Machine learning in Directed Energy Deposition (DED) additive manufacturing: A state-of-the-art review

Abstract

Directed Energy Deposition (DED) has become very popular for repair and rapid prototyping in metal manufacturing industries. However, as an anisotropic and defect-prone process, DED’s versatility and usability are currently limited. Machine Learning (ML) has been introduced to various Additive Manufacturing (AM) fields due to its functional ability to recognize complex process-structure–property (PSP) relationships. Yet, it has only been heavily employed in applications of DED very recently. Therefore, this work focuses on describing the different aspects related to DED in terms of ML and put forward a novel approach to summarize the different applications of ML approaches in DED. The methodology intends to catalog the three main aspects of the whole scenario, such as understanding the current problem domain of DED concerning the intricate phenomena and desired outcomes, visualizing the data stream, and finally, determining the suitable ML approach for the problem. This paper provides a state-of-art review of the defined problem domain based on properties, quality, defects, and process optimization, a list of external sensors, equipment, and material type required for the experiments, the ML approaches such as Supervised and Unsupervised learning with suitable algorithms and the available data types, as well as an initial detailed groundwork to provide an insight for the prospects of DED.