Identification and characterization of geological discontinuities is fundamental to understand rock mass behavior. Traditionally, structural information is gathered through field mapping of geological features, remote sensing, as well as core and borehole logging. Manual mapping is labor-intensive, requires significant expertise, and is subjective. Different experts may interpret and classify discontinuities differently, leading to inconsistencies in structural datasets. Even the same expert might change opinion over time due to personal biases and experience. These limitations highlight the need for standardized, efficient, and reproducible methodologies to improve the accuracy and reliability of rock mass characterization (Cheng et al. 2022; Wang et al. 2025). Furthermore, the development of automated approaches should be viewed as a complementary tool that augments human expertise. The outputs can provide preliminary interpretations that experts subsequently refine, thereby substantially improving overall workflow efficiency. Recent advances in computer science have led to the increasing application of Artificial Intelligence (AI) and Machine Learning (ML). Among these approaches, Deep Learning (DL) has emerged as one of the dominant paradigms in machine learning due to its ability to model complex patterns through deep neural network architectures. Although DL has made substantial progress in many fields, its application in geology, specifically structural geology, remains underdeveloped (Harvey Wang et al. 2018). Existing DL models for detecting geological structures have primarily focused on the identification of discontinuities but neglected to distinguish between different types of discontinuities, such as natural versus excavation-induced fractures (Alzubaidi et al. 2022; Liu et al. 2023). Moreover, the generalizability of DL models across different datasets remains a major challenge. Most existing models are trained and validated under highly specific conditions and often fail to perform reliably when applied to data acquired from different geological environments, imaging tools, or acquisition parameters. The primary objective of the “Differentiation of fractures and rock mass deformation in clay rocks by machine learning (ML)” project is to develop automated routines for geological datasets that can identify structural feature traces and surfaces, distinguish between different discontinuity types, and characterize rock mass behavior and deformation features. The Mont Terri rock laboratory (MTRL) provides a unique and comprehensive dataset spanning multiple spatial scales, from rock cores and boreholes to tunnel excavation surfaces, forming a robust foundation for the development and evaluation of machine learning methods across diverse geological datasets.
Wang et al. (Thu,) studied this question.