Novel Physics-Informed Machine Learning Approach to Estimate Safe Mud Window Using Petrophysical Logging Data

Abstract

Abstract The allowable limits of mud weights for drilling O&G wells, known as the safe mud window (SMW), play a crucial role in preventing wellbore instability issues and loss of circulation. The SMW consists of the minimum mud weight for shear failure (MWBO) and the maximum mud weight for tensile failure (MWBD), which are determined by the principal stresses of the formation, including the maximum (Shmax) and minimum (Shmin) horizontal stresses. Measuring these stresses accurately can be achieved through field tests or approximated using physics-based equations. However, obtaining the necessary in-situ geomechanical parameters for these equations, such as static Poisson's ratio and static elastic modulus, is not always feasible for all wells. Furthermore, the existing machine learning models rely on expensive and destructive tests. To address these challenges, this study investigated the feasibility of utilizing machine learning (ML) algorithms to predict these parameters in a time- and cost-effective manner. New ML-based models employing artificial neural networks (ANN) were developed to predict the SMW limits (MWBO and MWBD) using petrophysical well-log data as inputs. A comprehensive dataset consisting of field test data and petrophysical logging data was collected and extensively analyzed to train the models. The predictions generated by the developed ANN-based models exhibited a high degree of accuracy, with a mean absolute average error (MAPE) of less than 0.30% when compared to the actual output values. Thes developed models were validated using an unseen dataset, demonstrating remarkable agreement with the actual stress gradient and SMW limit values. The prediction accuracy exceeded 95%, and the MAPE was as low as 0.59%. The statistical analysis of the results confirmed the robustness of the developed equations in accurately predicting the SMW limits, provided that the logging data are available. The originality of this research lies in its ability to efficiently and affordably predict the safe mud window (SMW), thereby mitigating drilling problems such as borehole instability. The developed models provide a reliable tool for accurately determining the SMW, surpassing the conventional methods that are more time-consuming and costly.