INTEGRATED MULTI-PHYSICS WORKFLOW FOR AUTOMATIC ROCK CLASSIFICATION AND FORMATION EVALUATION USING MULTI-SCALE IMAGE ANALYSIS AND CONVENTIONAL WELL LOGS

摘要

Conventional well-log-based rock classification often overlooks rock fabric features (spatial distribution of solid and fluid rock components), which makes it not comparable against geologic facies, especially in formations with complex rock fabric. This challenge is usually addressed by identification of geological facies from core description and their integration with measured petrophysical properties. However, manual identification of geological facies using core data is a tedious and time-consuming process. In this paper, we propose an automatic workflow for joint interpretation of conventional well logs, computed tomography (CT) scan/core images, and routine core analysis (RCA) data for simultaneously optimizing rock classification and formation evaluation. First, we perform conventional well-log interpretation to obtain petrophysical properties of the evaluated depth intervals. Subsequently, we automatically extract rock-fabric related features derived from core photos and core CT-Scan images. Then, we use a clustering algorithm to obtain rock classes from the extracted rock-fabric features. We optimize the number of rock classes by iteratively increasing the number of rock classes from an initially assumed number until a permeability-based cost function converges below a predefined threshold. The proposed workflow can help expedite the process of geological facies classification by providing an overview of different lithologies and an overall stacking pattern. We successfully applied the proposed workflow to a sedimentary sequence with vertically variable rock fabric and lithology. Dual energy acquired core CT-Scan images were available along with core photos, RCA data, and conventional well logs. Image-based integrated rock classes were in agreement with the lithologies encountered in the evaluated depth interval. Class-byclass permeability models improved permeability estimates by 78% (decrease in mean relative error) in comparison to formation-by-formation permeability estimates. Furthermore, rock classes were consistently propagated to another well where core and CT-Scan images were not employed for rock classification. The detected rock classes were in agreement with lithofacies obtained from core description. Permeability estimates were also in good agreement with available RCA data.