Successful Application of Machine Learning to Generate Petrophysically-Consistent Acoustic Slowness During Logging While Drilling

Abstract Logging While Drilling (LWD) acoustic data has many geoscience applications, including pore pressure prediction, seismic tie-in, wellbore stability, completion design, and reservoir characterization. Accurate real-time extraction of compressional slowness from LWD acoustic data remains a key challenge due to downhole noise, tool mode effect, and formation heterogeneity. Traditional processing methods require manual intervention and may yield inconsistent results. This paper introduces an automated, machine learning–based processing to estimate a petrophysically-consistent compressional slowness in real time. The proposed Random Forest based approach is trained on a comprehensive set of labeled field data, capturing a wide range of petrophysical properties, borehole conditions, and tool configurations, sourced from wells logged globally. Both high and low frequency acquisition modes are required to cover the wide range of acoustic slowness expected from a well drilled through heterogeneous sequences. However, only a single consistent slowness correctly represents the formation property. An important element of our approach is to use formation evaluation (FE) data together with coherence peaks and quality control (QC) parameters as input features. Feature selection and model optimization techniques are applied to ensure robustness and generalization across different well environments. Field deployments results demonstrate that the new auto-processing workflow achieves ~ 99% accuracy in real-time compressional slowness picks, significantly outperforming traditional methods in both speed and reliability. The tested conditions include environments where the wellbores are drilled through challenging and highly variable formation sequences. The seamless integration of this AI-driven solution into real-time operations enhances data quality, reduces interpretation cycle time, and adds confidence to early decision-making at the wellsite. This work demonstrates the practical value of machine-learning in complex drilling environments.