Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results (2021)

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.

29   Research Approach Overview A refined PAM was established in this study by calibrating the kinetics model predictions of field aging using Equation (1) as a function of depth for a wide range of pavement sections, including both conventional HMA and modern materials (i.e., RAP, WMA, and PMA). Field aging levels were measured at different depths from binder extracted and recovered from in-service field cores. In addition, the component materials were used to prepare either loose mixtures that were aged in an oven at 95°C for a prolonged duration or USAT binder plates that also were long-term aged at 95°C. For either the loose mixture aging procedure or USAT, samples were removed periodically from the oven, and the binder AIPs were measured and used to derive the oxidation kinetics model parameters. Pavement hourly temperatures as a function of depth were used along with the kinetics model to predict field aging levels as a function of pavement depth. The predicted aging levels were compared against field aging that was measured from the field cores to calibrate the PAM at different depths. Figure 15 sum- marizes the experimental plan to conduct this task. Recall that the kinetics model established in the original NCHRP Project 09-54 ignores diffusion that is impacted by asphalt mixture morphology. Thus, a systematic study of the effect of asphalt mixture morphology on field aging was conducted using field cores obtained from projects that had systematic changes made to the mixtures’ air void and asphalt contents. The results were used to establish an adjustment to the PAM predictions to account for deviations from the Superpave® optimum asphalt contents. The PAM predictions were then compared against the Global Aging System (GAS) model. Test Materials and Field Projects Investigation into the Effects of Asphalt Mixture Morphology on Pavement Aging Table 4 summarizes the materials that were used to study the effects of asphalt mixture morph- ology on field aging. The field cores were obtained from the WesTrack project in Dayton, Nevada, and a project in Fundao, Brazil. These projects were ideal for studying the effects of morphology on aging as their materials offered systematic variations in asphalt content and air void content. The WesTrack project included field pavements constructed with three levels of asphalt content: Superpave optimum, low (Superpave optimum minus 0.7%), and high (Superpave optimum plus 0.7%) as well as three air void content levels: low (4%), medium (8%), and high (12%). In the naming system, WTFLM, for example, means a WesTrack field core (WTF) with low C H A P T E R 4 Refinement of the Pavement Aging Model (PAM)

30 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results asphalt content (L) and medium air void content (M). The Brazil field cores were 150-mm-diameter and 70-mm-height gyratory-compacted samples that were inserted in panels and subjected to 8 years of field aging (with no traffic loading) in Brazil, as shown in Figure 16. These Brazil field cores were prepared using four binders (Galp, LUBNOR, REPAR, and REPLAN) with different asphalt contents and air void contents, as indicated in Table 4. The inner 100 mm of the Brazil cores were extracted to study the variation in oxidation levels. Field Calibration of the Oxidation Kinetics Model The materials that were used to refine the CAI to prescribe the laboratory aging durations detailed in Table 3 also were used to calibrate the oxidation kinetics model. These test sections cover a wide range of pavement design, climatic conditions, ages, binder and aggregate char- acteristics, air void contents, asphalt contents, and gradations. Materials from these sections were used to recalibrate the laboratory aging durations as well as to recalibrate the PAM with depth, with the addition of the field cores from Fundao, Brazil. The systematic changes in the air void and asphalt contents of the WesTrack and Brazil field cores provided information to develop an adjustment to the PAM predictions to account for deviations from the Superpave optimum asphalt contents. The recalibration of the PAM builds upon the work completed under the original NCHRP Project 09-54 by incorporating WMA, RAP, and PMA sections along with the field core replicates from projects previously included in the initial calibration. Table 5 pre- sents detailed information regarding the test sections that were selected to calibrate and validate the PAM. The purpose of the validation sections is to have some confidence in the predictive capability of the PAM for sections that were not used in its calibration. It is important to note that the measurements from field cores exhibit a large variability. The scatter shown in Figure 8 (b) is just an example of how variable field core measurements can be. Understandably, such variability exists, though especially at the surface of the pavement, which is affected by UV oxi- dation and other surface wear mechanisms. Some sections were found to have log |G*| values at Figure 15. Experimental plan to calibrate PAM against field data.

Refinement of the Pavement Aging Model (PAM) 31   Project ID Location Age of Field Cores Section # Cores Tested Binder Grade WesTrack Dayton, Nevada 19 years (S4) Fine, Opt. ac%, L AV% (WTFOL) 4 PG 64-22 (S2) Fine, Low ac%, M AV% (WTFLM) 4 (S1) Fine, Opt. ac%, M AV% (WTFOM) 4 (S14) Fine, High ac%, M AV% (WTFHM) 4 (S17) Fine, Opt. ac%, H AV% (WTFOH) 4 (S18) Fine, High ac%, L AV% (WTFHL) 4 (S3) Fine, Low ac%, H AV% (WTFLH) 2 Brazil Fundao, Brazil 8 years Galp, 4.1% ac, 4% AV 1 PG 64-16 Galp, 4.1% ac, 8% AV 1 Galp, 4.6% ac, 4% AV 1 Galp, 5.1 % ac, 4% AV 1 Galp, 5.1 % ac, 8% AV 1 LUBNOR, 3.8% ac, 4% AV 1 PG 70-22 LUBNOR, 3.8% ac, 8% AV 1 LUBNOR, 4.3% ac, 4% AV 1 LUBNOR, 4.8% ac, 4% AV 1 LUBNOR, 4.8% ac, 8% AV 1 REPAR, 4.6% ac, 4% AV 1 PG 64-16 REPAR, 4.6% ac, 8% AV 1 REPAR, 4.1% ac, 8% AV 1 REPAR, 5.1% ac, 4% AV 1 REPAR, 5.1% ac, 8% AV 1 REPAR, 6% ac, 4% AV 1 REPLAN, 4.1% ac, 4% AV 1 PG 64-22 REPLAN, 4.1% ac, 8% AV 1 REPLAN, 4.6% ac, 4% AV 1 REPLAN, 4.6% ac, 8% AV 1 REPLAN, 5.1% ac, 8% AV 1 REPLAN, 5.1% ac, 8% AV 1 Note: ac% = asphalt content; AV% = air void content Table 4. Selected test sections for analysis of morphological properties. short-term aging that were approximately equal to or greater than the log |G*| values measured from field cores. These sections, thus, were not used in the calibration process or in deciding whether the model is validated but were used to check if the model is capable of making reason- able, long-term aging predictions despite potential laboratory short-term aging problems; these sections are designated with “STA Problem” in Table 5. Note that all the mixtures that contain RAP had the STA problem. Therefore, these RAP sections were not used for the calibration but were instead used for the verification of the calibrated PAM.

32 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results Figure 16. Setting up gyratory-compacted samples in panels for eld aging in Brazil.

