Scholarlink

<p>Process upsets are phenomena that occur more often than desired in aluminum electrolysis operations. In certain conditions, these can lead to cell tap-out. In previous work, the authors showed that latent variable models, built on process data, can detect abnormal situations. In this work, a principal component analysis (PCA) model was built on cell operation data and implemented in production at the Alcoa Fjardaal smelter. A cell ranking system was developed based on the model outcome to enable a standard prioritization method of the problematic cells. The contribution plots of the PCA analysis were also shown to help identify the potential issues. Operators investigated selected cells on the shop floor to assess the model performance. A comparison with existing tools at the plant was also carried out, showing the potential of advanced notice using a latent variables model. Examples of anomalies detected and investigation methods provided by the multivariate model are also presented.</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

Multivariate Monitoring of Reduction Cell Operations: Model Performance Assessment

  • 期刊论文

  • 发表日期：

    2025年2月7日

查看全文
  • Petre Manolescu,
  • Carl Duchesne,
  • Gudrun Saevarsdottir,
  • Julien Lauzon-Gauthier

摘要

Process upsets are phenomena that occur more often than desired in aluminum electrolysis operations. In certain conditions, these can lead to cell tap-out. In previous work, the authors showed that latent variable models, built on process data, can detect abnormal situations. In this work, a principal component analysis (PCA) model was built on cell operation data and implemented in production at the Alcoa Fjardaal smelter. A cell ranking system was developed based on the model outcome to enable a standard prioritization method of the problematic cells. The contribution plots of the PCA analysis were also shown to help identify the potential issues. Operators investigated selected cells on the shop floor to assess the model performance. A comparison with existing tools at the plant was also carried out, showing the potential of advanced notice using a latent variables model. Examples of anomalies detected and investigation methods provided by the multivariate model are also presented.