Performance-Based Mix Design for Porous Friction Courses (2018)

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.

50 Introduction Three aggregate types (limestone, traprock, and granite) were evaluated in Part 1. For each aggregate, one good mix design and one poor mix design were selected per agency recommenda- tions. Table 16 shows the properties of the original mix designs recommended by the Florida Department of Transportation (FDOT), New Jersey Department of Transportation (NJDOT), Virginia Department of Transportation (VDOT), Georgia Department of Transportation (GDOT), and South Carolina Department of Transportation (SCDOT), with a few exceptions mentioned below. The following changes were made to these mix designs for evaluation in this study: • It was decided to produce the laboratory mixes with the same asphalt binder so that binder grade could be eliminated as a variable in the test results. For that reason, PG 76-22 with SBS modifier was used in all mixes with one exception. Florida had a mix with SBS and one with GTR modifier. It was decided to keep the modifiers the same as what was used on the project for the FDOT mixes. Since the same aggregate source was used, the test results would provide a good comparison of performance with both modifiers. • The mixtures recommended by NJDOT, GDOT, and SCDOT contained hydrated lime. This was replaced in terms of filler with baghouse fines (BHF), and in terms of an anti-stripping agent with liquid anti-strip. Liquid anti-strip was used in this study at a dosage rate of 0.5% by weight of binder for all designs so that the type of anti-strip agent used would not be a factor in the test results. • The Georgia mixture (designed in 1995) used an AC20 base binder modified with SBS to produce PG 76-22. The Virginia mixture used a PG 82-22 RM and New Jersey used PG 76-22 modified with GTR. PG 76-22 asphalt binder with SBS modifier was used for these designs. Testing for gradations, specific gravity, and dry-rodded unit weight was conducted on the aggregates. Some of the stockpile gradations did not match the JMF gradations reported. This was to be expected since some of the designs were over 20 years old. The stockpile percentages were adjusted to match the original JMF blend gradation as closely as possible. Each adjusted grada- tion was verified to be within the agency gradation limit, and the differences from the original JMF blend gradation were within the agency’s mix production tolerance. A comparison of the original JMF and adjusted blends is shown in Table 17 for good mixture blends and Table 18 for poor mixture blends. The NCAT blends were used in this study. There were two differences between the Florida original mix designs. The Florida good design used a GTR modified binder and had 5% aggregate screenings in the design, while the poor design used an SBS-modified binder and had no screenings. The coarse aggregate sampled for the Florida designs had less than 3% passing the No. 8 sieve. A small amount of screenings were C H A P T E R 5 Part 1: Evaluation of Mix Designs

Part 1: Evaluation of Mix Designs 51 Mix Design Component Florida Florida New Jersey Virginia Georgia South Carolina Mixture Designation Good (cracking) Poor (cracking) Good (raveling) Poor (raveling) Good (raveling) Poor (raveling) Aggregate Mineralogy Limestone Traprock Granite Asphalt Type PG 67-22 PG 76-22 PG 76-22 PG 82-22 AC-20 PG 76-22 Binder Modifier 12% #30 GTR SBS GTR GTR SBS SBS Anti-strip 0.5% by weight of binder Fiber, % 0.4 0.4 0.3 0.3 0.4 0.3 AC, % 7.1 6.0 6.0 5.8 6.0 6.0 Total P-200, % 1.1 1.0 3.0 2.5 1.5 2.2 Table 16. Original mix design components. Percent Passing Sieve Florida “Good” Florida Limits Georgia “Good” Georgia Limits New Jersey “Good” New Jersey Limits JMF NCAT JMF NCAT JMF NCAT 25.0 mm, 1" 100 100 100 100 19.0 mm, 3/4" 100 100 100 100 100 100 12.5 mm, 1/2" 90 88 85 - 100 92 96 85 - 100 100 100 100 9.5 mm, 3/8" 66 68 55 - 75 66 66 55 – 75 92 91 80 – 100 4.75 mm, #4 24 26 15 - 25 25 21 15 – 25 34 36 30 – 50 2.36 mm, #8 10 8 5 - 10 8 8 5 – 10 13 11 5 – 15 1.18 mm, #16 6 6 5 6 8 8 5 – 10 0.600 mm, #30 4 5 4 5 6 6 0.300 mm, #50 3 3 3 4 5 5 0.150 mm, #100 2 2 2 3 4 5 0.075 mm, #200 1.1 0.9 2 - 4 1.5 2.0 2 - 4 3.0 3.9 2 - 5 AC, % 7.1 7.1 6.0 6.0 5.75-7.25 6.0 6.0 Table 17. “Good” mixture gradations (% passing) and ACs. Percent Passing Sieve Florida “Poor” Florida Limits Virginia “Poor” Virginia Limits South Carolina “Poor” South Carolina Limits JMF NCAT JMF NCAT JMF NCAT 25.0 mm, 1" 100 100 100 100 100 100 19.0 mm, 3/4" 100 100 100 100 100 100 100 100 100 12.5 mm, 1/2" 93 88 85 - 100 100 100 100 95 95 89 - 100 9.5 mm, 3/8" 69 68 55 - 75 86 87 85 - 100 70 74 63 - 75 4.75 mm, #4 23 26 15 - 25 21 25 20 - 40 21 21 15 - 25 2.36 mm, #8 9 8 5 - 10 9 8 5 - 10 8 8 5 - 10 1.18 mm, #16 5 6 6 5 0.600 mm, #30 4 5 5 5 3 0.300 mm, #50 3 3 4 3 0.150 mm, #100 3 2 3 5 2 0.075 mm, #200 1.0 0.9 2 - 4 2.5 2.6 2 - 4 2.2 1.7 0 - 4 AC, % 6.0 6.3 5.8 5.8 5.59 – 6.01 6.0 6.0 5.5 – 7.0 Table 18. “Poor” mixture gradations and ACs.

