Chemical Engineering Journal 284 (2016) 294304

Contents lists available at ScienceDirect

Chemical Engineering Journal

journal homepage: www.elsevier.com/locate/cej

Evaluation strategy for the control of the copper removal process based
on oxidationreduction potential
Bin Zhang, Chunhua Yang, Hongqiu Zhu , Yonggang Li, Weihua Gui
School of Information Science and Engineering, Central South University, Changsha, China

h i g h l i g h t s

A mechanistic model to estimate the relations between ORP and copper concentration is established.

A fuzzy process evaluation method based on the mechanistic model is proposed.

A control strategy for copper removal process is established integrated with the process evaluation method.

The effectiveness of the proposed evaluation method and control strategy is confirmed through industrial experiments.

a r t i c l e

i n f o

Article history:
Received 8 March 2015
Received in revised form 19 July 2015
Accepted 31 July 2015
Available online 6 September 2015
Keywords:
Process evaluation
Fuzzy logic
Copper removal
Oxidation reduction potential (ORP)
Control design

a b s t r a c t
The copper removal process purifies copper from leaching solutions with zinc powder in reactors. Due to
the complex reaction mechanism and unavailability of online measurements, zinc powder is usually
added inexactly, which easily leads to unstable production. This paper proposes an online evaluation
method based on oxidationreduction potential (ORP) for the control of the copper removal process. A
kinetic model is designed to translate the production requirements to evaluation indexes of ORP, and
the process is then graded by evaluating the fuzzified ORP and its trends according to these indexes.
By analyzing these evaluation grades, the process condition is divided into several classes, and each condition class corresponds to a control method set. The industrial experiments show that the copper
removal performance is improved by using the proposed evaluation and control strategy.
2015 Elsevier B.V. All rights reserved.

1. Introduction
Zinc hydrometallurgy is the main method of zinc production,
producing more than 80% of the worlds zinc [1]. This process is
divided into five steps: roasting, calcine leaching, solution purification, electrowinning and melting [2]. The leached zinc sulfate solution contains various impurities (e.g. copper, cobalt, nickel, and
cadmium) which lower the current efficiency and reduce the quality of the zinc ingot [3]. Therefore, the impurities need to be purified to an acceptable amount before electrowinning. Copper is
commonly removed by zinc powder in the first stage of purification process because it has a more negative oxidation potential
than the other impurities [4,5]. The purpose of copper removal process is to decrease the amount of copper ions in the leaching solution and reserve a portion of copper ions as activators used in the
cobalt removal stage. Copper removal is usually carried out in a
series of continuous stirred tank reactors (CSTRs), where ionic cop Corresponding author. Tel./fax: +86 731 88836876.
E-mail address: hqcsu@mail.csu.edu.cn (H. Zhu).
http://dx.doi.org/10.1016/j.cej.2015.07.094
1385-8947/ 2015 Elsevier B.V. All rights reserved.

per is deposited as metallic copper by adding zinc powder

(CuSO4 + Zn ? ZnSO4 + Cu). In some improved processes, a portion
of the deposited metallic copper is returned to the first reactor,
where the ionic copper goes on to react with metallic copper and
is precipitated as cuprous oxide (Cu + CuSO4 + H2O ? Cu2O +
H2SO4) [6].
These two reactions are mutually competitive as they consume
the same reactant, and the former reaction can promote the latter
one because the product of one reaction is the reactant of the other.
Due to the interactions between the reactions, the copper removal
process is sensitive to the reaction conditions [7]. Meanwhile, realtime concentration measurements are unavailable in the copper
removal process because on-line element analyzers are costly,
require complex maintenance, and are of questionable reliability
[8]. In addition, the variations in the removal reactions are incapable of being determined. Accordingly, excess or insufficient
amounts of zinc powder are usually added, making the entire
purification process unstable.
Recently, many researchers have considered oxidation reduction potential (ORP) for monitoring and controlling the industrial

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

chemical process [911]. ORP reflects the extent of the oxidation

reduction reaction and provides insight into the state of the reaction system that other parameters cannot reveal. Additionally,
ORP sensors are low-cost, available online, and industrially feasible
[12]. Due to these advantages, ORP has been applied in the
monitoring and control systems of many biological and chemical
processes such as biological nitrogen removal, wastewater treatment, and uranium extraction [1315]. Ruano et al. applied fuzzy
logic in the control of the biological nutrient removal process using
ORP and pH sensors [16]. In the control method, the ORP and pH
values are fuzzified, and the amounts of the control variables are
then adjusted based on the fuzzy controllers using the fuzzified
ORP and pH values are the inputs. Yu et al. presented the ORP
potential as one of the key parameters to control the E-Fenton
process for wastewater treatment by applying artificial neural network [17]. Won and Ra used moving slope changes of ORP to identify the condition of real-time control points for the aerobic and
anoxic phases in the wastewater process [18]. These applications
show that ORP has great potential for the industrial chemical processes control and provide useful advice for applying ORP in the
copper removal process.
Notably, in the studies mentioned above, the impact of ORP on
the control variables should be determined exactly before monitoring and controlling these processes. Sun et al. built a mathematical
model by applying the electrode reaction kinetics to determine the
relationships between ORP and the reaction rate in the cobalt
removal process of zinc purification [19]. The model was applied
to predict the cobalt concentration of the outlet purified solution,
which offers an effective way to model the copper removal process
with ORP in this paper. However, this method cannot be applied
directly in the studied process because the copper removal occurs
in the upstream stage of the cobalt removal. The contents of the
impurities in the copper removal stage are considerably higher
than those in the cobalt removal stage. Each of these impurities,
which are mostly unavailable, influences ORP. Consequently, ORP
is able to roughly estimate the copper removal reaction state,
whereas it can hardly estimate the reaction rate exactly.
Fuzzy logic offers an effective solution to solve this problem
with inherent difficulty and uncertainty [20,21]. It can transform
uncertain information and human linguistic data into mathematical formulas and vice versa, and it is also easily understood and
developed. Due to these advantages, fuzzy logic has been widely
applied in the control of industrial processes [2225]. It has been
introduced for solving uncertainty problems in the detection of
multiple combined faults on the online assessment of faulty conditions in an induction motor [26,27]. It has also been applied to con-

