ABSTRACT
Although various empirical models have been proposed to predict the resilient modulus (M r) from laboratory test parameters, less effort has been made in developing reliable empirical models using piezocone penetration test (CPTU) data. Moreover, the prediction accuracy of the existing empirical models is not high enough. In the present study, a novel empirical model was proposed to predict the Mr from CPTU data based on the polynomial neural network. To this end, a comprehensive database comprising 16 different sites in Jiangsu province, China, was firstly compiled, which contains 124 sets of Mr , cone tip resistance (qc ), sleeve friction (fs ), pore water pressure (u 2), moisture (w), and dry density (γd ) values at the in-situ stress condition. Taking the in-situ Mr values as reference values, three empirical models was developed using the group method of data handling (GMDH) neural network. The results show that proposed GMDH model 3 (GMDH method) with the input parameters of qc , fs , w, and γd can accurately predict the Mr . The obtained specific expressions for prediction of Mr further prove the reliability of the GMDH model. Overall, the new GMDH method can more accurately predict the Mr of subgrade soil and guide engineering practice.
