Artificial intelligence algorithms for prediction of cyclic stress ratio of soil for environment conservation

Abstract

Civil Engineering structures are constantly subjected to environmental forces and changes around the surroundings hence a strong need for assessment of the interaction and predictive measures that lead to pro-active actions towards the structural safety and health monitoring. Significant damage is caused to buildings and port infrastructure when soil is subjected to dynamic loadings and high-frequency vibrations as in earthquakes and events that involve the interaction of soil structures. Dynamic loadings, cause significant displacements and tilting in soil-supported structures. Sands, gravels, and other granular materials that support buildings, bridges, port infrastructure, and retaining structures have to withstand the impact of dynamic loadings, with the requirement of serious considerations, to account for liquefaction risk in the soil. Hence there is a need for careful assessments of soil under dynamic loadings. Data from the laboratory for cyclic triaxial experiments and cyclic direct shear experiments from LEAP-2020 (Liquefaction Experiment and Analysis Project) have been used to assess soil fatigue by comparing cyclic triaxial and direct shear strength data. CSR (Cyclic Stress ratio) had correlations with cycle count, axial strain, axial stress, load in KPa, void ratio. In this paper, machine learning (decision tree, extra tree, knn, random forest) and deep learning (artificial neural networks) techniques are applied to the laboratory dataset and results include prediction of the cyclic stress ratio of soil corresponding to the standard input parameters as per ASTM D 3999–91 and ASTM D 6528–07. This study predicts the CSR of soil that can be used for the predictive assessment of the soil. Additionally, CSR (cyclic stress ratio) is a crucial element in determining soil liquefaction