Minerals Engineering 18 (2005) 11231126

Technical note

This article is also available online at:

www.elsevier.com/locate/mineng

Control strategy for a Teniente Converter

L.G. Bergh
a

a,*

, P. Chacana b, C. Carrasco

Chemical Engineering Department, Santa Maria University, P.O. Box 110-V, Valparaiso, Chile
b
Caletones Smelter, El Teniente Division, Millan 1020, Rancagua, Codelco-Chile, Chile
Received 23 December 2004; accepted 14 February 2005
Available online 2 April 2005

Abstract
A diagnosis of the actual distributed control and measurements system for a Teniente Converter, processing copper concentrates,
is presented. Besides the importance of this converter in Caletones Smelter, there is a number of unsolved problems related to lack of
instrumentation, lack of process knowledge, odd operating practices, and lack of procedures to process data to aid management and
operating decisions. In general, process control of some local objectives are frequently achieved. A proposal to implement a supervisory control strategy for Teniente Converters, in harmony with other process units in the smelter, is presented.

2005 Elsevier Ltd. All rights reserved.
Keywords: Pyrometallurgy; Process control; Expert systems; Sulphide ores

1. Introduction
A diagnosis of the instrumentation and control system of a copper concentrate converter (Caletones Smelter, 1977) at Caletones Smelter, Chile, is presented. The
origin of large variations in copper content of white
metal and slug has been investigated. A control strategy
was developed, in order to organize and coordinate the
use of converter supplies to improve products quality.
The strategy also considered the interaction with other
process units and production goals. Similar works have
not been discussed in the literature (Jamsa-Jounela,
2001).

2. Teniente converters in Caletones Smelter

Caletones Smelter is located at the Andes Cordillera,
42 km from Rancagua, at 1556 m over sea level, in

Corresponding author.
E-mail address: luis.bergh@usm.cl (L.G. Bergh).

0892-6875/$ - see front matter
2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mineng.2005.02.007

Chile. Each day, over 3,800 tones of dried copper concentrate are fed into two Teniente converters (TC),
obtaining two products: white metal and slag. The white
metal is then processed in four Peirce Smith Converters
(PSC), obtaining blister copper. The slag from TCs is
processed in four slag cleaning furnaces (SCF). Two
products are obtained: the matte that is returned to
the TCs or PSCs, and the nal slag that is dumped in
the slag deposit.
Dried concentrate is pneumatically conveyed to the
TC. Annular air is supplied to aid in the distribution
of concentrate as it enters the converter. Reverts and
ux are fed into the converter through a small bin, aid
by blown air (garr gun). Air and oxygen are combined
to produce an enriched air stream, blown into the TC.
The calculation of the oxygen ow rate is based on the
combined air ow rates. A ow diagram showing all
the input and output variables, and the local control system, are illustrated in Fig. 1.
The slag and white metal are discontinuously removed
from opposite ends, and collected in ladles for further
processing. Slag and white metal level measurements
are manually taken every three hours. Approximately

1124

L.G. Bergh et al. / Minerals Engineering 18 (2005) 11231126

Reverts

Wet concentrate

Air garr-gun
FIC

Gases

Flux
FIC

FIC

Other
supplies

FIC

TI

Copper
Fe
SiO2
Fe3O4

Copper
TI

TI

AI

AI
LI

TI

Slag

LI
White
metal
FIC

FIC
FIC

Blown air
AI

Oxygen

FIC

FIC

FIC

Annular
Fuel
air
Dry concentrate
Transport air

FIC
Combustion air

Fig. 1. Simplied TC ow and instrumentation diagram.

17 ladles of slag and 7 ladles of white metal are removed

from the TC each shift. Hot gases exit the TC through an
open port at the top of the converter. Gases are cooled,
passed through an electrostatic precipitator and sent to
a gas handling plant for processing.
The TC process is a semi-continuous and autogenuous reactor. A chalcopyrite concentrate rst is thermally
decomposed (Eqs. (1) and (2)), and then iron sulphide
(Eqs. (3) and (4)) and sulphur (Eq. (5)) are oxidised,
and the iron oxide (Eqs. (6) and (7)) form a fayellite slag
in the presence of silica. The main reactions are:
4CuFeS2 ! 2Cu2 S FeS S2

