Page 53 - IJAMD-2-2

International Journal of AI for
Materials and Design ML-driven optimization in additive manufacturing

quality control, and enabling real-time monitoring. innovation in AM. Multi-modal data fusion techniques,
As AM continues to evolve into a key technology for incorporating thermal imaging, X-ray CT, and acoustic
next-generation manufacturing, the demand for higher sensors, will refine defect detection accuracy and improve
precision, material efficiency, and scalable production process stability. Moreover, the development of self-
methods has intensified. The application of ML in AM learning AI-driven AM systems, capable of autonomously
has addressed many of these challenges by providing data- optimizing print parameters and material compositions,
driven solutions for process parameter optimization, defect will enable fully automated, intelligent manufacturing
detection, and mechanical property prediction, thereby workflows.
reducing reliance on empirical trial-and-error methods.
Ultimately, the synergy between ML and AM is poised
This review explores the role of ML in optimizing 3D to revolutionize manufacturing by enabling higher
printing, highlighting advancements in process monitoring, precision, faster production cycles, and more sustainable
defect mitigation, material property prediction, and design fabrication processes. As interdisciplinary collaborations
optimization. In polymer-based AM, ML techniques have between materials science, computational modeling,
significantly enhanced printability assessment, process and AI engineering continue to grow, the vision of fully
stability, and surface quality prediction. Studies on FDM, functional, AI-optimized 3D printing for industrial-scale
DIW, and vat photopolymerization (SLA/DLP) have applications is becoming increasingly tangible. Continued
demonstrated how ML models, particularly CNNs, GANs, research efforts in scalable ML architectures, real-time
and time-series models, can improve printing accuracy adaptive control, and robust AM process simulations
by compensating for environmental variations, detecting will be essential for advancing next-generation intelligent
defects in real time, and optimizing material formulations. manufacturing ecosystems.
Similarly, in metal-based AM, ML applications in PBF
and DED have contributed to thermal control, melt pool Acknowledgments
monitoring, and mechanical property enhancement. None.
Reinforcement learning and hybrid physics-informed ML
models have facilitated adaptive process control, leading Funding
to improved consistency in microstructural integrity and
mechanical performance. This work was supported by the Technology Development
Program (Grant No. S3248116) funded by the Ministry
Despite these advancements, persistent technical of SMEs and Startups (MSS, Korea), and by the National
challenges remain in realizing the full potential of ML Research Foundation of Korea (NRF) grant, funded by
in AM. Data scarcity and variability continue to impede the Korean government (MSIT; Grant No. RS-2023-
model training, as high-quality labeled datasets are often 00211636).
limited in AM contexts. Models also struggle to generalize
across different machines, materials, and process settings, Conflict of interest
meaning an ML model tuned for one scenario may
perform poorly when applied elsewhere. Moreover, many Im Doo Jung is an Editorial Board Member of this journal,
advanced ML models function as “black boxes,” raising but was not in any way involved in the editorial and
interpretability concerns and limiting user trust in critical peer-review process conducted for this paper, directly or
manufacturing decisions. Achieving real-time, closed- indirectly. Separately, other authors declared that they
loop process control via ML is another hurdle due to have no known competing financial interests or personal
computational constraints, and current implementations relationships that could have influenced the work reported
still face latency and integration issues. Finally, there is a in this paper.
need to better integrate fundamental physics with data-
driven approaches to generate emerging physics-informed Author contributions
ML techniques that aim to improve prediction accuracy Conceptualization: Im Doo Jung, Hayeol Kim
and reliability while reducing reliance on massive training Visualization: Im Doo Jung, Hayeol Kim, Kyung-Hwan
data. Addressing these unresolved issues will be crucial Kim
for transitioning from promising prototypes to robust, Writing – original draft: All authors
industrial-grade intelligent AM systems. Writing – review & editing: Im Doo Jung, Hayeol Kim
In the future, the convergence of ML with advanced
sensing technologies, in situ monitoring systems, and Ethics approval and consent to participate
cloud-based manufacturing platforms will drive further Not applicable.

Volume 2 Issue 2 (2025) 47 doi: 10.36922/IJAMD025130010