Real-Time Machine-Learning Workflow for Log Reconstruction Reduces NPT and Unlocks Additional Reserves in a Niger Delta Brownfield

Abstract Real-time machine-learning workflows for log reconstruction can significantly reduce non-productive time (NPT) and unlock additional reserves in brownfield openhole operations, especially where drilling costs are high, and data quality is critical for accurately characterizing and targeting reservoir sweet spots. This case study from the "M" brownfield in the Niger Delta describes the deployment of a real-time machine-learning (ML) workflow during the drilling of a critical water injector that experienced sequential neutron and density tool failures, likely due to electronics malfunction, while drilling through primary and secondary target sands under depleted, wellbore-unstable conditions. Under conventional workflows, these failures would require at least two additional logging runs, prolonging open hole exposure, increasing NPT, and elevating the risk of well abandonment or sidetracks. To mitigate these risks, the integrated drilling and subsurface team implemented a supervised neural-network approach based on multi-layer perceptron technology and non-linear regression techniques. The workflow leveraged log data from the drilled interval and a set of nearby offset wells for training and validation, enabling robust prediction of unrecorded neutron and density responses across intervals affected by tool failures. The ML model was further extended to infer key petrophysical properties, including porosity (Φ) and permeability (K), thereby reconstructing a coherent petrophysical dataset in near real time despite acquisition gaps, poor hole conditions, and environmental noise. The ML-driven reconstruction eliminated two tool change-out and logging rerun trips, minimizing NPT while preserving business continuity. It also enabled full characterization of producing reservoirs where oil was encountered deeper than the pre-drill oil–water contact, providing additional reserves and supporting field-redevelopment opportunities. Post-drill pressure and production data confirmed the reliability of the reconstructed log and petrophysical responses. Key lessons learned include: (1) ML-based reconstruction can effectively compensate for logging tool failures in high-cost, mature environments; (2) integrating multiwell offset datasets is essential for robust model training, generalization, and uncertainty reduction; and (3) close collaboration between drilling, petrophysics, and data-science teams accelerates model deployment and operational uptake. Best practices emerging from this work advocate for embedding scalable, field-ready ML workflows into standard brownfield operating procedures as contingency options. These practices help safeguard data integrity, reduce unplanned NPT exposure, and enable safer, more informed decision-making during critical well construction and reservoir management activities.