Statistical Methods For Mineral Engineers — Validated & High-Quality

Introduction: Why Statistics Matter in Mineral Engineering For decades, mineral engineering was dominated by empirical rules of thumb, metallurgical “balance” calculations, and deterministic models. A plant metallurgist would take a grab sample, run a quick assay, and adjust the flotation pH based on instinct. While experience remains invaluable, the modern mining industry has realized a hard truth: mineral variability is the only constant.

Yet because of sampling errors, measurement errors (belt scales, flow meters), and time synchronization issues, closing a balance perfectly is rare. The adjusts each measurement minimally to satisfy constraints. 5.2 Maximum Likelihood Reconciliation (BILMAT algorithm) Modern practice uses weighted least squares, where each measurement is assigned a variance (from sampling and analytical error). Measurements with low variance receive small adjustments; bad actors receive large adjustments—flagging them for review.

[ s^2 = K \cdot d^3 \cdot \left( \frac1M_L - \frac1M_T \right) ] Statistical Methods For Mineral Engineers

Ore bodies are heterogeneous by nature. Grade fluctuates, liberation size changes, and gangue mineralogy shifts within meters. Without rigorous statistical methods, engineers risk making decisions based on noise, designing plants for averages that never occur, or failing to detect subtle but costly process drifts.

A reconciled feed grade that is statistically more reliable than any single direct measurement. Part 6: Advanced Methods – Multivariate Statistics Today’s mineral engineer has access to automated mineralogy (QEMSCAN, MLA), NIR sensors, and laser diffraction. This creates high-dimensional data. 6.1 Principal Component Analysis (PCA) PCA reduces dozens of variables (e.g., particle size bins, mineral abundance, XRD peaks) into a few uncorrelated “principal components.” Yet because of sampling errors, measurement errors (belt

Published under a Creative Commons Attribution License. Reproduce freely with attribution.

A copper-molybdenum plant used a ( 2^3 ) factorial design and discovered that the interaction between collector dosage and pH was statistically significant (p < 0.01), whereas neither factor alone was significant. The optimum was found at a combination previously dismissed by OFAT trials. 3.2 Response Surface Methodology (RSM) Once significant factors are identified, RSM (e.g., Central Composite Design, Box-Behnken) models curvature. This is essential for finding true maxima (recovery) or minima (cost, reagent consumption). Central Composite Design

A plant processing a complex sulfide ore used PCA on 25 QA/QC variables. Two components explained 78% of variance: PC1 (sulfide content) and PC2 (clay content). Monitoring just these two components instead of 25 separate charts simplified control. 6.2 Partial Least Squares (PLS) for Grade Prediction PLS is ideal when you have many collinear predictors (e.g., XRF elemental intensities) and want to predict an assayed grade. PLS finds latent variables that maximize covariance between predictors and responses.