SubsurfaceBreaks v. 1.0: a supervised detection of fault-related structures on triangulated models of subsurface homoclinal interfaces

Abstract

The study presents a novel approach for fault detection on subsurface geological homoclinal interfaces (slopes) using a supervised learning algorithm and careful input variable (feature) selection. Synthetic faulted slopes are generated using Delaunay triangulation via the Computational Geometry Algorithms Library (CGAL), allowing for adjustments of parameters. We introduce 24 features, including local geometric features and neighborhood analysis, for classification. Support Vector Machine (SVM) is employed as the classification algorithm, achieving high precision and recall rates for fault-related observations. Application to real borehole data (elevations of buried stratigraphic contacts) demonstrates the effectiveness of the method in detecting fault orientations; the challenges remain with respect to distinguishing faults with opposite dip directions. The study highlights the need to address 3D fault zone complexities and their identification. Despite limitations, the proposed supervised approach offers significant advancement over clustering-based methods, showing promise in detecting faults of various orientations. Future research directions include exploring more complex geological scenarios and refining fault detection methodologies.