295

struct fuzzy logic controllers when the mathematical models of the

controlled objects are undefined or uncertain [28,29]. Considering
these advantages, we introduce the fuzzy logic technique to evaluate the copper removal process for solving the uncertainty problem
in the kinetic model based on ORP. The amount of zinc powder
could then be determined in real time according to the evaluation
results. Therefore, the objectives of this work are as follows: (1) to
build a kinetic model for obtain quantitative acceptable ranges of
ORP with the impurity concentration limits; (2) to propose a fuzzy
evaluation method, based on ORP and its acceptable ranges, for the
copper removal process; and (3) to implement a control strategy
integrating with the evaluation method, thereby improving the
quality of the outlet copper concentration and stabilizing the production process.

2. Materials and methods

2.1. Pilot plant and monitoring system description
The evaluation and control strategy was developed and tested
in a zinc hydrometallurgy plant of China. The flow chart of the copper removal process in the plant is shown in Fig. 1. Zinc sulfate
solution fed into copper removal is pumped from the leaching process. The solution consists of various impurities among which the
content of copper is highest. Copper ions are removed in two fullfilled connected continuous stirred tank reactors (100 m3). The
amounts of zinc powder are added by weight belts. Most of the
copper ions are deposited in the first reactor by the bulk of the zinc
powder, while a small amount of zinc powder is fed into the last
reactor to maintain the outlet copper concentration within the
desired range. After the precipitation reaction is finished (approximately 50 min), the purified solution is sent to a thickener to separate the clean solution and the precipitate. The clean solution
from the top of the thickener is pumped to the next impurity
removal stage, and the part of the precipitate from the bottom of
the thickener is returned to the first reactor. The remaining precipitate is sent to the copper recovery process.
A series of sensors are installed in the copper removal process to
obtain on-line information about the process state. These sensors
are connected to a network system that includes several analog
digital converters (ADC), transmitters, a programmable logic controller (PLC), and an industrial computer. Most of the process
parameters (e.g. temperature, pH, ORP, etc.) are monitored online
every minute. The impurity concentrations in the inlet and outlet
solutions are measured manually every two hours. The amounts

Fig. 1. Flow diagram of copper removal in the zinc purification process.

296

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

Table 1
Average influent and effluent specifications of the copper removal process (over
300 days).
Parameter

Unit

Influent

Effluent

Flow rate of CuSO4 solution

Flow rate of underflow
Temperature
Solid content
pH
Cu2+

m3/h
m3/h
C
g/m3

g/L

180310
2831
6266
1.41.5
3.94.2
0.72.1

0.10.5

the rate of all electrode reactions by affecting the electron transfer

rate between the oxidant and reductant. The actual activation
energies of the anode and cathode reactions are calculated as
Eq. (7) [19]:

Eea E0ea  1  knFemix  eeq

Eec E0ec  knFemix  eeq

of zinc powder added to the reactors were adjusted manually

according with these industrial informations. The influent and
effluent characteristics of the copper removal process are presented in Table 1.

where E0ea and E0ec denote the standard activation energies of the
anode and cathode reactions, respectively, emix and eeq are the mixed
potential and the standard equilibrium potential, respectively, n is
the number of electron moles, F is the Faraday constant, and k is
the weight to describe the influence of electrode potential variations on the activation energies of the anode and cathode reactions.
The relationship between mixed potential and ORP was estimated
using a linear approximation method as [19]:

2.2. Relationship between concentration and ORP

emix peorp  q

The purpose of the copper removal process is to reduce the copper concentration to an acceptable amount by adding zinc powder.
Practically, the operators often assess the process and adjust the
amount of zinc powder according to the outlet copper concentration. However, due to the delayed measurements, the operation
is always delayed, making the process under unstable. Thus, ORP,
which timely and effectively describes the states of the reactions
in the process, is used to monitor and adjust the copper removal
process. Therefore, the relationship between ORP and the outlet
copper concentration should be analyzed before evaluating the
process according to ORP.
Copper removal is composed of two major oxidationreduction
reactions: copper cementation and comproportionation. Both of
these removal reactions are thought to follow first order kinetics
[7]. Therefore, the reactions can be described as:

r i ki C

where ri ; i 1; 2 is the reaction rate of the ith major reaction, C is

the concentration of copper ion, and ki ; i 1; 2 is the rate constant
of the ith major reaction. On the basis of the principle of independence of coexisting chemical reactions and material balance, the
rate of change of the copper concentration is described according
to Eq. (2):

dC
r 1 r 2 k1 k2 C
dt

The Arrhenius equation indicates that the rate constant ki is

affected by the activation energy Ea:i :


Ea:i
ki Ai exp 
RT

where Ai and Ea:i are the pre-exponential factor and activation

energy of the i th major reaction, respectively, R is the gas constant,
and T is the temperature in Kelvin.
Concerning the electrode characteristics of the materials in the
major reactions, the main electrochemical reactions occurring at
electrode surfaces during electrolysis in the anode are:

Zn ! Zn2 2e

E0 0:763 V

In the cathode, the reactions are:

2

Cu

2e ! Cu E0 0:337 V

2Cu2 H2 O 2e ! Cu2 O H

5
E0 0:203 V

According to the independence principle of parallel electrode

reactions, these reactions uniquely share the electrode potential
called the mixed potential, and the mixed potential determines

