Professional Documents
Culture Documents
On Scalable Analytical Models For Heap Leaching
On Scalable Analytical Models For Heap Leaching
. Infact, we consider
n ore particles, of radius R, numbered from 0 to n1, which forms
a heap section of height Z (see Fig. 1).
We denote by q
i
the quantity of valuable specie which can be
recovered fromthe particle i when time tends to innity, 0
i
1,
the fraction recovered from q
i
, i =0, n1. We consider that total
recovery of each particle it is not possible and does not happen
in practice. The reason for that is that in the heap the things does
not behaves ideally as phenomenological models assume. In fact, in
the heap there exists canalizations, clusters, and particles can not
behave ideally because they are difcult to reach for diffusions.
222 M.E. Mellado et al. / Computers and Chemical Engineering 35 (2011) 220225
Therefore,
=
n1
i=0
i
q
i
= q
n1
i=0
i
,
where q is the expected average value for the set of particles. Let ,
, R. We dene a real valued function f : R
4
R by
j (, , , Z):=
Z
, (1)
and dene,
i
= [1 j ||, , , Z||]
i
, i = 0, n 1.
The above choice will be clear in what follows. Then,
= q
n1
i=0
[1 j (, , , Z)]
i
= q
n1
i=0
_
1
Z
_
i
.
It is straightforward to show, by using the basic nite summation
formula, that,
(, , , Z) = q
_
1 (1 (Z
+]))
Z]2R
(Z
+])
_
.
The above equation can be normalized to obtain,
(, , , Z) =
Z
+
_
1
_
1
Z
_
Z]2R
_
,
which is our proposed constitutive equation for the recovery when
time tends to innity, E
, this is,
(, , , Z) =
Z
+
. (2)
It can be seen the above expression is a simple formula that allows
to scale-up a model inwhat concerning the recovery whenthe time
tends to innity. We consider this formula as a generalization for
the adjustment which is done in phenomenological models. It is
enough to consider that =0 and =+1. In fact, by using least
squares in the Non-linear regression module of Infostat software
(see footnote 1), we adjusted three data sets of experimental obser-
vations which involves recovery E(t) and time t, the recovery at
innity E
(Z) (1 qc
k (tt
0
)
).
The estimates for q and k coincides (q=1, k =0.14), but E
depends
on Z. This is, for Z =3, Z =6 and Z =9, the values are 78.77, 73.57 and
70.08, respectively. The effect on the recovery E(t) of the parameter
q is not signicant.
The choice of the function f is not so arbitrary. For particles at
a higher height, 1f is less than at lower level. This offers sev-
eral choices for f. On the other hand, the experimental data shows
that the relation between the recovery at innity E
(, , , Z) whent .
It is easy to obtain that,
(t) =
(, , , Z) (1 c
k((us]c
b
Z)t
k)
), (3)
where k,
k and, , , are constants tobe computed. Whenthe reac-
tion order is different to the unity, we need to state the Bernoulli
M.E. Mellado et al. / Computers and Chemical Engineering 35 (2011) 220225 223
equation,
dy
d0
= k
0
y
n
0
, (4)
where y is a dynamic variable and n
0
stands for a generalized reac-
tion order with respect to 0. It is important to remark that, for
instance, inPetersenandDixon(2007), the reactionorder of chemi-
cal reactions has been established trough a chemical analysis of the
inner reactions. Those formulae can be replaced here also. More-
over wepoint out that alsothat model needs tobecalibrated, but we
are interested in the global behaviour and overall in the scaling-up.
Without loss of generality, we dene y :=E
(, , , Z) E. Then,
the differential equation for modelling heap leaching is given by
d
d0
(
(, , , Z) ) = k
0
(
(, , , Z) )
n
0
. (5)
together with
(w) = 0.
