You are on page 1of 11

Analytica Chimica Acta 367 (1998) 123133

Modelling and optimization of cyanidation process.


A chemometric approach by regression analysis
Zhaoheng Fang1, Milan Meloun2, Mamoun Muhammed*
Department of Inorganic Chemistry, Royal Institute of Technology, S-100 44 Stockholm, Sweden

Received 15 January 1997; received in revised form 21 November 1997; accepted 16 December 1997

Abstract

A process of direct leaching of gold and silver from a sulde concentrate by cyanidation at ambient temperature was analyzed
by simulation and optimization, using a chemometrical approach based on non-linear regression. A set of models, describing
the inuence of the various relevant parameters on the dissolution process, including the initial concentration of sodium
cyanide, the oxygen partial pressure and the leaching time as the controllable variables in the process, were obtained based on
the kinetic study and mathematical proof of the statistical procedure. The parameters in the set of models were estimated from
the experimental data. Finally, the model was used to determine the optimum conditions for the leaching of gold and silver by
direct cyanidation. # 1998 Elsevier Science B.V. All rights reserved.

Keywords: Cyanidation process; Chemometric approach; Regression analysis

1. Introduction parameters that affect the rate (kinetics) and the extent
(yield) of the leaching. Laborious studies are then
Hydrometallurgical processes consist of several undertaken in order to estimate the `optimal' leaching
unit operations of which leaching is, in general, of conditions.
decisive importance for the economy of the integrated We have previously undertaken a detailed study for
process. This is especially true for the recovery of the leaching of gold and silver by cyanidation of a
precious metals by cyanide leaching. The dissolution copper concentrate [1]. The experimental data were
of metals is brought about by contacting the ore treated in order to determine the dependence of the
minerals with solutions with a given composition leaching process on the operating conditions studied
under suitable conditions and sufcient period of time. and thus to establish optimum operating conditions.
Therefore, there are several physical and chemical In searching for the optimal conditions in a leaching
step of a hydrometallurgical process, the economic
protability should be considered as the main criter-
*Corresponding author. Fax: 0046 8 7909072.
1 ion. In this paper, we present a mathematical model
On leave from Institute of Chemical Metallurgy, Beijing, PR
China. and a statistical analysis of the experimental data by
2
On leave from College of Chemical Technology, Purdubice, studying the system (surface) response to different
Czechoslovakia. variables (operating parameters). Several models were

0003-2670/98/$19.00 # 1998 Elsevier Science B.V. All rights reserved.


PII S0003-2670(98)00020-8
124 Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133

