Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures (2022)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

36 This chapter presents and discusses most of the technical effort of NCHRP 09-59. Imme- diately after this introduction, the results of the laboratory testing and analysis of the resulting data are presented. This includes mixture tests, binder tests, and a comparison of these two data sets. This is followed by a section on layered elastic analyses of hypothetical pavement structures to clarify some ramifications of the findings on flexible pavement systems. The following section discusses potential binder fatigue specification parameters. This includes a discussion of several serious problems resulting from the current protocol for selecting binder fatigue specification test temperatures. Validation test results and analysis are then presented, along with several related issues. The chapter ends with a summary discussion, including limitations to the findings. Results of Laboratory Testing and Data Analysis Appendix E contains tables summarizing in some detail the results of the various mixture and binder tests performed as part of NCHRP 09-59. The following sections summarize the results of the tests and analyses in various tables and graphics with some basic observations concerning the data. Toward the end of this section—one of the most important of this report— comparisons among mixture fatigue test results and various binder tests results are made. Results of the Mixture Fatigue Tests Results of the mixture fatigue tests—flexural and uniaxial—are included in Appendix E. This includes tests using both methods at multiple temperatures and strains. The mixture fatigue data were analyzed using both the GFTAB approach and traditional methods, as dis- cussed in Chapter 2. However, more emphasis is placed on the GFTAB analysis because it was designed specifically to develop relationships between mixture fatigue and binder test properties. The following sections present the result of testing and analysis of mixture and binder test data in the following order: (1) flexural fatigue, traditional analysis; (2) uniaxial fatigue, S-VECD analysis; and (3) flexural and uniaxial fatigue data, GFTAB analysis. Flexural Fatigue Testing—Traditional Analysis Correlations between the strain level applied in the BBFT and the mixture fatigue life for nine asphalt binders are shown in Figures 13 and 14 with the regression coefficients listed in Table 7. Detailed BBFT results are included in Appendix E. The following observations can be made on the basis of correlations between the applied strain level and mixture fatigue life. 1. The fatigue life curves representing asphalt mixtures with polymer-modified binders were generally on the right side of the plot, suggesting better fatigue resistance than those with unmodified or oxidized binders, especially at 20oC. C H A P T E R 3 Findings and Applications

Findings and Applications 37   Binder ID k1 k2 r2 (%) A 20°C 7.43 * 1022 5.848 82.8 10°C 2.69 * 1019 5.283 94.2 B 20°C 1.25 * 1028 8.630 93.1 10°C 1.63 * 1020 5.928 75.4 C 20°C 2.14 * 1020 5.024 94.5 10°C 2.11 * 1028 8.346 92.8 I 20°C 7.37 * 1020 5.723 91.7 10°C 1.69 * 1018 4.950 88.6 J 20°C 2.02 * 1022 6.006 94.9 10°C 2.38 * 1022 6.489 95.7 K 20°C 1.38 * 1022 6.074 92.8 10°C 2.23 * 1024 7.294 89.8 M 20°C 3.68 * 1021 5.374 99.5 10°C 1.33 * 1023 6.235 91.7 O 20°C 1.20 * 1023 6.342 98.6 10°C 4.25 * 1023 6.811 93.4 P 20°C 1.93 * 1022 6.597 92.6 10°C 1.99 * 1032 10.667 96.9 Table 7. Regression coefficients and r2 for bending beam fatigue transfer functions. 200 400 600 800 1,000 1,200 1,400 1.E+3 1.E+4 1.E+5 1.E+6 1.E+7 1.E+8 St ra in , µ ε Cycles to Failure A B C I J K M O P 200 400 600 800 1,000 1,200 1,400 1.0E+3 1.0E+4 1.0E+5 1.0E+6 1.0E+7 1.0E+8 St ra in , µ ε Cycles to Failure A B C I J K M O P Figure 13. Cycles to failure versus micro-strain at 20çC. Figure 14. Cycles to failure versus micro-strain at 10çC.

38 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures 2. The fatigue life curves at 10°C were closer together compared with those at 20°C, indicating that the fatigue lives of these mixes became less different as the test temperature decreased. 3. As shown in Table 7, asphalt mixtures with unmodified or oxidized binders generally had higher k2 coefficients (Equation 6). Such mixtures were less strain tolerant and had lower cycles to failure at higher strain levels compared with modified asphalt binders. Uniaxial Fatigue Testing—S-VECD Analysis Figure 15 shows the correlations between GR and Nf for all the mixtures tested at three temperatures for Binders 3, 8, and 13 and at two temperatures for the other binders. For a given GR, a lower Nf value indicated quicker failure and poorer resistance to fatigue cracking. Previous research found that the number of cycles to failure at GR = 100 correlated well with measured cracking in thick or thin pavements. Figure 16 shows the cycles to failure at GR = 100 for all the mixtures. Asphalt mixtures with polymer-modified asphalt binders generally had higher loading cycles to failure than those with unmodifed or oxidized binders. Detailed results of S-VECD analysis are presented in Appendix E. Coefficients C11 and C12 (Equation 7) at different temperatures were given for each mix in Table 8. Coefficient C12 was always higher at the lower test temperature than at the higher test temperatures, indicating that the average C-S curve was steeper at the lower test temperature for each mix. 1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 G R Cycles to Failure A B C D E F G H I J K L M N O P Figure 15. GR values versus cycles to failure. 0 5,000 10,000 15,000 20,000 25,000 A B C D E F G H I J K L M N O P C yc le s to F ai lu re Binder ID Figure 16. Cycles to failure at GR = 100.

Findings and Applications 39   Mixture Fatigue Testing—GFTAB Analysis In applying the GFTAB model to flexural fatigue data, statistical methods were applied to determine FFPR values as outlined in Chapter 2 and discussed in detail in Appendix D. Table 9 is a summary of the non-linear least squares analysis of the flexural fatigue data using the GFTAB model. Table 10 lists model parameter estimates and statistics for the failure envelope (a power law function) and the fatigue exponent coefficient. Additional statistics can be found in Appendix D. The R2 value for the model is good for a fatigue model at 90%. Binder Code Temp C11 C12 Difference of C12 between two temperatures A 6°C 0.001 0.582 12°C 0.003 0.465 0.117 between 6°C and 12°C 18°C 0.005 0.413 0.052 between 12°C and 18°C B 21°C 0.001 0.536 27°C 0.002 0.490 0.047 C 9°C 0.002 0.505 15°C 0.007 0.391 0.114 D 12°C 0.001 0.588 18°C 0.002 0.476 0.112 E 24°C 0.004 0.426 30°C 0.007 0.384 0.042 F 19.5°C 0.001 0.579 25.5°C 0.002 0.479 0.100 G 30°C 0.004 0.410 36°C 0.008 0.369 0.041 H 21°C 0.002 0.483 27°C 0.003 0.438 0.045 I 12°C 0.001 0.543 18°C 0.002 0.499 0.044 J 15°C 0.002 0.498 21°C 0.004 0.424 0.074 K 12°C 0.001 0.605 18°C 0.002 0.470 0.135 between 12°C and 18°C 24°C 0.003 0.447 0.023 between 18°C and 24°C L 27°C 0.002 0.490 33°C 0.003 0.456 0.034 M 24°C 0.006 0.401 30°C 0.007 0.379 0.022 N 21°C 0.008 0.380 27°C 0.012 0.351 0.028 O 15°C 0.001 0.522 21°C 0.002 0.488 0.035 between 15°C and 21°C 27°C 0.002 0.457 0.031 between 21°C and 27°C P 21°C 0.001 0.525 27°C 0.001 0.510 0.015 Table 8. Regression coefficients and C-S fit functions. Source Sum of Squares Degrees of Freedom Mean square F-value p-value Model 131.175 17 7.71616 113.4 < 0.001 Error 14.899 219 0.06803 Total 146.074 237 R2 89.8% Table 9. Summary of non-linear least squares model for flexural fatigue.

40 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures Figure 17 is a plot showing observed and predicted cycles to failure, coded for data source (NCHRP 09-59 or SHRP). Figure 18 is a plot of FFPR values determined from the flexural fatigue tests, whereas Figure 19 shows FFPR values determined from uniaxial fatigue tests. Both plots show confidence limits of 2 standard deviations for the FFPR values. The various binders appear to be well differentiated. Figure 20 is a plot comparing FFPR values from flexural fatigue with those determined from uniaxial fatigue testing. The plot includes confidence limits for the data points. Given the typical variability in mixture fatigue data, the FFPR values are in reasonable agreement. Confidence limits for seven of the nine data points overlap with the line of equality, and the r2 value is 75%. Binder Test Results Results of binder LAS and SDENT tests are summarized in Table 11, which lists FFPR values for the LAS and SDENT and binder R-values as determined from DSR testing of the binders. Also included in this table, for comparison purposes, are FFPR values from flexural and uniaxial fatigue testing. Note that only 9 of the 16 binders selected for NCHRP 09-59 were tested using flexural fatigue. These parameters are compared among each other and with other parameters in the next section of this report. Comparison of Mixture Fatigue with Binder Properties Figures 21 through 24 show the various relationships among the mixture fatigue FFPR values and the three binder parameters. In each figure, error bars are shown for both variables, representing ± 2 standard errors. In many cases, the error bars are so small that they become obscured by the data point. Figure 21 shows mix FFPR values as a function of R-value. The figures have been coded according to binder type and source of the mixture FFPR value (flexural or uniaxial fatigue). A good correlation (r2 = 83%) between mixture and binder FFPR is seen in this figure, indicating the R-value is in fact a good indicator of binder performance in both flexural and uniaxial fatigue. Parameter Coefficient Std. Dev. t-value p-value Log failure envelope coefficient 6.649 0.1669 33.84 0.000 Failure envelope exponent −0.806 0.0203 −39.62 0.000 Table 10. Flexural fatigue model parameter estimates. R² = 88% R² = 90% 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 1.E+02 1.E+04 1.E+06 1.E+08 Pr ed ic te d Cy cl es to F ai lu re Observed Cycles to Failure NCHRP 9-59 binders SHRP core asphalts Equality Pr ed ic te d Cy cl es to F ai lu re Figure 17. Plot of predicted and observed cycles to failure for mixture flexural fatigue tests.

Findings and Applications 41   0.0 0.5 1.0 1.5 2.0 2.5 A B C I J K M O P AAA AAB AAC AAD AAF AAG AAK AAM Flexural Fatigue FFPR Bi nd er C od e 0.0 0.5 1.0 1.5 2.0 A B C D E F G H I J K L M N O P Uniaxial Fatigue FFPR As ph al t B in de r C od e Figure 18. é 2s Confidence limits for flexural fatigue FFPR values. Figure 19. é 2s confidence limits for uniaxial fatigue FFPR values.

