Data-driven modeling of process, structure and property in additive manufacturing: A review and future directions

A thorough understanding of complex process-structure-property (P-S-P) relationships in additive manufacturing (AM) has long been pursued due to its paramount importance in achieving AM process optimization and quality control. Physics-based modeling and experimental approaches are usually time-consuming and/or costly. With the increasing availability of digital AM data and rapid development of data-driven modeling techniques, especially machine learning (ML), data-driven AM modeling is emerging as an effective approach towards this end. It allows for automatic discovery of patterns and trends in the AM data, construction of quantitative models of P-S-P relationships over the parameter space and prediction at unseen points without having to perform new physical modeling or experiments. A proliferation of researches on data-driven modeling of process, structure and property in AM have been witnessed in recent years. In this context, this paper aims to provide a systematic review of existing data-driven AM modeling with respect to different quantities of interest (QoI) along the process-structure-property chain. Specifically, this paper provides a summary of important information (i.e., input features, QoI-related output, data source and data-driven models) on existing data-driven AM modeling, as well as an in-depth analysis on relevant success achieved so far. Based on the comprehensive review, this paper also critically discusses the major limitations faced today and identifies some research directions that are promising for significantly advancing data-driven AM modeling in the future.

Automated summary

This paper provides a systematic review of existing data-driven additive manufacturing (AM) modeling, focusing on different quantities of interest (QoI) along the process-structure-property chain. It summarizes important information and analyzes the successes achieved so far. The paper also discusses the limitations and suggests promising research directions for advancing data-driven AM modeling.