Metallurgical Test Sample Selection using Nearest Neighbour Classification
The pre-feasibility stage of a resource characterisation project involved two drilling programs. Samples from the first program were used for metallurgical testing, which produced chemical assays for a range of size fractions. The objective of this project was to select a new set of samples from the second drilling program for more extensive metallurgical analysis. There is a risk that without considering the first round of results (combined with other geological data) the new subset of samples may not include the full range of metallurgical behaviour.
We approach the problem by performing multi-dimensional scaling and relatively simple K-means cluster analysis on the chemical assays from the first round of metallurgical testing. The results capture the main geometallurgical domains in the resource. The next step is to place the samples from each domain back in their relative geospatial locations. The 3D view (Figure 1) is encouraging, as it shows that the samples in each domain are not just chemically similar (as from the cluster analysis), but also form spatial clusters that likely relate to geological processes responsible for their formation. For example, domain 0 occurs predominantly at depth, while domain 3 forms a channel-like feature near the surface.
Figure 1: Geometallurgical domains and untested samples in their geospatial context.
The final step is to assign the head assays for samples from the second drilling program to geometallurgical domains using nearest neighbour classification. This measure of similarity essentially predicts the metallurgical behaviour of the untested samples. These results form the basis for choosing the most appropriate samples for further testing, thereby reducing technical risk to the project and helping to capture the full range of metallurgical behaviours.