Stress-Strain Modeling of Sands Using Artificial Neural Networks

Abstract

An attempt has been made to implement artificial neural networks (ANNs) for modeling the stress-strain relationship of sands with varying grain size distribution and stress history. A series of undrained triaxial compression tests for eight different sands was performed under controlled conditions to develop the database and was used for neural network training and testing. The investigation confirmed that a sequential ANN with feedback is more effective than a conventional ANN without feedback, to simulate the soil stress-strain relationship. The study shows that there is potential to develop a general ANN model that accounts for particle size distribution and stress history effects. The work presented in this paper also demonstrates the ability of neural networks to simulate unload-reload loops of the soil stress-strain characteristics. It is concluded from this study that artificial-neural-network-based soil models can be developed by proper training and learning algorithms based on a comprehensive data set, and that useful inferences can be made from such models.