Radial Basis Function Neural Network Predicting of Resilient Modulus for Shanxi Subgrade Soils

Abstract

Subgrade soils are very important materials to support highways. Current AASHTO pavement design procedures recommend the resilient modulus (Mr) of subgrade soils for pavement design and analysis. It is well known that resilient modulus of subgrade soils is dependent to in-situ state of stresses, moisture contents and compaction ratio. The primary objective of this study is to correlate the resilient modulus of commonly encountered subgrade soils in Shanxi, China with routine subgrade soil properties which include moisture content, dry density, plasticity index, and stress state using radial basis function based neural network approach. A tot al of 135 resilient modulus data were obtained using repeated triaxial loading tests. Two group data were used in the study: one for model development and another for model validation. Research results show that radial basis function based neural network is capable of predicting resilient modulus of subgrade soils using routine subgrade soil properties and stress state.