Calibration of Within Project Variabilities

As discussed earlier, WPV is tied to the rate of distress development for a given project: the higher the WPV, the more slower the distress progresses, indicating the wider range of project performance. The rate of distress development can be determined from observed performance history.

On the other hand, WPV can be expressed in terms of coefficients of variances for the selected set of random inputs. As discussed before, the following sets of variables for each layer are assumed to be random and follow certain statistical distributions:

•Thicknesses

•Moduli

•Fatigue resistances

•Rutting resistances

These variables are selected because their variances are found to have large effects on the range of simulated project performance. CalME uses Monte Carlo simulation to estimate the corresponding distribution of project performance, from which a performance history can be predicted. A predicted rate of distress development can in-turn be determined.

The way to account for WPV then lies in choosing the right coefficient of variances for the selected random inputs so that the predicted and observed rate of distress development matches. Note that resulting distributions for each of these random inputs should be similar but not the same as the actual industry practices. This is because the selected set of random variables is only a subset of all of the potential random factors affecting pavement performances.

As shown in Figure 3 under subsection Calibration of Transfer Function, the median value of the observed shape parameter  is about -5.0. The next step was to evaluate whether the use of the estimated median distributions of important mechanistic variables would result in the observed within-project variability (WPV) (i.e., the same result as using a fixed  of about -5.0.

The typical variance of total asphalt layer thickness came from data collected from cores and ground-penetrating radar stored in the Caltrans iGPR tool. Fourteen different projects built between 2000 and 2010 were analzyed, totaling 33 miles total length of paving. The conclusion was that the thickness variability found in the these projects matched those identified from the literature when developing CalME v2.0.

The typical variance of HMA stiffness and the fatigue damage equation parameter A was determined for each mix type and PG grade from laboratory flexural fatigue testing of 35 total HMA and RHMAG mixes.

A batch of Monte Carlo simulations was run with different combinations of asphalt layer thickness, stiffness, and fatigue parameter variabilities close to the values estimated as described above to evaluate whether the resulting equivalent shape factors were similar to the median observed shape parameter. Three combinations were found to result in observed shape parameters that are close to -5.0, as can be seen in Table 1. Based on these results, the first set was selected for use in the calibration WPV because these values are closest to those observed in the evaluation of thickness and asphalt properties described above.

Table 1: Variabilities of Asphalt Concrete Layer Properties and the Resulting Shape Parameter

*: SDF of a variable x is defined as the