881
A Preliminary Non-Linear Model Predictive Control for Pressing
Iron Ore Concentrates in an Industrial HPGR
Túlio M. Campos, Horacio A. Petit. Luís Marcelo M. Tavares
Universidade Federal do Rio de Janeiro
Ricardo O. de Freitas
Vale S.A.
ABSTRACT: Use of mathematical modeling of high-pressure grinding rolls (HPGRs) has achieved an
outstanding position in the minerals industry. In this regard, new dynamic modeling and simulation approaches
have been evolving to capture information of the process in real-time using online tools. Nevertheless, proper
application of these models as part of model predictive control algorithms is still missing. The present work
proposes a preliminary non-linear model predictive control (NMPC) algorithm based on the Modified Torres
and Casali model for pressing iron ore concentrates in an industrial HPGR. The Modified Torres and Casali
model is used to emulate the plant performance, besides being the internal model of the controller. Application
of the NMPC is demonstrated by assessing its ability to manipulate the operating pressure and roll peripheral
velocity in order to keep the HPGR throughput and the quality of product Blaine specific surface area (BSA) in
a reasonable setpoint value based on real-time variabilities in the HPGR feed BSA.
INTRODUCTION
Use of high-pressure grinding rolls (HPGRs) achieved sig-
nificant maturity in the minerals industry, with applica-
tions ranging from the cement industry (Aydoğan et al.,
2006) to fine iron ore concentrates (Van der Meer, 1997
Campos et al., 2021), and crushing stages with different
ores (Michaelis, 2009).
Given this level of flexibility to be applied in a wider
range of operations, proper process optimization and con-
trol strategies are required to keep the machine operating
at its most efficient mode. Usually, the control strategies
adopted in HPGR operations rely on speed and torque reg-
ulation (Jones, 2012). These attempt to ensure that HPGRs
will be working close to the maximum allowed torque, thus
being able to provide the finest possible product size dis-
tribution. For this strategy, hydraulic pressure or operat-
ing gap are the main manipulated variables used to achieve
the required target for size reduction, while roll peripheral
velocity is in charge of controlling the machine throughput.
Although the use of torque regulation has resulted in rea-
sonable improvements in some HPGR operations, the lack
of accuracy when capturing variations on roll surface wear
and uneven feed clearly show its limitations (Gardula et al.,
2015). For the case of HPGRs operated in pressing iron
ore concentrates in some pelletizing plants from Vale S.A.
(Brazil), machines are often controlled using fuzzy logic
with the aim of defining a setpoint of pressure to maxi-
mize the torque and to keep it in a given operational range,
A Preliminary Non-Linear Model Predictive Control for Pressing
Iron Ore Concentrates in an Industrial HPGR
Túlio M. Campos, Horacio A. Petit. Luís Marcelo M. Tavares
Universidade Federal do Rio de Janeiro
Ricardo O. de Freitas
Vale S.A.
ABSTRACT: Use of mathematical modeling of high-pressure grinding rolls (HPGRs) has achieved an
outstanding position in the minerals industry. In this regard, new dynamic modeling and simulation approaches
have been evolving to capture information of the process in real-time using online tools. Nevertheless, proper
application of these models as part of model predictive control algorithms is still missing. The present work
proposes a preliminary non-linear model predictive control (NMPC) algorithm based on the Modified Torres
and Casali model for pressing iron ore concentrates in an industrial HPGR. The Modified Torres and Casali
model is used to emulate the plant performance, besides being the internal model of the controller. Application
of the NMPC is demonstrated by assessing its ability to manipulate the operating pressure and roll peripheral
velocity in order to keep the HPGR throughput and the quality of product Blaine specific surface area (BSA) in
a reasonable setpoint value based on real-time variabilities in the HPGR feed BSA.
INTRODUCTION
Use of high-pressure grinding rolls (HPGRs) achieved sig-
nificant maturity in the minerals industry, with applica-
tions ranging from the cement industry (Aydoğan et al.,
2006) to fine iron ore concentrates (Van der Meer, 1997
Campos et al., 2021), and crushing stages with different
ores (Michaelis, 2009).
Given this level of flexibility to be applied in a wider
range of operations, proper process optimization and con-
trol strategies are required to keep the machine operating
at its most efficient mode. Usually, the control strategies
adopted in HPGR operations rely on speed and torque reg-
ulation (Jones, 2012). These attempt to ensure that HPGRs
will be working close to the maximum allowed torque, thus
being able to provide the finest possible product size dis-
tribution. For this strategy, hydraulic pressure or operat-
ing gap are the main manipulated variables used to achieve
the required target for size reduction, while roll peripheral
velocity is in charge of controlling the machine throughput.
Although the use of torque regulation has resulted in rea-
sonable improvements in some HPGR operations, the lack
of accuracy when capturing variations on roll surface wear
and uneven feed clearly show its limitations (Gardula et al.,
2015). For the case of HPGRs operated in pressing iron
ore concentrates in some pelletizing plants from Vale S.A.
(Brazil), machines are often controlled using fuzzy logic
with the aim of defining a setpoint of pressure to maxi-
mize the torque and to keep it in a given operational range,