Scholarlink

<p>The integration of machine learning in welding processes brought a transformative era, optimizing precision in parameter prediction. Regression techniques stand as the cornerstone of machine learning, offering a crucial toolset for modeling and understanding complex relationships, enabling the extraction of valuable insights from diverse datasets. In this matter, predicting parameters with small datasets presents a formidable challenge, requiring the study of innovative techniques. This study investigates the efficacy of data augmentation through Synthetic Data Vault (SDV) and Kriging in enhancing the prediction of GMAW parameters. The original dataset, composed of experimental welding data, includes inputs such as bead thickness, width, wire speed, stickout, and bevel type, and outputs including bead geometry characteristics. The dataset was augmented by 50%, 100%, and 150%, artificially. Before the regression analysis, a distance correlation method was used to ensure the use of all input–output parameters and ensure the quality of augmented data. Employing three regression models—linear, polynomial, and ridge multiple regression—reveals that ridge’s multiple regression consistently outperforms others, while linear regression demonstrates intermediary capabilities and polynomial regression exhibits the worst prediction values. The SDV provided the best results, with an improvement of 20.2% in the prediction error. This result was achieved with ridge multiple regression and a data augmentation of 150%.</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

Regression models in prediction of welding process parameters with artificial data augmentation

  • 期刊论文

  • 发表日期：

    2025年9月2日

查看全文
  • Vinícius Resende Rocha,
  • Fran Sergio Lobato,
  • Carlos Roberto Ribeiro,
  • Sebastião da SimõesCunha Jr.,
  • Louriel Oliveira Vilarinho,
  • Luiz Eduardo dos Santos Paes

摘要

The integration of machine learning in welding processes brought a transformative era, optimizing precision in parameter prediction. Regression techniques stand as the cornerstone of machine learning, offering a crucial toolset for modeling and understanding complex relationships, enabling the extraction of valuable insights from diverse datasets. In this matter, predicting parameters with small datasets presents a formidable challenge, requiring the study of innovative techniques. This study investigates the efficacy of data augmentation through Synthetic Data Vault (SDV) and Kriging in enhancing the prediction of GMAW parameters. The original dataset, composed of experimental welding data, includes inputs such as bead thickness, width, wire speed, stickout, and bevel type, and outputs including bead geometry characteristics. The dataset was augmented by 50%, 100%, and 150%, artificially. Before the regression analysis, a distance correlation method was used to ensure the use of all input–output parameters and ensure the quality of augmented data. Employing three regression models—linear, polynomial, and ridge multiple regression—reveals that ridge’s multiple regression consistently outperforms others, while linear regression demonstrates intermediary capabilities and polynomial regression exhibits the worst prediction values. The SDV provided the best results, with an improvement of 20.2% in the prediction error. This result was achieved with ridge multiple regression and a data augmentation of 150%.