Metal additive manufacturing (AM), widely regarded as a revolution in modern manufacturing for its ability to produce lightweight and geometrically complex components, has long faced a critical barrier to widespread adoption: microscopic internal defects that are invisible to the naked eye yet significantly compromise structural integrity. Now, a research team led by Professor Hyoung Seop Kim of POSTECH (Pohang University of Science and Technology) has harnessed the power of artificial intelligence (AI) to overcome this challenge, marking a major leap forward in the reliability of metal 3D printing technology.
Professor Hyoung Seop Kim and integrated M.S.-Ph.D. student Jeong Ah Lee, from the Graduate Institute of Ferrous & Eco Materials Technology and the Department of Materials Science and Engineering at POSTECH, collaborated with Dr. Jeong Min Park's research team at the Korea Institute of Materials Science (KIMS) to develop an AI-based predictive framework capable of accounting for microscopic defects in metal 3D printing processes. The findings were published in Acta Materialia, a leading international journal in materials science.
Metal 3D printing works by melting and layering metal powder using a laser, a process known as laser powder bed fusion (LPBF). During fabrication, small voids called pores can form inside the material, acting like air bubbles that substantially degrade the mechanical strength of finished components. In demanding applications such as aircraft structures and automotive parts, where materials are subjected to extreme conditions, even minor porosity can prove catastrophic.
Until now, assessing the effect of such defects required extensive experimentation and considerable time, posing a significant bottleneck in materials development and qualification for safety-critical industries.
Rather than attempting to eliminate defects entirely, the research team reframed the problem by focusing on understanding and predicting defects scientifically. The team integrated porosity data alongside process parameters, microstructural features, and mechanical property data to train an AI model. They applied a technique called Data Selection Machine Learning (DSML), which identifies only the most influential variables from the dataset, effectively filtering out noise and focusing the model on the factors that matter most.
The approach is analogous to a physician interpreting a CT scan to diagnose disease: the AI analyzes the internal microstructure and defect characteristics of metal components to anticipate their mechanical behavior before any physical testing is performed.
A key distinguishing feature of this work is its commitment to interpretability. Rather than delivering a "black box" AI that produces results without explanation, the team employed symbolic regression, an interpretable AI approach, to derive human-readable mathematical equations that describe the predictive model.
These equations reflect the underlying physical reality: as porosity increases, the effective load-bearing cross-sectional area decreases, thereby reducing strength. In this way, the AI does not merely predict outcomes; it explains why those outcomes occur, thereby enhancing scientific transparency and trust.
To validate the framework, the team fabricated AlSi10Mg alloy, one of the most widely used aluminum alloys in 3D-printed aerospace and automotive components, under a variety of process conditions. The AI-based model successfully predicted the yield strength of components with a Mean Absolute Error (MAE) of just 9.51 MPa within seconds, eliminating the need for complex experimental procedures.
This represents more than a 4-fold improvement in prediction accuracy compared to conventional approaches, demonstrating the framework's robustness and practical utility.
The research team envisions that this framework can serve as the basis for a "defect-aware design map," enabling the prediction and design of material performance ibased on process conditions in advance, thereby significantly reducing the trial-and-error typically associated with materials development.
By making it possible to predict and account for invisible defects at the design stage, this technology has the potential to dramatically accelerate the qualification and commercialization of metal 3D-printed components in industries where safety and reliability are paramount.
"We have demonstrated that AI can be used to scientifically understand and control defects," said Jeong Ah Lee, first author of the study. Professor Hyoung Seop Kim added, "This technology will enhance the reliability of metal 3D-printed components and significantly accelerate their commercialization in sectors such as aerospace and automotive industries."
This research was supported by the Leading Research Center Program of the National Research Foundation of Korea (NRF), the KIMS Institutional Research Program, and Hyundai Motor Group. Jeong Ah Lee received support from the Next Generation of Researchers Fellowship Program of the National Research Foundation of Korea.
