Minerals Engineering 22 (2009) 1350–1358

Contents lists available at ScienceDirect

Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng

Dynamic modelling of copper solvent extraction mixer–settler units

C.M. Moreno a, J.R. Pérez-Correa a,*, A. Otero b
a
Department of Chemical and Bioprocesses Engineering, Pontificia Universidad Católica de Chile, Casilla 306, Santiago 22, Chile
b
Mining Centre, Pontificia Universidad Católica de Chile, Casilla 306, Santiago 22, Chile

a r t i c l e i n f o a b s t r a c t

Article history: The copper Leaching, Solvent Extraction and Electrowinning circuit (LX–SX–EW) is one of the most effec-
Received 30 January 2009 tive processes for extracting copper from low grade ores. This work focuses on the liquid–liquid extrac-
Accepted 4 September 2009 tion SX sub-process, since many difficult to handle operational problems within LX–SX–EW circuits are
Available online 7 October 2009
related to SX malfunction. Controlling these problems better can reduce operational costs and increase
recovery and copper production. Realistic dynamic simulation is a standard tool nowadays to design
Keywords: and assess more effective control strategies. In this work we present a general dynamic model for SX
Extractive metallurgy
mixer–settlers, and applied it to two different units, one located in an extraction stage and the other
Hydrometallurgy
Simulation
in a stripping stage of a copper plant. The model includes a non-trivial settler hydrodynamics represented
by a set of nonlinear differential equations for both mixer and settler units. The mixer is modelled as a
continuous stirred tank reactor and the settler as a hydrodynamic circuit combining series and parallel
connections of continuous stirred tank and plug flow reactors. The model was calibrated with industrial
plant data, resulting in realistic simulations of outlet copper concentrations. Using the proposed model,
we obtained better fitting than that achieved with simpler settler models that include only a time delay.
The model fitting parameters provide sufficient flexibility to accurately reproduce the dynamics of differ-
ent units in industrial plants.
Ó 2009 Elsevier Ltd. All rights reserved.

1. Introduction in average, it is not surprising that process regulation is poor. As a

result, strong disturbances are unavoidable and many of them may
Due to increasing copper world demand, there is a strong incen- pass unnoticed and uncontrolled for at least 4 h. Hence, it can be
tive to apply different processes to copper extraction from low argued that many of the SX operational problems are related to
grade ores. The copper Leaching, Solvent Extraction and Electro- the lack of effective monitoring and control systems specific for
winning circuit (LX–SX–EW) (see Fig. 1) is one of the most effective this process.
processes for this purpose. In this work, we focus on the liquid–li- Consequently, on-line measurement of copper concentrations
quid extraction SX process, since many of the difficult to handle in the main process streams together with an effective automatic
operational problems within this circuit are related to SX malfunc- control strategy, should stabilize the process, improve cathodes
tion. Critical problems normally found in SX plants, which have a quality and homogeneity, and reduce operational costs. Unfortu-
considerable impact in extraction efficiency and selectivity, are: nately, it is not possible to measure on-line copper concentrations
crud formation, organic and aqueous phase entrainments, and var- with affordable and reliable instrumentation. In addition, it is
iable and unpredictable phase separation times in the settlers expensive and risky to develop and test control strategies directly
(Bergh and Yianatos, 2001). in the process plant. Alternatively, dynamic simulation can be used
For example, uncontrolled organic and aqueous flow-rates may effectively to develop and test estimation and control strategies
cause inefficient phase separation, leading to entrainment and pro- before they are implemented on-line, reducing the risks of
cess instability. These, in turn, reduce cathodes quality, increase catastrophic operational events, and the time and costs of
the consumption of valuable chemicals, contaminate the electro- development.
lyte solutions, cause premature corrosion of expensive anodes Despite that several SX process models have been presented in
and increase organic phase losses. Additionally, copper concentra- the literature, many of them are steady state or either too simple to
tion and pH are usually sampled and regulated manually by oper- represent realistic industrial data or too complex to calibrate.
ators. Since the samples are analyzed in the laboratory, taking 4 h Moreover, there are still many phenomena in SX processes that
are not well understood. For example, Van Bochove et al. (2000)
* Corresponding author. Tel.: +56 3544258; fax: +56 3545803. developed a thermodynamic model that predicts exactly the iso-
E-mail address: [email protected] (J.R. Pérez-Correa). therms of equilibrium. However, this model is very complex and

0892-6875/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mineng.2009.09.003
 C.M. Moreno et al. / Minerals Engineering 22 (2009) 1350–1358 1351

