A review of machine learning in additive manufacturing: design and process

Additive manufacturing (AM), owing to its unique manufacturing approach, can drive manufacturing towards higher levels of sophistication, flexibility, and intelligence within the context of Industry 4.0. However, the complexity of AM systems and their nature as data-intensive manufacturing domains present challenges in producing high-quality products. With the advancement of digital and computer technologies, data-driven machine learning (ML) has been widely applied in AM, as it provides effective methods for quality control, process optimization, and complex system modeling. This paper succinctly summarizes the various phases of utilizing ML to assist in AM processes. It elucidates the advantages of using ML over traditional methods in each phase, starting from the pre-processing phase of design for additive manufacturing (DfAM) and parameter optimization, through the processing phase of defect detection, to the post-processing phase of part quality assessment. The objective of DfAM is to optimize product design, taking into account the interactions between multiple variables. The nonlinear capability of ML can be effectively utilized in DfAM. Defect detection aims to ensure the repeatability, durability, and reliability of parts. Combining sensors with ML can guarantee part quality. Lastly, assessing part quality and inspecting various subsystems of AM are conducted from the perspective of the part's microstructure. This review explores the application of ML in addressing numerous issues related to the AM process workflow. It provides a comprehensive and systematic summary of the application of various ML models at different phases, investigates the potential of newer ML technologies to assist AM in the future, and concludes with the limitations of current ML applications in AM as well as future development directions. By intersecting these two dynamic fields, it aims to enrich the AM research domain.