Application of an Artificial Neural Network for Modeling the Mechanical Behavior of Carbonate Soils

Abstract

Carbonate soils have some distinctive features such as compressibility and skeletal particle crushability that make them distinguishable from other types of soils. Many experimental models have been developed to describe the complex behavior of carbonate soils, but despite these numerous works, there is no unified approach that can model the behavior of various types of these soils. In this paper, a new approach based on artificial neural networks is presented to predict the mechanical behavior of different carbonate soils. The network had five input neurons, namely, relative density, axial strain, maximum void ratio, calcium carbonate content, and confining pressure; ten neurons in the hidden layer; and two neurons in the output layer, namely, deviatoric stress and volumetric strain at the end of each increment. The network was trained and tested using a database that included results from a comprehensive set of triaxial tests on three carbonate soils. Comparison of the model prediction and experimental results revealed that the proposed approach was accurate and trustworthy in representing the mechanical behavior of various carbonate soils.