Summary of ME Design Software Local Calibration Efforts

Many state highway agencies have considered adopting the Mechanistic-Empirical Pavement Design Guide (MEPDG) and the accompanying AASHTOWare® Pavement ME Design software to replace the empirical AASHTO Pavement Design Guides and the widely used DARWin pavement design software. Based on the inputs and trial design information, the ME Design software “mechanistically” calculates pavement responses (stresses and strains) and uses those responses to compute incremental damage over time. It then utilizes the cumulative damage to “empirically” predict pavement distresses for each trial pavement structure.

Although the MEPDG was nationally calibrated using representative pavement test sites across North America, the application of the models to the full range of construction methods, materials, pavement preservation and maintenance practices, and climatic conditions is likely to significantly affect distress and performance. These local effects should be addressed through local calibration studies to adjust, if necessary, the calibration coefficients of transfer functions in the ME Design software.

Without properly conducted local calibration efforts, implementation of the MEPDG will not improve the pavement design process and may yield thicker, overdesigned asphalt pavements. Recognizing the importance of local calibration, NCAT engineers have been reviewing the general approach undertaken for state highway agencies, the results of those efforts, and recommendations for implementing the nationally or locally calibrated models.

While it is often referred to as “local calibration”, the process may include local verification, calibration, and validation of the MEPDG. As a minimum, the local verification process is needed to determine if state practices, policies, and conditions significantly affect design results. In this process, the distresses predicted by the Pavement ME Design software using the nationally calibrated coefficients are compared with measured distresses for selected pavement sections. If the difference between the predicted and measured distresses is not significant, the Pavement ME Design can be adopted using the default models and coefficients; otherwise, it should be calibrated to local conditions.

If local calibration is warranted, it is important that it addresses both the potential bias and precision of each transfer function in the Pavement ME Design software. Figure 1 illustrates how the bias and precision terms can be improved during the local calibration process. In practical terms, bias is the difference between the 50% reliability predicted distress and the measured distress. Precision dictates how far the predicted values at a specified design reliability level would be from the corresponding predicted values at the 50% reliability prediction. The locally calibrated models are then validated using an independent set of data. The models are considered successfully validated to local conditions if the bias and precision statistics of the models are similar to those obtained from model calibration when applied to a new dataset.

Improvement of bias and precision through local calibration
Figure 1: Improvement of Bias and Precision through Local Calibration.

While the AASHTO calibration guide details a step-by-step procedure for conducting local calibration, researchers have found that the actual procedures utilized vary from agency to agency. This is partially due to the timing of the publication relative to the initiation of such efforts and the release of new versions of the software. This presents challenges for state agencies, as local calibration is a cumbersome and intensive process and the software and embedded distress models are evolving faster than local calibrations can be completed.

Researchers found that calibration was typically attempted by looking at the predicted measured distress for a set of roadway segments and reducing the error between measured and predicted values by optimizing the local calibration coefficients. However, other approaches were taken. Although a minimum number of roadway segments necessary to conduct the local calibration for each distress model is provided in the AASHTO calibration guide, the step for estimating sample size for assessing the distress models was not always reported. For those efforts that did report a sample size, some were smaller than the minimum amount recommended.

The AASHTO calibration guide recommends conducting statistical analyses to determine goodness of fit, spread of the data, and the presence of bias in the model predictions. Three hypothesis tests are recommended: 1) to assess the slope, 2) to assess the intercept of the measured versus predicted plot, and 3) a paired t-test to determine if the measured and predictions populations are statistically different. Although some local calibration efforts included all three hypothesis tests, many only evaluated some of the statistical tests and others relied only on qualitative comparisons of measured versus predicted distresses.

As such, it is difficult to assess the quality of the predictions or effectiveness of local calibration for each model. For asphalt pavements, the rutting model was the most commonly calibrated model. Transverse and longitudinal cracking models were calibrated the least, with the longitudinal cracking model having the poorest precision. Given the significant spread reported for the default longitudinal cracking model, it should not be used for design. It is anticipated that a new model will be developed under the ongoing NCHRP 1-52 project.

Twelve studies conducted for state highway agencies were reviewed. Of the twelve studies, eleven aimed to calibrate performance models to state-specific conditions and one calibrated to conditions specific to a region. The table below denotes the number of verification, calibration, and validation efforts conducted for each performance model. Additionally, the results of the local calibration efforts are summarized. General improvements in predictions were realized with local calibration, however, the degree of improvement varied. Not all of the studies evaluated the presence of bias, but for those that did the results varied by model and study.

It is recommended that the statistical analyses outlined in the Guide for the Local Calibration of the MEPDG be utilized, as they enable a quantitative assessment of the calibration results. Specifically, such parameters help to determine if local calibration has reduced bias and improved precision and if implementation is warranted. This will also help identifying any weaknesses that may exist in the model that must be considered during the design process. The MEPDG and the AASHTOWare® Pavement ME software have the potential to improve pavement design. Local calibration and validation of the performance models are essential to the implementation of this design framework, however it is an ongoing effort as models continue to be refined, and unconventional materials utilized.

Table 1: Number of Verification/Calibration Studies and Summary of Calibration Results
Number of verification and calibration studies and summary of results