SubsurfaceBreaks: A supervised detection of fault-related structures on triangulated models of subsurface slopes

We present a novel supervised learning approach for fault detection in subsurface geological slopes. Synthetic faulted slopes were generated using Delaunay triangulation via the Computational Geometry Algorithms Library (CGAL), enabling precise control over model parameters. A total of 24 features, encompassing local geometric attributes and neighborhood analyses, were introduced for classification. A Support Vector Machine (SVM) classifier was employed, achieving high precision and recall in identifying fault-related features.

Application of the method to real borehole data, specifically elevations of buried stratigraphic contacts, demonstrated its effectiveness in detecting fault orientations. However, challenges remain in distinguishing faults with opposite dip directions. The study highlights the necessity of addressing 3D fault zone complexities for more robust fault identification.

Despite these challenges, the proposed supervised approach represents a significant advancement over traditional clustering-based methods, demonstrating its potential for detecting faults across diverse orientations. Future work will focus on incorporating more complex geological scenarios and refining fault detection methodologies to improve accuracy and applicability. This work underscores the promise of machine learning in advancing fault detection in geological studies.