Multi-sourced modelling for strip breakage using knowledge graph embeddings

Strip breakage is an undesired production failure in cold rolling. Typically, conventional studies focused on cause analyses, and existing data-driven approaches only rely on a single data source, resulting in a limited amount of information. Hence, we propose an approach for modelling breakage using multiple data sources. Many breakage-relevant features from multiple sources are identified and used, and these features are integrated using a breakage-centric ontology which is then used to create knowledge graphs. Through ontology construction and knowledge embedding, a real-world study using data from a cold-rolled strip manufacturer was conducted using the proposed approach.