Scholarlink

<p>Ore control after blasting is a vital part of open-pit mining operations, ensuring that the extracted ore meets economic thresholds and facilitating block model reconciliation. Effective ore control maximizes ore recovery while reducing waste dilution, yet errors at this stage can lead to substantial financial losses. Manual ore control methods are often time-consuming, costly, and prone to uncertainty due to sampling inconsistencies and delayed laboratory results. This study investigates the use of Portable X-ray Fluorescence (pXRF) technology as an efficient and economical method to support grade control in a gold mine. Rock samples collected from blasted muck piles were analyzed using pXRF, and after data preprocessing, 15 elements were identified as reliable predictors. Because geochemical datasets are compositional, the centered log-ratio (CLR) transformation was applied, followed by principal component analysis (PCA) to produce decorrelated component scores. Three feature sets, raw geochemical data, CLR-transformed data, and PCA, were used to train Random Forest classification models to distinguish ore from waste. The CLR-transformed model achieved the strongest overall performance with a weighted F1-score of 0.90, compared to 0.86 for the raw data model. The PCA-CLR model achieved a comparable weighted F1-score of 0.89 while delivering perfect ore precision (1.0) and greater robustness across multiple training datasets, at the cost of reduced ore recall (0.70). This work systematically compares these three feature representations for pXRF-based ore–waste classification applied directly to blasted muck pile samples from an active gold mine, offering a field-deployable complementary tool to support digline placement without replacing fire assay protocols.</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

Enhancing Ore Grade Control at the Mine Site Using pXRF Geochemical Data: A Random Forest–Based Approach

  • 期刊论文

  • 发表日期：

    2026年7月20日

查看全文
  • Mehdi Abdolmaleki,
  • Saleh Ghadernejad,
  • Juan Carlos Ordóñez-Calderón,
  • Kamran Esmaeili

摘要

Ore control after blasting is a vital part of open-pit mining operations, ensuring that the extracted ore meets economic thresholds and facilitating block model reconciliation. Effective ore control maximizes ore recovery while reducing waste dilution, yet errors at this stage can lead to substantial financial losses. Manual ore control methods are often time-consuming, costly, and prone to uncertainty due to sampling inconsistencies and delayed laboratory results. This study investigates the use of Portable X-ray Fluorescence (pXRF) technology as an efficient and economical method to support grade control in a gold mine. Rock samples collected from blasted muck piles were analyzed using pXRF, and after data preprocessing, 15 elements were identified as reliable predictors. Because geochemical datasets are compositional, the centered log-ratio (CLR) transformation was applied, followed by principal component analysis (PCA) to produce decorrelated component scores. Three feature sets, raw geochemical data, CLR-transformed data, and PCA, were used to train Random Forest classification models to distinguish ore from waste. The CLR-transformed model achieved the strongest overall performance with a weighted F1-score of 0.90, compared to 0.86 for the raw data model. The PCA-CLR model achieved a comparable weighted F1-score of 0.89 while delivering perfect ore precision (1.0) and greater robustness across multiple training datasets, at the cost of reduced ore recall (0.70). This work systematically compares these three feature representations for pXRF-based ore–waste classification applied directly to blasted muck pile samples from an active gold mine, offering a field-deployable complementary tool to support digline placement without replacing fire assay protocols.