IFAC MMM 2022 - XPS Advanced Modeling and Machine Learning in An Objective To Supplement Process Control Optimization

Advanced Modeling and Machine Learning in an objective to supplement
Process Control Optimization

Kurt Westhaver, Alan Hyde, Nicolas Lazare, Alexi Tracey, Tanai Marin, Usman Niaz
XPS Expert Process Solutions
6 Edison Road, Falconbridge, Ontario, Canada P0M 1S0
(e-mail: [email protected], [email protected], [email protected], [email protected],
[email protected], [email protected])
Abstract: XPS is continuing to enrich its Digital Twin and Artificial Intelligence portfolio with the
objective of producing key hybrid models for pyrometallurgical, hydrometallurgical and mineral processing
processes in order to respond to the current and future challenges in the mining and metals industry.
Keywords: Artificial intelligence, Digital twins, Industry 4.0, Machine learning, Metallurgy, Online
models, Prescriptive Maintenance, Process control, Simulation
1. INTRODUCTION 3. METALLURGICAL PROCESS MODELING

XPS is focused on optimizing metal recovery, plant XPS has developed an, in-house, expertise in developing
throughput, and emissions reduction. complex pyrometallurgical and hydrometallurgical models for
multiple clients. This expertise started with off-line models
In 2021, XPS presented a novel approach for real-time Fe end- when demand is exponentially growing to translate them into
point determination at Sudbury INO Smelter Finishing to real-time models (often labelled Digital Twins).
Converter using Thermodynamic Process Simulation during
the 5th International Symposium on Nickel and Cobalt. This
model since proved to improve target accuracy from 84% to
98% relative to target Fe end-point when it also demonstrated
strong benefices for reduction of emissions.
Building on the success of this project and multiple demands
in the industry, XPS is now developing new Digital Twins for
Smelter assets, Hydromet plants and Concentrators allowing
to generate more knowledge related to how these assets are
behaving.
However, some phenomenon can’t be simulated or modeled
and require other technics to be predicted or explained. In this
case Artificial Intelligence can be a desirable option to reach
the objective.
For this reason, the hybrid approach already developed in other Figure 1: A model showing the dynamic calculation of the
industries is beneficial for the mining and metals industry. normalized flow in a Smelter off-gas line.
2. ADVANCED PROCESS CONTROL

Since the commissioning of its first Advanced Process Control
(APC) project in the 90’s, XPS has continued to promote and
demonstrate the gain from this discipline for both
Concentrators and Smelters.
XPS successfully deployed Model Predictive Control (MPC),
Fuzzy Logic Control and Expert Systems for both
Concentrators and Smelters.
These models have shown a capability to highly reduce the
variability and permitted to be more aggressive on control
targets in different concentrator and smelter operations.
Figure 2: A model showing the dynamic calculation of SO2 in a
Smelter off-gas line. Figure 3: Predicting wear for a mineral processing asset using non-
linear regression and neural network techniques.
If dynamic models are leveraging the understanding of
operational staff covering steady-state conditions, process XPS has applied unsupervised machine learning techniques to
perturbations, transitions and anomaly detections they are also identify and study the contributing factors of process upsets. A
key for online control optimization of the assets. By principal component analysis (PCA) model was trained using
integrating those inside the control philosophy it allows to historical feed, chemistry, and temperature data for a reductive
create decision making applications that interact in real time smelting furnace to capture the correlation structure of furnace
with operators, technical teams and maintenance teams operating parameters, and to identify clusters within the
creating a new generation of Advisory and Expert systems. dataset representing abnormal operation. The model was
subsequently applied to a period of unstable operation
4. MACHINE LEARNING following a planned shutdown. Insights provided by the model
In recent years, XPS has continued to expand its expertise in helped to support more simplistic root cause analyses, identify
the fields of artificial intelligence and machine learning. This “no go” zones for the furnace, and to develop new strategies
expertise has been applied to forecast asset integrity, develop for furnace startup.
real-time predictions of unmeasured variables and identify
5. NEXT STEPS
factors contributing to undesirable process conditions.
XPS is developing Hybrid models using the best of APC,
Predictive maintenance is a key area of interest for many of advanced modeling and AI to push further the limit of plant
XPS’ clients. The use of machine learning algorithms to control optimization.
analyze the condition of an asset and predict when
maintenance may be required has the potential to significantly The vision is to merge these disciplines and techniques to
reduce the costs and downtime associated with equipment increase the accuracy, robustness and repeatability of the
failure. XPS has applied non-linear regression and neural automated controls and decision making systems.
network techniques to predict wear of mineral processing
equipment based on historical operating data as well as future 6. CONCLUSIONS
data retrieved from the plant’s production plan. The techniques XPS experts from all disciplines are proactively developing
used provide maintenance staff with forecast of the asset’s solutions to challenges related to new mine, concentrator,
remaining useful life that, combined with the dynamic smelter and refinery standards and to further optimize mineral
uncertainty values generated by the model, can be used to processing, metallurgy and maintainability.
schedule asset maintenance well in advance.
REFERENCES
Marin-Alvarado, T., MacKinnon, B., Zannier, N., Moraes, A.,
Westhaver, K., Doxtator, A., Nelson, P., Lazare, N.,
McIver, V., Almeida, T., Muinonen, S. (2021). Real-
time Fe end-point determination at Sudbury INO Smelter
Finishing Converter using Thermodynamic Process
Simulation. TMS NI-CO 2021 The 5th International
Symposium on Nickel and Cobalt.
Nelson, P., Reata, N. (2019). Selection and implementation of
a modern model predictive controller for furnace slag
composition control. CIM 2019 Convention & EXPO.

IFAC MMM 2022 - XPS Advanced Modeling and Machine Learning in An Objective To Supplement Process Control Optimization

Uploaded by

Copyright:

Available Formats

IFAC MMM 2022 - XPS Advanced Modeling and Machine Learning in An Objective To Supplement Process Control Optimization

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IFAC MMM 2022 - XPS Advanced Modeling and Machine Learning in An Objective To Supplement Process Control Optimization

Uploaded by

Copyright:

Available Formats

Advanced Modeling and Machine Learning in an objective to supplement

Process Control Optimization

1. INTRODUCTION 3. METALLURGICAL PROCESS MODELING

2. ADVANCED PROCESS CONTROL

You might also like