42 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures R² = 75% 0.40 0.60 0.80 1.00 1.20 0.40 0.60 0.80 1.00 1.20 FF PR fr om U ni ax ia l T es tin g FFPR from Flexural Fatigue Figure 20. Comparison of FFPR from flexural fatigue testing with FFPR from uniaxial fatigue testing. Error bars show pooled error limits of 2 standard deviations. Binder Code Mix FFPR/ Flexural Fatigue Mix FFPR/ Uniaxial Fatigue R-Value Binder FFPR/ SDENT Extension Binder FFPR/ LAS Test A 0.577 0.618 3.140 0.581 0.559 B 0.721 0.742 2.440 0.839 0.956 C 1.024 0.959 2.280 1.075 0.934 D 1.036 2.100 0.934 0.880 E 0.617 2.910 0.759 0.686 F 0.690 2.490 0.773 0.800 G 0.747 2.450 1.313 0.804 H 0.750 2.640 0.978 0.746 I 1.007 1.004 2.100 0.941 0.944 J 0.839 0.834 2.320 1.123 0.900 K 0.787 0.767 2.650 0.739 0.720 L 0.367 3.060 0.582 0.670 M 0.915 0.659 2.560 1.016 0.955 N 0.534 3.210 0.629 0.652 O 1.128 0.859 2.290 1.106 0.973 P 0.645 0.569 2.580 0.812 0.781 AAA 1.108 1.730 1.248 1.142 AAB 0.952 2.080 0.960 0.921 AAC 1.047 2.010 1.048 0.939 AAD 1.146 1.770 1.057 1.200 AAF 1.339 1.930 1.167 1.014 AAG 2.008 1.350 1.545 1.242 AAK 1.443 1.780 1.108 1.170 AAM 0.810 2.440 0.759 0.894 Est. Std. Error % 4.9 4.5 0.63 1.57 4.98 Table 11. Binder R-values and FFPR values.

Findings and Applications 43   R² = 83% 0.0 0.5 1.0 1.5 2.0 2.5 0.0 1.0 2.0 3.0 4.0 M ix tu re F ati gu e FF PR R-value SHRP (non-modified) BBF non-modified BBF polymer modified Uniaxial non-modified Uniaxial polymer modified R² = 81% R² = 96% 0.0 0.5 1.0 1.5 2.0 2.5 0.0 1.0 2.0 3.0 4.0 SD EN T Ex t. FF PR R-value SHRP (non-modified) NCHRP 9-59 non- modified NCHRP 9-59 polymer- modified R² = 63% 0.0 0.5 1.0 1.5 2.0 2.5 0.0 0.5 1.0 1.5 2.0 M ix tu re F ati gu e FF PR SDENT Extension FFPR SHRP (non-modified) BBF non-modified BBF polymer modified Uniaxial non-modified Uniaxial polymer modified Figure 21. Mixture FFPR as a function of Christensen-Anderson R-value. Error bars represent é 2 ë standard error. Figure 22. FFPR from SDENT extension as a function of R-value. Error bars represent é 2 ë standard error. Figure 23. Mixture FFPR as a function of FFPR from SDENT extension. Error bars represent é 2 ë standard error.

44 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures Furthermore, it appears to be valid for both polymer-modified and non-modified binders. This finding initially puzzled the research team, but they now believe that the effect of polymer modification on mixture fatigue in highly aged mixtures at intermediate to low temperatures is reduced compared with the effect in binder tests on lightly to moderately aged binders at intermediate temperatures. For instance, as will be discussed, polymer modification seems to be significantly more effective in improving SDENT FFPR values than those determined from mixture fatigue tests, probably because of the lower loading rates used in the SDENT test, which results in significantly lower binder modulus values and enhanced polymer effectiveness. When the strong correlations between R-value and mixture FFPR were first observed, they were feared to be an artifact of the GFTAB model: in GFTAB R-value affects the binder phase angle, which in turn affects the fatigue exponent, which then might affect the FFPR value. However, Figure 22, in which FFPR values based on SDENT extension are plotted against R-values, shows a similar relationship to that seen in Figure 21. The relationship here between FFPR from the SDENT test and R-value for non-modified binders is exceptionally good, while the correlation for polymer-modified binder is slightly weaker and shifted with respect to the relationship for binders that are not polymer modified. As discussed, the shift in the polymer- modified data is probably because the somewhat slow loading rate in the SDENT results in enhanced polymer effectiveness compared with the mixture fatigue tests. It is hypothesized that the effectiveness of polymer modifiers in the NCHRP 09-59 mixture tests was attenuated for three reasons: (1) the large number of stress concentrations present in asphalt concrete mixtures, (2) the heavy laboratory aging used to condition the mixes before testing, and (3) the resulting higher modulus values of the mixtures during fatigue testing. If the mixes had been only short-term aged and tested at slightly higher temperatures, the effect of the polymer modification would probably have been more pronounced. In any case, what is important in Figure 22 is that increasing R-value seemingly decreases the strain tolerance, even for the polymer-modified systems. That is, high R-values will always decrease the strain tolerance of a binder, even when it is polymer modified. In fact, Figure 22 suggests that high R-values reduce strain tolerance of modified binders more than non-modified ones. Figure 23 is a plot of FFPR from the SDENT extension test as a function of R-value. The overall r2 value here of 63% is only moderate, but that is partly because performance of the polymer-modified binders was different as measured by the SDENT test and as indicated in the mixture fatigue tests. If data from only non-polymer-modified binders are included, the r2 value increases to 79%. It appears that the SDENT test tends to overestimate performance for polymer-modified binders relative to their performance in mixture fatigue tests. R² = 98% R² = 73% R² = 56% R² = 91% 5 10 15 20 25 1.E+05 1.E+06 1.E+07 D EN T Ex te ns io n, m m |G*|, Pa R < 2.2 2.2 < R < 3.0 R > 3.0 Polymer modified Figure 24. DENT extension as a function of estimated binder modulus, coded for polymer modification and binder R-value.

Findings and Applications 45   As mentioned, there are several reasons for this but the most important is perhaps the relatively slow loading rate used in the SDENT test compared with mixture fatigue loading. The resulting relatively low binder stiffness in the SDENT test seems to improve the performance of most polymers used in the NCHRP 09-59 binders. This does not mean that the SDENT test is wrong or inaccurate—it simply means that the conditions used in the SDENT test are signifi- cantly different from those used in mixture fatigue tests, and as a result, the apparent perfor- mance of polymer-modified binders differs in these two tests. SDENT testing for NCHRP 09-59 was performed at as low a temperature as possible—lower test temperatures would produce quick, brittle fractures, providing little useful information on the binder. In many cases, testing at lower temperatures would be impossible using a water bath. One reason for the discrepancy between SDENT tests and mixture fatigue is made clear by Figure 24, which shows SDENT extension as a function of binder modulus, coded for polymer modification and R-value. As modulus increases, the effectiveness of polymer modification appears to decrease, until at approximately 30 MPa the modification no longer seems to influ- ence the results of the DENT test. The mixture fatigue tests were performed at binder modulus values ranging from about 2 to 85 MPa, so the effectiveness of the polymer modification was likely minimal because of binder modulus values were high during the tests. The extended loose-mix aging used in NCHRP 09-59 may also have had an adverse effect on the performance of polymer-modified binders. Had mixture fatigue tests been done at a significantly lower modulus range or using less aggressive laboratory aging methods, significant benefit may have been observed from polymer modification. R-value adequately characterizes fatigue/fracture performance for non-modified binders and for polymer-modified binders in the intermediate-high modulus range (especially when heavily aged), but probably does not adequately characterize the fatigue/fracture performance of polymer-modified binders in the intermediate-low modulus range. For a completely accurate characterization of fatigue/fracture performance, R-value should be controlled for all binders, and SDENT tests should be performed on polymer-modified binders. Both SDENT and R-value should be controlled for polymer-modified binders because at lower temperatures, especially for heavily aged mixes, polymer modification appears to become ineffective; then the fatigue/ fracture performance will be a function of R-value alone. If only SDENT test data are used in a specification, the fatigue and fracture performance of polymer-modified binders at low temperatures could become inadequate. Figure 25 shows the relationship between mixture fatigue FFPR values and LAS FFPR. The r2 value for this relationship, as in Figure 23, is only moderate (68%). In this case, the data appear to be similar for polymer-modified and non-modified binders; the reason for the relatively poor correlation appears to be the high variability of FFPR calculated from the LAS test. The relationship between mix FFPR and LAS FFPR also appears to follow a power law, while ideally it should be linear—in fact, the two sets of FFPR values should be approximately equal. This non-linear relationship is probably a result of the high strains and resulting non-linear behavior occurring during the LAS test. Comparisons for NCHRP 09-59 Binders The previous figures provide an overall picture of the relationships among binder and mixture test data. But, they are dense and complicated, so the more important relationships for the NCHRP 09-59 binders are hard to pick out. Figures 26 through 29 show the relationships for these binders only. Figure 26 shows FFPR from uniaxial fatigue as a function of R-value; the r2 value is moderate at 73%. Figure 27 shows uniaxial fatigue FFPR as a function of SDENT FFPR; in this case, the r2 value is even lower at only 38%. Figure 28 is a plot of FFPR from flexural

46 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures R² = 68% 0.0 0.5 1.0 1.5 2.0 2.5 0.0 0.5 1.0 1.5 M ix tu re F ati gu e FF PR LAS FFPR SHRP (non-modified) BBF non-modified BBF polymer modified Uniaxial non-modified Uniaxial polymer modified R² = 73% 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 2.0 2.5 3.0 3.5 M ix tu re F ati gu e FF PR R-value Uniaxial non-modified Uniaxial polymer modified R² = 38% 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.0 0.5 1.0 1.5 M ix tu re F ati gu e FF PR SDENT Extension FFPR Uniaxial non-modified Uniaxial polymer modified Figure 25. Mixture FFPR as a function of FFPR from LAS test data. Error bars represent é 2 ë standard error. Figure 26. Mixture FFPR from uniaxial fatigue as a function of Christensen-Anderson R-value. Figure 27. Mixture FFPR from uniaxial fatigue as a function of SDENT FFPR.

Findings and Applications 47   fatigue as a function of R-value; the r2 value in this case is moderate at 66%. The final plot in this series, Figure 29, shows flexural fatigue FFPR as a function of SDENT FFPR. The r2 value in this case is identical to that of Figure 28, at 66%. Correlations between LAS FFPR and uniaxial and flexural fatigue were 45% and 57%, respectively. Correlations once the SHRP binders are eliminated are lower, which is to be expected because this reduces the range in the data set. It is probably unrealistic to expect higher R-values than observed here because asphalt mixture fatigue testing is by nature highly variable. These plots confirm correlations between FSC and both binder R-value and SDENT FFPR. From these figures, it appears that the relationship between mixture fatigue performance as measured by FFPR and R-value is slightly better than for SDENT FFPR. This is possibly because the slow loading rate of the SDENT test results in an improvement in the performance of polymer- modified binders compared with what occurs at the faster loading rates during fatigue testing. Unfortunately, conducting the SDENT test at higher loading rates would be difficult, or even impossible. It should be emphasized that this finding is for heavily aged mixtures and binders. Materials conditioned with less-severe aging could produce different results because the effect of the polymer modification might then be more pronounced. In this case, SDENT FFPR might provide a better indication of fatigue performance because of its sensitivity to polymer modification. This is an important subject for further research. R² = 66% 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 2.0 2.5 3.0 3.5 M ix tu re F ati gu e FF PR R-value BBF non-modified BBF polymer modified R² = 66% 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.0 0.5 1.0 1.5 M ix tu re F ati gu e FF PR SDENT Extension FFPR BBF non-modified BBF polymer modified Figure 28. Mixture FFPR from flexural fatigue as a function of Christensen-Anderson R-value. Figure 29. Mixture FFPR from flexural fatigue as a function of SDENT FFPR.

