Multi-source data fusion monitoring system for super-elevation in laser powder bed fusion based on bi-stream cross-mode fusion network

Yingjie Zhang*, Canneng Fang, Jiong Zhang, Gang Chen, Zhangdong Chen, Honghong Du, Lang Cheng, Di Wang*

*Corresponding author for this work

Research output: Journal Publications and ReviewsRGC 21 - Publication in refereed journalpeer-review

Abstract

Precise and timely monitoring of super-elevation defects is crucial for maintaining powder bed uniformity and ensuring part quality in laser powder bed fusion (LPBF), as it prevents abnormal powder spreading and enhances recoater equipment reliability. To effectively analyze and fuse the complex signal features of different super-elevation severity under powder spreading and laser scanning scenarios, and overcome the accuracy and efficiency difficulty of multi-source data fusion monitoring, a bi-stream cross-mode fusion network (BCFNet) is proposed. First, this work analyzes the generation mechanisms of signals that are associated with super-elevation severity, and extracts high-quality and interpretable image and acoustic features to provide a reliable basis for subsequent fusion. Second, the multi-source features in the two scenarios are divided into recoater stream and laser stream for feature extraction and cross-attention fusion in parallel, improving the model efficiency and cross-mode interaction ability. Finally, a bi-linear feature fusion with residuals is used to efficiently fuse two streams into highly integrated key bi-stream fusion features, which improves the perception ability of complementary information. Experiment results show that BCFNet has a superior accuracy (99.08%), precision (99.74%), recall (97.50%), F1 (98.61%) and robustness compared with several data fusion methods. It can effectively mine the potential deep complementary features of multi-source data and provide more accurate and reliable monitoring. Furthermore, BCFNet demonstrates superior performance and robustness, even with limited training samples and across different noise levels. Notably, this work also verifies the spreading acoustic signal of recoater is a novel and reliable signal source for super-elevation monitoring. © 2024 Elsevier Ltd
Original languageEnglish
Article number108794
JournalOptics and Lasers in Engineering
Volume186
Online published2 Jan 2025
DOIs
Publication statusPublished - Mar 2025

Research Keywords

  • Bi-stream
  • Cross-mode
  • Data fusion
  • Laser powder bed fusion
  • Super-elevation

Fingerprint

Dive into the research topics of 'Multi-source data fusion monitoring system for super-elevation in laser powder bed fusion based on bi-stream cross-mode fusion network'. Together they form a unique fingerprint.

Cite this