Scholarlink

<p>With the continuous advancement of industrial automation, motors as essential driving components play a vital role in maintaining the stability and safety of the entire production system. To improve the accuracy and real-time performance of motor fault diagnosis, this paper proposes an intelligent multi-fault diagnosis method for motors based on deep learning and data fusion. Unlike existing approaches that typically rely on single-source signals or shallow feature representations, the proposed method introduces a unified framework that combines multisource data fusion with a wide-kernel CNN, residual learning, BiLSTM, and an adaptive attention mechanism to enhance spatiotemporal feature extraction. Focusing on bearing faults and inter-turn short circuit faults in stator windings, the method integrates current and vibration signals collected during motor operation. First, discrete wavelet transform is applied to denoise the raw signals and enhance their time-frequency features. Then, a wide kernel one-dimensional convolutional neural network (Wide Kernel 1D-CNN) is employed to extract spatial features, with residual connections introduced to mitigate gradient vanishing in deep networks. Subsequently, a bidirectional long short-term memory network (BiLSTM) models the temporal dynamics of the signals, and an adaptive attention mechanism is incorporated to enhance the focus on critical features. Finally, the trained model is deployed to edge computing devices and integrated with a cloud platform for remote monitoring and fault diagnosis. Experimental results show that the model achieves an accuracy of 98.16 % on unseen data and maintains over 95 % accuracy across various fault conditions, demonstrating strong robustness and promising potential for practical engineering applications.</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 motor multi-fault diagnosis algorithm based on wide kernel convolutional neural networks

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Lihuan Shao,
  • Hongtao Jiang,
  • Hezhan Sun,
  • Xinliang Zeng,
  • Yanjun Deng

摘要

With the continuous advancement of industrial automation, motors as essential driving components play a vital role in maintaining the stability and safety of the entire production system. To improve the accuracy and real-time performance of motor fault diagnosis, this paper proposes an intelligent multi-fault diagnosis method for motors based on deep learning and data fusion. Unlike existing approaches that typically rely on single-source signals or shallow feature representations, the proposed method introduces a unified framework that combines multisource data fusion with a wide-kernel CNN, residual learning, BiLSTM, and an adaptive attention mechanism to enhance spatiotemporal feature extraction. Focusing on bearing faults and inter-turn short circuit faults in stator windings, the method integrates current and vibration signals collected during motor operation. First, discrete wavelet transform is applied to denoise the raw signals and enhance their time-frequency features. Then, a wide kernel one-dimensional convolutional neural network (Wide Kernel 1D-CNN) is employed to extract spatial features, with residual connections introduced to mitigate gradient vanishing in deep networks. Subsequently, a bidirectional long short-term memory network (BiLSTM) models the temporal dynamics of the signals, and an adaptive attention mechanism is incorporated to enhance the focus on critical features. Finally, the trained model is deployed to edge computing devices and integrated with a cloud platform for remote monitoring and fault diagnosis. Experimental results show that the model achieves an accuracy of 98.16 % on unseen data and maintains over 95 % accuracy across various fault conditions, demonstrating strong robustness and promising potential for practical engineering applications.