Machine Learning on a Batch Distillation Plant

Batch distillation is an established process for the separation of mixtures. Batch distillation processes are highly dynamic and can be designed very flexibly; a batch distillation plant is thus an ideally suited research object for digitizing process engineering plants. Recently, a batch distillation plant was procured by LTD, which will now be used for a project in the field of machine learning as part of a cooperation between LTD, the Fraunhofer Institute for Industrial Mathematics (ITWM) in Kaiserslautern, and partners from computer science. As a first step, the batch distillation plant is to be extended, in particular regarding analytics; among other things, an online analysis with a benchtop NMR spectrometer is planned. Systematic experiments are to be carried out to generate knowledge on the process behavior of the plant, such as plate efficiencies and pressure losses, and on fluid properties. Additionally, data on malfunctions of the plant will be collected, some of which will be generated specifically. All data obtained with the plant are to be made permanentely available for machine learning purposes. One focus of using the data is on coupling machine learning with physical models, for which a dynamic physical model for batch distillations is available. A central aspect of the project is the detection of anomalies and their interpretation based on the analysis of time series data generated with the plant using machine learning methods.

download job opening (German version)

[back]