融合CBAM的迁移Xception模型在车刀缺陷检测中的应用
DOI:
CSTR:
作者:
作者单位:

1.太原理工大学机械工程学院太原030024;2.太原理工大学先进金属复合材料成形技术与 装备教育部工程研究中心太原030024;3.智奇铁路设备有限公司太原030032; 4.太原理工大学金属成形技术与重型装备全国重点实验室太原030024

作者简介:

通讯作者:

中图分类号:

TP274.3

基金项目:

中国博士后科学基金面上项目(2023M742576)资助


Application of transfer Xception model based on CBAM in turning tool blade defect detection
Author:
Affiliation:

1.School of Mechanical Engineering,Taiyuan University of Technology, Taiyuan 030024,China; 2.Engineering Research Center of Advanced Metal Composites Forming Technology and Equipment, Ministry of Education, Taiyuan University of Technology, Taiyuan 030024,China; 3.Zhiqi Railway Equipment Co., Ltd., Taiyuan 030032, China; 4.National Key Laboratory of Metal Forming Technology and Heavy Equipment, Taiyuan University of Technology,Taiyuan 030024,China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    圆形车刀在生产加工过程中容易使刀刃产生不同程度的缺陷,为了判断刀具是否能够满足加工要求,提出了一种基于刀具图像预处理与迁移Xception预训练模型的圆形机夹式车刀刀刃缺陷检测方法。首先设计刀具图像预处理流程,通过图像二值化转化实现像素灰度分层,结合连通域面积算法设定合理阈值,精准剔除刀具表面非功能性花纹轮廓,仅保留刀刃核心区域图像,从源头消除背景干扰;随后创新优化特征提取网络,以迁移Xception预训练模型为基础,引入卷积注意力模块(convolutional block attention module, CBAM),通过通道维度权重分配与空间维度特征聚焦的双重机制,强化模型对微小缺陷的特征捕捉能力,弥补传统迁移模型对特定缺陷特征关注度不足的短板。研究结果显示,本研究模型在包含不同损伤程度的圆形机夹式车刀数据集上,检测准确率达96.55%,相比未经过预处理及未添加CBAM模块的基准模型,精度提升显著。该方法通过预处理与注意力机制的协同优化,有效解决了花纹干扰与缺陷特征捕捉不足的核心问题,具备较高可靠性与实用性,可为工业生产中的刀具缺陷检测提供技术支持。

    Abstract:

    Circular turning tools are prone to produce different degrees of defects on the cutting-edge during production and processing. In order to judge whether the tool can meet the processing requirements, this paper proposes a circular clamp-on turning tool blade defect detection method based on tool image preprocessing and migration Xception pre-training model. Firstly, the tool image preprocessing process is designed, and the pixel grey release layering is realized through the binary transformation of the image. Combined with the connected domain area algorithm, a reasonable threshold is set to accurately remove the non-functional pattern contour of the tool surface, and only the image of the core area of the blade is retained to eliminate background interference from the source. Then, the feature extraction network is innovatively optimized, based on the migration Xception pre-trained model, and the CBAM attention module is introduced. Through the dual mechanism of channel dimension weight allocation and spatial dimension feature focusing, the feature capture ability of the model for minor defects is strengthened, and the shortcoming of the traditional migration model’s lack of attention to specific defect features is compensated. The research results show that the model proposed in this paper has a detection accuracy of 96.55% on the dataset of circular machine-clip turning tools with different degrees of damage, which is significantly improved compared with the benchmark model without pretreatment and no CBAM module. The method effectively solves the core problems of pattern interference and insufficient defect feature capture through the cooperative optimization of preprocessing and attention mechanism. It has high reliability and practicability, and can provide technical support for tool defect detection in industrial production.

    参考文献
    相似文献
    引证文献
引用本文

王钰杰,王振华,霍锋锋,张顺,刘元铭.融合CBAM的迁移Xception模型在车刀缺陷检测中的应用[J].电子测量与仪器学报,2026,40(9):285-300

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-11-13
  • 出版日期:
文章二维码
×
《电子测量与仪器学报》
关于防范虚假编辑部邮件的郑重公告