52 Performance-Based Mix Design of Porous Friction Courses added to the poor design in order to match the gradation of the original poor blend; therefore, the only real difference in the adjusted blends was the difference in binder type. Instead of using the 6.0% AC stated on the JMF for the Florida poor mix, it was decided to use the same AC as the good design (7.1%) and to subtract out the amount of GTR (12%). This changed the optimum binder content to 6.34%. With these changes, the mixtures were essentially the same but for type of asphalt binder modifier; but the additional asphalt may explain why the Florida poor mix ranked higher than the good mix in some test results. The mix modified with SBS ranked higher than the GTR mix in 10 of 14 tests conducted in this study. Specific gravity values were not provided on the JMF for some of the stockpiles. The specific gravity testing conducted for this project was primarily for the calculation of VMA, VCA, and film thickness. The aggregate blend bulk specific gravity (Gsb) values used for this study can be found in Table 19. Results and Discussion Cantabro Testing Cantabro testing was performed on both conditioned and unconditioned specimens to deter- mine if there was any statistical difference in the results based on the conditioning process. This was a preliminary step to determine if the specimens should be conditioned for the remainder of the experiment. The conditioned specimens were conditioned using the same vacuum saturation and freeze–thaw method as for TSR conditioning (AASHTO T 283). After conditioning, the specimens were air-dried for several days and then placed in a Core- Dry device to remove any trapped moisture. Specimens were produced at the optimum binder content and also at ± 1.0% of optimum. The conditioned versus unconditioned comparison was only conducted at the optimum AC. A minimum of three specimens were tested for each design point. A two-sample t-test was performed on each of the mix designs to determine if the conditioning had any significant effect on the Cantabro loss. The p-value for each of these comparisons (Table 20) showed that, with the exception of the Florida poor mixture, there was no statistical difference between the conditioned and unconditioned specimens. The average results for the conditioned and unconditioned specimens (Figure 37) showed that the conditioned specimens had less loss in most cases than unconditioned specimens. This is most likely due to the steric hardening that the binder incurred from the hot water bath. Since there was almost no statistical difference between the conditioned and unconditioned samples, all following Cantabro testing was performed on unconditioned specimens. A summary of the average results from all of the unconditioned Cantabro testing can be seen in Table 21, and a graphical depiction of the data can be seen in Figure 38. When considering only the design ACs, the well performing mixes from New Jersey and Georgia mixtures were the Mix Design Aggregate Mineralogy Blend Bulk Specific Gravity (Gsb) Georgia Granite 2.625 South Carolina Granite 2.615 Florida Limestone 2.410 Florida Limestone 2.410 Virginia Traprock 2.943 New Jersey Traprock 2.936 Table 19. Bulk specific gravity of aggregate blends.

Part 1: Evaluation of Mix Designs 53 Mix Design Designation p-Value Difference Georgia Good 0.426 Insignificant South Carolina Poor 0.818 Insignificant Florida Good 0.756 Insignificant Florida Poor 0.019 Significant Virginia Poor 0.126 Insignificant New Jersey Good 0.480 Insignificant Table 20. Cantabro conditioned vs. unconditioned t-test results (` = 0.05). 24.0 21.9 35.1 37.9 19.3 9.5 15.7 20.9 28.4 39.6 15.5 7.9 0 10 20 30 40 50 60 FL Poor - SBS FL Good - GTR VA Poor - SBS SC Poor - SBS GA Good - SBS NJ Good - SBS Av er ag e Ca nt ab ro L os s ( % ) No F/T 1 F/T Figure 37. Part 1: Conditioned versus unconditioned Cantabro loss results. Mix ID AC Content (%) Average Air Voids (%) Cantabro Loss (%) Average St. Dev. COV (%) FL Poor - SBS 5.3 19.4 35.4 6.3 17.8 FL Poor - SBS 6.3 17.7 24.0 3.8 16.0 FL Poor - SBS 7.3 16.0 15.2 2.6 16.7 FL Good - GTR 6.1 19.6 38.8 1.9 4.9 FL Good - GTR 7.1 17.1 21.9 2.5 11.2 FL Good - GTR 8.1 15.3 16.4 2.9 17.4 VA Poor - SBS 4.8 23.5 46.9 1.9 4.1 VA Poor - SBS 5.8 21.8 35.1 4.5 12.8 VA Poor - SBS 6.8 18.9 21.6 1.5 7.2 SC Poor - SBS 5.0 23.6 57.3 4.7 8.2 SC Poor - SBS 6.0 22.2 37.9 11.9 31.3 SC Poor - SBS 7.0 20.6 26.8 7.1 26.5 GA Good - SBS 5.0 17.5 25.4 1.9 7.6 GA Good - SBS 6.0 15.7 19.3 6.4 33.4 GA Good - SBS 7.0 12.5 12.8 3.2 24.7 NJ Good - SBS 5.0 21.9 19.7 3.9 20.0 NJ Good - SBS 6.0 19.0 9.5 2.6 27.6 NJ Good - SBS 7.0 17.2 4.3 1.2 27.2 Table 21. Summary of unconditioned Cantabro results.

54 Performance-Based Mix Design of Porous Friction Courses only specimens to pass the ASTM recommended 20% maximum loss criterion. If the AASHTO criterion (15%) is applied, only the New Jersey design passes. A relationship of specimen air voids to Cantabro loss was observed when considering all of the Cantabro data (Figure 39A). An exponential trend line was fitted to the data and a goodness of fit (R2) value of 0.32 was observed for all of the data; however, when the data are separated by NMAS, the trend line provides a much better fit. These results are represented in Figure 39 parts B, C, and D. 25.4 57.3 35.4 38.8 46.9 19.7 19.3 37.9 23.8 21.9 35.1 10.2 12.8 26.8 15.2 16.4 21.6 4.3 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 Georgia - Good South Carolina - Poor Florida - Poor Florida - Good Virginia - Poor New Jersey - Good Av er ag e Ca nt ab ro L os s ( % ) Optimum AC% -1.0% Optimum AC% Optimum AC% +1.0% Figure 38. Unconditioned Cantabro results for Part 1. 19.0 mm NMAS 0 10 20 30 40 50 60 70 10 15 20 25 Ca nt ab ro L os s ( % ) Specimen Air Voids (%) y = 0.6417e0.2023x R² = 0.65 All NMAS Specimens 0 10 20 30 40 50 60 70 10 15 20 25 Ca nt ab ro L os s ( % ) Specimen Air Voids (%) y = 2.278e0.1181x R² = 0.32 A) B) 12.5 mm NMAS 0 10 20 30 40 50 60 70 10 15 20 25 Ca nt ab ro L os s ( % ) Specimen Air Voids (%) y = 2.6237e0.1227x R² = 0.86 C) 9.5 mm NMAS 0 10 20 30 40 50 60 70 10 15 20 25 Ca nt ab ro L os s ( % ) Specimen Air Voids (%) D) y = 0.0325e0.2868x R² = 0.76 Figure 39. Air voids versus Cantabro loss relationship.

