Local Calibration for Mechanistic-Empirical Design using Genetic Algorithm

The Mechanistic Empirical Pavement Design Guide (MEPDG) is expected to be adopted by most transportation agencies and pavement engineers in the next few years. As a part of mechanistic-empirical pavement design procedure, it is required to locally calibrate distresses to match up analysis results with local measured data. However, it has been a challenging task for pavement practitioners and experts to calibrate distress models inherited in the design procedure due to the way the M-E design tool is processing the data. The literature review showed that the vast majority of calibration techniques currently in use are solely based on statistical analysis and trial and error approach for different combination of local calibration coefficients to find the best set that produces results closer enough to observed data in the field. This approach lack accuracy due to limited trials that can be evaluated and the absence of mathematical algorithm to guide the trial selection at the start of each MEPDG analysis cycle to find the optimum set of calibration coefficients. This study will investigate the possibility of using genetic algorithm (GA) to calibrate MEPDG distresses. Framework of calibration system will be designed to simulate the MEPDG calibration process within the genetic algorithm context. Site specific data from different locations will be used as inputs to MEPDG and initial calibration coefficient seeds will be presented to the system to produce initial distress output and compared to measured field data. The genetic algorithm will then be employed to guide the selection of new calibration set each time analysis cycle is performed and crossover and mutation processes will be used to produce new sets of chromosomes and presented to the calibration system for new evaluation cycle in an automated process to overcome drawbacks of the traditional trial and error approach. Calibration framework design and development will be discussed in this study along with results and advantages of using the genetic algorithm approach over traditional ones.