Scholarlink

<p>This study proposed a system for diagnosing the wear of drive bearings in gantry robots, which is essential for facilitating precise object transportation in industrial fields. The rotational speed of the bearing and the driving direction acceleration of the bearing housing were measured to diagnose the wear of the bearing. The measured data was converted into RPM-frequency spectrum images and used as training data for a CNN. The training data consisted of approximately 4500 images and included six different wear classes. To select the most suitable network architecture, the learning results of the pre-trained networks were compared. The selected ResNet-18 classified bearing wear classes with high accuracy. Further, to predict the wear conditions between a given six classes, we used CNN softMax output probabilities to define the expected value of the wear condition. As a result, the wear state of the bearing was predicted within an error of approximately +−5 %.</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

Wear diagnosis of gantry robot drive bearings using the expected value of convolutional neural network softmax output

  • 期刊论文

  • 发表日期：

    2026年7月11日

查看全文
  • Jeonghyeon Park,
  • Kwangsuck Boo

摘要

This study proposed a system for diagnosing the wear of drive bearings in gantry robots, which is essential for facilitating precise object transportation in industrial fields. The rotational speed of the bearing and the driving direction acceleration of the bearing housing were measured to diagnose the wear of the bearing. The measured data was converted into RPM-frequency spectrum images and used as training data for a CNN. The training data consisted of approximately 4500 images and included six different wear classes. To select the most suitable network architecture, the learning results of the pre-trained networks were compared. The selected ResNet-18 classified bearing wear classes with high accuracy. Further, to predict the wear conditions between a given six classes, we used CNN softMax output probabilities to define the expected value of the wear condition. As a result, the wear state of the bearing was predicted within an error of approximately +−5 %.