机电工程技术2025，Vol.54Issue(21)：15-20,6.DOI:10.3969/j.issn.1009-9492.2025.21.003
基于BP神经网络的全焊接球阀焊接质量预测
Welding Quality Prediction of Fully Welded Ball Valve Based on BP Neural Network
摘要
Abstract
Fully welded ball valves,distinguished by their superior integrated design,are extensively employed in pipeline transportation systems under high-pressure and other extreme operational conditions,where the welding quality of valve bodies constitutes a pivotal determinant of overall manufacturing performance.Addressing the challenge of predicting welding quality in fully welded ball valves,comprehensive evaluation metrics encompassing post-weld residual stress,deformation,weld bead width,and reinforcement height is established.An L25 orthogonal experimental design is employed to conduct high-precision numerical simulations of the submerged arc welding process using Abaqus,systematically investigating the effects of welding current,voltage,and speed on welding quality.A BP neural network-based predictive model for welding quality is subsequently developed.To enhance the fidelity of numerical simulations,the element birth and death technique is integrated to ensure precise thermal source modeling across diverse welding parameters.Recognizing that post-weld residual stress and deformation primarily stem from non-uniform distribution and abrupt variations in the temperature field during welding,temperature distribution gradients are incorporated into the predictive framework.Comparative experiments with control groups demonstrates significant improvements in prediction accuracy:deformation prediction accuracy increases by 8.11%,while residual stress prediction accuracy improves by 11.48%.关键词
全焊接球阀/焊接质量预测/生死单元技术/有限元分析/BP神经网络Key words
fully welded ball valve/welding quality prediction/element birth and death technique/finite element analysis/BP neural network分类
金属材料引用本文复制引用
何纳川,陈俊,黄志慧,陈刚,徐雷..基于BP神经网络的全焊接球阀焊接质量预测[J].机电工程技术,2025,54(21):15-20,6.基金项目
四川大学自贡校地科技合作项目(2023CDZG-10) （2023CDZG-10）
