الفهرس 
CHAPTER (1) INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 188 |  |  |
| | | from | 188 | | |
Abstract
”Resilient modulus (MR) is a main strength parameter used in most of the current pavement structural design methods to characterize unbound granular materials used for the construction of base and subbase layers as well as the subgrade soils. However, due to the complexity and cost of the MR test, other options i.e., regression modelling and machine learning are used to predict such important input taking into consideration the wide spectrum of factors that are known to influence the MR parameter. This study presents the results of modeling the resilient modulus (MR) of unbound granular base/subbase layers by means of the material index properties and stress state. The database, employed in this study, is collected from literature studies, which includes 50 unbound granular materials (UGMs) (28 from the Long Term Pavement Performance (LTPP) database and 9 from Virginia, US while the other 13 materials are representing different quarries in Egypt). The database includes (liquid limit, LL, plastic limit, PL, plasticity index, PI, optimum moisture content, OMC, %passing US standard sieve No. 4 and No 200). The total number of MR measurements in the collected data is 735. Two common literature MR-predictive models are used, the K-θ model and the Universal model as the base models. By using the fitting curve toolbox (CFTOOL) in the MATLAB program, the values of the regression coefficients of both models were recalibrated to predict the MR for each material individually. Both models’ regression coefficients (k-values) are correlated with the index properties of the soils (LL, PL, OMC, Pass#4 and Pass#200). The results of the Goodness of Fit Statistical Parameters (R2 values) showed that R2 for K-θ model was training set 0.67 and testing set 0.58, while for universal model 0.67 for training set and 0.65 for testing set. In addition, four different Least-squares support-vector machines (LS-SVM) are used to predict the resilient modulus (Particle Swarm Optimization model, Grey Wolf Optimizer model, Slime Mould Algorithm Model and Harris Hawks Optimization model). The data was divided into training set of 70% (515 datapoints) and testing set of 30% (220 datapoints). The results showed the ability of machine learning techniques to predict the MR with excellent results. The results of the Goodness of Fit Statistical Parameters (coefficient of determination, R2 and root mean square error, RMSE) show that the Grey Wolf Optimizer (GWO) model is the best model to predict the MR with R2 of 0.99 for training set and 0.95 for testing set.