Relative Significance of Coefficients and Local Calibration of PMED Rigid Performance Prediction Models

Abstract

The Pavement-ME is a modern approach to design new and rehabilitated pavements. The tool predicts the pavement distresses and roughness over the design life considering local traffic, material properties, and climate conditions. Several studies conducted in the past have optimized the performance prediction models by locally calibrating the transfer functions. However, those have primarily focused on calibration efforts by minimizing the standard error and bias. Moreover, the one-at-a-time (OAT) approach was used for sensitivity analysis by comparing the performance prediction with input change. The results ranked significant input variables based on coefficients sensitivity index (SI) or normalized sensitivity index (NSI). This paper presents Scaled Sensitivity Coefficients (SSCs) for sensitivity analysis of transfer function coefficients and calibration efforts for transverse cracking and IRI performance models of jointed plain concrete pavements (JPCP). About 65 pavement sections are used for cracking and IRI model calibrations in Michigan. SSCs identify the model's sensitivity to the parameters for performance prediction over the entire range of available sections and independent variables in the models. Both linear and non-linear approaches are used to calibrate the performance models while minimizing standard error and bias. The results show that SSCs are critical factors in parameter identifiability concerning the relative error of the coefficients. The transverse cracking model is more sensitive to C5. The sensitivity ranking for the IRI model is C1 (cracking), C4 (site factor), C3 (faulting), and C2 ( spalling). The results show that locally calibrated models significantly reduced standard error and bias compared to the global model.