proposed for different operating conditions of the sodium cyanide. The experimental data expressing
cyanidation process and were analyzed assuming that selected conditions of the process are presented in
the best model represents an optimum for the process Table 1.
operation.
For the optimization of the economic protability, 3.1. Regression data treatment
knowledge of a simple economic quantity, namely the
difference between the two major contributions: the At the stage of process development, the regression
value of the produced metals and the costs for leaching analysis of date {y, x} using kinetic equations yf(x, b)
process is required. The comparison between process to estimate parameters b is carried out according to the
alternatives is based on the optimum prot in each following steps:
case. The income depends mainly on the recovery of
1st step: Derivation of the initial hypothesis of
valuable metals, i.e. the yield of gold and silver in the
mathematical model to be proposed.
cyanidation of the AuCu sulphide concentrate. In
2nd step: Regression model building and testing by
general, the consumption of the reacting reagents,
the model fitting through the given data.
impure components in pregnant liquors and the oper-
3rd step: A search for a reliable computational
ating cost per unit time are the most important and
strategy by the use of simulation data.
determining parts of the process cost.
4th step: Proposed models of leaching process and
The goal of process modelling by a data simulation
an evaluation of experimental data.
is to nd a set of regression models that describe the
whole process and to estimate the unknown para-
3.1.1. Initial hypothesis of model proposed
meters of these models. Recently, our attention has
For the cyanidation process of the AuCu concen-
been focused on the regression modelling because the
trate, the variables describing the process are the
computer data simulation is a widely used tool in
extracted fraction of gold and silver, the content of
regression analysis. The mathematical model pro-
copper and iron dissolved in pregnant liquors, and the
posed describes the process and expresses the quanti-
consumption of cyanide. The results of the experi-
tative relation between variables. Therefore, our
mental kinetic study were used for deriving the initial
modelling, by data simulation, will also contain an
hypothesis of proposed models concerning the leach-
examination of correlation between the dependent
ing process with the following features and relations
variables representing the process (such as the yield
[1]:
of the valuable metals, the content of foreign compo-
nents (impurity), leaching time, and the reagent con- (a) The initial rate of gold dissolution is propor-
sumption) with individual controllable independent tional to the initial concentration of cyanide and
variables of the process conditions. the partial pressure of oxygen for low values, but
turns out to be independent at higher values.
(b) Dissolution of silver consists of two reaction.
2. Experimental The rate of first one is rather sensitive to the partial
pressure of oxygen, and little linked to the
The laboratory leaching was carried out in a glass concentration of cyanide while the rate of the
vessel of 500 cm3, the particle size of the concentrate second reaction is very slow and almost indepen-
was 120 mesh, and the stirring rate was kept at dent of the leaching time.
700 rpm. The reaction vessel was kept at 258C and (c) The kinetics of dissolution of copper is similar
the leaching time was varied. The detailed experi- to that of gold. The rate is a linear function of the
mental conditions are given elsewhere [1]. initial concentration of cyanide on the leaching
time.
3. Results (d) For the dissolution of iron, strong aeration
accelerates the precipitation reaction of the hydro-
The leaching was carried out either at a xed partial xide and causes the decrease of the concentration
pressure of oxygen or at a xed initial concentration of of ferrous ions. The reaction rate is proportional to
Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133 125

Table 1
Matrix of experimental data for leaching process containing the controlled independent variables x1i, x2i, x3i, i1,2. . .46, and measured
dependent variables yji, i1,2. . .46 for five proposed models, j1. . .5

