Conclusion and Outlook

The application of machine and deep learning in the process industry is an adequate way to predict or detect the flooding behavior in laboratory distillation and extraction columns. It was shown that both time-series data of process values and image recognition can be used for modelling. Parallel to this, examples were given, which enable the simple integration of AI-based monitoring systems into existing plants enabled by existing control architectures such as Module Type Package MTP. An adaptation of camera setups or the existing data structures, such as OPC-UA, are sufficient to provide an interface for a data science implementation. This results in a high potential for tooling up existing equipment with AI methods as part of the digital twin. It is possible to combine both analytical methods in order to specify the flooding behavior even more and to transfer the flooding detection from an AI-supported to an AI-controlled monitoring system.