ABSTRACT

Drilled cuttings from wells are useful sources of information to describe the subsurface geological properties for formation evaluation. However, the process of cuttings analysis and description is known to be labor and expert intensive and to lack consistency. In our initial work, we proposed a workflow based on object-based image analysis (OBIA) to analyze digital cuttings images, which is driven by geology and enabled by machine learning and computer vision techniques. The key challenge in lithology estimation from digital cuttings images is the visual variety of rocks; it is known that rocks from different types of lithology might look very similar or almost identical on the standard optical image. In this paper, we revisit our initial workflow and propose lithology OBIA (LiOBIA), which focuses on the lithology estimation computational pipeline composed of two steps. In the first step, we use an advanced data-driven technique to extract color and texture features from a cutting instance images. In the second step, we use k-nearest neighbor classification in the feature space (i.e., we retrieve similar cuttings from the reference dataset and estimate the lithology based on the retrieved cuttings). A strength of this workflow is its flexibility. The workflow design allows for multiple QC methods, extensions, and add-ons to the base workflow by incorporating any available data and information, whenever possible, such as integration of images under the UV light, lithology type filtering, and local cuttings datasets, to better estimate the lithology type. We evaluate the LiOBIA lithology estimation using our cuttings library dataset including five main lithology types. Our 2D manifold analysis confirms that the visually similar cuttings are close to each other in the high-dimensional feature space. The classification result showed an accuracy of more than 90%. We also applied LiOBIA to two wells in Australia. The cuttings property logs including lithology type are generated and show a good trend match with independently acquired expert description and downhole well logs. We demonstrate successful cases and discuss the limitations and potential improvements to our workflow.

This content is only available via PDF.
You can access this article if you purchase or spend a download.