Scholarlink

<p>Automated build-up welding processes often produce complex weld bead geometries that significantly influence weld quality, including cooling rate, residual stress, and distortion. In this study, a back-propagation (BP) neural network model is developed to predict key weld bead geometry parameters (such as bead width and height) based on welding process inputs in an automated build-up welding scenario. A series of welding experiments was conducted to provide training data covering a broad range of parameters. The trained BP neural network achieved high prediction accuracy for bead geometry, with a correlation coefficient of R = 0.989 and an average absolute relative error (AARE) of 2.65% for the training set, and R = 0.981 and AARE of 4.54% for the testing set<b>,</b> enabling reliable modeling of the nonlinear relationships between process variables and bead shape. Moreover, an optimization procedure was applied using the validated model to identify the optimal set of welding parameters that yield the desired bead geometry with improved quality. The results show that the BP neural network can accurately predict bead geometry with minimal error, and the model-driven optimization of welding parameters leads to improved weld bead formation and quality consis<b>tency.</b> This approach demonstrates not only high predictive performance but also practical applicability for automated process control. It provides a foundation for real-time weld quality optimization in industrial applications.</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

Prediction and Optimization of Weld Bead Geometry in Automated Build-Up Welding Using Back-Propagation Neural Networks

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • HaiTao Wang,
  • Guozheng Quan,
  • Zhangfan Tan,
  • Fanxing Meng,
  • Wenjing Rang

摘要

Automated build-up welding processes often produce complex weld bead geometries that significantly influence weld quality, including cooling rate, residual stress, and distortion. In this study, a back-propagation (BP) neural network model is developed to predict key weld bead geometry parameters (such as bead width and height) based on welding process inputs in an automated build-up welding scenario. A series of welding experiments was conducted to provide training data covering a broad range of parameters. The trained BP neural network achieved high prediction accuracy for bead geometry, with a correlation coefficient of R = 0.989 and an average absolute relative error (AARE) of 2.65% for the training set, and R = 0.981 and AARE of 4.54% for the testing set, enabling reliable modeling of the nonlinear relationships between process variables and bead shape. Moreover, an optimization procedure was applied using the validated model to identify the optimal set of welding parameters that yield the desired bead geometry with improved quality. The results show that the BP neural network can accurately predict bead geometry with minimal error, and the model-driven optimization of welding parameters leads to improved weld bead formation and quality consistency. This approach demonstrates not only high predictive performance but also practical applicability for automated process control. It provides a foundation for real-time weld quality optimization in industrial applications.