Structuring Maintenance Records of Manufacturing Lines Using Pre-trained LLM and a Small-scale Domain Ontology

Abstract

To identify the causes of defects in manufacturing systems, past maintenance records are important materials to utilize. Previous studies have attempted to construct a knowledge graph for utilizing these maintenance records. However, challenges have included the inconsistency and specificity of the vocabulary in the maintenance records, and the high cost of machine learning for constructing the knowledge graph. This study proposes a method for constructing a knowledge graph from maintenance records using a pre-trained large language model and a small domain ontology. In constructing a knowledge graph based on actual maintenance records, the proposed method has linked and structured a number of events. In the future, we plan to work on defect cause inference using the knowledge graph.