Scholarlink

<p>This study develops a machine learning (ML)-based decision-support model to assist in planning loading cycles for large-scale electric tempering furnaces in steel manufacturing. Addressing the limitations of experience-driven approaches, a data-driven framework was introduced using real-world production data from 1162 tempered forgings. The model uses eight input features covering dimensions, material grouping, and process parameters, and employs a multi-output Xtreme Gradient Boosting Regressor (XGBRegressor) algorithm. Cross-validation and test-set evaluations demonstrated strong predictive performance, with R<sup>2</sup> scores of 0.74, 0.88, and 0.87 across three critical batch-level targets: Total Weight, Forging Count, and Total Length. Post-processing analyses, including feature importance rankings and Partial Dependence Plots (PDPs), highlighted the dominant influence of forging shape and width on loading configurations, while predefined process parameters such as temperature and holding duration showed limited impact. The model also identified interpretable thresholds in block dimensions, offering practical guidance for managing new or irregular loading scenarios. This framework enhances planning consistency, reduces reliance on subjective decision-making, and provides a foundation for integrating ML-driven support into industrial furnace operations without replacing technician expertise.</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

Data-driven decision support for furnace loading in steel tempering: a machine learning approach based on industrial operations

  • 期刊论文

  • 发表日期：

    2025年11月24日

查看全文
  • Sajad Mirzaei,
  • Jean-Benoit Morin,
  • Mohammad Jahazi

摘要

This study develops a machine learning (ML)-based decision-support model to assist in planning loading cycles for large-scale electric tempering furnaces in steel manufacturing. Addressing the limitations of experience-driven approaches, a data-driven framework was introduced using real-world production data from 1162 tempered forgings. The model uses eight input features covering dimensions, material grouping, and process parameters, and employs a multi-output Xtreme Gradient Boosting Regressor (XGBRegressor) algorithm. Cross-validation and test-set evaluations demonstrated strong predictive performance, with R2 scores of 0.74, 0.88, and 0.87 across three critical batch-level targets: Total Weight, Forging Count, and Total Length. Post-processing analyses, including feature importance rankings and Partial Dependence Plots (PDPs), highlighted the dominant influence of forging shape and width on loading configurations, while predefined process parameters such as temperature and holding duration showed limited impact. The model also identified interpretable thresholds in block dimensions, offering practical guidance for managing new or irregular loading scenarios. This framework enhances planning consistency, reduces reliance on subjective decision-making, and provides a foundation for integrating ML-driven support into industrial furnace operations without replacing technician expertise.