Abstract
Accurate estimation of rock elastic and failure parameters plays a vital role in petroleum, civil and geotechnical engineering applications. During drilling operations, continuous logs of rock elastic and failure parameters are considered very helpful in optimizing geomechanical earth models. Commonly, rock elastic and failure parameters are estimated using well logs and empirical correlations. These are calibrated with rock mechanics laboratory experiments conducted on core samples. However, since these samples are expensive to get and time-consuming to test, artificial intelligence (AI) models based on available petrophysical well logs such as bulk density, compressional wave and shear wave travel times are utilized to predict the static Young's modulus (E-static) and the unconfined compressive strength (UCS) - with an emphasis on carbonate rocks. We present two AI techniques in this study: an artificial neural network (ANN) and an adaptive neuro-fuzzy inference system (ANFIS). The dataset used in this study contains 120 data points obtained from a Middle Eastern carbonate reservoir from which we develop an empirically correlated ANN model to predict E-static and an ANFIS model to predict the UCS. A comparison between the UCS, predicted by the proposed ANFIS model, and the published correlations show that the ANFIS model predicted the UCS with less error and with a high coefficient of determination. The error obtained from the ANFIS model was 4.5%, while other correlations resulted in up to 30% error on a published dataset. On the basis of the results obtained, we can say that the developed models will help geomechanical engineers to predict E-static and the UCS using well logs without the need to measure them in the laboratory.