The above equation have an analytical solution given by
(0) =
(, , , Z)
_
k
0
(n
0
1)(0 w) +
(, , , Z)
1n
0
1](1n
0
)
. (6)
An important issue from the mathematical and engineering point
of view is the fact that, if we take limit when n
0
tends to one in the
above equation, one recovers exactly the model for n
0
=1. There-
fore, we have a family of models which is continuous with respect
to n
0
. Therefore, in actual variables, we have,
(t) =
(, , , Z)
_
k (n
0
1)
_
u
s
c
b
Z
t
k
_
+
(, , , Z)
1n
0
_
1](1n
0
)
, (7)
where k,
k, n
0
and , , , are constants to be computed. As can
be seen, the above model can scale-up the height of the heap, the
acid ux and the nal recovery. The unknown parameters can be
computed by using optimization techniques.
4. The second analytical model suitable for scale-up
Now, we consider both 0 and z kinetics with a reaction order
equal to the unity, n
0
=n
z
=1. We state the differential equation
D(D +k
0
w
0
)(D +k
z
w
z
) = 0,
where E is the recovery at time t, k
0
is a constant related with
w
0
= (u
s
]c
b
Z) and k
z
is related to w
z
= (D
/c
]c
c
R
2
). The initial and
radiation condition are,
(w) = 0
and
=
(, , , Z) whent .
We obtain that,
(t) =
(, , , Z)
_
1 zc
k
0
_
us
c
b
Z
tw
_
(1 z) (1 c
kz
D
/c
R
2
c
0
(t
c
b
Z
us
w)
)
_
, (8)
where k
0
, k
z
, , , and z are constants to be computed.
We nowpresent the case whenthe reactionorders n
0
andn
z
can
be different to the unity. We include the order of reaction related
to 0, n
0
and the order of reaction n
z
as the induced by z. We have a
recuperation E
0
, with asymptotic behavior E
0,
, which is due to the
the kinetic induced by 0. Also, a recuperation E
z
, with asymptotic
behavior E
z,
, which is induced by z. Moreover, we consider the
total recovery E given by
=
0
+
z
and, consequently,
=
0,
+
z,
.
Here, E
0
and E
z
are the solutions of the ordinary differential equa-
tions,
d
d0
(
0,
0
) = k
0
(
0,
0
)
n
0
,
0
(w) = 0
and
d
dz
(
z,
z
) = k
z
(
z,
z
)
nz
,
z
_
D
/c
R
2
c
0
c
b
Z
u
s
w
_
= 0.
The solutions of these differential equations are given by,
0
=
0,
_
k
0
(n
0
1)
_
u
s
c
b
Z
t w
_
+(
0,
)
1n
0
_
1](1n
0
)
(9)
and
z
=
z,
_
k
z
(n
z
1)
D
/c
R
2
c
0
(t
c
b
Z
u
s
w) +(
z,
)
1nz
_
1](1nz )
. (10)
Therefore, by adding the equations we nd a heap leaching model
where we have assumed that the generalized reaction orders are
unknown. This is as follows,
(t) =
(, , , Z)
_
k
0
(n
0
1)
_
u
s
c
b
Z
t w
_
+(z
(, , , Z))
1n
0
_
1](1n
0
)
_
k
z
(n
z
1)
D
/c
R
2
c
0
_
t
c
b
Z
u
s
w
_
+((1z)
(, , , Z))
1nz
1](1nz )
. (11)
Finally, we observe that with a sample of observations, nonlinear
optimization methods can be used to adjust this scale-up analytical
model.
5. The third analytical model suitable for scale-up
As in the previous section, we consider both 0 and z kinetics
with a reaction order equal to the unity, n
0
=n
z
=1. We can state
the differential equation,
D(D +k
0
w
0
)(D +k
z
w
z
) = 0,
where E, k
0
, w
0
= (u
s
]c
b
Z), k
z
, w
z
= (D
/c
]c
c
R
2
) have the same
meaning as above. Now, we just only consider the radiation condi-
tion to obtain a degree of freedom in the model, i.e.,
=
(, , , Z) whent .
