IFAC Workshop on Automation in the

Mining, Mineral and Metal Industries

September 10-12, 2012. Gifu, Japan

Improving copper matte grade control in a concentrate flash furnace

Luis Bergh*, Iván Cornejo*, Fernando Romero+ and Cristian Sulzer°

*CASIM, Automation and Supervision Centre for Mineral Industry, Chemical Engineering Department,
Santa Maria University, Valparaiso, Chile, (e-mail: luis.bergh@usm.cl)
+
Process Control Manager, Anglo American Chile, Santiago, Chile
°Chagres Division, Anglo American Chile, Santiago, Chile

Abstract: The copper matte grade (CMG) variation in a flash furnace depends on bath temperature, phase
levels, ratio of oxygen to sulfur, iron and copper content on the feed, and tonnage ratio of silica flux and
feed, among others. The CMG is obtained from laboratory analysis of samples taken when the matte is
tapped out from the furnace. Then the CMG is known with a time delay of 1-2 hours, almost every one
hour. Since feed characteristics usually change during this period, any feedback control of CMG cannot
reduce enough the grade variability. On the other hand, it was shown that main disturbances are coming
with the composite feed tonnage and composition, being the powder tonnage control system the main
contribution. A new calibration system was proposed and implemented. During an evaluation period of
60 days, it was found that the standard deviation of the CMG was reduced in 25%. This improvement can
be explained considering two effects. First, the stabilization of the total feed tonnage that affect the mass
balance in the furnace. Second, even when the powder tonnage is only 7% of the total tonnage, its special
chemical composition contributed to increase the copper content in the matte. The stabilization of the
furnace operation also improved the phase separation, as well as reduced the variance of the gas SO2
composition and flow rate. The operators took advantage of this reduced variability to increase the fusion
rate and thus the mean generated gas flow rate in the furnace, without violating the plant constraints.
One important conclusion is that this result will permit to increase the average value of CMG, when the
development of the virtual sensor of feed composition is completed and when this estimation becomes
part of a FF-FB control. Copyright © 2012 IFAC

Keywords: flash furnace, control, industrial data, pyrometallurgy

frequency measured variables with targets. These high

1. INTRODUCTION
frequency target predictions permit the effective use of
Feedback control of local objectives is a powerful tool to feedback control during the absence of target measurements.
modify an input variable to compensate process disturbances Some complications may arise when the models are built on
or to closely follow set point changes. However, the and uses inaccurate measurements. Better estimation of some
successful implementation of this control mainly depends on key variables, as characteristics of the process feed, can be
process dynamics and on the accuracy of measurements and used to add a feed forward compensation. More details on
actuators. When process variables are frequently observed, control and optimization of flash smelting can be found in
and process disturbances are highly correlated, predictive Davenport et al. (2003).
control is a better alternative than conventional PID control,
especially when the process presents significant time delay. In this paper, the copper flash smelting process control is
Furthermore, if process dynamics is time variant, adaptive analyzed in the previous perspective. Typical historical
predictive control is the appropriate control tool. Knowledge- CMG, obtained from Chagres smelting plant for a period of
based expert control systems are an alternative when several 90 days, are shown in Figure 1. The dashed lines represent
objectives should be simultaneously met and process the upper and lower limits for this target. The mean CMG
dynamics is complex. The use of complementary tools to was 59.4% with a standard deviation of 2.2%. One can see
improve control is discussed by Hodouin (2009). that the target is at least 60% of the time outside the desired
band.
The variability of the targets cannot be significantly reduced
by any feedback control algorithm if significant time delay is A second objective is the matte temperature, since the
present and target variables are observed at low frequency, as efficiency of the phase separation is influenced by its
compared with process disturbances changes over time. One viscosity in the settler. The target temperature is 1255 °C, and
solution is to use models (virtual sensors) relating high an acceptable band is between 1230 and 1280 °C. Historical

