Utilizing the Monte-Carlo Capability in RS2 for Machine-Learning Applications

Due to the uncertainty that stems from the heterogeneous nature of geological materials, probabilistic tools have been incorporated into geotechnical practice. A notable example is the Monte Carlo analysis method that is available as a built-in feature in the program RS2 (Rocscience in Phase2 version 6.020. Rocscience Inc. Toronto, 2007). In this paper, we demonstrate how to utilize the Monte Carlo method for enhancement of geotechnical analysis. The procedure consists of two primary stages: 1) data generation, where numerous finite-element (FE) models are generated based on an estimated range of input parameters, and: 2) data analysis, where machine-learning models are used to correlate input parameters and results of interest. The verified ML algorithm can be referred to as a surrogate model. An example of the implementation of a surrogate model is illustrated through an anchored sheet pile wall problem. For this problem, the surrogate model correlates between the forces in the first and second row of anchors, thus allowing for prediction and optimization of ground support during construction.