Scholarlink

<p>The ultimate tensile strength (UTS) of high-strength steel (HSS) welded joints is a key indicator in welding procedure qualification (WPQ); however, traditional WPQ tests are inefficient, while existing predictive models suffer from limited generalization capability and interpretability. This study develops an efficient, accurate, and interpretable prediction model for UTS of HSS welded joints, utilizing various machine learning algorithms, optimization algorithms, and SHAP additive explanations, and further encapsulates them into a highly operable modeling system. Four machine learning algorithms—XGBoost, BPNN, SVR, and RF—were combined with K-fold cross-validation and Bayesian optimization to construct UTS prediction models using base metal composition, filler metal composition, and welding process parameters as input features. BPNN yielded the most accurate predictions on the test set and was further enhanced using particle swarm optimization (PSO), and SHAP analysis was carried out to interpret the PSO-BPNN model. HSS welding experiments were conducted to build an independent validation dataset. The comparison results showed that the PSO-BPNN model achieved R<sup>2</sup> = 0.9669, RMSE = 30.09&#xa0;MPa, and MAPE = 3.27% on the test set, with prediction errors below 6% on the validation set. The base metal composition affects UTS the most. Finally, the complex machine learning process is encapsulated into visual interactive interfaces based on front end, back end, and MySQL. This study aims to realize the prediction of UTS of HSS welded joints through machine learning, which will provide an efficient and accurate solution for WPQ, while promoting the digitalization and sustainable development of welding processes.</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

Ultimate Tensile Strength Prediction and Modeling System for High-Strength Steel Welded Joints Based on Interpretable Machine Learning

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • Xiaoya Yang,
  • Yuanyang Gao,
  • Shaogang Wang,
  • Yanhong Wei,
  • Liyu Fan,
  • Ping Ping,
  • Shengxuan Hu

摘要

The ultimate tensile strength (UTS) of high-strength steel (HSS) welded joints is a key indicator in welding procedure qualification (WPQ); however, traditional WPQ tests are inefficient, while existing predictive models suffer from limited generalization capability and interpretability. This study develops an efficient, accurate, and interpretable prediction model for UTS of HSS welded joints, utilizing various machine learning algorithms, optimization algorithms, and SHAP additive explanations, and further encapsulates them into a highly operable modeling system. Four machine learning algorithms—XGBoost, BPNN, SVR, and RF—were combined with K-fold cross-validation and Bayesian optimization to construct UTS prediction models using base metal composition, filler metal composition, and welding process parameters as input features. BPNN yielded the most accurate predictions on the test set and was further enhanced using particle swarm optimization (PSO), and SHAP analysis was carried out to interpret the PSO-BPNN model. HSS welding experiments were conducted to build an independent validation dataset. The comparison results showed that the PSO-BPNN model achieved R2 = 0.9669, RMSE = 30.09 MPa, and MAPE = 3.27% on the test set, with prediction errors below 6% on the validation set. The base metal composition affects UTS the most. Finally, the complex machine learning process is encapsulated into visual interactive interfaces based on front end, back end, and MySQL. This study aims to realize the prediction of UTS of HSS welded joints through machine learning, which will provide an efficient and accurate solution for WPQ, while promoting the digitalization and sustainable development of welding processes.