcz (residence
time in the collection zone). Similar to (2), RF refers to froth
recovery while ENT and Rw refers to degree of entrainment and
water recovery to the concentrate stream, respectively. Dos
Santos et al. (2014) showed how phenomenological models
can be used as part of (3) to model entrainment, water recovery
and froth recovery.
𝑘𝑘 ∙𝜏𝜏𝑐𝑐𝑐𝑐 ∙𝑅𝑅𝐹𝐹 ∙(1−𝑅𝑅𝑤𝑤 )+𝐸𝐸𝐸𝐸𝐸𝐸∙𝑅𝑅𝑤𝑤
𝑅𝑅𝐺𝐺 = (1+𝑘𝑘𝑐𝑐𝑐𝑐
𝑐𝑐𝑐𝑐 ∙𝜏𝜏𝑐𝑐𝑐𝑐 ∙𝑅𝑅𝐹𝐹 )∙(1−𝑅𝑅𝑤𝑤 )+𝐸𝐸𝐸𝐸𝐸𝐸∙𝑅𝑅𝑤𝑤
Putz & Cipriano (2015) used a hybrid Model Predictive
Control (HMPC) strategy on a simulator to control the final
tailings grade of a flotation circuit subject to level constraints.
The hybrid functionality was used to include scenarios where
pulp overflows and when froth flow is zero, in addition to
normal operation. The core of the model is a mass balance
performed on a number of defined granulometries (size
classes) for each species considered. Pulp levels in cells are
calculated using a similar approach as described in (1),
including valve dynamics and the effect of relative heights of
interacting cells. Collection rates (per granulometry class)
defines mass transfer between the pulp and froth phase, while
a drainage rates determines mass transfer between the froth
and pulp phase. Although the model includes parameters such
as air hold-up (in the level calculation) and reagent addition,
no attempt was made to manipulate aeration rate or reagent
addition for control purposes.
Operating flotation cells at their peak air recoveries have been
shown to maximise both grade and recovery (Smith, et al.,
2010). A theoretical model has been developed to calculate air
recovery, based on froth film characteristics (Neethling &
Cilliers, 2008). A controller implemented to operate a pilot
scale flotation cell at its peak air recovery point however used
a peak-seeking strategy without including any model (Shean,
et al., 2017).
Maldonado, et al. (2007) used phenomological models in a
dynamic programming application, with the aim of optimising
the froth level profile for a bank of cells. Seguel, et al. (2015)
used the same model, but with a different cost function
(maximising overall Cu recovery compared to minimising the
sum of squared Cu grades in tailings flows). In both cases the
model used a single flotation rate constant per species. The
flotation rate constant is modelled as a function of froth depth,
residence time and slurry grade, and was fitted using industrial
data. Concentrate flow is calculated as a function of froth
depth, and the rest of the model is based on mass balances.
Although the model was not used in an online control
application, it could potentially be used as a simulator in
parallel to the plant.
While this section covers a small portion of the available
flotation models used in control, it provides an overview of the
modelling techniques commonly employed and highlights the
scarcity of model based controllers in flotation circuits. The
MPC strategy described above was only implemented on a
pilot plant, and the HMPC strategy on a simulator, highlighting
the scarcity of model based controllers in industrial
applications. The key variables listed in 2.2 are mostly
modelled, with the exception being the large set of disturbance
variables.
4. INSTRUMENTATION
Reliable online measurements are essential in ensuring reliable
long-term use of model based control strategies (Hodouin,
2011). Wills & Napier-Munn (2006) stated that the key to
effective flotation control is online chemical analysis. In all the
(3) flotation models described above, the mineral compositions of
process streams form a core part of the models, supporting this
statement. On-line X-ray fluorescence (XRF) analysers can
provide assays on several elements as well as solids content,
but the sampling delay varies between 15 seconds and one
minute, and, depending on the number of samples analysed,
cycle time can vary between 5 and 15 minutes (Laurila, et al.,
2002). Visual and near-infrared reflectance spectroscopic
analysis can complement XRF devices and provide grade
analyses at a much higher frequency (Shean & Cilliers, 2011).
Remes et al. (2005) developed a dynamic flotation model to
study the influence of measurement accuracy and sampling
frequency of online XRF analysers on the economical
performance of the flotation process. It was shown that, in
order to reduce the error caused by the measurement delay, fast
basic measurement and control is necessary to complement
process analysers, and to keep the process stable until the next
assay arrives.