Independent variables Dependent variables of proposed model

i x1i a x2i b x3i c y1i d y2i e y3i f y4i g y5i h


PO2 (NaCN)l t NaCN (Cu) (Fe) XAu XAg
consumed (g/l) (g/l) leached leached
1 0.01 4.0 2.0 1.92 0.312 0.113 0.33 0.013
2 0.01 4.0 3.0 2.66 0.323 0.122 0.36 0.005
3 0.01 4.0 4.0 2.96 0.336 0.137 0.41 0.000
4 0.01 4.0 6.0 3.24 0.357 0.146 0.45 0.008
5 0.01 4.0 10.0 3.52 0.385 0.158 0.50 0.044
6 0.2 1.0 2.0 i 0.224 0.020 0.56 0.088
7 0.2 1.0 3.0 i 0.236 0.024 0.63 0.134
8 0.2 1.0 4.0 i 0.244 0.027 0.64 0.156
9 0.2 1.0 6.0 0.90 0.250 0.030 0.64 0.194
10 0.2 1.0 10.0 i 0.258 0.033 0.64 0.218
11 0.2 2.0 2.0 i 0.322 0.055 0.68 0.190
12 0.2 0.2 3.0 i 0.351 0.063 0.73 0.197
13 0.2 2.0 4.0 i 0.373 0.069 0.78 0.202
14 0.2 2.0 6.0 1.79 0.381 0.076 0.81 0.217
15 0.2 2.0 10.0 i 0.413 0.078 0.84 0.245
16 0.2 3.0 2.0 i 0.386 0.090 0.70 0.212
17 0.2 3.0 3.0 i 0.416 0.098 0.75 0.219
18 0.2 3.0 4.0 i 0.455 0.105 0.80 0.215
19 0.2 3.0 6.0 2.66 0.486 0.111 0.88 0.224
20 0.2 3.0 8.0 i 0.504 i i i
21 0.2 3.0 10.0 i 0.510 0.113 0.91 0.269
22 0.2 4.0 2.0 2.23 0.346 0.115 0.74 0.239
23 0.2 4.0 3.0 2.96 0.400 0.127 0.79 0.232
24 0.2 4.0 4.0 3.15 0.432 0.138 0.83 0.229
25 0.2 4.0 6.0 3.55 0.517 0.147 0.90 0.237
26 0.2 4.0 10.0 3.79 0.582 0.150 0.94 0.266
27 0.2 5.0 2.0 i i i 0.76 0.252
28 0.2 5.0 3.0 i i i 0.80 0.245
29 0.2 5.0 4.0 i i i 0.87 0.235
30 0.2 5.0 6.0 4.52 0.625 0.162 0.90 0.238
31 0.2 5.0 10.0 i i i 0.92 0.285
32 0.2 6.0 2.0 i i i 0.81 0.255
33 0.2 6.0 3.0 i i i 0.84 0.248
34 0.2 6.0 4.0 i i i 0.89 0.242
35 0.2 6.0 6.0 5.40 0.674 0.194 0.92 0.245
36 0.2 6.0 10.0 i i i 0.94 0.227
37 0.5 4.0 2.0 2.84 0.553 0.123 0.88 0.304
38 0.5 4.0 3.0 3.22 0.599 0.131 0.91 0.287
39 0.5 4.0 4.0 3.33 0.636 0.139 0.92 0.289
40 0.5 4.0 6.0 3.60 0.700 0.132 0.94 0.328
41 0.5 4.0 10.0 3.71 0.732 0.122 0.91 0.374
42 1.0 4.0 2.0 3.10 0.548 0.139 0.91 0.322
43 1.0 4.0 3.0 3.43 0.603 0.145 0.92 0.308
44 1.0 4.0 4.0 3.64 0.668 0.152 0.94 0.306
126 Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133

Table 1
(continued)

Independent variables Dependent variables of proposed model


a b c
i x1i x2i x3i y1i d y2i e y3i f y4i g y5i h
PO2 (NaCN)l t NaCN (Cu) (Fe) XAu XAg
consumed (g/l) (g/l) leached leached
45 1.0 4.0 6.0 3.79 0.714 0.145 0.95 0.351
46 1.0 4.0 10.0 3.97 0.783 0.140 0.95 0.403
a
The partial pressure of oxygen (atm).
b
The initial concentration of NaCN (g l1).
c
The leaching time (h).
d
The consumption of NaCN (g l1).
e
The copper concentration (g l1) in leaching solutions.
f
The iron concentration (g l1) in leaching solutions.
g
The extracted fraction of gold.
h
The extracted fraction of silver.
i
Missing quantities.

the initial concentration of cyanide and seems to be study and empirical model is
independent on the partial pressure of oxygen after
y a f x1 ; x2 expbx3 (1)
leaching for 2 h.
(e) The consumption of cyanide is correlated with where y stands for a dependent process variable, x1, x2,
the partial pressure of oxygen and also with the x3 are three independent process variables described
concentration of cyanide. above, b the unknown parameters to be estimated, a
(f) The leaching is done under conditions where no the constant and f(x1, x2) the function of two process
influence of stirring speed, particle size of the variables usually expressed as a product of polynomial
concentrate on the yield is observed and therefore of variables. According to the experimental strategy,
their influence was omitted here. ve various models may be proposed differing in
dependent variable, yj, j1. . .5, and in formulation
Therefore, the partial pressure of oxygen, the initial of the model proposed.
concentration of sodium cyanide and the leaching time
are considered to be the main controllable factors of
process kinetics and may be taken as the independent 3.1.2. Model building and testing
variables, x1, x2 and x3. The dependent variables, The data matrix {X, y} containing {xji, j1. . .3,
desired to specify the process itself, distinguishes five i1. . .n; yji, j1. . .5, i1. . .n} is tted by the
various models: y1 is the consumption of sodium initial hypothesis of model containing a vector of
cyanide, y2 the copper concentration in leaching solu- unknown parameters b{b1. . .bm}. To nd the
tions, y3 the iron concentration in leaching solutions, `best' model from several possible or proposed
y4 the extracted fraction of gold, and y5 the extracted ones, the residual-square sum function U is to be
fraction of silver. minimized,
To describe a leaching process, the most important X
n
model equations are those which describe a develop- U yji f x1i ; x2i ; x3i ; b2
ing period of the process at the beginning. The initial i1
hypothesis of the proposed models for the leaching X
n