Part 1: Evaluation of Mix Designs 55 An analysis of variance (ANOVA) analysis (α = 0.05) was conducted using a Tukey-Kramer grouping to determine whether statistical differences were observed between the mixture designs (Table 22). This was performed using statistical software. The analysis shows results for Cantabro loss at the optimum AC for each mixture. The Georgia and South Carolina designs (granite) were not grouped together, which shows that they were statistically different designs in regards to percent Cantabro loss. The New Jersey and Virginia designs (traprock) were also not grouped together showing they were statistically different as well. This is an important point to note since these mixtures share aggregate mineralogies. Volumetric Results Air Voids The volumetric properties calculated for this study were based on the Cantabro specimens, since Cantabro specimens were tested at multiple ACs. Each mixture had extra specimens fabricated at the optimum AC for permeability and conditioned Cantabro testing. Three specimens were fabricated for the other ACs. A summary of the volumetric properties can be found in Table 23. The CoreLok air voids were calculated for every specimen fabricated for this study. The aver- age value for each mixture and its variations were summarized. Air voids from Part 1 of the study can be found in a graphical depiction in Figure 40. All of the mix designs exceeded the anticipated minimum requirement of 15.0% air voids. However, both the ASTM and AASHTO specifications recommend a minimum air void content of 18.0%. The AASHTO specification also has a maximum limit of 22.0%. If these limits are used for design criteria, only the Virginia and New Jersey designs pass. An 18.0% design air void target may be more beneficial if it is determined that air void content has a direct correlation to mixture performance. Film Thickness The original expectations for some of these properties appeared to have minimal effect on the performance of the mixtures. The anticipated minimum film thickness requirement of 24 microns did not correspond to the mixture’s field performance. In Figure 41 it can be seen that the New Jersey good mix has a film thickness of 18.6 microns, while the South Carolina poor mix has a film thickness of 34.7 microns. This may indicate that film thickness is not as critical as first thought when designing a well performing PFC. VCA Analysis Determining the VCA of the mix and aggregate can help to determine if the design has stone-on-stone contact. The ratio of VCAMIX to VCADRC (VCAMIX/VCADRC) should be less than Mix ID Cantabro Loss, % N Mean Grouping South Carolina - Poor 3 37.9 A Virginia - Poor 3 35.1 A B Florida - Poor 3 23.8 A B C Florida - Good 3 21.9 A B C Georgia - Good 3 19.3 B C New Jersey - Good 3 10.2 C Table 22. ANOVA—unconditioned Cantabro loss at optimum AC.

56 Performance-Based Mix Design of Porous Friction Courses Mix ID Total AC (%) Avg. Va (%) Avg. VMA Avg. VCAMIX/VCADRC Avg. Film Thickness (microns) Florida - Poor 5.3 19.4 26.0 1.07 25.5 Florida - Poor 6.3 17.7 26.2 1.07 32.5 Florida - Poor 7.3 16.0 26.6 1.08 41.0 Florida - Good 6.1 19.6 26.8 1.08 28.3 Florida - Good 7.1 17.1 26.4 1.07 35.9 Florida - Good 8.1 15.3 26.6 1.08 43.7 Virginia - Poor 4.8 23.5 32.4 1.17 20.5 Virginia - Poor 5.8 21.8 32.9 1.18 25.4 Virginia - Poor 6.8 18.9 32.2 1.17 30.5 South Carolina - Poor 5.0 23.6 31.9 1.08 28.0 South Carolina - Poor 6.0 22.2 32.3 1.09 34.7 South Carolina - Poor 7.0 20.6 32.6 1.09 41.2 Georgia - Good 5.0 17.5 26.3 1.00 21.9 Georgia - Good 6.0 15.4 26.3 1.00 27.1 Georgia - Good 7.0 12.5 25.7 0.99 32.4 New Jersey - Good 5.0 21.9 31.3 1.28 15.1 New Jersey - Good 6.0 19.5 31.1 1.27 18.6 New Jersey - Good 7.0 17.2 31.0 1.27 22.2 Note: Avg. = average. Table 23. Mixture properties for Part 1. 19.4 19.6 23.5 23.6 17.5 21.9 17.7 17.1 21.8 22.2 15.4 19.5 16.0 15.3 18.9 20.6 12.5 17.2 0.0 5.0 10.0 15.0 20.0 25.0 Florida - Poor Florida - Good Virginia - Poor South Carolina - Poor Georgia - Good New Jersey - Good A ve ra ge C or eL ok A ir V oi d (% ) Optimum AC% - Minus 1.0% Optimum AC% Optimum AC% - Plus 1.0% Figure 40. Part 1 air void content using the CoreLok method. or equal to 1.00 in order for stone-on-stone contact of the coarse aggregate particles to occur. Most agencies do not require this limit for design although it is specified in both the AASHTO and ASTM OGFC design procedures. It was anticipated that this would be a critical design factor for PFC designs since this should be indicative of the mixture’s strength. Of the designs tested for this study, only New Jersey and Virginia require the calculation of VCA for design purposes. Figure 42 shows that the mixtures evaluated for this study all had values of 1.00 or greater. If the criteria is VCAMIX ≤ VCADRC, then the Georgia mix is the only design to pass the VCA

Part 1: Evaluation of Mix Designs 57 requirement. Since New Jersey and Virginia require the VCA calculation to be incorporated in their designs, the failing VCA ratios seemed to be an error. Further investigation and research was performed to determine if this data were in error or if the mix designs did in fact all have ratios of 1.00 or greater. The original concept for stone-on-stone contact to create an aggregate skeleton was defined in research conducted for NCHRP Project 09-08, which was research on stone matrix asphalt (SMA). This research was published in NCHRP Report 425 and this report recommended the VCA concept for use in SMA design. In this report the coarse aggregate in the VCADRC calculations is defined as the total aggregate blend material retained on the #4 sieve for 12.5 mm, 19 mm, and 25 mm NMAS mixtures. For the 9.5 mm NMAS mixtures, the coarse aggregate is defined as the aggregate blend material retained on the 2.36 mm (No. 8) sieve (Brown and Cooley, 1999). The original VCADRC calculation for this PFC study was based on ASTM D7064 and AASHTO PP 77. ASTM defines the coarse aggregate as material retained on the 4.75 mmm (No. 4) sieve; AASHTO does not define the coarse aggregate. In order to determine what effect changing the definition of coarse aggregate would have on the mixture’s VCADRC, the procedure recom- mended in NCHRP Report 425 was conducted on the New Jersey mixture. Since the New Jersey 26 28 21 28 22 15 33 36 25 35 27 19 41.0 44 30.5 41.2 32.4 22.2 0 5 10 15 20 25 30 35 40 45 50 Florida - Poor Florida - Good Virginia - Poor South Carolina - Poor Georgia - Good New Jersey - Good Av er ag e Fi lm T hi ck ne ss (m ic ro ns ) Optimum AC% - Minus 1.0% Optimum AC% Optimum AC% - Plus 1.0% Figure 41. Part 1 film thickness results. 1.07 1.07 1.18 1.09 1.00 1.27 0.95 1.00 1.05 1.10 1.15 1.20 1.25 1.30 Florida - Poor Florida - Good Virginia - Poor South Carolina - Poor Georgia - Good New Jersey - Good A ve ra ge V C A m ix /V C A dr c Mixture at Optimum Asphalt Content (%) No Stone on Stone Contact Stone on Stone Contact Figure 42. Part 1 VCA ratio using the No. 4 sieve to define coarse aggregate.

