Session
Machine Learning/Numerical Modelling, Finger Rock Room
Abstract Summary
A machine learning (ML) workflow was developed by SRK for geotechnical classification of core box photographs. The methodology involves an assessment of the degree of brokenness of the drill core, which has been defined as the Core Damage Index (CDI). Deep convolutional neural networks are trained and used to identify and map the representative rock mass character and attribute depth values to each interval. The model is trained by labelling a selection of core box photographs with a quantitative classification system, which is subsequently applied to the available core photo database. The ML workflow produces an interval table with the mapped CDI classes.
Geotechnical characterization of greenfield sites includes data from exploration drilling and dedicated geotechnical drilling. Typically site geologists log basic geotechnical parameters, including the Rock Quality Designation (RQD) and core recovery for all drillholes. A limited number of geotechnical drillholes are logged using dedicated classification systems including the Rock Mass Rating and Q systems. These datasets, of variable reliability, must be interpreted and applied to the geotechnical domains. Core photo reviews must be undertaken to understand the significance of low RQD intervals.
The CDI generates a high resolution dataset aimed at identifying the degree of fracturing, while not limited to set intervals like RQD or fracture frequency. This reduces the amount of effort required to characterize fractured domains that may only show a low RQD but can represent a range of conditions. The outputs from this classification inform and increase reliability of structural, hydrogeological, and geotechnical models.
Authors
Click here to view other presentations at this event.
