Prediction of California Bearing Ratio of Fine-grained Soil Stabilized with Admixtures

Abstract

The main focus of this study was to predict California bearing ratio (CBR) of stabilized soils with quarry dust (QD) and lime as well as rice husk ash (RHA) and lime. In the laboratory, the stabilized soils were prepared at varying mixing proportions of QD as 0, 10, 20, 30, 40 and 50%; lime of 2, 4 and 6% with varying curing periods of 0, 7 and 28 days. Moreover, the admixtures of RHA with 0, 4, 8, 12 and 16%; lime of 0, 3, 4 and 5% was used to stabilize soil with RHA and lime. In this study, the soft computing systems like simple linear regression (SLR), multiple linear regressions (MLR), back propagation artificial neural network (ANN) with different algorithms like Levenberg-Marquardt neural network (LMNN), bayesian regularization neural network (BRNN) and scaled conjugate gradient neural network (SCGNN) was implemented for the prediction of CBR of stabilized soils. Moreover, support vector machine (SVM) with different kernel functions like linear SVM (SVM-L), quadratic SVM (SVM-Q) and cubic SVM (SVM-C) were also performed. The result of ANN reveals that QD, lime and OMC were the best independent variables for the stabilization of soil with QD, while, RHA, lime, CP, OMC and MDD for the stabilization of soil with RHA. In addition, SVM proved QD and lime as well as RHA, lime, CP, OMC and MDD were the best independent variables for the stabilization of soil with QD and RHA, respectively. To check the performance of various models of soft computing systems, the prediction parameters like root means square error (RMSE), overfitting ratio (OR), coeficient of determination (R²) and mean absolute error (MAE) were considered. Result reveals the values of OMC of stabilized soil with QD and lime decreases, while, OMC increases in case of stabilized soil with RHA and lime. In addition, MDD of stabilized soil with QD and lime increases, while, decreases in case of stabilized soil with RHA and lime. The optimum content of QD was found 40% and lime 4% at varying curing periods to get better CBR of stabilized soil with QD and lime. Moreover, the optimum content of RHA was also found 12% and lime 4% at varying curing periods to get better CBR of stabilized soil with RHA and lime. The maximum CBR of stabilized soil with QD was found than that of stabilized soil with RHA for every curing period. The observed CBR and selected independent variables can be expressed by a series of developed equations with reasonable degree of accuracy and judgement from SLR and MLR analysis. These developed equations may be proposed to predict CBR of stabilized soils by knowing others independents variables in same cases. The model ANN showed comparatively the better values of CBR with satisfactory limits of prediction parameters (RMSE, OR, R2 and MAE) as compared to SLR, MLR and SVM for the prediction of CBR of stabilized soils. Therefore, the model ANN can be considered as the best fitted model in soft computing system for the prediction of CBR of stabilized soils. Finally, it might be concluded that the selected optimum content of admixtures and newly developed techniques of soft computing systems will further be used of other researchers to stabilize soil easily and then predict CBR of stabilized soils.