Figure 5: Queue based control of the quality analysis systems.
Orbit real-time database and data analysis
The processing platform with support systems discussed above generate a
lot of data in form of structural information, input and control
signals, detector signals, HPLC analysis, on-line software computations
etc. The on-line generated data is heterogeneous, i.e. different types,
asynchronous, i.e. in-line detectors vs advanced quality analysis, and
distributed, i.e. generated on multiple physical detectors and different
setups. A database was needed to handle real time process data and made
it practical to handle for both local and plant-wide data analysis.
General tools for automatic data analysis of real-time data, on-demand
at-line or off-line analysis, are important for process monitoring and
process control. One example is the peak analysis of the PCC operation
in the pilot-scale case study in Schwartz et al 2022. In Figure 6 the
first two switches in the PCC are shown together with the automatic peak
finder with associated peak measurement. The peak heights are plotted on
right where peak B is the interesting product peak. As seen the three
different PCC column do not behave exactly the same. They have slightly
different peak heights, about 3% difference.
This kind of data analysis is based on the data stored in the database,
which come from different sources in the process. In this case it was
actually only from one detector on one setup, but in the general case it
allows plant-wide data analysis using distributed data sources. Data
analysis processing can be performed locally at the setup with the
detector or at the setup needing the analysis results. An alternative
suggested here is a plant-wide computational extension, in an orbitX
controller, exploiting data in the database to perform more or less
advanced data analysis for usage somewhere in the processing system or
just for plant-wide monitoring. More advanced data analysis for
data-driven modelling and machine learning are often performed off-line,
but in some cases the results need to be validated in real-time. The
architecture suggested here allow both off-line development and on-line
validation.