Nomenclature

A extraction equilibrium isotherm parameter (g L1) X aL2 aqueous copper concentration in active zone of second
a1A aqueous active zone fraction of first Cholette–Cloutier Cholette–Cloutier unit (g L1)
unit X bL2 aqueous copper concentration in dead zone of second
a1O organic active zone fraction of first Cholette–Cloutier Cholette–Cloutier unit (g L1)
unit Xm aqueous copper concentration in mixer (g L1)
a2A aqueous active zone fraction of second Cholette–Clou- XP time delayed aqueous copper concentration in slow
tier unit flow branch (g L1)
a2O organic active zone fraction of second Cholette–Cloutier X ST aqueous equilibrium copper concentration in stripping
unit stage (g L1)
B extraction equilibrium isotherm parameter (g L1) XUi aqueous copper concentration in ith CSTR in fast flow
b1A aqueous dead zone fraction of first Cholette–Cloutier branch in settler (g L1)
unit Y* organic equilibrium copper concentration (g L1)
b1O organic dead zone fraction of first Cholette–Cloutier unit Yi organic inlet copper concentration (g L1)
b2A aqueous dead zone fraction of second Cholette–Cloutier Y aL1 organic copper concentration in active zone of first Cho-
unit lette–Cloutier unit (g L1)
b2O organic dead zone fraction of second Cholette–Cloutier Y bL1 organic copper concentration in dead zone of first Cho-
unit lette–Cloutier unit (g L1)
f plug flow fraction of slow flowing branch Y aL2 organic copper concentration in active zone of second
C stripping equilibrium isotherm parameter Cholette–Cloutier unit (g L1)
D stripping equilibrium isotherm parameter (g L1) Y bL2 organic copper concentration in dead zone of second
KE extraction mass transfer coefficient (h1) Cholette–Cloutier unit (g L1)
KS stripping mass transfer coefficient (h1) Ym organic copper concentration in mixer (g L1)
Qai aqueous mixer inlet flowrate (m3 h1) YP time delayed organic copper concentration in slow flow
QAL aqueous flowrate in slow flow branch in settler (m3 h1) branch (g L1)
Qam aqueous mixer outlet flowrate (m3 h1) Y ST organic equilibrium copper concentration in extraction
QAUi aqueous flowrate in ith CSTR outlet in fast flow branch stage (g L1)
in settler (m3 h1) YUi organic copper concentration in ith CSTR in fast flow
Qoi organic mixer inlet flowrate (m3 h1) branch in settler (g L1)
QOL organic flowrate in slow flow branch in settler (m3 h1) eE extraction stage organic copper concentration steady
Qom organic mixer outlet flowrate (m3 h1) state gradient (g L1)
QOUi organic flowrate in ith CSTR outlet in fast flow branch in eS stripping stage aqueous copper concentration steady
settler (m3 h1) state gradient (g L1)
VAL1 aqueous volume of first Cholette–Cloutier unit (m3) gE extraction efficiency coefficient
VAL2 aqueous volume of second Cholette–Cloutier unit (m3) gS stripping efficiency coefficient
Vam aqueous volume in mixer (m3) uA fraction of aqueous solution flowing by slow flow
VAUi aqueous volume of ith CSTR (m3) branch in settler
VOL1 organic volume of first Cholette–Cloutier unit (m3) uJ fraction of organic solution flowing by slow flow branch
VOL2 organic volume of second Cholette–Cloutier unit (m3) in settler
Vom organic volume in mixer (m3) k1A aqueous exchange rate between zones in first Cholette–
VOUi organic volume of ith CSTR (m3) Cloutier unit
VS settler volume (m3) k1O organic exchange rate between zones in first Cholette–
X* aqueous equilibrium copper concentration (g L1) Cloutier unit
Xi aqueous inlet copper concentration (g L1) k2A aqueous exchange rate between zones in second Cho-
X aL1 aqueous copper concentration in active zone of first lette–Cloutier unit
Cholette–Cloutier unit (g L1) k2O organic exchange rate between zones in second Cho-
X bL1 aqueous copper concentration in dead zone of first Cho- lette–Cloutier unit
lette–Cloutier unit (g L1)

