Optical Spectral Physics-Informed Attention Network for Condition Monitoring in WAAM

Online condition monitoring of wire arc additive manufacturing using plasma arc spectra is essential. Purely data-driven monitoring models often cannot provide reliable classifications for out-of-sample scenarios. This article proposes an optical spectral physics-informed global-local network (GLNet...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on industrial electronics (1982) 2024-08, Vol.71 (8), p.9708-9718
Hauptverfasser: Chen, Lin, Yang, Fei, Wang, Rui, Zhang, Yang, Diao, Zhaowei, Rong, Mingzhe
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Online condition monitoring of wire arc additive manufacturing using plasma arc spectra is essential. Purely data-driven monitoring models often cannot provide reliable classifications for out-of-sample scenarios. This article proposes an optical spectral physics-informed global-local network (GLNet) based on attention and optical spectral domain knowledge. Initially, an analysis of the physical characteristics of optical spectral signals is undertaken, with domain knowledge subsequently incorporated into the channel information during the network input construction. The architecture of GLNet includes a local branch (L-branch) to extract local structural features and a global branch (G-Branch) to consider how local patches interact globally. Additionally, this article devises a multiscale feature fusion block, which integrates multilevel features to enhance the discrimination ability of the fault condition monitoring model. In essence, the proposed network combines the power of deep learning with the generalizability of physics-based optical spectral features, enabling condition monitoring under varying conditions. Finally, the proposed method is experimentally verified on the copper and chrome zirconium copper dataset, which has considerable improvements in terms of classification accuracy.
ISSN:0278-0046
1557-9948
DOI:10.1109/TIE.2023.3325570