Then, according to Eqs. (2) and (8), the mathematic relationship

between ORP and copper ion concentration is built as Eq. (9):

ln

Ct
C0





Ee:1 2kFpeorp q  eeq:1
t A1 exp 
RT



Ee:2 kFpeorp q  eeq:2
A2 exp 
RT

Among these parameters, Ee:1 , Ee:2 , k, A1 , A2 , p, and q are

unknown and need to be identified, R, F, eeq:1 , and eeq:2 are constant,
and C 0 , t, and T are determined in the process.
Based on the complexity of the kinetic model and the numbers
of the unknown parameters, particle swarm optimization (PSO)
algorithm was applied to identify these parameters. The objective
of the parameter identification is to minimize the differences
between the predicted outlet copper concentration of the kinetic
model (C kinetic ) and the measured concentration (C real ) with nT test
samples.
T
1X
2
C kinetic k  C real k :
nT k1

min

10

2.3. ORP-based process evaluation

Due to the complexity of the copper removal process and the
influence of other impurities, it is difficult to estimate the outlet
copper concentration accurately based on the kinetic model with
ORP. However, ORP can effectively assess the state of the removal
process on line. For solving uncertainties in the fuzzy nature
between ORP and the process state, fuzzy logic is introduced to
evaluate the copper removal process. The value of ORP presents
the current state of the removal process, while its trend indicates
the underlying future state under the current operation. If only
the current value of ORP is considered in the evaluation process,
an improper evaluation result might be produced, which could
reduce the process stability. For example, when the value of ORP
is higher than the regular range, the process is considered to be
in a serious condition, and more zinc powder will be added to
improve the process condition. However, the operation might
cause the process to enter another extreme condition when there
is a sharp decrease in the copper concentration. In this section,
an evaluation method based on ORP and its trends is proposed
for the copper removal process. In the evaluation process, ORP
and its trends are classified into several fuzzy sets with several
fuzzy membership functions, and the parameters of these functions are defined using the kinetic model with the production limitation of the concentrations. According to the fuzzified ORP and its
trend, the process condition is graded by applying a set of fuzzy

297

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

Table 2
Parameter calculation and fuzzy language definition for ORP.
Label

Meaning

Function

Calculation of parameters

VL

Very low

LL

Little low

a1 gLMIN ; b1 a1  p2 w11
r11 gq1 LMIN 1  q1 LMAX ; w11 p1 r10  a1 ; m 2
r10 gq0 LMIN 1  q0 LMAX ; w10 p0 r11  r11 ; m 2
r11 gq1 LMIN 1  q1 LMAX ; w11 p1 d1  r10 ; m 2
d gLMAX ; c1 d p2 w11

Stable

LH

Little high

u x
l13 x
l10 x
l11 x

VH

Very high

w1 x

inference rules. The evaluation system mainly consists of the following four stages.
The first stage is the preparation of input variables, which provides a real-time trend extraction for ORP by applying a linear
regression method. The trend is extracted as follows:
Step 1: Determine the initial window size of the time series as
W, and set the threshold h.
Step 2: In the time window, build a linear model to approximate
the time series of ORP:

b
x k ak  k0 b

11

where k is the time step in the window, k0 is the initial time step,
b
x k is the approximate value, and a and b are the parameters in
the model. The linear model is fitted by minimizing the sum of
least-squares:

W
X
2
b
x ki  xki

12

i0

where xki is the real value of ORP at time step ki in the initial
window. The fitted parameter a is then considered as the initial
ORP trend. And singular value decomposition approach is used
to account for the singularity of matrix.
Step 3: Add the new data point to the window.
Step 4: Calculate the sum of least-squares (S) with the fitted
model over the new windows. If S is less than h, return to step
3, and a will be the trend of the new data point; otherwise,
return to step 2, and the new data will be treated as the initial
data of a new window.
Once the input variables including ORP and its trend are prepared, these quantitative variables are transformed into linguistic
variables in stage 2, which is named fuzzification. In this stage,
two types of membership functions are applied: the generalized
bell membership function, and Z-shaped and S-shaped membership functions. The former functions are defined by the following
equation:

lij x

1

2m



1  x  rij =wij 

13

where rij and wij (i 3; . . . ; 3; j 1; 2) are the center and amplitude of the curve, respectively, and m is a positive constant. The
middle membership functions are represented as Eq. (14):

8
1;
x 6 aj
>
>
>

2
>
j
j
>
aj
>
; a j 6 x 6 a b
< 1  2 bx
j
2
a j
j
u x
 j 2
>
a j b j
>
>
2 bxb
;
6 x 6 bj
j
>
2
a j
>
>
:
0;
x P bj

14

where a j and b j are the parameters that determine the slope of the
curve given by the jth functions. The last membership functions are
described as Eq. (15):

8
>
0;
x 6 cj
>
> 
2
>
>
j
j
j
>
>
;
c j 6 x 6 c d
< 2 dxc
j
2
c j
j
w x


j 2
j
j
j
>
j
>
1  2 x jd j ; c d
6x6d
>
>
2
d c
>
>
>
:
j
1;
xPd

15

where c j and d , similar to a j and b j , are the parameters that deterj

mine the slope of the curve given by w j x.