requires a great deal of specific experimental information, not nor- predictions of the steady states of a test SX–EW plant. The mixer
mally found in an industrial environment. Using simple isotherms was modelled as an ideal continuous stirred tank reactor, hereinaf-
calibrated with plant data, Aminian et al. (1998) obtained good ter, CSTR, and the settler as two compartments in parallel, an ideal
plug flow and an ideal CSTR. Using a more elaborated steady state
model, Pinto et al. (2004) improved the configuration of an SX Cu
plant applying multi-objective optimization. They included chem-
ical reaction kinetics and a non-ideal hydrodynamics in the mixer–
settler units. Then a multi-variable optimizing objective function
was defined that considered equipment geometry, residence time
and agitator speeds, among other design parameters. The optimi-
zation yielded the optimum unit and plant design, which can be
applied for the design of new plants, or for improving existing de-
signs. In spite of being extremely useful in SX plant design, that
model cannot be used to develop and test control systems.
Only dynamic models are useful to design control strategies by
Fig. 1. Typical LX–SX–EW copper plant. simulation. Most of these models presented in the literature
1352 C.M. Moreno et al. / Minerals Engineering 22 (2009) 1350–1358

include simple thermodynamics and rate expressions, where most 2.1. Mass balances
of the complexity is lumped into the hydrodynamics of the settler.
An example of this is the mixer–settler cascade model for rare The model was obtained by applying independent copper and
earths of Wichterlová and Rod (1999). Here, the hydrodynamics total mass balances on the mixer and the settler (Fig. 2) under
were modelled as a series of CSTRs. However, this model is not that the following assumptions:
useful to design control systems since it is too simple, not able to
reproduce the complex dynamics observed in real plants. Wilkin-  Mass transfer in the settler is significantly lower than in the
son and Ingham (1983) and Ingham et al. (2007) modelled the mix- mixer, therefore mass transfer is considered only in the mixer
er as a CSTR, but included entrainment factors in the outlet (Wilkinson and Ingham, 1983).
streams. Furthermore, the settler was modelled as two parallel  Equilibrium isotherms define composition gradients for mass
compartments, just as in the Aminian model. Although this is an transfer (Steiner and Hartland, 1983).
improvement over Wichterlová and Rod model, no comparison  Perfect mixer (Wilkinson and Ingham, 1983).
with industrial data has been presented so far. Komulainen et al.  Entrainments only occur in the settler (Aminian et al., 1998).
(2006) presented a dynamic model of a SX Cu process calibrated  Settler modelled with non-ideal hydrodynamics (Ingham et al.,
with plant data, however, the settler is modelled simply as a delay. 2007).
Hence, even though the model described the SX process dynamics  Constant density.
well, it will probably be difficult to fit that model to other plants
with more complex behaviour.
Here, we present a flexible SX–Cu unit model, able to reproduce
2.1.1. Mixer balances
the complex dynamics observed on any stage at any industrial
Applying total mass balances (see Fig. 2) to both, aqueous and
plant. The model includes McCabe–Thiele specific diagrams and
organic phases in the mixer tank, yield,
complex settler hydrodynamics. The model considers variables
normally measured at industrial facilities and simulates the dy- dV Am ðQ þ Q Om Þ
namic response of copper concentrations in both phases in the out- ¼ Q Ai  Am V Am ð1Þ
dt V Am þ V Om
let streams.
dV Om ðQ þ Q Om Þ
¼ Q Oi  Am V Om ð2Þ
dt V Am þ V Om
2. Process model
copper balances differ if the tank is in an extraction or in a stripping
stage. For an extraction unit,
Fig. 1 (Jackson, 1986) describes a standard LX–SX–EW industrial
plant. Leaching (LX) is the first stage, where copper is extracted by dX m Q
washing the ore with an acid aqueous solution (SX raffinate). The ¼ Ai ðX i  X m Þ  K E ðY E  Y m Þ ð3Þ
dt V Am
outlet stream of LX, the pregnant liquid solution (PLS), is the inlet
dY m Q
stream to the first stage of the SX process (extraction). In this stage, ¼ Oi ðY i  Y m Þ þ K E ðY E  Y m Þ ð4Þ
dt V Om
the PLS stream is mixed with the barren organic solution (BO), i.e.,
the outlet stream of the second SX stage (stripping), to selectively and for a stripping unit,
transfer copper from the aqueous phase to the organic phase. The
outlet loaded organic solution (LO) from the extraction stage, is dX m Q
mixed in the stripping stage with the lean electrolyte (LE) stream ¼ Ai ðX i  X m Þ þ K S ðX S  X m Þ ð5Þ
dt V Am
coming from the electrowinning process. Since there is a different dY m Q
pH in the stripping stage, copper transfers now from the organic ¼ Oi ðY i  Y m Þ  K S ðX S  X m Þ: ð6Þ
dt V Om
phase to the aqueous phase. Then, the outlet aqueous stream from
the stripping stage, rich electrolyte (RE), goes to the electrowinning In these equations, QAi and QOi are the aqueous (subindex A) and or-
stage, where copper is extracted in cathodes, bringing the lean ganic (subindex O) mixer inlet flowrates, VAm and VOm are the aque-
electrolyte back to the loop. ous and organic volumes in the mixer, Xm and Ym are the copper
In this work we developed a general dynamic model for mixer– concentrations in the aqueous and organic phases in the mixer,
settlers and fit it to two different units, one in the multi-units respectively, KE and KS are the copper mass transfer coefficients in
extraction stage and the other in the multi-units stripping stage the extraction and stripping units (subindex E stands for extraction
of an SX process of a LX–SX–EW copper plant from Molymet S.A., and S for stripping).
similar to that shown in Fig. 1. Although, this plant, with an annual X S and Y E are pseudo-equilibrium copper concentrations de-
production of 1 million lb of copper, is rather small compared to fined by empirical equilibrium isotherms (X* or Y*) and extraction
other industrial mining facilities. efficiencies (gE or gS) (Wilkinson and Ingham, 1983).

