Scholarlink

<p>Accurate monitoring of tool conditions is critical for optimizing productivity in modern manufacturing. This research presents a novel dual-method approach for tool state identification in milling operations, integrating transfer learning models with vision-based image processing to enhance tool conditions detection and classification. The study uses scalogram images derived from acoustic emission signals and input into several pretrained deep learning models, including a modified ResNet50 enhanced with dropout layers. The modified ResNet50 achieved the highest validation accuracy of 93%, while testing accuracy ranged from 51 to 72% across all models due to variations in cutting parameters. Notably, the modified ResNet50 demonstrated superior performance with a testing accuracy of 75%, showcasing its ability to handle noisy data through improved generalization. Despite these advances, sensor noise and cutting parameter variations significantly impacted model performance, particularly when distinguishing between worn and broken tools, with accuracy dropping below 50% for noisy datasets. To address these limitations, the study integrates a vision system that directly analyzes the tool’s physical condition by comparing individual cutting edges using pixel differences. This dual-method approach enhances the accuracy of tool condition identification, providing more reliable assessments that minimize production interruptions, reduce misclassifications, and optimize tool replacement decisions. This integration of sensor-based and vision-based tool condition monitoring offers a robust solution for modern manufacturing challenges.</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 novel approach of tool condition monitoring in milling operation with transfer learning models and vision system image processing

  • 期刊论文

  • 发表日期：

    2025年6月10日

查看全文
  • Ahmed Abdeltawab,
  • Zhang Xi,
  • Zhang Longjia

摘要

Accurate monitoring of tool conditions is critical for optimizing productivity in modern manufacturing. This research presents a novel dual-method approach for tool state identification in milling operations, integrating transfer learning models with vision-based image processing to enhance tool conditions detection and classification. The study uses scalogram images derived from acoustic emission signals and input into several pretrained deep learning models, including a modified ResNet50 enhanced with dropout layers. The modified ResNet50 achieved the highest validation accuracy of 93%, while testing accuracy ranged from 51 to 72% across all models due to variations in cutting parameters. Notably, the modified ResNet50 demonstrated superior performance with a testing accuracy of 75%, showcasing its ability to handle noisy data through improved generalization. Despite these advances, sensor noise and cutting parameter variations significantly impacted model performance, particularly when distinguishing between worn and broken tools, with accuracy dropping below 50% for noisy datasets. To address these limitations, the study integrates a vision system that directly analyzes the tool’s physical condition by comparing individual cutting edges using pixel differences. This dual-method approach enhances the accuracy of tool condition identification, providing more reliable assessments that minimize production interruptions, reduce misclassifications, and optimize tool replacement decisions. This integration of sensor-based and vision-based tool condition monitoring offers a robust solution for modern manufacturing challenges.