Scholarlink

<p>Industrial Computed Tomography (CT) is pivotal for detecting internal defects in materials. However, the acquisition of high-quality CT images is costly and labor-intensive, particularly when it comes to obtaining large datasets with accurate annotations. This limitation has hindered the application of Deep Learning (DL) image segmentation techniques in this field. This study presents an innovative simulation-based approach to overcome these limitations, effectively generating synthetic datasets for the training of DL models. By leveraging CAD models and simulating the CT scanning process with controlled defect introduction, we create a rich, labeled dataset that enables the development of accurate and robust DL algorithms for defect detection. This approach not only mitigates the limitations in data acquisition but also enhances the efficiency and accuracy of DL models in identifying defects in industrial CT scans. The proposed method represents a significant step forward in the field of quality assurance in manufacturing, offering a practical solution to the challenges of cost, labor, and data annotation in the application of DL to industrial CT imaging.</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

Simulation-Based Data Augmentation for Deep Learning in Industrial Computed Tomography Defect Detection

  • 期刊论文

  • 发表日期：

    2025年5月21日

查看全文
  • Haibin Lan,
  • Zhiyu Gao,
  • Wei Guan,
  • Changsheng Zhang,
  • Qianni Wang,
  • Chongwen Jiang,
  • Gang Zhao,
  • Jian Fu

摘要

Industrial Computed Tomography (CT) is pivotal for detecting internal defects in materials. However, the acquisition of high-quality CT images is costly and labor-intensive, particularly when it comes to obtaining large datasets with accurate annotations. This limitation has hindered the application of Deep Learning (DL) image segmentation techniques in this field. This study presents an innovative simulation-based approach to overcome these limitations, effectively generating synthetic datasets for the training of DL models. By leveraging CAD models and simulating the CT scanning process with controlled defect introduction, we create a rich, labeled dataset that enables the development of accurate and robust DL algorithms for defect detection. This approach not only mitigates the limitations in data acquisition but also enhances the efficiency and accuracy of DL models in identifying defects in industrial CT scans. The proposed method represents a significant step forward in the field of quality assurance in manufacturing, offering a practical solution to the challenges of cost, labor, and data annotation in the application of DL to industrial CT imaging.