Scholarlink

<p>The inherent strain method remains one of the most efficient approaches for rapidly predicting welding distortion in large and complex welded structures. However, its application is limited by the difficulty of obtaining inherent strain, which conventionally relies on either time-consuming experimental calibration or thermo-elasto-plastic finite element method (TEP-FEM) simulations. Here, a data-driven framework is proposed to predict inherent strain and enable rapid evaluation of welding-induced deformation. First, TEP-FEM was used to simulate the two-pass welding of Q960E steel, and the model was validated by residual stress and deformation measurements. Based on the validated model, the effects of geometric dimensions (length, width, and thickness) and process parameters (welding speed, preheating temperature, and heat input) on the inherent strain of butt welds were investigated. Subsequently, using inherent strain values obtained under different process conditions via the integral method, a back-propagation neural network model was established to predict the total residual plastic strain in butt joints. The model achieved high accuracy, with average error rates of 2.17% and 6.1% for transverse and longitudinal strains, respectively. Finally, the predicted inherent strains were used to calculate angular distortion under varying degrees of preset inverse deformation. Applying an inverse deformation equal to the original deformation reduced the final angular distortion to 0.46°, corresponding to a reduction of 91.6%. Compared with the deformation results simulated by TEP-FEM, the error remained within 10%, demonstrating that the inherent strain method can accurately capture the relationship between preset inverse deformation and post-weld angular deformation.</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

Data-driven inherent strain prediction and inverse deformation control for butt welds

  • 期刊论文

  • 发表日期：

    2026年5月2日

查看全文
  • Yi Zheng,
  • Mengyuan Lv,
  • Jieshi Chen,
  • Ziyu Tian,
  • Jian Luo

摘要

The inherent strain method remains one of the most efficient approaches for rapidly predicting welding distortion in large and complex welded structures. However, its application is limited by the difficulty of obtaining inherent strain, which conventionally relies on either time-consuming experimental calibration or thermo-elasto-plastic finite element method (TEP-FEM) simulations. Here, a data-driven framework is proposed to predict inherent strain and enable rapid evaluation of welding-induced deformation. First, TEP-FEM was used to simulate the two-pass welding of Q960E steel, and the model was validated by residual stress and deformation measurements. Based on the validated model, the effects of geometric dimensions (length, width, and thickness) and process parameters (welding speed, preheating temperature, and heat input) on the inherent strain of butt welds were investigated. Subsequently, using inherent strain values obtained under different process conditions via the integral method, a back-propagation neural network model was established to predict the total residual plastic strain in butt joints. The model achieved high accuracy, with average error rates of 2.17% and 6.1% for transverse and longitudinal strains, respectively. Finally, the predicted inherent strains were used to calculate angular distortion under varying degrees of preset inverse deformation. Applying an inverse deformation equal to the original deformation reduced the final angular distortion to 0.46°, corresponding to a reduction of 91.6%. Compared with the deformation results simulated by TEP-FEM, the error remained within 10%, demonstrating that the inherent strain method can accurately capture the relationship between preset inverse deformation and post-weld angular deformation.