Refinement of the Pavement Aging Model (PAM) 33   Project ID Location Age of Field Cores Section (Section ID) Binder Grade WMA Modified Binder Contains RAP Use FHWA ALF Virginia 8, 11 years Control-2002-Lane 2 (ACTRL) PG 70-22 Calibration 8 years SBS-LG-2002-Lane 4 (ASBS) PG 70-28  CalibrationCRTB-2002-Lane 5 (ACRTB) PG 76-28  Calibration 8, 11 years Terpolymer-2002-Lane 6 (ATerp) PG 70-28  Validation MIT Manitoba, Canada 4 years Control (MWC) Pen 150/200 Validation Advera-PTH 14 (MWA)  STA Problem Evotherm-PTH 14 (MWE)  STA Problem 15% RAP-PTH 8 (M15R)  STA Problem 50% RAP-PTH 8 (M50R)  STA Problem NCAT Alabama 4 years Control-S9 (NWC) PG 76-22 Calibration 50% RAP-N10 (N50R) PG 67-22  STA Problem Evotherm-S11 (NWE) PG 76-22  Validation 50% RAP with foam-N11 (N50RF) PG 67-22   STA Problem Foam-S10 (NWF) PG 76-22  Calibration LTPP Ohio 11 years SPS-1 ID: 39-0111 (LOH) AC-20 Validation California 8, 15 years SPS-8 ID: 06-A805 (LCA) AR-40 Validation New Mexico 10, 18 years SPS-8 ID: 35-0801 (LNM) AC-20 Calibration Texas 11, 18 years SPS-8 ID: 48-0802 (LTX) AC-20 Calibration Washington 11, 19 years SPS-8 ID: 53-0801 (LWA) AC-20 Validation Wisconsin 8, 17 years SPS-8 ID: 55-0806 (LWI) - Calibration South Dakota 14 years SPS-8 ID: 46-0804 (LSD) Pen 120/150 Calibration WesTrack Nevada 4, 19 years (S4) Fine, Opt. ac%, L AV% (WTFOL) PG 64-22 Calibration (S2) Fine, Low ac%, M AV% (WTFLM) Calibration (S1) Fine, Opt. ac%, M AV% (WTFOM) Calibration (S14) Fine, High ac%, M AV% (WTFHM) Calibration (S17) Fine, Opt. ac%, H AV% (WTFOH) Calibration (S18) Fine, High ac%, L AV% (WTFHL) Calibration (S3) Fine, Low ac%, H AV% (WTFLH) Calibration 17 years (S39) Coarse, Opt. ac%, L AV% (WTCOL) PG 64-22 Calibration(S36) Coarse, Opt. ac%, H AV% (WTCOH) Calibration MnROAD Minnesota 13 years S31 (MNS31) PG 64-34  Validation Note: SBS-LG = linear grafted styrene-butadiene-styrene; CRTB = crumb rubber terminal blend; SPS = specific pavement study; ac% = asphalt content; AV% = air void content; STA = short-term aging; L = low, M = medium, H = high. Table 5. Selected test sections for calibration and validation of PAM. Sample Preparation Methods The sample preparation methods used to refine the PAM largely coincide with those described in Chapter 3 but are repeated here for the convenience of the reader. Asphalt Mixture Aging All the asphalt mixtures aged in the laboratory were prepared using the component aggre- gate and binder that were used to construct the pavements from which the field cores were obtained. The mixtures were subjected to short-term aging at 135°C for 4 hours in accordance with AASHTO R 30 prior to long-term aging. The long-term oven aging of the loose mixtures was accomplished by separating the mix into several pans such that each pan had a relatively thin layer of loose mix that was approximately equal to the NMAS of the aged mix. The pans of loose mixture were conditioned in an oven at 95°C and systematically rotated to minimize any effects of an oven temperature gradient and/or draft on the degree of aging. After long-term aging, the materials were taken out of the oven and mixed to obtain a uniform mixture. Field Core Preparation Full-depth cores were acquired from in-service pavements and the FHWA’s Materials Ref- erence Library. The field cores were wrapped with plastic wrap and placed in a temperature- controlled room to minimize further aging during storage. Following storage, the upper 50 mm

34 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results of each field core were sliced to create four 12.5-mm-thick discs. The remainder of each field core was cut into 25-mm-thick discs. Special care was taken to avoid the tack coat and prime coat layers when slicing the field cores. The upper four discs were subjected to binder extraction and recovery, and the recovered binder was subsequently subjected to rheological testing. Testing only the upper four discs was found to be sufficient to determine the aging gradient at the top of the pavement. While the deeper sections of the pavement still undergo long-term aging com- pared to the short-term aging condition, an aging gradient is not as appreciable as that observed closer to the surface of the pavement. Microextraction and Recovery Microextraction and recovery of the asphalt binder from the asphalt mixtures and field cores were undertaken following the procedure proposed by Farrar et al. (2015). This procedure uses a solvent mixture of toluene and ethanol (85:15). The mixture sample size is limited to 200 g to produce approximately 10 g of asphalt binder per extraction, which is adequate for both ATR- FTIR spectrometry testing and DSR testing. To prevent further aging of the binder, the distilla- tion flask was subjected to vacuum pressure of 80.0 ± 0.7 kPa (600 ± 5 mm Hg) under nitrogen gas during the recovery procedure. The recovered samples were then placed in a degassing oven and heated to 130°C for 60 minutes under nitrogen to remove any remaining traces of the sol- vent. In this study, ATR-FTIR spectrometry testing was conducted following extraction and recovery to ensure that no detectable solvent was present prior to DSR testing. USAT The USAT, developed by Farrar et al. (2015) at the Western Research Institute, was employed in this study to derive the oxidation kinetics of asphalt binders as part of the aging model development. In the USAT, binder is placed in grooved plates to achieve a film thickness of 300 micrometers. In this study, the USAT plates were placed in an oven at 135°C for 4 hours to simulate the short-term aging of the loose mixtures. After this binder short-term aging process, the USAT plates were placed in an oven at 95°C to simulate long-term aging. The aging kinetics obtained through USAT were converted to loose mixture aging kinetics using the method briefly illustrated in Figure 4 and developed in the original NCHRP Project 09-54 under NCHRP Research Report 871 (Kim et al. 2018). The study on the use of USAT aging was conducted as part of the investigation aimed to eliminate the need for loose mixture aging in the use of the models developed in this project. The aging kinetics obtained through USAT aging were only used for the validation of model predictions. Determination of Rheological Aging Index Properties Frequency sweep tests of the extracted and recovered binders were conducted using an Anton Paar modular compact rheometer (MCR) 302 at frequencies ranging from 0.1 Hz to 30 Hz at 64°C using the 8-mm or 25-mm parallel plate geometry. A strain amplitude of 1% was applied at all frequencies. The log |G*| at 64°C and 10 rad/s was used as the rheological AIP for all analyses. Determination of Project-Specific, Hourly Pavement Temperature Histories The MERRA-2 hourly climatic data stations with coordinates and elevations closest to each project location were identified and used to obtain the most accurate weather information available. After the appropriate stations were chosen, 37 years of available weather data (1980 through 2017) were input and analyzed in the EICM to obtain hourly pavement temperature

Refinement of the Pavement Aging Model (PAM) 35   histories as a function of pavement depth. The sub-layering and nodal structure specified in the MEPDG were adopted for all EICM analyses (Applied Research Associates 2004). Prediction of Aging Index Properties for Variable Temperature History Using the Kinetics Model The kinetics model can directly predict log |G*| for any isothermal temperature for any duration. The calculations become more cumbersome when a non-isothermal temperature history is input into the model. An algorithm was developed to conduct these calculations for an hourly pavement temperature history. At each time step, a reduced time is calculated for the given temperature, which is used to determine the difference in log |G*| for that time step and temperature. The difference in log |G*| is then added to the log |G*| from the previous time step. This calculation is done incrementally where log |G*| acts as a state variable allowing the “jumps” between different kinetics curves corresponding to different temperatures. Figure 17 is an illustration of this calculation. T1 is the kinetics curve for the first temperature, and T2 is the kinetics curve for the second temperature, such that T1 > T2. For a temperature history of T1-T2-T1, log |G*| is calculated first based on the kinetics curve of T1. The difference in log |G*| is then calculated based on the kinetics curve of T2 for that time step and incremented to log |G*| of the previous time step. Finally, the kinetics curve of T1 is used again to calculate the dif- ference in log |G*| for that time step and incremented to the log |G*| of the previous time step. Findings and Applications Statistical Investigation into the Effects of Asphalt Mixture Morphology on Pavement Aging The log |G*| values of the field cores obtained from the WesTrack and Brazil projects as a function of depth were used to evaluate the effects of asphalt mixture morphology on field aging. These two projects offered systematic changes in asphalt content and air void content. Detailed WesTrack and Brazil data are shown in Appendix A. The analysis results were expected to show that the aging level, and therefore the log |G*| value, would decrease with depth for a given section given the temperature gradient and diffusion distance from the pavement surface. In addi- tion, the sections with lower asphalt contents were expected to have higher log |G*| values than sections with higher asphalt contents (given the same binder, aggregate, and air void content) due to the thinner film thickness that creates a shorter diffusion path for oxygen to cause oxida- tion. Also, an increase in air void content was expected to facilitate the percolation of the oxygen lo g |G *| a t 6 4° C, 1 0 ra d/ s (k Pa ) Aging Duration (days) T1 T2 T1-T2-T1 temperature history Figure 17. Illustration of the algorithm to calculate log |G*| using the kinetics model.