Y E ¼ ð1  gE ÞY i þ gE Y  ð7Þ
X S ¼ ð1  gS ÞX i þ gS X  ð8Þ

2.1.2. Settler balances

A non-ideal flow (see Fig. 3) for both phases was used to model
the hydrodynamics in this tank (Ingham et al., 2007). The basic
model assumes that the solution is split into two streams, one that
moves fast (fraction 1  u) and the other that moves slowly (frac-
tion u). The fast moving stream, represented by the upper branch
(subindex U) in Fig. 3, is modelled by N perfectly mixed tanks con-
nected in series. Then, in the upper branch, the aqueous phase cop-
Fig. 2. Flow diagram of a mixer–settler unit. per balances are,
 C.M. Moreno et al. / Minerals Engineering 22 (2009) 1350–1358 1353

(1-φA1) QAm1 QAUN

VUA1 VUAN
Xm1 XUN

QAm1,Xm1 QA1,X1

φA1 QAm1
Xm1 QAL1 a1AVL1A a2AVL2A
τ AL QAL1 QAL1
XP1 XL1A xL1A XL2A XL2A

λ1AQAL1 λ1AQAL1 λ1AQAL1

b1AVL1A b2AVL2A

XL1B XL2B

Fig. 3. Hydrodynamic model of aqueous solution in the settler; the hydrodynamic model for the organic solution is the same, but with sub-indexes ‘‘O” instead of ‘‘A”, and
copper concentrations ‘‘Y” instead of ‘‘X”.

Q OL1  a 
a
i¼1 dY L2
¼ Y L1 þ k2O Y bL2  ð1  k2O ÞY aL2 ð23Þ
dX Ui ð1  uA ÞQ Am dt a2O V OL2
¼ ðX m  X Ui Þ ð9Þ
dY L2 k2O Q OL1  a 
b
dt V AUi
¼ Y L2  Y bL2 ð24Þ
i ¼ 2...N dt b2O V OL2
dX Ui Q AUi1 Therefore, the final copper concentration at the exit of the set-
¼ ðX Ui1  X Ui Þ ð10Þ
dt V AUi tler is a weighted sum of both outlet branches,
and the organic phase copper balances are, X ¼ uA X aL2 þ ð1  uA ÞX UN ð25Þ
i¼1 Y ¼ uO Y aL2 þ ð1  uO ÞY UN ð26Þ
dY Ui ð1  uO ÞQ Om In these equations, a stands for active zone, b stands for dead zone
¼ ðY m  Y Ui Þ ð11Þ
dt V OUi and k stands for exchange rate between zones.
i ¼ 2...N
dY Ui Q OUi1 2.2. Constitutive equations
¼ ðY Ui1  Y Ui Þ ð12Þ
dt V OUi
2.2.1. Equilibrium
In turn, the slowly moving stream, represented by the lower
The equilibrium isotherm in an extraction unit (see Eq. (7)) is
branch in Fig. 3 (subindex L), is modelled as a plug flow tank in ser-
given by (Komulainen et al., 2006),
ies with two Cholette–Cloutier units (Aminian et al., 1998). Each of
these consists of two perfectly mixed zones, an active (top) and a AX 
dead one (bottom). Total and copper balances for both phases in Y ¼ ð27Þ
X þ B
the plug flow tank are,
A and B are empirical coefficients fitted with process measurements
Q AL ðtÞ ¼ uA Q Am Hðt  sAL Þ ð13Þ and extractant manufacturer data as follows:
X P ðtÞ ¼ X m Hðt  sAL Þ ð14Þ
A ¼ a  ML ð28Þ
Q OL ðtÞ ¼ uO Q Om Hðt  sOL Þ ð15Þ  b
Y P ðtÞ ¼ Y m Hðt  sOL Þ ð16Þ 10pH  h i h i 
B¼ c
PLS
d  Cuþ2 þ f  Cuþ2 ð29Þ
Here, H is the Heaviside step function, which represents a time ML PLS BO

delay. Copper balances for both phases in the two Cholette–Clou-

where ML is the maximum load (g L1) of Cu in the organic stream
tier units are given by,
(extractant manufacturer data), and pH and [Cu+2] are daily aver-
Q AL   aged off-line measured values. In these equations, a, b, c, d and f
a
dX L1
¼ X P þ k1A X bL1  ð1  k1A ÞX aL1 ð17Þ are fitting parameters (see Table 1).
dt a1A V AL1
k1A Q AL  a 
b
dX L1 Table 1
¼ X L1  X bL1 ð18Þ Isotherm fitting parameters.
dt b1A V AL1
Q AL1  a 
a Parameters Values
dX L2
¼ X L1 þ k2A X bL2  ð1  k2A ÞX aL2 ð19Þ a 0.99
dt a2A V AL2
b 1.02
dX L2 k2A Q AL1  a 
b
c 1.01
¼ X  X bL2 ð20Þ d 35.15
dt b2A V AL2 L2
f 27.15
Q OL  
a
dY L1 g 0.11
¼ Y P þ k1O Y bL1  ð1  k1O ÞY aL1 ð21Þ h 444.49
dt a1O V OL1
m 0.10
k1O Q OL  a 
b
dY L1 n 0.81
¼ Y  Y bL1 ð22Þ p 8.91
dt b1O V OL1 L1
1354 C.M. Moreno et al. / Minerals Engineering 22 (2009) 1350–1358

Similarly, for the stripping isotherm (see Eq. (8)) (Komulainen Table 2
et al., 2006), Measured variables in LX–SX–EW copper plant.

Measured variables Sampling time (h) Symbols

Y  ¼ C  X þ D ð30Þ
Inputs
C ¼ g  ML ð31Þ Aqueous inlet flowrate 1 Qai
vol:% m [Cu+2] aqueous inlet 1 Xi
D¼h þp ð32Þ Organic inlet flowrate 1 Qoi
½H2 SO4 nLE [Cu+2] organic inlet 1 Yi
Mixer volume 1 Vm
where vol.% is the % v/v of reactant in the organic and [H2SO4]LE is Settler volume 1 Vs
the free acidity of the lean electrolyte (g L1); g, h, m, n and p are fit- pH of PLS 1 pHPLS
ting parameters (see Table 1). Variables involved in Eqs. (28)–(32) Free acidity of the lean electrolyte 1 [H2SO4]LE
were measured from common shift samples in the laboratory by Outputs
chromatography (see Table 2). Fig. 4a shows daily averages of calcu- Volumes of phases in mixer (i = A, O) 1 Vim
Flowrates at mixer exit (i = A, O) 1 Qim
lated isotherm parameters A, B, C and D. Further details regarding
[Cu+2] in aqueous phase at mixer exit 1 Xm
fitting isotherm parameters are given in Appendix A. [Cu+2] in organic phase at mixer exit 1 Ym

2.2.2. Mass transfer

In this study, the extraction rate is modelled by a mass transfer
in the laboratory by chromatography from shift samples (see Table
expression obtained from the interface theory (Jackson, 1986),
2). Fig. 4b shows daily averages of calculated KE and KS coefficients.
Q Ai ðX i  X m Þ Q ðX i  X m Þ
KE ¼ ¼ Ai ð33Þ
V Am ðY E  Y m Þ V Am eE 2.2.3. Efficiencies
Q ðY i  Y m Þ Q Oi ðY i  Y m Þ Efficiencies for both, extraction and stripping units, are defined
K S ¼ Oi ¼ ð34Þ by,
V Om ðX S  X m Þ V Om eS
eE þ ðY m  Y i Þ
In these equations, the steady state copper concentration gradients gE ¼ ð35Þ
Y  Yi
(eE ¼ Y ST  Y m for an extraction unit and eE ¼ X ST  X m for a strip-
e þ ðX  X i Þ
ping unit) are fitting parameters (Komulainen et al., 2006). Copper gS ¼ S  m ð36Þ
X  Xi
concentrations in Eqs. (33) and (34) (Xi, Yi, Xm, Ym) were measured

1.2
(a)
Scaled values

0.4

-0.4

-1.2
0 168 336 504 672 840 1008 1176 1344
t (h) A B C D

1.80
(b)
K (scaled)