48 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures Healing Experiment Results of the healing experiment are shown in Figures 30 through 32. Figure 30 shows healing ratio as a function of binder phase angle at 0.13 rad/s; this frequency represents an approximate equivalent to the recovery time for the pulse loading. The correlations for both non-modified and polymer-modified binders are poor. Figure 31 shows healing ratio as a function of phase angle at 10 Hz, corresponding to the actual loading frequency. In this case, the correlation for the non-modified binders is good at 83%, but the correlation for the polymer-modified binders is poor at only 3%. Figure 32 shows healing ratio as a function of binder storage modulus at 10 Hz. In this case, the correlation for the non-modified binders is moderate (59%), but the correlation for polymer-modified binders is again weak (6%). NCHRP Project 09-44A, R² = 30% R² = 1% -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 20 30 40 50 60 70 H ea lin g Ra tio Phase Angle at 0.13 rad/s, degrees Non-modified Polymer-modified R² = 59% R² = 6% -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 H ea lin g Ra tio G' at 10 Hz, kPa Non-modified Polymer-modified Figure 30. Mixture healing as a function of binder phase angle at 0.13 rad/s. R² = 83% R² = 3% -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 10 20 30 40 50 60 H ea lin g Ra tio Phase Angle at 10 Hz, degrees Non-modified Polymer-modified Figure 31. Mixture healing as a function of binder phase angle at 10 Hz. Figure 32. Mixture healing as a function of binder storage modulus at 10 Hz.

Findings and Applications 49   “Validating an Endurance Limit for HMA Pavements: Laboratory Experiment and Algorithm Development,” found a relationship between mixture and binder stiffness and healing and endurance limit, in which increasing healing is associated with decreasing binder stiffness (see Witczak et al. 2013). Figure 33 shows healing ratio as a function of binder |G*|; the relationship is similar to that shown in Figure 32 for storage modulus, as should be expected. Figure 34 shows the relationship between healing ratio and initial mix modulus. In this case, the correlation is even worse— 42% for non-modified binders and only 4% for polymer-modified binders. Precise comparisons between the relationships shown in Figures 33 and 34 and the findings of NCHRP 09-44A (Witczak et al. 2013) are difficult because the rheology of the binders used in NCHRP 09-44A was not documented; only binder grades were reported. Also, the strength of the relationship between stiffness and healing was not directly described. It seems likely that the discrepancy in findings is because of differences between the two proj- ects in the rheology of the binders. NCHRP 09-44A used three binders, and it appears that none was polymer modified (Witczak et al. 2013). The rheological behavior of these three binders may have been similar, so that binder and mix stiffness would be highly correlated to binder phase angle. The inclusion of polymer-modified binders and non-modified binders with a wide range of rheological behaviors in NCHRP 09-59 means the relationship between binder phase angle and binder or mix stiffness is weak. The results of the healing experiment strongly suggest that healing is a function of phase angle for non-modified binders. For modified binders, phase angle does not appear to correlate R² = 57% R² = 6% -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 100 1,000 10,000 100,000 1,000,000 H ea lin g Ra tio |G*| at 10 Hz, kPa Non-modified Polymer-modified R² = 42% R² = 4% -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 10 100 1,000 10,000 H ea lin g Ra tio Initial Mix Modulus, ksi Non-modified Polymer-modified Figure 33. Mixture healing as a function of binder |G*| at 10 Hz. Figure 34. Mixture healing as a function of initial mix modulus at 10 Hz.

50 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures with healing. This is probably because of the more complex rheology and chemistry of polymer- modified binders. Directly measuring binder adhesion might provide a more reliable indication of healing potential for polymer-modified binders. That the phase angle at 10 Hz correlated better to healing than the phase angle at 0.13 rad/s is somewhat puzzling and seems to indicate that healing in asphalt binders has more to do with the nature of the fatigue damage as it is accumu- lating rather than the time allowed for recovery. In evaluating the practical significance of binder healing, it should be understood that heal- ing is more important during stress-controlled loading, that is, the sort of loading that occurs in thick pavement structures. Under these conditions, strains in the bound layers depend largely on their modulus rather than the underlying structure. As damage accumulates, the modulus of the bound layers will decrease, resulting in an even more rapid accumulation of damage. Healing will tend to reduce the magnitude of this feedback loop. In thin pavements, the loading is likely to be more strain controlled, that is, the deflections and strains in the bound layers are controlled by the subgrade and subbase and not by the bound layers. In this situation, the effect of healing on pavement strain and fatigue life will be less than for thicker, stress-controlled pavements. In this experiment, complete healing does not necessarily mean that a mix can be loaded indefinitely without failure—it means that the modulus of the mix does not significantly decrease during fatigue loading. Indications from this experiment suggest that the normal GFTAB model still applies to mixes that heal substantially during loading. Although complete or nearly complete healing will not confer an infinite fatigue life on a mixture, because the modulus does not decrease during loading, the strains will not increase and the fatigue life will improve significantly compared with a mixture with no healing. Calculation of the precise effect of healing on in situ pavement life is complicated because it depends on both the potential increase in strains that would occur in the pavement without healing and the degree of heal- ing that occurs, reducing this increase in strains. Furthermore, both factors depend strongly on loading frequency and temperature, as do many other factors affecting the fatigue life of flexible pavements. Disentangling the effects of pavement modulus, strain capacity, fatigue exponent, and healing in real pavement systems is a difficult task that cannot be addressed in NCHRP 09-59. Additional research is needed to better understand healing in asphalt binders and mixtures and how it affects the behavior of actual pavements. This is especially true for polymer-modified binders. Layered Elastic Analyses The results of the laboratory testing as described show that R-value is directly linked to the strain tolerance of asphalt binders. And because the fatigue exponent is proportional to the phase angle, which is also related to the R-value, the strain sensitivity of asphalt binders is also related to the R-value—as the R-value increases, strain sensitivity increases. Thus, a complex interaction of factors affects the fatigue performance of asphalt mixtures. As R-value increases, strain tolerance decreases, which decreases fatigue performance. At the same time, increasing R-values tend to increase the fatigue exponent, which will improve fatigue performance. In actual pavements, the situation becomes even more complicated because the strain in the pave- ment depends on the modulus of the mixture, which depends on the modulus of the binder. Strain levels will also depend on the overall structural stiffness of the binder. To understand how all these factors interact to affect fatigue performance, hypothetical pavement systems must be analyzed to see how binder rheology and pavement structure interact to affect fatigue performance.

Findings and Applications 51   A layered elastic analysis was performed on two simple hypothetical pavement systems. The “thin” pavement structure consisted of 100-mm of asphalt concrete over a 200-mm sub- base with a modulus of 150 MPa, resting on a subgrade with a modulus of 80 MPa. The “thick” pavement structure consisted of 200-mm asphalt concrete over a 300-mm subbase; the modulus of the subbase and subgrade were the same as for the thin structure. Asphalt concrete modulus values were estimated using the improved Hirsch model (Christensen and Bonaquist, 2015). The mixture was assumed to have a VMA value of 17% and a VFA (voids filled with asphalt) value of 60%, corresponding to a Vbe value of 10.2% and an air void content of 6.8%. Tensile trains at the bottom of the bound layers were determined using an Excel-based layered elastic analysis program (Levenburg, 2016). FFPR was estimated from the relationship between FFPR and R-value as shown in Figure 21. The fatigue life was determined using the GFTAB model described earlier in this report, specifically Equation 3, reproduced here for convenience: = × ε     ( )δ* (3) 901 N FFPR FSC f k where Nf = number of cycles to failure; FFPR = fatigue/fracture performance ratio, equal to the ratio of the strain capacity of a given binder to the average or typical strain capacity, = 3.19 R−1.56; FSC* = Average or typical fatigue strain capacity (%), analogous to failure strain, = 4.45 × 106 |G*|−0.806; ε = effective strain in binder (%), = mix strain/(Vbe/100); k1 = constant determined through statistical analysis to be 2; and δ = binder phase angle (degrees). Healing was not included in this analysis because it is not clear how to model healing when polymer-modified binders are used in a mixture or whether healing as modeled in laboratory experiments is reflected in field performance. The FSC* values were multiplied by 0.67 to provide reasonable values for cycles to failure (on the basis of engineering judgment). The range in binder modulus values was limited—from 1 to 300 MPa—to avoid extrapolating the model beyond the range of the data used in its development. The binders used in the analysis included four of the SHRP core asphalts (AAD-1, AAG-1, AAK-1, AAM-1) and the nine binders used in NCHRP 09-59 in the flexural fatigue experiment. The SHRP binders were included to provide several short-term (RTFOT) aged binders in addition to the long-term aged NCHRP 09-59 binders. The primary results of this analysis are shown in Figures 35 and 36. These plots show cycles to failure as a function of binder modulus for the thin and thick pavement systems. As shown in Figure 35, for thin pavement structure, the fatigue life decreases with increasing modulus. The fatigue life seems little affected by R-value when the modulus is below about 30 million MPa but decreases strongly with increasing R-value at higher modulus values. In Figure 36, for thick pavement structure, modulus has little effect on fatigue life up to about 30 million MPa, after which the fatigue life increases substantially with increasing modulus. In general, increasing R-values tend to increase fatigue life for the thick pavement. The effect of both modulus and R-value on pavement fatigue life depends on pavement structure, with the effects of these factors essentially reversing from thin structures to thick structures.

52 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures Layered Elastic Analysis and Potential Fatigue Specification Parameters Is there a rheological parameter that relates well to the fatigue life of a pavement? Examples are the current binder fatigue specification parameter, |G*| sin δ, and the GRP. Figure 37 shows fatigue life plotted against |G*| sin δ, while Figure 38 shows fatigue life plotted against GRP. Binders with extreme R-values (below 2.0 and above 3.0) have been removed from this plot, because these types of binders can both lead to poor fatigue performance, depending on the pavement structure. The GRP relates somewhat better to fatigue life than |G*| sin δ, although the difference does not appear to be that great. GFTAB theory suggests a third parameter for predicting the fatigue performance of asphalt binders: |G*| (R/2)2. This parameter suggests itself because of the inverse relationship between failure strain and modulus, and because of the relationship between FFPR and R-value. Figure 39 is a plot of fatigue life against |G*| (R/2)2, again for the thin pavement structure. This shows a slightly better relationship than for the GRP. This parameter, not surprisingly, does well in this analysis, considering the analysis was largely based on GFTAB theory. Figures 40 and 42 show the same series of plots as the previous three figures, but for the thick pavement structure. In this case, fatigue life increases dramatically at higher modulus values. The three parameters, |G*| sin δ, GRP, and |G*| (R/2)2, all seem to relate closely to fatigue life. Considered together, the layered elastic analysis suggests all three of these parameters are potentially useful as binder fatigue specification parameters, although |G*| sin δ does not seem to relate as well to fatigue life for thin pavements as the other two parameters. 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 Cy cl es to fa ilu re Binder |G*|, Pa R < 2.0 2 < R < 2.5 2.5 < R < 3 3 < R Figure 35. Layered elastic analysis: Thin pavement, cycles to failure as a function of binder modulus, coded for R-value. 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 Cy cl es to fa ilu re Binder |G*|, Pa R < 2.0 2 < R < 2.5 2.5 < R < 3 3 < R Figure 36. Layered elastic analysis: Thick pavement, cycles to failure as a function of binder modulus, coded for R-value.

Findings and Applications 53   1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Cy cl es to fa ilu re 2 < R < 2.5 2.5 < R < 3 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 |G*| sin δ, kPa 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 Cy cl es to fa ilu re 2 < R < 2.5 2.5 < R < 3 GRP, kPa 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Cy cl es to fa ilu re 2 < R < 2.5 2.5 < R < 3 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Binder |G*| (R/2)2, kPa Figure 37. Layered elastic analysis: Thin pavement, cycles to failure as a function of |G*| sin c, coded for R-Value. Figure 38. Layered elastic analysis: Thin pavement, cycles to failure as a function of Glover-Rowe parameter, coded for R-value. Figure 39. Layered elastic analysis: Thin pavement, cycles to failure as a function of |G*| (R/2)2, coded for R-value.

