Technological advances in sensor and material handling technology has enabled a step change in mine operations performance. Sensors capable of monitoring the grade of broken rock mass can be mounted on shovels, truck loading units or conveyors allowing highly precise ore/waste decisions to be made on small parcels of rock, increasing selectivity. This increased selectivity has the potential to remove sub-economic material from the product stream prior to incurring ore treatment costs. Bulk ore sorting is capable of intervening at <10t “pods” on run of mine conveyors, or alternatively at per bucket loading (5-100t) or truck scale (30-350t) parcels of ore. this grade engineering technique being developed by crcore (rutter, 2017) can improve mill feed grade, remove deleterious materials and ensure that the high capital intensity ore treatment facility is presented with the best ore feed strategy.>
While this technology can improve mine economics, estimating the performance of a bulk ore sorting system is challenging. Traditional resource estimation practices such as Ordinary Kriging and Inverse distance weighting (IDW) are based on weighted averages, a process that smooths the grade profile, reducing modelled variance and dispersion. Its purpose is to generate a globally unbiased result but is less suited to generating a model of adequate granularity to enable confidence in local grade heterogeneity. This is an essential component for bulk ore sorting success. Estimating ore sorting performance requires knowledge of the grade-tonnage curve at the scale of the sensor diversion unit (SDU). Evaluation based on an estimator that smooths the SDU distribution will not correctly predict ore/waste separation or economic performance.
Alternatives to traditional linear geostatistics have been tested on a variety of mineralisation styles. Additionally, a number of heterogeneity metrics have been developed to assist with evaluation and assessment. The difference between estimation for grade (metal) and modelling for traditional mining block models, versus methods used for predicting heterogeneity and variability are explored.