Artificial Neural Networks for Modeling Shear Modulus and Damping Behavior of Gravelly Materials

In this paper, the feasibility of developing and using artificial neural networks (ANNs) for modeling the shear modulus (G) and damping ratio (D) versus shear strain (γ) behaviors of dry and saturated isotropic and anisotropic highly compacted gravelly materials is investigated. The database used for development of the ANN models comprises a series of 172 large-scale dynamic triaxial tests on 12 different materials. The cyclic triaxial tests were carried out under different confining pressures and loading frequencies. A feed-forward model using multilayer perceptrons (MLPs) for predicting behavior of gravelly materials was developed, and the optimal ANN architectures were obtained by a trial-and-error approach in accordance with error indices and real data. The ability of ANNs to predict the confining pressure, loading frequency, anisotropy, and dry and saturation effects on G-γ and D-γ behaviors was investigated. Reasonable agreements between the simulated and test results were observed, indicating that the ANNs are capable of capturing the dynamic behavior of gravelly materials. Moreover, the generalization capability of the ANNs was also used to check the effects of items not tested, such as confining pressure, loading frequency, dry density, and grain-size distribution.