Mineral processing in order to ensure high ore recovery at the final product depends on many factors, among them, the low variability in the head grades feeding the plant. For this, homogenization piles have been an effective tool for significantly diminishing variability. This study tests the efficiency of geostatistical simulations to predict in situ grade variability and how this variability is transposed to the ore, forming blending piles. Traditionally, the methods used in designing these systems are not capable of properly incorporating the uncertainty associated with the estimated value. Ordinary kriging, for example, results in best local estimates, but with a smooth global variance. Conversely, geostatistical simulations provide equiprobable scenarios for the grades of a given deposit. These scenarios honor the sample values, their histogram and spatial distribution. They also provide a set of possible values for each block which forms each homogenization pile. Finally, it is possible to predict the in situ grade variability, given the real grades. The results were compared against the real (reference) mined grades through reconciliation. Results demonstrated that the methodology is accurate and precise enough to be applied in blending piles design.
In situ variability; homogenization pile; geostatistical simulation
