Toward Reliable Knowledge Graphs in Virtual Product Development: Data Strategies and Reproducibility Assessment

The integration of knowledge graphs into industrial product development workflows offers a promising solution for managing complex and heterogeneous data to enhance development efficiency. However, concrete strategies for their deployment, particularly with respect to data quality, integration, and framework implementation, remain underexplored. This study addresses this gap by investigating the data-related requirements and strategies essential for constructing robust and reproducible knowledge graphs within the context of virtual product development. Building on a proposed framework within the context of the BMW Group, we identify critical factors including the use of ontologies, standardized data formats, the implementation of unique and versioned entity identifiers, as well as data quality and integration strategies. To validate the reproducibility and correctness of the framework, we conduct a comparative analysis using two independent example datasets. Both knowledge graphs were constructed under identical conditions and evaluated in terms of completeness and accuracy. Our results confirm that the framework reliably generates high-quality knowledge graphs across varying domains with minimal manual intervention.