Local calibration of flexible performance models using maximum likelihood estimation approach

The pavement mechanistic-empirical design (PMED) is a widely used pavement analysis and design approach. The transfer functions are generally calibrated to implement the PMED for local conditions. The least square (LS) approach has been commonly used to calibrate these transfer functions. Although LS is a simplistic approach, the assumptions may not be valid, especially for non-normally distributed data. This paper uses the maximum likelihood estimation (MLE) approach to calibrate bottom-up cracking, total rutting, and international roughness index (IRI) transfer function for flexible pavements. The results show that overall, MLE outperforms the LS approach for synthetic and measured data. The difference is more evident in the case of bottom-up cracking data, which does not follow a normal distribution. Gamma distribution for bottom-up cracking and total rutting, whereas negative binomial for IRI is the most suitable distribution for the MLE approach. Overall, MLE using resampling methods provides a robust and better estimate than the LS approach.