ORP reflects vaguely the changes of copper concentration in the
reactors. It increases with increasing copper concentration and vice
versa. To make the fuzzified value meaningful and suitable for
practical control, the definition of these membership functions
should be considered of the influences the outlet copper concentration limitations. Assuming that the production limits of the outlet copper concentration are LMIN ; LMAX , the parameters in the
membership function of ORP can be calculated according to the
limits by applying eorp gC Cu2 , which denotes the inverse function of Eq. (10). To make the evaluation more accessible and intuitive, the fuzzy sets are labeled by the fuzzy languages in
accordance with their corresponding physical meanings in the
removal process. When the copper concentration is located in the
center of the production limits, ORP is in a stable situation, abbreviated S. The lower the concentration is, the lower the ORP is.
Accordingly, the fuzzified ORP can be labeled as little low (LL) or
very low (VL). When the concentration is beyond the minimum
limit, ORP reaches a considerably low range abbreviated VL. Correspondingly, ORP in a higher range can be classified as little high
(LL) and very high (VL).
The trend of ORP reveals its underlying variation, which influences the future state of ORP. A sharp trend can cause ORP to
return to the stable situation from an unstable situation in
minutes, whereas it can also cause ORP to depart from the stable
situation in a few seconds. Therefore, the parameters in the memberships of the ORP trend are defined according to its possible
influence on the ORP state. Denoting the time threshold as s
(min), if ORP might skip or drop from one situation to the next
higher or lower situation during s minutes, its trend is considered
as low positive or low negative, respectively; if ORP skips or drops
from the stable situation to the highest or lowest situation, its
trend is labeled as high positive or high negative, respectively.
Tables 2 and 3 present the calculation of each parameter for the
memberships and its corresponding physical meaning for ORP
and its trend. The shapes of the memberships are shown in Fig. 2
(a) and (b).
In stage 3, a set of rules is applied to the fuzzy sets obtained in
stage 2 for evaluating the copper removal process. The ORP trend
indicates the possible future situation of ORP. This possible outlook is thus considered in the proposed evaluation process. When
ORP is currently in the acceptable situation without fluctuation,
and its outlook is also acceptable, the copper process is definitely
in a stable condition. When the ORP is currently acceptable but
with a sharp trend that causes a cloudy outlook, the process is
not stable as expected. In contrast, when ORP is beyond the
limits but with a good trend that could cause it to return to the

298

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

Table 3
Parameter calculation and fuzzy language definition for the trend of ORP.
Grade

Meaning

Function

HN

High negative

u x

LN

Low negative

Zero

LP

Low positive

l21 x
l10 x
l11 x

HP

High positive

w2 x

VL

LL

LH

VH

Degree of membership

Degree of membership

1
0.8
0.6
0.4
0.2
0
0

20

Calculation of parameters


a2 a1  d1 =s; b2 a2  r2 w21

r21 r11  r11 =s; w21 r1 r20  a2 ; m 2

2

2
2
r0 0; w0 r0 r1  r21 ; m 2

r21 r11  r11 =s; w21 r1 d2  r20 ; m 2



2
1
2
d d  a1 =s; c2 d r 2 w21

40
60
ORP(mV)

80

LN

HP

LP

0.8
0.6
0.4
0.2
0

100

HN

-2

-1

0
dORP(mV/min)

Evaluation Grade

d
1

-1
2

100

0
dORP(mV/min)

50
-2

ORP(mV)

Fig. 2. Fuzzy logic evaluation based on ORP for: (a) and (b) the memberships of ORP and its trend, respectively; (c) the fuzzy interference rules; and (d) the relationships
among the evaluation grade, ORP, and its trend.

acceptable state, the process situation is not as terrible as it seems

to be. According to this idea, the specific evaluation standard, presented as reference rules for the copper removal process, is illustrated in Fig. 2(c). The evaluation linguistic variables are obtained
in by the Max-Prod operator following the Larsens fuzzy inference
method.
The linguistic variables are converted into a single numerical
output value in stage 4. The centroid defuzzification method is
applied where the crisp value of the output variable is computed
by finding the center of gravity of the membership function for
the fuzzy value. The value of the output is given by the algebraic
expression:

y

n
X

n
X

i1

i1

lyi yi =

lyi

16

where the non-fuzzy value y presents the evaluation grade of the

copper removal process. A more positive grade means a situation
with a higher concentration and vice versa; an evaluation grade closer to zero indicates a more stable process condition. Fig. 2(d)
shows the evaluation grade surface of the process, which displays
the relationship between two inputs (ORP and its trend) and the
response output (evaluation grade).
2.4. Real-time control strategy
In the copper removal process, most of the copper ions are precipitated in the first reactor; thus, the additive amount is mainly

added to this reactor. The last reactor is an auxiliary reactor for fine
tuning the outlet concentration. The amount added to this reactor
is small and is set manually. Practically, the added amount is
adjusted manually when the measurement impurity concentrations are delivered to the operators every two hours. If the ore
source is altered, or the process condition is not stable, the human
experience and the off-line measurements might not easy satisfy
the requirements of process production. In this section, a control
strategy is developed based on the process evaluation results provided by ORP to promote process stability, as shown in Fig. 3.
The system consists of an evaluation range estimation unit, a
process evaluation unit, an additive adjustment unit, and an additive control unit. The evaluation range estimation unit translates
the production limitation of the outlet concentration to an ORP
range based on the proposed kinetic model. The process evaluation
unit then assesses the process condition using ORP and its trend.
By analyzing the evaluation results, the adjustment mode selector
produces adjustment mode signal. And then the corresponding
adjustment mode is activated by the signal. If the model retraining
mode, which could also be chosen by judge whether pH and temperature exceed set ranges, is chosen, a new set of parameters
would be trained according with the resent industrial data from
database for the existed additive model. When the independent
adjustment mode is chosen, the adjustment amount of zinc powder added in the second reactor would be calculated based on
expert system. Similarly, if the associative adjustment mode is
selected, zinc amounts added in all the reactors would be adjusted.
In the additive control unit, the additive model set the major

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

299

Fig. 3. Evaluation and control system for the copper removal process: (a) the evaluation range estimation unit; (b) the process evaluation unit; (c) the additive adjustment
unit; and (d) the additive control unit.

