Scholarlink

<p>Rolling bearings being a critical element in rotating machines, their fault diagnosis is vital for the safe and reliable operations of mechanical systems. Deep learning techniques have emerged in vibration signal-based rolling bearing fault classification due to the paradigm shift towards data driven diagnosis with reduced human intervention. However, the noise present in the vibration signals reduces the efficiency of automated fault diagnosis especially in industrial applications. An automated denoising method for recurrence plot (RP) images of the vibration signals using cycleGAN is presented. The cycleGAN network is initially trained with variational mode decomposition (VMD) denoised RP images to automate the denoising process. Subsequently, the cycleGAN denoised RP images of vibration signals from different conditions of rolling bearings are classified using DenseNet and Inception networks. The performance of the proposed methodology is evaluated using datasets from a test rig and the open source dataset of the Case Western Reserve University (CWRU). Accuracy values achieved at 99.4 % for test rig and at 99.6 % for CWRU datasets validate that the proposed method outperform other methods in open literature.</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

CycleGAN denoising of recurrence plots of rolling bearing vibration signals for improved fault classification

  • 期刊论文

  • 发表日期：

    2025年1月31日

查看全文
  • Thomas Joseph,
  • Sudeep U.,
  • Keerthi Krishnan K.

摘要

Rolling bearings being a critical element in rotating machines, their fault diagnosis is vital for the safe and reliable operations of mechanical systems. Deep learning techniques have emerged in vibration signal-based rolling bearing fault classification due to the paradigm shift towards data driven diagnosis with reduced human intervention. However, the noise present in the vibration signals reduces the efficiency of automated fault diagnosis especially in industrial applications. An automated denoising method for recurrence plot (RP) images of the vibration signals using cycleGAN is presented. The cycleGAN network is initially trained with variational mode decomposition (VMD) denoised RP images to automate the denoising process. Subsequently, the cycleGAN denoised RP images of vibration signals from different conditions of rolling bearings are classified using DenseNet and Inception networks. The performance of the proposed methodology is evaluated using datasets from a test rig and the open source dataset of the Case Western Reserve University (CWRU). Accuracy values achieved at 99.4 % for test rig and at 99.6 % for CWRU datasets validate that the proposed method outperform other methods in open literature.