Performance Testing for Superpave and Structural Validation

CHAPTER 8. CONCLUSIONS AND RECOMMENDATIONS

This report provides a critical evaluation of the Superpave^® binder specifications |G*|/sinδ and |G*|sinδ as controlling parameters for rutting and fatigue cracking. Previously, the SHRP program explored a variety of theoretical principles and experimental techniques and ultimately selected these parameters because they are fundamental, easily measured in a DSR, and describe the dependence on temperature, frequency, and aging, all of which are relevant to factors that influence pavement performance. However, the SHRP binder validation was largely limited to unmodified asphalts. Since the implementation of Superpave^®, polymer modified asphalt binder usage has become prevalent. Key research studies identified shortcomings using |G*|/sinδ to control rutting of polymer modified asphalts; principally, the measurement of these material properties are in the small strain, linear viscoelastic region that does not mobilize polymer structures that impart beneficial performance. Research also showed shortcomings of fatigue parameter |G*|sinδ, as this seemed to be valid only for thinner, unmodified asphalt pavements.

Following exploratory laboratory studies of polymer modified asphalts (NCHRP 9-10, 90-07), a full-scale APT experiment was designed.^(20,22) This ALF experiment generated performance data that provided quantitative grounds to screen weaker tests and identify discriminating binder characterization parameters to potentially replace or enhance portions of the existing Superpave^® PG system. Asphalt test pavements were built with a single volumetric mix design using a variety of unmodified and polymer modified binders with practically equivalent high-temperature properties and different intermediate-temperature fatigue properties.

The experiment also included pavement sections to assess the performance of crumb rubber modified asphalt. A case study for FWD layer modulus back-calculation was provided and an assessment of emerging AMPT test equipment and NCHRP MEPDG software was made.

SUMMARY and CONCLUSIONS

Binder Performance Specification Parameters

Key findings are as follows:

• Polymer modified binders significantly improve the fatigue cracking performance compared to unmodified binders with similar high-temperature PG grades.
• There are more discriminating binder tests for fatigue cracking and rutting than standard Superpave^® |G*|sinδ and |G*|/sinδ, and these tests are poised for implementation or have already become standardized, including MSCR and oscillatory-based nonrecoverable stiffness for rutting and calculated CTOD for fatigue cracking.
• Increasing polymer content in relatively softer base asphalt binders to achieve higher temperature PG grades does not necessarily provide increased fatigue cracking resistance. An important caveat of this conclusion is that it may only be applicable for the particular structural configuration of the ALF pavements in this experiment. However, the rutting performance of this mixture was generally good and comparable to the other polymer modified mixtures until large rut depths (0.47 inches (12 mm)) were reached, and then rutting increased substantially.

Binder Test to Discriminate Permanent Deformation and Rutting

Key findings are as follows:

• The two most discriminating, implementable parameters for rutting quantify irrecoverable deformations that occur in asphalt binder. Oscillatory-based nonrecoverable stiffness is measured in the DSR in a similar manner as the familiar |G*|/sinδ. MSCR has already been developed as an AASHTO provisional standard and provides an added measure of recoverable measurements and, therefore, has advantages over oscillatory-based nonrecoverable stiffness.
• The performance data in this particular experiment demonstrated that |G*|/sinδ still has value as a specification parameter to control rutting.

The research literature discussed in chapter 1 documents the shortcomings of the Superpave^® high-temperature PG test. Namely, the test deforms the binder to small strains and has trouble capturing the performance-related effect of polymer modification. However, the rutting performance of the polymer modified and unmodified asphalt binders in this particular full-scale APT experiment do not reflect this deficiency. The binders used in this study were designed and selected to have the same high-temperature PG (see figure 15). Therefore, these binders should have exhibited the same rutting performance, which was essentially the case (see figure 28, figure 29, table 17, and table 19). There were quantifiable differences in the mean rutting, but variability in the measured performance diminished those differences. Therefore, one interpretation of this experiment’s rutting data is that |G*|/sinδ still has its merits as a specification parameter to control rutting.