It is easy to obtain that,
(t) =
(, , , Z) [1 +zc
k
0
((us]c
b
Z)tw)
+1c
kz (D
/c
]R
2
c
0
)(t(c
b
Z]us)w)
_
, (12)
where , , , k
0
, k
z
, z, 1 are constants to be computed.
We nowpresent the case whenthe reactionorders n
0
andn
z
can
be different to the unity. We include the order of reaction related
to 0, n
0
and the order of reaction n
z
as the induced by z. We have
a recuperation E
0
, with asymptotic behavior E
0,
, which is due to
the the kinetic induced by 0. Also, a recuperation E
z
, with asymp-
totic behavior E
z,
, which is caused by the kinetic induced by z.
Moreover, we consider the total ore recovery E given by
=
0
+
z
224 M.E. Mellado et al. / Computers and Chemical Engineering 35 (2011) 220225
and, consequently,
=
0,
+
z,
.
Here, E
0
and E
z
are the solutions of the ordinary differential equa-
tions
d
d0
(
0,
0
) = k
0
(
0,
0
)
n
0
,
and
d
dz
(
z,
z
) = k
z
(
z,
z
)
nz
.
The solutions of these differential equations are given by,
0
=
0,
_
k
0
(n
0
1)
_
u
s
c
b
Z
t
_
+(
0,
)
1n
0
_
1](1n
0
)
(13)
and
z
=
z,
_
k
z
(n
z
1)
D
/c
R
2
c
0
_
t
c
b
Z
u
s
w
_
+(
z,
)
1nz
_
1](1nz )
. (14)
Therefore, by adding the equations we nd a heap leaching model
where we have assumed that the generalized reaction orders are
unknown. This is as follows,
(t) =
(, , , Z)
_
k
0
(n
0
1)
_
u
s
c
b
Z
t w
_
+(z
(, , , Z))
1n
0
1](1n
0
)
_
k
z
(n
z
1)
D
/c
R
2
c
0
_
t
c
b
Z
u
s
w
_
+(1
(, , , Z))
1nz
1](1nz )
. (15)
Finally, we observe that with a sample of observations, nonlinear
optimization methods can be used to adjust this third model.
Before the applications, we summarize that we have proposed
three models. Model 1 described by Eqs. (3) and (7) considers just
the heap kinetic, model 2 presented by Eqs. (8) and (11) which
considers heap and particle kinetics and nally model 3, as a gen-
eralization of model 2, by means of Eqs. (12) and (15). For each
model wehavestatedtheequations byconsideringreactions orders
equal to the unity (exponential behaviour) and also, with reaction
orders as different tothe unity. Inthe next Section, we present some
application for each model.
6. Application examples
In this section, present simulations by using our three analytical
models under similar conditions to standard copper heap leaching
process in the north of Chile. We present numerical results for the
following situation: ore porosity c
0
:=0.03, r :=2.5 [cm], effective
pore diffusivity of reagent D
Ae
:=10
6
, bulk solution volume frac-
tion c
b
:=0.03, supercial bulk ow velocity u
s
:=0.000333, heap
depth Z :=3, 6, 9[m]. For all numerical experiments, we use
(, , , Z) =
Z
+
.
The sample of observations corresponds to a standard heap leach-
ing process under operation in the northern part of Chile. We
consider for the experiments, three heaps of 3, 6 and 9 [m]. Let
p=40 and R
p
, p=1, n be the sample of observations for the recu-
peration of each heap. Let
R
p
, p=1, n, the values obtained through
the model. The function to be minimized, which is the sum of the
squared discrete errors, is given by
c:=
p
i=1
(R
i
R
i
)
2
.
Fig. 2. Simple model prediction.
Fig. 3. Second model prediction.
As a performance measure we consider the deviation function,
d:=
_
c
p
.
