Transforming the Mining Value Chain
T2dA INTERPRETING STRUCTURAL AND GEOCHEMICAL PATTERNS USING MACHINE LEARNING
LEADERS: | |||
Matthew Cracknell, Anya Reading | |||
STUDENT: | |||
Shawn Hood (graduated) | |||
COLLABORATOR: | |||
Michael Gazley | CSIRO | ||
PROJECT SUMMARY
2019
2019 Shawn Hood submitted his PhD thesis titled ‘Machine-assisted modelling of lithology and metasomatism’. Shawn’s ground-breaking research culminated in the publication of two peer-reviewed papers. A paper was published in Chemical Geology with the support of Gold Fields Australasia, which combined the use of geological domain expertise, machine learning and 3D visualisation to model fluid flow pathways in Orogenic gold systems by tracing element depletion and enrichment. A second paper, published in the new journal Applied Computing and Geosciences, provides a practical guide to getting the most out of geochemical, geophysical and remote sensing data for automated mapping of basement geology in difficult to access and challenging environments. In his approach, Shawn demonstrated the important role experienced geologists play in computer-assisted geological modelling. Shawn has achieved his goal to develop practical advice and reproducible workflows that fit seamlessly into existing industry methodology for targeting and defining ore bodies. Well done Shawn! In-between publishing research and submitting his thesis Shawn also presented at PACRIM 2019 and co-hosted two conference-based workshops aimed at introducing machine learning to geologists. These have been well received and will be instrumental in fostering the rapid uptake of data science and machine learning approaches in the field of economic geology.
2018
This PhD project aims to develop novel approaches for interpreting large data sets of metalliferous-ore exploration and mining data. To achieve this, computational methods are being used to represent and interpret geological patterns in three dimensions. More specifically, machine learning methods are being developed and tested for the statistical inference of structural and geochemical patterns around hydrothermal ore deposits. This research encompasses three main aims:
- Investigation of metasomatic alteration using clustering and classification of geochemical data to construct models of element migration.
- Creatiion of models of permeability networks in complex structural environments using automated statistical inference of structural geology data.
- Creation of three-dimensional ore deposit prospectivity models by combining statistical geochemical and structural models.
These research activities involve automated workflows developed using open source software and aim to rapidly generate repeatable and objective three-dimensional models of geologically relevant features with input from geoscientists. A chief goal is to produce practical workflows that fit seamlessly into existing industry methodology for defining ore bodies, especially in brownfields environments.
During 2018, PhD student Shawn Hood published results in the Journal of Geochemical Exploration outlining a novel approach for the identification of protolith and altered equivalent lithologies. This work forms the basis for robust statistical estimation of element mobility in hydrothermally controlled ore deposits, research that Shawn Hood summarised in a presentation at RFG2018, Vancouver, Canada. Other outputs from this project include numerous reports to industry partners such as Saracen Mining Ltd and Gold Fields Australasia.
2017
This PhD project aims to develop novel approaches for interpreting large datasets of metalliferous-ore exploration and mining data. To achieve this, computational methods are being used to represent and interpret geological patterns in three dimensions. More specifically, Machine Learning methods are being developed and tested for the statistical inference of structural and geochemical patterns around hydrothermal ore deposits. This research encompasses three main aims:
- Investigating metasomatic alteration using clustering and classification of geochemical data to construct models of element migration
- Creating models of permeability networks in complex structural environments using automated statistical inference of structural geology data
- Creation of 3D ore deposit prospectivity models by combining statistical geochemical and structural models.
These research thrusts involve automated workflows developed using open source software, and aim to rapidly generate repeatable and objective three-dimensional models of geologically relevant features with input from geoscientists. A chief goal is to produce practical workflows that fit seamlessly into existing industry methodology for defining ore bodies, especially in brownfields environments. During 2017, preliminary results of these efforts were given as oral presentations at conferences in New Zealand and China, and supported by peer-reviewed extended abstracts. An initial manuscript is currently under review by the Journal of Geochemical Exploration.
2016
This PhD project aims to develop and combine machine learning algorithms and automated workflows to process and interpret patterns found in 3D structural and geochemical data. Automated workflows are being developed using open source software that aim to rapidly generate repeatable and objective 3D models of geologically relevant features with input from domain experts. These workflows will fit seamlessly into existing industry methodology for defining ore bodies in brownfields environments. This research supports positive exploration and ore extraction outcomes in complex geological structural settings, such as those found in orogenic gold deposits.
During the year, workflows that implement both unsupervised clustering and supervised classification algorithms have been developed to identify protolith groups and their altered counterparts from whole rock geochemical analyses. Outputs are then used in automated mass balance calculations that calculate element relative enrichment or depletion during hydrothermal metasomatism. The results can be plotted in 3D space to visualise fluid flow networks, which provide the opportunity to model structural domains that control ore deposit formation.