Heterogeneous hypergraph learning for analyzing surface defects in additive manufacturing process

Since surface quality is influenced by many factors in the additive manufacturing (AM) process and the relationships among these factors are so complex, it is difficult to represent and control them. With the rapid development of graph representation, graphs have become a popular method to represent complex relationships. Many embedding learning methods are correspondingly proposed to extract the information and discover new relationships. As a result, this paper presents a novel heterogeneous hypergraph learning framework to learn the embeddings and reconstruct the graph for AM process analysis and optimization. In the framework, an additive manufacturing experimental dataset is used to generate a heterogeneous hypergraph. Hence, a novel heterogeneous hypergraph embedding learning method, named exp2vec is proposed to obtain a low-dimensional representation of the graph, in which hyperedge embedding is added to improve the embedding learning performance. These embeddings are fed into a generative model, named variational graph auto-encoder with correction (VGAE-Corr) to reconstruct the graph for link prediction. A series of experiments on a heterogeneous hypergraph for AM are conducted. The results show the superiority of the proposed model regarding link prediction performance. A case study shows that the new model has the ability to analyze surface quality and process optimization.