High performance fluid dynamics on modern clusters produces large and heterogeneous logs and files. Without a shared vocabulary and a repeatable extraction path, key facts remain hard to find, access, interoperate and reuse (FAIR). This thesis focuses on JAX-Fluids, a CFD simulation software devoloped by TUM AER, and proposes a lightweight path that connects an engineering ontology Metadata4Ing, a metadata crawler HOMER, and an HPC workflow. First, guided by the modeling approach of Metadata4Ing_HPMC, we describe the simulation from a processing step-centric view: for each step we state which tool is used, by which method, who participates (person/organization), and what the inputs and outputs are. On this basis we derive a practical mapping from JAX-Fluids outputs to ontology terms. We then curate the classes and properties that are actually usable and explain how to instantiate and assign values from real JAX-Fluids results. Second, we address pain points observed during real runs of the HOMER crawler and optimize it for HPC use. We improve it with multi-format ontology input, clearer messages and status reports, an –exact-step execution mode, restored postprocessors, indirect property reachability with provenance, and a set of reusable, extensible configuration templates. On the HPC system at LRZ we integrate JAX-Fluids and HOMER in a single job and produce ontology-aligned JSON as an end-to-end example, including the final output.json. We further align the crawled fields with the HPC CFD Simulation profile in Coscine and outline how Terminology Service and MetaStore can complete the data lifecycle. The outcome shows that ontology-driven extraction can be embedded next to real CFD runs and can lay practical groundwork for FAIR and reproducible research.