amounts for the process; and the final amount of zinc powder
added in each reactor is set by summing the major and adjustment
amounts.
To determine the adjustment mode, it is necessary to observe
the process evaluation grade during a set time interval. Statistically, the distribution of these evaluation grades during the interval reveals the effect of the additive model on the removal
process. When most of the grade is unacceptable, the model is so
poor that it does not fit the process. Therefore, the quartiles of
the grade distribution are introduced as a statistical index to determine whether the additive model should be retrained to reset the
additive. The method of additive adjustment is determined according to the current grade when the model remains the same. The
specific procedure of this process is illustrated in Fig. 4.
Based on material balance of CSTR and the first kinetic model of
copper removal reactions, the additive model could be built to set
the major amounts of zinc powder added in the reactors [5,7]:

where f model denotes the additive model afforded by Eq. (17), pM are
the measurable industrial parameters, pE are the parameters that
need to be estimated. pE is usually unchanged, but if the model
retraining mode is activated, pE would be retrained with new industrial data from the database to fit the process accurately.
To amend the amounts of zinc powder, two sets of expert rules
are extracted and stored in the rule base: one is built for the independent adjustment mode; and the other is for the associate
adjustment mode. The evaluation grade (y ), the total amounts of
zinc powder (g Zn g Zn:1 g Zn:2 ), the total amount of the inlet copper ions (M F  C 0Cu2 ) and its change amount during the control
interval (DM) are selected as the influenced factors of the independent adjustment amounts. The general expression of the rules for
independent adjustment is illustrated as follow:

h
i
h
i
h
i
R1i : y 2 Y Ti ; Y Ti1 and M 2 MTi ; M Ti1 and DM 2 DM Ti ; DM Ti1
h
i

and g Zn 2 GTi ; GTi1 then g IZn:2 j1i y x1i DMg Zn =GTi1

8 
VC Cu2 FC 0Cu2 t  1 qtFC Cu2 t  Vr Cu2
>
>
>

<
h
i
2
6K D g O
g O
g Znunder sZn
Zn:1
Zn:2
S
r Cu2 
4pR2S k1G g1R
gk V 1 C Cu2
q
v
D
e
Zn
Zn
>
>
>
: O
g Zn:2 sg OZn:1

where Y Ti , MTi , DMTi and GTi denote the thresholds of y , M, DM and

g Zn , respectively; j1i and x1i are the parameters for calculating the
adjustment of zinc powder. The rule for associative adjustment

17

(R2i ) are similar to R1i , but more influenced factors are introduced

where C Cu2 is the rate of change of the copper concentration in

reactor;

C 0Cu2 ;i ;

i 1; 2 is the inlet copper concentration of the ith

reactor; F and q are the flow rates of the zinc sulfate solution and
the returned underflow; r Cu2 is the rate of the change of the copper
concentration; K D , kG , De , and k are the kinetic parameters that need
to be estimated; g OZn:1 , g OZn:1 and g Znunder denote the major amounts of
zinc powder added in the first and second rector and that taken
from the under flow, respectively; C Cu2 and V are the copper concentrations in the reactor and volume of the reactor, respectively;
q, sZn , and v Zn are the density, surface area, and volume of zinc powder, respectively, which can be measured and calculated in the process; and g and RS are the coefficient and particle radius of cuprous
oxide, respectively; s is the ratio of the amount added in the second
reactor to that added in the first reactor. Accordingly, the major
amounts set by the additive model can be simplified as Eq. (18):

g OZn:1
g OZn:2

f model pM ; pE
sg OZn:1

18

in R2i :

h
i
h
i
h
i
R2i : y 2 Y Ti ; Y Ti1 and M 2 MTi ; M Ti1 and DM 2 DM Ti ; DM Ti1
h
i
h
i
A
and g Zn:1 2 Z Ti ; Z Ti1 and g Zn:2 2 J Ti ; J Ti1 then g Zn:1
2 

A
ji y x2i DMg Zn:1 =Z Ti1 ; g Zn:2
j3i y x3i DMg Zn:2 =J Ti1

where g Zn:1 , c denote the amounts of zinc powder added in the first
and second reactors, respectively, Z Ti and JTi denote the thresholds of
g Zn:1 and g Zn:1 , respectively; j2i , j3i , x2i and x3i are he parameters for
calculating the associative adjustments of zinc powder.
The total amounts of zinc powder added in each reactor are calculated by Eq. (19).

gZn g Zn:1

8 O

O
I
Smode 1
>
< g Zn:1 g Zn:2 g Zn:2

A
A
g Zn:2 
Smode 2
g OZn:2 g Zn:2
g OZn:1 g Zn:1
>

: O
Smode 3
g Zn:1 g O
Zn:2

19

300

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

Begin

Calculating the process evaluation grade (gi)

Counting the distribution characteristic (Q1 ,

Q 2 and M) of G={gj |j=i-k o, ,i} during TO

Is [Q 1, Q 2] larger than [QT1, QT2]?

YES

NO
Dose M exceed [M 1, M 2] ?

YES

NO
Is the model retrained during TO?

NO

YES
YES

Is |g i| larger than g T?

Produce aadjustment model signal to

choose associative adjustment

Produce aadjustment model signal to

choose model retraining

NO
Produce aadjustment model signal to choose
independent adjustment

End
Fig. 4. Flow chart of adjustment mode selection: Q1, M, and Q2 denote the first, second, and third quartile of the grade distribution during TO, respectively, and QT1, M1, M2,
QT2, and gT are the evaluation thresholds.

where gZn denotes the vector of total amounts of zinc powder, Smode
denotes the adjustment mode signal, Smode = 1, 2, 3 represent the
independent adjustment, associative adjustment and mode retrainO
ing mode, respectively; g O
Zn:1 and g Zn:2 denote the amounts of zinc
powder, calculated by the new additive model, added in the first
and second reactors, respectively.