36 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results within a mixture and thus yield increased aging levels. However, as will be shown, the results deviated from these expectations in some cases. Statistical analysis was performed using the WesTrack and Brazil materials. For the WesTrack materials, analysis of the asphalt content and air void content was conducted. For the Brazil materials, analysis of binder type, asphalt content, and air void content was conducted. Table 6 and Table 7 present the statistical analysis results for the WesTrack and Brazil materials, respec- tively. In both tables, the variables under the column labeled “Tukey” are the variables that are statistically different from one another. Based on a 0.05 significance level, if the p-value given by the statistical coding software, “R,” and obtained from the input data is less than 0.05, then the values for average log |G*| will differ from one another. The WesTrack analysis, summarized in Table 6, was performed by pavement layer because each layer, designated as T1 through T4, corresponds to a different depth: 0.6 cm, 1.9 cm, 3.2 cm, Brazil p-value Tukey Binder Type Analysis T1 1.76E-05 REPAR-Galp LUBNOR-Galp LUBNOR-REPLAN LUBNOR- REPAR T2 1.30E-06 REPAR-Galp LUBNOR-Galp LUBNOR-REPLAN LUBNOR- REPAR T3 1.27E-04 REPAR-Galp LUBNOR-Galp LUBNOR-REPLAN T4 0.00952 REPAR-Galp LUBNOR-Galp Asphalt Content Analysis T1 0.0027 3.8–5.1%, 4.8–5.1%, 4.1–4.8% T2 0.0042 3.8–5.1%, 4.8–5.1%, 4.1–4.8% T3 0.0510 None T4 0.3620 None Air Void Content Analysis T1 0.9007 None T2 0.7016 None T3 0.3629 None T4 0.1213 None Table 7. Summary of statistical analysis for Brazil materials. WesTrack p-value Tukey Asphalt Content Analysis T1 0.3734 None T2 0.0325 Low-High T3 0.0053 Opt-HighLow-High T4 0.0889 None Air Void Content Analysis T1 0.4786 None T2 0.2132 None T3 0.1430 None T4 0.0182 Low-Medium Table 6. Summary of statistical analysis for WesTrack fine materials.

Refinement of the Pavement Aging Model (PAM) 37   and 4.5 cm, respectively. R was used to aid the analysis of the data sets. First, an analysis of vari- ance (ANOVA) test was run using the set of data for T1 (0.6 cm), grouped by binder type. The p-values indicate whether the average log |G*| values at 64°C and 10 rad/s differ for different binder types (or asphalt content/air void content). Then, the Tukey “honest significant differ- ence” (HSD) test was run to compare specific mean values of the different variables against each other. The R program output specifically identifies which variables differ (for example, high air void content versus low air void content). The WesTrack results presented in Table 6 show a more pronounced relationship between asphalt content and depth than between air void con- tent and depth. This can also be observed visually in Figure 18. The sections are labeled with their asphalt content and air void content. For example, section LM stands for low asphalt content (L) and medium air void content (M). From Figure 18, section HL (high asphalt content and low air void content) is obviously different when compared to the other sections. More statistically dif- ferent results are evident for T2 and T3 when looking at asphalt content and for T4 when looking at air void content. For both analyses (p-value and Tukey HSD), results would be significantly different if either or both of the asphalt content and air void content are high or low. The Brazil results presented in Table 7 show that, overall, the mean values for the LUBNOR binder seem to differ from the other means more significantly than for any other binder type. Galp seems to differ significantly as well, just not as often as LUBNOR. As the depth increases (T1→T4), fewer binders differ significantly from one another. These results make sense because aging affects the surface layers more than the layers deeper in the pavement. Asphalt content does not play a large role as the depth of the pavement increases (T3 and T4 versus T1 and T2). More significantly different results are evident for the upper layers of the pavement compared to the deeper layers. This conclusion is the same as that for the comparison of binder types. The asphalt contents of 3.8%, 4.1%, 5.1%, and 4.8% are significantly different for T1 and T2. When looking at air void content for each level of depth, none of the results differed significantly from one another (4% versus 8%). As the depth increased, the results did start to deviate from each other a bit more, but the difference still was not statistically different. This outcome has been observed in past analyses by looking at the graphical results of DSR data. In summary, the statistical analysis of the Brazil and WesTrack field cores showed that asphalt content has a significant effect on field aging. The Brazil data furthermore showed the significant 0 1 2 3 4 5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) LM LH HL HM OL OM OH Figure 18. All WesTrack sections delineated based on their asphalt content (first letter: L for low, H for high, O for optimum) and air void content (second letter: L for low, M for medium, H for high).

38 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results effect of asphalt type that is expected. However, the air void content was found to have an insignificant effect on field aging of the WesTrack and Brazil field cores. In later analysis, the WesTrack sections are analyzed to calibrate an adjustment to the PAM predictions as a function of deviation from the Superpave optimum asphalt content. Field Calibration of the Oxidation Kinetics Model The analyses of the effects of asphalt mixture morphology on the aging of the WesTrack and Brazil field cores provided insufficient insight to develop a rigorous diffusion model. In addition, rigorous diffusion models require detailed asphalt mixture information that is not typically available. For example, the Transport Model developed at Texas A&M University includes a fundamental diffusion model (Lunsford 1994; Prapaitrakul 2009; Han 2011; Jin, Cui, and Glover 2013). However, the Transport Model requires x-ray–computed tomography to quantify the accessible air void content of a given asphalt mixture and requires that the partial pressure-dependence of oxidation kinetics is quantified or assumed (Jin, Cui, and Glover 2013). Even with this detailed information, the Transport Model still requires field, project-specific calibration of the diffusivity (Jin, Cui, and Glover 2013), further limiting the practicality of implementation within ME pavement analysis frameworks. Therefore, instead of a diffusion model, depth-dependent field calibration of the kinetics model was carried out using all proj- ects designated as “Calibration” in Table 5; all of these asphalt mixtures were prepared at the optimum asphalt content. Equation (1) was applied to predict the log |G*| at 64°C and 10 rad/s in the field using hourly pavement temperature history data at different depths. The hourly pavement temperature history was obtained from EICM analysis of MERRA-2 weather files, consistent with the CAI analyses. The kinetics model-predicted log |G*| values were compared to those of binders extracted and recovered from field cores at different depths following a data smoothing process to remove spurious trends. The relationship between the kinetics model predictions and smoothed field core measurements was used to develop the depth-dependent field calibration function. The WesTrack field core results were then used to develop adjust- ments to model predictions as a function of deviations from the optimum asphalt content. No adjustments were proposed for air void content based on the analysis results for the WesTrack and Brazil field cores that demonstrated that air void content has a statistically insignificant effect on field aging. Field Core Data Quality and Smoothing A rational function was fitted to represent the relationship between the measured log |G*| values versus depth for each section (listed in Table 5) prior to field calibration to account for the inherent variability in the field core results as a function of depth. The data smoothing ensured reasonable trends with depth for each section prior to the field calibration. Looking at each data set, outliers were identified that clearly did not follow the expected data trend and thus were removed from the data smoothing process to improve the quality and precision of the rational function prediction of log |G*|. The MIT sections (MWA, MWE, M15R, and M50R) and NCAT sections (N50R and N50RF) exhibited log |G*| values at short-term aging that were approximately equal to or greater than the log |G*| values measured from field cores aged for 4 years in the field. This trend contradicts the expected increasing evolution of |G*| with time. This conflicting observation is attributed to the laboratory short-term aging procedure that overestimates field aging in cold climates. The laboratory short-term aging procedure that was used follows AASHTO R 30 that requires aging loose mixture in the oven at 135°C for 4 hours. Because of the questionable initial log |G*| values obtained, these sections were left out of the pool of calibration sections but were still included to check if the model was capable of making reasonable long-term aging predictions.