2FeS2 ! 2FeS S2

3FeS 5O2 ! Fe3 O4 3SO2

2FeS 3O2 ! 2FeO 2SO2

S2 2O2 ! 2SO2

2FeO SiO2 ! 2FeO
SiO2

6FeO O2 ! 2Fe3 O4

The copper quality targets for the white metal and slag
are 75% and 9% respectively. Approximately 3 samples
per shift of the white metal are taken and analyzed for
copper concentration. Slag samples are taken and analyzed each time slag is removed from the converter.
The sample is moisture and chemically analyzed.
The current control system is comprised of a number
of single loop controllers. Set points of the converter
manipulated variables are determined by operators
based on the current state of operations. The scheme
for controlling the white metal and slag copper content
is quite complex. Operators can manipulate several variables to aect the copper content in both the white metal

and slag, including: air, oxygen, concentrate, reverts and

ux ow rates.

3. Diagnosis
The current control scheme contains no coordination
of concentrate ow, air ow and oxygen ow rates. The
ux ow controller set point is determined by operators
based on results of the slag chemical analysis, in particular the Fe and SiO2 contents. A target Fe/SiO2 ratio of
approximately 1.6 is desirable in order to decrease slag
viscosity and improve downstream copper recovery.
The reverts controller set point is determined by operators based primarily on the slag temperature measurement. Reverts are used to cool the internal converter
load. There is no automatic strategy in the current control system to maximize concentrate ow rate or reverts
ow rate to maintain the copper concentrate in the white
metal and slag. In order to show the eect of operator
decisions, two sets of operating data for some months
ago are shown in Fig. 2. The rst set on the left was
for an abnormal operation with large variability on
the nal objective variables, while the second set, on
the right, corresponded to a better operation. Each period is two days long.
In general, it was found large product variability and
a considerable recycle of products. The nature of the
process propagates disturbances created in one process
operation to others, increasing unnecessarily the recycling of intermediate products o specications.

4. Supervisory control strategy

To improve the product quality and to decrease
material recycling a better administration of local con-

0.30
0.20
0.10
0.00
09:0

10:0

10:12

11:0

09:12

10:0

10:12

11:0

0.40
0.20
0.00
09:0

White metal Cu

09:12

Slag Cu

0.80
0.75
0.70
0.65
09:0

09:12

10:0

10:12

11:0

Time (d:h)

Slag magnetite

Slag magnetite

Slag Cu

White metal Cu

L.G. Bergh et al. / Minerals Engineering 18 (2005) 11231126

1125

0.80
0.75
0.70
0.65
15:0

15:12

16:0

16:12

17:0

0.30
0.20
0.10
0.00
15:0

15:12

16:0

16:12

17:0

15:12

16:0

16:12

17:0

0.40
0.20
0.00
15:0

Time (d:h)

Fig. 2. Main output data for two dierent operating conditions.

trol set points of input variables must be achieved. If all

input variables were continuously measured and a multivariate dynamic model were available, any classical
multivariate feedback control (for example, Ray, 1981)
could be implemented. However, this control problem
is more complex due to (i) the uncertainty on some
key measurements, (ii) the dierent frequency of some
measurements, (iii) the raw material characteristics
changes over time, (iv) the process complexity to develop phenomenological or empirical models to relate
input with objective variables. Therefore, as its becoming a general practice in mineral processing (Hodouin
et al., 2001), articial intelligence techniques appears
as the more ecient methodology to solve this kind of
problem. Examples of supervisory expert control development are given in Bergh et al. (1999).
To develop a supervisory control strategy the following approach was considered. Every variable can be
classied in one of three groups, representing:
Final objectives (white metal Cu %, slag Cu and magnetite %, inside a narrow band, with high and low
limits).
Intermediate objectives (white metal and slag levels,
bath temperature, slag Fe/SiO2 ratio, oxygen/concentrate ratio and oxygen enrichment, inside a broader
band, with high and low limits).
Basic resources (concentrate, reverts, blown air, oxygen and ux ow rate set points, white metal and slag
remotions, inside a wide band, with high and low
limits).
Final objectives are achieved by regulating intermediate objectives. For example, copper grade in white metal

