Local Calibration of the MEPDG Using Pavement Management Systems
Chapter 9. Recommendations
Based on the review and analysis conducted under this study, the research team has the following recommendations to assist in the MEPDG calibration process using pavement management data.
- Evaluate the potential differences between the SHA pavement condition survey methodology and that conducted in accordance with LTPP Distress Identification Manual. Since the MEPDG performance prediction models are based on the noted distress of the LTPP pavement sections, differences in pavement condition assessment can influence the accuracy of the distress prediction.
- For this study, many of the design inputs for the included project sections are the default values contained in the MEPDG software. An in-depth review of the inputs should be conducted to make certain the values apply to the design process and materials for the DOT. Recalibration of the models should be performed if additional design information is obtained.
- A larger sample size is needed for a statistically meaningful calibration. However, many agencies are expected to have to deal with limitations to the amount of material properties and traffic data that can be correlated to the condition data within a SHAs pavement management system. For those states that are utilizing data contained within the LTPP, acquiring data from adjacent states could be considered as one way of providing additional data sets.
- NCDOT is currently measuring pavement roughness using a high-speed profiling device; however, results of rut depths measured from the profiler were not provided to the research team. It is recommended that NCDOT, and other SHAs that modify their data collection procedures, evaluate the use of new data in the calibration process. Having additional pavement sections whose performance is monitored over consecutive years is expected to greatly improve the accuracy of the calibrated MEPDG prediction model.
- Although it likely would improve the accuracy of the thermal crack prediction, it is not expected that many SHAs will go to the effort of measuring the depth of all thermal cracks; however, it would be beneficial to at least measure the depth of the transverse crack at each significant change of a cooling cycle. NCDOT currently measures thermal cracking based on the predominant distress severity and extent over the pavement sections. To improve the thermal crack prediction, it is recommended that agencies also consider measuring crack width, crack length, and the number of thermal cracks over the pavement section.
- Asphalt mixtures should be characterized in accordance with dynamic modulus and tensile strength. The dynamic modulus and tensile strength properties are needed to determine the asphalt mixture thermal properties (m, n, and A).
- Recalibration should take place when additional data are available near the end of the pavement service life. Data for the JCP projects only spanned approximately half of the anticipated service life. Data with distress measurements near the end of the performance period are expected to improve the calibration of the performance models.
- Many of the JCP distresses were not observed on the sections included in the calibration study. Model calibration should be reviewed when distress data approaching the selected failure criteria are available. Alternatively, if performance data does not approach the selected design criteria, the selection of this value should be re-assessed for local performance.
- The NCDOT distress surveys are completed by March of each year, but the exact time of the survey is not recorded in the pavement management system. With the MEPDG predicting distresses on a monthly basis, the survey month should be compared. While the difference of 1 month may be minimal in the calculations, having unknown dates adds to the prediction error.
- Due to the requirement of a large number of pavement sections, this study did not calibrate the IRI model for either HMA or PCC pavements. It appears that NCDOT has the necessary measurement equipment and pavement sections to adequately calibrate these models. Calibration of the IRI models appears to be stop
Another challenge with the MEPDG calibration process will be in obtaining, evaluating, and analyzing the large quantity of data inputs. For the NCDOT evaluation, a small number of pavement sections still resulted in a considerable length of time to evaluate, analyze, and format for calibration purposes. The availability of only limited performance data and the presence of data that did not meet the LTPP data definitions required additional efforts by the research team to correlate the two survey approaches. Other agencies that do not use survey procedures that are identical to the LTPP methodology will have to go through similar steps.
Nevertheless, it has been demonstrated that the MEPDG models can be calibrated using pavement management data even if only limited data sets are used for the initial calibration activities. Based on the calibration steps outlined in this report, a suggested timeline for calibrating the MEPDG performance prediction models is presented in table 65. The actual timeline will be dependent on the availability of performance data, the size of the state pavement network, the variation of pavement categories and designs, and the number and experience of personnel.
Table 65. Estimated timeline for local calibration.
||Comments (NCHRP 2009)
|Select hierarchical input level
||This should reflect how an agency intends to use the MEPDG on a day-to-day basis.
|Develop experimental plan
||Determine the number of pavement categories (e.g., full-depth HMA, HMA overlays, PCC, PCC overlays). Consider grouping by similarities in material type, climate, subgrade soil, and traffic loadings.
|Estimate sample size
||Identify which performance models will be calibrated. Establish performance criteria for each distress type (e.g., rut depth, load-related cracking, faulting). Determine acceptable bias. In addition, each pavement category should also include replicate sections.
|Select roadway segments
||Selected segments should have a range of distress for pavements of similar age. Consider excluding pavements with premature failure or extremely superior performance. Selected pavement sections should have similar types and extent of distresses present over a similar length of time. This will help in minimizing the potential for bias.
|Evaluate project and distress data
||Access correlation between LTPP and State pavement distress definitions. Check, confirm, and remove outliers. Confirm that selected pavement sections include values close to the selected performance criteria. Confirm pavement sections have initial IRI data (IRI prediction model is highly dependent on initial IRI), construction history data, traffic data, rehabilitation data, and materials data.
|Conduct field testing and forensic investigation
||Confirm rut depth in various pavement layers through extensive coring or trench studies. Confirm location of crack initiation (top-down or bottom-up). This step should be based on the SHAs acceptance of the assumptions and conditions contained within the MEPDG; therefore, it is difficult to estimate the length of time to conduct this investigation.
|Access and reduce/eliminate bias
||Includes the evaluation of predicted versus measured pavement distress. Adjust calibration coefficients if the model precision is reasonable, but the accuracy is poor. If the prediction is reasonable, but the precision is poor the calibration coefficient is likely dependent on a site feature, material property, and/or design feature. If the precision is poor and the model accuracy is dependent on time or number of load cycles (i.e., poor correlation between measured and predicted distress), the exponent on the number of load cycles needs to be considered.
|Assess and reduce standard error
||Assess the relationship between the standard error from the local calibration process to that in the MEPDG software. If significantly different, determine if the standard error is dependent on some other parameter or material/layer property. If no dependency is determined, accept the local calibration coefficient.
||Determine whether or not to accept the locally calibrated models or use the nationally calibrated models. Identify any major differences between LTPP projects and SHA standard practice. Determine whether or not the calibration coefficients explain any of these differences. Ensure engineering reasonableness.
||This is an estimate of maximum length of time required for the calibration process. However, depending on the available data contained within the pavement management system, this process could be shortened considerably.