Liquefaction Strength of Ottawa Sand: CDSS Experiments and ANN Modeling

Abstract

Characterization of the liquefaction strength of sandy soils is essential in modeling geotechnical engineering problems involving liquefiable soils. This paper investigates the liquefaction strength of Ottawa F65 sand through an extensive series of undrained, stress-controlled, cyclic direct simple shear (CDSS) tests performed at different densities, overburden pressures, and static shear stresses prior to cyclic shearing. The relatively large number of CDSS tests is used to develop an artificial neural network (ANN) model in order to predict Ottawa F65 liquefaction strength for densities and loading conditions that are not available in the experimental results. The predictive capability of the ANN model is assessed using blind predictions of the cyclic strength in new CDSS tests for a relative density and vertical effective stress not available in the training data set. Afterwards, CDSS tests under similar conditions were carried out. The comparisons of the predictions with the experimental results suggest that the ANN model provides reasonably good and conservative predictions of the soil liquefaction strength and shows sensitivity to changes in vertical effective stress, soil density, and cyclic stress ratio.