Development of a geotechnical rock mass model using geochemical data and machine learning

In mining, the development of a realistic and well-informed engineering geological model can be challenging. Mining projects typically present with sparse geotechnical data and variable ground conditions, resulting in challenges with the interpretation of geotechnical unit boundaries, and the development of a three-dimensional model. This paper presents a case study where geochemical data collected for orebody knowledge definition was used to develop a rock mass model to inform pit slope design. During model development, borehole domaining assisted by machine learning was trialed to improve model reliability and enhance efficiency. This paper has demonstrated that an automated approach to rock mass classification using geochemical data is an effective and useful method which can be incorporated into the geotechnical model development process.

The methodology adopted for rock mass classification and model development is summarised below:

• Rock mass classification of geotechnical boreholes based on logging data and core photographs.
• Selection of ‘twin’ reverse circulation (RC) boreholes (adjacent to geotechnical boreholes).
• Comparison of downhole geochemical data against the logged rock mass unit intervals.
• Identification of geochemical signatures associated with each rock mass unit.
• Trial of machine learning methods to identify and classify rock mass units using geochemical data.
• Apply the results of the machine learning to a wider dataset with the objective of producing a 3D rock mass model.