0.80

-0.20

-1.20
0 168 336 504 672 840 1008 1176 1344
t (h) E S

0.30
(c)
0.00
β (scaled)

-0.30

-0.60
0 168 336 504 672 840 1008 1176 1344
t (h) E S

Fig. 4. Calculated model parameters: (a) Isotherm parameters. (b) Mass transfer coefficients. (c) Efficiencies.
 C.M. Moreno et al. / Minerals Engineering 22 (2009) 1350–1358 1355

In the mixers, Ym and Xm are the outlet copper concentrations, Yi with magnetic flowmeters (Siemens MAG 3100) and standard
and Xi are the inlet concentrations, and Y* and X* are the equilib- pH-meters, respectively. Copper and acid concentrations were
rium concentrations. Fitting parameters, eE1 and eS1, were set by measured in the laboratory by chromatography. In the actual pro-
trial and error to minimize the sum of squared errors between sim- cess at Molymet S.A., inlet copper concentration standard devia-
ulated and measured outlet copper concentrations. Note that an tions ranged between 1 and 2 g L1, flowrates between 0.03 and
efficiency 1 means that the respective steady state copper concen- 0.3 m3 h1, and pH varied between 0.8 and 1.8. To keep confiden-
tration gradient is 0; hence, Y E ¼ Y  or X S ¼ X  . Fig. 4c shows daily tiality, no further process details can be disclosed.
averaged calculated efficiencies. In these calculations, copper con-
centration values averaged every 8 h in the laboratory by chroma-
tography, were used. Nomenclature provides a complete 3. Results and discussion
description of the symbols used.
Simulations were performed with SimulinkÒ software, using First, a parameter sensitivity analysis was carried out to aid
calculated model parameters (shown in Fig. 4) updated daily, input model calibration. After calibration, i.e., setting model parameters
variables (shown in Fig. 5) measured hourly, and main model at fixed values, simulations were compared with measured outputs
parameters (see Table 3) which were fixed during the whole sim- of the actual process, shown in Fig. 6. The same analysis was car-
ulation run (56 days). Flowrates and pH were measured on-line ried out for both extraction and stripping units.

9.6
(a)
X (scaled)

4.8

0.0

-4.8
0 168 336 504 672 840 1008 1176 1344
PLS LE
t (h)
1.4
(b)
Y(scaled)

0.0

-1.4

-2.8
0 168 336 504 672 840 1008 1176 1344
BO LO
t (h)

0.08
(c)
QA(scaled)

-0.08

-0.24

-0.40
0 168 336 504 672 840 1008 1176 1344
t (h) PLS LE

0.03
(d)
QO(scaled)

0.01

-0.01

-0.03
0 168 336 504 672 840 1008 1176 1344
BO LO
t (h)

Fig. 5. Scaled measured input variables: (a) Copper concentration in aqueous inlet streams. (b) Copper concentration in organic inlet streams. (c) Aqueous inlet flowrate. (d)
Organic inlet flowrate.
1356 C.M. Moreno et al. / Minerals Engineering 22 (2009) 1350–1358

Table 3
Model parameters for extraction unit (E) and stripping unit (S). Sub-indexes A and O stand for aqueous and organic phases, respectively. C–C stands for Cholette–Cloutier units.

Parameters Symbols Nominal values (E) Nominal values (S) Units

Fitting parameters
Fraction of plug flow in lower branch f 0.3 0.4 –
Steady state concentration gradient eE 0.7 1.9 g L1
Fraction going through lower branch (i = A, O) ui 0.8 0.9 –
Fraction in C–C units active zones (i = 1, 2; j = A, O) aij 0.6 0.7 –
Exchange rate between C–C zones (i = A, O; j = 1, 2) kij 1.8 1.3 –
Number of upper branch CSTR tanks N 2.0 2.0 –
Calculated from fitting parameters (see Appendix A)
Upper branch CSTR volume (i = A, O; j = 1, N) ViUj 5.3 2.7 m3
Plug flow residence time (i = A, O) sLi 0.9 1.3 h
Volume of C–C units (i = A, O; j = 1, 2) ViLj 14.2 14.3 m3
Fraction of dead volume in C–C units (i = 1, 2; j = A, O) bij 0.4 0.3 –

2.10
(a)
X (scaled)

1.10

0.10

-0.90
0 168 336 504 672 840 1008 1176 1344

t (h) Simulated-E Measured-E

0.85
(b)
Y (scaled)

0.45

0.05

-0.35
0 168 336 504 672 840 1008 1176 1344
t (h) Simulated Measured

2.10
(c)
1.10
X (scaled)