Refinement of the Pavement Aging Model (PAM) 39   Relationship Between Kinetics Model Predictions and Smoothed Field Core Measurements The relationship between the kinetics model predictions and the smoothed field core mea- surements of log |G*| values at different depths was evaluated to inform the field calibration of the PAM. Figure 19 shows the relationship between the smoothed log |G*| values measured from field cores and those predicted from the kinetics model at all depths. The sections are labeled as “mixture ID-field age.” For example, ACTRL-8Y stands for the ALF Control section aged in the field for 8 years. In addition, • ALF sections are shown in yellow, • MIT sections in red, • NCAT sections in blue, • LTPP sections in green, • WesTrack sections in pink, and • MnRd section in black. Note that the MIT sections (MWA, MWE, M15R, and M50R) and NCAT sections (N50R and N50RF) were not included in the field calibration of the PAM but are shown here in gray. According to this figure, the only clear outlier is M50R. The predicted log |G*| value is highly dependent on the initial log |G*| value (i.e., the log |G*| value of the short-term aged loose mixture). Because the initial log |G*| is too high, the predicted log |G*| is similarly biased in comparison to the field-measured log |G*|. The other RAP sections and WMA sections seemed to follow the same trend as the con- ventional sections, which indicates that a separate field calibration of the kinetics model for the RAP and WMA sections is not required. However, insufficient data for the RAP and WMA sections due to the high initial log |G*| values obtained using the laboratory short-term aging procedure in AASHTO R 30 rendered the data untrustworthy. Figure 19. Comparison between smoothed field core measurements and kinetics model predictions of log |G*| values for all field sections at all pavement depths.

40 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results Field Calibration of the Pavement Aging Model The field calibration of the PAM is intended to yield a log |G*| gradient with depth and time that meets a number of conditions. These conditions are established based on field core obser- vations and engineering judgment. It is meant to provide a log |G*| gradient with depth and time that satisfies boundary conditions to be used in a pavement performance analysis software. Figure 20 shows log |G*| measurements from four field sections with field cores obtained at two aging durations. The sections are labeled as “mixture ID-field age,” e.g., ACTRL-8Y stands for the ALF Control section aged in the field for 8 years. For each section, the aging gradient with depth and with time is shown. The field core measurements from these sections are used to infer the conditions listed in the next paragraph. The results of the field-calibrated PAM should comply with the following conditions: 1. At time zero, log |G*| is equal to log |G*| 0 at all depths. 2. The depth gradient becomes steeper as the depth gets closer to the pavement surface. 3. The depth gradient reaches to an asymptote as the pavement depth increases. 4. The asymptote increases with increasing time. 5. The change of the depth gradient with time experiences an initial rapid evolution followed by a slower evolution toward an asymptote with time. 6. The log |G*| value at the pavement surface does not exceed a maximum log |G*| value, deter- mined to be equal to 4.5 kPa through a long-term binder aging study presented later in this section. 7. The change in asymptote as a function of time decreases with increasing depth. 8. PAM gradient predictions must match field core measurements of log |G*| visually and quan- titatively through R2 and root mean square error (RMSE). The conditions described in 1–7 above are depicted in Figure 21. The field calibration of the PAM is based on what will henceforth be called the “Aging Param- eter,” abbreviated as AP and defined in Equation (16). The AP is derived from Equation (1) to provide the predicted change in log |G*| due to long-term aging, normalized by the material- specific kinetics model parameter M. The right side of Equation (16) shows that, according to the kinetics model given in Equation (1), AP is only a function of pavement temperature and time since the material-specific parameters (i.e., log |G*| and M) only appear in the middle portion of the equation. ( )( )= − = −      − − + log * log * 1 1 exp (16)0AP G G M k k k t k tkinetics c f f c The AP values are calculated for all sections using the hourly pavement temperature histories and kc and kf parameters in Equation (1). These values are compared to APfield values calculated using the smoothed field core log |G*| measurements, combined with log |G*|0 and M obtained from laboratory loose mixture aging. An example of AP calculated using Equation (16) for multiple pavement depths is shown in Figure 22 (b). AP is calculated using hourly pavement temperature history. An example of only the hourly pavement surface temperature history is shown in Figure 22 (a). The corresponding pavement temperature histories are used for the calculation of AP at other pavement depths. As demonstrated in Figure 22 (b), AP (and consequently log |G*|kinetics) does not exhibit a rapid initial evolution nor does it slow with time to reach an asymptote (conditions 5, 6, and 7). This trend can be problematic as log |G*|kinetics would continuously increase with time, which is unrealistic and contrary to the field core measurements shown in the right column of Figure 20.

Refinement of the Pavement Aging Model (PAM) 41   0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) LWI-8Y LWI-17Y (a) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) LNM-10Y LNM-18Y (c) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) WTFOM-4Y WTFOM-19Y (e) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) ACTRL-8Y ACTRL-11Y (g) 0 1 2 3 4 5 0 5 10 15 20 25 30 Time (years) LWI-8Y LWI-17Y (b) 0 1 2 3 4 5 0 5 10 15 20 25 30 Time (years) LNM-10Y LNM-18Y (d) 0 1 2 3 4 5 0 5 10 15 20 25 30 Time (years) WTFOM-4Y WTFOM-19Y (f) 0 1 2 3 4 5 0 5 10 15 20 25 30 Time (years) ACTRL-8Y ACTRL-11Y (h) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) lo g |G *| at 6 4° C , 1 0 ra d/ s lo g |G *| at 6 4° C , 1 0 ra d/ s lo g |G *| at 6 4° C , 1 0 ra d/ s lo g |G *| at 6 4° C , 1 0 ra d/ s Figure 20. Field core log |G*| measurements plotted versus depth and versus time, respectively, for: (a) and (b) LWI, (c) and (d) LNM, (e) and (f) WTFOM, and (g) and (h) ACTRL.

42 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results Figure 23 shows AP plotted versus depth starting at time zero and plotted for time increments of 2 years. The hourly pavement temperature history for each depth and time increment is used to calculate AP in this graph. The effect of temperature change across the pavement depth results in a mild-sloped gradient, shown in Figure 23, with a higher AP value and thus log |G*|kinetics at the pavement surface compared to that at a depth of 5 cm. It is expected, though, that the gradient has a steeper slope in reality, with an even higher value near the surface and a lower value at 5 cm depth, as shown in left column figure of Figure 20. Also, as demonstrated, the gradient seems to increase at a constant rate at each time increment and does not experience a slower evolution with longer field aging durations. This is also demonstrated by overall linear-increasing behavior 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) 0 1 2 3 4 5 0 5 10 15 20 25 30 Time (years) lo g |G *| at 6 4° C , 1 0 ra d/ s 1 1 2 3 4 5 6 2 7 4 6 5 (a) (b) log |G*| at 64°C, 10 rad/s (kPa) Figure 21. Illustrations of aging gradient intended from the field calibration function as a function of (a) pavement depth and (b) time. The numbered circles illustrate the conditions 1 through 7. Time (hours) 0 2 4 6 8 10 12 14 16 18 x104 -20 0 20 40 60 80 Te m pe ra tu re (° C) (a) Time (hours) 0 2 4 6 8 10 12 14 16 18 x104 0 1 2 3 4 5 6 AP (k Pa ) Pavement Surface Pavement Depth 1 Pavement Depth 2 (b) Figure 22. AP evolution with time, based on pavement hourly pavement temperature history.

Refinement of the Pavement Aging Model (PAM) 43   shown in Figure 22, which is contrary to field core measurements shown in the right column of figures in Figure 20. Therefore, this trend violates conditions 2, 5, 6, and 7. In order to achieve the trend with time shown in Figure 21 (b) and satisfy condition 5 (i.e., an initial rapid evolution followed by a slower evolution toward an asymptote with time), AP must be transformed as shown in Equation (17) to yield what is referred to as APfield. Only AP at the pavement surface is used to generate the gradient with depth. As time increases to infinity, and thus APt increases to infinity, APfield reaches the asymptote n1. ( )= − − ×1 (17)1 2AP n efield n APt where n1, n2 = regression parameters, and APt = Aging Parameter at the pavement surface at time t, kPa. The asymptote n1 must be a function of depth to satisfy condition 7. The function with depth must also provide a steep gradient close to the pavement surface (condition 2) and must have the ability to define a fixed maximum value of log |G*| at the surface (condition 6). Equation (18) shows the rational function selected to define n1. = + +1 (18)1 1 2n b z b z where b1, b2 = regression parameters, and z = depth (cm). The rational function provides the desired gradient shape with depth (condition 2). At the pavement surface (depth of 0), n1 becomes equal to the constant regression parameter b2. Thus, b2 represents the maximum log |G*| value that can be achieved at the pavement surface (condi- tion 6). For infinite depth, n1 becomes equal to b1 (using L’Hôpital’s Rule to evaluate the limit at the indeterminate form). Accordingly, b1 acts as the asymptote with depth (condition 3) for n1. With increasing time (i.e., with increasing APt), APfield at infinite depth increases until it reaches a maximum at infinite time. Therefore, the overall asymptote with depth for APfield increases with time, satisfying condition 4. AP (kPa) D ep th (c m ) 0 2 4 6 8 101 3 5 7 9 0 1 2 3 4 5 Time = 0 years Time = 2 years Figure 23. AP plotted versus depth for time increments of 2 years.