58 Performance-Based Mix Design of Porous Friction Courses design is a 9.5 mm NMAS mix, this seemed appropriate. The results provide a decrease in the VCA ratio to a value of 0.92, which passes the recommended criterion. Upon observing the significant affect changing the definition of coarse aggregate had on the VCA results, further investigation was conducted on the VCA test procedure. Watson et al. (2004b) conducted research on the VCA technique to ensure that method was suitable for use in OGFC pavements. Digital imaging techniques were used to determine if stone-on-stone contact was occurring in several mix designs. The report concluded that the use of VCA was valid; however, there were instances where the VCA ratio (using the No. 4 sieve to define coarse aggregate) was greater than 1.00 but X-ray images showed stone-on-stone contact existed. For these mixtures, redefining the coarse aggregate by using the No. 8 sieve, in some cases, gave a passing VCA ratio. The recommendation was to determine the critical breakpoint sieve and use that sieve to define the coarse aggregate. The critical breakpoint sieve is defined as the finest sieve size for which at least 10% of the total aggregate is retained (Watson et al., 2004b). In NCHRP Report 640, Cooley et al. also defines the coarse aggregate by using the breakpoint sieve. If the breakpoint sieve method is used for this study, all of the designs would define coarse aggregate as material retained on the No. 8 sieve. In order to ensure a comprehensive investigation into the VCA ratio, this method was also performed. These results can be seen in Figure 43. Using the breakpoint sieve method, all of the designs passed the VCA criterion. The breakpoint sieve for all designs was the No. 8 sieve. It was anticipated that these designs would pass due to the fact that the percent coarse aggregate ranged from 83.1 (New Jersey) to 86.5 (South Carolina). VMA Results The voids in mineral aggregate (VMA) of each mixture changed very little for the varying ACs. Typical VMA data over a range of ACs has a vertical parabolic shape. As can be seen in Figure 44, the “curves” are basically nonexistent. As shown previously in Equation 2, the Gmb and the percent stone in the mixture are the changing factors. PFC specimens show relative little change in Gmb with varying ACs. The reason that differences in air void content are observed is due pri- marily to the change in Gmm. This change in Gmb is sometimes so small that there is no noticeable difference between ACs. In Figure 44 and Table 23, the Virginia mix is shown to have a higher VMA at the optimum AC than it does at the other ACs. This is due to the average Gmb of the optimum AC (5.8%) showing little change from the lower AC (4.8%). This variability is not 0.81 0.82 0.93 0.93 0.82 0.92 0.50 0.60 0.70 0.80 0.90 1.00 1.10 1.20 Florida - Poor Florida - Good Virginia - Poor South Carolina - Poor Georgia - Good New Jersey - Good A ve ra ge V C A m ix /V C A dr c U si ng B re ak po in t Si ev e Mixture at Optimum Asphalt Content (%) No Stone on Stone Contact Stone on Stone Contact Figure 43. Part 1 VCA ratio using breakpoint sieve (No. 8) to define coarse aggregate.

Part 1: Evaluation of Mix Designs 59 unexpected when fabricating PFC specimens. A change in VMA of less than 1.0% for a design shows no relative change to the mixture. This seems to indicate that the VMA of PFC mixes is not sensitive to change in AC. Permeability The permeability testing showed a direct correlation between the permeability and air void content of the specimens (Figure 45). A summary of the data (Table 24) shows that two of the poor mixtures have higher air voids and corresponding higher permeability values. The New Jersey design has good performance, a high air void content, and a high permeability rate. This may be due to the 9.5 mm NMAS of the mixture. The design requirements in the ASTM and AASHTO specifications state that the mixtures must have a minimum air void content of 18.0%. In Figure 45, the optimum air void content associated with the recommended permeability rate of 100 meters/day is close to 17.0%. Two of the good designs (Georgia and Florida GTR) had a permeability rate lower than the recommended 100 meters/day criterion. Since these mixtures had such good performance in the field, the recommended permeability rate needs to be reevaluated. With Mississippi specifying as low as 35 meters/day, a compromise of 50–60 meters/day may be a better criterion. The lowest average rate recorded for this part of the study was 77 meters/day. However, both the Georgia and Florida well performing mixes had individual permeability values as low as 69 m/day. Based on this information, and a minimum recommended air voids of 15%, a minimum permeability value of 50 m/day is recommended based on Figure 45. 25 26 27 28 29 30 31 32 33 34 4.00 5.00 6.00 7.00 8.00 9.00 A ve ra ge V M A Asphalt Content (%) Florida - Poor Florida - Good Virginia - Poor South Carolina - Poor Georgia - Good New Jersey - Good Figure 44. VMA curves for Part 1 design data. y = 24.87x - 324.88 R² = 0.9384 0 50 100 150 200 250 14 16 18 20 22 24 P er m ea bi lit y (k ) m et er /d ay Air Voids (%) Figure 45. Part 1 permeability to air void correlation.

60 Performance-Based Mix Design of Porous Friction Courses Performance Testing Draindown The draindown testing was performed on all of the JMF designs using a 2.36 mm (#8) mesh basket. All of the designs used a final grade PG 76-22 binder. The mixing temperature range for this binder was 320°F–330°F. In order to assure that draindown was not going to occur, the higher end of the range (330°F) was used for the mixing temperature (lower test temperature). The amount of fiber content varied between 0.3% and 0.4% for this testing based on the mix design provided by the agency. The results in (Table 25) show that no draindown occurred for any of the designs. Hamburg Wheel-Tracking Test Results The HWTT data for this testing proved to be difficult to analyze. The mixtures had a wide range of performance and made it impractical to perform statistical analysis on the data. The simplest and most comprehensive way to explain the performance of these specimens was through a graphical depiction (Figure 46). The figure shows that most of the mixtures performed reasonably well, with the exception of South Carolina’s granite mix design. There were three sets of data (six specimens fabricated) for each mix design. The rut depth of the specimens was recorded every 200 passes and three sets of data were aver- aged to form the graphs depicted in Figure 46. Table 26 shows the variability in the three sets of data as well as the pertinent mix design information. It should be noted that while the “Greatest Rut Depth Recorded” is being reported, this is a slight misnomer. The HWTT machine records rut data until the LVDT reaches its maximum limit. MIX ID TOTAL AC (%) FIBER (%) AVERAGE AIR VOIDS (%) PERMEABILITY (K) METERS/DAY Average St. Dev. COV (%) Georgia - Good 6.0 0.4 15.7 80 10.5 13.1 Florida - Good 7.1 0.4 17.1 77 13.3 17.2 New Jersey - Good 6.0 0.3 20.3 186 37.0 19.9 South Carolina - Poor 6.0 0.3 22.2 209 17.1 8.2 Florida - Poor 6.3 0.4 17.3 107 13.4 12.4 Virginia - Poor 5.8 0.3 21.9 237 8.2 3.5 Table 24. Part 1 permeability data summary. MIX ID TOTAL AC (%) FIBER (%) DRAINDOWN (%) Test Temp, °F 330 357 FL Poor - SBS 6.3 0.4 0.0 0.0 FL Good - GTR 7.1 0.4 0.0 0.0 VA Poor - SBS 5.8 0.3 0.0 0.0 SC Poor - SBS 6.0 0.3 0.0 0.0 GA Good - SBS 6.0 0.4 0.0 0.0 NJ Good - SBS 6.0 0.3 0.0 0.0 Table 25. Part 1 Draindown results using a 2.36 mm (#8) mesh basket.