54 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Cy cl es to fa ilu re 2 < R < 2.5 2.5 < R < 3 |G*| sin δ, kPa 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 Cy cl es to fa ilu re 2 < R < 2.5 2.5 < R < 3 GRP, kPa 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Cy cl es to fa ilu re Binder |G*| (R/2)2, kPa 2 < R < 2.5 2.5 < R < 3 Figure 40. Layered elastic analysis: thick pavement, cycles to failure as a function of |G*| sin c, coded for R-value. Figure 41. Layered elastic analysis: thick pavement, cycles to failure as a function of Glover-Rowe parameter, coded for R-value. Figure 42. Layered elastic analysis: thick pavement, cycles to failure as a function of |G*| (R/2)2, coded for R-value.

Findings and Applications 55   In the previous discussion, the terms “thin” and “thick” pavement refer to the specific structures used in the layered elastic analysis: “thin” meaning the pavement with 100 mm of bound material over a 200-mm subbase, and “thick” meaning the pavement with 200 mm of bound material over a 300-mm subbase. In the remainder of this report, these terms are for the most part used in a more general sense: “thin” pavements are those in which the strains are relatively high and mostly controlled by the pavement subgrade, while “thick” pavements are those in which the strains are relatively low and controlled by the stiffness of the bound layers. As a general guideline, pavements with bound material less than 100 mm thick probably act like thin pavements, while those with 200 mm or more of bound material probably behave like thick pavements. Pavement of intermediate thickness probably exhibits behavior in between these extremes. However, how a pavement responds to traffic loads will also depend on the quality of the subgrade. Additional research is needed to better define the factors that deter- mine whether a pavement is functioning in a “thin” or a “thick” manner. Effect of Improper Grade Selection on Fatigue Life In implementing the current binder grade specification, some state highway departments select binder grades that do not necessarily reflect the grades the FHWA LTPPBind software recommends. The reasons for such decisions are not addressed in this report. However, engi- neers need to understand the potential impact of such decisions on the fatigue life of flexible pavements. If the asphalt binder used in pavements is changed by selecting a stiffer binder grade, fatigue life will tend to decrease because the binder will have overall lower FSC. This can, however, be mostly offset by using a thicker layer of bound material in the pavement. It would be useful to have a rough understanding of how much of an increase in thickness would be needed to offset a given increase in binder grade. The layered elastic analysis was used to answer this question by comparing the relationship between fatigue life and temperature, relative to the estimated cracking temperature for the normal case, with cases for which the temperature has been decreased by 6°C and 12°C. This would be equivalent to specifying a stiffer binder over the optimal grade by one and two grades, respectively. Trial and error was then used to determine how much of an increase in pavement thickness would be needed to offset the resulting loss in fatigue life. As in the previous analysis, binders with extremely high and low R-values have been excluded from this analysis. Figures 43 and 44 show examples of this process. Figure 43 compares fatigue life for the thin pavement structure at normal temperatures with fatigue life at a temperature decreased by 6°C 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Fa tig ue L ife , 1 00 m m Fatigue Life, -6 C and 120 mm 2 < R < 2.5 2.5 < R < 3 Equality Figure 43. Fatigue life at normal temperature versus fatigue life with 6çC decrease in temperature and 20 mm increase in pavement thickness (thin pavement structure).

56 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures and a pavement thickness (bound material) increased by 20 mm. Figure 44 is a similar plot but for a 12°C decrease in temperature and a 30-mm increase in thickness. The analysis for the thick pavement structure showed a different effect—in most cases, a decrease in temperature showed an improvement in fatigue life. The conclusion here is that for thin pavement tempera- tures, an increase in the low or intermediate temperature grade (equivalent to a decrease in temperature) should be accompanied by a significant increase in pavement thickness if pre- mature fatigue failures are to be avoided. This finding is particularly important in colder climates, where commonly used binder grades often do not strictly meet the requirements given in the LTPPBind software. The relationship between low- to intermediate-temperature binder grade and fatigue life is important in understanding the shortcomings in the current system. Any substantial mis- match between the low to intermediate binder grade selection and what is warranted on the basis of local climate will affect the fatigue performance of a thin asphalt concrete pavement significantly—similar to a change in pavement thickness of 20 to 30 mm. Such grade mismatches appear to be common in cold climates, where the LTPPBind software suggests a low design temperature of −40°C, but highway departments specify PG 58-34 or even PG 58-28 binders. It is not surprising that premature fatigue occurs in such situations. Highway departments should understand that if they choose to use binder grades stiffer than warranted by their local climate, their pavements will be prone not only to transverse cracking but also to premature fatigue cracking, unless their typical pavement thicknesses are increased. Grade mismatch also occurs because the current system calculates the intermediate temperature base on the average of the high and the low grading temperatures. This works well if the upper binder grade has not been adjusted for traffic; however, if the upper grade is adjusted for traffic, the intermediate specification temperature will be elevated above what is needed to control fatigue properties. The effect of such grading errors is probably more severe than that caused by using a less-than- optimal fatigue specification parameter, |G*| sin δ, as will be discussed. Analysis of Potential Binder Fatigue Specification Parameters The previous analysis suggests that the current binder fatigue specification should be revised. Both GRP and |G*| (R/2)2 would appear to be improvements over |G*| sin δ in indicating fatigue performance. These could both be implemented easily by simply replacing the cur- rent |G*| sin δ maximum value of 5,000 kPa with an appropriate maximum value for the 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Fa tig ue L ife , 1 00 m m Fatigue Life, -12 C and 130 mm 2 < R < 2.5 2.5 < R < 3 Equality Figure 44. Fatigue life at normal temperature versus fatigue life with 12çC decrease in temperature and 30-mm increase in pavement thickness (thin pavement structure).

Findings and Applications 57   selected new fatigue parameter. Significant evidence has also shown that fatigue performance is related to R-value and that an allowable range for R-value would also help ensure good fatigue performance. The previous analyses excluded binders with extreme R-values because these can lead to poor performance: binders with high R-values perform poorly in thin pavements, and binders with low R-values perform poorly in thick pavements. In the previous analyses, binders with R-values below 2 or above 3 were eliminated. However, several polymer-modified binders tested during NCHRP 09-59 had R-values over 3.0; only one had a value over 3.2. A specification range of 2.0 to 3.2 would allow most polymer-modified binders while eliminating several binders selected as examples associated with poor performance. The proposed range for R would produce good control over fatigue life when another parameter, such as GRP or |G*| (R/2)2 is also controlled. This range in R-values is probably appropriate for binder aged with RTFOT and 40-hour PAV. However, the range is probably too high for binders aged using RTFOT and 20-hour PAV. For example, among the SHRP binders, this would be the Boscan binder (AAK-1), which has a record of good performance. It would not eliminate the West Texas intermediate binder (AAM-1), which has a history of problems with fatigue and non-load associated surface cracking. Shifting the allowable range for R-value to a range of 1.50 to 2.50 for RTFOT/20-hour PAV aging would correct this problem. In setting tentative limits on binder R-value (or ΔTc values), some consideration should be given to general experience over the past few years with premature failures in asphalt pavements caused by cracking, especially nonload-associated cracking. Essentially, all recent research has shown that highly negative ΔTc values (corresponding to high R-values) are asso- ciated with increased nonload-associated cracking in flexible pavements, while positive ΔTc values (low R-values) are associated with adequate or even good performance. This suggests that high R-values and high negative ΔTc values are from a practical standpoint a more serious problem than low R-values and positive ΔTc values. This is perhaps because more thin asphalt pavements are being built than thick, full-depth pavements. Also, the described layered elastic analysis suggests that the failure of thin flexible pavements made with binders that have high R-values can sometimes be rapid. That is, these failures are generally more dramatic than failures resulting from the use of binders with low R-values in thick pavements. In any case, although theoretical evidence indicates that there should be both high and low limits on R or ΔTc, recent experience suggests that an upper limit on R and a lower limit on ΔTc is more important. Another concern with setting tentative limits on R-value is whether such limits should be the same for polymer-modified binders, or if such limits should be established at all for adequately modified binders. The primary issue, as discussed in the preceding paragraph, is the potentially poor performance of binders with high R-values when used in high-strain applica- tions (such as thin pavements) at low temperatures. Polymer-modified binders, in general, may exhibit higher failure strains at low temperatures compared with similar non-modified binders, although NCHRP 09-59 did not find evidence of this. If this is the case, then higher R-values can and should be permitted for polymer-modified binders. One way of accomplishing this would be through the AASHTO M 332 specification (using the multiple-stress creep recovery test for high-temperature binder grading). The maximum binder R-value could be increased for the binder meeting the “V” and “E” requirements, assum- ing that substantial polymer modification would be required for such binders to meet this performance level. However, there are other concerns with binders having high R-values. Recent research conducted as part of NCHRP 09-60, for instance, shows a clear increase in physical hardening with decreasing (more negative) ΔTc values (Pascal et al., 2019). This finding would argue against

58 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures relaxing R-value requirements for polymer-modified binders. Another reason for maintaining the same R-value limits for polymer-modified binders is that these materials are expected to exhibit superior performance—if the maximum R-value is relaxed for polymer-modified binders, some materials might exhibit no better fatigue resistance than non-modified binders. Considering these various factors, the NCHRP 09-59 research team proposes that the issue of polymer-modified binders and the use of R-value (or ΔTc or some other similar parameter) in an improved binder specification be reevaluated after the completion of NCHRP 09-60, which is collecting and analyzing significantly more data on binder rheology and performance-related test data. Before suggested values for GRP and |G*| (R/2)2 can be addressed, the physical significance of these parameters needs to be understood. Both are good indicators of the failure strain of a binder under a given loading temperature and loading frequency. This is clear from Figure 45, which shows calculated FSC as a function of GRP for the flexural fatigue experiment data. GRP is calculated at the fatigue loading frequency of 10 Hz and at the test temperature for each specimen. The r2 value of 96% is extremely high, demonstrating that GRP is an excellent indicator of binder strain tolerance. This shouldn’t be surprising because GRP was originally developed as a surrogate for ductility. Figure 46 shows FSC as a function of |G*| (R/2)2, with a similarly strong correlation. Both GRP and |G*| (R/2)2 are effective as fatigue specification parameters because they are good predictors of binder failure strain. Figure 47 shows FSC as a function of the current fatigue specification parameter, |G*| sin δ, and the r2 value in this case is lower than for the other two functions at 77%. As a comparison with Figure 45, Figure 46, and Figure 47, Figure 48 shows R² = 96% 0.1 1.0 10.0 100.0 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Fa tig ue S tr ai n Ca pa ci ty , % Glover-Rowe Parameter, kPa Figure 45. FSC as a function of GRP for flexural fatigue experiment data. R² = 98% 0.1 1.0 10.0 100.0 1.E+03 1.E+04 1.E+05 1.E+06 Fa tig ue S tr ai n Ca pa ci ty , % |G*| (R/2)2, kPa Figure 46. FSC as a function of |G*| (R/2)2 for flexural fatigue experiment data.

