Subprojects

F Measure

Management and Quantification of Maturity Improvement

F defines measures of process maturity and records the demonstrator's data and their provenance in a knowledge graph.

Illustration. Each forming run is linked to its material, machines, part and 3D scan. Models learn from the collected evidence, and the schematic maturity gauge rises.

Work packages

Results of the first funding phase, by work package.

  1. F.1 Formal process maturity measure

    Elucidability, Forcability and Supervisability translate observability, controllability and quality tolerance into probabilistic maturity measures. Forcability was made computable as a stochastic reach-avoid problem, solved with approximate dynamic programming and estimated by Monte Carlo simulation on an electric arc furnace example. Supervisability and Elucidability exist so far in simplified form.

    Diagram of a production process under uncertainty with controller, measurement and final product quality, linked to the three maturity measures Forcability, Elucidability and Supervisability.
    The three maturity measures around a production process under uncertainty. Forcability concerns steering into a target set, Elucidability state estimation from observations, and Supervisability the satisfaction of quality tolerances. Figure: KI-FOR 5339.

    Publications

  2. F.2 Virtual Process Dossier (VPD)

    The VPD is a process-aware data catalogue that adds a knowledge-graph layer above the raw data and captures prospective and retrospective workflow provenance, with a schema that reuses DCAT, PROV, SOSA/SSN, QUDT and WiLD. A provenance-capturing framework and a web interface were implemented, and the example process was modelled in the VPD. The publication is under review.

    Diagram of the research data infrastructure, with the VPD user interface, the VPD ontology and framework, and a knowledge graph above the source data and the manufacturing environment.
    High-level overview of the knowledge-graph-based FAIR research data infrastructure. Figure: KI-FOR 5339.

    Publications

  3. F.3 Hybrid semantic-qualitative-numerical question answering

    RDFdL integrates RDF with differential dynamic logic, verifies transitions with KeYmaera X and returns the verified results as queryable RDF. Graph-based retrieval-augmented generation was evaluated on airport data, a shape-based SPARQL generator placed among the top three in several sub-challenges of the Text2SPARQL challenge at ESWC 2025, and a bounded LLM decision layer improved a fixed baseline in 200 simulated scenarios of a composite-forming workflow (CASE 2026).

    RDFdL overview. Metadata and simulation models are stored as RDF, translated into a differential dynamic logic model, proven with KeYmaera X, and the verified safety properties are returned to the RDF store for SPARQL queries.
    RDFdL system overview. Figure: KI-FOR 5339.

    Publications

About the subproject

F spans all subprojects. It defines when a process counts as mature, and its Virtual Process Dossier records experiments, simulations and decisions with their provenance.

Principal investigators
Researchers
  • Lukas Kubelka
  • Yuyang Li
  • Zahra Nasrollah
Former members
  • Negar Arabizadeh
Host institute
Vision and Fusion Laboratory (IES), KIT
People in F

Publications 8