It must be strongly emphasized that there were notable numerical and statistical challenges in making defensible comparisons between the varieties of binder candidates to identify the best tests. Ideally, such analyses use more data points with larger spread to overcome scatter. Simply using R² was not appropriate. The Kendall’s tau measure of association method was utilized because it is better suited for small datasets and quantifies the quality of a ranking. Statistical significance (probability) was computed as well. The different quantities were combined into a single composite score. Laboratory scale performance tests were integrated to strengthen the scoring because the binder contributions are reflected in laboratory tests in addition to full-scale pavement tests, which provide the analysis with a degree of balance with the full-scale ALF performance.

Some candidate binder parameters such as LSV and ZSV were found to have a strong relation to laboratory-scale and full-scale performance. However, these parameters can be susceptible to registering improved performance from additives or modification (i.e., acid), which is not as legitimate as the beneficial improvements from polymer modification used in this study. Thus, they were not recommended.

Binder Tests to Discriminate Fatigue and Cracking

There were fewer numerical and statistical challenges in the datasets to rank weaker and stronger binder tests for fatigue cracking. The strongest analysis came from the thinner, 4-inch (100-mm) ALF lanes because two of the thicker, polymer modified mixtures in the 5.8-inch (150-mm) lanes did not exhibit any fatigue cracking. Nondestructive testing and extrapolated estimates were used to provide a reasonable measure of ranking for the whole set of 5.8-inch (150‑mm) lanes.

The key finding is as follows:

• The analysis from both 4- and 5.8-inch (100- and 150-mm) lanes plus a third verification from an Ontario test site showed calculated CTOD was consistently the most discriminating. Binder yield energy was the next strongest. These results confirm tests that deform binder to very large strains discriminate fatigue resistance from polymer modification. CTOD can be considered implementable because it can be measured using a force-ductility apparatus that is available in some pavement testing laboratories.

Crumb Rubber Modified Asphalt

The key finding is as follows:

• Gap-graded crumb rubber modified asphalt mix (Arizona wet process) placed in a composite pavement structure exhibited excellent fatigue cracking resistance to bottom-up fatigue cracks. Fatigue cracks initiated and propagated up through 2 inches (50 mm) of conventional dense-graded asphalt on the bottom but did not progress through any of the 2 inches (50 mm) of gap-graded crumb rubber mix on top.

The CR-AZ mix was formulated using a PG58-22 base binder plus 17 percent crumb rubber by weight following ASTM D6114.⁽²⁴⁾ The estimated high and intermediate PG grades were 194 and 74.1 °F (90.1 and 23.4 °C), respectively using techniques developed in research based on DSR measurements on unaged binder. The CR-AZ binder could be characterized in the RTFO by tilting the oven back and in the 0.975-inch (25-mm) plate DSR using a 0.078-inch (2-mm) gap rather than a 0.039-inch (1-mm) gap. The aggregate gradation was gap-graded to accommodate the crumb rubber particles, and the binder content was 7.1 percent. The mix was produced using an onsite shearing mill at the asphalt plant. The rutting performance was similar to the dense-graded mixtures in the experiment, although the mixture has a higher binder content and somewhat lower mix dynamic modulus. The fatigue performance was the best of the 4-inch (100-mm) lanes. The ability to arrest the bottom-up fatigue cracks is attributable to the rich rubber modified binder and possibly because the stress concentration is in the lower control lift.

The CR-TB binder did not require onsite mills at the plant and was delivered and handled like conventional asphalt binder. The CR-TB mix performed well. The rutting performance was the best of all the 4-inch (100-mm) lanes because it had slightly stiffer Superpave^® PG than the others, and the fatigue performance was slightly less than the SBS modified binder.

Fiber Reinforced HMA

Key findings are as follows:

• Very good fatigue cracking resistance was observed in the dense-graded mixture reinforced with polyester fibers. The fatigue cracking of this section was measurably better than those of the polymer modified sections even though a less-resistant unmodified asphalt binder was used in the mix. This was the second best performer of its thickness group behind the composite, gap-graded crumb rubber asphalt pavement.
• The presence of fiber had no significant beneficial or negative impact on rutting performance.

Mixture Performance Characterization Tests

Key AMPT findings are as follows:

• AMPT flow number and SST RSCH were the two strongest laboratory indicators of ALF rutting. The AMPT flow number test is a stronger predictor and more implementable.
• Most flow number tests did not achieve tertiary flow and showed simpler two-stage curves but still adequately discriminated performance.

