Evaluation of Different Methods for Populating the LTPP Materials Database with the Dynamic Modulus
Acquiring the |E*| from existing pavements is difficult due to the standardized dimensions of the test specimen. Other geometries, indirect tension specimens and prismatic specimens, have been tested to determine if the measured |E*| is statistically different from the modulus obtained from the AASHTO TP 62 protocol. This study provides a comparison of the effects of a non-uniform state of stress and anisotropy. These effects are isolated by comparing specimens prepared by Superpave gyratory compaction and vibratory steel-wheel compaction. The results in the thesis are verified using four 12.5 mm surface course mixtures with different aggregate types and binder types, and one 25.0 mm base mixture. The results are verified using volumetric variations such as different percentage of aphalt cement and air voids. The results show that the difference between the |E*| values obtained from different geometries is statistically insignificant. The results provide justification for using alternative methods for acquiring the |E*| experimentally, specifically from previously constructed pavements.
Practioners prefer a mathematical model since measuring the |E*| is a time and labor intensive. The second section of this thesis presents an artificial neural network (ANN) to predict the |E*| from the measured MR. The first step is an analytical method of calculating the MR from the |E*|. It involves the application of multiaxial linear viscoelastic theory to linear elastic solutions for the indirect tension test developed by Hondros (1959). The results show that the predicted and measured MR values are in close agreement. The results provide a forward model for the back-calculation of the |E*| from MR. Using this forward model, a database of measured dynamic moduli is populated with corresponding predicted resilient moduli to train an ANN. The trained ANN is the back calculation model used to predict |E*| from measured MR. The dynamic moduli predicted from the measured resilient moduli using the trained ANN are found to be reasonable compared to the measured dynamic moduli.
Advisor:Dr. Murthy Guddati; Dr. Y. Richard Kim; Dr. Roy H. Borden
School Location:USA - North Carolina
Source Type:Master's Thesis
Date of Publication:12/11/2007