The Fig. 2 shows the least squares adjustment by using Excel Solver
tool
2
. Some guess values of the parameters k
0
, w, , and , with
Z =6, were used as initial values in the iterative process for the esti-
mation of the parameters in the rst model (3), which requires to
nd 5 independent parameters (see Fig. 2). The estimated values
for that model were used to adjust the second model that requires
to nd 7 independent parameters (see Fig. 3). Finally, these last
estimations were used to nd the 8 independent parameters for
the third model (see Fig. 4). Due to the scalability of the models,
in each case the parameter estimations were done just with the
data corresponding withZ =6andthe interpolations andextrapola-
tions were made by preserving the same set of parameters. Finally,
for each model we report the sum of the squared errors SSE, the
standard deviation o =(SSE/n) and the determination coefcient
R
2
= [(R(t),
R(t))]
2
. The Fig. 2 shows the scalability of the model
(3). We strongly recall the behaviour at innity is no 100% as the
pure phenomenological models predicts.
Fig. 2 shows the results for the three heights where =91.76,
=0.03, =0.101, k =1999.47 and
k = 0.000462. This adjustment
presents SSE=94.8, o =1.540 and
R
2
= 99.75%, i.e., the model (3)
explains the 99.75% of the variability produced in the process.
Fig. 3 shows the results for the three heights where, =90.00,
=0.03, =0.091, k
0
=1976.63, w = 0.000462, k
z
=0.043486 and
2
Microsoft Excel uses the non-linear optimization code GRG2, Leon Lasdon, Uni-
versity of Texas, Austin and Alan Waren, Cleveland State University.
M.E. Mellado et al. / Computers and Chemical Engineering 35 (2011) 220225 225
Fig. 4. Third model prediction.
z=0.99999. With respect to the model (3), it has been reduced the
sum of squared errors to SSE=90.7 and the standard deviation to
o =1.506. The determination coefcient is almost not affected
R
2
=
99.74%. This shows that the model explain 99.74% of the variability
in the recovery. It is important to mention that the reduction in the
variability implies more reliable estimates where, the inclusion of
the second exponential is the key point.
Finally, Fig. 4 shows the results for three heights, where the esti-
mations corresponds to =90.00, =0.03, =0.091, k
0
=1976.63,
w = 0.000461, k
z
=0.043471, z=0.99999 and 1=0.00861. It is
important to remark that, although the model (8) works in a very
goodmanner, themodel (12) reduces evenmorethesumof squared
errors SSE=86.9 and the standard deviation o =1.474 (although
does not affect the determination coefcient
R
2
= 99.74%, explain-
ing the 99.74% of the variability in the recovery). In this case,
the variability reduction implies reliable estimates than the one
obtained with the other two previous models. This is almost
entirely explained as far one note that there is not included the
initial condition for the delay time giving to the model one degree
of freedom which benets the optimization procedure. Finally, we
conclude remarking that all three models behaves very good, and
one canthinkthat is enoughwiththe rst model. The reasontocon-
sider the three models here is that our industrial data is extremely
dominant intheheapandnot at aparticlelevel. Theresults arecom-
pletely different, for instance, ROM heap leaching operations. We
can cite Mellado et al. (2009), for other proofs in different operation
conditions for instance, the radius of the particles.
7. Concluding remarks
A scale-up methodology for the heap leaching modelling has
been proposed. The mathematical theory is consistent and show
that one can scale-up by parts several kinds of models. The pre-
sented three models were developed by using the combination of
an underlying Bernoulli-type model together with phenomenolog-
ical relations in order to obtain analytical models able to scale-up
the heap leaching process. The results in predicting and scaling
the recovery under rather big changes in the height are good
enough to consider these models for the analysis, design, control
and optimization of the heap leaching process. Usually, in industry
applications, one have to deal with a number of heaps. The heap
leaching planning, to reduce costs and increase utilities, must be
done with optimization techniques. In this sense, the analytical
models can be used because they are simple but enough accurate.
Moreover, depending on the choice of the optimization algorithm,
the model structure is a key point to be considered because a not
suitable choice of the model structure can lead to non global opti-
mal solutions.
