Machine learning on assembly guidance system ontology for manufacturing assembly

In high-mix, low-volume (HMLV) manual assembly environments, significant product variation presents considerable challenges for operators. Multimodal digital work instruction systems, often termed assembly guidance systems, have emerged as a crucial solution to provide adaptive and dynamic support, thereby minimising operator cognitive load. Ontologies, as formal semantic tools, are vital for organising complex manufacturing and assembly information, which enables the development of adaptive guidance systems. This study investigates how machine learning (ML) techniques, specifically link prediction, applied to manufacturing ontologies can assist in the authoring of these guidance systems by determining suitable instruction delivery methods. A novel assembly guidance ontology was developed using a dataset of 200 electronics repair manuals, comprising 4754 distinct steps. The study utilised these steps to train various ML methods for relational prediction, aiming to assign optimal graphic visualisations within the work instructions. This research demonstrates the applicability of ML techniques, commonly employed in domains like bioinformatics, to manufacturing ontologies, thereby addressing the current scarcity of large datasets for link prediction in this field.