Curation and transformation of science data originating from real-world experiments to fuel semantic representations is a complex task. In this presentation, we demonstrate how vast amounts of raw data in various heterogeneous digital formats can be (semi-)automatically post-processed step-by-step. A large spectrum of lessons learned is elaborated on and discussed for each individual step. Furthermore, the transformation output is mapped to two different types of ontologies (the PMD Core and the BWMD ontology) so that synergies of the processes are highlighted as well as some caveats and adoption thresholds. The results are FAIR (findable, accessible, interoperable, and re-usable) material science and engineering (MSE) data modeled in comprehensive knowledge-graphs with complementary competency questions to demonstrate their applicability and usefulness for common use cases. To render the presented results reproducible, all corresponding non-confidential original data as well as the data acquisition pipeline and other mapping procedures used are available online.