978-3-902823-12-0/12/$20.00 © 2012 IFAC 13 10.3182/20120910-3-JP-4023.00035

IFAC MMM 2012
September 10-12, 2012. Gifu, Japan

data was centered in 1270 °C with deviations between 1250 A typical flash furnace is described in Figure 3. In flash
and 1300 °C. smelting the fine, dried copper sulphide concentrate and silica
flux with air, oxygen enriched air or pure oxygen blast and
70
recycled flue dust are fed through a concentrate burner into
68
Copper matte grade (%)

the flash furnace reaction shaft. In the reaction shaft the

66
suspension of gas, solid concentrate particles and solid flux
64
particles, are formed and one important reaction is:
62
60 2CuFeS2 + 2.5O2 + SiO2 = Cu2S.FeS + FeO.SiO2 + 2SO2 +
58 energy (1)
56
54 Parallel reactions are the oxidation of FeS and FeS2:
52
50 FeS + 1.5O2 = FeO + SO2 (2)

FeS2 + 2.5 O2 = FeO + 2SO2 (3)

Date [day-month]
  Concentrate
Fig. 1. Historical copper matte grade. O2 N2
Flux
A third objective is to maintain the magnetite concentration
in the slag phase in 6%, and not to exceed 12%. In Figure 2
Dryer 
some typical magnetite concentrations are shown, with a
mean value of 12% and deviations between 8 and 16%.
22 TI
Magnetite concentration [%]

20
18
AI
16
14 AI
12
10 Fig. 3. A typical flash furnace.
8
6
4 The formed FeO then migrates from matte to slag phase due
2 to a change in density. The extension of both reactions is
0 controlled by the amount of oxygen, as a limiting reactive.
The powder is composed of oxidized sulfur, copper and iron,
sulfated and reduced copper, and other materials that do not
participate in oxidation reactions, and they only have the
Date [day-month]
effect of cooling the bath.
Fig. 2. Typical magnetite concentration in slag. When the suspension leaves the reaction shaft, the reacted
molten concentrate particles and inert flux particles are
In the next sections the process characteristics, the main separated from the gas stream and hit towards molten slag
disturbances, measurements and control systems are and matte. In the molten bath, the reactions are completed
described to understand the causes of this poor control. Some and matte and slag will be settled at the settler of the furnace
solutions are first proposed and then successfully as respective layers due to their different density.
implemented to reduce the observed variance and to permit
displacing the mean value inside the control band. The matte and slag are tapped out through the holes located at
the side or at the end walls of the settler. The waste gases
2. COPPER FLASH SMELTING IN CHAGRES generated at the process are ducted through the uptake shaft
into waste-heat boiler and electrostatic precipitator, and then
Outokumpu flash smelting is a pyrometallurgical process for processed on the sulphuric acid plant.
smelting metal sulphide concentrates (Jamsa-Jounela et al.,
2003). The Outokumpu flash smelting process consists of a 2.1 Flash furnace operation and control
flash furnace, waste-heat boiler and electrostatic precipitator.
A flash smelter usually includes the following auxiliary units: The flash furnace operator must smelt concentrate at a steady,
feed mixture preparation and drying, converters, slag specified rate while: a) producing matte of specified Cu grade
treatment system, SO2 fixation system, anode furnace and b) producing slag of specified Fe3O4 content c) producing
anode casting. slag at specified temperature, d) producing waste gases of

