930 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
• Minimise: a penalty is given proportional to the
value of a variable
• Maximise: a reward proportional to the value of a
variable
• Minimise to target: a reward proportional to the dif-
ference between the target and actual values
Each reward has a weight associated with it, so the final
reward value can be calculated as the sum of the weighted
partial rewards. Additionally, limits are introduced for each
manipulated variable, including limiting the rate of change
of manipulated variables.
Finding the optimal set of manipulated variables
that maximise the final reward is performed based on the
Covariance Matrix Adaptation Evolution Strategy (Hansen
2019).
Integration
For the optimization process to occur in real-time there
must be data integration between existing IT/OT solutions
of the plant and our solution. Figure 2 shows an integra-
tion diagram of our optimization solutions, of which the
Flotation Optimization App is one example. Of impor-
tance is the capability to signal back into the Advanced
Process Control (APC) and base layer control systems the
outputs of our Decision Optimizer and Virtual Sensor out-
puts. In open-loop mode, recommended setpoints by the
Optimizer are initially fed into the control system by opera-
tors, in a decision-support system fashion. Once trust in the
optimization solution has been established, recommended
setpoints can be fed automatically into the control system
(closed-loop mode).
User Interface
brains.app is IntelliSense.io’s real-time decision-making
platform that helps mineral processing operations to:
1. Describe what is happening, with insight into
unmeasurable areas thanks to Virtual Sensors
2. Diagnose the root cause of behaviour because hid-
den states are exposed with Virtual Sensors
3. Predict what will happen in the future using
Scientific AI
4. Simulate and test the AI output against operator-
recommended setpoints
5. Optimise and close the loop with direct feeds into
a control system
In the case of flotation optimization an operator screen has
been developed to help operators identify opportunities for
performance improvement. Figure 3 shows a screenshot of
the Flotation Optimizer Screen. This screen is divided into
three sections, the leftmost summarises feed conditions,
the central section shows the current setpoint and recom-
mended setpoints per cell (for each airflow and level) and
per section (incl. each reagent dosing point), and the right-
most summarises the current plant performance against the
optimised performance. If the difference between the cur-
rent operation (setpoint and/or performance) deviates sig-
nificantly from the optimised operation the user interface
Figure 2. Integration of Virtual Sensor and Process Optimization solutions for mineral processing
• Minimise: a penalty is given proportional to the
value of a variable
• Maximise: a reward proportional to the value of a
variable
• Minimise to target: a reward proportional to the dif-
ference between the target and actual values
Each reward has a weight associated with it, so the final
reward value can be calculated as the sum of the weighted
partial rewards. Additionally, limits are introduced for each
manipulated variable, including limiting the rate of change
of manipulated variables.
Finding the optimal set of manipulated variables
that maximise the final reward is performed based on the
Covariance Matrix Adaptation Evolution Strategy (Hansen
2019).
Integration
For the optimization process to occur in real-time there
must be data integration between existing IT/OT solutions
of the plant and our solution. Figure 2 shows an integra-
tion diagram of our optimization solutions, of which the
Flotation Optimization App is one example. Of impor-
tance is the capability to signal back into the Advanced
Process Control (APC) and base layer control systems the
outputs of our Decision Optimizer and Virtual Sensor out-
puts. In open-loop mode, recommended setpoints by the
Optimizer are initially fed into the control system by opera-
tors, in a decision-support system fashion. Once trust in the
optimization solution has been established, recommended
setpoints can be fed automatically into the control system
(closed-loop mode).
User Interface
brains.app is IntelliSense.io’s real-time decision-making
platform that helps mineral processing operations to:
1. Describe what is happening, with insight into
unmeasurable areas thanks to Virtual Sensors
2. Diagnose the root cause of behaviour because hid-
den states are exposed with Virtual Sensors
3. Predict what will happen in the future using
Scientific AI
4. Simulate and test the AI output against operator-
recommended setpoints
5. Optimise and close the loop with direct feeds into
a control system
In the case of flotation optimization an operator screen has
been developed to help operators identify opportunities for
performance improvement. Figure 3 shows a screenshot of
the Flotation Optimizer Screen. This screen is divided into
three sections, the leftmost summarises feed conditions,
the central section shows the current setpoint and recom-
mended setpoints per cell (for each airflow and level) and
per section (incl. each reagent dosing point), and the right-
most summarises the current plant performance against the
optimised performance. If the difference between the cur-
rent operation (setpoint and/or performance) deviates sig-
nificantly from the optimised operation the user interface
Figure 2. Integration of Virtual Sensor and Process Optimization solutions for mineral processing