Findings and Applications 59   R² = 77% 0.1 1.0 10.0 100.0 1.E+03 1.E+04 1.E+05 Fa tig ue S tr ai n Ca pa ci ty , % |G*| sin δ, kPa R² = 74% 0 5 10 15 20 0 20 40 60 80 100 FS C, % SDENT Extension, mm Figure 47. FSC as a function of |G*| sin c for flexural fatigue experiment data. Figure 48. FSC as a function of SDENT extension for flexural fatigue experiment data. FSC from the flexural fatigue experiment as a function of SDENT extension; the r2 value in this case is similar to that for |G*| sin δ, at 74%. There are two complimentary approaches to determining appropriate values for either the GRP or |G*| (R/2)2 in a revised fatigue specification. One approach is to calculate the equivalent values to the current limit on |G*| sin δ of 5,000 kPa. However, this equivalency will depend on the R-value of the binder; if selecting 2.2 as a typical R-value for a PAV-aged binder, the equivalent value of the GRP would be 5,300 kPa. The equivalent value for |G*| (R/2)2 would be 8,700 kPa. The second approach, as discussed in detail in the next section, is to examine test data and performance from pavement test sections and ALFs to determine whether the specification value has some inherent limit above which performance tends to deteriorate. Figure 49 shows phase angle as a function of |G*| for the three potential specification parameters at their respec- tive equivalent limits. Also shown on this plot are lines of equal estimated FSC. This shows a serious shortcoming of |G*| sin δ compared with the other two parameters—it fails to closely control FSC. The GRP and |G*| (R/2)2 both exert good control of binder FSC and should therefore serve as better fatigue specification parameters. A third possible approach to determining appropriate specification values is based on the suggested limits for the GRP of 180 and 450 kPa, representing possible cracking and likely cracking. However, these limits were based on research on thermal cracking and related pave- ment distress by Kandhal (1977) and did not in any way involve fatigue cracking. For this reason, these limits are not considered appropriate for use in developing an improved binder fatigue specification, although they may be useful in addressing thermal cracking, block cracking, and raveling (Kandhal, 1977; Glover et al., 2005).

60 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures Two significant problems in the current binder fatigue specification have been identified in this report to this point: (1) the current parameter, |G*| sin δ, appears to promote the use of binders that have relatively low strain tolerance, and (2) agencies sometimes specify that low-temperature binder grades be stiffer than what is optimal for a given climate, which also increases intermediate stiffness for the binder, potentially decreasing its strain tolerance and fatigue performance. The first problem can be addressed by selecting a fatigue parameter that exerts better control over binder strain tolerance—either GRP or |G*| (R/2)2 would seem to be suitable. The second problem can be addressed by having a clearer way of defining the fatigue test temperature and by making sure state highway departments understand the ramifications of using stiffer binders than are optimal for their local climate. One potential change that would reduce unintentional grade mismatch would be to link the fatigue (intermediate) test temperature to the low PG binder grade, rather than the average of the low and high grading temperatures. This would eliminate the problem of grade bumping artificially increasing the fatigue test temperature. To evaluate the current fatigue test tempera- ture, the LTPPBind software was used to determine the base binder grades for 20 cities across the United States and Canada. The average air temperature was also determined, and this was used to estimate the average pavement temperature using an equation proposed by Huang (1993): 1 1 4 34 4 6 (15)M M z z p a= + +     − + + where Mp = mean average pavement temperature (°F), Ma = mean annual air temperature (°F), and z = depth of desired location within pavement (in.), assumed to be 2.0 in. Table 12 shows the 19 cities used in this analysis, the average pavement temperature deter- mined from Equation 15, the base binder grades as determined from the LTPPBind software, the current binder fatigue test temperature, and the proposed binder fatigue test temperature. The proposed scheme for binder fatigue test temperatures is given in Table 13. This approach was developed to provide an average fatigue test temperature approximately 4°C above the average pavement temperature—approximately what the current system does. However, as shown in Figure 50, the current system is not consistent in the relationship between fatigue Figure 49. Phase angle as a function of |G*| for potential binder fatigue parameters. Dashed lines are lines of equal fatigue strain capacity. |G*| sin c is calculated at 5,000 kPa, GRP at 5,300 kPa, and |G*| (R/2)2 at 8,700 kPa. 20 30 40 50 60 6,000 7,000 8,000 9,000 10,000 Ph as e An gl e, d eg re es |G*|, kPa |G*| sin δ GRP |G*| (R/2)^2 5% 20% 15% 10%

Findings and Applications 61   test temperature and average pavement temperature; the test temperature is too high for harder binder grades and too low for softer grades. Figure 51 shows the relationship for the proposed system. Although more scatter appears, the relationship between average pavement temperature and binder fatigue test temperature is more consistent. In fact, the standard error between the target test temperature (4.0°C higher than the average pavement temperature) and the binder fatigue test temperature is 3.0°C for the proposed system, and 4.0°C for the current system. Not only does the proposed system avoid the problem of high-temperature grade bumping, it provides grades with a more consistent relationship to average pavement temperature. Figure 50 also illustrates a third problem with the current grading system—the fatigue/ intermediate test temperature does not relate well to the average pavement temperature. The City Avg. Pavement Temp. °C PG High- Temp. Grade °C PG Low- Temp. Grade °C Current Fatigue Test Temp. °C Proposed Fatigue Test Temp. °C Montreal, QC 11.8 52 −46 7 15 Augusta, ME 12.0 52 −34 13 19 New York, NY 18.2 58 −28 19 22 Washington, DC 18.5 64 −28 22 22 Raleigh, NC 21.8 70 −28 25 22 Jacksonville, FL 26.4 76 −10 37 29 New Orleans, LA 27.6 76 −10 37 29 Memphis, TN 23.2 70 −28 25 22 Chicago, IL 14.6 58 −40 13 17 Fargo, ND 9.8 58 −46 10 15 Kansas City, MO 19.2 64 −28 22 22 Oklahoma City, OK 22.3 70 −22 28 25 Reno, NV 17.3 52 −28 16 22 Albuquerque, NM 19.4 64 −22 25 25 Denver, CO 14.9 58 −28 19 22 Boise, ID 16.4 58 −28 19 22 Seattle, WA 16.1 52 −22 19 25 Phoenix, AZ 31.0 82 −10 40 29 San Francisco, CA 19.5 58 −10 28 29 Table 12. Binder grading information for Selected North American cities. Low PG Grade °C Proposed Binder Fatigue Test Temp. °C −46 15 −40 17 −34 19 −28 22 −22 25 −16 27 −10 29 Table 13. Proposed binder fatigue test temperatures. Figure 50. Average pavement temperature as a function of binder fatigue test temperature (current grading system). 0 10 20 30 40 50 0 10 20 30 40 50 Av g. P av em en t T em p. , C Current Binder Fatigue Test Temperature, °C

62 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures proposed system would at least partially correct this problem. A better match between fatigue test temperature and average test temperature would probably involve a more complicated system and a significant increase in the number of grades producers would have to supply. As mentioned, because highway agencies may decide to use stiffer binder grades than demanded by their local climate, the proposed specification should explain that this might lead to premature fatigue failures unless thicker pavements are specified. A related problem can occur in mild climates such as the Pacific Coastal climate because the average pavement temperature in such regions is closer to the low PG grade than is usually the case. Under the proposed system, such areas would have to either specify some other intermediate test temperature or simply specify binder grades rated to a lower temperature than would otherwise be required in their climate. For example, on the California coast, a PG 64-10 might meet the requirements to resist thermal cracking but would not provide adequate insurance against fatigue cracking. Therefore, a PG 64-16 should be specified. One way to check for such problems is to calculate the average pavement temperature; if the binder fatigue test temperature is more than 7°C higher, the low PG grade should be lowered. A note addressing this problem should also be included in the proposed specification change. A fourth, important problem in the current grading system is not being addressed by NCHRP 09-59—shortcomings in laboratory aging procedures. First, the current PAV aging protocol appears to provide less aging than needed for accurate evaluation of a binder’s fatigue performance. Second, the current PAV aging protocol also may not accurately reflect differ- ences in binder aging in different climates. These problems have been addressed by NCHRP Project 09-61, “Short- and Long-Term Binder Aging Methods to Accurately Reflect Aging in Asphalt Mixtures” (Bonaquist et al., 2021). Because the binder-aging protocol has been addressed by that project and is likely to be revised in the future, proposals for NCHRP 09-59 are based on the current aging protocol so that they can be implemented quickly. Once a revised protocol is established, additional research may be needed to revise the binder fatigue testing protocol. The information in this report should provide useful guidance in making these revisions. Validation Testing Validation testing was performed on three sets of pavement test sections and five sections from the second FHWA ALF fatigue experiment. The field sites and ALF sections are considered separately because the ALF sections were tested under constant temperature conditions and so cannot be compared with actual pavement test sections. However, the ALF test sections are still useful in evaluating the effectiveness of potential binder fatigue parameters. 0 10 20 30 40 50 0 10 20 30 40 50 Av g. P av em en t T em p. , C Proposed Binder Fatigue Test Temperature, °C Figure 51. Average pavement temperature as a function of binder fatigue test temperature (proposed grading system).

Findings and Applications 63   Field Validation Sites Three sets of test pavements were included in validation testing: (1) four test sections on US-93 about 50 miles north of Wikieup, Arizona; (2) four test sections on Highway 112 near Rochester, Minnesota (Olmsted County); and (3) four test sections from the NCAT test track. The Arizona and Minnesota test pavements were conducted as part of the Asphalt Research Consortium, while the NCAT test track is an ongoing activity at the National Center for Asphalt Technology. Condition surveys were regularly made on these pavements, including the extent of fatigue cracking. Characteristics of these test sections are summarized in Table 14. Binders from the Arizona and Minnesota test sections were received in the original state then aged using the RTFOT and 40-hour PAV at 100°C. Data on recovered binders from the Minnesota test sections were acquired courtesy of Gerald Reinke of MTE Services, Inc. Binders from the NCAT test sections were recovered by NCAT and forwarded to Advanced Asphalt Technologies for testing. Climate data for the test pavements along with binder grading data from the LTPPBind soft- ware are given in Table 15. The pavement structure for each validation site is given in Table 16, while the cracking data are given in Table 17 (Corrigan 2016; Farrar et al. 2016). Additional details on the test sections from the NCAT test track can be found on the About the Test Track section of their website (http://eng.auburn.edu/research/centers/ncat/testtrack/index.html). The first series of comparisons, given in Figures 52 through 55, shows the relationship between potential specification parameters and observed fatigue cracking for the three validation sites. The specification parameters were calculated at 4°C above the estimated average pavement temperature, except for R-value, which was calculated at 9°C or 10°C and 10 rad/s. As discussed, Project Date Constructed Date of Condition Survey Section Code Binder AZ US-93 near Wikieup 2001 2012 1-1 PG 76-16, air blown 1-2 PG 76-16, Venezuelan 1-3 PG 76-16, Rocky Mountain blend 1-4 PG 76-16, Canadian blend MN US-112 near Rochester/ Olmsted Co. 2006 2014 1-2 PG 58-34, terpolymer-modified 1-3 PG 58-28, Canadian blend 1-4 PG 58-28, Canadian, 8% REOB 1-5 PG 58-28, Venezuelan blend NCAT test track, Auburn, AL July–August 2015 June 2018 N1 PG 67-22, 20% RAP N8 PG 67-22, 20% RAP, 5% RAS S5 PG 64-28, 35% RAP S6 PG 88-22, 20% RAP Table 14. General characteristics of field validation sites. Project LTPPBind, 98% Reliability (Continuous) LTPP Binder Base PG Grade Local Binder Base PG Grade Estimated Average Pavement Temp. °C Binder Fatigue Test Temperature °C High PG Grade Low PG Grade Current Proposed AZ US-93 69.5 −9.6 PG 70-10 PG 76-16 22.9 34 29 MN US-112 52.5 −40.0 PG 58-40 PG 58-34 11.7 16 17 NCAT track 69.3 −15.4 PG 70-16 PG 67-22 22.6 26.5 27 Table 15. Climate data and binder grading information for field validation sites.