44 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results The parameter b2 represents the maximum log |G*| value that can be achieved at the pavement surface. The material-independent maximum log |G*| would result in a material-dependent AP as shown in Equation (19). = −log * log * (19)max max 0AP G G M where APmax = maximum Aging Parameter at the pavement surface for each material (kPa), and log |G*|max = maximum log |G*| value at the pavement surface for all materials (kPa). In an effort to determine the maximum log |G*| value that a binder might reach in the field while still maintaining any integrity, two binders, LTX and LSD, were aged for prolonged periods of time in the oven using thin film. Samples of the binders were taken out at dif- ferent durations and tested using the DSR to obtain the log |G*|. The highest log |G*| value at 64°C, 10 rad/s that could be obtained through direct testing was between 4.0 kPa and 4.2 kPa. The binder aged for longer durations crumbled into pieces upon handling and could not be formed into a sample for testing. Although the modulus of such binder might theoretically continue increasing upon aging, the binder loses integrity and ceases to have any meaningful performance. Thus, a log |G*| value of 4.5 kPa is selected as the maximum value that can be reached under field conditions for the purpose of the PAM used in pavement performance simulations. In putting Equations (17), (18), and (19) together, and rewriting log |G*| in terms of APt, Equation (20) is obtained. ( )= + + +     − − ×log * log * 1 1 (20), 0 1 max 2G G M b z AP z et z n APt where log |G*|t,z = dynamic shear modulus at time t and depth z, kPa. The log |G*| values obtained from Equation (20) and the log |G*| measured directly from field cores are compared in an LOE plot. Application of Equation (20) improved on the original relationship between the predicted and the measured log |G*| (Figure 19), but left room for further improvement. Comparisons between the field-calibrated PAM predictions and field core measurements of log |G*| for the calibration sections and sections with STA problems are shown in Figure 24 for all depths. Figure 24 shows that, excluding the points in gray, the PAM underpredicts the field measurements for a few field sections (LWI, LSD, WesTrack), which appear above the LOE. Note that two of the three sections (LWI and LSD) happen to be in colder climates than other sections. The PAM also seems to overpredict the field measurements for other sections (NWC and NWF), which appear below the LOE and happen to be in warmer climates than other sections. To account for the climate-dependent bias observed in the predicted log |G*| values, a climate- based term is introduced in the parameter n2 as shown in Equation (21). As shown, n2 is now a function of AP calculated at the pavement surface and at 10 years of aging using Equation (16) (using the hourly pavement temperature history). The AP value calculated based on Equa- tion (16) is a function of time and temperature only, and thus, considering AP at a fixed time would distinguish the sections based on their climatic conditions. = − × = (21)2 1 2 10n a e a APt

Refinement of the Pavement Aging Model (PAM) 45   where a1 and a2 = regression parameters, and APt=10 = Aging Parameter at the pavement surface at 10 years of aging, kPa. By incorporating Equation (21) into Equation (20), the final calibration function for PAM appears in Equation (22). ( )= + + +     − − ×log * log * 0.916 1 1 (22), 0 maxG G M z AP z et z N APt where APt = −log * log *, 0G G M kinetics t APmax = −4.5 log * 0G M N = − × =e APt0.477 0.226 10 |G*|t,z = dynamic shear modulus at time t and depth z, kPa, |G*|0 = dynamic shear modulus at STA condition, kPa, |G*|kinetics,t = dynamic shear modulus at the pavement surface at time t, kPa, APt = Aging Parameter at the pavement surface at time t, kPa, APt=10 = Aging Parameter at the pavement surface at 10 years of aging, kPa, APmax = Maximum Aging Parameter at the pavement surface (kPa), N = material-dependent parameter, M = material-dependent constant, z = depth, cm, and t = time, days. 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 Predicted & Calibrated log |G*| at 64°C, 10 rad/s (kPa) ACTRL-8Y ACTRL-11Y ASBS-8Y ACRTB-8Y NWC-4Y NWF-4Y LNM-10Y LNM-18Y LTX-11Y LWI-8Y LWI-17Y LSD-14Y WTFOM-4Y WTFOM-19Y WTC-17Y MWA-4Y MWE-4E M15R-4Y M50R-4Y N50R-4Y N50RF-4Y LOE All Depths Yellow: ALF Blue: NCAT Green: LTPP Pink: WesTrack Gray: Sections w. STA Problems R2= 0.5988 RMSE = 0.2856 M ea su re d lo g |G *| af te r S m oo th in g at 6 4° C , 1 0 ra d/ s (k Pa ) Figure 24. Comparison between smoothed field core measurements and initial field-calibrated PAM predictions for calibration sections and sections with STA problems at all depths.

46 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results The field-calibrated PAM predictions of the log |G*| values were determined for each project and field core depth using Equation (22). Comparisons between the field-calibrated PAM pre- dictions and field core measurements of log |G*| for the calibration sections and sections with STA problems are shown in Figure 25 for all depths. Recall that some sections were found to have log |G*| values at short-term aging that were approximately equal to or greater than the log |G*| values measured from field cores. These sections, thus, were not used in the calibration pro- cess or in making the decision whether the model was validated, but were simply used to check if the model was capable of making reasonable, long-term aging predictions despite potential laboratory short-term aging problems; these sections are designated with “STA Problem.” The results demonstrate generally good agreement between the measured and predicted values, indicating the field-calibrated PAM yields reasonable predictions in terms of R2 and RMSE. The sections are labeled as “mixture ID-field age,” e.g., LWI-8Y stands for the LTPP Wisconsin section aged in the field for 8 years. In addition, • ALF sections are shown in yellow, • NCAT sections are shown in blue, • LTPP sections are shown in green, • WesTrack sections are shown in pink, and • STA problems are shown in gray for completeness, although they were not used in the calibration. For most of the sections with STA problems, the PAM overpredicts log |G*|. The predicted log |G*| value is highly dependent on the initial log |G*| value (i.e., the log |G*| value of the short-term aged loose mixture). As mentioned earlier, because the initial log |G*| is too high for these problematic sections, the predicted log |G*| is similarly biased in comparison to the 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 ACTRL-8Y ACTRL-11Y ASBS-8Y ACRTB-8Y NWC-4Y NWF-4Y LNM-10Y LNM-18Y LTX-11Y LWI-8Y LWI-17Y LSD-14Y WTFOM-4Y WTFOM-19Y WTC-17Y MWA-4Y MWE-4E M15R-4Y M50R-4Y N50R-4Y N50RF-4Y LOE All Depths Yellow: ALF Blue: NCAT Green: LTPP Pink: WesTrack Gray: Sections w. STA Problems R2 = 0.7687 RMSE = 0.2168 Predicted & Calibrated log |G*| at 64°C, 10 rad/s (kPa) M ea su re d lo g |G *| af te r S m oo th in g at 6 4° C , 1 0 ra d/ s (k Pa ) Figure 25. Comparison between smoothed field core measurements and field- calibrated PAM predictions for calibration sections and sections with STA problems at all depths.

