Scholarlink

<p>Non-stationary noise under complex operating conditions makes it challenging to ensure stable accuracy in predicting the remaining useful life (RUL) of rolling bearings. To address this issue, this paper proposes a hybrid prediction method integrating signal processing, degradation modeling, and deep learning. First, a joint denoising strategy complete ensemble empirical mode decomposition with adaptive noise symbolic entropy wavelet threshold denoising (CEEMDAN-SE-WTD) is developed to enhance the quality of vibration signals. Second, a comprehensive feature evaluation index maximal information coefficient diversity entropy&#xa0;(MIC-DE) is constructed based on the Gamma degradation model, and an adaptive 3σ criterion is proposed to improve the consistency between features and degradation trends as well as the reliability of health stage division. Finally, a deep temporal model self-attention bidirectional temporal convolutional network and bidirectional gated recurrent unit&#xa0;(SA-BiTG) is designed, with its parameters optimized by the LVY algorithm (LVYA)&#xa0;to boost prediction performance. Experimental results demonstrate that the proposed method validates the effectiveness of the framework in complex degradation scenarios on both the XJTU-SY and PHM2012 datasets.</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

A hybrid model-driven approach for remaining useful life prediction of rolling bearings

  • 期刊论文

  • 发表日期：

    2026年5月23日

查看全文
  • Hongchuan Cheng,
  • He Zhang,
  • Guilong Fu,
  • Yirui Xu,
  • Zhaoyang Shi,
  • Weicai Gong,
  • Zhiwu Shang,
  • Xiafei Shi

摘要

Non-stationary noise under complex operating conditions makes it challenging to ensure stable accuracy in predicting the remaining useful life (RUL) of rolling bearings. To address this issue, this paper proposes a hybrid prediction method integrating signal processing, degradation modeling, and deep learning. First, a joint denoising strategy complete ensemble empirical mode decomposition with adaptive noise symbolic entropy wavelet threshold denoising (CEEMDAN-SE-WTD) is developed to enhance the quality of vibration signals. Second, a comprehensive feature evaluation index maximal information coefficient diversity entropy (MIC-DE) is constructed based on the Gamma degradation model, and an adaptive 3σ criterion is proposed to improve the consistency between features and degradation trends as well as the reliability of health stage division. Finally, a deep temporal model self-attention bidirectional temporal convolutional network and bidirectional gated recurrent unit (SA-BiTG) is designed, with its parameters optimized by the LVY algorithm (LVYA) to boost prediction performance. Experimental results demonstrate that the proposed method validates the effectiveness of the framework in complex degradation scenarios on both the XJTU-SY and PHM2012 datasets.