A data-driven approach for predicting nepheline crystallization in high-level waste glasses


Creative Commons License

Sargin I., Lonergan C. E. , Vienna J. D. , McCloy J. S. , Beckman S. P.

JOURNAL OF THE AMERICAN CERAMIC SOCIETY, vol.103, no.9, pp.4913-4924, 2020 (Peer-Reviewed Journal) identifier identifier

  • Publication Type: Article / Article
  • Volume: 103 Issue: 9
  • Publication Date: 2020
  • Doi Number: 10.1111/jace.17122
  • Journal Name: JOURNAL OF THE AMERICAN CERAMIC SOCIETY
  • Journal Indexes: Science Citation Index Expanded, Scopus, Academic Search Premier, Aerospace Database, Applied Science & Technology Source, Art Source, Chimica, Communication Abstracts, Compendex, Computer & Applied Sciences, EBSCO Education Source, INSPEC, Metadex, Civil Engineering Abstracts
  • Page Numbers: pp.4913-4924
  • Keywords: data-science, classification, high-level waste glass, nepheline, THERMODYNAMIC ASSESSMENT, LIQUIDUS, SODIUM, IRON, SPECIATION

Abstract

High-level waste (HLW) glasses with high alumina content are prone to nepheline crystallization during the slow canister cooling that is experienced during large-scale production. Because of its detrimental effects on glass durability, nepheline (NaAlSiO4) precipitation must be avoided; however, developing robust, predictive models for nepheline crystallization behavior in compositionally complex HLW glasses is difficult. Using overly conservative constraints to predict nepheline formation can limit the waste loading to lower than the achievable capacity. In this study, a robust data-driven model using five compositional features has been developed to predict nepheline formation. A new descriptor is introduced called the "difference based on correlation", which has higher accuracy compared to previous descriptors and also has more balanced false positive and false negative rates. The analysis of the model and the data show an overlap, instead of a distinct compositional boundary, between glasses that form and do not form nepheline. As a result, the model's predictive accuracy is not the same throughout the feature space and instead is dependent on the location of the glass composition in the dimensionally reduced feature space.