process is derived on the base of process kinetics and a e2i  minimum (2)
i1
resulting model is then modied according to the
procedure of regression model building. A general for each of ve model hypotheses, j1. . .5. When the
phenomenological expression on the basis of a kinetic U function reaches a minimum and the unknown
Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133 127

parameters b{b1. . .bm} are found, two tests are the ill-conditional parameters cause a rather at shape
carried out: of the hyperparaboloid U in (m1)-dimensional space
and therefore their estimation is uncertain. The ill-
(i) The first test makes a statistical analysis of
conditional parameters usually exhibit quite large
residuals (e) to examine the quality of achieved
standard deviation s(bi) and are seldom a matter
fitness of the regression curve through data points.
of estimation. It is recommended to keep these
The calculated regression diagnostics usually
parameters constant on the xed value of initial
contain the following statistical characteristics
guess.
of the
P residual set: the residual meanPE^e Either tests or an examination of conditioning
1=n ni1 ^ei ; the mean residual jej 1=n ni1 j^ej;
parameters in the model enable a modication of
the residual standard deviation
the initial hypothesis of the proposed model. After
!1=2
X n
ei e2 testing all parameter estimates and leaving out insig-
s^e (3) nicant parameters (as they are equal to zero), the
n1
i1 proposed model of a leaching process for a given data
and two parameters describing the shape of the set can be simplied. For example, the initial hypoth-
residual distribution, i.e. the skewness g1(e) and esis of the model proposed for gold yield (Model (4))
the curtosis g2(e). The goodness-of-fit may be contains nine parameters but four of them may be left
considered not only according to the measures out from estimation
of location and dispersion but also according to 5 x1i
some measures of distribution shape. The best fit Model 4 : y4 1
1 2 x1i
is achieved when the residual distribution is of
Gaussian normal nature. Examination of an 1 6 x2i 1 7 x3i
 (4)
actual residual distribution is carried out by the 1 8 1 3 x2i 1 4 x3i 9 x23i
skewness g1(e), the curtosis g2(e) and Pearson
c2-test. For normal distribution it should be valid with the initial guess of parameters for the rst
that g1(e)0 and g2(e)3 while c2exp c21a m, hypothesis of model proposed
where c2exp is the test criterion of Pearson c2-test b1 0:26  0:06;
and c21a m is the quantile of c2-distribution;
(for details cf. Ref. [3]). Regression diagnostics b2 27:2  7:1;
are the only criteria according to which a
b3 1:6  0:3;
selection among several proposed models may
be performed because the `best' model always b4 2:2  0:4;
reaches the best curve fitting through a given b5 1:2  0:5;
data set.
(ii) The second test concerns the statistical b6 0:15  0:06;
significance of the parameter estimates b j , b7 102  87;
j1. . .m. The parameter j is significant in b8 0:20  0:8;
the model if its estimate bj is greater than three
times its standard deviation, bj>3 sj, i.e. `3 sigma' b9 0:0025  0:0015
rule. Parameter estimates bj, not fulfilling this rule, and rened values of parameters for the simplied
are then taken as being equal to zero and the nal hypothesis of model proposed
parameter j is considered to be insignificant in the
model. b1 0:28  0:01;
Parameters in a model may be distinguished as b2 24:2  2:2;
well-conditioned parameters and ill-conditioned ones.
While the well-conditioned parameters can be pre- b3 1:5  0:2;
cisely estimated by most of the regression programs, b4 2:2  0:4;
128 Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133

