Semantic web technologies and relational databases have been in the portfolio of data management tools since the 1990s and earlier. By then, systems of heterogeneous data sources had a different complexity, and those that model a constantly updating complex system had not yet been termed digital twins. Today, integrating computer aided simulations of physical processes into data ecosystems arises as a new compatibility and interoperability challenge, as these ecosystems work on types of data from many disciplines, such as measurement data or manufacturing design data. One pursued option is standardizing simulation data storage and harmonizing the data structures with the other data domains. The STEP standard (3rd edition upgrade of AP209) currently fails to perform despite its vast capabilities bridging design and simulation data. Closed-world descriptions and overly redundant and re-implementing architectures do not seem fruitful, and alternative methods that allow openness, such as ontologies, dominate the digital twin discussions. So, the integration of simulation results needs to be reimagined. This contribution describes a new software architecture and the corresponding application methodology for integrating large quantities of structured simulation data into a semantic data structure. The semantic structure is considered to be the meta-data of the entire heterogeneous system, not limited to the meta-information corresponding to the simulation data. A blow molding simulation chain is used to derive requirements for a case of industrial relevance. We present the integration of VMAP files alongside VMAP-specific processes into an ontological framework by discussing the required semantics and software classes, using a VMAP to OWL translator and VMAP-specific routines. The semantics of these routines is represented itself in a separate process ontology. We demonstrate the capability of the architecture to assure data permeability from the global digital twin level to the detailed numerical values of data entries and even new key indicators in a three-step approach after instantiating a knowledge graph: querying metadata, querying a routine, and extracting new information.