2285
Machine Learning Forcefields for Cutting-Edge Description of
Mineral-Water Interfaces
D. Dell’Angelo, M. Badawi
Université de Lorraine, CNRS, Laboratoire Lorrain de Chimie Moléculaire, Metz, France
J. Lainé
ArcelorMittal Global R&D, Maizières-lès-Metz, France
Y. Foucaud
GeoRessources, Université de Lorraine, CNRS, Nancy, France
ABSTRACT: Improve the separation contrast between minerals is the key to improve the flotation processes.
In order to process more complex ores containing more and more fine particles, new combinations of flotation
reagents must be designed. In this line, molecular modeling could be an attractive tool to predict the adsorption
affinity between dozen of formulations and various minerals. Especially when extended, it provides access to
detailed mechanistic information on solvent configurations and may ascertain crucial dynamical events over the
adsorption process. However, a better compromise between cost of calculation and accuracy must be met. Here
we introduce our recent advancements in the building of new accurate mineral-water interfaces based on active
learning of ab initio molecular dynamics trajectories. The case of kaolinite, quartz and calcite will be taken as
examples.
Keywords: Machine learning, molecular simulations, flotation, solid/liquid interface
INTRODUCTION
Motivation: The Quest for Sustainable Separation
Techniques
The worldwide demand for raw materials has increased
considerably over the last decade and following the strong
economic growth it is expected to double by 2030 (Grohol
and Veeh 2023). However, the progressive depletion of the
deposits could make the supply of the resources themselves
difficult (Sun and Yao 2023). As a matter of fact, a decline
in the mineral content leads to an increase in the energy
required for extraction, a greater consumption of water and
a greater production of waste, given by the inevitably lower
efficiencies of the separation processes (Lèbre and Corder
2015). From this perspective, mining waste can be seen as
an acceptable source of raw materials, making their recy-
cling essential (Kinnunen and Kaksonen 2019). A thin line
defines the classification of an ore as a waste or a resource,
which depends on three critical factors: time, extraction
strategy and economic context. Time allows the research
and evolution of technological processes that allow the
extraction of valuable minerals from what were once sim-
ply considered waste. The extraction strategy used defines
Machine Learning Forcefields for Cutting-Edge Description of
Mineral-Water Interfaces
D. Dell’Angelo, M. Badawi
Université de Lorraine, CNRS, Laboratoire Lorrain de Chimie Moléculaire, Metz, France
J. Lainé
ArcelorMittal Global R&D, Maizières-lès-Metz, France
Y. Foucaud
GeoRessources, Université de Lorraine, CNRS, Nancy, France
ABSTRACT: Improve the separation contrast between minerals is the key to improve the flotation processes.
In order to process more complex ores containing more and more fine particles, new combinations of flotation
reagents must be designed. In this line, molecular modeling could be an attractive tool to predict the adsorption
affinity between dozen of formulations and various minerals. Especially when extended, it provides access to
detailed mechanistic information on solvent configurations and may ascertain crucial dynamical events over the
adsorption process. However, a better compromise between cost of calculation and accuracy must be met. Here
we introduce our recent advancements in the building of new accurate mineral-water interfaces based on active
learning of ab initio molecular dynamics trajectories. The case of kaolinite, quartz and calcite will be taken as
examples.
Keywords: Machine learning, molecular simulations, flotation, solid/liquid interface
INTRODUCTION
Motivation: The Quest for Sustainable Separation
Techniques
The worldwide demand for raw materials has increased
considerably over the last decade and following the strong
economic growth it is expected to double by 2030 (Grohol
and Veeh 2023). However, the progressive depletion of the
deposits could make the supply of the resources themselves
difficult (Sun and Yao 2023). As a matter of fact, a decline
in the mineral content leads to an increase in the energy
required for extraction, a greater consumption of water and
a greater production of waste, given by the inevitably lower
efficiencies of the separation processes (Lèbre and Corder
2015). From this perspective, mining waste can be seen as
an acceptable source of raw materials, making their recy-
cling essential (Kinnunen and Kaksonen 2019). A thin line
defines the classification of an ore as a waste or a resource,
which depends on three critical factors: time, extraction
strategy and economic context. Time allows the research
and evolution of technological processes that allow the
extraction of valuable minerals from what were once sim-
ply considered waste. The extraction strategy used defines