Scholarlink

<p>Assessment of atomic structure of material, depending on the task, usually require computationally expensive numerical simulations or some form of direct human participation. One possible alternative approach is involving different kinds of artificial neural networks, which may, in certain problems, offer a&#xa0;satisfactory compromise between precision, performance and reproducibility. Sample of differently compressed aluminum crystals was considered. Based on generated pictures of residual distribution of defects in their post-compression relaxed state, two tasks of predicting spall strength and identifying prevailing type of lattice at maximum compression were solved by means of convolutional neural networks (CNNs). It is shown that machine learning based method that described in paper is efficient for solving such tasks, even if the algorithm and form of data representation are probably not optimal and dataset size is relatively small, and have perspective for further improvements via pure quantitative enhancements.</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

Application of convolutional neural networks for analysis of aluminum microstructure

  • 期刊论文

  • 发表日期：

    2026年5月4日

查看全文
  • Arthur I. Fatkhullin,
  • Alexander E. Mayer

摘要

Assessment of atomic structure of material, depending on the task, usually require computationally expensive numerical simulations or some form of direct human participation. One possible alternative approach is involving different kinds of artificial neural networks, which may, in certain problems, offer a satisfactory compromise between precision, performance and reproducibility. Sample of differently compressed aluminum crystals was considered. Based on generated pictures of residual distribution of defects in their post-compression relaxed state, two tasks of predicting spall strength and identifying prevailing type of lattice at maximum compression were solved by means of convolutional neural networks (CNNs). It is shown that machine learning based method that described in paper is efficient for solving such tasks, even if the algorithm and form of data representation are probably not optimal and dataset size is relatively small, and have perspective for further improvements via pure quantitative enhancements.