See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/233487271

An Operator Training Simulator System for MMM Comminution and

Classification Circuits

Conference Paper · September 2012

DOI: 10.13140/RG.2.1.2248.1686

CITATIONS READS

5 585

3 authors, including:

Rodrigo Toro
Honeywell
8 PUBLICATIONS 86 CITATIONS

SEE PROFILE

All content following this page was uploaded by Rodrigo Toro on 04 June 2014.

The user has requested enhancement of the downloaded file.

 An Operator Training Simulator System for
MMM Comminution and Classification
Circuits 
Rodrigo Toro ∗ José Manuel Ortiz ∗ Iván Yutronic ∗∗
∗
Honeywell Chile S.A., Santiago, Chile
(e-mail: [email protected], [email protected])
∗∗
Compañı́a Minera Doña Inés de Collahuasi, Iquique, Chile
(e-mail: [email protected])

Abstract: The high copper prices, together with the lack of experienced personnel in mineral
processing operations, have lead to high investments in technology for this industry. Simulation
based training is becoming a necessary phase for new projects and adoption of new technologies –
like APC (Advanced Process Control)– because it is an instance to transmit the best operational
practices over a safe environment. In this paper, an Operator Training Simulator (OTS) system
–for grinding and classification circuits– is presented. A case study, in which a complete SAG
and secondary grinding/classification circuit has been simulated; supervisors and operators have
been trained in order to educate them on the best operational practices, using APC applications.
The OTS sessions have lead to an increase in the utilisation of APC applications from 20% to
70%, in average.

Keywords: Simulators, population balance models, advanced process control.

