Developing geometallurgical proxies for ore hardness - A mineral sands case study

Raw data from: Sikushumane, M. (2024). Developing geometallurgical proxies for ore hardness - A mineral sands case study. Master of Science thesis. Department of Chemical Engineering, University of Cape Town.

Raw data contained in spreadsheets includes: XRF, QEMSCAN, XRD, XCT porosity, Equotip HLS and A*b measurements for the complete sample set.

Abstract: The breakage response of an ore during comminution is directly linked to its hardness. Thus, understanding ore hardness variability is paramount to effective mining and processing. Traditionally, ore hardness is measured using single particle breakage tests, such as the JK drop weight test. However, direct hardness testing is costly, time-consuming, and requires large sample masses, and is thus not suitable for routine measurement of ore variability. Consequently, extensive research has been conducted to develop geometallurgical proxies to predict ore hardness. Geometallurgical proxies refer to indirect data that can be correlated with direct hardness measurements. Literature shows that the correlations between proxy data and direct hardness measurements vary depending on the characteristics of the ore, and need to be developed for each ore based on its geological and mineralogical characteristics.

This study aims to develop geometallurgical proxies to predict the hardness of the cemented hard layers (CHLs) found in a local mineral sands deposit. To achieve this, 43 samples were collected from 11 locations across the current mining pits in one of the two orebodies called the West orebody. Proxy data, inclusive of chemical assays, bulk mineralogy, porosity, and Equotip Leeb hardness (HLS) were collected, as well as the direct hardness index (A*b) from the hardness index tester. Correlations were investigated between the proxy data and the A*b results (A*b and 100/A*b) to identify potential proxies to predict ore hardness. Categorical data consisting of location (pit) and layer type were also investigated as potential indicators of hardness variability. Once the variables with statistically significant correlations with either A*b or 100/A*b had been identified, linear regressions were run using the least squares method.

A total of 14 statistically significant regression models were developed as potential proxies. The 14 models were compared against each other and ranked based on how well they predicted A*b or 100/A*b based on the R² values and relative standard errors (RSEs), and the cost of proxy data acquisition and turnaround time to identify the most suitable proxy model(s) for the CHLs. The most suitable proxy model identified predicts A*b from chemical assays and Equotip HLS, and incorporates pit information, with an R² of 0.67 and RSE of 36.9 %. However, because Equotip measurements are not routinely collected at this site yet, the top two chemical assay-based proxy models were also identified as suitable proxies to use in cases where the Equotip dataset is not available. These ranked fourth based on the above-mentioned criteria and have R² and RSE, respectively, of 0.53 and 45.2 % (A*b model) and 0.65 and 88.2 % (100/A*b model). The high RSE in the 100/A*b model is attributed to the fewer number of hard samples in the current dataset. The proxies developed in this study are preliminary and serve as basis for further geometallurgical research aimed at increasing orebody knowledge to improve the mining and processing efficiency of the CHLs.