Machine-Learning Solutions for Mitigating Drilling Environment Impacts on NMR Logs in Brazilian Presalt Carbonates

Nuclear magnetic resonance (NMR) logging while drilling (LWD) has gained prominence in evaluating pre- salt carbonates in the Santos Basin, offshore southeastern Brazil, and became a key solution in porosity determination in calcite, dolomite, and silica-bearing matrix rocks. This article presents the challenges faced in well drilling environments in the Santos Basin pre- salt, such as severe lateral vibrations, uncertainties in temperature estimation, wellbore instabilities, and well enlargement by vugs (centimeters to metric) and super K zones. These issues affect NMR data acquisition, occasionally cause tool failures, and highlight how a machine-learning approach has been mitigating these impacts. To face these challenges, the geological data acquisition team in Petrobras has been applying a two-step approach. Firstly, the NMR log quality control automatization, flagging discrepancies between acquired data from other porosities logs comparison, such as density porosity (PhiDen), neutron porosity (PhiNeu), and sonic porosity (PhiSon). Thus, potential issues like instrumental deviations, measurement errors, or external interferences are real time (or quasi) signalized if differences exceed a user-defined threshold. This automated quality control enhances problem detection, significantly reducing the time required for manual data analysis and troubleshooting. Nevertheless, due to technical or operational constraints in LWD scenarios, NMR data can be insufficiently suitable, such as in case of tool failure or severe drilling environment. If so, the NMR porosity is estimated by machine-learning algorithms. These supervised models were built based on two data sets: gamma ray (GR), density (RHOB), neutron porosity (NPHI), compressional slowness (DTC) and photoelectric factor (PEF) or GR, DTC and shear slowness (DTS). With data acquisition covering these logging combinations, these models generate synthetic logs that resemble acquired NMR logs, consistently coherent to those in the Brazilian pre-salt context. Synthetic NMR logs have been used as an alternative to the choice of pressure test points and to avoid relogs and new dedicated runs due to the failure of LWD NMR tools. These automated solutions have been applied in more than forty wells, enhancing operational efficiency, reducing costs, decreasing rig time, ensuring data integrity, and supporting faster decision making. Some remarkable examples will be discussed in detail here.