Key axial fatigue findings are as follows:

• An alternative test for flexural beam fatigue that used axial, DT-compression cyclic loading to capture fatigue damage modulus reduction was assessed.
• Axial fatigue with VECD can be used to generate fatigue properties at multiple conditions with a smaller experimental program than beam fatigue.
• This test was the strongest, most implementable indicator of fatigue cracking, and correcting the tests results for true strain control using VECD theory further strengthened the test’s abilities.

Other key findings are as follows:

• The importance of testing asphalt mixtures to confirm performance cannot be understated. Such testing should not rely entirely on binder tests because additives such as fibers will challenge specification tests at the binder scale, and only mixture tests are suited to accommodate pavement structural attributes and volumetric mix design characteristics.
• The fiber mix presented some challenges for the three best mixture tests for fatigue cracking; none were able to reflect higher fatigue resistance.

A variety of materials characterization tests were conducted on ALF cores, plant-produced mixtures, and laboratory-produced mixtures. Some were established tests while others were emerging tests. The following tests were utilized in this research:

• AMPT flow number: More convenient flow number tests using less confinement can produce permanent strain magnitudes and rankings similar to more confined tests so long as the deviator stress is smaller in less-confined tests. Also, predicted rutting using flow number curves from more convenient, less-confined tests can be similar to predictions using confined flow number inputs because the less-confined deviator stress is accordingly smaller.
• Axial fatigue: Although the axial fatigue tests require added steps for gluing caps onto the end of the specimen, it still offers advantages over flexural beam fatigue because samples can be made in the Superpave^® gyratory compactor. Axial cyclic fatigue tests in stress control rank performance in opposite manners, with stress control favoring stiffer mixtures and strain control favoring softer mixtures.
• Dynamic modulus: Dynamic modulus tests presented the best set of data to understand differences between cores versus plant-produced lab-compacted mixtures and laboratory-produced mixtures. Trends were essentially the same in all three versions where cores were measurably softer. Dynamic modulus and associated parameters that incorporate the viscous phase angle were not consistently strong indicators of rutting or fatigue cracking. However, the modulus is still absolutely necessary and an important component for mechanistic-empirical performance prediction and VECD analysis.
• TTI OT: The TTI OT captured the rank order of ALF fatigue cracking very well for a smaller group of materials but only provides an index and not an engineering property that is useful by itself for pavement design. TTI is continuing development of the test for fatigue cracking in pavement design.
• DENT: DENT mixture testing for EWF and calculated CTOD was also a fairly strong indicator of fatigue cracking, but the required number of replicates and the scatter in the analysis are drawbacks. The test also does not yield an engineering property.
• Wheel tracking: The French PRT and HWT were unable to satisfactorily capture the rank order of the full-scale ALF rutting.
• IDT strength: IDT strength alone is not a strong indicator to identify fatigue cracking resistance.

FWD

FWD testing revealed variation in the stiffness of the unbound layer within each binder’s test lane and between test lanes. Back-calculation programs EVERCALC and MODCOMP, having different optimization algorithms, were evaluated and found to provide the same trends. The average stiffness of the crushed stone base was between 11.9 and 9.6 ksi (82 and 66 MPa), slightly stiffer than the average subgrade stiffness ranging between 11.4 and 11.1 ksi (79 and 77 MPa). Known depth to bedrock was easily detected. EVERCALC moduli were chosen because the base was generally stiffer and the subgrade was generally softer. Predicted HMA tensile strain using laboratory-measured modulus and FWD back-calculated base and subgrade stiffness gave reasonable magnitudes compared to what was measured from embedded strain gauges.

The use of MDDs was key in evaluating the quality of the back-calculations. Forward predictions of the base and subgrade layer deflections using the back-calculated modulus revealed very good agreement on the top of the base layer, fair agreement in the center of the base layer, and less agreement at the bottom of the base layer. This indicated that the back-calculation could have been additionally optimized. However, the focus of the experiment was the HMA layers, and the back-calculated moduli were deemed sufficient given that the critical strains and stresses for the purposes of this study were located within the asphalt layers. A seasonal monitoring program enables the depth to the seasonally-affected and unaffected subgrade layer to be estimated and identified more accurately.

Mechanistic–Empirical Pavement Performance Analysis

The key finding is as follows:

• Additional mixture-specific characterization inputs are needed above and beyond the |E*| dynamic modulus to be able to better discriminate and rank performance of modified and unmodified asphalt.

