Advancing Additive Manufacturing through Deep Learning: A Comprehensive Review of Current Progress and Future Challenges

TL;DR

This paper provides the first PRISMA-guided, comprehensive review of deep learning applications in additive manufacturing, organizing the literature into design for AM (DfAM), DL-driven AM modeling, and DL-driven process monitoring and control across seven AM categories. It identifies key DL trends, including generative models for design, physics-informed networks to reduce data requirements, and 3D point-cloud approaches for in-situ monitoring, while highlighting data-related challenges and uncertainty. The authors propose a conceptual framework and actionable directions to generalize DL to complex geometries, manage data quality and uncertainty, and improve interpretability. The work aims to guide researchers and industry practitioners in deploying DL across the AM lifecycle and to illuminate gaps for future research and standardization.

Abstract

This paper presents the first comprehensive literature review of deep learning (DL) applications in additive manufacturing (AM). It addresses the need for a thorough analysis in this rapidly growing yet scattered field, aiming to bring together existing knowledge and encourage further development. Our research questions cover three major areas of AM: (i) design for AM, (ii) AM modeling, and (iii) monitoring and control in AM. We use a step-by-step approach following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines to select papers from Scopus and Web of Science databases, aligning with our research questions. We include only those papers that implement DL across seven major AM categories - binder jetting, directed energy deposition, material extrusion, material jetting, powder bed fusion, sheet lamination, and vat photopolymerization. Our analysis reveals a trend towards using deep generative models, such as generative adversarial networks, for generative design in AM. It also highlights an increasing effort to incorporate process physics into DL models to improve AM process modeling and reduce data requirements. Additionally, there is growing interest in using 3D point cloud data for AM process monitoring, alongside traditional 1D and 2D formats. Finally, this paper summarizes the current challenges and recommends some of the promising opportunities in this domain for further investigation with a special focus on (i) generalizing DL models for a wide range of geometry types, (ii) managing uncertainties both in AM data and DL models, (iii) overcoming limited, imbalanced, and noisy AM data issues by incorporating deep generative models, and (iv) unveiling the potential of interpretable DL for AM.

Figures (7)

• Figure 1: PRISMA flowchart detailing the article selection process, adapted from the guidelines by page2021prisma_Prisma_Reference
• Figure 2: Summary of AM data types extracted from reviewed literature. These data types serve as inputs for DL models which leverage the inherent temporal, spatial, and structural information to enhance AM performance and quality.
• Figure 3: Proposed methodology framework for deep learning applications in various aspects of additive manufacturing
• Figure 4: Proposed framework of DL-enabled design for AM
• Figure 5: Generalized high-level DL-driven AM modeling framework