News | Physics rocks! Rock physics and machine learning pave a way for future exploration
The Platform’s first patent has been granted, a new rock classification technique to help aid resource exploration.
Read on to find out more.
Figure 1. An example of processed data output from using the automated rock characterisation method Roman has developed.
Multimillion-dollar drilling programs are an essential part of modern resource exploration. Huge drills bore down into the earth – 1,800 m is a typical depth in oil and gas exploration – to sample the rocks present. Sensors that detect the properties and composition of the rocks are attached to the drills. These sensors quickly and effectively collect large amounts of data about the rock types, providing essential information for exploration geologists. However, there is a problem: analysing the data is a considerably slower process than collecting the data, causing a bottleneck in the information gathering process. Furthermore, the modelling frameworks used to process the data
are often inefficient and can be affected by human bias,
and unless the data is processed it cannot be used.
Roman Beloborodov, a Deep Earth Imaging FSP postdoc, in collaboration with colleagues from CSIRO Energy and Mineral Resources, is researching objective automated techniques to process and characterise drill core rock data effectively and
rapidly. He has developed a method to automate rock characterisation using
a machine learning algorithm combined with refined rock physics models.
The results are consistent with expert (human) geological interpretations of the
rocks and can be obtained in a fraction of the time compared with conventional manual approaches.
The method has been patented and its adoption by industry will allow for rock physics information to be more easily included in exploration, offering the clear potential to increase resource exploration success.