Table 2
The accuracy expressed by the relative derivation rel,i and the precision s(bi) of parameter estimates for simplified Model (4) and simulated
data in dependence on the instrumental error, Sinst(y): b10.28, b224.2, b31.5, b42.2, b51.0, b60.15, b7100, b80.2 and b90.0025

Set sinst(y) b1 b2 b3 b4

rel,1 s(b1) rel,2 s(b2) rel,3 s(b3) rel,4 s(b4)


5 8 5 6 5 7 5
1 0.000 1.010 2.010 1.610 4.010 2.710 4.010 2.310 5.0107
2 0.005 0.145 0.0022 1.26 0.33 0.56 0.031 0.52 0.048
3 0.025 0.651 0.0121 6.93 1.75 4.08 0.163 13.10 0.306
4 0.050 7.67 0.0211 7.13 2.88 9.99 0.320 7.3 0.402
5 0.100 8.79 0.0396 8.34 6.90 37.87 0.849 28.16 0.611

while other ve parameters are kept at constant values parameter estimation was examined in dependence
of initial guess; on the instrumental error sinst(y). For increasing
value of sinst(y) starting from 0, and 0.005 through
5 1:000;
0.100, the relative deviation of individual parameter
6 0:150; estimates, Drel,i, (accuracy) and the parametric stan-
7 100:0; dard deviations, s(bi), (precision) are presented in
Table 2.
8 0:200;
For known value of an instrumental error sinst(y)
and achieved by the law of errors propagation, the preci-
sion and accuracy of parameter estimates may be
9 0:00250:
guessed on the base of results from Table 2.

3.1.3. A search of computational strategy 3.1.4. Proposed models of leaching process


Examining reliability, i.e. accuracy and precision of Regression analysis of simulated data and the pro-
parameter estimates bi, at rst, a search for optional cedure of a model building and testing lead to the nal
computational strategy with the use of simulated data forms of some models proposed for the process of
is carried out. For a given set of parameters bj, cyanidation of AuCu concentrate,
j1. . .m, the `theoretical points' are calculated using
Model (1):
the model proposed. Each theoretical point is then
transformed into the simulated one by the addition of a 1 10x1i
y1 1 x2i exp 3 x3i 0:7 (6)
random error ei, yiyth,iei. The set of random errors e 1 b2 x1i
exhibits Gaussian normal distribution and has been
calculated with the use of a random number generator Model (2):
and a known value of the instrumental error sinst(y). 1 b2 x1i
The resulting set of simulated data is quite close to the y 2 1 1 3 x2i 1 4 exp0:25x3i
1 b5 x1i
set of real experimental data. (7)
The accuracy may be classied by the relative
deviation of the parameter estimate from the true Model (3):
value of parameter question,
y3 1 1 20x2i 1 2 x3i (8)
i bi
rel;i  100% (5) Model (4):
i
while the precision of each parameter estimate is x1i
y 4 1
considered according to the parametric standard 1 2 x1i
deviation s(bi). For gold yield described by the sim- 1 0:15x2i 1 100x3i
 (9)
plied Model (4), for example, the reliability of 1 0:21 3 x2i 1 4 x3i 0:0025x23i
Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133 129

Table 3
Parameter estimates and goodness-of-fit test by regression analysis of experimental data from Table 1 carried out with the STATGRAPHICS
program [4]

