Developing Master Curve Predictive Equation Models for Local Conditions: A Case Study for Hawaii

One of the critical inputs for mechanistic-empirical pavement design and analysis is the Hot Mix Asphalt (HMA) stiffness, which is needed to compute the strains used in different transfer functions to predict damage and pavement distresses. The recently developed Mechanistic-Empirical Pavement Design Guide (MEPDG) uses the dynamic modulus (|E*|) as the design stiffness parameter. In the MEPDG, three input levels are available for determining the input values for most of the material parameters. In general, the input parameters are measured directly for level 1, estimated from correlations or regression equations for level 2, or estimated based on best estimate or default values for level 3. In the case of |E*| for new HMA materials, the MEPDG makes use of either Witczak’s model (NCHRP 1-37A) or the enhanced Witczak’s model (NCHRP 1-40D) for input levels 2 and 3. These models were developed with extensive databases and provide a good fit to the data used in their development and use information about gradation, mixture volumetrics, and binder characteristics. To obtain more representative results, it is still desirable that states develop their own libraries and models representative of their respective mixes. That information can then be used to provide pseudo input level 1 |E*| data. A potential challenge to accomplish this in the short term, however, is that there may not be enough variation in the local databases to estimate many of the parameters of the models if only a few gradations and binders are studied comprehensively. This paper presents a case study with this situation. The paper first compares the predictions of the Witczak model adopted in NCHRP 1-37A with measurements of dynamic modulus for a few Hawaiian mixes. Most of the data is from 79 Superpave mix specimens with 12.5 mm Nominal Maximum Aggregate Size (NMAS) that were prepared with two different gradations and three binder types (one unmodified and two polymer modified) at different asphalt content and air voids levels. The database also contains information from five additional field specimens. It is found that although Witczak’s model provides reasonable values, it provides biased predictions for the mixes analyzed. The paper also describes the statistical estimation of model parameters of a master curve model similar to the Witczak’s models but considering the limited variation in the dataset of gradation and binder characteristics. A total of 2,447 valid observations were used to estimate the model parameters. Some of the model explanatory variables are indicator or dummy variables (i.e., equal to 0 or 1) used to account for unobserved differences in gradation and binder type. It is shown that an excellent fit can be obtained using this type of variable combined with mixture volumetric and other binder information. A detailed analysis is also made of the effects of mix volumetrics on |E*|.

Language

  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01328018
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Jan 20 2011 11:22AM