Scholarlink

<p>Accurate vibration measurement is vital for tool condition monitoring in CNC lathe operations. Traditional accelerometers are expensive, noise-sensitive, and limited in displacement measurement. Non-contact methods like digital image correlation (DIC) offer high precision and full-field capabilities without the environmental constraints of laser systems; however, conventional DIC struggles with high-frequency and complex deformations due to its limited ability to capture high-order gradients. This paper introduces AT-DICNet, a novel speckle pattern-based framework that integrates deep learning with DIC to achieve micron-level accuracy in non-contact tool vibration deformation measurement. AT-DICNet employs convolutional neural networks enhanced with a pyramid pooling module and attention mechanisms to focus on critical deformation features. A specialized training dataset simulating real tool conditions was developed to evaluate the network’s efficacy. In synthetic tests, AT-DICNet achieved a mean absolute error of 0.0221 and a root mean square error of 0.0281. In real-world CNC lathe applications, it demonstrated only a 1.18<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10845_2025_2723_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> </InlineEquation> error in micron-level measurements. The framework offers real-time processing capabilities, handling images at 0.0093&#xa0;s per image during actual tool vibration deformation measurements. Evaluations indicate that AT-DICNet outperforms traditional DIC methods and existing networks in measuring complex tool vibration deformations, providing an advanced approach for real-time tool condition monitoring.</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

At-dicnet: a novel framework based speckle pattern for non-contact micron-level tool vibration deformation precision measurement

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Jing Lei,
  • Shuang Mei,
  • Quan Zhao,
  • WangDe Qiu,
  • LeiBin Wan,
  • GuoJun Wen

摘要

Accurate vibration measurement is vital for tool condition monitoring in CNC lathe operations. Traditional accelerometers are expensive, noise-sensitive, and limited in displacement measurement. Non-contact methods like digital image correlation (DIC) offer high precision and full-field capabilities without the environmental constraints of laser systems; however, conventional DIC struggles with high-frequency and complex deformations due to its limited ability to capture high-order gradients. This paper introduces AT-DICNet, a novel speckle pattern-based framework that integrates deep learning with DIC to achieve micron-level accuracy in non-contact tool vibration deformation measurement. AT-DICNet employs convolutional neural networks enhanced with a pyramid pooling module and attention mechanisms to focus on critical deformation features. A specialized training dataset simulating real tool conditions was developed to evaluate the network’s efficacy. In synthetic tests, AT-DICNet achieved a mean absolute error of 0.0221 and a root mean square error of 0.0281. In real-world CNC lathe applications, it demonstrated only a 1.18 \(\%\) error in micron-level measurements. The framework offers real-time processing capabilities, handling images at 0.0093 s per image during actual tool vibration deformation measurements. Evaluations indicate that AT-DICNet outperforms traditional DIC methods and existing networks in measuring complex tool vibration deformations, providing an advanced approach for real-time tool condition monitoring.