Parameter estimates

Model(1) Model(2) Model(3) Model(4) Model(5)


b1s(b1) 2.1660.076 0.14820.0149 0.0015180.000062 0.27630.147 7.6770.799
b2s(b2) 8.2520.288 7.3321.587 0.018940.00583 24.172.23 2.0450.176
b3s(b3) 0.0088640.00144 0.40890.0508 1.5060.206 0.047890.00659
b4s(b4) 0.50720.0719 2.5060.318
b5s(b5) 2.9530.634

Goodness-of-fit test
Data size n 25 37 37 45 45
Residual mean E(e) 0.139 0.0067 1.29E-3 0.00051 0.00049
Mean residual oeo 0.058 0.0049 0.0014 0.0048 0.0028
Standard deviation s(e) 0.288 0.030 0.0084 0.032 0.019
Skewness g1(e) 2.16 0.812 0.326 0.065 1.41
Curtosis g2(e)-3 5.49 0.167 0.019 0.127 4.42
2exp 16.0 7.76 4.76 3.18 14.8
21a (d.f.) 16.7 9.21 9.21 6.25 14.9

Model (5): 3.2. Optimization of the leaching


process
0:8x1i 7x2i 1
y5 1 3 x3i expx3i
1 1 x1i 1 2 x2i The principle for the optimization of a process is the
(10) maximum of its economic protability. For the pro-
where bi are unknown parameters to be estimated by cess under study, the signicant factor is the difference
regression analysis of experimental data from Table 1. between the income and cost of process operation. In
Found parameter estimates and a proof of their relia- cyanidation of the AuCu concentrate, the income
bility by statistics of a goodness-of-fit test are given in only includes the value of gold and silver recovery.
Table 3. The cost could include many terms. Among
The 2 statistic used in this case in dened by them, the most important part for the evaluation of
[2] a process is the cost of direct operation of the
process, i.e. the reagent consumption for the leaching
X
k
Oi Ei 2 process and further renement owing to removal of the
2exp (11) impurities entered in the process. Therefore, it is
i1
Ei
reasonable to suppose that the cost for the removal
where Oi is the observed residual frequency, Ei the of copper and iron from the pregnant liquors depends
expected residual frequency, k the number of classes, on their content in the liquors, whereas operating
and k1 the freedom degrees. The computed 2 value cost depends mainly on the operating time of the
is 3.18. Using the Pearson's 2 test to examine that process. The other costs may be considered as a
the residual set comes from a normal distribution, constant.
the degrees of freedom is decreased by c, where c is In estimating the economic protability, the follow-
the number of parameters estimated from the data. ing basic value are considered:
In this case, the total 2 has (81)43 degrees of
freedom. Thus for Model (4) the value 23;0:90 6:25 Gold 100 SEK/g
is obtained, and we do not reject the model expression Silver 5 SEK/g
for gold yield. For all ve models, the 2 test proved a NaCN 1213 SEK/kg
Gaussian normal distribution of residuals. where SEK is the Swedish currency,
130 Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133

Fig. 1. Contours of the process optimization model (from Eq. (13), i0.25).

