Scholarlink

To replenish aluminaAlumina consumed during the electrolysis process, aluminaAlumina doses are periodically injected into the bath following a predetermined sequence. The feeding period is generally the same for all feeders and varies according to overfeeding and underfeeding cyclesCycle. This procedure is crucial for monitoring the average concentration of dissolved aluminaAlumina in the bath through global cell voltageVoltage variationsVariations. However, this approach does not prevent local variationsVariations in dissolved aluminaAlumina content, which can lead to anode effectsAnode effects in case of local aluminaAlumina deficits, or sludge formationSludge formation in the case of local surpluses. To enhance homogenizationHomogenization of the dissolved aluminaAlumina concentration, we propose to optimize the injection frequency of each feeder taking into account the physical phenomena involved in the process. The AlucellAlucell software is first used to describe magnetohydrodynamics flows, as well as the injection, transport, dissolution, and consumption of aluminaAlumina in the electrolytic bath. For a given operating point, simulationsSimulation are carried out for various sets of feeder injection frequencies. A deep-learning surrogate model, using a graph neural network architecture, is then trained against the numerical results, and is used to provide near instantaneous predictions of the dissolved aluminaAlumina distribution forDistribution any set of feeder injection frequencies. Ultimately, the use of a multiphysics model combined with a deep learningDeep learning surrogate makes it possible to identify the optimal feeder parameters for maximizing alumina homogeneityAlumina homogeneity in the bath, or to target aluminaAlumina injection in a specific region where a deficiency is suspected. This innovative approach opens up promising opportunities for pot process controlProcess control, offering potential improvements in efficiency and carbon footprintCarbon footprint reductionReduction, notably by limiting the frequency of anode effectsAnode effects.

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

Optimization of Alumina Feeding in Electrolysis Cells Using Multi-physics Modeling and Deep Learning Surrogate

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kévin Patouillet,
  • Nadia Chailly,
  • Bertrand Allano,
  • Alan Clark,
  • John Perry,
  • Matías Vázquez

摘要

To replenish aluminaAlumina consumed during the electrolysis process, aluminaAlumina doses are periodically injected into the bath following a predetermined sequence. The feeding period is generally the same for all feeders and varies according to overfeeding and underfeeding cyclesCycle. This procedure is crucial for monitoring the average concentration of dissolved aluminaAlumina in the bath through global cell voltageVoltage variationsVariations. However, this approach does not prevent local variationsVariations in dissolved aluminaAlumina content, which can lead to anode effectsAnode effects in case of local aluminaAlumina deficits, or sludge formationSludge formation in the case of local surpluses. To enhance homogenizationHomogenization of the dissolved aluminaAlumina concentration, we propose to optimize the injection frequency of each feeder taking into account the physical phenomena involved in the process. The AlucellAlucell software is first used to describe magnetohydrodynamics flows, as well as the injection, transport, dissolution, and consumption of aluminaAlumina in the electrolytic bath. For a given operating point, simulationsSimulation are carried out for various sets of feeder injection frequencies. A deep-learning surrogate model, using a graph neural network architecture, is then trained against the numerical results, and is used to provide near instantaneous predictions of the dissolved aluminaAlumina distribution forDistribution any set of feeder injection frequencies. Ultimately, the use of a multiphysics model combined with a deep learningDeep learning surrogate makes it possible to identify the optimal feeder parameters for maximizing alumina homogeneityAlumina homogeneity in the bath, or to target aluminaAlumina injection in a specific region where a deficiency is suspected. This innovative approach opens up promising opportunities for pot process controlProcess control, offering potential improvements in efficiency and carbon footprintCarbon footprint reductionReduction, notably by limiting the frequency of anode effectsAnode effects.