Acknowledgments
Mario E. Mellado, Luis A. Cisternas and Edelmira D. Glvez
wish to thank CONICYT for nancial support, through Fondecyt
Project 1090406. Mara P. Casanova wishes to thank Universi-
dad de Concepcin for nancial support, through DIUC Project
208.014.016-1.0.
References
de Andrade Lima, L. R. P. (2004). A mathematical model for isothermal heap and
column leaching. Brazilian Journal of Chemical Engineering, 21, 435447.
Bouffard, S. C., & Dixon, D. G. (2001). Investigative study into the hydrodynamics of
heapleaching processes. Metallurgical andMaterial Transactions B, 32B, 763776.
Bouffard, S. C., & Dixon, D. G. (2006). Evaluation of kinetic and diffusion phenomena
in cyanide leaching of crushed and run-of-mine gold ores. Hydrometallurgy, 86,
6371.
Bouffard, S. C., & Dixon, D. G. (2009). Modeling pyrite bioleaching in isothermal test
columns with the HeapSim model. Hydrometallurgy, 95(34), 215226.
Cross, M., Bennett, C. R., Croft, T. N., McBride, D., & Gebhardt, J. E. (2006). Computa-
tional modeling of reactive multi-phase ows in porous media: Applications to
metals extraction and environmental recovery processes. Minerals Engineering,
19, 10981108.
Dixon, D. G., &Hendrix, J. L. (1993a). A mathematical model for heap leaching of one
or more solid reactants from porous ore pellets. Metallurgical Transactions, 24B,
10871102.
Dixon, D. G., &Hendrix, J. L. (1993b). General model for leaching of one or more solid
reactants from porous ore pellets. Metallurgical Transactions, 24B, 157168.
Lizama, H. M., Harlamovs, J. R., McKay, D. J., & Dai, Z. (2005). Heap leaching kinetics
are proportional to the irrigation rate divided by heap height. Minerals Engineer-
ing, 18, 623630.
Mellado, M. E., & Cisternas, L. A. (2008). An analyticalnumerical method for solving
heap leaching of one or more solid reactants fromporous pellets. Computers and
Chemical Engineering, 32(10), 23952402.
Mellado, M. E., Cisternas, L. A., & Glvez, E. D. (2009). An analytical approach to heap
leaching. Hydrometallurgy, 95(12), 3338.
Mousavi, S. M., Jafari, A., Yaghmaei, S., Vossoughi, M., & Sarkomaa, P. (2006). Com-
puter simulation of uid motion in a porous bed using a volume of uid method:
Application in heap leaching. Minerals Engineering, 19, 10771083.
Petersen, J., &Dixon, D. G. (2007). Modelling zinc heap bioleaching. Hydrometallurgy,
85, 127143.
Reverberi, A. P., Esposito, A., & Vegli, F. (2002). Solid dissolution in a batch process
with intraparticle diffusion and lumped kinetics. Chemical Engineering Science,
57(17), 37333744.
Roman, R. J., Benner, B. R., & Becker, G. W. (1974). Diffusion model for heap leaching
and its application to scale-up. AIME Transactions, 256, 247252.
Sheikhzadeh, G. A., Mehrabian, M. A., Mansouri, S. H., & Sarra, A. (2005). Compu-
tational modelling of unsaturated ow of liquid in heap leaching using the
results of column tests to calibrate the model. International Journal of Heat and
Mass Transfer, 48, 279292.
Sidborn, M., Casas, J., Martinez, J., & Moreno, L. (2003). Two-dimensional dynamic
model of a copper sulphide ore bed. Hydrometallurgy, 71, 6774.
Valencia, J. A., Mndez, D. A., Cueto, J. Y., &Cisternas, L. A. (2007). Saltpeter extraction
and modelling of caliche mineral heap leaching. Hydrometallurgy, 90, 103114.
Wu, Ai-xiang., Liu, Jin-zhi., & Tang, Ling-yan. (2007). Simulation of coupled owing-
reaction-deformation with mass transfer in heap leaching processes. Applied
Mathematics and Mechanics, 28, 327335.