14:30 〜 14:45
[HRE13-03] タングステン鉱物から製品材料までの高次元化学組成データの次元圧縮によるトレーサビリティ技術の開発
キーワード：トレーサビリティ、機械学習、レアアース
In the development of metal resources, illegal mining problems involving environmental issues and human rights violations have arisen, but traceability technology for mineral resource materials that can deal with these problems has not yet been established. In order to establish traceability technology, this study focuses on tungsten, which has limited production areas, and proposes a traceability technology using machine learning to identify intermediate products from the place of origin to industrial products by the patterns of contained elements (impurities). Using machine learning, we aim to develop a traceability technology for metallic materials by identifying the place of origin from the REE (rare earth elements) patterns of tungsten minerals and finally from the elemental patterns of raw materials.
In this study, REEs of two tungsten minerals, Scheelite and Wolframite, from multiple localities were measured by LA-ICP-MS, and 260 published REE pattern data of tungsten minerals (30 localities) were also used. These REE patterns were characterized and identified by unsupervised machine learning (PCA and UMAP) to identify the origin and material.
The REE patterns differ between Scheelite and Wolframite and can be further distinguished using PCA and UMAP. The REE patterns and PCA results varied depending on the mineral type, but there were no significant differences based on provenance, and PCA was unable to identify it. On the other hand, UMAP can classify into different clusters depending on the attribute, and it is considered to have better discrimination performance than PCA.
The REE concentrations in the WC material were ultra-trace (10<ppb) and could not be accurately measured. This suggests that the concentration of REEs decreases during the material manufacturing process. PCA and UMAP were utilized for dimensional compression.
Machine learning (UMAP) for the REE patterns of tungsten minerals and tungsten carbide revealed that the various REE patterns in the mineral stage converge to a certain region during the manufacturing process. This could be applied to traceability of tungsten from mineral production to material manufacturing.
In this study, REEs of two tungsten minerals, Scheelite and Wolframite, from multiple localities were measured by LA-ICP-MS, and 260 published REE pattern data of tungsten minerals (30 localities) were also used. These REE patterns were characterized and identified by unsupervised machine learning (PCA and UMAP) to identify the origin and material.
The REE patterns differ between Scheelite and Wolframite and can be further distinguished using PCA and UMAP. The REE patterns and PCA results varied depending on the mineral type, but there were no significant differences based on provenance, and PCA was unable to identify it. On the other hand, UMAP can classify into different clusters depending on the attribute, and it is considered to have better discrimination performance than PCA.
The REE concentrations in the WC material were ultra-trace (10<ppb) and could not be accurately measured. This suggests that the concentration of REEs decreases during the material manufacturing process. PCA and UMAP were utilized for dimensional compression.
Machine learning (UMAP) for the REE patterns of tungsten minerals and tungsten carbide revealed that the various REE patterns in the mineral stage converge to a certain region during the manufacturing process. This could be applied to traceability of tungsten from mineral production to material manufacturing.