EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Multi-Representation Dual-Stream Feature Fusion Lightweight Model for Mechanical Fault Identification

Xinyue Pei, Jiliang Mu, Xiujian Chou

AI · 13.11.2026 · 7 (11) · 486

DOI: 10.3390/ai7110486 ↗

Kurz gefasst

Fachartikel in der Zeitschrift AI, online erschienen am 13.11.2026. Verfasst von Xinyue Pei und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In the field of mechanical fault diagnosis, single-modal representations often fail to capture comprehensive fault characteristics of complex systems, leading to limited diagnostic accuracy and poor robustness against environmental interference. This paper proposes MobileNetV2_CS+DSCNN, a synergistic dual-stream learning framework fusing raw time-domain vibration waveforms and their CWT time–frequency representations. One-dimensional (1D) vibration signals are converted into two-dimensional (2D) time–frequency maps via Continuous Wavelet Transform (CWT) to extract multi-scale localized time–frequency features. The framework adopts a parallel dual-stream architecture: a Channel Shuffle (CS)-embedded MobileNetV2 for 2D time–frequency map extraction, and a depthwise separable convolutional neural network (DSCNN) for 1D raw signal feature learning. Distinct from the original Channel Shuffle designed for grouped convolution, CS is deployed to boost inter-channel information exchange inside the MobileNetV2 branch and enhance cross-stream feature interaction. A dynamic weighted fusion module adaptively integrates these heterogeneous features for fault classification. Experimental results demonstrate that the proposed method outperforms conventional single-representation approaches under diverse operating conditions. The lightweight framework achieves near-100% test accuracy and average Macro F1-score over 98.69%, with favorable computational efficiency and strong anti-noise performance.

Zitieren

APA

Pei, X., Mu, J., & Chou, X. (2026). Multi-Representation Dual-Stream Feature Fusion Lightweight Model for Mechanical Fault Identification. AI, 7(11), 486. https://doi.org/10.3390/ai7110486

Vancouver

Pei X, Mu J, Chou X. Multi-Representation Dual-Stream Feature Fusion Lightweight Model for Mechanical Fault Identification. AI. 2026;7(11):486. doi:10.3390/ai7110486

BibTeX
@article{Pei2026,
  author = {Pei, Xinyue and Mu, Jiliang and Chou, Xiujian},
  title = {Multi-Representation Dual-Stream Feature Fusion Lightweight Model for Mechanical Fault Identification},
  journal = {AI},
  year = {2026},
  volume = {7},
  number = {11},
  pages = {486},
  doi = {10.3390/ai7110486}
}
Kostenlos starten →