858 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
and formulated innovative strategies and methodologies
that are applicable in industrial settings.
Summary of Findings
The core findings of this research demonstrate the signifi-
cant potential of online reinforcement learning, a branch
of machine learning, for intelligent control in industrial
processes:
• Identification of Key Variables: The research suc-
cessfully identified critical operational parameters
and control features, laying a robust foundation for
subsequent modeling and control strategies.
• Digital Twin Development: The creation of a data-
based digital twin has been pivotal. This high-fidelity
simulation mirrors the intricate dynamics of grinding
processes, offering a platform for predictive analysis
and operational experimentation without real-world
interruptions.
• Training Environment for Reinforcement Learning:
A specialized training environment was developed,
tailored to the unique requirements of the grinding
circuit. This environment’s adaptability and scalabil-
ity highlight its capability to handle diverse opera-
tional data effectively.
• Training of Reinforcement Learning Models: The
research meticulously trained and evaluated both
on-policy and off-policy reinforcement learning
algorithms, which demonstrated high efficacy and
robustness, indicating their substantial potential for
real-world applications.
These contributions not only advance the field of mineral
processing but also set a precedent for future research and
development in intelligent industrial control systems. The
exploration of advanced machine learning techniques,
particularly in autonomous and adaptive control, opens
Figure 6. Trends showcasing changes in recirculating load and other operational parameters during a shift in target product size
and formulated innovative strategies and methodologies
that are applicable in industrial settings.
Summary of Findings
The core findings of this research demonstrate the signifi-
cant potential of online reinforcement learning, a branch
of machine learning, for intelligent control in industrial
processes:
• Identification of Key Variables: The research suc-
cessfully identified critical operational parameters
and control features, laying a robust foundation for
subsequent modeling and control strategies.
• Digital Twin Development: The creation of a data-
based digital twin has been pivotal. This high-fidelity
simulation mirrors the intricate dynamics of grinding
processes, offering a platform for predictive analysis
and operational experimentation without real-world
interruptions.
• Training Environment for Reinforcement Learning:
A specialized training environment was developed,
tailored to the unique requirements of the grinding
circuit. This environment’s adaptability and scalabil-
ity highlight its capability to handle diverse opera-
tional data effectively.
• Training of Reinforcement Learning Models: The
research meticulously trained and evaluated both
on-policy and off-policy reinforcement learning
algorithms, which demonstrated high efficacy and
robustness, indicating their substantial potential for
real-world applications.
These contributions not only advance the field of mineral
processing but also set a precedent for future research and
development in intelligent industrial control systems. The
exploration of advanced machine learning techniques,
particularly in autonomous and adaptive control, opens
Figure 6. Trends showcasing changes in recirculating load and other operational parameters during a shift in target product size