1130 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
and impeller resistance and boundary conditions. With
this detailed information along with a network solver that
iterates within model time steps to find a continuous solu-
tion for the whole network, the model contains the capa-
bilities to predict equipment performance.
Additionally, solids slurries are defined with material
concentrations, densities, and viscosities.
With all this defined information, the plant design and
current conditions can be incorporated to have an impact
pump power.
ADAPTING MODEL TO HISTORICAL
PLANT PERFORMANCE
The process in discussion has been running for multiple
years, and installations never perfectly match vendor data,
so historical plant data is needed to refine the model per-
formance further. Besides equipment wear and installation
differences, there’s further impacts in the process that can
be interpreted from data trends. Even though pump perfor-
mance has variable speed curves and uses density, historical
performance had further relationships of differing perfor-
mance based on flow and solids percentage, especially in
the pump-in-series configuration.
Relationships observed in the historical data were
implemented in the model. A head derating value is sent
into the pump based on historical correlation with flow and
solids content. At low-flow conditions for the pumps in
series, the pump-in-series have a relationship that prorates
the pump speed of the variable speed pump.
For the first set of pumps in series, the measured values
coming from the live plant include pumps’ on/off status,
pump speeds, pump currents, slurry solids content, bound-
ary conditions of tank levels and elevations. This is enough
information for the system that the number of measured
values exceeds the number of inputs needed, equating
to a degree of freedom. Both the flow data and the vari-
able speed data provide redundancy. With the redundant
entries, there’s the opportunity to use that value to flag
when the inputs don’t agree. This flag can lead to retuning
the model or to an alert to operations to investigate in the
plant where the values disagree.
The second set of pumps at the booster station pump
out to different locations of different distance and eleva-
tion. The destination is changed in the field and is rarely
if ever communicated to the historian. In this case, the
speed and flow are not redundant because the final leg of
the pipeline is not fully defined. The tactic in this case is to
use the speed signal for the pumps and the flowrate is used
as an indirect input to back-calculate the current pressure
drop in the final pipeline leg.
ONLINE DIGITAL TWIN
For the online Digital Twin, ANDRITZ has long experience
in connecting our simulation models to control systems via
OPC to retrieve real-time plant data. When connecting to
live plant signals, an added layer is often required to check
the communication status and filter the incoming live data.
Such filtering includes output limits, generating an alarm
if the value has not changed in extended period of time, a
rate of change alarm that recognizes irregular spikes, and
mathematical options such as moving average or first order
filters to reduce measurement noise. The outputs from the
Digital Twin are communicated back to the control systems
and operator screens via OPC. The outputs include calcu-
lated values, virtual instruments, setpoints, alarms, warn-
ings, recommendations to operators, etc.
Part of the model follows closely what is happening
in the live plant, capturing the state of the process with
its real-time dynamics. This section of the model can be
used for the alarming, filtering, and potential virtual instru-
ments. The second part of the model takes the validated
data and has the same equipment blocks to run scenarios
with the optimizer and find optimal power consumption.
OPTIMIZING OPERATION
The optimizing goal for this application is minimizing
power consumption. Table 1 provides a summary of the
optimization problem, followed by further explanation.
The main focus for each set of parallel pipelines, the
Thickener Sump and Booster Station sections each solved
separately (refer to the block diagram in Figure 1), is deter-
mining the load split between the pipelines that has the
minimum power consumption while achieving the desired
combined flowrate. The load division is primarily decided
by the VSD speed setting for the variable speed pumps.
The optimization problem is solved in IDEAS using an
optimizer block that implements an inter-point nonlinear
programming algorithm referred to as IPOPT (Wächter
Table 1. Overview of Optimization Problem Definition
Objective Function Inputs Constraints
Minimize Total Power for a set of parallel
lines of pumps in series
Speed setting for Line A Total Flow within error margin of Flow Setpoint
Speed setting for Line B VSD load split between Line A and Line B within a set range
Step Change in VSD load split limited to a set value
and impeller resistance and boundary conditions. With
this detailed information along with a network solver that
iterates within model time steps to find a continuous solu-
tion for the whole network, the model contains the capa-
bilities to predict equipment performance.