64 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures Project Age at Survey years Section Code Fatigue Cracking AZ US-93 near Wikieup 11 1-1 7.9 m 1-2 17.8 m 1-3 64.7 m 1-4 66.1 m MN US-112 near Rochester/ Olmsted Co. 8 1-2 0.8 m 1-3 18.8 m 1-4 39.2 m 1-5 0 m NCAT test track, Auburn, AL 3 N1 22.5% N8 17.0% S5 0% S6 0% Table 17. Cracking data for field validation sites. Project Base Course Thickness Surface Course Thickness Layer Containing Study Binders Commentsmm mm AZ US-93 126 19 Base Asphalt rubber surface course MN US-112 100 38 Surface NCAT track 127 38 Surface Table 16. Structure of field validation sites. 0 5 10 15 20 25 0 20 40 60 80 100 0 5,000 10,000 Cr ac ki ng , % a re a Cr ac ki ng , s qu ar e m et er |G*| sin δ, kPa AZ (left), 93% MN/MTE (left), 0% NCAT (right), 66% NCAT fit MN fit AZ fit 0 5 10 15 20 25 0 20 40 60 80 100 0 20,000 40,000 Cr ac ki ng , % a re a Cr ac ki ng , s qu ar e m et er GRP, kPa AZ (left), 91% MN/MTE (left), 85% NCAT (right), 52% NCAT fit MN Fit AZ fit Figure 52. Fatigue cracking as a function of |G*| sin c for field validation sites, at 4çC above average pavement temperature and 10 rad/s. Figure 53. Fatigue cracking as a function of Glover-Rowe Parameter for field validation sites, at 4çC above average pavement temperature and 10 rad/s.

Findings and Applications 65   0 5 10 15 20 25 0 20 40 60 80 100 0 50,000 100,000 Cr ac ki ng , % a re a Cr ac ki ng , s qu ar e m et er |G*|(R/2)2, (kPa) AZ (left), 53% MN/MTE (left), 96% NCAT (right), 47% NCAT fit MN fit AZ fit Figure 54. Fatigue cracking as a function of |G*| (R/2)2 for field validation sites, at 4çC above average pavement temperature and 10 rad/s. 0 5 10 15 20 25 0 20 40 60 80 100 0.0 2.0 4.0 Cr ac ki ng , % a re a Cr ac ki ng , s qu ar e m et er AZ (left), 49% MN/MTE (left), 60% NCAT (right), 2%MN fit NCAT fit AZ fit Binder R-Value Figure 55. Fatigue cracking as a function of R-value for field validation sites. 4°C above the average pavement temperature represents the target temperature for both the current and proposed fatigue test temperature. Some care should be taken in interpreting these plots; the cracking data and binder samples have been taken at different ages; two binders have been extracted and one was an original binder subjected to laboratory aging. What is important in evaluating these plots are two factors: (1) the observed correlation for a given site between the selected parameter and the amount of fatigue cracking and (2) whether the current specification limit—or the equivalent for proposed parameters—appears to be at least approxi- mately at an appropriate level. The current specification limit is 5,000 kPa for |G*| sin δ; as discussed, equivalent values for GRP and |G*| (R/2)2 are 5,300 kPa and 8,700 kPa, respectively. In Figure 52, fatigue cracking is plotted as a function of |G*| sin δ. The correlation between these variables is good for the Arizona test site (93%), moderate for the NCAT sections (66%), and nonexistent for the Minnesota test site. Figure 53 is the plot of cracking versus GRP. In this case, both the Arizona and Minnesota sites show good correlations (91% and 85%, respectively), while the correlation for the NCAT section is moderate at 52%. Figure 54 is a plot of fatigue cracking as a function of |G*| (R/2)2. In this case, the correlation is good for the Minnesota test site (96%) and moderate for the Arizona test site and the NCAT test track (53% and 47%, respectively). Figure 55 is the final plot in this series, showing fatigue cracking as a function of R-value. The correlation in this case is moderate for the Arizona and Minnesota test sites, 49% and 60%, respectively, but poor for the NCAT test sections with an r2 of only 2%. However, the sense of

66 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures the relationship for the Arizona and NCAT test sites is reversed from what is expected and from that observed for the Minnesota test site. For the Arizona and NCAT test sections, cracking decreases with increasing R-value rather than increasing. Based on the results of the layered elastic analysis, this should not be surprising. In thin pavements at lower temperature, fatigue life tends to decrease with increasing R-value. However, this relationship begins to change and eventually reverses as temperature and pavement thickness increase. The Minnesota test sec- tions are the thinnest and use the hardest binder grades relative to that required according to the LTPPBind software, so cracking, not surprisingly, increases with R-value. The Arizona and NCAT sections are thicker and use softer binder grades relative to the local climate, factors that would tend to lead to a decrease in cracking with increasing R-value. This shows why R-value is probably not a good parameter for an overall fatigue specification, although it does suggest that a limit on R-value might be appropriate for thin pavements when stiffer binders are used. For Figures 52 through 54, the relationship between fatigue cracking and the three param- eters seems roughly consistent with the tentative specification limits. Generally, the position of the three test sites on the plot seems consistent with the relative aging. The NCAT test sections generally appear furthest to the left on the plot—toward lower parameter values and suggesting less aging. Because of the accelerated loading, these are the least aged of the three sections. The Minnesota sections represent 6-year-old pavement in a cold climate. The Arizona sections have been laboratory aged but using a harsh procedure involving RTFOT aging followed by 40 hours of PAV aging. The second set of plots (Figures 56 through 59) is similar to the first, but the parameters are calculated at the proposed fatigue specification temperature. The correlations for the indi- vidual data sets are nearly identical, as would be expected, because these figures only represent a relatively minor change in temperature. Furthermore, the relative positions of the data for each site are also similar and consistent with the existing specification limit or the equivalent values for alternative parameters. This is encouraging, as it suggests the proposed scheme is effective in achieving the desired target test temperature. These figures can be compared with Figure 52, which shows fatigue cracking as a function of |G*| sin δ for the three validation sites, using the fatigue test temperature as determined under the current scheme. What is notable here is that the relative position of the three data sets has changed significantly, with the Arizona and NCAT data shifting toward higher values, and the Minnesota data shifting toward lower values. The overall appearance of the plot is different from that for testing at 4°C over the average pave- ment temperature, suggesting that the current scheme is not consistent in identifying an effec- tive test temperature for binder fatigue testing. This is a serious concern and is perhaps more important than selecting the best binder fatigue specification parameter. For thin pavements, 0 5 10 15 20 25 0 20 40 60 80 100 0 5,000 10,000 Cr ac ki ng , % a re a Cr ac ki ng , s qu ar e m et er |G*| sin δ, kPa AZ (left), 94% MN/MTE (left), 0% NCAT (right), 62% Figure 56. Fatigue cracking as a function of |G*| sin c for field validation sites, at proposed binder fatigue test temperature and 10 rad/s.

Findings and Applications 67   0 5 10 15 20 25 0 20 40 60 80 100 0 10,000 20,000 30,000 40,000 Cr ac ki ng , % a re a Cr ac ki ng , s qu ar e m et er AZ (left), 78% MN/MTE (left), 85% NCAT (right), 49% GRP, kPa 0 5 10 15 20 25 0 20 40 60 80 100 1,000 10,000 100,000 Cr ac ki ng , % a re a Cr ac ki ng , s qu ar e m et er |G*|(R/2)2, kPa AZ (left), 47% MN/MTE (left), 96% NCAT (right), 43% 0 5 10 15 20 25 0 20 40 60 80 100 0 5,000 10,000 Cr ac ki ng , % a re a Cr ac ki ng , s qu ar e m et er |G*| sin δ, kPa AZ (left), 92% MN/MTE (left), 0% NCAT (right), 60% Figure 57. Fatigue cracking as a function of Glover-Rowe parameter for field validation sites, at proposed binder fatigue test temperature and 10 rad/s. Figure 58. Fatigue cracking as a function of |G*| (R/2)2 for field validation sites, at proposed binder fatigue test temperature and 10 rad/s. Figure 59. Fatigue cracking as a function of |G*| sin c for field validation sites, at current binder fatigue test temperature and 10 rad/s.

68 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures fatigue performance is mostly a function of binder modulus, with R-value having a secondary effect. Modulus is mostly a function of temperature, so if the fatigue test temperature is not selected consistently, the resulting parameter values will not be effective in controlling fatigue performance. Analysis of Second FHWA ALF Fatigue Experiment The second validation testing involves data from the second FHWA ALF fatigue experiment (Gibson et al., 2012). In this experiment, test sections of 100 mm and 150 mm thickness were placed at the FHWA ALF facility at Turner-Fairbank Highway Research Center. The mixes were essentially identical, except for using a range of binders. Figures 60 through 63 show fatigue cracking as a function of the possible binder fatigue parameters for the 100-mm thick pavement sections. The temperature used for calculating these parameters was about 4°C above the test temperature of 19°C (23.1°C), the test tempera- ture in this case coincidentally being essentially identical to the average pavement temperature for this location. Data for these plots were from FHWA testing on recovered binders subjected to standard PAV aging. These data were thought to better represent the condition of the binder in situ compared with the heavily aged (RTFOT/40-hour PAV) binders tested as part of NCHRP 09-59. Correlations for all considered binder fatigue parameters—except for R-value—are good to excellent, ranging from 82% (|G*| sin δ) to 94% (GRP). The correlation for R-value is poor at only 8%. As with the field validation sites, the ALF data seem to be roughly consistent with the tentative specification limits, on the basis of the current maximum value for |G*| sin δ of 5,000 kPa. Correlations between the proposed parameters and cracking for the thicker 150-mm pavements were in general poor, with all values being below 25% except for R-value, which was moderately strong at 65%. However only one tested section (the air-blown binder) showed any significant cracking, so it is difficult to make firm conclusions on the basis of results for the 150-mm sections. Furthermore, as pointed out in the section on layered elastic analysis, correlations between binder parameters and fatigue cracking should be strongest for thin pavement sections. Values of Proposed Specification Parameters for NCHRP 09-59 Binder An important issue is to discover values for the proposed specification parameters that apply to the NCHRP 09-59 binders. Are these values consistent with projected binder performance? R² = 82% 0 20 40 60 80 100 0 2,000 4,000 6,000 8,000 Pe rc en t C ra ck ed |G*| sin δ, kPa Figure 60. Fatigue cracking as a function of |G*| sin c for FHWA ALF 100-mm sections, at 23.1çC and 10 rad/s.

