--> Characterize Millimeter-Scale Unconventional Rock Using Micron-Scale Sample Imaging and Machine Learning: Pore Properties in Eagle Ford, Marcellus and Wolfcamp

AAPG Annual Convention and Exhibition

Datapages, Inc.Print this page

Characterize Millimeter-Scale Unconventional Rock Using Micron-Scale Sample Imaging and Machine Learning: Pore Properties in Eagle Ford, Marcellus and Wolfcamp

Abstract

A thorough knowledge about rock properties is crucial for economic oil and gas exploration. Unconventional resources are known to exhibit multi-scale and highly heterogeneous pore structure. Imaging techniques are widely used to visualize and study rock properties. Due to limitations in imaging technology, less than 1% of a rock sample area can be sampled and studied at high resolution. Obviously, relevancy and representativeness of sampling area as well as obtained rock properties are often considered unreliable. We present an in-house developed technology that combines imaging techniques with machine learning to characterize multi-scale rock properties. The technology is based on the understanding that a rock consists of building blocks, i.e. fabrics, intermixed spatially at various scales. Detailed knowledge of all fabrics at its representative scale will lead to an improved characterization of the rock sample. A fabric possesses a set of properties, e.g. porosity, fraction of organic matter, pore size distribution and others. Unsupervised machine learning is used to learn about fabrics present in a sample. It also recommends optimum sub-sampling areas for smaller-scale higher resolution image acquisition and properties upscaling. The required resources in this approach are several orders of magnitude lower than acquiring mosaics of small-scale images covering a similar area at the large scale. We apply the present technology to acquire images and characterize pore properties of rock samples from Eagle Ford, Marcellus and Wolfcamp. The rock sample area is approximately 0.5 × 0.5 [mm] with a resolution of 244 [nm]. Small-scale images with an area of approximately 30 ×18 [micron] with a resolution of 10 [nm] are used to characterize pores within fabrics. The upscaled pore properties and fractions of organic matter are compared with that derived from small-scale mosaic covering a similar area. The comparison shows a very good agreement confirming the accuracy and reliability of the present technology. The present work demonstrates a novel approach to characterize multi-scale rock properties using machine learning technique. This is a step towards bridging knowledge from pore scale (where hydrocarbon lives) to the reservoir scale (where production takes place) for an economic oil and gas exploration.