3. Results and discussion

To observe the behavior of the evaluation and control strategy
under the specified disturbances, three different cases are chosen
from the practical process. The ranges of the inlet parameters for
these cases are shown in Table 4, and the inlet flow rate and copper
concentration are illustrated in Fig. 5. In the first case, the flow rate
of the inlet solution is mostly constant, and its changes are discontinuous and in a proper range. In the second and third cases, the
inlet flow rate is changed frequently, but the flow rate fluctuates
more sharply, and the amplitude is larger in the last case compared
with that in the second case.
We first present the evaluation grade obtained by the proposed
method for the three cases based on ORP values sampled from the

practical process (Fig. 6). Fig. 6(a) shows the evaluation grade
under a considerably stable situation. Most the inlet copper concentration was not change sharply during most of the measurement intervals, and the flow rate was mostly constant. It is easy
to determine the amounts of zinc powder based on human experience. Correspondingly, ORP varies in a narrow range close to a
proper value, and the evaluation grades are thus mostly acceptable.
Fig. 6(b) illustrates the evaluation grade under the situation with
some fluctuations. In this case, ORP is kept at a higher level so that
the evaluation grade is mostly above 0.5. However, when the ORP
drops to the lower limitation (at the 38th hour), the evaluation
grade predicts the decline before ORP reaches a lower level.
Similarly, the evaluation grade also forecasts the ascent of ORP
after two hours. The evaluation results of the third case are
similar to that of the second one. Due to the fluctuation of the
flow rate and inlet copper concentration, human operators could
not set enough amounts of zinc powder and adjust them timely.
ORP mostly exceed the acceptable range, and the grade is thus
mostly high, as shown in Fig. 6(c). It is observed that the grade
can reveal the outlook of the removal state before it occurs,
especially in the fluctuant situation (e.g. the second and third
cases).

Table 4
The range of inlet parameters in the three cases.
Case

Inlet copper concentration (g/L)

Inlet flow rate (m3/h)

Underflow rate (m3/h)

Solid content in underflow (kg/L)

pH value

Temperature (C)

1
2
3

0.911.43
0.91.24
1.051.37

202289
182283
176308

2931
2830
2930

1.41.5
1.41.5
1.41.5

3.94.2
3.94.2
3.94.2

6266
6266
6266

301

1.5

200

100
0
0

16

24
Time (hour)

Flow Rate (m3/h)

32

40

2
300
1.5

200

100
0
0

16

24
Time (hour)

c
Flow Rate (m3/h)

0.5
48

32

40

0.5
48
2

300
1.5

200

100
0
0

16

24
Time (hour)

Inlet flow rate

32

40

0.5
48

Cu concentration (g/L)

2
300

Cu concentration (g/L)

Flow Rate (m3/h)

Cu concentration (g/L)

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

Inlet copper concentration

85

1
0.5

65

0
45

-0.5

25
0

16

24
Time (hour)

32

40

120

1
0.5

90

0
55

-0.5

20
0

ORP(mV)

-1
48

16

24
Time (hour)

32

40

150

-1
48

Condition Grade

ORP(mV)

1
0.5

100

0
50

-0.5

0
0

16

24
Time (hour)
ORP

32

40

-1
48

Condition Grade

ORP(mV)

Condition Grade

Fig. 5. Inlet flow rate and copper concentration for the first (a), second (b), and third (c) cases.

Evaluation results

Fig. 6. ORP values and the evaluation grades for the first (a), second (b), and third (c) cases.

For comparison, we also use the controller used in current practice and a conventional control strategy. In current practice, the
amounts of zinc powder are set manually based on the inlet and
outlet copper concentrations and the flow rate. The conventional
control strategy is similar to the proposed control strategy; however, the amount of zinc added is adjusted according to the current
ORP value rather than the evaluation grades. Fig. 7 shows the
amounts of additive set by three different control strategies. The
adjustment modes, in the proposed approach, differ from each
other in these different cases. In the first case, mostly, only the

amount added in the second reactor needs to be slightly adjusted

because the evaluation grades are generally acceptable. However,
in the other cases, all of the amounts added in the reactors would
be adjusted in most of the time for the fluctuation of evaluation
grade. In the practical controller, the additive amount is adjusted
three or four times in a shift by the workers when the manually
measured concentration is delivered. The other two control strategies set the additive amounts more timely compared to the manual
operation. Compared with the other control strategies, the
amounts set by the proposed strategy change more smoothly when

302

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

200
150
100
50
0

16

32

40

16

24
Time (hour)

32

40

f
Zinc amount (kg/h)

200

16

24
Time (hour)

16

24
Time (hour)

32

40

48

16

24
Time (hour)

32

40

48

32

16

24
Time (hour)

32

40

48

100

50

48

400

20

100

40

48

200

e
Zinc amount (kg/h)

24
Time (hour)

Zinc amount (kg/h)

Zinc amount (kg/h)

60

Zinc amount (kg/h)

Zinc amount (kg/h)

40

150
100

48

Conventional control

50
0

Evaluation based control

Manual control

b
0.4

20
Number

Cu concentration (g/L)

Fig. 7. Zinc amounts setting by different control strategies for three cases: (a) and (b) are the additive amounts added in the first and second reactors in the first case,
respectively; similarly, (c) and (d) for the second case, respectively; and (e) and (f) for the third case, respectively.

