Scholarlink

<p>The precision of CNC machining and tool life are significantly impacted by turning tool wear, and it is crucial to accurately identify wear regions under challenging operating conditions in order to ensure machining quality and lower manufacturing costs. Therefore, a region detection framework combining transfer learning-based GoogleNet for lathe tool wear state detection and improved DeepLab V3 semantic segmentation network is constructed. First, the feature extraction capability of GoogleNet was optimized based on migration learning to achieve the classification of lathe tool wear state. Second, an improved densely-interconnected atrous pooling module was introduced to construct the multi-scale enhanced DeepLab V3 Network. The recognition accuracy of the improved GoogleNet under the optimal parameter configuration reached 90.3%, the misclassification concentration under various noise conditions was kept below 18.5%, and the computational efficiency was less than 3.3 ms/sample. The proposed method improves wear detection accuracy. Specifically, the improved DeepLab V3 achieved an 8.91% increase in mean Intersection over Union and a 29.76% reduction in boundary error. After applying error compensation, the cutting and feed errors were reduced to 0.062&#xa0;mm and 0.049&#xa0;mm, respectively, with a contour error standard deviation of just 0.010. The research results show that this method provides a feasible solution for the wear detection of turning tools under complex working conditions, which is especially suitable for precision manufacturing industries such as CNC machining, automotive parts manufacturing, aerospace, and so on. It has certain significance in ensuring machining quality and reducing operation and maintenance costs.</p> Graphical abstract <p></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

Turning tool wear detection and error analysis based on DeepLab V3 semantic segmentation networks

  • 期刊论文

  • 发表日期：

    2025年7月29日

查看全文
  • Haifang Yin,
  • Zhenhua Wang

摘要

The precision of CNC machining and tool life are significantly impacted by turning tool wear, and it is crucial to accurately identify wear regions under challenging operating conditions in order to ensure machining quality and lower manufacturing costs. Therefore, a region detection framework combining transfer learning-based GoogleNet for lathe tool wear state detection and improved DeepLab V3 semantic segmentation network is constructed. First, the feature extraction capability of GoogleNet was optimized based on migration learning to achieve the classification of lathe tool wear state. Second, an improved densely-interconnected atrous pooling module was introduced to construct the multi-scale enhanced DeepLab V3 Network. The recognition accuracy of the improved GoogleNet under the optimal parameter configuration reached 90.3%, the misclassification concentration under various noise conditions was kept below 18.5%, and the computational efficiency was less than 3.3 ms/sample. The proposed method improves wear detection accuracy. Specifically, the improved DeepLab V3 achieved an 8.91% increase in mean Intersection over Union and a 29.76% reduction in boundary error. After applying error compensation, the cutting and feed errors were reduced to 0.062 mm and 0.049 mm, respectively, with a contour error standard deviation of just 0.010. The research results show that this method provides a feasible solution for the wear detection of turning tools under complex working conditions, which is especially suitable for precision manufacturing industries such as CNC machining, automotive parts manufacturing, aerospace, and so on. It has certain significance in ensuring machining quality and reducing operation and maintenance costs.

Graphical abstract