Skip to main content
U.S. flag

An official website of the United States government

Accurate keyhole instability prediction in metal additive manufacturing through machine learning-aided numerical simulation

Published

Author(s)

Jiahui Zhang, Runbo Jiang, Kangming Li, Pengyu Chen, Xiao Shang, Zhiying Liu, Brian Simonds, Qianglong Wei, Hongze Wang, Jason Hattrick-Simpers, Tao Sun, Anthony Rollet, Yu Zou

Abstract

A primary obstacle impeding the use of metal additive manufacturing technologies in fatigue-sensitive applications is the presence of porosity, primarily caused by keyhole instability. To tackle this challenge, it is imperative to accurately forecast keyhole behaviors, characterized by keyhole depth, when employing different printing parameters such as laser power, scan speed, laser spot size, and material composition. In this study, we demonstrate the feasibility of employing computational fluid dynamics simulation model in conjunction with experimental laser absorptance data to quantitatively predict keyhole depth. The accuracy of the predictions is validated by experimental results derived from X-ray imaging. To predict keyhole depth under new printing parameters, we develop and compare three distinct approaches: (1) An iterative loop that combines a forward simulation model with a backward predictive model for automated predictions. (2) A simulation model based on laser absorptance predicted by an established machine learning model. (3) Machine learning models developed for direct mean keyhole depth prediction and then converted to keyhole instability. These methods advance our understanding and control of the metal additive manufacturing process across various materials and conditions, ultimately facilitating the fabrication of defect-free samples.
Citation
Science Advances

Citation

Zhang, J. , Jiang, R. , LI, K. , Chen, P. , Shang, X. , Liu, Z. , Simonds, B. , Wei, Q. , Wang, H. , Hattrick-Simpers, J. , Sun, T. , Rollet, A. and Zou, Y. (2025), Accurate keyhole instability prediction in metal additive manufacturing through machine learning-aided numerical simulation, Science Advances (Accessed April 2, 2025)

Issues

If you have any questions about this publication or are having problems accessing it, please contact reflib@nist.gov.

Created February 20, 2025