Subject of this
contribution
The aim of this work is to build upon these good practices and provide
additional guidance for the implementation of ML methods based on
another use case, i.e. flooding prevention in a spinning band
distillation and an extraction column. The main focus is set on an
easy-to-follow procedure for the integration of ML solutions following
the example of Min et al. , which has been described in the
following in more detail. For a distillation column, sensor data is
evaluated to forecast the pressure drop with supervised learning
methods. Subsequently, the operating state is classified based on the
forecast combined with a clustering algorithm to form an early flooding
warning system. The distillation column is designed according to the MTP
(module type package) concept, which provides easy access to all sensor
data from OPC/UA servers via a Python script, demonstrating the
advantages of standardized interfaces. Flooding in the extraction column
is analyzed via computer-vision and a convolutional neural network to
design smarter equipment that can identify its own operating state.
Finally, an online control strategy based on the ML models is discussed
and its usability within the MTP architecture is evaluated.