and suppose: called the simplied benet. In the case of cyanidation


under air bubbling (oxygen partial pressure0.2 atm),
1. The cost for removal of 1 kg copper or iron from
the contours of simplied protability I1 calculated
1 m3 pregnant liquor is j% of the total value of
from Eq. (12) are shown in Figs. 13, respectively for
gold.
i0.5, 1.0 and 2.0. The area of the highest value
2. The operating costs in 1 h equals to i% of total
indicates the optimum operation conditions. It can
value of the gold.
be seen that the optimum initial concentration of
To keep it simple, the case in which the costs for the NaCN depends on the consumption of cyanide. In
removal of copper and iron are omitted is discussed the case of our experimental conditions, the optimum
first. The economic profitability for the treatment of initial concentration is 4.55 g/l of NaCN. The opti-
one ton of the AuCu concentrate can be estimated by mum operating time is sensitive to the operation cost
the equation: in unit time. With an increase of the operating costs
from 0.5 to 2% the optimum leaching time decreases
B 1000 500 y4 i%0 x3 50 150y5 120 10y1 from 8 to 4 h.
A SEK 50000 y4 0:15y5 0:024y1 So far the costs for the removal of copper and
i%x3 A 5000I1 A (12) iron from leached liquors is concerned, the prot-
ability can be estimated with the following equa-
where tion:
I1 y4 0:15y5 0:024y1 i%x3 (13) B 1000 50y4 i%x3 j%y2 y3 50 150y5
where x3 is the leach time, h; and A the constant, 120 10y1 A 5000y4 0:15y5 0:024y1
representing all other costs. i%x3 j%y2 y3 A 5000I2 A (14)
The basic operation conditions included in the
derivation of the equation are: the ratio of solid and where I2y40.15y50.024y1i%x3j%(y2y3).
liquor is 1 : 10, the grade of gold and silver in the Similarly, Eq. (14) is the base of the process
concentrate are 50 and 150 mg l1, respectively. optimization. The calculated results are given with
Since Eqs. (12) and (13) are mathematically the contours at several groups of i and j (Figs. 46).
equivalent in maximization, the Eq. (13) is considered The optimum initial concentration of NaCN depends
as the basic function for process optimization, and I1 is mainly on the consumption of cyanide, and alters
Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133 131

Fig. 2. Contours of the process optimization model (from Eq. (13), i1).

Fig. 3. Contours of the process optimization model (from Eq. (13), i2).

little due to the change of the cost of copper and 4. Conclusions


iron removal (e.g. j2 to 6), and the best value is
ca. 3.5 g/l. The optimum leaching time is also The following conclusions can be drawn:
sensitive to the operating cost in unit operating
time. 1. The cyanidation process of the AuCu sulphide
Eq. (13) or Eq. (14) can be used to predict the concentrate can be described by a set of simula-
optimum area of process operation under pure oxygen tion models for the yield of gold and silver, the
bubbling. The optimum area is little different to that concentration of copper and iron in liquors as well
obtained under air bubbling. as the consumption of cyanide with the initial
132 Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133

Fig. 4. Contours of the process optimization model (from Eq. (14), i0.25, j2).

Fig. 5. Contours of the process optimization model (from Eq. (14), i0.5, j1).

concentration of sodium cyanide, the oxygen process is worked out. Operating conditions have
partial pressure and operating time as the been estimated from the model. The concentration
controlled variables of the process. Reliability of of sodium cyanide is the main factor affecting the
the models proposed are proven mathematically process optimization, and its optimized initial con-
by a statistical procedure, and parameters are centration ranges from 3.5 to 4 g/l under aeration
estimated by regression analysis and examined by with compressed air. The content of iron and
a goodness-of-t test. copper in pregnant liquors, considered as impuri-
2. The process optimization has been performed on ties and removed in later refining steps, has less
the basis of an estimation of the process economic effect on the optimization of the process profit-
profitability, and the optimization model for the ability under our fundamental assumptions. The
Z. Fang et al. / Analytica Chimica Acta 367 (1998) 123133 133

Fig. 6. Contours of the process optimization model (from Eq. (14), i2.5, j2).

optimum leaching time is sensitive to the operating [2] R.M. Bethen, Statistical methods for engineers and scientists,
John Wiley and Sons, New York, 1985, p. 75.
costs in unit time.
[3] M.A. Sharaf, D.L. Illman, B.R. Kowalski, Chemometrics.,
3. The proposed optimization function can be used to John Wiley and Sons, New York, 1986, pp. 179296, Chap. 6.
predict the optimum operation conditions that have [4] Statgraphics, A statistical graphics system, Ver 3.0. Statistical
not been investigated in detail. Graphics Corporation, STSC. Incorp. USA.

References

[1] Z. Fang, M. Muhammed, Leaching of precious metals from


complex sulphide ores. A kinetic study on cyanide leaching of
gold and silver, to be published.

You might also like