927
Real-Time Flotation Optimization Using Scientific AI
Francisco Reyes, Adam Griffin, Grant Kopec, Niel Knoblauch,
Mark De Geus and Cristian González
IntelliSense.io, Cambridge, United Kingdom
ABSTRACT: Annually more than a billion tonnes of ore are treated by froth flotation worldwide. Therefore,
improvements in its operation can make significant contributions towards achieving production and sustainability
targets, especially for critical minerals.
We present a flotation optimization solution that combines first-principle and artificial intelligence models:
Scientific AI. These models are linked to an Optimizer that evaluates the current process performance and
suggests optimised setpoints for airflows, froth depths and reagent dosage to control room operators. These
setpoints maximise the Value Driver (a multi-variable reward-based function) that can also incorporate process
constraints. In this paper, we present a case study for an iron ore mine in South America.
INTRODUCTION
The mining industry is currently between a rock and a hard
place. It has to simultaneously achieve challenging sustain-
ability and production targets, while facing dwindling head
grades in increasingly complex deposits. Froth flotation
is one of the cornerstones of mineral processing, annu-
ally treating more than a billion tonnes of ore worldwide
(Pashkevich et al., 2023). As such, any improvements in
efficiency can have considerable impact on the industry in
terms of both sustainability and production.
Froth flotation is a physico-chemical process that uti-
lises surface properties to separate the minerals of interest
from gangue minerals. The separation process occurs in a
three-phase system (water, air, solids) with particles being
selected based on their hydrophobicity and buoyancy.
Reagents are commonly used to assist this process by modi-
fying surface properties and stabilising the froth (Wills and
Napier-Munn 2005). Control room operators and metal-
lurgists have the complex task of needing to pay attention
to froth depth, airflow, feed rate and reagent dosage across a
multi-bank and multi-cell circuit to maximise metal recov-
ery without sacrificing the final product grade.
The task of optimising froth flotation is therefore com-
plex, always changing and multivariate. A problem that has
been tackled mostly by 1) expensive and lengthy site cam-
paigns to collect data and perform offline simulations, or
2) the use of expert control systems to code the operators’
knowledge of the process into a set of rules. The main prob-
lem with the first approach is that the optimised solution
might come months after the sampling campaign, with
current conditions probably being different from the data
collected during the campaign. Moreover, simulations are
‘offline’ (not run in real time) so they cannot easily adapt
to new conditions. The second approach, albeit real-time,
does its decision-making based on local setpoints without
the consideration of the process interdependencies and
constraints (Quintanilla et al., 2021).
In light of this, we present our Flotation Optimization
App, a solution based on the use of Scientific AI that is,
the combination of first-principle and artificial intelligence
Real-Time Flotation Optimization Using Scientific AI
Francisco Reyes, Adam Griffin, Grant Kopec, Niel Knoblauch,
Mark De Geus and Cristian González
IntelliSense.io, Cambridge, United Kingdom
ABSTRACT: Annually more than a billion tonnes of ore are treated by froth flotation worldwide. Therefore,
improvements in its operation can make significant contributions towards achieving production and sustainability
targets, especially for critical minerals.
We present a flotation optimization solution that combines first-principle and artificial intelligence models:
Scientific AI. These models are linked to an Optimizer that evaluates the current process performance and
suggests optimised setpoints for airflows, froth depths and reagent dosage to control room operators. These
setpoints maximise the Value Driver (a multi-variable reward-based function) that can also incorporate process
constraints. In this paper, we present a case study for an iron ore mine in South America.
INTRODUCTION
The mining industry is currently between a rock and a hard
place. It has to simultaneously achieve challenging sustain-
ability and production targets, while facing dwindling head
grades in increasingly complex deposits. Froth flotation
is one of the cornerstones of mineral processing, annu-
ally treating more than a billion tonnes of ore worldwide
(Pashkevich et al., 2023). As such, any improvements in
efficiency can have considerable impact on the industry in
terms of both sustainability and production.
Froth flotation is a physico-chemical process that uti-
lises surface properties to separate the minerals of interest
from gangue minerals. The separation process occurs in a
three-phase system (water, air, solids) with particles being
selected based on their hydrophobicity and buoyancy.
Reagents are commonly used to assist this process by modi-
fying surface properties and stabilising the froth (Wills and
Napier-Munn 2005). Control room operators and metal-
lurgists have the complex task of needing to pay attention
to froth depth, airflow, feed rate and reagent dosage across a
multi-bank and multi-cell circuit to maximise metal recov-
ery without sacrificing the final product grade.
The task of optimising froth flotation is therefore com-
plex, always changing and multivariate. A problem that has
been tackled mostly by 1) expensive and lengthy site cam-
paigns to collect data and perform offline simulations, or
2) the use of expert control systems to code the operators’
knowledge of the process into a set of rules. The main prob-
lem with the first approach is that the optimised solution
might come months after the sampling campaign, with
current conditions probably being different from the data
collected during the campaign. Moreover, simulations are
‘offline’ (not run in real time) so they cannot easily adapt
to new conditions. The second approach, albeit real-time,
does its decision-making based on local setpoints without
the consideration of the process interdependencies and
constraints (Quintanilla et al., 2021).
In light of this, we present our Flotation Optimization
App, a solution based on the use of Scientific AI that is,
the combination of first-principle and artificial intelligence