Additionally, solids slurries are defined with material
concentrations, densities, and viscosities.
With all this defined information, the plant design and
current conditions can be incorporated to have an impact
pump power.
ADAPTING MODEL TO HISTORICAL
PLANT PERFORMANCE
The process in discussion has been running for multiple
years, and installations never perfectly match vendor data,
so historical plant data is needed to refine the model per-
formance further. Besides equipment wear and installation
differences, there’s further impacts in the process that can
be interpreted from data trends. Even though pump perfor-
mance has variable speed curves and uses density, historical
performance had further relationships of differing perfor-
mance based on flow and solids percentage, especially in
the pump-in-series configuration.
Relationships observed in the historical data were
implemented in the model. A head derating value is sent
into the pump based on historical correlation with flow and
solids content. At low-flow conditions for the pumps in
series, the pump-in-series have a relationship that prorates
the pump speed of the variable speed pump.
For the first set of pumps in series, the measured values
coming from the live plant include pumps’ on/off status,
pump speeds, pump currents, slurry solids content, bound-
ary conditions of tank levels and elevations. This is enough
information for the system that the number of measured
values exceeds the number of inputs needed, equating
to a degree of freedom. Both the flow data and the vari-
able speed data provide redundancy. With the redundant
entries, there’s the opportunity to use that value to flag
when the inputs don’t agree. This flag can lead to retuning
the model or to an alert to operations to investigate in the
plant where the values disagree.
The second set of pumps at the booster station pump
out to different locations of different distance and eleva-
tion. The destination is changed in the field and is rarely
if ever communicated to the historian. In this case, the
speed and flow are not redundant because the final leg of
the pipeline is not fully defined. The tactic in this case is to
use the speed signal for the pumps and the flowrate is used
as an indirect input to back-calculate the current pressure
drop in the final pipeline leg.
ONLINE DIGITAL TWIN
For the online Digital Twin, ANDRITZ has long experience
in connecting our simulation models to control systems via
OPC to retrieve real-time plant data. When connecting to
live plant signals, an added layer is often required to check
the communication status and filter the incoming live data.
Such filtering includes output limits, generating an alarm
if the value has not changed in extended period of time, a
rate of change alarm that recognizes irregular spikes, and
mathematical options such as moving average or first order
filters to reduce measurement noise. The outputs from the
Digital Twin are communicated back to the control systems
and operator screens via OPC. The outputs include calcu-
lated values, virtual instruments, setpoints, alarms, warn-
ings, recommendations to operators, etc.
Part of the model follows closely what is happening
in the live plant, capturing the state of the process with
its real-time dynamics. This section of the model can be
used for the alarming, filtering, and potential virtual instru-
ments. The second part of the model takes the validated
data and has the same equipment blocks to run scenarios
with the optimizer and find optimal power consumption.
OPTIMIZING OPERATION
The optimizing goal for this application is minimizing
power consumption. Table 1 provides a summary of the
optimization problem, followed by further explanation.
The main focus for each set of parallel pipelines, the
Thickener Sump and Booster Station sections each solved
separately (refer to the block diagram in Figure 1), is deter-
mining the load split between the pipelines that has the
minimum power consumption while achieving the desired
combined flowrate. The load division is primarily decided
by the VSD speed setting for the variable speed pumps.
The optimization problem is solved in IDEAS using an
optimizer block that implements an inter-point nonlinear
programming algorithm referred to as IPOPT (Wächter
Table 1. Overview of Optimization Problem Definition
Objective Function Inputs Constraints
Minimize Total Power for a set of parallel
lines of pumps in series
Speed setting for Line A Total Flow within error margin of Flow Setpoint
Speed setting for Line B VSD load split between Line A and Line B within a set range
Step Change in VSD load split limited to a set value