Nuclear analytical techniques applied to determine ore hardness using artificial intelligence

The use of ore hardness metrics, known as comminution parameters, is essential in mineral processing for planning mills and optimizing their power and energy consumption during operation. The parameter used depends on the type of mill and power model used, and the industry standards are the Drop Weight Index (DWI) and the Bond Work Index (BWI). Traditional tests that calculate these parameters for mineral samples are complex and expensive, making them prohibitive to be carried out in large quantities by mining companies, leading to design flaws due to the lack of information about hardness variability. However, due to the subsequent requirements of mineral processing, nuclear analytical techniques are carried out in large quantities, including X-ray Fluorescence (XRF), Inductively Coupled Plasma Atomic Emission Spectroscopy (ICP-AES), and X-ray Diffraction (XRD). Machine learning algorithms trained with a small number of comminution tests in conjunction with data from nuclear analytical techniques can be used to estimate the parameters of samples whose hardness has not been tested [4], providing an additional and practical way for comminution projects to obtain more data on the variability of the mineral body, reducing the energy consumption of mills and the emission of polluting gases. The present work aims to investigate the results of using machine learning methods for this application, evaluating the individual performance of two nuclear analytical techniques: XRD, which provides information about the crystalline structure of a sample [5], and ICP-AES, which is a spectroscopy technique to measure the concentration of each chemical element [6]. The objective is to simulate practical scenarios where only one of the two techniques was performed and verify whether crystallography or spectroscopy provides a more satisfactory result in conditions of database limitation. (author)