XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 849
mineral processing. The integration of digital twin technol-
ogy and reinforcement learning in grinding circuits repre-
sents a significant technological leap. Digital twins allow
for real-time monitoring and control, providing a platform
for operational experimentation and optimization. When
combined with reinforcement learning algorithms, these
systems offer a dynamic, adaptive approach to process con-
trol, surpassing traditional methods.
The aim of this study is to create an autonomous con-
trol system for dry grinding circuits through a data-driven,
machine learning approach. The research encompasses the
identification of key variables, the construction of a digital
twin for the grinding circuit, the design of a reinforcement-
learning model, and discusses some initial thoughts about
the development of an edge-computing layout for indus-
trial implementation. In doing so, this research addresses
the pressing need for energy efficiency, operational safety,
and sustainability in the mineral processing industry.
By bridging the gap between theoretical research and
practical application, this study not only contributes to the
field of intelligent process control but also provides action-
able solutions to enhance the efficiency and sustainability of
mineral processing operations, leading the industry towards
a more sustainable and more efficient future.
LITERATURE REVIEW
Introduction: The Necessity of Grinding Circuit
Optimization
This research presents an in-depth exploration of dry grind-
ing circuits in mineral processing, with a particular focus
on a configuration that is frequently encountered in cement
production. As demonstrated in Figure 1, the study con-
centrates on a specific setup: a two-compartment ball mill
paired with a dynamic separator. This choice is based on the
widespread use of this configuration in the industry and its
significant potential for operational optimization.
The methodologies developed in this research, while
tailored to this specific layout, are designed to be modular
and adaptable. They could be extended to various circuit
layouts in mineral processing, underlining the versatil-
ity of the proposed approach. This adaptability makes the
research valuable not just for the specific grinding circuit
under study but also for broader applications in mineral
processing.
The study intentionally omits the analysis of pre-grind-
ing equipment such as roller presses, which are commonly
integrated into modern grinding circuits. Roller presses
induce micro-fractures in the material, thereby reducing the
energy required for grinding and enhancing overall energy
efficiency. However, incorporating a roller press would
require specific evaluations tailored to individual plants, a
complexity beyond the scope of this study. This aspect of
comminution technology, and its implications for energy
efficiency in grinding circuits, is extensively discussed in
the work of (Forssberg &Yanmin, 2003). Additionally,
this research does not delve into other physical optimiza-
tion methods like optimizing the liners or grinding media
distribution, which can also significantly improve grinding
efficiency. Instead, the focus is on optimizing operational
parameters within the preexisting setup.
Evolution of Control Strategies in Grinding Circuits
The realm of control systems in grinding circuits has wit-
nessed a significant evolution, marked by the transition
from manual adjustments to more complex automated sys-
tems. This shift is reflective of the ongoing efforts to address
the challenges posed by the intricate nature of grinding
operations.
Proportional-Integral-Derivative (PID) Controllers
Historically, the workhorse of control systems in these
circuits has been the Proportional-Integral-Derivative
(PID) controllers. These systems have been extensively uti-
lized due to their simplicity and ease of implementation.
Studies by (Edwards, Vien, &Perry, 2002 Wei &Craig,
2009a) underline the widespread adoption of PID control-
lers in the industry. However, the inherent limitations of
PID controllers, particularly their inability to adapt to the
dynamic and complex nature of grinding processes, have
been a point of concern.
Alternative Control Methodologies
In addition to PID controllers, there are alternative control
methodologies that can be paired with them to enhance
their capability, each with its unique strengths and chal-
lenges. The literature highlights several of these strategies:
• Step Controllers: Designed to maintain system
variables within a specified range, making discrete
adjustments in response to deviations from a set
threshold. Their application is advantageous in sys-
tems not designed for continuous updates to the
control value.
• Fuzzy Logic Controllers: These controllers use
“If-Then” rules to manage uncertainties and approxi-
mate reasoning, suitable for complex systems where
variables do not conform to strict binaries (Radha
Krishna &Biswal, 2016 Zadeh, 1973).
