Process data provisioning with reference to product and semantics for data-driven quality optimization
In manufacturing and chemical processes, a lot of data is collected and made available to IT systems. Most process data is recorded in the form of time series of status or
measurement values. In addition, there is variously structured data, e.g. the current configuration of control systems, image and video recordings, and time-discrete events
such as alarms.
For the optimization of the product quality by means of statistical or machine learning methods, it is necessary to map this data from different parts of the plant to the production process of individual product units. This mapping should be done in an automatic way on the basis of existing engineering documents, in order to reduce the
effort during the construction of new plants or the reconstruction of existing plants or in case of product changes.
Research will be carried out into how product units can be tracked across several production steps, either - as far as possible - by means of identification features attached to them or on the basis of flow measurement values and a model of material flow paths. For this purpose, a generic identification model for different types of product units
is to be developed, which can also represent the joining/mixing of multiple intermediate products into one final product or separation of products, as well as the stochastic mixing of product units.
In addition, the technical realization of the integration and efficient provision of the data is to be researched, whereby their semantics and context of origin are to be preserved. For this purpose, databases for raw data and a metadata repository for data context and semantics will be used.