Scholarlink

<p>The viscosity of metallurgical slag is a critical parameter influencing process efficiency and product quality. Traditional experimental methods are time-consuming and complex, while existing predictive models often suffer from limited accuracy and generalizability. A highly accurate and robust viscosity prediction and optimization method for the CaO–SiO<sub>2</sub>–MgO–Al<sub>2</sub>O<sub>3</sub> slag system was developed by integrating thermodynamic simulation, experimental measurements, and machine learning. A hybrid dataset was constructed by generating 3000 theoretical viscosity data points using FactSage software and incorporating 255 representative experimental values. To reconcile the discrepancy between simulated and measured data, a nearest-neighbor error compensation strategy was employed, yielding a corrected dataset with improved agreement to experimental observations. Based on this dataset, 28 regression algorithms were evaluated. Among them, exponential Gaussian process regression achieves the best performance, with a root mean square error of 0.0195&#xa0;Pa&#xa0;s and a coefficient of determination (<i>R</i><sup>2</sup>) of 0.9681 on the test set—representing improvements of 89.5% and 61.8% compared to uncorrected and purely experimental models, respectively. The model demonstrates strong generalization ability and resistance to overfitting. Furthermore, a genetic algorithm was applied to optimize slag composition and temperature, achieving a minimum predicted viscosity of 0.3298&#xa0;Pa&#xa0;s under specified constraints. By combining the generalizability of thermodynamic simulations with the precision of experimental data, this method provides a reliable strategy for viscosity prediction and process optimization in the complex slag system and offers potential for broader applications in predicting high-temperature molten material properties.</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

Viscosity prediction and optimization of CaO–SiO2–MgO–Al2O3 slag system based on integrated calibration of multi-source data

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • En-Ze Shi,
  • Jue Tang,
  • Man-Sheng Chu,
  • Hong-Yu Tian

摘要

The viscosity of metallurgical slag is a critical parameter influencing process efficiency and product quality. Traditional experimental methods are time-consuming and complex, while existing predictive models often suffer from limited accuracy and generalizability. A highly accurate and robust viscosity prediction and optimization method for the CaO–SiO2–MgO–Al2O3 slag system was developed by integrating thermodynamic simulation, experimental measurements, and machine learning. A hybrid dataset was constructed by generating 3000 theoretical viscosity data points using FactSage software and incorporating 255 representative experimental values. To reconcile the discrepancy between simulated and measured data, a nearest-neighbor error compensation strategy was employed, yielding a corrected dataset with improved agreement to experimental observations. Based on this dataset, 28 regression algorithms were evaluated. Among them, exponential Gaussian process regression achieves the best performance, with a root mean square error of 0.0195 Pa s and a coefficient of determination (R2) of 0.9681 on the test set—representing improvements of 89.5% and 61.8% compared to uncorrected and purely experimental models, respectively. The model demonstrates strong generalization ability and resistance to overfitting. Furthermore, a genetic algorithm was applied to optimize slag composition and temperature, achieving a minimum predicted viscosity of 0.3298 Pa s under specified constraints. By combining the generalizability of thermodynamic simulations with the precision of experimental data, this method provides a reliable strategy for viscosity prediction and process optimization in the complex slag system and offers potential for broader applications in predicting high-temperature molten material properties.