An Image-Based Artificial Intelligence Approach for the Determination of Analog Petrophysical Rock Properties From Drill Cuttings

A thin-section (TS) image-based artificial intelligence (AI) model, leveraging an associated database of 100 conventional core samples, has been developed and validated for the determination of analog petrophysical properties when conventional core is unavailable. The AI model continues to be updated, resulting in an ongoing learning process. In this study, we focus on applying this technology to synthetic drill cuttings and present gained insight into additional factors that may also impact the AI-based analog drill cuttings results. A subset of 20 samples, consisting of multiple lithologies of conventional core-based synthetic cuttings, were evaluated in three size classes of 5 mm, 4 mm, and down to 2 mm (60 samples total) in order to compare analog petrophysical properties (porosity, Klinkenberg permeability, matrix density, electrical properties, and capillary pressure) to their physically measured properties. This was accomplished in the following steps: (1) Each synthetic cutting sample to be evaluated was created by disaggregating a conventional core plug, whose data were used in the creation of the AI model and sieving the results into the three size classes noted above. (2) A TS was then prepared for each of the sample’s three size classes. (3) The thin sections (TSs) were subsequently scanned in high resolution (0.44 μm/pixel), and (4) each scanned image was submitted to the AI model for analysis, and the petrophysical data sets associated with each AI-based analog match were returned from the database for evaluation and comparison to the physically measured core data. Results of the AI analysis indicate that rock pore structure plays a major role in the analog synthetic cuttings’ results. Analog petrophysical properties of the clastic synthetic cuttings reasonably matched the physically measured properties at a frequency of 85%, while the carbonate synthetic cuttings matched at a low frequency of 38%. Analysis of the three size fractions—5, 4, and 2 mm—found that the size of the cuttings, and thus the amount of sample heterogeneity available for analysis, was less of an influencing factor in the results. This is good news for drill cutting applications. This finding reflects well on the resolution (0.44 μm/pixel) of the TS images used in the model, though analysis of cuttings less than 2 mm in size should be further evaluated to find the effective lower limit on the effect of size. The above observations are being evaluated further, including the potential loss of heterogeneity during the process of creating cutting samples.