1. INTRODUCTION The main contribution of this paper is a confirmation of

the fact that the use of OTS systems helps to transmit
Operator training simulators have been widely used in the the best practices and reduce the adoption time of new
chemical industry (see Coker and Ludwig (2007)), and technologies–such as APC–, giving the student the confi-
are being recently incorporated in the mineral processing dence to confront complicated operational events through
industry (see Malhotra et al. (2009) and Sbárbaro and the experimentation in a safe environment.
Villar (2010)). There are many simulators for the mining This paper has been structured in the following way: In
industry, mainly focused on the plant design and therefore Section 2, a brief description of the dynamic models used in
based on steady-state models (see Roine et al. (2011)), the simulation is presented. In Section 3, the OTS system
rather than dynamic models, which are the ones used architecture is reviewed. In Section 4, a case study is
in OTS systems due to the need to simulate transient presented, after, in Section 5, some results, taken from the
responses and automatic control strategies. case study, are discussed. Finally, in Section 6, conclusions
The expected benefits of the use of OTS systems are: faster and future work are presented.
and safer ramp-up for new projects, enhancements in the
learning process effectiveness and faster adoption of new 2. DESCRIPTION OF COMMINUTION MODELS
control technologies, among others.
The population balance model was introduced by Epstein
On this paper, an OTS system for a copper grinding and (1947) with further development by many authors, in-
classification process is presented. The simulated equip- cluding Kellsall et al. (1969), Whiten (1974), Herbst and
ment was modelled using dynamic population balance Fuerstenau (1968, 1973), and Austin and Weller (1982).
models, implemented on Matlab R
S-Functions. The OTS The model concepts are widely used in modern process
system was deployed using OPC connections between the simulation.
simulation server and the operator stations. Advanced pro-
cess control–APC–applications, were installed in operator The models used to simulate the process are based
stations with the purpose of instructing students in the in the Modern Comminution and Classification Theory
correct use of those applications. An automatic evaluation (see Gutiérrez and Sepúlveda (1986)). For comminu-
method has been implemented in order to have indicators tion/classification simulations, n + 1 size intervals have
of the student performance through the pre-programmed been defined, where the ith interval is delimited by the
operational events or cases. upper bound of size i − 1, named di−1 , and delimited by
the lower bound of size i, named di , in such a way that,
 Work of R. Toro and J. M. Ortiz has been supported by Honeywell d0 > d1 > . . . > di > di+1 > . . . > dn > dn+1 , (1)
Chile S.A. where,
Work of I. Yutronic has been supported by Compañı́a Minera Doña
Inés de Collahuasi. • n, is the number of size classes; and
 • dn+1 = 0. the fraction of the total volume that is used by the mass
holdup (including balls); and α is the lift angle of the load.
At every size interval, the mass holdup in the comminution
equipment has been modelled using a general population The lift angle varies with the density of the slurry. A linear
balance model, defined as follows: relation is adjusted in such a away that the lift angle is
i−1
 lower for higher values of slurry density. As a result, the
dHi
= −Si Hi + Si bij Hj + MiF − MiP , (2) denser the slurry, the lower will the power draw be. This
dt j=1 has correspondence with what is observed in real plants.
where,
2.2 SAG Mill
• Hi , is the mass holdup in the ith size class. Hence,
the total mass holdup is calculated as H = i Hi ; SAG mills have a discharge grate, which works as a
• Si , represents the ith term of the selection function, classification device.
that is, the disappearance rate particles of size i; The product mass flow is also calculated considering
• bij , represents the breakage function term, that is, the perfect mix of the slurry in the mill, but in this case the
probability that a particle of size i becomes of size j discharge grate adds the following term to equation (3),
as it breaks;
Hi
• MiF , is the mass flow of particles of size i being fed MiP = (1 − Cgi ) q P , (7)
into the comminution equipment; Vs
• MiP , is the mass flow of particles of size i that come where Cgi is the term of the discharge grate’s classification
out as product. function corresponding to size i.
For every plant, parameters Si and bij are calculated from In this case, qP depends on the friction imposed by the
a desired steady-state result. That is, from a particle size grate and the pressure difference on the discharge. As a
distribution of the feed and product (e.g. samples taken consequence, Vs is calculated from the total mass holdup
from the real plant), these parameters are adjusted using and the components’ densities.
optimization techniques, in such a way that the desired The water mass holdup depends on its own mass balance
product is obtained from that particular feed. equation,
The product mass flow value MiP , is calculated differently dHw
= MwF − MwP , (8)
depending on the type of comminution equipment. The dt
F
most common ones will be analysed next. where Mw is the water mass flow fed into the mill.
The product’s water mass flow is calculated assuming
2.1 Ball Mill perfect mix, yielding:
Hw P
The product mass flow is calculated considering perfect MwP = q , (9)
mix of the slurry in the mill. Hence, Vs
Hi P
MiP = q , (3) 2.3 Hydrocyclones
Vs
where Vs is the total slurry volume contained in ball voids The Hydrocyclone model is based on Plitt’s model. The
and above balls level, and q P is the volume flow of product. classification curve depends on the feed pressure on the
Since a ball mill discharges by overflow, Vs is constant and cyclone’s inlet, and the percent of solids of the slurry being
hence, the product volume flow is equal to the feed volume fed.
flow.
The inlet pressure will influence on how steep the classifi-
Furthermore, given that the load volume is considered con- cation curve will result. The higher the pressure, the closer
stant, the water mass holdup is calculated as by difference: to the ideal classifier. The pressure calculation depends on
H the feed volumetric flow, the geometry of the cyclone and
Hw = ρw (Vs − ), (4) the percent of solids of the feed.
ρo
where ρw is the water density and ρo is the ore density. The percent of solids in the feed influences on the fines
and water by-pass, that is, the percentage of water and
Similarly, the water mass flow of product is given by fine particles that appear on the cyclone’s underflow.
n P
M Moreover, the percent of solids determines the cut size
Mw = ρw (q P − i i ), (5) of the classification curve.
ρo
where ρo is the ore density. For more details on the Hydrociclone model, see Gutiérrez
The power drawn by the mill is calculated as follows: and Sepúlveda (1986).
PN et = 0.238 · D2.5 · L · nc · ρap (Jt − 1.065Jt2 ) sin α, (6) 2.4 Other Models
where D and L are the diameter and Length of the mill Other models used on this training include:
respectively, nc is the percentage of critical speed that
the mill is operating on; ρap is the apparent density, that • Sump: A simple tank model (basic mass balance).
is, the density of what is contained in the mill, including • Pump: Basic time lag, with speed as input and
balls and slurry; Jt is the load fraction of the mill, that is, volumetric flow as output.
 • Conveyor Belt: implemented using a time transport. HMI Station 01
+
Other models developed by the group include: flotation APC

