An advanced local calibration method for mechanistic-empirical pavement design

Model calibration and validation is an important step of empirical modeling. The current local calibration (LC) method suggested by the AASHTOWare mechanistic-empirical pavement design guide (MEPDG) is defective and can result in a completely distorted prediction of pavement reliability in design. This paper proposes an advanced LC method that integrates a jackknife sampling procedure and an iteratively weighted least squares technique into one coherent LC process. The jackknife sampling method, which has been recommended by AASHTO for model validation, is now used for identifying outliers of the data in order to ensure a homogeneous population. Meanwhile, an iteratively weighted least squares method is proposed to simultaneously estimate the LC coefficients and the standard deviation functions. Two LC case studies for Ontario roads are presented: one for the bottom-up fatigue cracking models of asphalt concrete pavements and the other for the joint faulting models of Portland cement concrete pavements. The comparison with the traditional split-data method has shown the effectiveness and efficiency of the proposed method. With little modification, the method can be introduced as a general calibration-validation process for all empirical and mechanistic-empirical models in engineering disciplines.