The NCHRP 1-37A and NCHRP 1-40D MEPDG models are largely calibrated using data from LTPP.⁽¹⁾ These do not include any significant amounts of polymer modified asphalt data, whereas the ALF experiment included several different polymer modified asphalts. Consequently, the rank order and the magnitude of predicted rutting and fatigue cracking were not accurate. In fact, the actual MEPDG software was not able to be used for the predictions because it was not developed to accommodate specialized APT conditions (i.e., controlled temperatures, heavy wheel loads). A standalone application that emulates the MEDPG had to be used.

Just as with the binder parameters previously discussed, modulus measured at small strains (binder |G*| and mixture |E*|) does not mobilize the material into regions that force the particular performance aspects to be revealed, whether permanent deformation or fatigue crack resistance. Along these lines, the NCHRP 9-30A project is pursuing this type of material characterization input and recalibration of the guide for improved rutting prediction.⁽⁹⁰⁾ This study illustrated an approach similar to the methodologies that will come out of NCHRP 9-30A, which use flow number permanent deformation curves to yield mixture-specific coefficients for the MEPDG model framework rather than a single national calibration for all HMA. Magnitudes of the predicted rutting were improved drastically, but the ranking was still not captured. However, the statistical similarities and variability in the measured rutting bracket the improved predictions. Additional insight gained from the flow number-based rutting prediction analysis is that the value of characterization at more than one temperature would allow a better prediction of rutting over a range of temperatures.

Two aspects of damage and fatigue cracking that the MEPDG does not provide were highlighted. Seismic PSPA tests on two ALF lanes illustrated that HMA modulus decreases by as much as 50 percent before fatigue cracks show at the surface. HMA modulus in the MEPDG is not recursively updated to reflect damage and modulus reduction. The pattern in which actual bottom-up fatigue cracking develops was not captured whereby the surface remains crack free until surface cracking is initiated and propagates.

Despite these shortcomings, the MEPDG was useful in analyzing the uniformity of ALF construction. The national calibration containing many different material characteristics and structural features taken from LTPP can be assumed to adequately capture the impact variations in layer thickness, density (modulus), and base and subgrade support have on predicted rutting and fatigue. Three scenarios were evaluated: as-built, as-designed, and as-designed with fixed (average) base and subgrade stiffness. The very similar rankings revealed from the exercise indicated no significant changes in the rank order.

RECCOMENDATIONS

The following recommendations are made based on the conclusions from this research:

• Adopt CTOD for the next generation of binder specification tests for fatigue cracking resistance.
• Support continued refinement and implementation of cyclic axial fatigue protocols in AMPT equipment. This is ongoing in two FHWA-supported research activities: performance-related specifications and Western Research Institute Asphalt Research Consortium Technology Deployment activity.
• Consider application of gap-graded crumb rubber in thicker layers and pavement structures that take advantage of placing the bottom of the layer at or near a neutral axis to arrest the propagation of fatigue cracks.
• Review literature and survey agencies to evaluate performance and cost benefits of other fiber-modified HMA projects’ performance to compare and contrast with ALF performance.

A number of follow-up activities are recommended, as follows:

• Develop draft CTOD specifications in AASHTO format beginning with Ontario Method LS-299 and AASHTO T 300.^(71,106)
• Fabricate molds and tabs for the DENT test and distribute to participating agencies.
• Present strengths of CTOD to AASHTO and the FWHA Asphalt ETGs to gather written comments on implementability and applicability as specification test (some presentations have been made to ETG already).
• Organize an exploratory round-robin study in preparation for ruggedness evaluation of CTOD.
• Explore the LTTP database to identify sites with different levels of fatigue cracking performance and characterize the binders for CTOD for added validation (ongoing).
• Conduct flow number tests at temperatures other than 147 °F (64 °C) to refine the flow number-based rutting performance of ALF mixture following the recommendations of NCHRP 9-30A as they are developed.
• Quantify the cost and benefits of fiber and crumb rubber for the particular ALF pavements structure and performance.
• Expand upon PSPA characterization of ALF lanes in future experiments.
• Ensure that absorption of cores for density is measured in future quality assurance during ALF construction and consider sampling from behind the paver rather than from within trucks.