Deep learning-based prediction of particle breakage and friction angle of water-degradable geomaterials

Abstract

Crushed soft rocks are becoming inevitable geotechnical materials for construction of foundations, fill material for transportation infrastructures, and earthen dams for economic and environmental reasons. The contemporary issue is to predict the water-induced disintegration of these non-traditional granular materials and its correlation with the friction angle of these soils. Machine learning techniques offer a viable solution in this context as they are widely used to understand and predict complex phenomena across various applications. In this work, a deep learning technique, artificial neural networks (ANN), was employed to an experimental dataset obtained from torsional shear tests on dry and saturated crushed soft rocks to predict the water-induced particle breakage potential (ID), peak friction angle (PFA), and residual friction angles (RFA). The experimental ID – PFA and ID – RFA correlations were also predicted using ANN model for all the three outputs (ID, PFA, RFA) with very low errors.