Refinement of the Pavement Aging Model (PAM) 47   field-measured log |G*|. The overall prediction accuracy at all depths has increased, as illustrated by an increase in R2 between Figure 19 and Figure 25 from 0.6 to 0.77. The results show that within the limited data used in this study, the calibrated PAM seems to predict the aging of the conventional HMA mixtures, WMA mixtures, RAP mixtures, and PMA mixtures reasonably well. However, these findings should be confirmed by analyzing additional WMA and RAP materials in future work after resolving the issue of overestimation of the log |G*|0 values when using the current AASHTO R 30 short-term aging procedure. Figure 25 shows that the field-calibrated PAM yields predictions as the data is well centered along the LOE (condition 7). However, it is difficult to assess the accuracy of the predicted depth gradient of individual sections from Figure 25 (conditions 1 through 6). Therefore, the field- calibrated PAM is further evaluated by examining the measured and predicted depth gradients of individual sections. Figure 26 shows the evolution of log |G*| for 24 years with 1-year increments as shown in gray lines as well as the measured field core data in symbols and the corresponding predictions in dashed/dotted lines for all calibration sections. Note that the measured field core data presented in Figure 26 are not smoothened. Figure 26 shows that recognizing the variability of aging in field cores, the predictions from the calibrated PAM, in terms of the aging gradient shape and the level of aging, match those measured from the field cores from various climatic regions reasonably well. Previous discussions about the development of the field calibration function of PAM explained how conditions 1 through 6 are met. Figure 26, though, shows graphically the resemblance of the obtained gradient of each section to that shown in Figure 21. Figure 27 shows the evolution of log |G*| for ACTRL versus time and depth prior to calibra- tion and after calibration. The gray lines plotted against time represent the log |G*| prediction at depths corresponding to the field core measurements using data points obtained at 1-year increments. Figure 27 clearly shows the improvement in log |G*| prediction. This demonstra- tion of the prediction capability is important to understand the predictive behavior outside the range of measured data. The predicted log |G*| prior to calibration shows a continuous increase, which can be unrealistic after 20 to 25 years of aging. This trend is opposed by the field core measurements that show a slower evolution of log |G*| with time for ACTRL in Figure 27 and three more sections in Figure 20. Field Validation of the Pavement Aging Model Some field sections were left out of the calibration pool for independent validation. The field- calibrated PAM predictions of the log |G*| values were determined for each project and field core depth using Equation (22). Figure 28 presents a comparison of the smoothed field core measure- ments and field-calibrated PAM predictions of log |G*| for all the validation sections at all depths. The results do not indicate bias in the PAM predictions with depth, suggesting that the depth- dependent calibration is appropriate. The results show that within the limited data used in this study, the calibrated PAM may be able to predict the aging of the conventional HMA mixtures (MWC, LOH, LCA, LWA, MNS31), WMA mixtures (NWE), and PMA mixtures (ATerp). The overall R2 of 0.72 that corresponds to the aggregated data suggests promising prediction accu- racy, given all of the variables that impact field aging that are not taken into account directly in the PAM (morphology, UV aging, etc.). Figure 29 shows the evolution of log |G*| for 24 years with 1-year increments as shown in gray lines as well as the measured field core data in symbols and the corresponding prediction in dashed/dotted lines for all validation sections. Note that the measured field core data presented in Figure 29 are not smoothened. Recognizing the variability of aging in field cores, the predic- tions from the calibrated PAM in terms of the aging gradient shape and the level of aging match those measured from the field cores from various climatic regions reasonably well.

0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) NWC-4Y Pred. NWC-4Y (a) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) NWF-4Y Pred. NWF-4Y (b) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) WTFOM-4Y Pred. WTFOM-19Y Pred. WTFOM-4Y WTFOM-19Y (e) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) WTC-17Y Pred. WTC-17Y (f) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) LNM-10Y Pred. LNM-18Y Pred. LNM-10Y LNM-18Y (h) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) LTX-11Y Pred. LTX-11Y (c) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) LWI-8Y Pred. LWI-17Y Pred. LWI-8Y LWI-17Y (d) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) LSD-14Y Pred. LSD-14Y (g) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) ACTRL-8Y Pred. ACTRL-11Y Pred. ACTRL-8Y ACTRL-11Y (i) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) ASBS-8Y Pred. ASBS-8Y (j) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) ACRTB-8Y Pred. ACRTB-8Y (k) Figure 26. Predicted evolution of log |G*| with time and depth and measured field core data for the calibration sections: (a) NCAT Control, (b) NCAT Foam, (c) LTPP Texas, (d) LTPP Wisconsin, (e) WesTrack Fine (opt. %AC, medium %AV), (f) WesTrack Coarse, (g) LTPP South Dakota, (h) LTPP New Mexico, (i) ALF Control, (j) ALF SBS, and (k) ALF CRTB.

Refinement of the Pavement Aging Model (PAM) 49   0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 D ep th (c m ) (c) 0 1 2 3 4 5 0 5 10 15 20 25 30 Time (years) (a) 0 1 2 3 4 5 0 5 10 15 20 25 30 lo g |G *| at 6 4° C , 1 0 ra d/ s Time (years) (b) 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 D ep th (c m ) (d) lo g |G *| at 6 4° C , 1 0 ra d/ s log |G*| at 64°C, 10 rad/slog |G*| at 64°C, 10 rad/s ACTRL-8Y Pred. ACTRL-11Y Pred. ACTRL-8Y ACTRL-11Y Figure 27. Predicted evolution of log |G*| with time for multiple depths and the corresponding measured field core data for the ACTRL section: (a) prior to calibration and (b) after calibration, and the predicted evolution of log |G*| with depth for multiple time durations and the corresponding measured field core data for ACTRL section: (c) prior to calibration and (d) after calibration. 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 ATerp-8Y ATerp-11Y MWC-4Y NWE-4Y LOH-11Y LCA-8Y LCA-15Y LWA-11Y LWA-19Y MNS31-13Y LOE All Depths Yellow: ALF Blue: NCAT Green: LTPP Red: MIT Black: MnROAD R2= 0.7176 RMSE = 0.2515 Predicted & Calibrated log |G*| at 64°C, 10 rad/s (kPa) M ea su re d lo g |G *| af te r S m oo th in g at 6 4° C , 1 0 ra d/ s (k Pa ) Figure 28. Validation of smoothed field core measurements and field-calibrated PAM predictions for validation sections at all depths.

50 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) ATerp-8Y Pred. ATerp-11Y Pred. ATerp-8Y ATerp-11Y (a) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) MWC-4Y Pred. MWC-4Y (b) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) NWE-4Y Pred. NWE-4Y (c) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) LOH-11Y Pred. LOH-11Y (d) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) LCA-8Y Pred. LCA-15Y Pred. LCA-8Y LCA-15Y (e) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) LWA-11Y Pred. LWA-19Y Pred. LWA-11Y LWA-19Y (f) 0 1 2 3 4 5 6 0 1 2 3 4 5 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) MnROAD-13Y Pred. MnROAD-13Y (g) Figure 29. Predicted evolution of log |G*| with time and depth and measured field core data for the validation sections: (a) ALF Terpolymer, (b) MIT Control, (c) NCAT Evotherm, (d) LTPP Ohio, (e) LTPP California, (f) LTPP Washington, and (g) MnROAD.

Renement of the Pavement Aging Model (PAM) 51   0 1 2 3 4 5 0.4 0.6 0.8 1.0 1.2 1.4 1.6 D ep th (c m ) log |G*| / log |G*| (opt) (kPa) -0.70% AC (Medium AV) 0.70% AC (Medium AV) -0.7% AC (High AV) +0.7% AC (Low AV) Figure 30. Effect of deviations from the optimum asphalt content on smoothed log |G*| values of WesTrack ne eld cores that correspond to medium air void content. Field Calibration Adjustment Factor for Deviations from the Optimum Asphalt Content Because the results of the statistical analysis of the Brazil and WesTrack eld cores showed that asphalt content has a signicant eect on eld aging, the WesTrack sections that were used in the systematic study of mixture morphology were analyzed. e purpose was to calibrate an adjustment to the PAM predictions as a function of deviation from the Superpave optimum asphalt content. An adjustment for air void content was deemed not necessary because air void content was found to have an insignicant eect on eld aging of the WesTrack and Brazil eld cores. e smoothed eld core measurements of the WesTrack sections were used to establish the adjustment to the PAM predictions as a function of deviation from the Superpave optimum asphalt content. e relationship between the ratio of the log |G*| value that corresponds to the asphalt content of interest to the log |G*| value that corresponds to the optimum asphalt content [i.e., log |G*| (opt)] and asphalt content at each depth was rst investigated for all the WesTrack ne sections with asphalt contents that deviated from the optimum. Figure 30 pres- ents a summary graph of the results. Note that the WesTrack low and high asphalt contents correspond to −0.7% and +0.7% from the optimum asphalt content, respectively. Figure 30 reveals that the log |G*| ratios of the sections that correspond to asphalt contents that are lower than the Superpave optimum are greater than 1, and the log |G*| ratios of the sections that correspond to asphalt contents that exceed the Superpave optimum are generally less than 1. is outcome suggests that an increase in asphalt content decreases oxidative aging, which matches expectations because an increase in lm thickness increases the diusion path for oxygen. However, except for the WesTrack ne section that was prepared with a high asphalt content and low air void content [i.e., +0.7% ac (Low AV)], the log |G*| ratios are relatively close to 1, indicating only moderate sensitivity to asphalt content. Figure 30 reveals no clear trends in the log |G*| ratios with depth. erefore, the proposed adjustment to the PAM predictions to account for deviations from the Superpave optimum asphalt content is depth- independent. e corresponding adjustment is given in Equation (23).

