The design and development of complex aerospace systems pose significant challenges due to their growing complexity. Iterative design processes, guided by formal specifications, strive to refine initially vague characteristics through multiple stages. Despite these efforts, the integration of diverse disciplinary knowledge into system models often remains incomplete. This study tackles the challenge of incomplete knowledge in MBSE system models by introducing a graph-based machine learning approach to uncover and address missing links. SysML models are transformed into knowledge graphs, where relational graph convolutional networks are applied to identify missing connections that may be overlooked by human analysts. By utilizing the graphical structure of system models, this approach automates link prediction and enhances model completeness. The results demonstrate that graph neural networks effectively recovers missing links, achieving 72\% accuracy, although dataset availability remains a key challenge for large-scale implementation. The findings emphasize the need for extensive open-source datasets to fully understand the impact of machine learning in systems engineering frameworks. Keywords --- model-based systems engineering, knowledge graph, graphical neural networks