Page 75 - MSAM-4-3

Materials Science in

Additive Manufacturing

ORIGINAL RESEARCH ARTICLE
Gaussian process-based interpretable prediction

of melt track morphology through melt pool in
additive manufacturing

1,2
Xin Lin 1 , Shilin Liu 1 , Haodong Chen 1 , Jinrong Mao 1 , and Kunpeng Zhu *
1 Precision Manufacturing Institute, Wuhan University of Science and Technology, Wuhan, Hubei,
China
2 Institute of Intelligent Machines, Hefei Institutes of Physical Science, Chinese Academy of Sciences,
Hefei, Anhui, China
(This article belongs to the Special Issue: Smart Additive Manufacturing: Product and Process
Qualification through Innovation in Design, Modeling, Monitoring, Machine Learning, Metrology, and
Materials Science)

Abstract

Melt track monitoring in the laser powder bed fusion (LPBF) process is crucial for
preventing internal defects in as-printed parts. Uncontrollable melt pool dynamic
behavior easily leads to melt track morphology defects. Existing monitoring methods
face challenges in balancing modeling accuracy and physical interpretability.
Specifically, traditional physics-based models typically require complex monitoring
equipment, extensive simulation data, and empirical formulas, resulting in high costs
and limited applicability. Meanwhile, conventional data-driven models lack physical
*Corresponding author:
Kunpeng Zhu constraints, leading to insufficient interpretability, process parameter sensitivity,
(zhukp@iamt.ac.cn) and poor generalization. To address these challenges, this article proposes a deep
Citation: Lin X, Liu S, Chen H, Gaussian process-based method for LPBF melt track morphology prediction. The
Mao J, Zhu K. Gaussian process- proposed model employs kernel functions in the first layer to learn melt pool evolution
based interpretable prediction of patterns and embeds the Rosenthal equation into the second-layer kernel function
melt track morphology through melt
pool in additive manufacturing. as a physical constraint, constructing a physically interpretable multilayer Gaussian
Mater Sci Add Manuf. process framework. Finally, a softmax classifier based on melt track geometric
2025;4(3):025200030. deviation achieves five-category melt track morphology recognition. Multi-condition
doi: 10.36922/MSAM025200030
experimental results demonstrated that the proposed method achieved root mean
Received: May 12, 2025 square errors of 0.069, 0.020, and 0.039 for melt track geometry, outperforming
Revised: June 5, 2025 traditional data-driven models in prediction accuracy. The classification accuracy
reached 90.76%. Furthermore, the influence of different features on melt track
Accepted: June 8, 2025
morphology is quantified through time-lagged mutual information analysis and
Published online: July 17, 2025 other visualization methods. This study provides an effective solution for achieving
Copyright: © 2025 Author(s). quality monitoring and defect prediction in the LPBF process.
This is an Open-Access article
distributed under the terms of the
Creative Commons Attribution Keywords: Laser powder bed fusion; Deep Gaussian process; Morphology prediction;
License, permitting distribution, Physical constraint; Melt pool monitoring
and reproduction in any medium,
provided the original work is
properly cited.
Publisher’s Note: AccScience
Publishing remains neutral with 1. Introduction
regard to jurisdictional claims in
published maps and institutional Additive manufacturing (AM) technology has made remarkable developments in
affiliations. the manufacturing field in recent years and is an innovative approach that surpassed

Volume 4 Issue 3 (2025) 1 doi: 10.36922/MSAM025200030