Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.603473
Title: The application of a near-infrared sensor to the sorting of minerals
Author: Gaydon, James William
Awarding Body: University of Exeter
Current Institution: University of Exeter
Date of Award: 2011
Availability of Full Text:
Access through EThOS:
Abstract:
The aim of this research was to investigate the potential of a Near-Infrared (NIR) sensor for use in the automatic sorting of minerals and ores. The NIR region of the electromagnetic spectrum has been used for mineral mapping of core logs and to characterise mineral samples in the minerals industry. NIR sensors are used in automatic sorting applications in the food and recycling industry but minimal work has been carried out to apply this technology to the sorting of minerals. The investigation involved two aspects: optimisation of a NJR sensor and the development of methodologies to sort a copper ore. The experimental setup consisted of a NIR sensor situated above a belt, with the belt illuminated by two infrared halogen lamps. The sensor used an Acousto-ptic Tuneable Filter (AOTF) crystal to filter wavelengths of light and measured up to 371 wavelengths between 1308 nm and 2405 nm. Two methods were assessed as a means to convert the raw data to reflectance values. Optimisation of the sensor involved physical changes to the setup with each change assessed by examining the effect on the signal-ta-noise ratio of the spectra measured. The potential application of a smoothing algorithm was investigated and looked at in terms of noise removal and spectral feature retention. Methods of sorting a copper ore based on NIR readings were investigated. The ore originated from the Mantos Blancos mine, Chile and had been split into thirteen fractions. The aim was to separate the high-grade material from the low-grade and lo remove the carbonates from the product material. It was also desired that the low-grade oxide fraction be retained in the product fraction of the ore and not lost to waste. The ore was examined using chemical analysis and NIR measurements of each of the ore fractions. Peaks within the NIR spectrum of the rock types were assigned to various chemical species and spectral analysis was carried out with a view to the potential sorting of the ore. The significant contributions to the NIR spectra of the different fractions of ore by iron sulphide minerals. copper-rich minerals such as chrysocolla. malachite (and associated minerals), muscovite and calcite were noted. It was also noted that the high-grade copper carbonate fraction particles were very similar spectrally to the other high-grade copper fraction particles and that the low-grade oxide particles shared a number of characteristics with the other low-grade material. This would complicate any separation of these fractions into the desired product/waste classification. With an understanding of the characteristics of the ore spectra. methods were investigated to separate product and waste material. A number of methods were explored and compared and contrasted with one another. An initial assessment was made using Principal Component Analysis (PCA) which is widely used as a means of detecting underlying patterns in complex data. Transects of particles were analysed using PCA and the effects of particle homogeneity and heterogeneity on the results explored. To sort the ore using PCA, a novel method was developed which inserted two standard pixels into particle transects prior to PCA and which were used as markers to classify pixels and particles as product or waste. The first of the methods developed split the ore into product and waste fractions but a subsequent method was developed to divide the ore into more detailed classifications such as iron sulphide-rich, copper-rich, atacamite-rich and low-grade copper. The use of a k-means clustering algorithm, a method often used in automatic sorting data analysis was investigated. A training data set was used to establish a number of cluster centres. Initial clustering was based on differences in reflectance between pixel spectra. Normalisation of the sample spectra prior to clustering led the resulting clusters to be separated on the basis of the shape of the spectra. The cluster centres were used to sort particles of a Second data set. Both reflectance-based and spectral shape-based methods exhibited strengths and weaknesses when sorting the ore and a method was developed that used aspects of each. This method found success with an accuracy of 88 % (i.e. 88% particles in the test data set were correctly identified as product or waste). A particular success was the ability of the method to separate the low-grade oxide ore from the other low-grade material and the high-grade carbonate material from other high-grade copper materials. The application of mineral mapping algorithms were also investigated as a means of sorting the ore. Despite wide usage in remote sensing and core logging applications, these algorithms have not found their way to automatic sorting application due to time constraints on measurement and analysis. Methods ranging from the simple (correlations with library spectra) to complex (continuum-removal following by peak mapping) were examined. Ultimately, noise was found to be a problem for these types of techniques, complicating the identification of mineral s. To solve these problems a large, complex library of spectrum data would be required and this, along with the complexity in mapping and comparing peaks. suggested that the data processing requirements would be too large for a time-sensitive application such as automatic sorting. Following the exploration of these methods of sorting the ore, the results were compared and contrasted. The potential usage of these methods in an industrial setting was explored including factors such as the effect of dust and moisture on the spectra observed. The effect of particle movement on spectra measurement was also considered. A methodology for NIR sorting of minerals/ore was proposed. Future work should focus on applying these methods to non-stationary particles and the development of a sensor capable of a high signal-to-noise. ratio at low measurement intervals. The methods developed during the course of this work should be trialled on other types of ore and optimised in each case and the potential use of wavelengths beyond the range used in this study investigated.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.603473  DOI: Not available
Share: