Machine learning and knowledge graph based design rule construction for additive manufacturing

Abstract

Additive Manufacturing (AM) is becoming data-intensive while increasingly generating newly available data. The availability of AM data provides Design for AM (DfAM) with a newfound opportunity to construct AM design rules with improved understanding of AM’s influence on part qualities. To seize the opportunity, this paper proposes a novel approach for AM design rule construction based on machine learning and knowledge graph. First, this paper presents a framework that enables i) deploying machine learning for extracting knowledge on predictive additive manufacturability from data, ii) adopting ontology with knowledge graphs as a knowledge base for storing both a priori and newfound AM knowledge, and iii) reasoning with knowledge for deriving data-driven prescriptive AM design rules. Second, this paper presents a methodology that constructs knowledge on predictive additive manufacturability and prescriptive AM design rules. In the methodology, we formalize knowledge representations, extractions, and reasoning, which enhances automated and autonomous construction and improvements of AM design rules. The methodology then employs a machine learning algorithm of Classification and Regression Tree on measurement data from National Institute of Standards and Technology for construction of a Laser Powder Bed Fusion-specific design rule for overhang features. This work supports AI related decision-making in additive manufacturability analysis and (re-)design for AM and guides machine learning to addressing problems related to AM design rules. This work is also meaningful as it provides sharable AM design rule knowledge with the AM society.