Local Area Network

Simulation Server

cells, thickeners, and crushers. HMI Station 02

+ OPC
APC Server

...
3. THE OPERATOR TRAINING SYSTEM HMI Station 06
+ Instructor Station
APC
3.1 Models Library

Based on the above described models, a modular Matlab Fig. 2. OTS System Architecture
& SimulinkR
library was developed. Each module can be run on virtual machines installed on dual-core CPU work
parameterised to represent a specific site and connected stations with 2GB of RAM.
depending on the desired plant configuration.
With the above configuration, the simulation server can
The implementation was based on Matlab R
S-Functions. handle 6 instances of the simulated process, running in
These functions can represent any dynamic model by the real-time, using a backward euler integrator scheme.
use of the following structure:
3.3 Evaluation Strategy
ẋ = F(x, u)
y = G(x, u) Each case is evaluated with a score, which is calculated
using an excursion-based method, in which the student
where u is the input, y the output, and x the model’s
starts with the maximum score and every time a KPI
states.
has an excursion from its pre-defined limits the score
In the case of Ball and SAG Mills, the states are exactly is penalised proportionally to the excursion time. The
the mass holdups in each size interval plus the water mass excursion time of every KPI is recorded, and the final score
holdup. is calculated as follows: 
(ωi · texi )
Score = i  · 100, (10)
ttotal · i ωi
where,
• Score ∈ [0, 100], is the final score for a training
exercise;
• texi , is time where the ith KPI was outside the high
or low limit;
• ttotal , is the total simulation time; and
• ω i ∈ R+ 0 , is a weighting factor, that may be greater
than 1.

 
 



Fig. 1. View of Simulink

R
developed Library.  

 
For supporting the simulation, some extra modules were  
developed, including: feeders, mineral and water sources,

PID controllers with PV tracking, manual and auto mode,
among others. On Fig. 1, a view of the developed library
for grinding and classification is shown. Fig. 3. Excursion-based Evaluation

3.2 System Architecture The idea behind the weighting factor is to give the pos-
sibility to prioritise one KPI above another. In Fig.3,
The OTS system architecture is presented on Fig.2. It is a graphical representation of the evaluation method is
composed of one simulation server, in which 6 simulation presented.
instances are running in parallel. Inside the simulation
server, an OPC server is configured, acting as a simulation 4. CASE STUDY
data hub. Student stations are based on virtual machines,
in which a .NET HMI of the process is installed, together The case study of this paper is the grinding and classifi-
with the APC platform, based on Honeywell’s Profit cation circuit – Line 3 of Compañı́a Minera Doña Inés de
ControllerR
. Collahuasi (CMDIC).
The simulation server has an 8-core Intel
R
Xeon
R
CPU, Operations personnel started a project in which they
running at 2.66Ghz with 8GB of RAM. Operator stations want to educate their engineers and operators on the best
operational practices, as well as encouraging the utilisation • Measured Disturbances: fresh ore mass flow (fed to
of APC applications installed on the SAG and secondary SAG Mill), and fresh water added to the centralised
grinding-classification units. sump.

