EUVIMEDEUVIMED entdecken
Crossref

A Physics-Guided Spatiotemporal Feature Cross-Fusion Network for Damage Localization in CFRP

Chuntao Wei, Xiaorui Chu, Chunlei Jiang, Quanhui Sha

IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 1-19

DOI: 10.1109/tim.2026.3680183 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Chuntao Wei und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 54 Literaturangaben. Gefördert von: 2024 Science and Technology Plan Fund Project of Liangshan Prefecture, China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Ultrasonic guided-wave technology has attracted increasing interest for damage localization in carbon fiber-reinforced polymer (CFRP) structures. An accurate damage localization requires the integrated utilization of ultrasonic guided wave (UGW) signals and the physical spatial information from sensor networks, relying on a dense and well-designed sensor layout. However, most existing deep learning-based approaches still lack the capability to explicitly encode and leverage physical spatial information within a unified neural network framework. Moreover, in densely deployed sensor networks, these methods often introduce substantial high-dimensional redundancy and considerable computational overhead, thereby limiting both model performance and practical applicability in engineering scenarios. To solve these challenges, this article proposes an UGW-based damage monitoring method using a physics-guided spatiotemporal feature cross-fusion network (PSTCFN) for accurate damage localization in CFRP. First, based on a predeployed sparse sensor network, the guided wave signals are converted into a graph structure representation. Sensor paths serve as nodes in the graph, while the physical spatial distance between sensor paths, the interpath propagation direction, and signal similarity jointly determine the edges and their weights. This establishes a geometric-signal-associated graph convolutional network (GS-GCN) model guided by physical mechanism information, linking geometric space with signal associations to extract complex spatial features from sensor-path signals. Second, the time-dependent features of the sensor path signals are mined with bidirectional gated recurrent unit (BiGRU), which are concatenated with the spatial features mined by GS-GCN to obtain the implied spatiotemporal dependent features of the signals. Finally, a graph-based multihead bidirectional cross-attention mechanism is proposed to achieve cross fusion of spatial and temporal features, enabling the automatic extraction of critical damage-related features for accurate damage localization. The performance of the method is evaluated using datasets from the open-guided wave (OGW) platform. Experimental results demonstrate that, with only four piezoelectric transducers (PZTs), the proposed method achieves 100% accuracy in damage region identification and delivers excellent precision and fit in coordinate-level regression localization, while exhibiting superior robustness and generalization over all baseline models—even in small-sample scenarios.

Förderung

  • 2024 Science and Technology Plan Fund Project of Liangshan Prefecture, China

Zitieren

APA

Wei, C., Chu, X., Jiang, C., & Sha, Q. (2026). A Physics-Guided Spatiotemporal Feature Cross-Fusion Network for Damage Localization in CFRP. IEEE Transactions on Instrumentation and Measurement, 75, 1-19. https://doi.org/10.1109/tim.2026.3680183

Vancouver

Wei C, Chu X, Jiang C, Sha Q. A Physics-Guided Spatiotemporal Feature Cross-Fusion Network for Damage Localization in CFRP. IEEE Transactions on Instrumentation and Measurement. 2026;75:1-19. doi:10.1109/tim.2026.3680183

BibTeX
@article{Wei2026,
  author = {Wei, Chuntao and Chu, Xiaorui and Jiang, Chunlei and Sha, Quanhui},
  title = {A Physics-Guided Spatiotemporal Feature Cross-Fusion Network for Damage Localization in CFRP},
  journal = {IEEE Transactions on Instrumentation and Measurement},
  year = {2026},
  volume = {75},
  pages = {1--19},
  doi = {10.1109/tim.2026.3680183}
}
Kostenlos starten →