Figure 6: General peak data analysis used to monitor the PCC
performance, left) the first two switches of the PCC together with
automatic peak finding, right) PCC peak heights after about 40 hour
operation.
Model based simulation, monitoring and control
The concept of digital twins in real-time applications means to use
digital representations and mathematical models for monitoring and
control of the processing system. The architecture of the proposed
system with Orbit supervisory controllers and network of Orbit
controllers allows the user to implement, study, design and validate
these real-time applications of digital solutions.
The generation and validation of mathematical models is as important as
it is time consuming, requiring a lot of engineering resources, see
Nilsson and Andersson 2017. An attractive idea is to automatically
generate the process model and to have automatic methods to calibrate
and validate the model to experimental data. One such idea was studied
and presented in Tallvod et al (2022A). In Orbit, every object in the
digital configuration of the physical system has a mathematical
representation. It means that a complete model of the setup is
automatically generated based on the described configuration. Many of
the objects can be described quite well, because the behavior is
dominated by measurable physical attributes, like volume, flow rate,
concentrations etc. Other objects need to be calibrated to experimental
data and therefore needs a design of experiments to generate high
quality data for parameter estimation. Off-line experiment plans are
straightforward to perform together with computational extensions for
parameter estimations of different models, see Tallvod et al (2022A).
During continuous operation it is much harder to do experiments for
parameter estimations and model calibration, but limited disturbances
are often allowed if the control system can handle the disturbances.
One example of real-time usage of a mathematical model is in Kalman
filter based monitoring. Extended Kalman filter tools are implemented as
computational extensions in Orbit, with the possibility to use the build
in automatic model generation. A simplified example is illustrated in
Figure 7, which is a size exclusion chromatographic separation of two
components with vary large difference in size. This case is actually
linear, which simplifies the Kalman filter design in the case, but it is
still a dynamic filter with covariance update algorithm. In Figure 8,
column concentration profiles are estimated using the Kalman filter at
four different time instances. The Kalman filter predicts the profile
based on the model and corrects their value using the measurement
signal. The example illustrates the power of these kinds of real-time
algorithms for soft sensing, parameter estimation and on-line
prediction.