0.10

-0.90
0 168 336 504 672 840 1008 1176 1344

t (h) Simulated-S Measured-S

0.20
(d)
-0.20
Y (scaled)

-0.60

-1.00
0 168 336 504 672 840 1008 1176 1344
t (h) Simulated-S Measured-S

Fig. 6. Simulated vs. measured output variables: (a) Copper concentration in aqueous extraction outlet stream. (b) Copper concentration in organic extraction outlet stream.
(c) Copper concentration in aqueous stripping outlet stream. (d) Copper concentration in organic stripping outlet stream.
 C.M. Moreno et al. / Minerals Engineering 22 (2009) 1350–1358 1357

0.85

Y (scaled)
0.45

0.05

-0.35
0 168 336 504 672 840 1008 1176 1344
t (h) Model Delay Real

Fig. 7. Comparison between our model and a simplified pure delay settler model. Copper concentration in extraction organic-outlet stream.

3.1. Sensitivity analysis so they can take opportune corrective actions. To develop these
associations though, exhaustive process operation and data analy-
The impact of changing model parameters in the simulation of sis are necessary, after running the model in parallel with the pro-
output variables was assessed. We were particularly interested in cess for long periods. It is worthwhile noting that the model has a
detecting high impact parameters that could then be used to fit prediction horizon of 24 h, since calculated parameters (see Fig. 4)
the model to reproduce actual process outputs. Starting from a are updated daily based on previous day laboratory data.
nominal case, defined by arbitrarily assigned values of input vari- The model, as it is can be used to implement on-line state and
ables and model parameters, the model was simulated several parameter estimations, using for example an Extended Kalman Fil-
times applying one step change at a time in a given parameter. ter. These estimations can provide further tools to improve process
By trial and error, we found that for both units (extraction and supervision, and even can help to implement automatic control
stripping), the split fractions, us, and steady state copper concen- strategies. Furthermore, a dynamic model for the entire extraction
tration gradients, es, were high impact parameters. Therefore, they SX sub-process can be developed to assess the impact of better
were used to calibrate the model for the extraction and the strip- operational practices or alternative automatic control strategies
ping units, running a large number of trial and error simulations. on process performance, before implementing them in the real
The rest of model parameters, since they did not affect much sim- plant, saving time, resources and minimizing the risk of cata-
ulation results, were defined after few simulations. The final model strophic events.
parameter values are given in Table 3.
4. Conclusions
3.2. Comparing simulations with measured values
The developed model is a relatively simple form to represent
Fig. 6 shows simulated model outputs compared with plant non-ideal flow in mixer settler tanks. We chose this model struc-
copper concentration measurements in both phases. It can be seen ture since it represents a good compromise between accuracy
that good agreement between simulations and measurements has and complexity. The main contribution of this work is to show that,
been achieved for both units. For the extraction unit, relative mean with few sensitive fitting parameters, the proposed model provides
squared errors for the outlet stream copper concentrations were realistic dynamic simulations of industrial mixer–settler units.
0.03% for the aqueous phase and 6.76% for the organic phase. In Also, the model can be calibrated with standard measurements
turn, for the stripping unit, relative mean squared errors were and information normally available in current industrial LX–SX–
0.07% for the aqueous phase and 2.89% for the organic phase. These EW plants; therefore, specific dynamic models for a given indus-
differences in the observed errors are probably due to inherent trial SX plant can be developed with moderate effort.
sampling and laboratory measurements variability. For example,
measurement errors in the organic phase are larger. In addition,
Acknowledgements
the extraction unit shows much more variability than the stripping
unit, which is expected since its feed comes directly from the heap
J.R.P. appreciates the support of AGAUR from the Generalitat de
Leaching stage. In turn, this variability is attenuated at the outlet of
Catalunya through Grant 2007PIV-00017 and Pontificia Universi-
the stripping unit due to the damping effect of the mixing tanks.
dad Católica de Chile for financial support for a sabbatical stay at
Despite this difference, the model is flexible enough to reproduce
Department D’Enginyeria Química at Universitat Rovira i Virgili.
the dynamics of both units well. Furthermore, we compared the
C.M.M appreciates the support of Molymet S.A. for providing
performance of our model against a simple delay model (repre-
experimental information and Pontificia Universidad Católica de
sented as a plug flow) for the settler. The simpler model performs
Chile for granting a PhD scholarship. We appreciate the comments
much worse, yielding a relative mean square error of 15.0% for the
of the anonymous referees that help us to significantly improve the
aqueous phase (not shown) and 44.8% for the organic phase (see
readability of this paper.
Fig. 7). Consequently, it is highly beneficial to model the settler
with a more complex hydrodynamic model.
Fig. 4 also shows interesting features, like the extreme variation Appendix A. Model calibration
of the estimated mass transfer coefficients and efficiencies. For in-
stance, the significant drop in the efficiency (see Fig. 4c) of the A.1. Equilibrium isotherm calculations
extraction unit can be attributed to an operational problem and
may be useful to identify its cause. Hence, the model can be refined For a collection of pH, and [Cu+2]PLS values, the extractant man-
more by associating mass transfer and efficiencies with specific ufacturer provides several McCabe diagrams, that can be used to
operational problems like entrainment, crud formation and vari- obtain isotherm coefficients, A and B, as a correlation of these vari-
able phase separation times. These would be highly desirable to ables. Additionally, plant measurements of ML, pH, [Cu+2]PLS and
improve process supervision by alerting process engineers earlier, [Cu+2]BO (normally available at industrial facilities) were averaged
1358 C.M. Moreno et al. / Minerals Engineering 22 (2009) 1350–1358

