Scholarlink

<p>The existing intelligent fault diagnosis methods are usually designed only for a single diagnostic task, and their performance significantly deteriorates in multitasking scenarios. Therefore, this paper proposes a Multi-scale Residual Multi-Task Depthwise Separable Network (MSRTDSN) for bearing fault diagnosis. Firstly, a shared network is constructed through a depthwise separable network to extract universal features; Secondly, a task branch network is established using multi-scale residual modules to diagnose fault types, degrees, and operating conditions in parallel, and a dynamic loss weighting algorithm is used to adaptively balance task learning; Finally, convert the one-dimensional vibration signal into a two-dimensional Gram Angular Difference Field (GADF) image as input. Experiments on the CWRU and PU bearing datasets show that MSRTDSN outperforms the comparison model in multi task diagnostic accuracy, demonstrating strong feature extraction and generalization abilities.</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

Bearing fault diagnosis method based on multi-scale residual and multi task depthwise separable network

  • 期刊论文

  • 发表日期：

    2025年6月23日

查看全文
  • Chengfei Yang,
  • Xing Shao,
  • Jun Gao,
  • Cuixiang Wang,
  • Chen Qian

摘要

The existing intelligent fault diagnosis methods are usually designed only for a single diagnostic task, and their performance significantly deteriorates in multitasking scenarios. Therefore, this paper proposes a Multi-scale Residual Multi-Task Depthwise Separable Network (MSRTDSN) for bearing fault diagnosis. Firstly, a shared network is constructed through a depthwise separable network to extract universal features; Secondly, a task branch network is established using multi-scale residual modules to diagnose fault types, degrees, and operating conditions in parallel, and a dynamic loss weighting algorithm is used to adaptively balance task learning; Finally, convert the one-dimensional vibration signal into a two-dimensional Gram Angular Difference Field (GADF) image as input. Experiments on the CWRU and PU bearing datasets show that MSRTDSN outperforms the comparison model in multi task diagnostic accuracy, demonstrating strong feature extraction and generalization abilities.