14
IFAC MMM 2012
September 10-12, 2012. Gifu, Japan

specified SO2 content and e) maintaining a protective coating (b) keeping enough Fe and S in the matte so that subsequent
of magnetite-rich slag into the furnace. convertions can operate autogenously while smelting some
other recycled materials.
The most important phenomenon that affects process control
is the quality of feed material. The overall quality of the feed The sampling points are located, one at the exit of the drying
mixture is influenced by quality and amount of the feed (dry system, and two in each of the powder belts. The revert is
concentrate, recycled slag, revert and powder) and the sampled before it is charged to the silo. All samples are taken
amount of silica flux. A schematic of the feed material and analysed for copper, iron, sulfur and other contents, 5-6
preparation is shown in Figure 4. times per day. Total dry feed load and composition are not
measured on line.
Dry concentrate
and silica flux 2.2. Problems in CMG control
Revert Powder
Feedback CMG control can work when the dry feed is a
T40 T100 T50 prepared mixture of a given and known composition that last
for several hours. In Chagres, a time variant composition feed
is produced mainly because of constraints in space and
7 tph 3 tph concentrates coming from different plants. The matte grade is
D D obtained by laboratory analysis from samples taken every one
hour in average, when the matte is tapped out of the furnace.
70 tph 80 tph The sample analysis can take 1-2 hours. When this delayed
FF information is available the ratio between the total oxygen
and concentrate can be adjusted. Most of the time, the feed
Fig. 4. Schematic of feed materials to the furnace. quality has changed during this period and the new ratio do
not help to reduce the variance in CMG. A large ratio gives
Dry copper concentrate is mixed with a dosed amount of extensive Fe and S oxidation and high CMG. Physically, the
silica flux, in a 400 tons silo, discharging in a dosing system. ratio is controlled by adjusting the rates at which air and
The total amount of dry feed is set by screws discharging on oxygen enter the furnace, at constant concentrate feed rate.
a conveyor. Test performed contrasting the tonnage delivered
against the set point of the dry feed control, showed that the Similarly, when the slag contents are available the silica flux
maximum tonnage error was less than 3%. can be adjusted. The iron oxide formed by concentrate
oxidation is fluxed with SiO2 to form liquid slag. The amount
Revert is transported from a 100 tons silo by a series of lift of SiO2 is based upon the slag having (a) a low solubility for
mats and conveyor belts to a 3 m3 silo that discharges into a Cu and (b) sufficient fluidity for easy tapping and a clean
weight meter conveyor set to a precise load. The revert matte/slag separation. A SiO2/Fe mass ratio of 0.7 to 1.0 is
loading control system started its operation in 2010, and test used. It is controlled by adjusting the rate at which silica flux
performed to contrast the tonnage delivered against the revert is fed.
tonnage set point showed that the maximum tonnage error
was less than 0.5 tons/h. Matte and slag temperatures are measured as matte and slag
flow from the furnace. Disposable thermocouple probes and
A pair of conveyor drags, from the electrostatic precipitators optical pyrometers are used. Matte and slag temperatures are
and the waste heat boiler, feed a 50 tons silo with powder. controlled by adjusting the air to concentrate and to oil
The load of powder is adjusted by setting the drags velocity. burners. Slag temperature is adjusted with the oil burner
Some tests were performed to the calibration system operation. Matte and slag temperatures are typically around
employed since 2006. Significant differences were found 1250°C. Excessive temperatures are avoided to minimize
between the tonnage set point value and the powder tonnage refractory and cooling jacket wear.
delivered. When the control system required 2.6 tons/h the
drags really delivered 3.9, and when the control system 2.3. Improvements in CMG control
required 2,9 tons/h the drags really delivered 5.6.
Furthermore, if the drags velocity was set to 0 rpm, the CMG control can be improved if a virtual sensor of the
SCADA system displayed 1.9 tons/h. Consequently, the objective replaces the lack of direct measurements on time.
powder tonnage control system must be reviewed. However, due to the sparse, delayed, and inaccurate
measurements available, even PLS methods were not able to
Chagres flash furnace strategy is to charge dried concentrate obtain a useful predictive model of CMG as a function of the
mix to the furnace at a prescribed rate and to base all other measured variables. An alternative is the incorporation of
controls on this rate. Having chosen concentrate feed rate, the feed forward control to adjust on-line the oxygen/concentrate
flash furnace operator must next select the matte copper ratio, based on stoichiometric calculations. To implement
grade by estimating the extent of Fe and S oxidation. It is this, the actual composition of the total feed to the furnace
selected as a compromise between: (a) maximizing SO2 must be estimated by using a virtual sensor based on delayed
evolution in the flash furnace (to be efficiently captured) and measurements from each kind of solid inputs.

