Extracting Physical Properties from Thin Section: Another Neural Network Contribution in Rock Physics

Thin section observation is generally carried out to study the petrographic properties of a rock e.g; pores or grain shapes, sizes, distribution, pore connectivity, or eventually 2D optical porosity of the rock. In this paper wider use of thin section analysis is explored to tell more about the physical properties of rock. This work is purely computeraided theoretical modeling, in which from thin section image, the effective values of elastic, electrical, and thermal properties are drawn based on the existing rock physics models. Stained and pore impregnated carbonate thin section which is partially saturated by heavy oil is used in this study. The pattern recognition algorithm of artificial neural network is employed as the rock constituent classfier in this work. It is programmed to replace the RGB pixel value of each constituent with known elastic, electric, and thermal properties value. In other word, it generates the distribution (map) of acoustic velocity and density, electrical conductivity and dielectric permittivity, and thermal conductivity in the thin section image. Values assigned to generate the map are obtained from known physical constants of minerals obtained in rock physics literature. From the generated map, pixel manipulation is performed to compute the fraction of the constituents and other properties of the rock (e.g. porosity, mineral composition). These results are then used to estimate the effective value of those physical properties. The estimation is performed by using theoretical rock physics models e.g. Voigt-Reuss-Hill model, Hashin-Sthrikman bound model, Kuster-Toksoz model, and Clausius-mossotti model. Consequently, the result brings a wider scientific view of the thin section image which now becomes both geologically and physically meaningfull. The neural network yields the nearly perfect map depending on the variation of the image color pattern. The effective value of each physical property depends on the accuracy of fraction calculation. The estimated effective value might be different from the core scale measured value since other carbonate minerals (aragonite and siderite) are assumed to be absent in the samples due difficulties in distinguishing them. Also, it depends on the staining technique in the thin sectioning process which in turn affects the captured image quality. Using this method one may obtain a brief estimate of physical properties in a rock from thin section before conducting laboratory measurement in core scale. If the rock is sufficiently homogeneous and fairly isotropic, the estimated results might be close to the laboratory measurements. In application, the generated acoustic map is tested to dynamically simulate the acoustic wave propagation and to measure the corresponding effective velocity. The effective velocity can be simultaneously estimated and measured in the thin sections of carbonate. Finally, it introduces the so called “thin section rock physics” in which physical properties modeling and measurement are undertaken in thin sections of rocks.