52 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results ( )= × − ×log * log * 1 0.149 % (23), , ,G G ACt z adj t z dev where %ACdev = the asphalt content of interest minus the Superpave optimum asphalt content (%), and log |G*|t,z = the log |G*| value determined from Equation (22). The fitting parameter used in Equation (23) was determined using a least squares optimi- zation of the difference in log |G*|/log |G*|(opt) values determined from smoothed field core measurements and the log |G*|field,adj values/log |G*|field values predicted by Equation (22). Thus, the adjustment for asphalt content could be calibrated using field core measurements alone to isolate the effect of asphalt content without smearing in the errors in the PAM predictions. Figure 31, Figure 32, and Figure 33 show the effects of the adjustment for asphalt content on the PAM predictions for four WesTrack sections that correspond to the medium, high, and low air void contents, respectively. The results for the pavement sections that correspond to the optimum asphalt content at each air void content are shown for reference. For three of them—Figure 31 (b), Figure 32 (b), and Figure 33 (b)—the adjustment for asphalt content led to good agreement between the field core measurements and PAM predictions. As shown in Figure 31 (c), the PAM predictions without adjustment for asphalt content led to good predic- tions of log |G*| values of the WesTrack fine section with medium air void content and high asphalt content. Figure 33 (b) demonstrates that the field core measurements that correspond to the WesTrack fine section with low air void content and high asphalt content indicate a 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) log |G*| at 64°C, 10 rad/s WTFOM-19Y Predicted without Adj for %AC (a) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) WTFLM-19Y Predicted without Adj for %AC Predicted with Adj for %AC (b) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) WTFHM-19Y Predicted without Adj for %AC Predicted with Adj for %AC (c) log |G*| at 64°C, 10 rad/slog |G*| at 64°C, 10 rad/s Figure 31. Comparison between field core measurements and PAM predictions of log |G*| of WesTrack fine sections with medium air void content at (a) optimum asphalt content, (b) optimum asphalt content −0.7% (low), and (c) optimum asphalt content +0.7% (high).

Refinement of the Pavement Aging Model (PAM) 53   greater gradient with depth than all the other sections, which explains the reason that the PAM predictions do not align very well with the field core measurements. However, the adjustment for asphalt content generally improved the PAM predictions of the WesTrack fine section with high air void content and high asphalt content. It should be emphasized that limited data were used to generate the adjustment for asphalt content and, therefore, the effect of asphalt content merits additional consideration in future research. Comparisons Among Field Core Measurements, the Calibrated Pavement Aging Model Predictions, and GAS Model Predictions Pavement ME, which is a current pavement performance prediction software used by state agencies in the United States, incorporates the aging effect via the GAS model (Mirza and Witczak 1995, Applied Research Associates 2004). The accuracy of the PAM developed under this project should be evaluated against the accuracy of the GAS model since it reflects the current state-of-practice model. Table 8 presents a summary of the GAS model. The model predicts viscosity at a given long- term aging level at 0.6 cm below the pavement surface (ηaged) using the mix/lay-down viscosity (ηt = 0) and the mean annual air temperature (MAAT), which then can be used to predict the viscosity at any depth at that long-term aging level (ηt,z), as shown in Step 2 of Table 8. If the 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) WTFHL-19Y Predicted without Adj for %AC Predicted with Adj for %AC (b) log |G*| at 64°C, 10 rad/s 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) WTFOL-19Y Predicted without Adj for %AC (a) log |G*| at 64°C, 10 rad/s Figure 33. Comparison between field core measurements and PAM predictions of log |G*| of WesTrack fine sections with low air void content at (a) optimum asphalt content and (b) optimum asphalt content +0.7% (high). 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) WTFLH-19Y Predicted without Adj for %AC Predicted with Adj for %AC (b) log |G*| at 64°C, 10 rad/s 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) WTFOH-19Y Predicted without Adj for %AC (a) log |G*| at 64°C, 10 rad/s Figure 32. Comparison between field core measurements and PAM predictions of log |G*| of WesTrack fine sections with high air void content at (a) optimum asphalt content and (b) optimum asphalt content −0.7% (low).

54 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results original viscosity (ηorig) is available instead of ηt=0, a conversion can be done to obtain ηt=0, as shown in Step 1 of Table 8. In order to compare the measured data with the predictions obtained from both the PAM and the GAS model, a single AIP should be used. The measured dynamic shear modulus (|G*|) and phase angle values used in DSR testing can be converted to viscosity, as shown in Equation (24) (Applied Research Associates 2004). Therefore, the measured |G*| and phase angle values obtained from binder extracted and recovered from STA loose mixture can be converted to viscosity and used as inputs to the GAS model along with the MAAT for the location of the pave- ment section of interest. The GAS model was used to predict the viscosity at the same age level of the field cores that correspond to each pavement section. To convert the predicted viscosity back to |G*|, the measured phase angles from the field cores were used along with the predicted viscosity in Equation (24) to obtain |G*|. Note that the inputs for the GAS model are less cum- bersome to obtain than the inputs for the PAM. The former requires only the short-term aged viscosity value and MAAT, whereas the latter requires the short-term aged log |G*| value, the material-specific parameter M, and the pavement temperature history obtained from the EICM. M is obtained by measuring log |G*| at short-term aging and at multiple laboratory long-term aging conditions. Recall that any laboratory long-term aging duration can be considered in order to obtain log |G*| to calibrate M as long as it provides binder AIPs that are well dispersed on the oxidation timescale. 100 * 1 sin (24) 4.8628 Gη= × δ     where η = viscosity [centipoise (cP)], |G*| = dynamic shear modulus [pascal (Pa)], and δ = phase angle (°). Figure 34 shows the predictions obtained using the PAM and the GAS model plotted against the measured data points obtained from field cores at different depths. The PAM demonstrates Pr ed ic tio n of v is co si ty a t a ny a gi ng le ve l a nd d ep th u si ng th e G lo ba l A gi ng S ys te m (G A S) m od el 1. Viscosity Prediction at Short-Term Aging Level 0 0 1 0 1 log log( ) log log( ) 0.054405 0.004082 0.972035 0.010886 t origa a a code a code = mix/lay-down viscosity, cP, = original viscosity, cP, and = hardening ratio (0 for average). 2. Viscosity Prediction at Long-Term Aging Level and with Depth 0log log( )log log( ) 1 t aged At Bt 0 , (4 ) ( )(1 4 ) 4(1 ) t t t z E E z Ez = aged viscosity, cP, = parameters as a function of , MAAT, and test temperature, MAAT = mean annual air temperature, °F, = time, months, = depth, inches, = aged viscosity at time t and depth z, MPoise, and = aged surface viscosity, MPoise. Table 8. Summary of GAS model predictive equations.

Refinement of the Pavement Aging Model (PAM) 55   greater predictive accuracy than the GAS model, as illustrated by R2 = 0.7061 compared to R2 = 0.3792. R2 was calculated excluding M50R because it is a clear outlier due to its high initial log |G*|. Figure 35 through Figure 39 present comparisons between the field core measurements versus the field-calibrated PAM predictions and GAS model predictions of log |G*| as a function of depth for selected pavement sections. Recall that the GAS model predicts viscosity; thus, to 0 1 2 3 4 0 1 2 3 4 0 0.5 1 1.5 2 2.5 3 3.5 4 M ea su re d lo g |G *| at 6 4° C, 10 ra d/ s (k Pa ) Predicted log |G*| at 64°C, 10 rad/s (kPa) All Depths R2 = 0.7061 RMSE = 0.3032 (a) M50R 0 0.5 1 1.5 2 2.5 3 3.5 4 M ea su re d lo g |G *| at 6 4° C, 10 ra d/ s (k Pa ) Predicted log |G*| at 64°C, 10 rad/s (kPa) All Depths R2 = 0.3792 RMSE = 0.4406 (b) M50R Figure 34. Comparison between field core measurements and predictions of (a) PAM and (b) GAS model at all depths. 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) log |G*| at 64°C, 10 rad/s (kPa) ACTRL-8Y Pavement Aging Model GAS(a) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) ASBS-8Y Pavement Aging Model GAS (b) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) ACRTB-8Y Pavement Aging Model GAS (c) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) ATerp-8Y Pavement Aging Model GAS(d) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) Figure 35. Comparison between field core measurements versus field-calibrated PAM predictions and GAS model predictions of log |G*| for (a) ALF CTRL, (b) ALF SBS, (c) ALF CRTB, and (d) ALF Terpolymer.