may be regulated by modifying the oxygen/concentrate

ratio. Furthermore, intermediate objectives are achieved
by regulating basic resources. Following the example,
the oxygen/concentrate ratio may be adjusted by changing the ow rate set point of either the concentrate or the
oxygen. As a multivariate model is not available, the
decision is taken accordingly to other information, in
the form of logic rules in an expert shell.
The supervisory control strategy is then based on an
expert system structure, where the inputs to each procedure come from distributed control system (DCS) internal and external alarms. The output of each procedure
determines the set points values of the DCS basic control loops. The principal objective of the expert system
is to nd at any time the maximum concentrate ow rate
that can be processed considering all the operating internal and external constraints.
The external alarms are the deviation of the processing capacity of other operating units, such as PSC, SCF,
Gas and Euent Treatment Plants and Production and
Maintenance Programs. The state of these alarms is expected to occasionally change. For example, if one PSC
is out of operation, the production of white metal must
be restricted to the actual PSC processing capacity. This
implies that the set point of the concentrate ow rate
must be decreased. Otherwise, an unnecessary material
recycle will begin and the quality of intermediate products will be dicult to be maintained. If the excess of
white metal were temporarily stored in the TC, then
the increment in levels will produce an odd operating
condition and the specications of slag and white metal
will not be met. If the white metal were transferred to
the available PSCs, then the odd operating condition
is propagated to this unit. Similar analysis can be done,

1126

L.G. Bergh et al. / Minerals Engineering 18 (2005) 11231126

for example, when one Gas Treatment Plant has temporarily a lower capacity. If this alarm is ignored then SO2
emissions will increase or the plant operating o specications will be damaged.
The internal alarms are divided in two groups:
Continuous measurements (concentrate, air, oxygen,
ux and reverts ow rates, bath and gas temperature)
and calculated variables (oxygen enrichment, oxygen/
concentrate and ux/concentrate ratios).
Sampled measurements (white metal and slag copper
percentages, slag magnetite, iron and silica percentages, and slag and white metal levels).
The internal alarm procedures are executed every
time the information is available. For continuous measurements they are executed on variable averages over
a period of time. Each procedure consists of processing
logic rules which outputs use a very low number of
parameters to be tuned. The main internal procedures
considered in the system are:
Continuous:
maximization of concentrate ow rate, considering
also the availability of air and oxygen ow rates;
bath temperature control, considering also the
reverts ow rate, the gas temperature and the oxygen enrichment, to modify either the reverts ow
rate, the oxygen enrichment or the concentrate
ow rate.
Discontinuous:
white metal copper concentration control, considering also the oxygen/concentrate ratio to modify
the oxygen/concentrate ratio;
slag copper concentration control, sending warning messages to operator;
slag magnetite concentration control, considering
also the slag Fe/SiO2 ratio to modify this ratio or
to send warning messages;
white metal level control, sending warning messages to start remotion of white metal;
slag level control, sending warning messages to
start remotion of slag.
The optimization routine to maximize the concentrate ow rate set point considered all the alarm states.
However, external alarms override decisions based on

internal alarms, as a way to insure that product recycling is minimised. The output of these procedures also
species the set points of all the intermediate objective
variables. The next stage is the calculus of the basic resource set points, based on stoichiometric relationships.

5. Conclusions
The Teniente Converter is a very complex process
where the key variables are measured infrequently and
sometimes with considerable errors. The input variables
are all under DSC control. However, operating decisions usually lead to large variability of product quality
and hence to unnecessary recycling of material between
the units in the smelter.
A supervisory control strategy has been proposed
based on experimental studies in the form of an expert
system. Complex procedures has been developed to consider internal abnormal situations and to consider external constraints. These procedures provide a consistent
methodology to administrate the local control loops in
harmony with the temporal capacity of the other units
in the smelter.

Acknowledgement
The authors would like to thanks Conicyt (Project
Fondecyt 1020215) and Santa Maria University (Project
270322) for their nancial support and to El Teniente
for allowing the presentation of this work.

References
Bergh, L.G., Yianatos, J., Acuna, C., Perez, H., Lopez, F., 1999.
Supervisory Control at Salvador Flotation Columns. Minerals
Engineering 12 (7), 733744.
Fundicion Caletones, 1977. Division El Teniente, Codelco-Chile,
Patent 30226, A device for continuous fusion and conversion of
copper concentrates, October 1977.
Hodouin, D., Jamsa-Jounela, S-L., Carvalho, M.T., Bergh, L.G.,
2001. State of art and challenges in mineral processing control.
Control Engineering Practice 9 (9), 10071012.
Jamsa-Jounela, S-L., 2001. Current status and future trends in the
automation of Mineral and Metal Processing. Control Engineering
Practice 9 (9), 10191024.
Ray, W.H., 1981. Advanced Process Control. MacGraw-Hill Inc.