Feeders
An HMI, based on the DCS schematic of CMDIC, was
created in order to give the student a working environment
SAG
Pebbles
as similar to the real one as possible. Fig.5, presents
WIC
1011
Crusher
the HMI created for CMDIC OTS sessions. The main
regulatory control loops have been reproduced in the
simulation platform, giving the opportunity to transmit
To Stock
Pile
basic control theory concepts to the students, such as
wind-up situations, and how to carry out regulatory-
SECONDARY GRINDING APC control evaluations, through the APC operation interface.
Dilution
FIC
Water
LI

To To
Flotation AI AI Flotation
DI DI
PI PI

AI AI
DI DI

PI PI

JI JI

BallMill BallMill
1012 1013

Fig. 4. Case Study Layout

Line 3 of grinding and classification circuit is composed

of a 40x22 feet gearless SAG Mill (one of the biggest
SAG mills in the world). Its product enters a screen, from
where the oversized particles are transported to a pebbles
crusher plant and the undersize to a common sump. Four Fig. 5. OTS System HMI
pumps take the slurry from the sump to four hydrocyclone
batteries–with 10 hydrocyclones per battery–, overflow
goes to rougher flotation and undersize to two 26x38 feet 4.1 OTS Scenarios
Ball Mills. The process layout is presented on Fig.4.
For the training sessions, the first scenarios were centred
There are two APC applications on that unit. One is the in SAG Mill operations, followed by secondary grinding
SAG mill, presented on Toro and Yutronic (2010), and plus classification operations. The final scenario was about
the other is the secondary grinding and classification ap- coordinated operation. The defined scenarios were the
plication, reviewed on Yutronic et al. (2011). The control following:
variables associated to APC applications are the following:
SAG Mill Application (1) SAG Mill overload due to an increase on the ore
coarseness;
• Controlled Variables: Mill weight (as a measure of the (2) Low Stock-Pile level and its effects on the plant;
internal load), noise (as a measure of impacts inside (3) Constrained operation, with low water network pres-
the mill), power consumption, torque, and mass flow sure;
of produced pebbles (screen oversize), (4) Secondary grinding. Constrained operation with sud-
• Manipulated Variables: fresh ore feed mass flow, Mill den pump malfunction events and low water avail-
speed, and dilution water in terms a solid percentage ability;
ratio (controlled by a PI regulator), (5) Operational optimisation of the integrated circuit.
• Measured Disturbances: percentage of fine ore in the
fresh feed, and returned pebbles. 4.2 Evaluation and APC Operational Reports
Secondary Grinding and Classification APC Application
For CMDIC training sessions, two types of evaluations
• Controlled Variables: hydrocyclone batteries pres-
were carried out. The first one is the automatic evaluation,
sure, level of centralised sump, +65# percentage (on
explained in Section 3.3. Also, a theoretical evaluation
hydrocyclone batteries overflow), solid percentage of
has been carried out at the end of the practical sessions,
hydrocyclone batteries feed, and power consumption
with the purpose of checking if the relevant ideas were
of ball mills,
transmitted correctly to the students.
• Manipulated Variables: speed of each pump attached
to the centralised sump, mass flow of process water Another feature that helped to reinforce key concepts was
(recovered) added to the centralised sump, the adaptation of the APC Operational Reports, used to
evaluate operational shifts in a daily basis. Those reports simulation platform. Additionally, crushing, flotation and
were analysed after the application of each scenario, and thickening models are being developed, in order to cover
extrapolated to a complete shift in order to transmit to the the complete sulphur-concentration process.
students the impact of their decisions in the operational
KPIs. ACKNOWLEDGEMENTS
5. RESULTS The authors would like to acknowledge the support and
contributions of the people involved in the development of
The evaluation of the impact of an OTS training is not the CMDIC project, specially to Claudio Zamora, Héctor