Part 1: Evaluation of Mix Designs 61 Figure 46. Part 1 HWTT results. Mix ID Total AC (%) Total P-200 (%) Fiber (%) Avg. Air Voids (%) Greatest Rut Depth (mm) RRI Average (mm) St. Dev. (mm) COV (%) Avg. (mm) St. Dev. (mm) COV (%) Georgia - Good 6 2.0 0.4 14.3 8.99 2.88 32 12921 2267 18 South Carolina - Poor 6 1.7 0.3 21.7 15.26a 1.84 11.6 1270 500 39 Florida - Poor 6.3 0.9 0.4 17.9 6.81 0.26 3.8 14638 202 1 Florida - Good 7.1 0.9 0.4 17.2 8.47 1.04 12.2 13328 816 6 Virginia - Poor 5.8 2.6 0.3 22.2 7.04 0.51 7.2 14457 400 3 New Jersey - Good 6 3.9 0.3 19.7 6.39 1.17 18.3 14971 921 6 aRut depth recorded at maximum limit of LVDT, which occurred at 3200 passes. Table 26. Part 1 HWTT summary. The South Carolina design specimens failed so quickly that the maximum rut depth was reached prior to the machine reaching 4,000 passes. So while the average maximum rut depth recorded for the South Carolina mix is reported as 15.85 mm, it is not wholly representative of the sample’s performance because that value does not show how poorly the specimens per- formed. The graphs along with the accompanying data in the table provide a comprehensive view of the HWTT. The RRI was also determined for each HWTT replicate according to Equation 6 and sum- marized in Table 26. All the mixtures, except for South Carolina’s granite mix design, yielded an RRI greater than 10,000, suggesting good rutting performance.

62 Performance-Based Mix Design of Porous Friction Courses Moisture Susceptibility Testing The moisture susceptibility testing for Part 1 provided both indirect tensile strength (ITS) of the conditioned and unconditioned specimens, along with the TSR for each mixture. All of the designs met the expected AASHTO PP 77 criterion of 0.70 retained tensile strength. The ASTM D7064 specification requires a TSR of 0.80 or better. If the ASTM criterion were applied to these designs, the Florida poor and the Georgia good mixtures would fail. The lower TSR could have been due to the change in type or amount of anti-stripping agent used. The low tensile strengths for the South Carolina mix may have been affected by the difference in anti-strip agent as well. The effect of replacing hydrated lime with liquid anti-strip on the Georgia, South Carolina, and New Jersey mixes is unknown. Even though the Georgia and Florida poor mixes failed the TSR requirement according to ASTM, they have two of the highest unconditioned strengths. The Georgia design had the second highest conditioned ITS values for this part of the study (Table 27). The ITS of the mixtures is essential because it directly relates to the asphalt’s cohesive poten- tial. So while it is important to note that two of the mixtures failed the TSR according to one specification, it is equally important to take into account the unconditioned ITS. The TSR can be improved most of the time by increasing the amount of liquid anti-strip or by switching to hydrated lime, but the ITS is based on the mixture properties and the corresponding adhesion of mixture components. An ANOVA was conducted to determine if any of the mixtures were significantly different. Tukey-Kramer statistical groupings were also included in the analysis to determine which mixtures were significantly different on a mix-by-mix basis. Table 28 shows the results of the analysis based on the ITS of each mixture. The data were grouped into two separate sets for the analysis: conditioned and unconditioned. Means that do not share the same letter are significantly different. Mix ID Total AC (%) Fiber (%) Avg. Specimen Air Voids (%) Avg. ITS (psi) TSR Conditioned Unconditioned Conditioned Unconditioned FL - Poor 6.3 0.4 17.2 17.2 52.8 72.5 0.73 FL - Good 7.1 0.4 17.4 17.5 54.0 50.1 1.08 VA - Poor 5.8 0.3 20.8 20.9 53.2 59.5 0.89 SC - Poor 6.0 0.3 21.2 21.2 36.7 45.2 0.81 GA - Good 6.0 0.4 13.9 14.0 57.7 74.3 0.78 NJ - Good 6.0 0.3 18.2 18.2 64.5 76.2 0.85 Table 27. Part 1—Moisture susceptibility testing summary. Mix ID Conditioned Unconditioned N Mean Grouping N Mean Grouping New Jersey - Good 3 64.5 A 3 76.2 A Georgia - Good 3 57.7 A B 3 74.3 A B Florida - Poor 3 54.0 B 3 72.5 A B Virginia - Poor 3 53.2 B 3 59.5 B C Florida - Good 3 52.8 B 3 50.1 C South Carolina - Poor 3 36.7 C 3 45.2 C Note: N = number of observations. Table 28. Part 1—ANOVA statistical comparisons of ITS (` = 0.05).

