Authors: Tilley, E; Yang, B; Otto, SA; Sinaga, Y; Bewick, R

Open access courtesy of:


Cite As:
Tilley, E, Yang, B, Otto, SA, Sinaga, Y & Bewick, R 2022, 'Objectivity through mineralogy: application of spectrographic data and machine learning for improved geotechnical characterisation of porphyries', in Y Potvin (ed.), Caving 2022: Fifth International Conference on Block and Sublevel Caving, Australian Centre for Geomechanics, Perth, pp. 1293-1302,

Download citation as:   ris   bibtex   endnote   text   Zotero

A primary objective of geotechnical characterisation is the progressive reduction of uncertainty in our understanding of rock mass character to support the development of reliable mine design and operation. A critical part of this process is the investigation of the sources of natural variability in rock strength parameters. The development and increasing use of spectrographic core imaging as part of resource drilling programs and accessibility to machine learning (ML) tools for data processing presents an opportunity for improved understanding of the sources of variability by allowing for more robust interrogation of correlations between mineralogy and associated influences on rock strength. Ongoing studies for an underground porphyry project in Indonesia provided an opportunity to investigate the use of a Random Forest (RF) algorithm to develop a high density, downhole record of Point Load Index (PLI) strength. This RF tool combines the output of multiple decision trees to reach a single result based on two data sources: 1) comprehensive mineral composition from hyperspectral core imaging, and 2) rock hardness from high density dynamic rebound index testing. Through this approach, it was possible to augment the spatial coverage of available strength index testing and provide sufficient data density to support the development of a block model of rock strength in three-dimensional space. In this paper, the authors outline the approach taken to develop and train the RF algorithm, limitations in the current methodology, the results and associated confidence in the outputs, and learnings to be considered in future applications.

Keywords: machine learning, rock mass characterisation, rock strength, spectrographic data

ASTM 2017, Standard Test Method for Determination of the Point Load Strength Index of Rock and Application to Rock Strength Classifications (ASTM D5731–16), ASTM International, West Conshohocken.
Brown, ET (ed.) 1981, Rock characterization, testing & monitoring — ISRM suggested methods, Pergamon Press, Oxford.
Louppe, G 2014, ‘Understanding random forests: From theory to practice’, arXiv preprint arXiv:1407.7502.
Pedregosa, F, Varoquaux, G, Gramfort, A, Michel, V, Thirion, B, Grisel, O, … Duchesnay, E 2011, ‘Scikit-learn: Machine Learning in Python’, JMLR 12, pp. 2825–2830.
Ridley, J 2013, Ore Deposit Geology, Cambridge University Press, New York. p. 114.

© Copyright 2022, Australian Centre for Geomechanics (ACG), The University of Western Australia. All rights reserved.
Please direct any queries or error reports to