Intelligent prediction of coefficients of curvature and uniformity of hybrid cement modified unsaturated soil with NQF inclusion
Onyelowe, Kennedy C.
MetadataShow full item record
The cost and sophisticated equipment required to conduct earthwork laboratory experiments have been of concern to the design and performance monitoring of infrastructures in recent times. Lateritic soils especially those under unsaturated conditions are erratic and deserve close attention in terms of laboratory studies. In order to overcome the rigors and time consumed during experimental procedures, soft computing has been used to predict soil parameters for the purpose of design and construction. In this work, the ANN, GEP and LMR were employed to predict the coefficients of curvature and uniformity of lateritic soil treated with multiple binders locally generated, which were hybrid cement (HC) and nanostructured quarry fines (NQF). The effect of the varying dosages of HC and NQF added to the soil were studied and the behavior of clay activity, clay content, frictional angle, coefficients of curvature and uniformity were measure. 121 datasets were generated from the experimental exercise for the selected parameter both for the predictors and for the targets. These datasets were deployed in the ratio of 70 is to 30% for training and testing of the models predictions respectively. The performances of the models were evaluated using error analysis (VAF, RMSE, MAE) and accuracy (R2) indices and it was observed that the ANN outclassed both GEP and LMR due to its speed and robustness in adopting backpropagation and feed-forward algorithms. Furthermore, the sensitivity analysis showed that F, C, H (HC), NQF and Ac in that order of most influential to least influential influenced the behavior of the Cc model with H (HC) and NQF showing equal effect on the Cc. Also, H (HC), NQF, F, C and Ac in that order of influence from most to least affected the behavior of the Cu predicted model also with HC and NQF having equal effect on the Cu. Generally, the learning techniques showed good performance in predicting the outputs hence are good techniques to be utilized in design and performance evaluation.