Scholarlink

<p>Fault diagnosis of rotating machinery is essential for ensuring safety and improving efficiency in the industrial field. However, existing methods encounter challenges in fault diagnosis under complex conditions, such as the impact of variable loads and noise interference on diagnostic performance. To address these issues, this work proposes a model named CBMA for diagnosing rotating machinery faults based on spatiotemporal feature fusion. The model integrates convolutional neural networks (CNN), bidirectional temporal convolutional networks (BITCN), and multi-head attention (MHA) mechanisms, enhancing the ability to capture and fuse spatiotemporal features through an optimized architecture. Initially, vibration signals are preprocessed using variational mode decomposition (VMD) and discrete wavelet transform (DWT) to extract multilayer time-frequency features. Subsequently, CNN and BITCN are for spatial and bidirectional temporal feature extraction, and MHA achieves effective feature fusion. The introduction of a local time response enhancement (LTRE) mechanism and an adaptive feature transformation (AFT) module further enhances the model’s adaptability and precision. Experimental results demonstrate that CBMA achieves an average accuracy as high as 99.65 % and 98.82 % on two public datasets, surpassing other models by 1.80–18.84 % under conditions of variable speeds and loads. Remarkably, CBMA still maintains an accuracy of 97.8 % at a -6 dB signal-to-noise ratio. Furthermore, ablation studies confirm the positive contributions of each component to the overall performance of the model.</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 fault diagnosis method for rotating machinery based on spatiotemporal feature fusion

  • 期刊论文

  • 发表日期：

    2025年8月4日

查看全文
  • Chengjun Wang,
  • Mingxin Wang

摘要

Fault diagnosis of rotating machinery is essential for ensuring safety and improving efficiency in the industrial field. However, existing methods encounter challenges in fault diagnosis under complex conditions, such as the impact of variable loads and noise interference on diagnostic performance. To address these issues, this work proposes a model named CBMA for diagnosing rotating machinery faults based on spatiotemporal feature fusion. The model integrates convolutional neural networks (CNN), bidirectional temporal convolutional networks (BITCN), and multi-head attention (MHA) mechanisms, enhancing the ability to capture and fuse spatiotemporal features through an optimized architecture. Initially, vibration signals are preprocessed using variational mode decomposition (VMD) and discrete wavelet transform (DWT) to extract multilayer time-frequency features. Subsequently, CNN and BITCN are for spatial and bidirectional temporal feature extraction, and MHA achieves effective feature fusion. The introduction of a local time response enhancement (LTRE) mechanism and an adaptive feature transformation (AFT) module further enhances the model’s adaptability and precision. Experimental results demonstrate that CBMA achieves an average accuracy as high as 99.65 % and 98.82 % on two public datasets, surpassing other models by 1.80–18.84 % under conditions of variable speeds and loads. Remarkably, CBMA still maintains an accuracy of 97.8 % at a -6 dB signal-to-noise ratio. Furthermore, ablation studies confirm the positive contributions of each component to the overall performance of the model.