Prediction of Soil-Water Characteristic Curve Using Artificial Neural Network Approach
Conference Paper
Overview
Identity
Additional Document Info
Other
View All
Overview
abstract
Copyright 2018 by the American Society of Civil Engineers. All Rights Reserved. In pavement ME design, the soil-water characteristic curve (SWCC) is used to estimate the resilient modulus of unbound material at different saturation conditions based on the enhanced integrated climatic model (EICM). The Fredlund-Xing equation is employed to generate the SWCC and the fitting parameters of the equation are correlated to the soil physical properties using regression models. However, the prediction accuracy of this method needs to be improved. To solve this problem, this study employs an artificial neural network (ANN) approach to improve the prediction accuracy of the SWCC fitting parameters. A new set of SWCC-related soil physical properties are selected to predict the SWCC. A large number of soil datasets collected from the NCHRP 9-23A project are used to develop the ANN models. The developed ANN models of the SWCC fitting parameters are used to predict the suction-water content relationship. The R2 values were found to be 0.9 and 0.95 for predicting the SWCC curves of plastic and non-plastic soils respectively, which are significantly higher than existing correlation models. The prediction accuracy of the developed ANN models are validated by comparing the measured and predicted matric suction of new soils datasets that are not used to train the ANN models.
