Scholarlink

<p>Accurate magnesium (Mg) control is critical for the industrial recycling of 5xxx-series aluminum alloys. This study demonstrates a SCADA-integrated, machine learning-based predictive control strategy for Mg regulation in 5182 Al–Mg–Mn alloy during full-scale furnace operation. A gradient boosting regressor (GBR) was trained using a time-resolved intra-heat reference campaign (23 compositional measurements) and more than 20,000 SCADA-recorded thermal and combustion variables. Using temporally consistent blocked time-series validation, the model achieved stable predictive performance (<i>R</i><sup>2</sup> ~ 0.96, RMSE ~ 0.11 wt.% Mg) across an operating range of 1.78-4.91 wt.%. Mg. The exploratory warm-start transfer to a second furnace preserved trend fidelity under similar control logic. ML-assisted supervision reduced magnesium dosing variance by 28.7% and post-alloying stabilization time by 12.4%, yielding a 3-4% reduction in net energy consumption and CO<sub>2</sub>-equivalent savings of approximately 230-260 tons per furnace per year. The results demonstrate the industrial feasibility of data-driven alloying control under real production constraints.</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

Predictive Control of Magnesium Content in Industrial 5182 Aluminum Alloy Recycling Using SCADA-Guided Gradient Boosting

  • 期刊论文

  • 发表日期：

    2026年3月26日

查看全文
  • Mengya Wang,
  • Jiahui Xu,
  • Xiaohu Wang,
  • Farid Wirawan,
  • Mouhamadou Aziz Diop

摘要

Accurate magnesium (Mg) control is critical for the industrial recycling of 5xxx-series aluminum alloys. This study demonstrates a SCADA-integrated, machine learning-based predictive control strategy for Mg regulation in 5182 Al–Mg–Mn alloy during full-scale furnace operation. A gradient boosting regressor (GBR) was trained using a time-resolved intra-heat reference campaign (23 compositional measurements) and more than 20,000 SCADA-recorded thermal and combustion variables. Using temporally consistent blocked time-series validation, the model achieved stable predictive performance (R2 ~ 0.96, RMSE ~ 0.11 wt.% Mg) across an operating range of 1.78-4.91 wt.%. Mg. The exploratory warm-start transfer to a second furnace preserved trend fidelity under similar control logic. ML-assisted supervision reduced magnesium dosing variance by 28.7% and post-alloying stabilization time by 12.4%, yielding a 3-4% reduction in net energy consumption and CO2-equivalent savings of approximately 230-260 tons per furnace per year. The results demonstrate the industrial feasibility of data-driven alloying control under real production constraints.