Scholarlink

<p>Due to the fewer fault samples, it is difficult to diagnose the fault of slewing bearings in complex working conditions. For this reason, a model based on Time-series Generative Adversarial Networks (Time GAN) combined with Synergistic Similarity Graph Construction (SSGC) and Graph Attention Network (GAT) is proposed. Time GAN is introduced to generate new training sample features while preserving the unique temporal correlation of its samples. SSGC method is utilized to construct graph structure data for the newly generated training samples and put them into the GAT model with multi-head attention mechanism for classification. This solves the problem that traditional deep learning methods cannot fully utilize the spatial relationship between training sample features under different working conditions. The experimental results show that the proposed method can effectively recognize each health state of slewing bearing with classification accuracy of up to 90%, which is better than other methods.</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 data enhanced algorithm for fault diagnosis of slewing bearings based on times-series generative adversarial networks

  • 期刊论文

  • 发表日期：

    2025年2月22日

查看全文
  • Li Sun,
  • Jun Wu,
  • Guochao Li,
  • Xiaodie Ren,
  • Jinjun Wang,
  • Sizhao Wen

摘要

Due to the fewer fault samples, it is difficult to diagnose the fault of slewing bearings in complex working conditions. For this reason, a model based on Time-series Generative Adversarial Networks (Time GAN) combined with Synergistic Similarity Graph Construction (SSGC) and Graph Attention Network (GAT) is proposed. Time GAN is introduced to generate new training sample features while preserving the unique temporal correlation of its samples. SSGC method is utilized to construct graph structure data for the newly generated training samples and put them into the GAT model with multi-head attention mechanism for classification. This solves the problem that traditional deep learning methods cannot fully utilize the spatial relationship between training sample features under different working conditions. The experimental results show that the proposed method can effectively recognize each health state of slewing bearing with classification accuracy of up to 90%, which is better than other methods.