Scholarlink

<p>Many intelligent bearing fault diagnosis approaches have been explored by researchers. However, how to realize accurate compound fault diagnosis for rolling bearings, particularly under variable working conditions remains a major challenge. To this end, a novel parallel multi-scale convolutional neural network with incremental learning (PMSCNN-IL) framework is developed for efficient intelligent diagnosis. Initially, a denoising approach is employed to maximally mitigate noise contamination in vibration signal analysis. Subsequently, incorporating the periodicity of vibration signals while simultaneously capturing detailed features across various scales through parrallel multi-scle convolution block, an attention mechanism (AM) is adopted for feature fusion, thereby enhancing feature extraction capabilities. Lastly, to facilitate adaptation to variable working conditions, incremental learning (IL) is introduced, granting PMSCNN-IL the ability to continuously learn new fault features from data streams. The effectiveness of the PMSCNN-IL is validated on two experimental datasets: Case Western Reserve University (CWRU) dataset and Self-Collected dataset. Compound fault diagnosis experiments demonstrate that PMSCNN-IL achieves a high accuracy, exceeding 98%, for rolling bearings. PMSCNN-IL maintains high diagnostic accuracy on both datasets even under the variable working conditions. Furthermore, to assess the generalization capability of the PMSCNN-IL, we conducted industrial experiments using Wind Turbine Dataset. Experimental results demonstrate that PMSCNN-IL exhibits robust diagnostic performance even when applied to complexity and variability of real-world operating conditions. This study highlights the effectiveness and reliability of integrating artificial intelligence with signal processing for bearing fault diagnosis, thereby facilitating advancements in intelligent fault diagnosis technologies.</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

Parallel multi-scale convolutional neural network with incremental learning for bearing compound fault diagnosis under variable working conditions

  • 期刊论文

  • 发表日期：

    2025年9月2日

查看全文
  • Shuzhen Han,
  • Shengke Sun,
  • Ke Pang,
  • Jianfei Li,
  • Fujun Tian,
  • Dong Zhen,
  • Guojin Feng,
  • Pingjuan Niu

摘要

Many intelligent bearing fault diagnosis approaches have been explored by researchers. However, how to realize accurate compound fault diagnosis for rolling bearings, particularly under variable working conditions remains a major challenge. To this end, a novel parallel multi-scale convolutional neural network with incremental learning (PMSCNN-IL) framework is developed for efficient intelligent diagnosis. Initially, a denoising approach is employed to maximally mitigate noise contamination in vibration signal analysis. Subsequently, incorporating the periodicity of vibration signals while simultaneously capturing detailed features across various scales through parrallel multi-scle convolution block, an attention mechanism (AM) is adopted for feature fusion, thereby enhancing feature extraction capabilities. Lastly, to facilitate adaptation to variable working conditions, incremental learning (IL) is introduced, granting PMSCNN-IL the ability to continuously learn new fault features from data streams. The effectiveness of the PMSCNN-IL is validated on two experimental datasets: Case Western Reserve University (CWRU) dataset and Self-Collected dataset. Compound fault diagnosis experiments demonstrate that PMSCNN-IL achieves a high accuracy, exceeding 98%, for rolling bearings. PMSCNN-IL maintains high diagnostic accuracy on both datasets even under the variable working conditions. Furthermore, to assess the generalization capability of the PMSCNN-IL, we conducted industrial experiments using Wind Turbine Dataset. Experimental results demonstrate that PMSCNN-IL exhibits robust diagnostic performance even when applied to complexity and variability of real-world operating conditions. This study highlights the effectiveness and reliability of integrating artificial intelligence with signal processing for bearing fault diagnosis, thereby facilitating advancements in intelligent fault diagnosis technologies.