Part 1: Evaluation of Mix Designs 63 Based on unconditioned strengths, it can be seen that there is a gap between the mean strength of the mixtures. The New Jersey, Georgia, and Florida poor designs have strengths greater than 70 psi, while the Virginia, Florida good, and South Carolina designs have strengths of less than 60 psi. This distinct split between the unconditioned ITS values seems to indicate the need for a specified minimum ITS in the design procedure. The Florida poor and good designs appear to be swapped when analyzing these results based on field performance; however, the only difference between these designs was the binder modifier and total AC. The GTR design had an AC of 7.1%, which may have had a lubricating effect on the sample when it was broken in indirect tension. A comparison of the conditioned ITS values shows that the poor South Carolina mix had the lowest strengths. All the other mixes had tensile strengths of at least 50 psi. One reason for the low tensile strength may be due to the fact that the South Carolina mix also had the highest air voids. It is reasonable to expect the cohesion properties measured by tensile strength will decrease as the void structure between aggregate particles increases. Since many agencies typically design PFC mixes with up to 22% air voids based on the dimensional method (about 20% air voids based on the CoreLok method), there was a need to evaluate the effect of air voids on tensile strength for all the standard mixes in Part 1. The results (Figure 47) indicate that the minimum conditioned tensile strength should be about 50 psi at 20% air voids using the CoreLok procedure. The unconditioned tensile strength results tend to converge toward the conditioned results as air voids increase, which indicates that a minimum of 70 psi for unconditioned samples may not be practical at high air void levels. If a minimum conditioned strength and a minimum TSR are implemented, there is no need to specify a minimum unconditioned tensile strength. Shear Strength At least three replicates for each mix design were tested in the bond strength test device by shearing through the compacted sample. The results are an indication of shear resistance within the mix created by the combination of binder and inter-particle cohesion. Mixtures with higher shear strength should be better resistant to the shear effects that result in raveling. Table 29 shows that the poorly performing mixes from South Carolina and Virginia had the lowest shear strength. Those same two mixes, along with the Florida good mix, ranked lowest in area to peak results. ANOVA results with Tukey-Kramer grouping shown in Table 30 also show the South Carolina poor and Virginia poor are statistically in a separate group as far as shear strength. The Florida poor mix is also in a separate grouping since it had the highest shear strength of all six mix designs. The Florida good and South Carolina poor mixes had the lowest values for area to peak. Conditioned y = -1.9624x + 88.688 R² = 0.3064 Unconditioned y = -3.1256x + 119.8 R² = 0.3455 20 30 40 50 60 70 80 90 12.0 14.0 16.0 18.0 20.0 22.0 24.0 Te ns ile S tr en gt h, p si Va, % (based on CoreLok method) TS, Conditioned TS, Unconditioned Figure 47. Relationship of tensile strength to air voids.

64 Performance-Based Mix Design of Porous Friction Courses Figure 48 shows that there is a relationship between air voids and shear strength (R2 = 0.60) such that the shear strength of the mix decreases as air voids increase. This relationship was expected due to earlier comparisons between Cantabro loss and air voids. These results indicate there may need to be a maximum limit on air voids in order to improve PFC durability. At a maximum of 20% air voids using the CoreLok method, the minimum shear strength would be 130 psi. Cracking Susceptibility Test Results Overlay Test Results When running the OT, results are normally extremely variable. There were a few results in the OT testing for this study that seemed to be outliers, so the use of ASTM E178, Standard Practice for Dealing with Outlying Observations, was implemented. The standard is used to test the statistical significance of the results from a study to determine if an “outlier” is present within the data set. Equation 10 uses the average, standard deviation, and the number of observations to compare to a confidence interval value. The confidence interval value is provided in a table in the standard. A one-sided test with a confidence interval of 90% was chosen for this evaluation. Equation 10T X X S n n( )= − where Tn = test criterion X n = anticipated outlier X – = arithmetic average of all n values S = estimate of the population standard deviation based on the sample data Mix ID Va, (%) Shear Strength (psi) Area to Peak (lb/in.) Avg. Std. Dev. COV (%) Avg. Std. Dev. COV (%) FL Poor 17.5 170.4 19.9 11.7 541.9 56.0 10.3 FL Good 16.5 149.2 22.8 15.3 427.2 35.8 8.4 GA Good 15.1 162.0 12.6 7.8 510.9 40.2 7.9 NJ Good 19.6 138.6 11.9 8.6 606.5 66.2 10.9 SC Poor 21.5 122.2 6.8 5.6 461.6 52.2 11.3 VA Poor 21.4 121.1 9.4 7.8 492.5 67.6 13.7 Table 29. Shear strength results. Mix ID Shear Strength (psi) Mix ID Area to Peak (lb/in.) Grouping Grouping Florida - Poor A New Jersey - Good A Georgia - Good A B Florida - Poor A B Florida - Good A B Georgia - Good A B New Jersey - Good A B Virginia - Poor A B South Carolina - Poor B South Carolina - Poor B Virginia - Poor B Florida – Good B Table 30. ANOVA of shear strength and area to peak.

Part 1: Evaluation of Mix Designs 65 There was a single outlier in three of the specimen sets. Since there were at least four samples tested per mix design, this left at least three specimens on which statistical analysis could be per- formed. The cycles to failure was based on a 93% load reduction (Tx-248-F). The test terminates once the specimens reach 93% load reduction from the peak load or at 1,000 cycles. Some of the specimens went the full 1,000 cycles and never reached the 93% reduction prior to the test termi- nating. For these specimens, the data was extrapolated to determine the number of cycles it would have taken to reach to the 93% load reduction. A summary of the data sets and their properties can be found in Table 31, while a figure depicting the cycles to failure can be found in Figure 49. The coefficient of variance (COV) for the OT is normally extremely high (approximately 50%). The average COV for this testing was 20.5%, which indicates that these data results show less variability than typical OT specimens. This may be due in part to the ASTM outlier specification being implemented. Using a one-way ANOVA (α = 0.05), it was determined that the mixtures were statistically different. The grouping according to the Tukey-Kramer method can be found in Table 32. The model fit was good (R2 = 84.22%), which is most likely due to the COV of cycles to failure being lower than typical for this testing. A clear difference between the mixtures can be observed when looking at the means. The New Jersey, South Carolina, and Virginia designs all had to be extrapolated because they exceeded the 1,000 cycle test limit. This can be seen in the groupings as well, since they are grouped separately from the remaining three designs. The Florida good design performed extremely poorly (mean cycle to failure of 67). It is not clear why this was the case, especially when this mixture performed well in the field. y = -6.9422x + 272.47 R² = 0.603 100 110 120 130 140 150 160 170 180 190 200 12.00 14.00 16.00 18.00 20.00 22.00 24.00 Sh ea r S tr en gt h, p si Air Voids, % (based on CoreLok method) Figure 48. Relationship of shear strength and air voids. Mix ID Replicates Average Air Voids (%) Average Peak Load (kN) Cycles to Failure Average St. Dev. COV (%) FL - Poor 3 18.7 2.093 370 57 15.4 FL - Good 3 17.8 1.731 67 2 2.3 VA - Poor 4 19.2 1.818 1,291 431 33.4 SC - Poor 6 19.2 1.798 1,491 388 26.0 GA - Good 3 12.8 2.621 583 166 28.4 NJ - Good 4 18.5 1.993 1,866 324 17.3 Table 31. Part 1—overlay tester summary.