Findings and Applications 69   R² = 94% 0 20 40 60 80 100 0 2,000 4,000 6,000 8,000 Pe rc en t C ra ck ed GRP, kPa R² = 91% 0 20 40 60 80 100 0 5,000 10,000 15,000 Pe rc en t C ra ck ed |G*| (R/2)2, kPa R² = 8% 0 20 40 60 80 100 1.00 1.50 2.00 2.50 3.00 Pe rc en t C ra ck ed R-Value, kPa Figure 61. Fatigue cracking as a function of Glover-Rowe parameter for FHWA ALF 100-mm sections, at 23.1°C and 10 rad/s. Figure 62. Fatigue cracking as a function of |G*| (R/2)2 for FHWA ALF 100-mm sections, at 23.1çC and 10 rad/s. Figure 63. Fatigue cracking as a function of R-Value for FHWA ALF 100-mm sections, at 23.1çC and 10 rad/s. Figure 64 is a plot of GRP as a function of R-value for the 16 NCHRP 09-59 binders, coded for the projected fatigue performance: good, moderate, or poor. This projected performance was based on the judgment of the panel and the research team during binder selection and indicates the typical perceived fatigue performance for binders of a given type and grade. In this figure, a maximum GRP value of 8,000 kPa has been shown, along with the proposed allowable R-value range from 2.0 to 3.2. These binders were aged using RTFOT/40-hour PAV rather than RTFOT/20-hour PAV. The GRP maximum has been raised from 5,000 kPa to

70 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures 8,000 kPa to account for the more severe aging and for where the various binders appear on this plot. Of the seven binders with estimated good fatigue performance, six pass the proposed specification. For the four binders with projected moderate performance, three pass. For the five binders with estimated poor fatigue performance, only one passes. This binder was projected to be a poor performer because it contained REOB, but the rheological characteristics of this binder are not typical of binders containing REOB. Most binders containing REOB have unusually high R-values; this “REOB-containing” binder has the lowest R-value of all the NCHRP 09-59 binders. All considered, the proposed specification parameters and limits seem reasonably consistent with the estimated performance of the NCHRP 09-59 binders. Discussion This chapter has presented a substantial amount of complex data concerning the relationship between various binder parameters and the fatigue performance of asphalt concrete mixtures. Asphalt mixture fatigue response is a complex function of various parameters. As modulus increases, strain capacity decreases, which decreases fatigue life. At the same time, increasing modulus will result in decreasing phase angle, which will increase the fatigue exponent, increas- ing fatigue life. Healing is probably a significant factor in asphalt mixture fatigue performance, with increasing phase angles leading to better healing and improved fatigue life for binders that are not polymer modified; for polymer-modified binders, the relationship between rheology and healing is unclear. To further complicate matters, in real pavement systems softer binder and mixes will in general exhibit higher failure strains but will also tend to result in higher strains in situ, potentially offsetting the beneficial effects of lower modulus values. This would seem at first to be an impossibly complex problem, but the task of developing an effective binder fatigue specification is simplified when the critical situation is considered: high strains relative to the binder strain tolerance, which will lead to rapid fatigue failures. Such a situation can exist in thin or otherwise weak pavement structures, and potentially at the surface of thicker pavements as a result of tire-pavement interactions and thermal stresses. Under these conditions, the most important factor controlling fatigue life is the strain tolerance of the asphalt binder. Under high-strain conditions, as binder failure strain decreases, fatigue life will also decrease. An especially important correlation to this finding is that in general, softer binders will tend to produce pavements with better fatigue performance compared with those made with stiff binders. The failure strain or FSC of asphalt binders tends to follow a well-defined envelope with respect to modulus; a power law model was used in NCHRP 09-59 to define this relationship. 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 1.0 2.0 3.0 4.0 G RP a t T es t T em p. , k Pa R-Value Good Moderate Poor Spec. fail fail fail pass Figure 64. Proposed binder fatigue specification as applied to NCHRP 09-59 binders after RTFOT/40-hour PAV aging. Coded for estimated fatigue performance.

Findings and Applications 71   However, this standard envelope varies significantly among different binders. A fundamental question addressed in NCHRP 09-59 is which test or tests are good indicators of this inherent strain tolerance. Laboratory testing suggests that binder R-value is a good predictor of overall FSC. Although the SDENT test also correlates to inherent FSC, the correlation is not as good as for R-value, possibly because the rate of loading in the SDENT test is slower than in mixture fatigue tests. Considering the difficulty and cost of implementing a specification including the SDENT test, it cannot be recommended on the basis of research conducted as part of NCHRP 09-59. It should be noted that NCHRP 09-59 used heavily aged binders and mixture in its testing program. If the materials had not been as thoroughly aged, the findings might have been different. This could be a fruitful topic for further research. Although R-value is a good indicator of overall strain tolerance, modulus has an even stronger effect on failure strain and fatigue life. The Glover-Rowe parameter appears to relate well to binder failure strain, as does a newly proposed parameter, |G*| (R/2)2. These parameters appear to account for both the effect of modulus and the effect of R-value on failure strain and are good candidates for use in a revised binder fatigue specification. The current binder fatigue parameter, |G*| sin δ, does not correlate as well to binder failure strain, and in fact can allow binders with relatively poor strain tolerance to be used in pavement applications. SDENT extension, like |G*| sin δ, correlates to binder failure strain, but not as well as GRP or |G*| (R/2)2. Because numerous researchers and engineers have used the GRP over the past 5 years to indi- cate nonload-associated cracking and fatigue cracking, and because validation testing suggests the GRP indicates fatigue performance in thin pavements slightly better than |G*| (R/2)2, this parameter is proposed in this report as the best overall choice for a binder fatigue specification parameter to replace |G*| sin δ. It would be useful here to revisit the ratings of the tests and parameters discussed at the beginning of this report and in Appendix A; Table 18 shows a revised version of Table A-2, which summarizes preliminary ratings of a variety of tests. Table 18 includes fewer tests, and some numerical ratings have been revised. The rating for the LAS test for correlation to perfor- mance (and in Table 18, performance-related tests) has been lowered from 5 to 3. The rating of the DENT test for technical difficulty and ease of implementation has been increased from 1 to 2 because the test was found easier to perform once the details were worked out. Yet, there are still concerns about how many laboratories currently have ductility devices and about the amount of material needed. The current specification parameter, |G*| sin δ, has also been included in Table 18 for comparison with alternative tests and parameters. In this rating system, GRP and R-value appear to be the best parameters for controlling the fatigue behavior of asphalt binders in an improved specification. Criterion Weight LAS DENT Glover- Rowe Parameter and R-Value |G*| sin δ Equipment cost (additional/new) 10 5 3 5 5 Active time requirement 20 3 1 3 3 Correlation with performance and performance-related tests 50 3 5 5 3 Engineering soundness 5 3 3 5 4 Technical difficulty/ease of implementation 15 5 2 5 5 Total 100 3.5 3.5 4.6 3.6 Table 18. Revised ratings of candidate binder fatigue tests.

72 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures A serious concern—probably even more significant than the concern with |G*| sin δ—is the ineffective way the current specification addresses fatigue test temperature. Ideally, the fatigue test temperature should be closely tied to average pavement temperature, but the current speci- fication does not appear to do this well. This is partly because the test temperature is tied to both the low and the high binder temperature grades; unless an agency carefully applies the specification, the fatigue test temperature can be significantly higher than intended. Even when the binder is applied correctly (to base binder grades before adjustments for traffic and vehicle speed), the relationship between the current binder fatigue test temperature and average pave- ment temperature is not ideal. A different approach is suggested in this report, in which the fatigue test temperature is tied to the low PG grade, rather than the average of the low and high PG grades. Furthermore, several notes are suggested to make the intent of the specification clear. For example, some areas commonly use binders rated as much as two grades higher than the low PG grade given by the LTPPBind software. In such cases, the pavement will not only be prone to increased thermal cracking but will also be subject to significantly greater fatigue cracking unless the pavement thickness is increased. Including a note to this effect would ensure that highway agencies are aware of this potential problem. Replacing |G*| sin δ with GRP and using an improved protocol for specifying binder fatigue test temperature would solve many problems with the current binder fatigue specification. However, in addition to these changes, an allowable range for R-value (or an equivalent param- eter such as ΔTc) should be established. There are two reasons for this limitation: 1. In thin pavements at low temperatures, binders with high R-values can result in rapid accu- mulation of fatigue damage. This is likely a major reason for recently observed premature failures of pavements in Ontario and the Northern United States made with binders contain- ing REOB, which tend to have high R-values. 2. In thick pavements, binders with low R-values can show relatively poor fatigue performance. By limiting high and low R-values, both problems are addressed. Tentative allowable ranges for R-value are from 1.5 to 2.5 for binders aged with RTFOT and 20-hour PAV, and from 2.0 to 3.2 for binders aged with RTFOT followed by 40-hour PAV. An important additional consideration is how to determine R-value for specification purposes. In NCHRP 09-59, DSR data at high modulus values (|G*| values typically from 10 to 20 MPa) were used in calculating R. High modulus values are needed for calculating R because for some binders, especially those heavily modified with polymers, the phase angle becomes distorted at low to intermediate modulus values. The higher the modulus value, the more likely this dis- tortion can be avoided. For specification purposes, R-values calculated from BBR data would probably be more suitable—the modulus values at the specification temperature would easily be high enough to prevent these problems. The intermediate temperature DSR measurement, on the other hand, will often involve |G*| measurements well below 10 MPa. This problem could be addressed by adding a short temperature sweep to the DSR intermediate temperature procedure, but calculating R from BBR data would prevent adding this complexity to the current specifica- tion. R can be calculated from BBR data using the following equation, similar to Equation 4. 2 3,000 1 (16)( ) ( ) ( ) = − R log log S log m where R = Christensen-Anderson R-value (rheologic index), S = BBR creep stiffness at 60 seconds (MPa), and m = BBR m-value at 60 seconds.

Findings and Applications 73   As discussed previously in this report, parameters other than R-value exert a similar control over binder rheology. These include ΔTc and BBR stiffness at m = 0.30 (S (0.3)). The parameter ΔTc has become widely used in evaluating pavement failures from nonload-associated cracking. Although there is a good correlation between R and ΔTc, the latter reflects both time dependence and temperature dependence, while R is strictly a function of the time-dependent behavior of a binder. The parameter S (0.3) is simply calculated as the BBR stiffness at the point where m = 0.30, regardless of the loading time. Theoretically, the correlation between R and S (0.3) should be close to 100%, because these are both ways of characterizing the shape of the binder creep response with respect to time without consideration of temperature dependency. Use of any of these three parameters would exert similar effects on binder rheology and would be equal—or nearly equal—in their effectiveness. This report has focused on the use of R in a binder specification because of its fundamental nature, but an effective, improved binder specification could be implemented with any of these parameters. Although the analysis in this report suggests that an improved binder fatigue specification should include both minimum and maximum values of R (or some equivalent parameter), this may not be practical if the precision of the parameter is too poor. Having both upper and lower limits in this case would produce an overly restrictive specification. In such a case, an upper limit on R (or minimum ΔTc) could be implemented without a minimum value. Most evidence currently indicates that binders with high R-values or highly negative ΔTc values are more likely to fail prematurely than binders with low R-values or highly positive ΔTc values. Completion of NCHRP 09-60 should provide some indication of the precision of R, ΔTc, and S (0.3) and a better idea of whether a lower limit on R is advisable. The current laboratory aging method for binder fatigue testing uses RTFOT conditioning following by PAV aging for 20 hours. In NCHRP 09-59, the aging protocol was more severe, using RTFOT conditioning and 40 hours of PAV aging. For any binder fatigue test to be effective, it must include a laboratory aging method that mimics aging in the field with reason- able accuracy. Much research has recently been conducted to address this problem, including NCHRP 09-61, which directly addresses laboratory aging of binders (Bonaquist et al., 2021). Because future changes in the binder-aging protocol are likely, and because addressing prob- lems in the binder fatigue specification as quickly as possible is important, it appears findings of NCHRP 09-59 should be implemented with the current aging protocol. A final critical issue is what value the chosen specification parameters should have. The current binder fatigue parameter, |G*| sin δ, has a maximum value of 5,000 kPa. A simple approach and one easy to implement would be to use an equivalent limit on the new speci- fication parameter. For GRP, this would be 5,300 kPa; for |G*| (R/2)2, the value would be 8,700 kPa. For practical purposes, the maximum value for GRP could probably be rounded down to 5,000 kPa—identical to the current maximum for |G*| sin δ. Using these equivalent values assumes that the aging protocol would initially be unchanged (RTFOT followed by 20-hour PAV). Using an equivalent value (and similar but more effective test temperature) would help ensure that the new specification is not overly restrictive or too lax. Furthermore, the previous analyses in this paper suggest that the primary problems with the current binder fatigue specification do not involve the exact specification value. Instead, the problems are the result of the weak relationship between |G*| sin δ and strain capacity, poorly defined and implemented test temperatures, and an aging protocol that may not accurately reflect in situ aging. Maintaining a similarity to the current specification during initial implementation of the proposed binder fatigue specification should allow rapid implementation of a change that addresses two of these problems. The third issue—binder laboratory aging—has been addressed in NCHRP 09-61 (Bonaquist et al., 2021). A significant effort was made in NCHRP 09-59 to correlate binder rheology to healing. As mentioned at the beginning of this summary, the results show, for binders that are not polymer

