850 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
(Qin &Badgwell, 2003 Rawlings &Mayne, 2009).
Despite its capabilities, the computational intensity
of MPC and the need for continuous model updates
present practical challenges.
A comparative analysis of conventional control meth-
ods (such as PID, step and fuzzy-logic controllers) and
advanced methods (like Model Predictive Control (MPC))
reveals the limitations of traditional approaches in adapting
to the dynamic nature of grinding processes. The literature
suggests that while conventional controllers are simple and
widely used, they fall short in managing complex systems
(Costea, et al., 2014 Pomerleau, Hodouin, Desbiens,
&Gagnon, 2000). Advanced methods like MPC have
historically faced challenges in computational intensity
and model upkeep (Qin &Badgwell, 2003 Rawlings &
Mayne, 2009), but recent advancements in computational
hardware and machine learning technologies have begun to
overcome these hurdles.
The increasing power of modern processors and the
advent of state-of-the-art machine learning techniques have
greatly reduced concerns around computational demands,
making real-time analysis and model adjustment feasible
even in complex industrial environments. Furthermore,
machine learning’s inherent adaptability offers a promising
1. Fresh feed bunker
2. Ball mill
3. a: dynamic separator b: mill dedusting filter &fan
4. product filter
5. CEOPS particle size distribution measurement
Figure 1. Schematic representation of a typical dry grinding circuit as considered in this research, emphasizing the specific
configuration of a two-compartment ball mill paired with a dynamic separator
(Qin &Badgwell, 2003 Rawlings &Mayne, 2009).
Despite its capabilities, the computational intensity
of MPC and the need for continuous model updates
present practical challenges.
A comparative analysis of conventional control meth-
ods (such as PID, step and fuzzy-logic controllers) and
advanced methods (like Model Predictive Control (MPC))
reveals the limitations of traditional approaches in adapting
to the dynamic nature of grinding processes. The literature
suggests that while conventional controllers are simple and
widely used, they fall short in managing complex systems
(Costea, et al., 2014 Pomerleau, Hodouin, Desbiens,
&Gagnon, 2000). Advanced methods like MPC have
historically faced challenges in computational intensity
and model upkeep (Qin &Badgwell, 2003 Rawlings &
Mayne, 2009), but recent advancements in computational
hardware and machine learning technologies have begun to
overcome these hurdles.
The increasing power of modern processors and the
advent of state-of-the-art machine learning techniques have
greatly reduced concerns around computational demands,
making real-time analysis and model adjustment feasible
even in complex industrial environments. Furthermore,
machine learning’s inherent adaptability offers a promising
1. Fresh feed bunker
2. Ball mill
3. a: dynamic separator b: mill dedusting filter &fan
4. product filter
5. CEOPS particle size distribution measurement
Figure 1. Schematic representation of a typical dry grinding circuit as considered in this research, emphasizing the specific
configuration of a two-compartment ball mill paired with a dynamic separator