Scholarlink

<p>The efficiency and stability of top-blown bath smelting depend critically on the performance of the top-submerged lance (TSL). Conventional design methods struggle to balance accuracy and computational cost in high-dimensional parameter spaces. To address this limitation, a CFD-assisted, machine learning optimization framework was established. A dataset of 1024 CFD cases was employed to train and benchmark several surrogate models, and the artificial neural network (ANN) model delivered the highest prediction accuracy and extrapolation capability. SHAP feature importance and sensitivity analysis identified the vane rotation angle as the dominant factor affecting <i>SN</i>. The validated ANN was then coupled with the crested porcupine optimizer (CPO) algorithm for high-precision, high-efficiency optimization of lance structural parameters, achieving an optimized <i>SN</i> value of 0.9349, with a relative error of only 2.56&#xa0;pct compared with CFD simulation results. These results demonstrate the efficiency and reliability of the ANN + CPO strategy for TSL swirler optimization.</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

AI-Based Modeling and Optimization of Top-Submerged Swirling Lance in Bath Smelting Furnace for Non-ferrous Metals

  • 期刊论文

  • 发表日期：

    2025年6月2日

查看全文
  • Siyuan Chen,
  • Shiliang Yang,
  • Hua Wang

摘要

The efficiency and stability of top-blown bath smelting depend critically on the performance of the top-submerged lance (TSL). Conventional design methods struggle to balance accuracy and computational cost in high-dimensional parameter spaces. To address this limitation, a CFD-assisted, machine learning optimization framework was established. A dataset of 1024 CFD cases was employed to train and benchmark several surrogate models, and the artificial neural network (ANN) model delivered the highest prediction accuracy and extrapolation capability. SHAP feature importance and sensitivity analysis identified the vane rotation angle as the dominant factor affecting SN. The validated ANN was then coupled with the crested porcupine optimizer (CPO) algorithm for high-precision, high-efficiency optimization of lance structural parameters, achieving an optimized SN value of 0.9349, with a relative error of only 2.56 pct compared with CFD simulation results. These results demonstrate the efficiency and reliability of the ANN + CPO strategy for TSL swirler optimization.