Scholarlink

<p>The resolution of mechanical contact between two bodies using numerical methods, such as the finite element method, requires the prior assignment of contacting roles–slave or master–to each body, which is performed before solving the contact problem. When a body is designated as the slave, non-penetration constraints are imposed to prevent penetration into the master. To ensure accurate results, each body must be assigned the correct role, a task that is complex and often impractical to perform manually. To address this, artificial neural networks (ANNs) are proposed in this paper for three 2D mechanical contact problems to automatically assign the appropriate roles, thereby improving accuracy while reducing computational cost compared to the symmetric formulation, where no role assignment is required. A data generation strategy essential for training the ANNs is also presented. Beyond contact mechanics, the proposed approach can be extended to other physics problems involving interface conditions on non-conforming meshes, such as magnetostatics and thermal conduction.</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

An artificial neural network for predicting the roles of contacting bodies in computational contact mechanics

  • 期刊论文

  • 发表日期：

    2025年9月16日

查看全文
  • Houssam Houssein

摘要

The resolution of mechanical contact between two bodies using numerical methods, such as the finite element method, requires the prior assignment of contacting roles–slave or master–to each body, which is performed before solving the contact problem. When a body is designated as the slave, non-penetration constraints are imposed to prevent penetration into the master. To ensure accurate results, each body must be assigned the correct role, a task that is complex and often impractical to perform manually. To address this, artificial neural networks (ANNs) are proposed in this paper for three 2D mechanical contact problems to automatically assign the appropriate roles, thereby improving accuracy while reducing computational cost compared to the symmetric formulation, where no role assignment is required. A data generation strategy essential for training the ANNs is also presented. Beyond contact mechanics, the proposed approach can be extended to other physics problems involving interface conditions on non-conforming meshes, such as magnetostatics and thermal conduction.