Knowledge Graphs for the use of capturing engineering expertise in industrial settings

Knowledge elicitation is a time-consuming element of building expert systems, where expertise captured from domain experts is typically hard coded into software solutions or used only for dataset labelling, This means that said expertise cannot be reused with similar systems are developed at later dates in the same domain, especially if those future solutions are built in different platforms of software development environments.
We propose capturing knowledge in an implementation-agnostic form using dynamic knowledge graphs. These graphs not only represent complex relationships through interconnected nodes and edges but also encapsulate functionality by integrating software methods that can process and update information within the graph structure.
In this implementation we employ the graph to sort, label and preprocess data automatically using encoded expert knowledge to guide the training of a model used to measure degradation of an asset within a Nuclear Reactor. Working within the NeuroSymbolic cycle the graph’s reasoning guides the model’s training, the model’s output is fed back into the graph to be reasoned about, we keep all the expert knowledge held within the graph and provide explainability through traceability since the graph is a queryable record of all training decisions.