Scholarlink

<p>Conventional diameter-based measurements cannot capture the complex crack morphologies governing damage progression in thermo-mechanically loaded steel bores. This study presents an image-based workflow that converts in situ microscopic images into quantitative crack density maps. Images were acquired from a retired large-caliber gun barrel forcing cone at 20 × magnification (8 μm/pixel) and processed using an improved U-Net model incorporating MultiScale blocks and spatial attention mechanisms. The model achieved 93.89% pixel accuracy, a recall of 93.71%, and an IoU of 72.43% on an independent test set. Crack skeletonization enabled length per area density computation, and sliding window analysis (300 × 300 pixels, 2.4 × 2.4 mm) established a three-tier erosion classification: non-pitted (&gt;1.8 mm/mm<sup>2</sup>), transition (1.2-1.8 mm/mm<sup>2</sup>), and pitted (&lt;1.2 mm/mm<sup>2</sup>). Notably, pitted regions exhibited significantly lower crack density than non-pitted regions (Mann–Whitney U test, <i>p</i> &lt; 0.001), consistent with the hypothesis that micro-spallation removes secondary cracks. Geometric analysis of crack-bounded "island blocks" further revealed that concave polygons-representing incompletely separated fragments-exhibited disproportionately large area contributions, serving as transitional indicators in damage evolution. By explicitly documenting acquisition parameters, window dimensions, and classification thresholds, this workflow provides reproducible morphology metrics for condition assessment and life prediction modeling.</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

Image-Based Crack Density Quantification and Damage Classification in Gun Barrel Forcing Cones

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Mengran Zhu,
  • Jinghua Cao,
  • Yang Yang,
  • Ying Liu,
  • Zheng Li,
  • Yao Jiang,
  • Jiawei Fu,
  • Jingtao Wang,
  • Linfang Qian

摘要

Conventional diameter-based measurements cannot capture the complex crack morphologies governing damage progression in thermo-mechanically loaded steel bores. This study presents an image-based workflow that converts in situ microscopic images into quantitative crack density maps. Images were acquired from a retired large-caliber gun barrel forcing cone at 20 × magnification (8 μm/pixel) and processed using an improved U-Net model incorporating MultiScale blocks and spatial attention mechanisms. The model achieved 93.89% pixel accuracy, a recall of 93.71%, and an IoU of 72.43% on an independent test set. Crack skeletonization enabled length per area density computation, and sliding window analysis (300 × 300 pixels, 2.4 × 2.4 mm) established a three-tier erosion classification: non-pitted (>1.8 mm/mm2), transition (1.2-1.8 mm/mm2), and pitted (<1.2 mm/mm2). Notably, pitted regions exhibited significantly lower crack density than non-pitted regions (Mann–Whitney U test, p < 0.001), consistent with the hypothesis that micro-spallation removes secondary cracks. Geometric analysis of crack-bounded "island blocks" further revealed that concave polygons-representing incompletely separated fragments-exhibited disproportionately large area contributions, serving as transitional indicators in damage evolution. By explicitly documenting acquisition parameters, window dimensions, and classification thresholds, this workflow provides reproducible morphology metrics for condition assessment and life prediction modeling.