Facies and permeability prediction based on analysis of core images

More Info
expand_more

Abstract

The present standards for core interpretation do not contain the acquisition of high resolution images from core slabs; images are taken on a very low resolution under a poor light source for administrative purposes only. The advantages of taking high resolution images and subsequent analysis of these images could be substantial and are investigated in this project. Besides the possible advantages image analysis could have, these images provide a safe way to store core information, as they are not prone to deterioration over time, which cores themselves are. Obtaining a high resolution description of facies and permeability by means of image analysis is a promising new technique, which could ease the operation of core analysis. This technique is relatively new because of the trend of increasing resolution of digital cameras and increasing processing power of computers, which make it possible to obtain high resolution images and process them without losing detail. In this project a routine is developed to analyze the images and the routines accuracy is compared to the present day standards of core interpretation. The proposed routine in this project first segments the core on a centimeter scale parallel to bedding, which is performed by a correlation scheme. The segments are subsequently subjected to an image analysis algorithm. Image analysis was based on RGBD color data and its auto-covariance properties, to enable the mapping of color and texture of the core. The results of this image analysis are used to classify the core based on lithology and grain size and produce a permeability model of the core. To enhance separation between facies in terms of the RGBD color data and Auto-Covariance properties, the data undergoes a Centered Log Ratio Transformation resulting in a continuous data space. The data subsequently undergoes a Principal Component Analysis to detect the properties that are potentially informative about the facies and permeability. Initially classification between sandstone and other lithologies was performed on the log-transformed RGBD color data, by means of a quadratic decision boundary. Subsequently analysis of the Auto-Covariance properties was performed to extract a permeability model of sandstone, which was calibrated with plug data. The resulting classification of lithology showed to be accurate for 84 % of the segments, where the largest misinterpretation occurred between very fine sandstone, siltstone and mudstone. All but the finest sandstones grain size classes showed an accuracy of classification above 95 %. Grain size was classified into the correct class or a similar class in terms of permeability for 55 % of the fine to medium grained sandstone. For mudstone, siltstone and coarse sandstone this percentage ranged between 93 % and 100%. The root mean squared error of the permeability model was an order of magnitude. This error is 30 % larger than the root mean squared error of the null model, which is a model that averages permeability over the facies as interpreted by the geologist. These results imply that image analysis could potentially be a good source of information and especially when combined with other reliable methods. The areas where image analysis is prone to misclassification could be classified by other reliable methods; Misclassification between mudstone and sandstone could easily be extracted with a gamma ray log, for example. The resulting map of grain size, lithology and permeability could aid the geologist during his core analysis. The initial estimation of the core’s characteristics is digitalized by the image analysis routine, reducing the job of the geologist to verifying the results and adjusting them where necessary.