15
IFAC MMM 2012
September 10-12, 2012. Gifu, Japan

For a constant setting of total dry feed a large variance has This new system reduced the average powder tonnage set
been observed. A study (Cornejo, 2012) shows that the large point deviations from 2 to 0.25 tons/h, as it is illustrated in
variability contribution came from the powder loading system Figure 6.
(72%), and then from the dry feed loading system (21%) and
from the revert loading system (7%). Then the first step in
14 Using old Using new
this project was to reduce the feed bias and variance, by

Powder tonnage (ton/h)

checking each of the load estimation procedure. The main 12 calibration calibration
results are presented and discussed in the next section of this 10
paper. The next step will be the estimation of time delays 8
from the local measurements to the furnace injection burner. 6
When these results were obtained a virtual sensor to predict 4
the feed composition will be built. The improved predictions
2
of feed composition will be used to adjust the
oxygen/concentrate ratio. 0

3. RESULTS AND DISCUSSIONS

Date [day-month]
The powder loading system was studied and some pitfalls
were detected. The feed to the powder hopper was interrupted Fig. 6. Effect of changing the powder tonnage control.
and the velocity of the dragging system was set for a given
period of time to different values. The initial and final hopper One can also see the effect of stabilizing the same average
weights were registered for each period. Using this data the total feed on the CMG in Figure 7. A 25% reduction in
tonnage calculated for each drugging system velocity was variability was observed. Analyses of gases showed a
compared to the value registered in the SCADA system. The reduction in oxygen concentration variation. One question
results are shown in Figure 5. The main explanation for the arises at this time, why the smelting process is so sensitive to
observed calibration errors was that the last calibration this part of the feed which represents only 3 tons/h for a total
procedure was executed before some physical changes in the feed rate of 80 tons/h. The improvements in the new control
dragging system were performed and that the correlation system are to deliver the right amount of powder to the
between tonnage and drags velocity did not included the burner and to significantly reduce its variation.
point zero tonnage at zero drags velocity. Any time the set
point of the powder tonnage was changed the control system 70
over reacted producing large overshoots and offsets. The 68 Before After
Copper matte grade (%)

effect of the powder hopper level was also investigated. A 66

new correlation between tonnage and drags velocity was 64
found, as it is shown in Figure 5. 62
60
58
8 56
54
7 52
Effective powder tonnage [t/h]

50
6
Date [day-month]
5
Fig. 7 Effect on copper matte grade.
4
New This unexpected result is explained in part because the total
3 feed tonnage variation was considerably decreased and
Old because the fraction of powder in the feed was reduced. The
2 particular effect of powder addition can be explained by the
2 3 4 5 6 7 8 chemical reactions occurring in the flash furnace. The powder
contains mainly copper sulfate and hematite, and therefore
Powder tonnage set point [t/h] the powder is decomposed, as follows:
Fig. 5. Comparison of new and old calibration systems. CuSO4 = 1/2Cu2O + SO2 + 1/4O2 (4)
Since this model is effective only when the powder hopper is Fe2O3 = 2/3Fe3O4 + 1/6O2 (5)
working normally, an algorithm to detect discharge problems
from the hopper was developed, by using the change in Both reactions liberate oxygen to be used in the oxidation of
tonnage available in the hopper. iron, and thus in the concentration of the copper in the matte.

16
IFAC MMM 2012
September 10-12, 2012. Gifu, Japan

Additionally, the copper oxide produced in Eq. 4 reacts with A secondary effect of this result is that the standard deviation
the iron sulfur transferring the iron from the matte to the slag: of the gas flow rate generated at the flash furnace was also
reduced in almost 20%, as it is shown in Figure 10.
Cu2O + FeS = Cu2S + FeO (6)
45
Then by reducing the amount and variability of powder
addition, the copper matte grade average remained in 59.6%, 40
but its variability is significantly reduced in 25%. Finally, the

Gas flow rate [KNm3/h]

first goal of reducing CMG variability was achieved. This 35
small variability will allow to finding a better reactive
30
dosage, for example increasing the powder rate, to push the
CMG inside the target band. 25
The slag magnetite concentration changes are shown in 20
Figure 8. One can see that the concentration mean value was
decreased from 13% to 10%, while its standard deviation was
reduced in 50%. Date [day-month]

Fig. 10. Effect on generated gas flow rate in flash furnace.

