19 Estimating Resilient Modulus Using Neural Network Models

Resilient modulus (M_R) is an important material property in pavement design. Although generally determined by cyclic triaxial testing, the M_R values can be estimated from correlations with other material properties. Consequently, a combined laboratory and modeling study was undertaken to develop artificial neural network (ANN) models for subgrade soils. Sixty-three soil samples from different sites were tested for Atterberg limits, wet sieving, dry sieving, moisture-density, M_R, and unconfined compressive strength. The following parameters were used in the development of the ANN models: moisture content, dry density, plasticity index, percent passing No. 200 sieve, and compressive strength. Bulk stress and deviatoric stress were also used as model parameters. Six different ANN models were considered, namely Linear Network (LN), Generalized Regression Neural Network (GRNN), Radial Basis Function Network (RBFN), and Multi-Layer Perceptrons Network (MLPN) with one (MLPN-1), two (MLPN-2) and three (MLPN-3) hidden layers. Based on the R² values, the MLPN-2 model was found to be most effective. The proposed MLPN-2 model can be used in the design of pavements involving common subgrade soils in Oklahoma.