Local Calibration of Joint Faulting Model by Using Resampling Techniques

Abstract

The local calibration of the performance models in the Pavement-ME analysis and design procedure is a challenging task mainly because of (a) data needed to characterize the existing pavement sections, (b) nature of the performance models, and (c) inadequate number of pavement sections for a robust calibration. The transverse joint faulting model in the Pavement-ME uses a complex incremental approach. The model has eight (8) different calibration coefficients which can be adjusted to fit the measured faulting. Further, calculating predicted faulting using the published model equations is difficult. The determination of faulting outside the Pavement-ME is essential to calibrate all the model coefficients simultaneously. This paper documents the process of calculating faulting outside of the Pavement-ME by utilizing the inputs specific to the pavement sections. The first step consists of documenting the impact of each calibration coefficient on predicted faulting. This sensitivity assists in determining the most important calibration coefficient in the model. The second step is to calibrate the model by using all the available pavement sections in the calibration dataset. Multiple sampling approaches were utilized to calibrate the faulting model in the State of Michigan. First, the traditional approach uses about 70% of the data for calibration and the remaining 30% for validation. However, most of the States have a limited number of pavement sections for local calibration. Therefore, there is a need to employ statistical methodologies that are more efficient and robust for model calibrations. Second, bootstrapping and jackknifing resampling methods are used to efficiently estimate the model coefficients. The results in the paper show the effect of different sampling approaches on model parameter estimations and compare the role of such parameters in reducing the bias and the standard error of the faulting model.