Scholarlink

<p>Effective modification of alumina (Al<sub>2</sub>O<sub>3</sub>) non-metallic inclusion (NMI) is essential for improving medium-carbon steel quality. However, accurately predicting Al<sub>2</sub>O<sub>3</sub> modification efficiency under small-sample industrial conditions remains challenging due to limited labeled data and low information density. This study proposes a hybrid framework that integrates mechanism-driven feature engineering with a KNN-guided Gaussian perturbation (KNN-GNP) data synthesis method to predict the liquid NMI fraction in medium-carbon steel. Based on 25 industrial heats, 6 engineered features were constructed according to metallurgical mechanisms of Al<sub>2</sub>O<sub>3</sub> formation, Ca-S/Ti competition, and calcium deviation (Ca_dev), enhancing feature representation for inclusion modification prediction. The KNN-GNP approach synthetically expanded the dataset while preserving statistical consistency and metallurgical validity. Machine learning models were trained and compared using R<sup>2</sup> and RMSE metrics, and interpretability was analyzed through feature importance and partial dependence plots. The proposed method achieved an R<sup>2</sup> of 0.99 and an RMSE of 0.007, outperforming conventional regression and generative models. Key variables (Al_mul_O, Ca_dev_plus_Ti, and Ca_dev_mul_Ti) were identified as dominant factors influencing inclusion modification efficiency. This framework provides a reliable and interpretable data-driven approach for optimizing metallurgical processes under limited data conditions, contributing to intelligent and sustainable steel manufacturing.</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

Mechanism-Driven Feature Engineering and KNN-Guided Gaussian Perturbation Data Synthesis for Prediction of Al2O3 Inclusion Modification Efficiency in Medium-Carbon Steel

  • 期刊论文

  • 发表日期：

    2026年2月4日

查看全文
  • Hao Wu,
  • Shuangli Liu,
  • Zhuosuo Zhou,
  • Yibo Ai,
  • Weidong Zhang

摘要

Effective modification of alumina (Al2O3) non-metallic inclusion (NMI) is essential for improving medium-carbon steel quality. However, accurately predicting Al2O3 modification efficiency under small-sample industrial conditions remains challenging due to limited labeled data and low information density. This study proposes a hybrid framework that integrates mechanism-driven feature engineering with a KNN-guided Gaussian perturbation (KNN-GNP) data synthesis method to predict the liquid NMI fraction in medium-carbon steel. Based on 25 industrial heats, 6 engineered features were constructed according to metallurgical mechanisms of Al2O3 formation, Ca-S/Ti competition, and calcium deviation (Ca_dev), enhancing feature representation for inclusion modification prediction. The KNN-GNP approach synthetically expanded the dataset while preserving statistical consistency and metallurgical validity. Machine learning models were trained and compared using R2 and RMSE metrics, and interpretability was analyzed through feature importance and partial dependence plots. The proposed method achieved an R2 of 0.99 and an RMSE of 0.007, outperforming conventional regression and generative models. Key variables (Al_mul_O, Ca_dev_plus_Ti, and Ca_dev_mul_Ti) were identified as dominant factors influencing inclusion modification efficiency. This framework provides a reliable and interpretable data-driven approach for optimizing metallurgical processes under limited data conditions, contributing to intelligent and sustainable steel manufacturing.