Patrik ThomasMichalski

NFDIxCS develops a shared research-data infrastructure for computer science. Its scope includes datasets as well as software and the technical context required to understand or reproduce a result. The consortium works on FAIR principles, persistent citation and interoperable services for complex, domain-specific research objects.

A central concept is the Research Data Management Container: related artefacts are bundled instead of being published as disconnected files. Metadata has to describe their relationships precisely enough for other researchers and services to find, interpret and reuse them.

My contribution focused on extending schemas for empirical research knowledge. I worked on representing heterogeneous research objects and their execution context in a structured form, so that data, software and documented dependencies remain connected and citable.

KI@CAU Datencampus Kiel was established to make data and AI methods usable across disciplinary boundaries at Kiel University. The state funded the three-year initiative with about two million euros. Interdisciplinary tandems linked computer scientists with researchers from other fields and worked from data acquisition through analysis to later reuse.

The practical challenge is that disciplines use different instruments, formats and quality criteria. A shared strategy therefore cannot consist of one model or one platform. It needs transferable workflows, documented interfaces and enough domain context for results to remain interpretable.

During my involvement, I worked in this interdisciplinary setting, including exchanges with electrical engineering and physics. The focus was on connecting data-science methods with the way data is actually produced and evaluated in the participating disciplines.

SPP 2422 combines process data, expert knowledge and solution spaces from simulations to model metal-forming operations. Institutes from forming technology, automation and data science work together so that generic mathematical representations can be tested against concrete manufacturing processes.

For machine learning, the available labelled measurements are a bottleneck. Real forming trials are expensive, process parameters interact and a model must still respect the physical character of a force–displacement curve. Synthetic data is useful only when these constraints remain controllable and traceable.

I developed a controllable generator for synthetic force–displacement curves. The generated process data supports the training and evaluation of models for active tool-surface optimisation without presenting synthetic curves as measured observations.

Marispace-X created a maritime data space based on Gaia-X principles such as data sovereignty, security, interoperability and modularity. Its architecture connects edge, fog and cloud processing so that large sensor datasets can be exchanged and analysed across organisational boundaries.

The use cases range from underwater sensor networks and offshore wind to the monitoring of critical infrastructure and historic munitions. They share a dependence on spatial data that arrives from different systems, at different resolutions and under different access conditions.

My work concerned collision-risk-aware ship routing and the management of the underlying spatial data. I investigated how route planning can account for other moving vessels instead of treating the sea as a static cost surface.