Maximise Orebody Value Through the Automation of Resource Model Development Using Machine Learning
By Samuel Oliver and David Willingham, MathWorks
Although a resource model is central to the mineral resource value estimation process (Glacken & Snowden, 2001), creating it is a labour intensive task and in the end presents a limited representation of the actual orebody. Producing this resource model requires inputs from geology, mining, metallurgical, and commercial disciplines. It requires thousands of samples from hundreds of drill holes to be verified, grouped in geological domain, interpolated, and then valued. Even after all this effort, a model is only an estimation; further significant effort has gone into quantifying uncertainty in the model, and subsequently risk, around any value estimates derived from the model. Further risks to model quality are introduced through poorly conditioned data or incorrect assumptions.
This paper proposes the application of machine learning to automate the resource model development. Machine learning is applied to the traditionally manual tasks of geological formation, domain identification, and validation of the block model mineralogy. Through automation, the resource estimation process can be accelerated, allowing more drill holes or a larger resource body to be processed in a given timeframe, and allowing the process to be more agile to changes in input data and assumptions. A case study based on drill hole data from a Western Australian iron ore deposit (Government of Western Australia, Department of Mines and Petroleum, 2015) is used to demonstrate the application of machine learning in this process.
This paper was presented at the Third International Geometallurgy Conference 2016.
Published 2016 - 80845v00