Knowledge graph embedding learning system for defect diagnosis in additive manufacturing

Since additive manufacturing involves complex physical processes and is influenced by many process parameters, it is challenging to manufacture defect-free products. To explore the causes of defects and obtain better product quality, exploring the relations among these factors is necessary. The knowledge graph is a powerful technique to express the relationships between entities and has been applied to many real-world tasks, such as disease diagnosis, recommendation systems, and traffic prediction. In this paper, a novel knowledge graph embedding learning system is presented, in which a designed knowledge graph ontology for AM and fact triples collected from the literature are utilized for the construction of a knowledge graph for defect diagnosis in additive manufacturing (KDDAM). A relational enhanced graph convolutional network is proposed to uncover new hidden relations among parameters, effects, defects, and product quality in the KDDAM. To validate the performance of the proposed model, we conducted many experiments on KDDAM and compared it with other baseline and state-of-the-art models. Our model yielded better results in two traditional link prediction metrics, with an MRR of 0.316 and 0.7, respectively. In addition, a case study shows that our model can discover new causes of defects to help us optimize the process.