Abstract
Metallurgical accounting in Platinum Group Metal (PGM) concentrator plants relies on representative sampling and reliable assay data; however, measurement system variability associated with slurry feed sampling often dominates total uncertainty. This study evaluates the capability of a feed stream measurement system in a PGM concentrator plant. It uses an integrated, Theory of Sampling (TOS) based framework combining replication experiments, variographic analysis and sampling scheme optimisation.
A replication experiment was conducted to quantify short-term sampling and analytical variability across the complete lot-to-aliquot pathway, incorporating primary sampling, secondary sampling, sample preparation and laboratory analysis. The results yielded Relative Standard Variability (RSV) values of 7.5% for primary sampling, 3.9% for secondary sampling and 1.7% for laboratory analysis, demonstrating that the measurement system can deliver acceptable precision under controlled operating conditions and confirming that primary sampling is the chief contributor to total measurement uncertainty.
To assess long-term measurement system performance under routine plant operation, variographic analysis was applied to a six-month time series of 12 hourly feed assays. The absolute variogram revealed a low overall sill and negligible cyclic variability, indicating stable process behaviour. However, a dominant nugget effect accounting for approximately 58% of the total variance was identified, demonstrating that short-range measurement system variability, rather than true process variation, governs long-term data quality. This finding highlight that while the sampling system is capable in principle, sustained operational performance is limited by residual sampling related errors.
Variography-based Total Sampling Error (TSE) simulation was subsequently used to optimise the sampling scheme. The results show that increasing the number of increments per composite sample and reducing the sampling interval systematically reduced TSE. An optimal, operationally feasible configuration of Q = 12 increments at a sampling rate of r = lag 1 was identified, yielding an estimated TSE of approximately 2.9%. This configuration directly targets the dominant short-...