Scholarlink

<p>Gradient porous metals, widely used in aerospace and automotive industries, must meet multiple performance requirements such as lightweight, energy absorption, and heat transfer, making design and analysis challenging. Combining fully connected neural network (FCNN) and genetic algorithm (GA), this study proposes the FCNN-GA framework for the multi-objective design of gradient porous metals. The method starts with parametric modeling, where the structure is defined by three key parameters: center pore radius, radius gradient, and pore wall thickness. This is followed by dataset generation through finite element simulations. Then, an FCNN model captures the nonlinear relationship between these parameters and performance. Finally, by integrating a GA, the FCNN-GA framework enables dual- and triple-objective optimization. Results show that the FCNN model achieves a goodness of fit greater than 0.97, providing a reliable surrogate for finite element analysis. The framework successfully balances conflicting performance requirements, offering significant potential for aerospace and energy applications.</p> Graphical Abstract <p></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

Multi-objective optimization of lightweight, energy absorption, and heat transfer in gradient porous metals via deep learning and genetic algorithm

  • 期刊论文

  • 发表日期：

    2025年11月5日

查看全文
  • Lei Wang,
  • Minghai Tang,
  • Donghui Yang,
  • Huayuan Tang

摘要

Gradient porous metals, widely used in aerospace and automotive industries, must meet multiple performance requirements such as lightweight, energy absorption, and heat transfer, making design and analysis challenging. Combining fully connected neural network (FCNN) and genetic algorithm (GA), this study proposes the FCNN-GA framework for the multi-objective design of gradient porous metals. The method starts with parametric modeling, where the structure is defined by three key parameters: center pore radius, radius gradient, and pore wall thickness. This is followed by dataset generation through finite element simulations. Then, an FCNN model captures the nonlinear relationship between these parameters and performance. Finally, by integrating a GA, the FCNN-GA framework enables dual- and triple-objective optimization. Results show that the FCNN model achieves a goodness of fit greater than 0.97, providing a reliable surrogate for finite element analysis. The framework successfully balances conflicting performance requirements, offering significant potential for aerospace and energy applications.

Graphical Abstract