Employing the ANN technique to predict the limits of mud weights for minimum shear failure and induced tensile fracture: a case study in an Iraqi oilfield

Determining the minimum (Min MW) and maximum (Max MW) mud weights is essential for mitigating shear and tensile rock failures, and defining a safe operating window, which requires precise calculations of the stresses around the wellbore. However, it is frequently difficult to construct geomechanical models because of the lack and uncertain data, including direct stress measurements, core samples, and image logs. To overcome these limitations, this study used field data from southern Iraq to create artificial neural network (ANN) models that predict limits of Min MW and Max MW. Model training was conducted using datasets from one well, and validation against the Mogi–Coulomb failure criterion was conducted using datasets from another well. Strong predictive ability was shown by the ANN models, which obtained R2 values of 0.87 and 0.98 and RMSE values of 0.95 and 0.49 for Min MW and Max MW, respectively. The ANN models reduced the RMSE by 30–40% and increased prediction accuracy by roughly 15–25% in terms of R2 when compared to the conventional Mogi–Coulomb failure criterion. According to validation using caliper log data, both ANN and Mogi–Coulomb techniques successfully detected wellbore instabilities; however, the ANN models exhibit better generalization to new wells. These findings demonstrate that the ANN approach is a useful and effective method for mud weight design optimization and wellbore stability prediction in situations with limited data.