Scholarlink

<p>To address the limitations of traditional plastic constitutive models in accurately describing the complex elastoplastic mechanical behavior of dual-phase (DP) steel, this paper proposes a data-driven modeling approach integrating Representative Volume Elements (RVE) and Back-Propagation Neural Networks (BPNN). First, a 3D micro-FE RVE model based on real microstructure is constructed to generate high-fidelity macroscopic mechanical response data under multi-axial loading paths. Subsequently, a BPNN-based constitutive model incorporating physical consistency constraints (e.g., positive definiteness of plastic work) is developed. The elastoplastic Jacobian matrix is derived through backpropagation of neural network gradients and embedded into finite element software via a user-defined material subroutine (UMAT). Validation results demonstrate that the proposed model controls the average deviation of stress prediction within 20&#xa0;MPa, with an equivalent stress fitting accuracy exceeding 92%. In three-point bending simulations, the relative error of springback prediction is reduced from over 10% (traditional models) to less than 3.1%, effectively capturing the residual stress distribution induced by the deformation incompatibility between ferrite and bainite phases. This study breaks through the bottlenecks of traditional models, such as cumbersome parameter calibration, providing a precise numerical tool for the forming process of high-performance materials like X80 pipeline steel. In industrial applications, this model enables digital optimization of forming processes. By accurately predicting and pre-compensating die profiles, it addresses the challenges of springback control in high-strength steel, significantly reducing physical trial-and-error costs while enhancing forming precision and manufacturing efficiency.</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

Research on the Neural Network-Based Elastoplastic Constitutive Model of Dual-Phase Steel Using RVE and Its Application in Springback Calculation

  • 期刊论文

  • 发表日期：

    2026年4月22日

查看全文
  • Hao Wang,
  • Kai Sun

摘要

To address the limitations of traditional plastic constitutive models in accurately describing the complex elastoplastic mechanical behavior of dual-phase (DP) steel, this paper proposes a data-driven modeling approach integrating Representative Volume Elements (RVE) and Back-Propagation Neural Networks (BPNN). First, a 3D micro-FE RVE model based on real microstructure is constructed to generate high-fidelity macroscopic mechanical response data under multi-axial loading paths. Subsequently, a BPNN-based constitutive model incorporating physical consistency constraints (e.g., positive definiteness of plastic work) is developed. The elastoplastic Jacobian matrix is derived through backpropagation of neural network gradients and embedded into finite element software via a user-defined material subroutine (UMAT). Validation results demonstrate that the proposed model controls the average deviation of stress prediction within 20 MPa, with an equivalent stress fitting accuracy exceeding 92%. In three-point bending simulations, the relative error of springback prediction is reduced from over 10% (traditional models) to less than 3.1%, effectively capturing the residual stress distribution induced by the deformation incompatibility between ferrite and bainite phases. This study breaks through the bottlenecks of traditional models, such as cumbersome parameter calibration, providing a precise numerical tool for the forming process of high-performance materials like X80 pipeline steel. In industrial applications, this model enables digital optimization of forming processes. By accurately predicting and pre-compensating die profiles, it addresses the challenges of springback control in high-strength steel, significantly reducing physical trial-and-error costs while enhancing forming precision and manufacturing efficiency.