Powder spreading anomaly detection in laser powder bed fusion additive manufacturing using the full-scale feature adaptive UNet++ network

Zhu, Min, Han, Quanquan, Zhang, Zhenhua, Wu, Defan, Zhao, Peng, Wang, Xiaodan, Liu, Yue, Ji, Ze ORCID: https://orcid.org/0000-0002-8968-9902 and Yang, Shoufeng 2026. Powder spreading anomaly detection in laser powder bed fusion additive manufacturing using the full-scale feature adaptive UNet++ network. Journal of Manufacturing Processes 157 , pp. 1274-1289. 10.1016/j.jmapro.2025.12.051

Full text not available from this repository.

Abstract

Powder bed anomalies in the laser powder bed fusion (LPBF) additive manufacturing process may cause various defects, potentially reducing the mechanical properties of the fabricated components. Although deep learning methods have been successfully employed to identify various powder bed anomalies, the categorization of “powder spreading anomaly” has been oversimplified into a single class, neglecting multi-layer insufficient powder spreading conditions. The correlation between insufficient powder spreading and mechanical properties also remains unclear, and the threshold at which insufficient powder spreading leads to a decline in mechanical properties has not been investigated. This study develops a novel powder spreading anomaly detection system (PSADS) for LPBF that includes a deep learning–based segmentation algorithm, known as Full-scale Feature Adaptive UNet++ (FFA-UNet++). FFA-UNet++ addresses challenges such as brightness homogenization, texture homogenization, and boundary blurring caused by multi-layer insufficient powder spreading. The algorithm enables the detection of six types of insufficient powder spreading, achieving a mean intersection over union (mIoU) close to 57 %, representing a 4 %–11 % improvement over six widely used segmentation networks, with a maximum IoU exceeding 90 % for individual anomaly categories. The study also investigates the effects of the six types of insufficient powder spreading on the mechanical properties of LPBF-fabricated 316 L stainless steel and proposes a process control guideline: The cumulative number of insufficient powder layers per part should not exceed three layers. This study provides a novel vision-based powder spreading anomaly detection method used for LPBF process and furnishes valuable insights for LPBF quality control within industrial applications.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering
Publisher:	Elsevier
ISSN:	1526-6125
Date of Acceptance:	18 December 2025
Last Modified:	08 Jan 2026 14:15
URI:	https://orca.cardiff.ac.uk/id/eprint/183732

Actions (repository staff only)

Edit Item