A review of deep learning in metal additive manufacturing: Impact on process, structure, and properties

Deep learning (DL) is increasingly used to predict and control the formation of microstructures, optimize properties, and reduce defects in additively manufactured metallic components. This review examines the specific applications of deep learning in additive manufacturing (AM), such as part design and architecture, in-situ process sensing and monitoring, microstructure and property control, defect detection, and the mitigation of residual stress and distortion. The review emphasizes the significance of computational resources, data requirements, and the role of physics-informed deep learning in advancing these applications. Additionally, best practices for algorithm selection and dataset suitability are addressed, along with current research gaps that hinder progress, including challenges in understanding AM processes and enhancing computational efficiency. Finally, the outlook presents future directions for research, underscoring the importance of real-time implementation and model interpretability. This work aims to provide a foundational framework for researchers and practitioners looking to leverage deep learning in the evolving field of additive manufacturing.