Assessment of Soaked California Bearing Ratio of Clay-Gravel Mixtures Using Artificial Neural Network Modeling

Abstract

The flexible pavement consists of different thickness of layers of different materials. CBR, elastic modulus, moisture condition, unit weight are the basic characters of subgrade for the design of pavement components. Characterization of CBR is of primary importance for all flexible pavement-related tasks. CBR is a laborious and time-consuming test, therefore many researchers have suggested ANN techniques to predict CBR because it provides much better alternative. As per the IRC recommendation, CBR should be found for a soaked specimen of subgrade soil. The study presents the application of ANN for estimation of CBR of clay-gravel mixture under soaked condition. Nine different clay percentage and five different moisture range were selected for conducting CBR tests. In the analysis, dry unit weight parameter was made standardized using mean particle size of the mixtures and surcharge weight which was kept on the mould to provide vertical confinement while penetration of the plunger during testing. To determine the contribution of each independent variable, many methods have been proposed for sensitivity analysis. In the present study, Garson, Olden and Lek's profile models were developed using neural network to assess the influence of the parameters. From both the models contribution of independent variables is visualized. CBR values are found to reduce continuously when blending with clay fraction, but moisture content has more negative impact than clay on CBR values. The unit weight has strongest relationship with CBR. The strength of model that was developed has been examined in terms of standard error and co-efficient of determination. The standard error is 0.08800434 and R2 values are 0.9316485 of the generated model. It shows that the ANN technique is able to learn the relation between CBR and soil mass.