56 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) M15R-4Y Pavement Aging Model GAS (a) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) M50R-4Y Pavement Aging Model GAS (b) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) MWA-4Y Pavement Aging Model GAS (c) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) MWE-4Y Pavement Aging Model GAS (d) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) Figure 36. Comparison between field core measurements and field-calibrated PAM predictions and GAS model predictions of log |G*| for (a) MIT 15% RAP, (b) MIT 50% RAP, (c) MIT Advera, and (d) MIT Evotherm. 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) N50R-4Y Pavement Aging Model GAS (a) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) NWF-4Y Pavement Aging Model GAS (b) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) NWE-4Y Pavement Aging Model GAS (c) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) N50RF-4Y Pavement Aging Model GAS (d) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa) Figure 37. Comparison between field core measurements and field-calibrated PAM predictions and GAS model predictions of log |G*| for (a) NCAT 50% RAP, (b) NCAT Foam, (c) NCAT Evotherm, and (d) NCAT 50% RAP with Foam.

Refinement of the Pavement Aging Model (PAM) 57   0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) LSD-14Y Pavement Aging Model GAS (a) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) LNM-10Y Pavement Aging Model GAS (b) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) LWI-8Y Pavement Aging Model GAS (c) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) LTX-11Y Pavement Aging Model GAS (d) log |G*| at 64°C, 10 rad/s (kPa)log |G*| at 64°C, 10 rad/s (kPa) log |G*| at 64°C, 10 rad/s (kPa)log |G*| at 64°C, 10 rad/s (kPa) Figure 38. Comparison between field core measurements and field- calibrated PAM predictions and GAS model predictions of log |G*| for (a) LTPP South Dakota, (b) LTPP New Mexico, (c) LTPP Wisconsin, and (d) LTPP Texas. 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) WTFHL-19Y WTFHM-19Y Pavement Aging Model GAS (c) log |G*| at 64°C, 10 rad/s (kPa) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) WTFLM-19Y WTFLH-19Y Pavement Aging Model GAS (b) log |G*| at 64°C, 10 rad/s (kPa) 0 1 2 3 4 5 6 0 1 2 3 4 D ep th (c m ) WTFOL-19Y WTFOM-19Y WTFOH-19Y Pavement Aging Model GAS (a) log |G*| at 64°C, 10 rad/s (kPa) Figure 39. Comparison between field core measurements and field- calibrated PAM predictions and GAS model predictions of log |G*| for (a) WesTrack Fine (optimum %AC; low, medium & high %AV), (b) WesTrack Fine (low %AC, medium & high %AV), and (c) WesTrack Fine (high %AC, low & medium %AV).

58 Long-Term Aging of Asphalt Mixtures for Performance Testing and Prediction: Phase III Results convert viscosity to |G*|, the phase angle measured from field cores was utilized along with Equa- tion (24). The use of the measured phase angle from field cores led to the occasional increasing trend of |G*| with depth for the GAS model predictions. For most of the sections evaluated, the PAM predictions agree reasonably well with the field core measurements, except for some cases (e.g., MIT 15% RAP, MIT 50% RAP, NCAT 50% RAP, NCAT 50% RAP with Foam, and LTPP Wisconsin). The PAM generally predicts higher log | G*| values for depths closer to the pavement surface and in some cases, lower log |G*| values deeper in the pavement compared to the GAS model. Note that some of the sections were used to develop the depth-dependent calibration for the PAM, which thus constitutes, in some sense, a circle in the analysis. ATerp, M15R, M50R, MWA, MWE, N50R, N50RF, and NWE, however, were not used in the depth- dependent calibration of the PAM. All of these sections were not involved in the development or calibration of the GAS model, thereby constituting a pure validation of the model. The trend observed in the GAS predictions shown in Figure 35 through Figure 39 is attributed to the use of measured phase angle values obtained from tested field cores at different depths. The authors acknowledge that the GAS model predictions shown in this work could be confounded by the two-time use of Equation (24) and any ensuing uncertainties that could emerge. Future work to refine the prediction of the PAM for RAP and WMA sections should include a wider variety of sections for both categories. Summary The PAM equations and definitions are summarized in Table 9. The equations are shown here for the convenience of the reader. The kinetics model shown in Equations (1) through (3) was refined as a function of pavement depth for a wide range of pavement sections, including sections that contain both conventional HMA and other materials (i.e., RAP, WMA, and PMA). The depth-dependent field calibration of the kinetics model was carried out using all sections with asphalt mixtures prepared at the optimum asphalt content. Sections used in the systematic study of mixture morphology were analyzed to calibrate an adjustment to the PAM predictions as a function of deviation from the Superpave optimum asphalt content. An adjustment for air void content was not established because air void content was found to have an insignificant effect on field aging. The PMA results generally align with the rest of the data, suggesting that a separate field cali- bration of the PAM for PMA materials is unnecessary. Some of the RAP and WMA data were clear outliers and were excluded from the field cali- bration of the PAM. The outlying behavior is attributed to the laboratory short-term aging procedure that overestimates field aging in cold climates. Other RAP sections and WMA sections seemed to follow the same trend as the conventional sections, which indicates that a separate field calibration of the kinetics model for RAP and WMA sections was not required. For most of the sections evaluated, the PAM predictions agree reasonably well with the field core measurements, except for some of the field sections that contained RAP and/or WMA. The PAM predictions generally were found to outmatch the GAS model predictions, with some exceptions.

Refinement of the Pavement Aging Model (PAM) 59   Description Equation(s) Eqn(s)Ref. Definition of Terms Kinetics Model 0log | * | log | * | 11kinetics cf k tc f G G M e k t k k exp aff f E k A RT exp acc c E k A RT (1) (2) (3) |G*|kinetics = long-term aged binder shear modulus at 64°C, 10 rad/s (kPa) |G*|0 = short-term aged binder shear modulus at 64°C, 10 rad/s (kPa), kf = rate of fast reaction, kc = rate of constant reaction, Af = regression parameter equal to 1.25×103, Ac = regression parameter equal to 3.68×107, Eaf = regression parameter equal to 95.04, Eac = regression parameter equal to 62.21, R = universal gas constant or ideal gas constant equal to 0.008314 (kJ/mol K), T = pavement temperature (Kelvin), t = reaction time (days), M = fitting parameter related to fast reaction reactive material Field Calibration with Respect to Depth max , 0 0.916 log | * | log | * | 1 1 tN AP t z z AP G G M e z , 0log | * | log | * |kinetics t t G G AP M 0 max 4.5 log | * |G AP M 100.2260.477 tAPN e (22) |G*| t,z = long-term aged binder shear modulus after depth- dependent calibration at 64°C and 10 rad/s at time t (kPa), |G*| kinetics,t = long-term aged binder shear modulus calculated at the pavement surface at time t (kPa), APt = Aging Parameter at the pavement surface at time t (kPa), APt=10 = Aging Parameter at the pavement surface calculated at 10 years of aging (kPa), APmax = Maximum Aging Parameter at the pavement surface (kPa), N = material-dependent parameter, and z = pavement depth (cm). Field Calibration Adjustment Factor for Deviation from the Optimum Asphalt Content , , ,log | * | log | * | 1 0.149 %t z adj t z devG G AC (23)      |G*| t, z, adj = long-term aged binder shear modulus after deviation from optimum asphalt content calibration at 64°C and 10 rad/s (kPa), and %ACdev = asphalt content of interest minus the Superpave optimum asphalt content (%). Table 9. Equation summary of the PAM.