Scholarlink

<p>A machine learning-assisted topology optimization framework for designing 2D structures is developed. A micropolar elasticity-based finite element model is formulated and integrated into this framework to compute the material compliance, which accounts for microstructure effects. The topology optimization (TO) procedure is based on the modified SIMP method and begins with generating three intermediate material layouts with distinct density profiles. These layouts serve as inputs for a machine learning (ML) model trained to predict the final optimal layout for given material properties and prescribed loading and boundary conditions. Three ML models—feedforward neural networks (FFNN), convolutional neural networks (CNN), and generative adversarial networks (GAN)—are trained and implemented to execute the ML-assisted TO framework. Numerical results reveal that the microstructure effects, as represented by the two micropolar material constants, can significantly influence the optimal topology and structural stiffness. Compared to the traditional TO approach, the newly developed ML-assisted TO framework effectively reduces computation time and lowers computational energy consumption. The new ML-assisted TO framework provides an accurate, efficient, and computationally viable tool for structural and material designs.</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

Machine learning-assisted topology optimization framework for designing 2D structures based on micropolar elasticity

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月11日

查看全文
  • H. W. Zhou,
  • M. Shaat,
  • X.-L. Gao

摘要

A machine learning-assisted topology optimization framework for designing 2D structures is developed. A micropolar elasticity-based finite element model is formulated and integrated into this framework to compute the material compliance, which accounts for microstructure effects. The topology optimization (TO) procedure is based on the modified SIMP method and begins with generating three intermediate material layouts with distinct density profiles. These layouts serve as inputs for a machine learning (ML) model trained to predict the final optimal layout for given material properties and prescribed loading and boundary conditions. Three ML models—feedforward neural networks (FFNN), convolutional neural networks (CNN), and generative adversarial networks (GAN)—are trained and implemented to execute the ML-assisted TO framework. Numerical results reveal that the microstructure effects, as represented by the two micropolar material constants, can significantly influence the optimal topology and structural stiffness. Compared to the traditional TO approach, the newly developed ML-assisted TO framework effectively reduces computation time and lowers computational energy consumption. The new ML-assisted TO framework provides an accurate, efficient, and computationally viable tool for structural and material designs.