Generation of Semantic Knowledge Graphs from Maintenance Work Order Data

Industrial maintenance activity data is typically stored in unstructured form within the databases of maintenance man-agement systems. For this reason, effectively exploring the data and uncovering valuable patterns concealed within it is often highly challenging. Consequently, historical maintenance data is seldom analyzed or reused for purposes such as failure prevention, maintenance history reconstruction, or maintenance diagnostics. If the knowledge embedded in maintenance data is liberated and formalized, it can significantly improve the intelligence of maintenance management systems by enabling knowledge reuse. This research aims to help advance the progression from data to information and knowledge through data-driven creation of knowledge graphs built from the unstructured data available in maintenance work orders. A Simple Knowledge Organization System (SKOS) thesaurus is used to support automated entity extrac-tion from text. The thesaurus is extended with the aid of a fine-tuned Large Language Model (LLM). A formal ontology provides the semantics of the knowledge graph. A software tool is developed to streamline the semi-automated text-to-graph translation process. The proposed framework was validated based on 100 work orders extracted from the com-puterized maintenance management system of a construction equipment manufacturer. The experimental validation proved that graph-based representation of work order data could effectively enhance information retrieval, analysis, and pattern extraction particularly if it is supported by formal ontology and rule-based reasoning methods.