A stochastic wellbore stability framework is presented where uncertainties in wellbore stability models are derived through a Bayesian inversion of borehole geomechanical observations. We achieve this by treating all wellbore stability inputs as uncertain, allowing us to estimate the probabilities of different borehole failure mechanisms given a set of borehole measurements. By comparing predictions over the model space with specific observations along a well, we generate stochastic safe drilling mud windows by only propagating the errors of the subset of model space samples that satisfy a minimum posterior probability with regards to the observed data. We illustrate the stochastic framework with a case study from a data-rich wellbore in the Vaca Muerta shale and show the impact of different measurements and their uncertainties, the choice of the forward model hypothesis-using isotropic and anisotropic elasticity- and the choice of failure criterion on the stress and critical mud weight posterior distributions. We also incorporate the effects of stress regime constraints and number of observations on the predicted drilling mud weight window uncertainties. Our work demonstrates how a rigorous inclusion of observations in the stress inversion process allows a more representative uncertainty quantification workflow for drilling risk assessments.
The prediction of safe drilling mud windows is an essential component of the well planning process. As operators continue to target deeper and more challenging environments, the prediction of in-situ geomechanical conditions will require streamlining the steps involved in assessing how to efficiently drill wells taking into account both economic and environmental considerations. The level of rigor in which geomechanics is incorporated during the early stages of well design is commonly constrained, not only by the availability of offset well information and subsurface measurements, but also by the pace at which teams can perform the necessary analyses required for a wellbore stability prediction.