Oil & Gas Industry Use Cases

Reservoir Rock & Mineral Segmentation & Rock Type Classification (via microscopy imaging & CT)

Reservoir rock core plugs and thin sections are imaged via numerous 2D & 3D microscopic and CT imaging modalities: SEM, micro-CT, medical CT (full core), FIB-SEM, and polarized light microscopy.

An important use case is the accurate pixel-by-pixel (or voxel-by-voxel) 2D or 3D segmentation of the rock image, classifying each pixel by some known type such as: empty pore space, pore filled with kerogen-oil (especially in shale), and any number of different base rock mineral types. Deep learning can be used (based on a relatively small amount of expert-labelled ("ground truth") data to automatically perform accurate segmentation.  Mineral identification in some cases can be accurately performed based on texture and other present information, without the need for costly, time consuming EDS mapping.

Such imaging data can be combined with other data (such as NMR, XRD, etc.) to determine rock types and/or probability of yield for resource extraction, by applying machine learning models.

Well-log Analysis

Machine learning can very effectively be used with well log data and supervised learning to make predictions (rock types, likely yield, flow permeability, etc..) based on past examples used in training. In some cases missing instrumentation can be synthetically generated from past data even when missing from a given log.

Seismic Imaging Resolution Enhancement

Well known for its challenging nature, deep learning has been used with labelled data to better identify hard-to-discriminate critical features in seismic imaging.

Drilling-rate Optimization

Machine learning can be used based on historical data to specify optimum drilling parameters in a specific case. Even small optimizations in efficiency of course lead to large ROI.