Machine Learning to Estimate the Brittleness Index of Organic Shale: Case Study of Marcellus Shale Formation

A rock’s brittleness index (BI) aids in pinpointing the optimal formation for hydraulic fracturing in shale reservoirs. The two main methods for estimating BI in the petroleum industry are mineral composition and elastic parameters. Nevertheless, not all wells can use these methods to determine BI because of the lack of mineralogical composition and shear wave slowness data. This paper introduces a machine learning (ML) method for forecasting BI utilizing easily accessible Marcellus shale drilling cuttings. This study attempted to develop an approach that uses the bulk elemental composition of crushed samples measured by X-ray fluorescence (XRF) to predict the BI. The study first used the elemental composition of XRF data to infer mineralogy, and then mineral composition to predict the BI of the target formation along the wellbore. In comparing the neural network algorithm model, random forest algorithm model, and ensemble model, the prediction effect of the ensemble model is more accurate than the single model, and the BI depth curve obtained conforms to the previous understanding of mineralogy and mechanics. The study revealed that the difference between ML prediction results and the effectiveness of standardized calculations is relatively small. However, ML demonstrated acceptable prediction performance with higher efficiency, and the prediction accuracy can reach 83.3%. The method proposed can be widely used to establish the BI of unconventional formations by utilizing XRF tests on drill cutting samples, which is a low-cost, reliable, and fast method. Also, drilling cuttings are often widely available through the whole lateral interval.