Leveraging knowledge from historic engineering drawings

In the nuclear power industry, paper documents are in regular use as part of the life cycle of asset management, maintenance, and operations. Despite their importance, fully digitizing these documents is problematic due to being expensive and time consuming, while automation is made difficult by a combination of factors, primarily a significant time cost and required expertise by both software engineers and nuclear engineers to design digitization systems, label training data and validate the outputs of untrusted black-box solutions. In addition, a direct digitization of such documents, converting them to an equivalent digital format, is not necessarily the most useful representation of the contained knowledge. In this paper we present a framework for extracting and encoding both knowledge and data utilizing Knowledge Graphs (KGs); extracting the knowledge more efficiently than fully manual digitization, while still keeping the human supervision in place using a Human-in-the-loop approach. We then present a case study on using these techniques to convert engineering drawings into KGs, demonstrating the insights allowed by this form, as well as the advantages it provides to the development of intelligent systems. We show how this form can still be used to rebuild the original document in an equivalent format if desired, as well as how the accessible format makes it easier to analyze the data and develop a variety of other useful applications, such as tying in to other software packages or automatically suggesting redesigns.