Japan Geoscience Union Meeting 2023

Presentation information

[J] Oral

H (Human Geosciences ) » H-TT Technology & Techniques

[H-TT15] Development and application of environmental traceability methods

Tue. May 23, 2023 10:45 AM - 12:00 PM 201B (International Conference Hall, Makuhari Messe)

convener:Ichiro Tayasu(Research Institute for Humanity and Nature), Ki-Cheol Shin(Research Institute for Humanity and Nature), Nobuhito Ohte(Department of Social Informatics, Graduate School of Informatics, Kyoto University), Chairperson:Ichiro Tayasu(Research Institute for Humanity and Nature)

10:45 AM - 11:00 AM

[HTT15-06] Metal Traceability Technology Using REE Pattern Identification Rare Earth Patterns by Machine Learning : From Tungsten Minerals to Raw Materials

*Yuya Kato1, Noriyoshi Tsuchiya1, Diana Mindaleva1, Matsuno Satoshi1, Geri Agroli1 (1.Graduate School of Environmental Studies, TOHOKU University)


Keywords:traceability, machine learning

Traceability is the process of making a product traceable from the procurement of raw materials through production to consumption or disposal. Traceability of metal materials contributes to the quality, safety, and sustainability of metal products, but it is generally difficult to trace the supply chain because the primary mining mine and the final production site are far apart. Illegal mining is a typical example, causing environmental destruction and human rights abuses. Therefore, a method of identifying mineral resources and products is needed to ensure traceability of concentrates. In this study, we focused on tungsten, a metal material with a limited origin, and verified a new method using machine learning to identify the origin.
To establish a traceability technique, we analyzed the patterns of rare earth elements in tungsten-bearing minerals (primary samples) and tungsten carbide (WC) (final product). Seven primary samples were collected: Scheelite : CaWO4 (Japan: Kiwada Mine, Date Nagai Mine, Kobushin Mine, Kaneuchi Mine; China: Xuebaoding Mine; USA: Merrill Prospect) and Wolframite : (Fe,Mn)WO4 (Japan: Akenobe Mine). Three WC samples from different manufacturers were used as final products. Rare earths in the samples were determined by LA-ICP-MS for tungsten-bearing minerals, dissolved in 95% H2O2 and 5% royal water, and ICP-MS for the WC samples. Elemental mapping was performed using XRF to identify tungsten minerals, and five points were measured per sample. The obtained rare-earth patterns were identified on two dimensions using two dimensional compression methods (PCA and UMAP) in machine learning. The data set used the respective ratios of each of the 15 rare earth elements (105 in total) for a total of 15 lanthanides and yttrium in each sample to identify by the rare earth patterns.
The rare earth patterns of all samples followed Odd Harkins' law. Eu values were higher in the samples from the Kiwada, Kaneuchi, and Date Nagai mines. Only the Merrill Prospect sample showed different REE patterns at different points within the same sample. The Akenobe Mine sample showed a large difference in concentration (about 10,000-fold), but the REE pattern was similar from point to point. but they show different rare-earth patterns. The Date Nagai Mine was a skarn deposit, and the Xuebaoding Mine was a pegmatite deposit, but the rare earth patterns were similar. From the above, it was found that the rare earth patterns were independent of deposit type and minerals.
PCA was performed on the obtained REE patterns, and the PCA results (fig.1) were plotted in two dimensions (PC1 and PC2) for the first and second principal components, since the cumulative contribution was about 80% up to the second principal component. The ratio of Eu to Sm and Gd was different when comparing the rare-earth patterns of the Xuebao ding and Date Nagai mines, where PC1 was similar and only PC2 was distant, and the ratio of Eu to Sm and Gd was different. On the other hand, comparing the rare earth patterns of the Kiwada Mine and the Date Nagai Mine, where PC2 is similar but only PC1 is separated, the ratio of heavy REEs increased in the Kiwada Mine, but decreased in the Date Nagai Mine, suggesting that the behavior of heavy REEs was affected.
Since the PCA results could not clearly classify each sample, UMAP, which specializes in classification, was used to analyze the results. The UMAP results were more clearly clustered than the PCA results, with a large division between the Date Nagai Mine, the Xuebao ding Mine, and other samples. The UMAP results are divided into two major groups: the Kiwada Mine and the Date Nagai Mine, and the other samples.
In this study, it was found that the REE patterns in the minerals can be discriminated for each producing region by using machine learning. This can be used to identify the origin of the minerals and for traceability.