Due to the close relationship between visual froth surface
features and flotation performance, a lot of research has been
done on the use of flotation cameras, as a soft sensor for
grades, to provide measurements at a faster rate than what can
be provided by XRF analysers. Significant progress has been
made in understanding froth behaviour, and to quantify the
impact of manipulated and disturbance variables on froth
characteristics, for example the effect of reagent dosage on
bubble size (Zhu, et al., 2014). He et al. (2013) utilized a
probability density function (PDF) of the froth colour texture
unit number to characterise froths based on colour and texture.
A nonparametric estimation method based on the fixed normal
kernel basis was proposed to describe this distribution. Xu et
al. (2015) proposed a complex network-based texture
extraction and classification method for froth imaging to
extract the distinctive froth texture features in different
production states.
Liu & MacGregor (2008) developed a control strategy to
achieve desired froth image properties that are related to froth
stability, by manipulating reagent addition rates. In Zhu, et al.
(2016), a B-spline estimator is used to describe the bubble size
PDF, in order to classify bubble sizes with non-Gaussian
features. A multi-output least square support vector machine
(MLS-SVM) is then applied to establish a dynamical
22
 CCA 2017
Johannesburg, South Africa
December 7-8, 2017 D.J. Oosthuizen et al. / IFAC PapersOnLine 50-2 (2017) 19–24 23
relationship between the weights of the B-spline estimator and
the reagent dosage. Based on this structure, a reagent addition
control strategy was implemented to track a desired bubble
size PDF.
Aldrich, et al. (2010) however concluded that, despite several
advances in machine vision on flotation froths, conflicting
results were obtained on linking image features to froth grade,
and that no long term fully automated control system based on
machine vision have been developed to date.
There may however be scope in using froth image properties
in combination with a flotation model to calculate grade, rather
than searching for a direct link. Other soft-sensing applications
also exist using froth images. Xu et al. (2016) proposed a
multi-model soft measurement method to estimate the froth
layer thickness based on the visual features. The froth layer
thickness was established by the kernel extreme learning
machine (KELM) models under different working conditions.
An experimental device to measure froth recovery is described
by Rahman, et al. (2013). Although still in an early stage of
development, it could potentially be used in future to measure
the froth recovery, which is used in several models.
Yianatos, et al. (2008) describes a manual gas-holdup sensor
and bubble-load sensor. Both however require manual
interaction and cannot be used as online measurements. The
bubble load sensor described by Moys, et al. (2010) shows
good accuracy, but also requires manual interaction. A gas
holdup sensor described by Vinnett, et al. (2016) uses a
combination of images and superficial gas velocity, and has
the potential to provide an online gas holdup measurement.
There is also ongoing research on bubble size measurement, as
detailed in Bhondayi & Moys (2014).
Laurila, et al. (2002) describes several other measurements
that may be available on a flotation plant, including their
limitations.
- Slurry flow measurements are typically performed using
magnetic flow meters, but these are generally considered
problematic due to solid particles and suspended air bubbles
causing inaccuracies.
- If other techniques are not viable, concentrate flow rates in
open channels with known dimensions can be calculated
using ultrasonic level transmitters, but are not accurate.
- Density measurements may be provided by some XRF
devices, or alternatively, nuclear density meters may be
used. Installation is critical in preventing inaccuracies
caused by bubbles.
- Slurry level measurements using a float with a target plate
and ultrasonic level transmitter are commonly used, while
measurements based on hydrostatic pressure or direct
ultrasonic measurements are often troublesome or sensitive
to variations in density.
- A number of airflow measurement techniques are available
and are generally considered accurate.
- pH measurements are often problematic due to
contamination of the electrode. In some cases, conductivity
probes can however be used as substitute.
23
The fact that measurement devices are available however does
not imply that every process stream would be instrumented.
Flow rates and on-stream analyses would typically only be
available on critical streams, and pH measurements only in the
conditioning tank (Laurila, et al., 2002).
5. CONCLUSIONS
Several good dynamic froth flotation models have been
developed that can potentially be used in control and
optimisation applications. The long-term reliability of these
models however depends on the availability of accurate, fast
and reliable online measurements, which still seem to be
lagging.
ACKNOWLEDGEMENTS
This work is based on research supported in part by the
National Research Foundation of South Africa (IRC grant
number 103666).
REFERENCES
Aldrich, C., Marais, C., Shean, B. J. & Cilliers, J. J., 2010.
Online monitoring and control of froth flotation systems
with machine vision: A review. International Journal of
Mineral Processing, Volume 96, pp. 1-13.
