Sedimentary Analysis Via Automatic Image Segmentation and Clustering with the LWD OBM Resistivity Image: A Case Study from Gulf of Mexico

Abstract

Abstract Microfacies analysis is the first step for depositional environment interpretation and sand body prediction. Textural details from borehole images are building blocks for facies analysis, representing different paleo sediimentation conditions. Associated workflows have been applied on high resolution borehole images by geologists and log analysts manually. Automation via machine learning solutions provides an opportunity to improve the working efficiency and accuracy. Such an approach has given satisfactory results with post-drilling wireline images. In this paper, the improved workflow for sedimentary analysis was applied and validated with a logging-while-drilling (LWD) resistivity imager in oil-based mud environment (OBM). The OBM LWD resistivity image in oil-based mud provides 72 data points at single depth from 4 different frequencies of electromagnetic measurements with a patented processing. The non-gap resistivity image gives more confident texture characterization. The continuous histogram and correlogram derived from image data were used for image segmentation. In each image segmentation, multiple vector properties were extracted from image data representing different texture features including adoptive variogram horizontally. Agglomerative clustering was selected for its stability and repeatability. The internally built dendrogram allows to automatically determine the number of clusters by finding a stable distance between the clusters’ hierarchy branches. In addition to the features extracted from image data, optional petrophysical logs with variable weights may be fed to the algorithm for a better classification. A case study from Gulf of Mexico is being used to demonstrate this workflow with Hi-Res LWD image. More than 10 different sedimentary geometries were classified automatically from image and petrophysical logs. The microfacies were named manually from sedimentary geometries with the related geological concept accordingly. The fluvial channel and delta sedimentary environment were interpretated finally from microfacies association. The interpretation results were compared and validated with published dips-based solution as well. This is the first time for the automatic borehole image segmentation with LWD OBM images. The working efficiency was improved a lot through this workflow and the accuracy of microfacies interpretation was guaranteed by machine learning solution.