Variability of Ignition Furnace Correction Factors (2017)

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.

34 5.1 Sensitivity Experiments The results of the sensitivity study were analyzed using the linear regression method. For each combination of variables (furnace type, test temperature, air flow, sample mass, and asphalt content), the data were fitted to a linear model as follows: = + + + + + + ε Equation 21 1 2 2 3 3 4 4 5 5Y B B X B X B X B X B Xi o i where Yi = ith observation of the dependent vari- able (observed value), B0, B1, B2, B3, B4, B5 = model coefficients, X1, X2, X3, X4, X5 = independent variables (factors included in the sensitivity study), and ei = error term. Following this approach, the statistical significance of the model coefficients was determined. For statistically signifi- cant factors, the model coefficients can be used to estimate the values for each factor that will keep their effect below a specified level. 5.1.1 Experiment 1 – Three Furnaces 5.1.1.1 Factors Affecting Asphalt Content Correction Factor The results of the asphalt content correction factor sensi- tivity study are included in Appendix C. For this experiment, factors listed in Table 19 (A to E) were included. Regression equations for the asphalt content correction factor were devel- oped including factors A to E. A summary of these results is presented in Table 26. This table summarizes the probability values (p-values) indicating the significance of the regression coefficient for each of the factors included in the sensitivity experiment. The p-value is the probability of rejecting the null hypoth- esis when it is, in fact, true. For this evaluation, it is the prob- ability that the regression coefficient for a factor is 0 when the analysis shows that it is either greater than or less than 0. A low p-value indicates that the regression coefficient is statisti- cally significant and the factor has an effect on the test results. The analysis consists of selecting a critical p-value above which the regression coefficient is not significant, so it can be concluded that the sensitivity factors do not affect the test results. In general, a prediction model of the form presented in Equation 1 must include variables that are highly corre- lated to the predicted outcome. Therefore, a p-value of 0.05 is usually selected, but for a sensitivity evaluation, selecting a p-value that is too small may result in an incorrect conclusion that may exclude one or more factors and affect the results. For this study, a p-value of ≤0.10 was considered a statistically significant difference. Table 26 presents the factors with their corresponding p-values; when p-values are less than or equal to 0.10, they are shown in bold. This table summarizes the results by mixture. Table 27 summarizes the average asphalt correction fac- tors for each variable at each level of analysis by mix. Based on the average asphalt correction factor tests results, it was con- sidered important to evaluate whether the differences were large enough to be practically significant. Since the current AASHTO standard specifies an acceptable range of two test results equal to 0.196, a difference higher than 0.10 was con- sidered a good threshold for a significant difference that would be accepted by agencies. If a difference in asphalt correction factor of more than 0.10 between the levels of each variable is considered practically significant, some of the results that were found statistically significant may not be practically significant. Table 27 includes a column to indicate when there was a sta- tistically significant difference in the results; when there was a statistically significant difference, it was determined whether there was a practical difference (final column). C h a p t e r 5 Results and Analysis of the Experiments