22
20
Magnetite concentration [%]

The recommended maximum target for the gas flow rate is 35

18
KNm3/h. This became an important result since one
16
14 bottleneck is the capacity of the acid plant to process the
12 gases. The small variability observed allow the operators to
10 push the flow rate close to the limit most of the time. The
8 mean gas flow rate was increased from 31.9 to 33.3 KNm3/h.
6 The increased stability of the generated gas flow rate favored
4 also the heat recovery at the boiler and the increment of the
2 fusion rate, in order to take advantage of the new space to
0 generate more gases, as it is shown in Figure 11. The mean
fusion rate was increased from 68.2 to 73.6 ton/h.

Then the next step is to shift the average CMG to a better

Date [day-month] value by proper adjustment of the oxygen enrichment
coefficient by feed forward control, based on the virtual
Fig. 8. Effect on slag magnetite concentration. sensor of composition of the total feed. This work will be
done in the near future. The matte temperature control and
Since the oxygen/concentrate ratio is used to manually the control of the magnetite concentration in the slag can also
control the CMG, the stabilization of the CMG was used by take advantage of the initiated stabilization process.
operators to reduce the variation of this ratio, as it is shown in
Figure 9. The standard deviation of the ratio was reduced
from 9.2 to 2.9 Nm3/ton, while its mean value remained 110
Fusion rate (ton/h)

around 153 Nm3/ton. 100

90
80
190
70
Before After
180 60
Oxygen/concentrate ratio

50
170
40
[Nm3/ton]

160
150 Date [day-month]
140
Fig. 11. Effect on concentrate fusion rate.
130

CONCLUSIONS
Date [day-month]
Feed back control of CMG has found to be inefficient
Fig. 9. Effect on oxygen/concentrate ratio. because of significant variation in the load preparation and

17
IFAC MMM 2012
September 10-12, 2012. Gifu, Japan

direct measurement of targets with long time delays. Since

efforts to build a virtual sensor to predict the CMG were
unfruitful, due to sparse and inaccurate measurements, two
proposals are made. First, to reduce the bias and variance of
the total furnace load, and second, to develop a virtual sensor
to have a better estimation of the composition of the total
feed entering into the furnace, to be used in a feed forward
control of CMG.

A preliminary study showed that the powder loading control

system was the main contribution of the bias and variability
in the total feed entering into the furnace.

The bias and variance reductions in the powder loading

system allowed a reduction in CMG variability of 25%. This
improvement can be explained considering two effects. First,
the stabilization of the total feed tonnage that affect the mass
balance in the furnace. Second, even when the powder
tonnage is only 7% of the total tonnage, its special chemical
composition contributed to increase the copper content in the
matte through the described reactions.

The stabilization of the furnace operation also improved the

phase separation, as well as reduced the variance of the gas
SO2 composition and flow rate. The operators took advantage
of this reduced variability to increase the fusion rate and thus
the mean generated gas flow rate in the furnace, without
violating the plant constraints.

One important conclusion is that this result will permit to

increase the average value of CMG, when the development of
the virtual sensor of feed composition is completed and when
this estimation becomes part of a FF-FB control.

ACKNOWLEDGEMENTS

The author would like to thanks Santa Maria University (Project

271203), NEIM, Project P07-087-F, ICM-Mideplan and
Fondecyt (Project 1100854) for their financial support, and
Anglo American Chile for the work support and allowing the
presentation of this paper.

REFERENCES

Davenport W., D. Jones, M. King and E. Partelpoeg, 2003.

Flash Smelting Analysis, Control and Optimization,
Second Edition, John Wiley and Sons.
Hodouin D., 2009. Automatic Control in Mineral Processing
Plants: An Overview, Pre-prints IFACMMM2009
Workshop, (Ed. L.G. Bergh), Viña del Mar, Chile, 14-16
October, pp. 1-12.
Jämsä-Jounela S. -L., M. Vermasvuori, P. Endén, S.
Haavisto, 2003. A Process Monitoring System Based on
the Kohonen Self-organizing Maps, Control
Engineering Practice, 11, 1, pp. 83-92.
Cornejo, I., 2012, Improvements on the control system of a
flash smelting furnace (in spanish), M.Sc. Thesis,
Chemical Engineering Department, Santa Maria
University, Valparaiso, Chile.

18