Owing to the advancements in the field of machine-learning (ML), the prospects of coupling ML with engineering analyses are currently being realized for various applications. Such applications can be particularly useful for geological engineering, where the ground properties are heterogeneous and hard to estimate accurately. In this regard, surrogate models are being increasingly used as tools to accelerate learning for both research and practical applications. In this paper, we describe the application of a surrogate model for the analysis of a slope stability problem using the limit equilibrium slice method. For setting up the surrogate model, two primary stages are required: 1) artificial data generation, where numerous results are computed, and: 2) data learning, where ML is used for building the correlation between the problem inputs and results of interest. Through this relatively simple example, we demonstrate how routine engineering tasks can be readily automated.
Given the state of current knowledge in rock engineering, most problems do not allow for accurate predictions of rock mass response to excavations (Elmo et al., 2022). In rock and soil engineering projects, owing to the heterogeneous nature of the geological materials, it is generally advised to examine a range of input parameters, even after rigorous site investigations (Lees, 2013). For this reason, probabilistic tools have been coupled with geotechnical analysis (Abdulai & Sharifzadeh, 2021). For some problems, it is advantageous to use approximated closed-from solutions for rigorous probabilistic analysis that consider the range of inputs and results, rather than invest in a small number of highly complicated numerical models (Mitelman & Elmo, 2019). Indeed, commercial codes that rely on simpler solutions offer built-in probabilistic analysis capabilities. Examples include commercial codes such as RocSupport, for tunnel support design, and Slide, for slope stability problems (Rocscience, 2004). Due to the rapid advancements in computing power, it has become more reachable to couple probabilistic analysis with numerical modeling codes. An example for this is the Monte Carlo analysis feature available in the elastoplastic numerical code RS2 (Rocscience, 2007). Probabilistic analysis allows the engineers to extract valuable insights for the investigated problem, primarily, by allowing for the assessment of the range of anticipated outcomes. In addition, probabilistic analysis can be used for computing the probability of failure (PoF), which provides a better means for determining the reliability of a system compared to the traditional factor of safety (FoS) (Phoon et al., 2022).
