Machine-vision-based automation of geotechnical core logging: rock quality designation workflow and comparative assessment at Dund Sukhait

Authors: Jaroling, J; Sarrazin, C; Mulshaw, H; Jackson, A; Hanžl , P; Tumurkhuu, G Download Paper Open access courtesy of:

DOI https://doi.org/10.36487/ACG_repo/2645_74

Cite As:
Jaroling, J, Sarrazin, C, Mulshaw, H, Jackson, A, Hanžl , P & Tumurkhuu, G 2026, 'Machine-vision-based automation of geotechnical core logging: rock quality designation workflow and comparative assessment at Dund Sukhait ', in A van As, D Cumming-Potvin & J Wesseloo (eds), Caving 2026: Proceedings of the Sixth International Conference on Block and Sublevel Caving, Australian Centre for Geomechanics, Perth, pp. 1-10, https://doi.org/10.36487/ACG_repo/2645_74



Abstract:
Recent advances in artificial intelligence (AI) and machine learning (ML) are creating new opportunities to automate structural geological data collection in mining through machine-vision-based drillcore logging. These technologies offer improved speed, repeatability, and consistency compared to conventional manual logging. However, industry adoption remains limited, partly due to constraints in legacy data capture systems and inconsistencies in historical geological datasets. This paper presents the application of an AI-based image analysis platform to structurally interpret drillcore from the Dund Sukhait project in Mongolia, a structurally complex deposit characterised by faulted and variably fractured ground conditions. Four drill holes were processed using automated structural logging workflows, and the resulting datasets were compared against manually logged records. The comparison focused on rock quality designation (RQD), fracture frequency (FF), and classification performance metrics. Results demonstrate variable statistical correlation between AI-derived outputs and manual logging; however, automated interpretations remain geologically coherent and internally consistent at the domain scale. Interval-scale discrepancies occur both in highly fractured (low-RQD) zones and locally within competent rock, reflecting break-type discrimination and interval segmentation effects. Automated processing achieved an average logging rate of approximately one hour per kilometre of drillcore processed, representing a substantial improvement in throughput relative to conventional logging. While the manual datasets reflected a high level of professional expertise, reliance on spreadsheet-based data capture introduces vulnerabilities that limit reproducibility and integration with advanced analytical workflows. The findings indicate that both interval-scale break discrimination challenges and traditional data management frameworks influence AI evaluation metrics. For projects such as Dund Sukhait, the adoption of structured databases and purpose-built digital logging systems with integrated quality assurance and quality control (QA/QC), combined with continued refinement of break-type classification algorithms, is essential to enable scalable and reliable ML-assisted structural interpretation.

Keywords: machine learning, structural geology, drillcore logging, geotechnical data, rock quality designation

References: