Scholarlink

This study presents an approach for diagnosing bearing damage using vibration signal analysis combined with a Convolutional Neural Network (CNN). Vibration signals collected from an electric generator under various operating conditions are first transformed into spectrogram images through the Short-Time Fourier Transform (STFT). These images are then utilized as inputs for a custom-designed CNN architecture, termed Bearing-CNN, which integrates both sequential and parallel layers to effectively extract and classify features associated with common bearing faults such as retainer cracks, ball roller scratches, and outer ring cracks. The experimental results demonstrate that the proposed Bearing-CNN achieves a test accuracy of 99.96% with significantly reduced computational complexity and training time compared to well-known models like GoogLeNet and SqueezeNet. The promising performance of Bearing-CNN underlines its potential for real-time fault diagnosis in industrial applications, providing a robust and efficient solution for early fault detection and maintenance planning.

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

Diagnose Bearing Damage Through Vibration Signals Based on CNN Network

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ngo Thi Hoa,
  • Tang Ha Minh Quan,
  • Tran Minh Ket

摘要

This study presents an approach for diagnosing bearing damage using vibration signal analysis combined with a Convolutional Neural Network (CNN). Vibration signals collected from an electric generator under various operating conditions are first transformed into spectrogram images through the Short-Time Fourier Transform (STFT). These images are then utilized as inputs for a custom-designed CNN architecture, termed Bearing-CNN, which integrates both sequential and parallel layers to effectively extract and classify features associated with common bearing faults such as retainer cracks, ball roller scratches, and outer ring cracks. The experimental results demonstrate that the proposed Bearing-CNN achieves a test accuracy of 99.96% with significantly reduced computational complexity and training time compared to well-known models like GoogLeNet and SqueezeNet. The promising performance of Bearing-CNN underlines its potential for real-time fault diagnosis in industrial applications, providing a robust and efficient solution for early fault detection and maintenance planning.