Baas, D., Hille, S. & Karageorgos, J., 2007. Improved
Flotation Process Control at Newcrest's Telfer Operation.
Ninth AUSIMM Mill Operators’ Conference, pp. 87-93.
Bascur, O., 2005. Example of a Dynamic Flotation
framework. Centenary of Flotation Symposium, pp. 85-91.
Bergh, L. G. & Yianatos, J. B., 2011. The long way toward
multivariate predictive control of flotation processes.
Journal of Process Control, Volume 21, pp. 226-234.
Bergh, L. & Yianatos, J., 2013. Control of rougher flotation
circuits aided by industrial simulator. Journal of Process
Control, Volume 23, pp. 140-147.
Bhondayi, C. & Moys, M. H., 2014. Measurement of a proxy
for froth phase bubble sizes as a function of froth depth in
flotation machines Part 1. Theoretical development and
testing of a new technique. International Journal of Mineral
Processing, Volume 130, pp. 8-19.
Bouchard, J., Desbiens, A., Villar, R. & del Nunez, E., 2009.
Column flotation simulation and control: An overview.
Minerals Engineering, Volume 22, pp. 519-529.
Conradie, A. v. E., Bascur, O., Aldrich, C. & Nieuwoudt, I.,
2003. Integrated comminution and flotation neurocontrol
using evolutionary reinforcement learning. Application of
Computers and Operations Research in the Minerals
Industries, SAIMM, pp. 209-216.
Craig, I. K. & Henning, R. G. D., 2000. Evaluation of
advanced industrial control projects: a framework for
determining economic benefits. Control Engineering
Practice, 8(7), pp. 769-780.
Craig, I. K. & Koch, I., 2003. Experimental design for the
economic performance evaluation of industrial controllers.
Control Engineering Practice, 11(1), pp. 57-66.
Dos Santos, N. A., Savassi, O., Peres, A. E. C. & Martins, A.
H., 2014. Modelling flotation with a flexible approach-
Integrating different models to the compartment model.
Minerals Engineering, Volume 66-68, pp. 68-76.
CCA 2017
Johannesburg, South Africa
December 7-8, 2017
24 D.J. Oosthuizen et al. / IFAC PapersOnLine 50-2 (2017) 19–24
Du Preez, N. B., Crundwell, F. K. & Knights, B. D. H., 2013.
Flotation of PGM-containing minerals: Plant-wide
regression and prediction of circuit performance. Minerals
Engineering, Volume 54, pp. 116-123.
Finch, J. A. & Dobby, G. S., 1990. Column Flotation.
Oxford: Pergamon Press.
Gaudin, A. M., 1957. Flotation. Second ed. New York:
McGraw-Hill.
Hadler, K. & Cilliers, J. J., 2009. The relationship between
the peak in air recovery and flotation bank performance.
Minerals Engineering, pp. 451-455.
Hadler, K., Smith, C. D. & J, C. J., 2010. Recovery vs. mass
pull: The link to air recovery. Minerals Engineering,
Volume 23, p. 994–1002.
He, M., Yang, C., Wang, X., Gui, W. & Wei, L., 2013.
Nonparametric density estimation of froth colour texture
distribution for monitoring sulphur flotation process.
Minerals Engineering, Volume 53, pp. 203-212.
Hodouin, D., Jämsä-Jounela, S. L., Carvalhoc, M. T. &
Bergh, L., 2001. State of the art and challenges in mineral
processing control. Control Engineering Practice, Volume
9, pp. 995-1005.
Jämsä-Jounela, S.L., 1992. Simulation study of self-tuning
adaptive control for rougher flotation. Powder technology,
Volume 69, pp. 33-46.
Jämsä-Jounela, S. L., Dietrich, M., Halmevaara, K. & Tiili,
O., 2003. Control of pulp levels in flotation cells. Control
Engineering Practice, Volume 11, pp. 73-81.
Kämpjärvi, P. & Jämsä-Jounela, S. L., 2003. Level control
strategies for flotation cells. Minerals Engineering, 16(11),
pp. 1061-1068.
Kewe, T., Moffat, N. & Schaffer, M., 2014. Porgera Flotation
Circuit Upgrade and Expert System Installation. 12th
AusIMM Mill Operators' Conference, pp. 345-355.
King, R. P., 1982. Principles of Flotation. Johannesburg:
SAIMM.
Laurila, H., Karesvuori, J. & Tiili, O., 2002. Strategies for
instrumentation and control of Flotation Circuits.
Vancouver, Society for Mining, Metallurgy, and
Exploration, p. 2174–2195.