66 Performance-Based Mix Design of Porous Friction Courses I-FIT The I-FIT procedure was used to determine the mixture susceptibility to intermediate temperature cracking. While most dense-graded mixtures have an FI ranging from 0 to 20, the values for the PFC mixtures were much larger due to the large slope, post-peak. Since most of the testing performed to date has been primarily on dense-graded mixtures with varying amounts of asphalt binder replacement, the use of the FI to distinguish the good and poor mixes for this project will be subjective. The peak load, Gf, and FI were all analyzed in order to see which prop- erty would provide the best model for distinguishing the mixtures. Figure 50 through Figure 52 show a comparison of the mixtures based on peak load, Gf, and FI. The determination of possible outliers was also performed for this part of the testing; and based on the results, three specimens were determined to be outliers and were removed from the test data (1 specimen each from Georgia, Florida good, and New Jersey) prior to performing any analysis. Using the Minitab software, an ANOVA statistical analysis with Tukey-Kramer grouping was performed on the peak load, Gf, and FI (Table 33, Table 34, and Table 35). While all of the analyses showed significant differences between the mixtures, the FI analysis provided the best fit with an R2 = 71.41%. The South Carolina and Virginia designs were statistically different from the other designs. Since FI is an indication of the mixture’s resistance to cracking, these results were not surprising since those projects were replaced due to raveling and not because of crack- ing. Both the South Carolina and Virginia mixtures showed ample flexibility in both the I-FIT and OT. However, durability testing (Cantabro) showed these mixes to perform very poorly. This shows that while a mix may be exceedingly flexible, it may not necessarily be durable. In terms of Gf, both of the Florida designs were grouped together. The Gf of these specimens was significantly less than that of the other mixtures tested. 370 67 1,291 1,491 583 1,866 0 500 1,000 1,500 2,000 2,500 FL Poor SBS FL Good GTR VA Poor SBS SC Poor SBS GA Good SBS NJ Good SBS A ve ra ge O T C yc le s to F ai lu re Figure 49. Part 1—overlay tester results. Mix ID N Mean Grouping New Jersey - Good 4 1866 A South Carolina - Poor 6 1491 A Virginia - Poor 4 1291 A B Georgia - Good 3 583 B C Florida - Poor 3 370 C Florida - Good 3 67 C p < 0.001 R2 = 84%. Table 32. Part 1—ANOVA statistical comparisons of OT results (` = 0.05).

Part 1: Evaluation of Mix Designs 67 1.694 1.263 1.556 1.297 1.255 1.582 0.000 0.500 1.000 1.500 2.000 2.500 Georgia - Good South Carolina - Poor Florida - Poor Florida - Good Virginia - Poor New Jersey - Good kN Figure 50. Part 1 I-FIT—average peak load. 1,828 1,924 1,491 1,193 1,850 1,929 0 500 1000 1500 2000 2500 3000 Georgia - Good South Carolina - Poor Florida - Poor Florida - Good Virginia - Poor New Jersey - Good A ve ra ge F ra ct ur e E ne rg y (J /m 2) Figure 51. Part 1 I-FIT—average fracture energy (Gf). 28.7 57.7 23.5 25.2 57.5 35.6 0.0 20.0 40.0 60.0 80.0 100.0 Georgia - Good South Carolina - Poor Florida - Poor Florida - Good Virginia - Poor New Jersey - Good A ve ra ge F le xi bi lit y In de x (F I) Figure 52. Part 1 I-FIT—average flexibility index (FI). Mix ID Peak Load (kN) N Mean Grouping Georgia - Good 5 1.694 A New Jersey - Good 5 1.582 A B Florida - Poor 6 1.556 A B Florida - Good 5 1.297 B South Carolina - Poor 4 1.263 B Virginia - Poor 6 1.255 B p = 0.001 R2 = 54%. Table 33. Part 1 I-FIT ANOVA for peak load.

68 Performance-Based Mix Design of Porous Friction Courses Wet Track Abrasion The Wet Track Abrasion Test was originally performed according to the ISSA TB-100 test procedure and the specimens were submerged in water and tested for 5.25 minutes. This produced no visible wear on the specimen, so the testing time was increased to 30 minutes. After the 30 minute test, there was still no visible wear on the specimen. After drying the specimen and reweighing it, it was noted that there was no abrasion loss (Figure 53). The rubber hose had begun to abrade during the 30 minute test cycle, so it was decided that this procedure would not be valid for determining the durability of the PFC designs. The purpose of this test was to determine the cohesiveness of the mixtures by abrading them with a rubber hose. After failure of the Wet Track Abrasion Test to simulate raveling of aggregate particles, an alternative test was chosen to replace the Wet Track Abrasion Test. An experiment using a modified version of the I-FIT for cracking was conducted to measure the amount of shear force needed to break the cohesive bond of the mixture. I-FIT (No-Notch Variant) Given the shortcomings of the Wet Track Abrasion Test to measure the cohesion of PFC, a modification of the I-FIT was pursued as another test for mixture cohesion. The I-FIT specimen was fabricated according to the test procedure, but a notch was not cut into the specimen. This allowed the crack to form at the path of least resistance. The data analysis method was the same as the original I-FIT procedure; however, since it is a modification to the procedure, the normal criteria may not apply. This was solely used to try and differentiate the good from the poor mixtures. There were three outliers removed from the data prior to performing the analysis (one from Georgia, Virginia, and Florida poor). The average peak load (Figure 54) recorded for this test Mix ID Fracture Energy (J/m2) N Mean Grouping New Jersey - Good 5 1929 A South Carolina - Poor 4 1924 A Virginia - Poor 6 1850 A Georgia - Good 5 1828 A Florida - Poor 6 1491 A B Florida - Good 5 1193 B p = 0.001 R2 = 56% Table 34. Part 1 I-FIT ANOVA for fracture energy. Mix ID FI N Mean Grouping South Carolina - Poor 4 57.7 A Virginia - Poor 6 57.5 A New Jersey - Good 5 35.6 B Georgia - Good 5 28.7 B Florida - Good 5 25.2 B Florida - Poor 6 23.5 B p < 0.001 R2 = 71% Table 35. Part 1 I-FIT ANOVA for flexibility index.

