Scholarlink

<p>The vibration time-series signals generated during hoist operation are often accompanied by complex noise, which severely hampers the accurate identification of bearing fault features. Therefore, denoising the time-series signals from hoist operation is crucial to improving the reliability of fault diagnosis. This paper proposes a vibration signal denoising model that integrates U-Net and Transformer architectures. The core design involves inserting multi-layer Transformer encoders into the bottleneck of the U-Net structure. This fusion mechanism leverages the U-Net’s skip connections to preserve high-resolution local signal details while simultaneously utilizing the Transformer’s self-attention mechanism to capture long-range temporal dependencies in the down-sampled features, enabling end-to-end signal denoising that effectively distinguishes periodic fault signatures from noise. Furthermore, a two-stage hoist fault diagnosis method is introduced. First, the proposed denoising model denoises the raw 1D vibration signal. Second, the denoised signal is transformed into a 2D time-frequency representation via Continuous Wavelet Transform (CWT), which is then fed into a ResNet18 model for final fault classification. Comparative evaluations with various common denoising and fault diagnosis algorithms confirm the superiority of the proposed approach. Under both low and high noise conditions, the proposed fault diagnosis method achieves an average accuracy of 99.8%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hoist fault diagnosis method integrating time series signal denoising model

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Xiaojie Yu,
  • Chaowei Zang,
  • Huayu Shou,
  • Qiang Niu

摘要

The vibration time-series signals generated during hoist operation are often accompanied by complex noise, which severely hampers the accurate identification of bearing fault features. Therefore, denoising the time-series signals from hoist operation is crucial to improving the reliability of fault diagnosis. This paper proposes a vibration signal denoising model that integrates U-Net and Transformer architectures. The core design involves inserting multi-layer Transformer encoders into the bottleneck of the U-Net structure. This fusion mechanism leverages the U-Net’s skip connections to preserve high-resolution local signal details while simultaneously utilizing the Transformer’s self-attention mechanism to capture long-range temporal dependencies in the down-sampled features, enabling end-to-end signal denoising that effectively distinguishes periodic fault signatures from noise. Furthermore, a two-stage hoist fault diagnosis method is introduced. First, the proposed denoising model denoises the raw 1D vibration signal. Second, the denoised signal is transformed into a 2D time-frequency representation via Continuous Wavelet Transform (CWT), which is then fed into a ResNet18 model for final fault classification. Comparative evaluations with various common denoising and fault diagnosis algorithms confirm the superiority of the proposed approach. Under both low and high noise conditions, the proposed fault diagnosis method achieves an average accuracy of 99.8%.