daily and, through Eqs. (28) and (29), estimated values of isotherm V S ð1  uÞ
^ and B
^ were obtained. Then, parameters a, b, c, d and f V Uij ¼ ðA3Þ
coefficients A N
were fitted to minimize the following sum of squared differences, f uV S
sLi ¼ ðA4Þ
56 h i2 h i2  Q im
X
Min ^i
Ai  A ^i
þ Bi  B ðA1Þ ð1  f ÞuV S
V Lij ¼ ðA5Þ
i¼1 2
bij ¼ 1  aij ðA6Þ
where subindex ‘i’ refers to the respective day. The fitted values are
given in Table 1.
Similarly, for a collection of [H2SO4] values, a set of McCabe dia- References
grams was given by the manufacturer, yielding a set of values for C
Aminian, H., Bazin, C., Hodouin, D., 1998. Residence time distributions in SX–EW
and D. Additionally, a set of plant measurements of [H2SO4]LE and equipment. International Journal of Mineral Processing, 235–242.
vol.% were averaged daily and, through Eqs. (31) and (32), esti- Bergh, L.G., Yianatos, J.B., 2001. Current status and limitations of copper SX/EW
mated values of isotherm coefficients C ^ and D ^ were obtained. plants control. Minerals Engineering 14 (9), 975–985.
Ingham, J., Dunn, I.J., Heinzle, E., Prenosil, J.E., 2007. Chemical Engineering
Then, parameters h and p were fitted to minimize the following
Dynamics. An Introduction to Modelling and Computer Simulation. Wiley-
sum of squared differences, VCH. pp. 174–183.
Jackson, E., 1986. Hydrometallurgical Extraction and Reclamation. John Wiley &
56 h
X i2 h i2 
^i ^i Sons. pp. 109–138.
Min Ci  C þ Di  D ðA2Þ Komulainen, T., Pekkala, P., Rantala, A., Jamsa-Jounela, S.L., 2006. Dynamic
i¼1 modelling of an industrial copper solvent extraction process.
Hydrometallurgy 81 (1), 52–61.
Pinto, G.A., Durao, F.O., Fiuza, A.M.A., Guimaraes, M.M.B.L., Madureira, C.M.N., 2004.
A.2. Mass transfer and efficiency calculations Design optimisation study of solvent extraction: chemical reaction, mass
transfer and mixer–settler hydrodynamics. Hydrometallurgy 74, 131–147.
Steiner, L., Hartland, S., 1983. Unsteady-state extraction. In: Lo, T.C., Baird, M.H.I.,
Mass transfer and efficiency coefficients were obtained using Hanson, C. (Eds.), Handbook of Solvent Extraction. John Wiley & Sons, pp. 249–
Eqs. (33)–(36). Values of eE and eS were calibrated by trial and error 264.
after several simulations. The final values of eE and eS were those Van Bochove, G.H., Krooshoff, G.J.P., de Loos, T.W., 2000. Modelling of liquid–liquid
equilibria of mixed solvent electrolyte systems using the extended electrolyte
that minimized the mean squared error between the simulated NRTL. Fluid Phase Equilibria 171, 45–58.
and the measured response. Wichterlová, J., Rod, V., 1999. Dynamic behaviour of the mixer–settler cascade.
Extractive separation of the rare earths. Chemical Engineering Science, 4041–
4051.
A.3. Calculated parameters in Table 3 Wilkinson, W.L., Ingham, J., 1983. Dynamic behavior and control. In: Lo, T.C., Baird,
M.H.I., Hanson, C. (Eds.), Handbook of Solvent Extraction. John Wiley & Sons, pp.
They were obtained from, 853–886.