Machine Learning Powered Approach for Sonic Slowness Extraction and Interpretation

Dipole sonic data is vital for estimating formation shear slowness and identifying anisotropy in subsurface formations. Traditionally, the classification of anisotropy types—such as intrinsic, fracture-based, or stress-induced—relied on manual visual inspection of dispersion analysis and was disconnected from lithological interpretation. This paper utilizes a machine learning (ML)-enabled workflow that automates sonic slowness extraction and anisotropy classification, drastically improving efficiency and consistency. Using an established neural network classifier, cross-dipole sonic data was categorized into dispersion-based classes: crossover, noncrossover, high-frequency split, and parallel separation. To enhance accuracy, lithological context was added via elemental spectroscopy interpreted through the SpectroLith technique, enabling clay volume-based lithology identification. This integration allowed further classification of anisotropy into VTI (vertical transverse isotropy), stress-induced, and fracture-based types using clay volume thresholds. Applied to a Newark Basin stratigraphic interval, this novel approach enabled rapid and geologically informed anisotropy interpretation, cutting analysis time by over 90%. The results directly informed geomechanical modeling and stress testing, supporting real-time decisions and reducing rig wait times. This paper presents the full methodology, including quality control steps, demonstrating how ML-driven interpretation can streamline workflows and improve formation evaluation for applications such as carbon storage suitability.