Modeling resilient modulus of fine-grained materials using different statistical techniques

Abstract

For the realistic prediction of pavement performance, it is very important to accurately characterize the mechanical behavior of unbound material layers and subgrade soils. In pavement analysis using the elastic layered theory, material properties in terms of dynamic elastic modulus and Poisson’s ratio are the major input parameters. The dynamic elastic modulus of pavement materials or resilient modulus (MR) is measured by conducting repeated load triaxial compression tests typically not available to highway authorities due to the high costs involved in acquiring such high-performance equipment and/or lack of knowledge on how to operate this specialized equipment. Therefore, pavement engineers are obligated to use pavement design methodologies based on empirical tests. The main objective of this study is to use multiple linear regression (MLR), nonlinear regression and backpropagation artificial neural network algorithms to develop models to predict the resilient modulus of fine-grained materials based on 3709 soil samples collected from the Long-Term Pavement Performance (LTPP) website. The key input parameters selected for this study are: the confining pressure (σ3), nominal maximum axial stress (σ1), percent of silt (S), Liquid Limit (LL), Plasticity Index (PI), percent passing number 200 sieve (P#200), maximum dry density (ρmax-dry), percent of clay (C), optimum moisture content (wopt), and laboratory-determined resilient modulus. Results revealed that the change of wopt has a higher effect on MR compared with the change in density, the percent fines, percent of silt, percent of clay, confining pressure, PI and LL but less than the effect of changing axial stress. In addition, the best modeling technique to predict the resilient modulus of fine-grained soil was found to be the Artificial Neural Network (ANN) followed by the nonlinear regression and finally the MLR.