0.3

10

0.2
0

16

24

32

40

0
0.1

48

0.15

0.2

0.35

0.4

0.45

0.5

0.4

0.45

0.5

0.4

0.45

0.5

20

0.3

10

0.2
0

16

24

32

40

0
0.1

48

0.15

0.2

0.25

0.3

0.35

Cu concentration (g/L)

f
0.4

20
Number

Cu concentration (g/L)

0.3

d
0.4

Time (hour)

0.25

Cu concentration (g/L)

Number

Cu concentration (g/L)

Time (hour)

0.3

10

0.2
0

16

24

32

40

48

0
0.1

Time (hour)
Manual control

0.15

0.2

0.25

0.3

0.35

Cu concentration (g/L)
ORP based control

Evaluation based control

Fig. 8. Comparison of the running results for three cases: (a) and (b) for the outlet copper concentration and the corresponding concentration distribution in the first case,
respectively; similarly, (c) and (d) for t the second case, respectively; and (e) and (f) for the third case, respectively.

ORP is out of the optimal range. This operation not only saves
energy but also avoids a sudden change in the outlet copper concentration, which might lead continuous oscillations in production.

The running results of these three control strategies are illustrated in Fig. 8. From Fig. 8, it is observed that a portion of the outlet copper concentrations obtained in the practical and

303

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

Table 5
Assessment indexes for the running results.
Cases

The outlet copper concentration

Average additive amount (kg/h)

Mean value (g/L)

Qualification ratio (%)

Oscillation range (g/L)

Median (g/L)

The 1st case

Manual control
Conventional control
Evaluation based control

0.298
0.274
0.306

100
100
100

[0.200,0.390]
[0.200,0.350]
[0.260,0.340]

0.310
0.270
0.300

112.3
119.8
105.2

The 2nd case

Manual control
Conventional control
Evaluation based control

0.332
0.268
0.304

92
96
100

[0.260,0.410]
[0.190,0.390]
[0.250,0.380]

0.320
0.250
0.300

141.0
169.5
149.7

The 3rd case

Manual control
Conventional control
Evaluation based control

0.361
0.330
0.337

84
92
96

[0.230,0.410]
[0.190,0.410]
[0.220,0.405]

0.380
0.340
0.340

267.7
297.9
281.7

conventional control strategies either too high to reduce the production efficiency of next processes, or too low to afford enough
activators for the cobalt removal stage. However, by applying the
proposed approach, the oscillation ranges of the outlet copper concentrations become narrow; and most of the concentrations locate
in the center of the production limitation (0.20.4 g/L).
To accurately assess the running results of the control strategies, five indexes are introduced in this paper: qualification ratio,
median, mean concentration, oscillation range, and average additive amount (Table 5). From the mean concentration and minimum
values, it is observe that the outlet copper concentrations in the
practical process are usually too low to afford enough activator
for the cobalt removal stage, and fluctuate in a large range. Both
of the other control strategies could improve this situation. By
applying the proposed control strategy, nearly all of the running
results are qualified and the outlet concentrations are high enough
to afford an appropriate amount of the activator for the next
removal stage, whereas the results obtained by the other method
occasionally exceed the production limitation. From the simulation
results, it is observed that the stability of the copper removal process under the evaluation-based control is superior to those under
the conventional and the practical controls regardless of the situation. Therefore, the proposed approach thus could not only reduce
the outlet copper concentration and zinc consumption, but also
reserve an appropriate amounts of copper ions as activators for
the next removal stage.

4. Conclusions
Based on the kinetic and Arrhenius laws, a mechanistic model is
established to estimate the optimal range of ORP relevant for the
production limitation of the outlet impurity concentration. An
evaluation method is then proposed to grade the copper removal
process. In this method, ORP and its trends are fuzzified according
to the estimated range. It affords quantitative feedback information for process control that represents the current and underlying
states of the removal reactions. Based on the evaluation method, a
control strategy is designed. In this strategy, three types of control
methods are prepared for different situations. The control methods
are chosen by analyzing the statistical characteristics of the evaluation grade. This control system can improve the qualified ratio
and the process stability more than using the conventional control
method. Furthermore, it minimizes energy consumption and guarantees an appropriate activator amounts for the next process.
However, there are still many problems to study, including associative control when more ORP information is afforded in all reactors. Overall, the simulation results show that the proposed control
strategy has promising applications for industrial-scale impurity
removal processes in hydrometallurgy.

Acknowledgments
This work is financially supported by the innovation-driven
plan in Central South University (Grant No. 2015cx007), Science
Fund for Creative Research Groups of the National Natural Science
Foundation of China (Grant No. 61321003), and the National Natural Science Foundation of China (Grant Nos. 61273185 and
61174133).

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.cej.2015.07.094.

