Civil Engineering ETDs

Publication Date

Fall 12-15-2017

Abstract

The recently developed mechanistic-empirical pavement design guide (MEPDG, also known as Pavement M-E design method) uses the nationally calibrated, binder viscosity-based dynamic modulus predictive model for the design and analysis of asphalt pavements. In this study, this model is assessed for its appropriateness for asphalt-aggregate mixtures typically used in New Mexico. In essence, this study investigates the predictability issue of complex modulus of New Mexico mixes. A total of 54 Superpave mixes with different aggregate gradations, air voids, and binder grades were collected from the mixing plants and from the pavement construction sites. The loose asphalt mixtures were then compacted, cored, and sawed to cylindrical specimens and tested for dynamic modulus and phase angle in the laboratory. Independence assurance testing was performed to assess the precision and accuracy of the test results. The time-temperature superposition principle was applied to develop mastercurves of complex modulus and phase angle functions of the asphalt concrete samples. The evaluated complex modulus mastercurve parameters were then used to calibrate the viscosity-based Witczak model for predicting dynamic or complex modulus of local asphalt concrete materials. The assessment of this model indicated significant underprediction and bias of the model in its current form for predicting complex modulus of the Superpave asphalt-aggregate mixes of New Mexico.

To this end, a new set of regression-based models for predicting dynamic modulus and phase angle functions of local asphalt mixtures were developed and validated in this study. Material properties such as mix volumetrics, aggregate gradations, and asphalt binder characteristics are the main factors that affect the viscoelastic material functions, such as the complex modulus of asphalt concrete. A goal is to examine the effects of these mixture variables on the complex modulus of asphalt concrete and thus modify existing predictive models or develop a new model to predict the complex modulus of asphalt concrete more accurately. With the aim at hand, the effects of aggregate gradation parameters on the complex modulus function of asphalt concrete were determined. To characterize various aggregate gradations, the two well-known gradation parameters of the aggregate blend, namely, the fineness modulus and the uniformity coefficient, were considered. Next, the effects of these two parameters on the complex modulus and phase angle functions were determined, and used in developing new predictive models. Statistical evaluation showed that fairly accurate estimations of dynamic modulus and phase angle of the local mixes can be possible by using these new predictive models.

While the above models use binder’s viscosity, a new set of models were developed using binder’s shear modulus. Indeed, M-E design has a binder shear modulus based dynamic modulus model. Various researcher claimed that the binder shear modulus based model is more biased and inaccurate when compared to the tested data or the predicted data from viscosity based model. The new binder shear modulus based model also uses gradation parameters and mixture volumetrics. To develop this model, asphalt binders were tested for complex shear modulus and phase angle using dynamic shear rheometer. Dynamic shear modulus and phase angle functions were generated by applying time-temperature superposition principle. Non-linear optimization technique was used to correlate the model parameters to the material properties to develop the final form of the model. Statistical evaluation showed good accuracy of the predictions made by these models.

Apart from the regression-based modeling and to improve the accuracy of the characterization problem, an advanced dynamic modulus and phase angle predictive model is developed in this study based on the artificial neural network methodology. A database containing 1,620 dynamic moduli with phase angle were used to develop this artificial neural network. A neural architecture with two hidden layers, each with 12 nodes was found to be suitable for predicting the dynamic modulus and phase angle of asphalt concrete. Statistical evaluation showed an excellent prediction ability of this model.

It is known that viscoelastic time-domain material functions, such as, relaxation modulus and creep compliance, or frequency domain function, such as, complex modulus can be used to characterize the linear viscoelastic behavior of asphalt concrete in modeling of pavement structure. Among these, the complex modulus has been adopted in the recent pavement M-E design method. However, for advanced analysis of pavement, such as, use of finite element method requires that the complex modulus function to be converted into relaxation modulus or creep compliance functions. There are a number of exact or approximate methods available in the literature to convert one linear viscoelastic material function to another. All these methods (i.e. exact or approximate methods) are applicable for any linear viscoelastic material up to a certain level of accuracy. However, the applicability and accuracy of these interconversion methods for asphalt concrete material were not studied very much in the past. Thus, a question arises if these methods are even applicable in case of asphalt concrete, and if so, what is the precision level of the interconversion method being used. To investigate these facts, this study has undertaken an effort to validate a numerical interconversion technique by conducting representative laboratory tests. The method was previously used in asphalt industry with adopting a simplification of considering time constants to be identical in generalized Maxwell and generalized Voigt model. However, in the present context, the assumption is regarded as over-simplification and therefore, an exact approach to estimate the time constants in these two mechanical models is developed. For validation, cylindrical asphalt concrete specimens were tested for complex modulus, relaxation modulus, and creep compliance at different test temperatures and loading rates. The time-temperature superposition principle was applied to develop linear viscoelastic material functions. The numerical interconversion technique was used to convert one material function to another, and hence, were compared to the laboratory tested material functions. The conversion showed good agreement with the laboratory test data. A statistical evaluation was conducted to determine if the interconverted material functions are similar to the laboratory tested material functions. Besides finite element modeling technique was also used to validate the interconversion method.

Keywords

Dynamic Modulus, Asphalt Concrete, Complex Modulus, Modeling, Characterization, Viscoelastic, Finite Element Method, Neural Network

Document Type

Dissertation

Language

English

Degree Name

Civil Engineering

Level of Degree

Doctoral

Department Name

Civil Engineering

First Committee Member (Chair)

Rafiqul A Tarefder

Second Committee Member

Arup K Maji

Third Committee Member

Tang-Tat “Percy” Ng

Fourth Committee Member

Tariq Khraishi

Available for download on Monday, December 16, 2019

Share

COinS