• Model Predictive Control (MPC): MPC represents
a significant advancement, utilizing dynamic mod-
els to predict and optimize future system behaviors
mineral processing. The integration of digital twin technol-
ogy and reinforcement learning in grinding circuits repre-
sents a significant technological leap. Digital twins allow
for real-time monitoring and control, providing a platform
for operational experimentation and optimization. When
combined with reinforcement learning algorithms, these
systems offer a dynamic, adaptive approach to process con-
trol, surpassing traditional methods.
The aim of this study is to create an autonomous con-
trol system for dry grinding circuits through a data-driven,
machine learning approach. The research encompasses the
identification of key variables, the construction of a digital
twin for the grinding circuit, the design of a reinforcement-
learning model, and discusses some initial thoughts about
the development of an edge-computing layout for indus-
trial implementation. In doing so, this research addresses
the pressing need for energy efficiency, operational safety,
and sustainability in the mineral processing industry.
By bridging the gap between theoretical research and
practical application, this study not only contributes to the
field of intelligent process control but also provides action-
able solutions to enhance the efficiency and sustainability of
mineral processing operations, leading the industry towards
a more sustainable and more efficient future.
LITERATURE REVIEW
Introduction: The Necessity of Grinding Circuit
Optimization
This research presents an in-depth exploration of dry grind-
ing circuits in mineral processing, with a particular focus
on a configuration that is frequently encountered in cement
production. As demonstrated in Figure 1, the study con-
centrates on a specific setup: a two-compartment ball mill
paired with a dynamic separator. This choice is based on the
widespread use of this configuration in the industry and its
significant potential for operational optimization.
The methodologies developed in this research, while
tailored to this specific layout, are designed to be modular
and adaptable. They could be extended to various circuit
layouts in mineral processing, underlining the versatil-
ity of the proposed approach. This adaptability makes the
research valuable not just for the specific grinding circuit
under study but also for broader applications in mineral
processing.
The study intentionally omits the analysis of pre-grind-
ing equipment such as roller presses, which are commonly
integrated into modern grinding circuits. Roller presses
induce micro-fractures in the material, thereby reducing the
energy required for grinding and enhancing overall energy
efficiency. However, incorporating a roller press would
require specific evaluations tailored to individual plants, a
complexity beyond the scope of this study. This aspect of
comminution technology, and its implications for energy
efficiency in grinding circuits, is extensively discussed in
the work of (Forssberg &Yanmin, 2003). Additionally,
this research does not delve into other physical optimiza-
tion methods like optimizing the liners or grinding media
distribution, which can also significantly improve grinding
efficiency. Instead, the focus is on optimizing operational
parameters within the preexisting setup.
Evolution of Control Strategies in Grinding Circuits
The realm of control systems in grinding circuits has wit-
nessed a significant evolution, marked by the transition
from manual adjustments to more complex automated sys-
tems. This shift is reflective of the ongoing efforts to address
the challenges posed by the intricate nature of grinding
operations.
Proportional-Integral-Derivative (PID) Controllers
Historically, the workhorse of control systems in these
circuits has been the Proportional-Integral-Derivative
(PID) controllers. These systems have been extensively uti-
lized due to their simplicity and ease of implementation.
Studies by (Edwards, Vien, &Perry, 2002 Wei &Craig,
2009a) underline the widespread adoption of PID control-
lers in the industry. However, the inherent limitations of
PID controllers, particularly their inability to adapt to the
dynamic and complex nature of grinding processes, have
been a point of concern.
Alternative Control Methodologies
In addition to PID controllers, there are alternative control
methodologies that can be paired with them to enhance
their capability, each with its unique strengths and chal-
lenges. The literature highlights several of these strategies:
• Step Controllers: Designed to maintain system
variables within a specified range, making discrete
adjustments in response to deviations from a set
threshold. Their application is advantageous in sys-
tems not designed for continuous updates to the
control value.
• Fuzzy Logic Controllers: These controllers use
“If-Then” rules to manage uncertainties and approxi-
mate reasoning, suitable for complex systems where
variables do not conform to strict binaries (Radha
Krishna &Biswal, 2016 Zadeh, 1973).
• Model Predictive Control (MPC): MPC represents
a significant advancement, utilizing dynamic mod-
els to predict and optimize future system behaviors