An Analysis of plant performance indicators for enhancing recovery efficiency in Platinum Group Metals processing.

Abstract

This study addressed the challenge of inconsistent platinum recovery at Mimosa Mining Company, where operational decisions were often based on assumptions rather than empirical evidence. A log-linear (log-normal) regression model was developed using historical plant data on feed grade, concentrate grade, tail grade, and grind size to quantify their influence on recovery. The model satisfied normality assumptions on the log scale, corrected for mild heteroscedasticity with robust standard errors, and was validated out of sample. Key findings included a negative elasticity for feed grade (higher feed does not always improve recovery) and a positive elasticity for grind size (coarser grinding may increase recovery), challenging conventional industry beliefs. These insights were translated into actionable recommendations for feed blending and grind optimization, providing process engineers with an evidence-based framework to stabilize and enhance recovery. The study demonstrates that a well-specified statistical model can replace guesswork, reduce operational risks, and support data-informed decision-making in PGM processing.