Scholarlink

<p>Accurate prediction of weld pool morphology remains challenging in laser-arc hybrid welding (LAHW) due to complex multi-parameter coupling effects. The performance of three machine learning (ML) architectures—artificial neural network (ANN), recurrent neural network (RNN), and generative adversarial network (GAN)—was systematically evaluated for melt pool morphology prediction. The results indicated that the ANN model exhibited good prediction stability and generalization capability compared to RNN and GAN. This performance can be attributed to its ability to accurately capture the complex nonlinear relationship between welding parameters and molten pool characteristics. The ANN model achieved an optimal coefficient of determination (<i>R</i><sup>2</sup> = 0.854) in weld pool prediction, with a Willmott Index (WI) of 0.9661. The root mean square error (RMSE) and mean absolute error (MAE) values were 0.176 and 0.153&#xa0;mm, respectively. Two independent controlled experiments confirmed that the deviations between predicted and experimentally measured weld pool dimensions remained within 0.2&#xa0;mm. The study provides a reliable ML-driven framework for LAHW optimization, enhancing process reliability and weld quality assurance.</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-Based Prediction of Weld Pool Features in Laser-Arc Hybrid Welding of Ultra-High-Strength Wear-Resistant Steels

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • Bin Zhang,
  • Zhe Gao,
  • Sitong Li,
  • Haijun Pan,
  • Zhenguang Liu,
  • Zigang Chen,
  • Hongyan Liu,
  • Xinjun Shen,
  • Zhenxing Li,
  • Xiaonan Wang

摘要

Accurate prediction of weld pool morphology remains challenging in laser-arc hybrid welding (LAHW) due to complex multi-parameter coupling effects. The performance of three machine learning (ML) architectures—artificial neural network (ANN), recurrent neural network (RNN), and generative adversarial network (GAN)—was systematically evaluated for melt pool morphology prediction. The results indicated that the ANN model exhibited good prediction stability and generalization capability compared to RNN and GAN. This performance can be attributed to its ability to accurately capture the complex nonlinear relationship between welding parameters and molten pool characteristics. The ANN model achieved an optimal coefficient of determination (R2 = 0.854) in weld pool prediction, with a Willmott Index (WI) of 0.9661. The root mean square error (RMSE) and mean absolute error (MAE) values were 0.176 and 0.153 mm, respectively. Two independent controlled experiments confirmed that the deviations between predicted and experimentally measured weld pool dimensions remained within 0.2 mm. The study provides a reliable ML-driven framework for LAHW optimization, enhancing process reliability and weld quality assurance.