Scholarlink

<p>To address the limitations of traditional electromechanical fault detection methods, which typically rely on component-level modeling and focus on single-fault scenarios with limited accuracy for compound faults. This paper proposes an integrated detection framework combining Function-Movement-Action-Part (FMAP), Three-dimensional Convolutional Block Attention Module Convolutional Neural Networks (3DCBAM-CNN), and Two-dimensional Gated Recurrent Unit (2DGRU), referred to as the FCCG model. The approach begins by decomposing the electromechanical system using FMAP to extract Meta-actions (MAs) and Meta-action Units (MAUs), which form the basis for fault data modeling. A sliding window (SW) is then applied to extract temporal features and convert them into two-dimensional (2D) images. These are processed through convolution layers to construct three-dimensional (3D) representations, within which CBAM attention modules are embedded to enhance channel and spatial feature representation. A transition layer fuses the 3D output with the original image to form the input of the 2DGRU, enabling spatiotemporal feature extraction. Finally, fault probabilities are obtained using Softmax and Argmax functions to determine the fault category. Experiments conducted on a custom-built dataset show that the proposed method achieves 99.14% in both accuracy and recall. Comparative results demonstrate that FCCG performs particularly well in multi-state fault classification, indicating its strong potential for real-time, high-precision, and comprehensive fault detection in complex electromechanical systems.</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

Data-driven fault collaborative diagnosis in mechatronic equipment: a meta-action based spatiotemporal fusion approach

  • 期刊论文

  • 发表日期：

    2025年8月27日

查看全文
  • Minghui Gu,
  • Yulong Li,
  • Lijun Yu,
  • Xiaogang Zhang,
  • Hui Long,
  • Wenkai Yong

摘要

To address the limitations of traditional electromechanical fault detection methods, which typically rely on component-level modeling and focus on single-fault scenarios with limited accuracy for compound faults. This paper proposes an integrated detection framework combining Function-Movement-Action-Part (FMAP), Three-dimensional Convolutional Block Attention Module Convolutional Neural Networks (3DCBAM-CNN), and Two-dimensional Gated Recurrent Unit (2DGRU), referred to as the FCCG model. The approach begins by decomposing the electromechanical system using FMAP to extract Meta-actions (MAs) and Meta-action Units (MAUs), which form the basis for fault data modeling. A sliding window (SW) is then applied to extract temporal features and convert them into two-dimensional (2D) images. These are processed through convolution layers to construct three-dimensional (3D) representations, within which CBAM attention modules are embedded to enhance channel and spatial feature representation. A transition layer fuses the 3D output with the original image to form the input of the 2DGRU, enabling spatiotemporal feature extraction. Finally, fault probabilities are obtained using Softmax and Argmax functions to determine the fault category. Experiments conducted on a custom-built dataset show that the proposed method achieves 99.14% in both accuracy and recall. Comparative results demonstrate that FCCG performs particularly well in multi-state fault classification, indicating its strong potential for real-time, high-precision, and comprehensive fault detection in complex electromechanical systems.