Machine Learning Assisting in Estimating Brittleness Index of Middle Bakken Formation from Drilling Cuttings

Conventional brittleness index (BI) evaluation requires laboratory measurement on intact core plugs. These plugs are expensive to obtain and rarely available for horizontal well sections. This study sought to develop a machine learning (ML)-assisted approach using the bulk elemental composition of crushed samples measured by x-ray fluorescence (XRF) to estimate the BI of the target formation. This study consisted of two phases: the first phase used the elemental composition of XRF data to infer mineralogy, and the second phase used the mineral composition to predict the BI of the formation along the wellbore. The calculation and validation in two phases were based on previously generated data sets of multiple (>50) Middle Bakken samples tested using XRF, x-ray diffraction (XRD), and rock mechanics. In the first phase, supervised ML, a neural network algorithm specifically was employed, and the results were compared to XRD data to determine how reliable the ML method could predict mineralogy of reservoir rock. The result demonstrated an acceptable prediction performance with a high efficiency, which will evolve as the database is enriched. In the second phase, the BI was predicted based on inferred mineral composition, from which the calculated BI was comparable to the measured values from previous studies, and the results were also validated with experimentally measured elastic properties in this study. The method proposed in this study can be widely used to establish a BI of unconventional formations by utilizing XRF tests on drill cutting samples and forgoing multiple other test methods with core plugs. This approach only requires the XRF measurements of drill cuttings, which is highly advantageous as XRF testing is a low-cost, reliable, and fast method. Also, cuttings are often widely available through a whole lateral interval. A timely and efficient ML-assisted evaluation of geomechanical information significantly contributes to the design of effective well stimulation and the prediction of a fracture network induced by hydraulic fracturing, and it is especially useful if conventional geomechanical tests that require core samples cannot be conducted.