Liu, J. J. & MacGregor, J. F., 2008. Froth-based modeling
and control of flotation processes. Minerals Engineering 21
(2008) 642–651, Volume 21, pp. 642-651.
Lynch, A. J., Johnson, N. W., Manlapig, E. V. & Thorne, C.
G., 1981. Mineral and Coal Flotation Circuits - Their
Simulation and Control. Amsterdam: Elsevier Scientific
Publishing Company.
Maldonado, M., Desbiens, A. & del Villar, R., 2009.
Potential use of model predictive control for optimizing the
column flotation process. Int. J. Miner. Process, Volume
93, pp. 26-33.
Maldonado, M., Sbarbaro, D. & Lizama, E., 2007. Optimal
control of a rougher flotation process based. Minerals
Engineering, Volume 20, pp. 221-232.
Moys, M. H., Yianatos, J. & Larenas, J., 2010. Measurement
of particle loading on bubbles in the flotation process.
Minerals Engineering, Volume 23, p. 131–136.
Neethling, S. J. & Cilliers, J. J., 2008. Predicting air recovery
in flotation cells. Minerals Engineering, Volume 21, p.
937–943.
24
Putz, E. & Cipriano, A., 2015. Hybrid model predictive
control for flotation plants. Minerals Engineering, Volume
70, pp. 26-35.
Rahman, R. M., Ata, S. & Jameson, G. J., 2013. Froth
recovery measurements in an industrial flotation cell.
Minerals Engineering, Volume 53, pp. 193-202.
Remes, A., Kongas, M., Saloheimo, K. & Jämsä-Jounela, S.
L., 2005. Economical effects of on-line elemental analysis
performance on flotation control. IFAC Proceedings
Volumes, 28(1), pp. 235-240.
Savassi, O. N., 2005. A compartment model for the mass
transfer. Int. J. Miner. Process. 77 (2005) , Volume 77, p.
65–79.
Schubert, J.H., Henning, R.G.D., Hulbert, D.G. & Craig, I.K.,
1995. Flotation control - a multivariable stabilizer. XIX
IMPC, San Francisco.
Schwarz, S., Alexander, D., Whiten, W.J., Franzidis, J.P. &
Harris, M.C., 2006. JKSimFloat V6: improving flotation
circuit performance and understanding. IMPC, Instanbul,
Turkey.
Seguel, F., Soto, I., Krommenacker, N., Maldonado, M. &
Becerra-Yoma, N., 2015. Optimizing flotation bank
performance through froth depth profiling: Revisited.
Minerals Engineering, Volume 77, pp. 179-184.
Shean, B., Hadler, K. & Cilliers, J. J., 2017. A flotation
control system to optimise performance using peak air
recovery. Chemical Engineering Research and Design,
Volume 117, pp. 57-65.
Shean, B. J. & Cilliers, J. J., 2011. A review of froth flotation
control. International Journal of Mineral Processing,
Volume 100, pp. 57-71.
Smith, C. D., Hadler, K. & Cilliers, J. J., 2010. Flotation bank
air addition and distribution for optimal performance.
Minerals Engineering, Volume 23, pp. 1023-1029.
Vinnett, L., Ledezma, T., Alvarez-Silva, M. & Waters, K.,
2016. Gas holdup estimation in flotation machines using
image techniques and superficial gas velocity. Minerals
Engineering, Volume 96-97, pp. 26-32.
Wills, B. A. & Napier-Munn, T. J., 2006. Wills' Mineral
Processing Technology. 7th ed., Elsevier Science &
Technology Books.
Xu, D. et al., 2015. Complex networks-based texture
extraction and classification method for mineral flotation
froth images. Minerals Engineering, Volume 83, pp. 105-
116.
Xu, D. et al., 2016. Multi-model soft measurement method of
the froth layer thickness based on visual features.
Chemometrics and Intelligent Laboratory Systems, Volume
153, pp. 112-121.
Yianatos, J. B., Moys, M. H., Contreras, F. & Villanueva, A.,
2008. Froth recovery of industrial flotation cells. Minerals
Engineering, Volume 21, pp. 817-825.
Zhu, J. et al., 2016. Combined fuzzy based feedforward and
bubble size distribution based feedback control for reagent
dosage in copper roughing process. Journal of Process
Control, Volume 39, pp. 50-63.
Zhu, J. et al., 2014. Probability density function of bubble
size based reagent dosage predictive control for copper
roughing flotation. Control EngineeringPractice, Volume
29, pp. 1-12.