Predictive models of mineralogy from whole-rock assay data: case study from Productora Cu-Au-Mo deposit, Chile

Mineralogy is a fundamental characteristic of a given rock mass throughout the mining value chain. Understanding bulk mineralogy is critical when making predictions on processing performance. However, current methods for estimating complex bulk mineralogy are typically slow and expensive. Whole-rock geochemical data can be utilized to estimate bulk mineralogy using a combination of ternary diagrams and bivariate plots to classify alteration assemblages (alteration mapping), a qualitative approach, or through calculated mineralogy, a predictive quantitative approach. Both these techniques were tested using a data set of multielement geochemistry and mineralogy measured by semiquantitative X-ray diffraction data from the Productora Cu-AuMo deposit, Chile.

Using geochemistry, samples from Productora were classified into populations based on their dominant alteration assemblage, including quartz-rich, Fe-oxide, sodic, potassic, muscovite (sericite) and clay-alteration, and least altered. Samples were also classified by their dominant sulfide mineralogy. Results indicate that alteration mapping through a range of graphical plots provides a rapid and simple appraisal of dominant mineral assemblage in volcanic rocks, which closely matches the measured mineralogy.

In this example, calculated mineralogy using linear programming generated robust quantitative estimates for major mineral phases, including quartz and total feldspars, as well as pyrite, iron oxides, chalcopyrite, and molybdenite, which matched the measured mineralogy data extremely well (R² values greater than 0.78, lowmoderate root mean square error). The results demonstrate that calculated mineralogy can be applied in the mining environment to significantly increase bulk mineralogy data and quantitatively map mineralogical variability. This was useful even though several minerals were challenging to model due to compositional similarities, and clays and carbonates could not be predicted accurately.

Fig. 1. Comparison of mineralogy estimates by semiquantitative X-ray diffraction data corrected to assays by weighted least squares (WLSQ; measured mineralogy) against calculated mineralogy by linear programming. Regression line and R² also shown. A) Quartz. B) K-feldspar. C) Plagioclase (including sodic and calcic plagioclase). D) Pyrite. E) Chalcopyrite. F) Molybdenite. G) Total chlorite and micas (including chlorite, muscovite, phengite and biotites). H) Total clay (including kaolinite and smectite-montmorillonite). I) Fe-oxides (including magnetite, hematite, goethite). Major minerals, including quartz, feldspars, pyrite and Fe-oxides show excellent correlation between measured (WLSQ) and calculated (linear programming).