Predicting Excavation-Induced Damage Depth Through MLP

ABSTRACT: Being able to predict the dimensions of excavation damage zones (EDZs) is crucial for the design of permeability-sensitive structures. This study focuses on predicting excavation-induced damage depth using a machine learning method called multi-layer perceptron (MLP). The inputs of numerical models, used to model EDZs, were all used as features in the MLP to examine the influence on the predicted EDZ. Permutation analysis was employed to assess input feature importance. By, using Pearson and Spearman correlation matrices, the numerical input parameters were classified into three categories. After individually excluding each category, the stress components were found to be the most important model input. Next, the variance inflation factor (VIF) approach was employed to introduce medium-sized and small-sized models, comprising six models each. Comparing the MLP results from the large model group, which achieved an R2 score of 84% and a minimal mean absolute error (MAE) of 0.06, with the best performance of the small and medium sized VIF results, it was found that the smaller model which incorporated only three parameters (Young's modulus, crack initiation, σ1) exhibited similar performance with an R2 score of 83% and an MAE of 0.07, making it a viable alternative relying on fewer parameters. 1. INTRODUCTION The exploration and utilization of deep geological formations for the disposal of radioactive waste and the excavation of underground openings present complex challenges that demand a comprehensive understanding of the excavation-induced changes to the host rock. Various global research programs, including those at the Underground Research Laboratory (URL) in Canada, HRL (Hard Rock Laboratory) in Sweden, Mont Terri in Switzerland, and Yucca Mountain in the USA among many others, have contributed significantly to unraveling the mechanical, thermal, hydraulic, and geochemical intricacies of these geological environments (Cai & Kaiser, 2005) and the impact of excavation therein. Excavation of underground openings is a dominant subject in rock mechanics and many studies have focused on the brittle failure characteristics around deep excavations (Duan et al., 2019; Hsiao & Chi, 2013; Lee et al., 2012). During excavating, the in-situ stresses undergo redistribution, resulting in a gradual increase in tangential stress and a decrease in radial stress at specific locations on the excavation surface. This alteration in stress levels can induce brittle damage, surpassing the crack initiation (CI) stress threshold. The cracks initiated during excavation extend into the rock mass as the stress concentration surpasses the rock's failure strength. A new equilibrium is established, forming excavation damage zones (EDZs) within the affected region (Perras & Diederichs, 2016). The formation of the EDZ is a complex phenomenon, with researchers emphasizing stress redistribution, blasting disturbances, unloading stress waves, and the coupling effects of loading waves as key contributors to its evolution (Cai, 2008; Diederichs et al., 2004).

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.