Scholarlink

<p>Replica metallography offers a rapid, nondestructive approach for assessing the microstructural states of stainless steel in manufacturing. However, accurately distinguishing between cast and forged conditions remains challenging due to subtle variations in grain-boundary curvature, dendritic morphologies, and twinning features. To address this challenge, this paper proposes a data-driven intelligent classification framework that integrates hybrid feature fusion with support vector machine (SVM) optimization. By combining traditional texture descriptors with deep convolutional features, the framework captures both global boundary morphologies and fine-scale texture patterns. A support vector machine classifier, optimized by the gray wolf optimization (GWO) algorithm, is employed to enhance the precision of the decision boundary. Experimental results show that the proposed framework achieves high classification accuracy while retaining practical training and inference efficiency. The method also provides an objective tool for nondestructive manufacturing-state verification of 304 stainless steel components. By reducing reliance on subjective metallographic interpretation, it supports field inspection, quality traceability, and reliability assessment of components where cast/forged substitution may affect service performance. These results demonstrate that integrating microstructural knowledge with data-driven learning can improve the objectivity and reliability of practical materials inspection.</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

Replica Metallography-Based Classification of Cast Versus Forged Austenitic Stainless Steel Using Hybrid Deep Features and Optimized Support Vector Machine

  • 期刊论文

  • 发表日期：

    2026年7月21日

查看全文
  • Hengxu Zhao,
  • Yanbing Guo,
  • Jian Zhang,
  • Ning Zhong,
  • Wang Zhang,
  • Wei Liao,
  • Yanhua Lei,
  • Zhikang Shen

摘要

Replica metallography offers a rapid, nondestructive approach for assessing the microstructural states of stainless steel in manufacturing. However, accurately distinguishing between cast and forged conditions remains challenging due to subtle variations in grain-boundary curvature, dendritic morphologies, and twinning features. To address this challenge, this paper proposes a data-driven intelligent classification framework that integrates hybrid feature fusion with support vector machine (SVM) optimization. By combining traditional texture descriptors with deep convolutional features, the framework captures both global boundary morphologies and fine-scale texture patterns. A support vector machine classifier, optimized by the gray wolf optimization (GWO) algorithm, is employed to enhance the precision of the decision boundary. Experimental results show that the proposed framework achieves high classification accuracy while retaining practical training and inference efficiency. The method also provides an objective tool for nondestructive manufacturing-state verification of 304 stainless steel components. By reducing reliance on subjective metallographic interpretation, it supports field inspection, quality traceability, and reliability assessment of components where cast/forged substitution may affect service performance. These results demonstrate that integrating microstructural knowledge with data-driven learning can improve the objectivity and reliability of practical materials inspection.