Local Calibration of Flexible Performance Models Using Maximum Likelihood Estimation Approach

Abstract

This paper uses maximum likelihood estimation (MLE) approach to calibrate bottom-up cracking, total rutting, and international roughness index (IRI) transfer function for flexible pavements. It used four distributions: gamma, exponential, negative binomial, and log-normal, and results are compared with the LS approach. Initially, synthetic data is generated for bottom-up cracking to demonstrate the effectiveness of MLE over the LS approach. Finally, measured data for two hundred and fifty-six new flexible pavements is used from MDOT’s PMS database to calibrate and validate transfer functions. Resampling methods are combined with MLE to improve its robustness. 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.