straight forward. Therefore, the results presented in this Slater, René Fuenzalida, Eduado–Lalo–Cubillos, Carmen
section are an indirect measure of this impact. Gloria Cortés, José Olivares and Sandy Ramı́rez.
One of the consequences of OTS sessions is an increase
in the utilisation of the APC applications, mainly be- REFERENCES
cause training gives the student a better understanding
Austin, L.G. and Weller, K.R. (1982). Simulation and
of how advanced control strategies work. In Fig.6, the
scale-up of wet ball milling. In XIV Int Min Proces
daily effective utilisation of the secondary grinding APC
Congress.
application, for a six month period starting one month
Coker, A. and Ludwig, E. (2007). Ludwig’s applied process
before the first training session took place, is presented
design for chemical and petrochemical plants. Elsevier
in grey bars. A 28-day moving average of the effective
Gulf Professional Pub.
utilisation is presented on dashed black lines. OTS sessions
Epstein, B. (1947). The material description of certain
are presented as black bars, with a 140% value, labeled
breakage mechanisms leading to the logaritmic-normal
“Training”, and the maintenance stops, in a brick-like red
distribution. Journal of the Franklin Institute, 244, 471–
colour, with a 120% value, are labelled as “Maintenance”.
477.
Note that after the first three sessions the utilisation of
Gutiérrez, L. and Sepúlveda, J. (1986). Dimension-
the APC application increased substantially. In January,
amiento y optimización de plantas concentradoras me-
a previously scheduled maintenance in the dynamic models
diante técnicas de modelación matemática. CIMM.
of the APC applications was carried out, which explains
Herbst, J.A. and Fuerstenau, D.W. (1968). The zero order
the utilisation decrease.
production of fines in comminution and its implication
Training
in simulation. Trans SME/AIME, 241, 531–549.
Maintenance
Herbst, J.A. and Fuerstenau, D.W. (1973). Mathemati-
cal simulation of dry ball milling using specific power
100 information. Trans SME/AIME, 254, 348–354.
Kellsall, D.F., Reid, K.J., and Stewart, P.S.B. (1969). The
80 study of grinding processes by dynamic modelling. Elec
Eng Trans Eng, EE5(1), 155–169.
60 Malhotra, D., Taylor, P., Spiller, E., and LeVier, M.
(2009). Recent Advances in Mineral Processing Plant
40
Design. Society for Mining Metallurgy.
Roine, T., Kaartinen, J., and Lamberg, P. (2011). Training
20
simulator for flotation process operators. In Preprints
0
of the 18th IFAC World Congress.
Jul−06 Aug−01 Sep−01 Oct−01 Nov−01 Dec−01 Jan−01 Sbárbaro, D. and Villar, R. (2010). Advanced Control and
Supervision of Mineral Processing Plants. Advances in
Fig. 6. Effective APC Utilisation, Maintenances and OTS Sessions Industrial Control. Springer.
Toro, R. and Yutronic, I. (2010). Design and implemen-
6. CONCLUSIONS AND FUTURE WORK tation of advanced automatic control strategy based on
dynamic models for high capacity sag mill. In Proceed-
A confirmation of the positive impact that the OTS ings of The Canadian Mineral Processors 42nd Annual
sessions had on the operation is the APC applications Conference.
utilisation increase. The confidence given to the students Whiten, W.J. (1974). A matrix theory of comminution
to face difficult decisions in every-day opperation allowed machines. Chem Eng Sci, 29, 588–599.
for safer driving of the plant towards the optimum. Yutronic, I., Espinoza, P., and Olivares, J. (2011). Sag
OTS sessions gave operators the opportunity to become & ball mill control by model predictive controllers on
familiar with the best practices, defined by operations and 3 lines at collahuasi. In Proceedings of The SAG 2011
metallurgical supervisors. Those practices have helped in- Congress.
creasing APC applications utilisation, minimise the events
of constrained controllers by prompt detection of equip-
ment malfunction and water constraints, get an integrated
view of the process, avoid production losses and maintain
product quality.
Currently, the simulation team is migrating the Matlab &
SimulinkR
models to UniSim R
, a Honeywell’s commercial

View publication stats