Meta-knowledge triple driven multi-modal knowledge graph construction method and application in production line control with Gantt charts

Abstract

Digital manufacturing involves complex and multidimensional interactions among production line resources, resulting in massive multi-modal knowledge. The knowledge often lacks correlation and contextual readability, leading to data silos. The rapid development of knowledge graphs (KGs) has rekindled interest in manufacturing knowledge engineering. Investigating the framework of multi-modal manufacturing data assets in enterprises and transforming them into a general-purpose KG database to support manufacturing processes is of significant importance. Guided by the principle of using KG as a manufacturing database, this study developed a multi-modal production line manufacturing knowledge graph (PLMKG) to support dynamic manufacturing on production lines. Firstly, the schema layer of the PLMKG is constructed using the Entity-Relationship model and a manufacturing knowledge pattern framework, with meta-knowledge triples proposed for schema data expression. Secondly, an event-state trigger dynamic instantiation method based on triples binding is proposed to enable self-growth. Third, a method integrating dynamic Gantt charts is introduced to synchronize the control of PLMKG and the manufacturing process. The anomaly detection model is employed to detect production, with the results stored in the PLMKG and Gantt charts for process control. Finally, a PLMKG prototype system for data management and process visualization is developed, with a 3D printing production line case study validating the construction and application of PLMKG. The results indicate that the proposed PLMKG integrates multi-modal manufacturing knowledge structurally and provides AI readiness for manufacturing, finally supporting the production line operation as a database.