Comparison of Performance Models’ Local Calibration Approaches—MLE vs. Least Square

Pavement mechanistic-empirical design (PMED) is a state-of-the-art analysis and design tool for flexible and rigid pavements. The transfer functions in the PMED are nationally calibrated and may provide unrealistic predictions for the local conditions, leading to under-design or over-design of pavement layer thicknesses. Therefore, the PMED models need recalibration for local conditions. Several studies have recalibrated the transfer functions for local conditions using the least square (LS) approach. Although LS is a widely used approach, it requires certain assumptions that may not be valid for non-normal distributions. This study uses the maximum likelihood estimate (MLE) approach to calibrate the transfer functions for PMED v2.6. The MLE approach optimizes a given function by matching predictions with a known distribution. MLE used four probability distributions: exponential, gamma, log-normal, and negative binomial. The paper presents the calibration of local transverse and bottom-up cracking for rigid and flexible, respectively. A total of 54 and 78 pavement sections for rigid and flexible pavements, respectively, are selected from the Michigan Department of Transportation (MDOT) Pavement Management System (PMS) database. The selection is based on the observed performance and availability of the PMED inputs. The results show that MLE outperformed the LS approach for both cracking models. MLE provides more robust calibration coefficients, especially for non-normal data distributions.