74 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures modified, healing increases with increasing phase angle. This means that healing will increase with increasing temperature and decreasing values of R or increasingly positive values of ΔTc. For polymer-modified binders, no clear relationship was observed between binder rheology and mixture healing. Although these findings seem to disagree with those of NCHRP 09-44A (Witczak et al., 2013), they can be reconciled when it is realized that NCHRP 09-44A did not use any polymer-modified binders, and if it is assumed that there were not large differences in the rheological behavior (R-value) of the three binders used in this research. Under these conditions, healing would be expected to increase with decreasing binder and mixture modulus, as observed in NCHRP 09-44A, because binder modulus and phase angle are directly related when non-polymer-modified binders with a similar R-value are considered. Therefore, the findings of NCHRP 09-59 and NCHRP 09-44A are not necessarily inconsistent. For non-polymer-modified binders, establishing a maximum R-value should be effective in ensuring adequate healing. Limiting R, while also controlling modulus through the limits on GRP, will help ensure that these materials have reasonably high phase angles and good healing properties. This is because decreasing R-values will result in higher phase angles at any given modulus level. If an alternative assumption is made—that healing increases with decreasing binder and mixture modulus (as proposed in NCHRP 09-44A, see Witczak et al., 2013)—the proposed maximum value of GRP would help to ensure adequate healing, because GRP is closely related to binder modulus, and hence mixture modulus. Addressing binder healing potential more directly is impossible given the results of the work done as part of NCHRP 09-59 because mixes made with polymer-modified binders unfortunately do not show a clear relation- ship between rheology and healing. Future research should probably focus on a test that directly measures the adhesive and cohesive properties of binders and relates this to observed mixture healing. An alternate approach to a specification test addressing healing is the direct measure- ment of binder fatigue life with and without rest periods. Unfortunately, the only test that currently attempts to characterize binder fatigue life is the LAS test, which in NCHRP 09-59 did not relate well to observed mixture fatigue performance. Further modifications in the LAS test or the analysis of LAS data, however, might improve the test, in which case it might be suitable for directly measuring healing potential. Observed Performance of Polymer-Modified Binders in NCHRP 09-59 Because polymer-modified binders are widely accepted as having superior fatigue and fracture properties, they were expected to perform better than non-modified binders in NCHRP 09-59. Most polymer-modified binders evaluated using the SDENT test as part of NCHRP 09-59 exhibited greater extension to failure at a given stiffness value than non-polymer-modified binders did. That is, in this test, they exhibited greater inherent strain tolerance and their FFPR values were higher than for binders containing no polymer. However, this increase in inherent strain tolerance was not observed in FSC values calculated from mixture fatigue tests. In fatigue testing, at a given modulus level and R-value (or similar GRP value) the strain capacity of polymer- modified and non-modified binders appeared similar. This finding was unexpected, and it is not clear at this time whether it applies to a wider range of tests and conditions or it is an artifact of the specific procedures used in NCHRP 09-59. One main area of concern is the extended loose-mix aging used before mixture testing. This aging method may disrupt the effectiveness of polymer-modified binders, for example, by degrading the bond between the aggregate and the polymer network in ways that would not occur in an actual pavement. Extended aging, in general, may also result in lower performance of polymer-modified binders than standard aging procedures, producing a somewhat skewed picture of the relative performance of these materials. For instance, through the first 3 to 5 years

Findings and Applications 75   of service, pavements made with polymer-modified binders might exhibit significantly better fatigue performance than those made with non-modified binders. Even if this benefit were to disappear after 5 to 10 years of service, the initial improvement in performance should still result in substantially longer pavement life when polymer-modified binders are used. A laboratory test based only on extended aging may be evaluating the mixture in a condition for which the effectiveness of the polymer-modified binder has been diminished to a level that does not represent the entire life of the pavement. Because both polymer-modified binders and extended loose-mix aging are increasingly used, these topics warrant prompt additional and thorough research. If the performance of polymer-modified binders is found to be unrealisti- cally degraded during extended loose-mix aging, additional research may be needed to develop alternative mixture-aging methods. It might also be necessary to further refine and evaluate the SDENT test if it seems to better reflect the strain tolerance of modified binders than R-value and GRP. Another related issue is the potential use of R in the improved binder specification, proposed primarily to prevent premature failure in thin pavements made with binders having high R-values. Polymer-modified binders often have relatively high R-values, suggesting somewhat lower failure strains than typical asphalts and also suggesting that they might be subject to early failure when used for thin pavements in cold climates (relative to the low PG grade). However, if the aging protocols used in NCHRP 09-59 did in fact result in unrealistically poor fatigue performance for the polymer-modified binders, it would mean that high R-values are not as much of a concern for these materials. That would suggest that the improved binder specifica- tion should relax or even eliminate the R-value requirements for polymer-modified binders. This relaxation could be implemented in a modified version of the AASHTO M 332 binder specification by having a higher maximum R-value for the two highest traffic levels, for which polymer modification would usually be required to obtain the necessary properties to pass the specification. Additional information on R-value, ΔTc, and various other performance-related test data is being gathered as part of NCHRP 09-60. The issue of polymer-modified binders and how best to address them in an improved binder fatigue specification should be reevaluated at the conclusion of this project. The similarity of FFPR values for polymer-modified and non-modified binders in mixture fatigue tests after extended loose-mix aging does not mean that the polymer-modified binders evaluated in NCHRP 09-59 would not exhibit better fatigue performance in situ than non- modified binders. This is because the FSC value and fatigue performance of a binder are a function of both inherent strain tolerance (FFPR) and stiffness in the climate where the binder is used. A binder too stiff for a given climate will have inadequate strain tolerance and will therefore exhibit poor fatigue performance. A binder that is softer in a given climate, on the other hand, will have high strain tolerance and will exhibit good fatigue performance. In fact, the effect of modulus on FSC and fatigue performance is probably just as important as the effect of inherent strain tolerance (FFPR). Data from the uniaxial fatigue tests show that the difference in FFPR from the worst to the best performing binder is equivalent to about a 6°C change in temperature. If the difference between the best and worst binders, as indicated by SDENT extension, is instead used, this difference increases to 9°C. That is, the difference between the FFPR for a binder exhibiting average inherent strain tolerance and for a binder exhibiting either good or poor performance is equivalent to about a 3°C to 5°C change in temperature. This suggests that proper grading and selection of binders is just as important to ensuring good fatigue performance as the inherent strain tolerance, as indicated by fatigue testing or SDENT testing. It also means that binders relatively soft at the intermediate grading temperature may perform well in fatigue even if their inherent strain tolerance is not exceptional. This is in fact the case for most

76 Relationships Between the Fatigue Properties of Asphalt Binders and the Fatigue Performance of Asphalt Mixtures polymer-modified binders evaluated in NCHRP 09-59—that is, even though their FFPR values are not much different from the non-modified binders, their expected fatigue perfor- mance as indicated by GRP is generally better than the non-modified binders. This expected performance is clear from Figure 65, which shows GRP at the proposed fatigue test temperature and 10 rad/s for the 16 NCHRP 09-59 binders. The four binders with the best (lowest) GRP are all polymer modified; 7 of the 10 binders with the lowest GRP are polymer modified. The GRP values for the two REOB binders are surprising in that they are relatively low—lower than for three of the polymer-modified binders. The precise composition of these binders, however, is not clear, and a binder containing REOB might exhibit acceptable fatigue performance if it contains a modest amount of REOB and a good-quality base binder. Further- more, one REOB binder exhibited a high R-value, suggesting performance issues independent of GRP. Limitations of Findings The findings described in this chapter are based on a review of available literature dealing with the relationship between binder properties and asphalt concrete fatigue performance, and on laboratory testing and analysis conducted as part of NCHRP 09-59. There are several limita- tions to these findings. Probably the most significant is the aging protocols used in the study. The primary binders were aged using the RTFOT followed by 40-hour PAV aging. Mixtures were aged in the loose condition for 5 days at 95°C. These are relatively severe aging protocols. Results of the testing and analysis, particularly with the polymer-modified binders, might have been different had less-severe aging methods been used. Laboratory aging in general is a critical facet of the binder fatigue specification but was not extensively addressed in NCHRP 09-59. Some problems with the current specification are probably caused by inability of the current aging methods to accurately predict binder rheo- logical changes in different climates. The relatively severe mixture-aging protocol used in NCHRP 09-59 might produce an unrealistic degradation in fatigue and fracture performance of mixtures containing polymer- modified binders. Additional research is needed to determine how different aging procedures affect the fatigue performance of polymer-modified binders. A closely related issue is that of the effect of binder and mixture stiffness on the fracture and fatigue response of polymer-modified mixtures. In NCHRP 09-59, most fatigue tests were done at temperatures and frequencies corresponding to binder modulus values ranging from about 7 to 80 MPa, somewhat higher 0 5,000 10,000 15,000 G RP a t 1 0 ra d/ s, k Pa Figure 65. Values for Glover-Rowe parameter at proposed binder fatigue test temperature and 10 rad/s for NCHRP 09-59 binders.

Findings and Applications 77   than commonly used in mixture fatigue tests. Most mixture fatigue tests done during SHRP, for example, involved conditions corresponding to binder modulus values ranging from about 3 to 30 MPa. Effectiveness of polymer modification in improving fatigue and fracture proper- ties may decrease with increasing binder modulus. This is another area requiring additional research. The NCHRP 09-59 problem statement made clear that the binder test procedures consid- ered during the project should be ones currently used as specification tests or otherwise widely used and accepted so that findings could be easily implemented. The research team used this approach and as a result, many potential binder test procedures were not considered during this research project. One or more of these tests might in the long run be a better specification test than those evaluated in NCHRP 09-59, but the tests considered in this research were those that could be more easily implemented. Further research on improved binder fatigue tests is encouraged. Much of the analysis presented in this report relies on the newly developed GFTAB model, which provides a rational framework for describing mixture fatigue performance over a wide range of conditions. The basic features of this model are that fatigue performance is a func- tion of binder failure strain and the fatigue exponent. Failure strain in turn is related to binder modulus and the inherent strain tolerance of a given binder. The fatigue exponent is inversely proportional to the binder phase angle. The model appears to work well and largely explains why establishing a consistent relationship between binder properties and mixture fatigue perfor- mance is difficult. Although the research team thinks the GFTAB model helped establish useful relationships between binder properties and mixture fatigue performance, it has not been fully developed and calibrated for use in pavement design and should not be used for that purpose without further development, calibration, and review. Completing the testing and analyses in NCHRP 09-59 was difficult, reflecting a complicated set of relationships contributing to mixture fatigue performance. The NCHRP 09-59 research team believes the findings, once implemented, will lead to significant improvements in the fatigue performance of flexible pavements. However, fully eliminating all incidences of pre mature failure caused by fatigue damage is probably impossible, partly because the fatigue phenomenon in asphalt concrete pavements is complex, but also because many factors other than asphalt binder properties can affect pavement performance. Factors include binder content; pavement com- paction; mixture segregation during construction; and mixture temperature during production, storage, and transport. Binder fatigue performance is only one of many factors that affect the fatigue life of a flexible pavement.