Hyperspectral Imaging Applications to Geometallurgy: Utilizing Blast Hole Mineralogy to Predict Au-Cu Recovery and Throughput at the Phoenix Mine, Nevada

Abstract

Abstract The Phoenix mine and predecessor operations in north-central Nevada have produced an aggregate of 5.2 Moz of gold and 550 million pounds of copper from an Eocene-aged Au-Cu porphyry-related skarn. The complex skarn mineralogy intimately associated with ore-grade mineralization poses significant challenges to blasting, mining, comminution, and process operations. These challenges are rooted in highly variable silicate mineralogy, which manifests as fine-grained, submillimeter grain-size, generally green colored rocks that inhibit accurate identification in the field. Prior to this study, all mineralogical data utilized in Phoenix mine ore control were sourced from blast hole cuttings mapped by ore control geologists in the field—the standard practice at many modern mine sites. At Phoenix, a direct link between mineralogy and mill performance was recognized; however, mineralogical data captured in the field was not sufficient to optimize process operations. To address this, it was determined that analytical work was necessary to quantify fine-grained mineralogy of variable ore types. A visible-near and short-wave infrared (VNIR-SWIR) hyperspectral imaging system provided the ideal tool, as it allows near real-time mineralogical data acquisition and semiquantitative determination of mineral abundances. Multiple iterative studies were conducted to prove that hyperspectral imaging of Phoenix ore types provides results suitable for process optimization. This six-month study described here included hyperspectral imaging of 3,008 blast hole cuttings samples from three pits, and 877 crusher feed, rougher feed, and rougher tails samples. Hyperspectral feature extractions derived from mill samples were paired with associated mill performance data and used to build predictive Au-Cu recovery, grade, and throughput models using multiple linear regression, partial least squares, and deep learning techniques with R-correlation values to observed data of 0.56 to 0.71. Blast hole hyperspectral data were then applied to recovery, grade, and throughput models to calculate predicted recoveries and throughputs that were spatially kriged with excellent correlations to geologic features. The application of VNIR-SWIR hyperspectral imaging to blast hole cuttings is a powerful predictive and diagnostic geometallurgical tool in operations where silicate mineralogy has a strong impact on process operations.