References
[1] J.C. Balarini, L. de, O. Polli, T.L.S. Miranda, R.M.Z. Castro, A. Salum, Importance of
roasted sulphide concentrates characterization in the hydrometallurgical
extraction of zinc, Miner. Eng. 21 (2008) 100110.
[2] K. Laatikainen, M. Lahtinen, M. Laatikainen, E. Paatero, Copper removal by
chelating adsorption in solution purification of hydrometallurgical zinc
production, Hydrometallurgy 104 (2010) 1419.
[3] Y. Xie, S. Xie, X. Chen, W. Guia, C. Yang, L. Caccett, An integrated predictive
model with an on-line updating strategy for iron precipitation in zinc
hydrometallurgy, Hydrometallurgy 151 (2015) 6272.
[4] B. Sun, W.H. Gui, Y.L.C.H. Wang, Yang, Intelligent optimal setting control of a
cobalt removal process, J. Process Control 24 (2014) 586599.
[5] Y.G. Li, W.H. Gui, K.L. Teo, H.Q. Zhu, Q.Q. Chai, Optimal control for zinc solution
purification based on interacting CSTR models, J. Process Control 22 (2012)
18781889.
[6] J. Nsi, Statistical analysis of cobalt removal from zinc electrolyte using the
arsenic-activated process, Hydrometallurgy 73 (2004) 123132.
[7] B. Zhang, C.H. Yang, H.Q. Zhu, Y.G. Li, W.H. Gui, Kinetic modeling and
parameter estimation for competing reactions in copper removal process from
zinc sulfate solution, Ind. Eng. Chem. Res. 52 (2013) 1707417086.
[8] M.V. Ruano, J. Ribes, A. Seco, J. Ferrer, Low cost-sensors as a real alternative to
on-line nitrogen analysers in continuous systems, Water Sci. Technol. 60
(2009) 32613268.
[9] C. Chang, Y. Ma, C. Lo, Application of oxidationreduction potential as a
controlling parameter in waste activated sludge hydrolysis, Chem. Eng. J. 90
(2002) 273281.
[10] N. Kishimoto, Y. Nakamura, M. Kato, H. Otsu, Effect of oxidationreduction
potential on an electrochemical Fenton-type process, Chem. Eng. J. 260 (2015)
590595.
[11] S. Lackner, C. Lindenblatt, H. Horn, Swinging ORP as operation strategy for
stable reject water treatment by nitritationanammox in sequencing batch
reactors, Chem. Eng. J. 180 (2012) 190196.
[12] J. Claros, J. Serralta, A. Seco, J. Ferrer, D. Aguado, Real-time control strategy for
nitrogen removal via nitrite in a SHARON reactor using pH and ORP sensors,
Process Biochem. 47 (2012) 15101515.
[13] R. Yu, F. Chi, W. Cheng, J. Chang, Application of pH, ORP, and DO monitoring to
evaluate chromium(VI) removal from wastewater by the nanoscale zerovalent iron (nZVI) process, Chem. Eng. J. 255 (2014) 568576.
[14] C. Antileo, H. Medina, C. Bornhardt, C. Muoz, F. Jaramillo, J. Proal, Actuators
monitoring system for real-time control of nitrificationdenitrification via
nitrite on long term operation, Chem. Eng. J. 223 (2013) 467478.
[15] S. Lackner, H. Horn, Evaluating operation strategies and process stability of a
single stage nitritationanammox SBR by use of the oxidationreduction
potential (ORP), Bioresour. Technol. 107 (2012) 7077.

304

B. Zhang et al. / Chemical Engineering Journal 284 (2016) 294304

[16] M.V. Ruano, J. Ribes, A. Seco, J. Ferrer, An advanced control strategy for
biological nutrient removal in continuous systems based on pH and ORP
sensors, Chem. Eng. J. 185 (2012) 212221.
[17] R. Yu, C. Lin, H. Chen, W.P. Cheng, M. Kao, Possible control approaches of the
electro-fenton process for textile wastewater treatment using on-line
monitoring of DO and ORP, Chem. Eng. J. 218 (2013) 341349.
[18] S.G. Won, C.S. Ra, Biological nitrogen removal with a real-time control strategy
using moving slope changes of pH(mV)- and ORP-time profiles, Water Res. 45
(2011) 171178.
[19] B. Sun, W.H. Gui, Y.L. Wang, C.H. Yang, An integrated prediction model of
cobalt ion concentration based on oxidationreduction potential,
Hydrometallurgy 140 (2013) 102110.
[20] P.S. Londhe, B.M. Patre, A.P. Tiwari, Design of single-input fuzzy logic
controller for spatial control of advanced heavy water reactor, IEEE Trans.
Nucl. Sci. 61 (2014) 901911.
[21] R.J. Romero-Troncoso, E. Cabal-Yepez, R.A. Osornio-Rios, H. Miranda-Vidales,
FPGA-based online detection of multiple combined faults in induction motors
through information entropy and fuzzy inference, IEEE Trans. Ind. Electron. 58
(2011) 52635270.
[22] Z. Li, Z. Qin, Z. Wu, S. Li, Y. Zhang, M. Dong, W. Fan, J. Wang, Fuzzy logic control
of a reverse flow reactor for catalytic oxidation of ventilation air methane,
Control Eng. Pract. 25 (2014) 112122.

[23] J.J. Carbajal-Hernndez, L.P. Snchez-Fernndez, J.A. Carrasco-Ochoa, J.F.

Martnez-Trinidad, Immediate water quality assessment in shrimp culture
using fuzzy inference systems, Expert Syst. Appl. 39 (2012) 1057110582.
[24] M.C. Heredia-Molinero, J. Snchez-Prieto, J.V. Briongos, M.C. Palancar,
Feedback PID-like fuzzy controller for pH regulatory control near the
equivalence point, J. Process Control 24 (2014) 10231037.
[25] R.M. Kamel, A. Chaouachi, K. Nagasaka, Wind power smoothing using fuzzy
logic pitch controller and energy capacitor system for improvement microgrid performance in islanding mode, Energy 35 (2010) 21192129.
[26] D.O. Ferraro, Fuzzy knowledge-based model for soil condition assessment in
Argentinean cropping systems, Environ. Modell. Softw. 24 (2009) 359370.
[27] Leposava.B. Ristic, Borislav.I. Jeftenic, Implementation of fuzzy control to
improve energy efficiency of variable speed bulk material transportation, IEEE
Trans. Ind. Electron. 54 (2007) 250258.
[28] X.D. Sun, K.H. Koh, B.G. Yu, M. Matsui, Fuzzy-logic-based V/f control of an
induction motor for a dc grid power-leveling system using flywheel energy
storage equipment, IEEE Trans. Ind. Electron. 56 (2009) 31613168.
[29] D. de Almeida Souza, W.C.P. de Aragao Filho, G.C.D. Sousa, Adaptive fuzzy
controller for efficiency optimization of induction motors, IEEE Trans. Ind.
Electron. 54 (2007) 21572164.