852 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
Development of Data Collection Infrastructure
Edge Device Selection
An industrial-grade OnLogic Helix 511 fanless edge-PC,
equipped with an Intel Core i5 processor and 32 GB of
RAM, was integrated into the plant’s control system. This
edge computing device, running Ubuntu Desktop 22.04
LTS and utilizing the OPC-UA communication protocol,
served as a central repository for all operational data, crucial
for the modeling process. It was selected for its capability to
support advanced machine learning operations and poten-
tial future control algorithm deployment, ensuring seam-
less and secure data handling.
Data Integrity and Frequency
Data collection integrity and frequency were optimized by
setting a one-minute sampling interval. This interval was
chosen to balance the need for timely data against potential
noise and fluctuations in measurements, with data being
averaged over this period to mitigate short-term variabil-
ity. This setup ensures the data reflects the dynamic nature
of the grinding process accurately and reliably, supporting
robust modeling and analysis.
Data Preparation and Feature Engineering
To ensure the effectiveness of the machine learning models,
the data preparation and feature engineering phases were
meticulously structured. These phases included:
• Data Merging and Normalization: Conducted a
comprehensive merging and normalization process
to align measurement values and operational set-
points from various sources, standardizing data for-
mats and scales for analytical coherence.
• Feature Selection and Cleaning: Identified and
selected key features impacting the grinding process,
rigorously cleaning the dataset to remove inconsis-
tencies and applying action smoothing techniques
for real-time operational stability.
• Feature Engineering: Employed advanced tech-
niques to transform raw data into formats optimized
for machine learning, enhancing predictive power
Initial Process Analysis
A meticulous analysis was conducted to assess data quality and
frequency, establishing a robust understanding of the existing
grinding circuit. This critical examination laid the groundwork for
future data-driven optimization strategies.
Data Types and Acquisition
Identification of key data types—measurement values, operational
setpoints, and design characteristics—was crucial. This phase also
covered pilot testing at Cemtec’s small-scale facility to test initial
models and explore data interdependencies, setting the stage for
subsequent scale-up to industrial settings.
Scaling from Pilot Plant to Industrial Scale
The project initiated with pilot testing at Cemtec's facility, equipped
with a grinding setup (1.2 x 3.2 meters, 35 kW drive, 1.0 tph
throughput). This stage focused on developing and refining initial
models, assessing data interdependencies under controlled
conditions. Following pilot success, the study expanded to three
industrial-scale plants in the cement and minerals sector to test
model scalability and robustness. This expansion aims to validate
models across diverse operational environments, ensuring
adaptability and effectiveness.
Addressing Data Gaps
Recognizing infrequent analysis of particle size in industrial setups,
the decision was made to implement the CEOPS (Cemtec Online
Particle Measurement System). This allowed for real-time
measurements and significantly enhanced dynamic process
modeling capabilities.
Development of Data Collection Infrastructure
Edge Device Selection
An industrial-grade OnLogic Helix 511 fanless edge-PC,
equipped with an Intel Core i5 processor and 32 GB of
RAM, was integrated into the plant’s control system. This
edge computing device, running Ubuntu Desktop 22.04
LTS and utilizing the OPC-UA communication protocol,
served as a central repository for all operational data, crucial
for the modeling process. It was selected for its capability to
support advanced machine learning operations and poten-
tial future control algorithm deployment, ensuring seam-
less and secure data handling.
Data Integrity and Frequency
Data collection integrity and frequency were optimized by
setting a one-minute sampling interval. This interval was
chosen to balance the need for timely data against potential
noise and fluctuations in measurements, with data being
averaged over this period to mitigate short-term variabil-
ity. This setup ensures the data reflects the dynamic nature
of the grinding process accurately and reliably, supporting
robust modeling and analysis.
Data Preparation and Feature Engineering
To ensure the effectiveness of the machine learning models,
the data preparation and feature engineering phases were
meticulously structured. These phases included:
• Data Merging and Normalization: Conducted a
comprehensive merging and normalization process
to align measurement values and operational set-
points from various sources, standardizing data for-
mats and scales for analytical coherence.
• Feature Selection and Cleaning: Identified and
selected key features impacting the grinding process,
rigorously cleaning the dataset to remove inconsis-
tencies and applying action smoothing techniques
for real-time operational stability.
• Feature Engineering: Employed advanced tech-
niques to transform raw data into formats optimized
for machine learning, enhancing predictive power
Initial Process Analysis
A meticulous analysis was conducted to assess data quality and
frequency, establishing a robust understanding of the existing
grinding circuit. This critical examination laid the groundwork for
future data-driven optimization strategies.
Data Types and Acquisition
Identification of key data types—measurement values, operational
setpoints, and design characteristics—was crucial. This phase also
covered pilot testing at Cemtec’s small-scale facility to test initial
models and explore data interdependencies, setting the stage for
subsequent scale-up to industrial settings.
Scaling from Pilot Plant to Industrial Scale
The project initiated with pilot testing at Cemtec's facility, equipped
with a grinding setup (1.2 x 3.2 meters, 35 kW drive, 1.0 tph
throughput). This stage focused on developing and refining initial
models, assessing data interdependencies under controlled
conditions. Following pilot success, the study expanded to three
industrial-scale plants in the cement and minerals sector to test
model scalability and robustness. This expansion aims to validate
models across diverse operational environments, ensuring
adaptability and effectiveness.
Addressing Data Gaps
Recognizing infrequent analysis of particle size in industrial setups,
the decision was made to implement the CEOPS (Cemtec Online
Particle Measurement System). This allowed for real-time
measurements and significantly enhanced dynamic process
modeling capabilities.