Part 1: Evaluation of Mix Designs 69 shows a difference in the mixtures. From the results of the ANOVA (Table 36) it can be seen that the Georgia, Florida poor, and the New Jersey mixtures are significantly different from the other mixtures. The model fit is good (82.53%), and there is a distinct numerical separation of the means. Virginia and South Carolina are statistically different designs, while Florida good was not statistically different from either. The Gf of the designs did not provide as much separation of the mixtures as the peak load, but the Florida good design was statistically different from all of the other designs. The New Jersey design with a Gf of 3871 J/m 2 was statistically different from the South Carolina and Florida good designs. The results of this testing can be found in Figure 55, with the corresponding analysis in Table 37. The FI of the designs trended the same as the notched I-FIT specimens. The results (Figure 56) showed that South Carolina and Virginia still had the highest FI, while the Florida designs had the lowest. Georgia and both of the Florida designs were statistically different from the other designs, while South Carolina was statistically different from all of the designs (Table 38). A comparison of the notch and no-notch I-FIT data was conducted to determine if the no- notch modification was a valid method for determining cohesion of the mixture or if the results were not significantly different from the notched data. The peak load, Gf. and FI were analyzed for each mix design and the notch and no-notch results were compared using a t-test. Equal variance was assumed, and α of 0.05 was used for a confidence interval of 95%. Graphical com- parisons of these properties can be seen in Figure 57 through Figure 59, and a summary table with the results (p-value) of the t-tests can be found in Table 39. A ranking of mix performance by test procedure can be found in Table 40. Sample 146 – Prior to Testing Sample 146 – After 30 Minutes of Abrasion Figure 53. Wet Track Abrasion sample showing no abrasion for PFC mixture. 3.111 2.057 2.951 1.645 3.286 2.295 0.000 0.500 1.000 1.500 2.000 2.500 3.000 3.500 4.000 Georgia - Good Florida - Good New Jersey - Good South Carolina - Poor Florida - Poor Virginia - Poor A ve ra ge P ea k L oa d (k N ) Figure 54. Part 1 I-FIT no-notch peak load chart.

70 Performance-Based Mix Design of Porous Friction Courses Mix ID Peak Load N Mean Grouping Florida - Poor 5 3.286 A Georgia - Good 5 3.111 A New Jersey - Good 6 2.951 A Virginia - Poor 5 2.295 B Florida - Good 6 2.057 B C South Carolina - Poor 6 1.645 C p < 0.001 R2 = 83% Table 36. ANOVA for Peak load of Part 1 no-notch I-FIT specimens. 3017 1646 3871 2572 3193 3348 0 500 1000 1500 2000 2500 3000 3500 4000 Georgia - Good Florida - Good New Jersey - Good South Carolina - Poor Florida - Poor Virginia - Poor Av er ag e Fr ac tu re E ne rg y (J /m 2) Figure 55. Part 1 no-notch I-FIT fracture energy chart. Mix ID Fracture Energy N Mean Grouping New Jersey - Good 6 3871 A Virginia - Poor 5 3348 A B Florida - Poor 5 3193 A B Georgia - Good 5 3017 A B South Carolina - Poor 6 2572 B C Florida - Good 6 1646 C p < 0.001 R2 = 62% Table 37. ANOVA for fracture energy of Part 1 no-notch I-FIT specimens. 21.6 19.5 36.1 62.7 21.1 49.6 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 Georgia - Good Florida - Good New Jersey - Good South Carolina - Poor Florida - Poor Virginia - Poor Av er ag e Fl ex ib ili ty In de x (F I) Figure 56. Part 1 no-notch I-FIT FI chart.

Part 1: Evaluation of Mix Designs 71 Mix ID FI N Mean Grouping South Carolina - Poor 6 62.7 A Virginia - Poor 5 49.6 A B New Jersey - Good 6 36.1 B C Georgia - Good 5 21.6 C Florida - Poor 5 21.1 C Florida - Good 6 19.5 C p < 0.001 R2 = 72% Table 38. ANOVA for FI of Part 1 no-notch I-FIT specimens. 1.556 1.249 1.255 1.263 1.694 1.694 3.286 2.057 2.295 1.645 3.111 2.951 0.000 0.500 1.000 1.500 2.000 2.500 3.000 3.500 4.000 FL Poor FL Good VA Poor SC Poor GA Good NJ Good Av er ag e Le ak L oa d kN Peak Load - Notch Peak Load - No Notch Figure 57. I-FIT notch versus no-notch comparison for peak load. 1,491 1,205 1,850 1,924 1,828 1,929 3,193 1,646 3,348 2,572 3,017 3,871 0 1,000 2,000 3,000 4,000 5,000 6,000 FL Poor FL Good VA Poor SC Poor GA Good NJ Good Av er ag e Fr ac tu re E ne rg y (J /m 2) FE - Notch FE - no notch Figure 58. I-FIT notch versus no-notch comparison for fracture energy.

72 Performance-Based Mix Design of Porous Friction Courses 23.5 27.9 57.5 57.7 28.7 35.6 21.1 19.5 49.6 62.7 21.6 36.1 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 FL Poor FL Good VA Poor SC Poor GA Good NJ Good Av er ag e Fl ex ib ili ty In de x FI - Notch FI - No Notch Figure 59. I-FIT notch versus no-notch comparison for FI. Mix ID p-Value Peak Load Gf FI FL Poor SBS 0.000 0.000 0.575 FL Good GTR 0.015 0.059 0.052 VA Poor SBS 0.000 0.000 0.268 SC Poor SBS 0.013 0.032 0.641 GA Good SBS 0.000 0.019 0.212 NJ Good SBS 0.000 0.002 0.949 Table 39. Significance of I-FIT notch versus no-notch specimens. Test Procedure NJ GA FL Good FL Poor VA SC Cantabro 1 2 3 4 5 6 Hamburg 1 5 4 2 3 6 TSR 3 5 1 6 2 4 ITS, Conditioned 1 2 3 5 4 6 ITS, Unconditioned 1 2 5 3 4 6 Shear 4 2 3 1 6 5 Shear, Area to Peak 1 3 6 2 4 5 Overlay Tester 1 4 6 5 3 2 I-FIT, Peak Load (Notched) 2 1 4 3 6 5 I-FIT, Fracture Energy (Notched) 1 4 6 5 3 2 I-FIT, Flexibility Index (Notched) 3 4 5 6 2 1 I-FIT, Peak Load (No Notch) 3 2 5 1 4 6 I-FIT, Fracture Energy (No Notch) 1 4 6 3 2 5 I-FIT, Flexibility Index (No Notch) 3 4 6 5 2 1 Table 40. Ranking of mix performance by test procedure.

Part 1: Evaluation of Mix Designs 73 All of the designs when analyzed for peak load were significantly different. All of the designs except Florida poor were also significantly different for the Gf. The Florida good mix was close (0.059), which is suggestive of it being statistically different, but conclusive results most likely cannot be drawn from this value. All of the FI comparisons show that none of the designs are affected by the notch in the specimens. The FI data for each group are not statistically different. Since peak load and almost all of the fracture energy comparisons are statistically different for the regular I-FIT and no-notch I-FIT specimens, both methods are used for comparison of the remaining mixtures for this study. Since FI was shown to be statistically equivalent for both the regular I-FIT and no-notch I-FIT, only the FI of the regular I-FIT specimens will be reported for the remaining testing.