EUVIMEDEUVIMED entdecken
Crossref

Wind Turbine Blade Fault Diagnosis Based on Adaptive Cross-Channel Learning With Fine-Tuned KAN-Transformer Model

Miao Yu, Runxin Zeng

IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 3513214-3513214

DOI: 10.1109/tim.2026.3690907 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Miao Yu und 1 weiteren Autor:innen. Die Arbeit nennt 19 Literaturangaben. Gefördert von: Beijing Natural Science Foundation, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate turbine blade fault diagnosis is crucial for wind farm operation and maintenance as well as power system stability. However, wind turbine blade vibration signals are usually interfered with by strong noise, which affects the extraction and identification of fault features. Therefore, this article introduces multichannel signals, explores mine inter-channel correlations, and proposes a fault diagnosis model based on adaptive cross-channel learning and fine-tuning the Kolmogorov–Arnold network (KAN)-Transformer. First, a 1-D CNN as a time-domain filter is combined with a wavelet transform as a frequency domain filter, and the input signal is blindly deconvoluted to enhance the signal processing capability. Meanwhile, in order to optimize the filter parameters to minimize the loss function, a lemming population-based optimization algorithm is introduced to dynamically adjust the parameters to enhance the noise reduction effect. In addition, Sin–Wavelet–spline is used instead of B-spline in the KAN to enhance the nonlinear fitting ability of the model. KAN layer is used instead of a multilayer perceptron (MLP) layer in the Transformer structure to improve the feature expression ability and fault identification accuracy, and the signal processing ability is enhanced by an adaptive multichannel tokenizer to adaptively perceive the feature distribution of each channel to enhance the characterization ability of the original input. Subsequently, the hierarchical cross-channel transformer encoder is designed to mine the deep dependencies between channels to enhance the robustness of the identification of key fault features. Finally, the attention-guided fault classifier integrates extracted features to accurately classify distinct blade health states. Compared to existing methods, this proposed method can effectively denoise and improve the accuracy of wind turbine blade fault diagnosis.

Förderung

  • Beijing Natural Science Foundation
  • National Natural Science Foundation of China

Zitieren

APA

Yu, M., & Zeng, R. (2026). Wind Turbine Blade Fault Diagnosis Based on Adaptive Cross-Channel Learning With Fine-Tuned KAN-Transformer Model. IEEE Transactions on Instrumentation and Measurement, 75, 3513214-3513214. https://doi.org/10.1109/tim.2026.3690907

Vancouver

Yu M, Zeng R. Wind Turbine Blade Fault Diagnosis Based on Adaptive Cross-Channel Learning With Fine-Tuned KAN-Transformer Model. IEEE Transactions on Instrumentation and Measurement. 2026;75:3513214-3513214. doi:10.1109/tim.2026.3690907

BibTeX
@article{Yu2026,
  author = {Yu, Miao and Zeng, Runxin},
  title = {Wind Turbine Blade Fault Diagnosis Based on Adaptive Cross-Channel Learning With Fine-Tuned KAN-Transformer Model},
  journal = {IEEE Transactions on Instrumentation and Measurement},
  year = {2026},
  volume = {75},
  pages = {3513214--3513214},
  doi = {10.1109/tim.2026.3690907}
}
Kostenlos starten →