Scholarlink

<p>The multi-stage manufacturing process of non-oriented silicon steel (NOSS) involves complex interactions among composition, processing parameters and properties. When quality deviations arise, pinpointing critical process variables from extensive production data poses significant technical challenges. To enable rapid root-cause diagnosis of property anomalies, this study proposes a data-driven machine learning framework for decoding composition–process–property (CPP) correlations in NOSS. A data preprocessing strategy is developed, and a Bayesian-optimized categorical gradient boosting (BO-CatBoost) algorithm is applied to construct predictive models for CPP relationships, achieving high-precision performance prediction. Additionally, an intelligent optimization framework combining data analytics and particle swarm optimization (PSO) is implemented to identify dominant process parameters influencing property deviations and formulate corresponding optimization strategies. Results show titanium (Ti) content exhibits the strongest correlation with core loss under specified processing conditions. A controlled reduction of Ti content by 0.0007 mass% is shown to effectively decrease core loss by 0.2&#xa0;W·kg⁻<sup>1</sup>.</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

Prediction and Optimization of Core Loss for Non-oriented Silicon Steel Based on Machine Learning

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Tianlong Su,
  • Siwei Wu,
  • Yuduo Zhao,
  • Yuqiao Wang,
  • Tai Zhang,
  • Guodong Wang

摘要

The multi-stage manufacturing process of non-oriented silicon steel (NOSS) involves complex interactions among composition, processing parameters and properties. When quality deviations arise, pinpointing critical process variables from extensive production data poses significant technical challenges. To enable rapid root-cause diagnosis of property anomalies, this study proposes a data-driven machine learning framework for decoding composition–process–property (CPP) correlations in NOSS. A data preprocessing strategy is developed, and a Bayesian-optimized categorical gradient boosting (BO-CatBoost) algorithm is applied to construct predictive models for CPP relationships, achieving high-precision performance prediction. Additionally, an intelligent optimization framework combining data analytics and particle swarm optimization (PSO) is implemented to identify dominant process parameters influencing property deviations and formulate corresponding optimization strategies. Results show titanium (Ti) content exhibits the strongest correlation with core loss under specified processing conditions. A controlled reduction of Ti content by 0.0007 mass% is shown to effectively decrease core loss by 0.2 W·kg⁻1.