2025 - Sustainable Industrial Processing Summit
SIPS2025 Volume 1. Dutrow Intl. Symp. / Geochemistry
| Editors: | F. Kongoli, G. Artioli, M. Asta, S. Hayun, A. Navrotsky, R. Riedel, N. Ross, A. Simon, B. Tsikouras, S. Webb |
| Publisher: | Flogen Star OUTREACH |
| Publication Year: | 2025 |
| Pages: | 156 pages |
| ISBN: | 978-1-998384-38-9 (CD) |
| ISSN: | 2291-1227 (Metals and Materials Processing in a Clean Environment Series) |
CD shopping page
PETROGRAPHIC INTELLIGENCE: PREDICTING CATALYTIC POTENTIAL FROM ROCKS
Elena Ifandi1; Daphne Teck Ching Lai1; Haezan Jangarun1; Stavros Kalaitzidis2;1UNIVERSITI BRUNEI DARUSSALAM, Bandar Seri Begawan, Brunei Darussalam; 2UNIVERITY OF PATRAS, Rio, Greece;
Type of Paper: Regular
Id Paper: 348
Topic: 67
Abstract:
Chromitites uniquely catalyse low‑temperature methane formation [1], motivating data‑driven exploration of naturally occurring catalysts for CO₂ methanation aligned with sustainable energy goals [2]. Using 12 Greek chromitite samples with quantified methane, petrographic point counting and grain‑size distributions were encoded via percentile ranks and modeled with a super‑ensemble that stacks Multinomial Naive Bayes with XGBoost, guided by automated model search and manual tuning. The selected classifier achieved 75% training and 71% test accuracy, outperforming alternative algorithms evaluated on these data. Model interpretability using SHAP [3] and partial dependence plots identified several key predictors of high methane levels. Large olivine crystals within chromitites emerged as the strongest positive predictor. Medium-sized veins showed a positive association, while large veins had adverse effects. Large spinel crystals acted as a secondary, though weaker, indicator. The workflow converts petrographic observations—often visible at hand‑specimen scale—into practical field criteria for targeting chromitites that host mineral catalysts, thereby reducing reliance on synthetic catalysts and mitigating pressure on noble and critical metal supply chains. This first application of machine learning to field exploration of mineral catalysts demonstrates how tree‑based, interpretable ensembles can capture complex relationships in multivariate petrographic data and enable precision exploration for carbon‑neutral energy materials.