35 Table 26. Significance of asphalt content correction factor test parameters based on p-values – sensitivity study – Experiment 1 by mixture. Material Ignition Unit Test Temperature (°F) Air Flow AC Content(%) Sample Mass (grams) Mix 1 0.00 0.00 0.34 0.00 0.40 Mix 2 0.00 0.00 0.69 0.00 0.30 Mix 3 0.00 0.00 0.27 0.89 0.29 Mix 4 0.01 0.00 0.08 0.59 0.48 Note: Bold type indicates p-values that are less than or equal to 0.10. Table 27. Average asphalt content correction factors by mixture for all variables at each level – Experiment 1. Mix # Variable Levels Average Correction Factor (%) Statistically Significant (Y/N) Practically Significant (Y/N) Mix 1 Ignition unit Thermolyne Y Y Troxler Gilson Temperature 800°F Y Y 1,000°F Air flow Damper 30% N –Damper 100% AC content Opt. AC − 1% Y N Opt. AC + 1% Sample mass 1,500 g N –2,000 g Mix 2 Ignition unit Thermolyne Y N Troxler Gilson Temperature 800°F Y N 1,000°F Air flow Damper 30% N –Damper 100% AC content Opt. AC − 1% N –Opt. AC + 1% Sample mass 1,500 g N —2,000 g Mix 3 Ignition unit Thermolyne Y Y Troxler Gilson Temperature 800°F Y Y 1,000°F Air flow Damper 30% N —Damper 100% AC content Opt. AC − 1% N —Opt. AC + 1% Sample mass 1,500 g N —2,000 g Mix 4 Ignition unit Thermolyne Y Y Troxler Gilson Temperature 800°F Y Y 1,000°F Air flow Damper 30% Y Y Damper 100% AC content Opt. AC − 1% N —Opt. AC + 1% Sample mass 1,500 g N —2,000 g 0.07 0.12 −0.03 −0.02 0.12 0.06 0.04 0.10 0.00 0.04 0.06 −0.26 −0.21 −0.36 −0.33 −0.23 −0.27 −0.26 −0.24 −0.31 −0.27 −0.28 0.55 0.82 0.33 0.35 0.78 0.60 0.53 0.57 0.56 0.53 0.60 2.07 1.90 1.24 0.87 2.60 1.94 1.53 1.67 1.80 1.65 1.82 Note: Opt. = optimum.

36 From Table 26 and Table 27, results for the correction fac- tors are discussed in the following. For Mix 1. Three factors were found to be statistically significant: • Ignition furnace, with average correction factors of 0.07%, 0.12%, and -0.03% (for Thermolyne, Troxler, and Gilson furnaces, respectively). • Test temperature, with average correction factors of -0.02% and 0.12% (for 800 and 1,000°F for Thermolyne and Gilson; default and Option 1 for Troxler). • Asphalt content, with average correction factors of 0.10% and 0% (for optimum AC - 1% and optimum AC + 1%). When the practical difference was evaluated (>0.10), only ignition type and test temperature were found to be practi- cally significant. For Mix 2. Two factors were found to be significant: • Ignition furnace, with average correction factors of -0.26%, -0.21%, and -0.36% (Thermolyne, Troxler, and Gilson). • Test temperature, with average correction factors of -0.33% and -0.23% (800 and 1,000°F; default and Option 1 for Troxler). The differences in the tests results were determined to be statistically significant but are not considered practically significant. For Mix 3. Two factors were found to be significant: • Ignition furnace, with average correction factors of 0.55%, 0.82%, and 0.33% (Thermolyne, Troxler, and Gilson). • Test temperature, with average correction factors of 0.35% and 0.78% (for 800 and 1,000°F for Thermolyne and Gilson; default and Option 1 for Troxler). These results were considered to be statistically and practi- cally significant. For Mix 4. Three factors were found significant: • Ignition furnace, with average correction factors of 2.07%, 1.9%, and 1.24% (Thermolyne, Troxler, and Gilson). • Test temperature, with average correction factors of 0.87% and 2.6% (for 800 and 1,000°F for Thermolyne and Gilson; default and Option 1 for Troxler). • Air flow, with average correction factors of 1.94% and 1.53% (for damper 30% open and 100% open). These results were considered to be statistically and practi- cally significant. Based on these results, the following observations have been made regarding asphalt content correction factors for different furnaces and aggregates. 1. Type of furnace and test temperature were both statistically significantly significant for the four mixtures, but practi- cally significant only for Mixes 1, 3, and 4. For Mix 1, the maximum difference was equal to 0.15. For Mixes 3 and 4 with aggregates that experienced additional mass loss (in this case more than 0.5% mass loss) during ignition, causing higher correction factors, the type of furnace and test temperature had a bigger impact on the test results. 2. Reducing the test temperature from 1,000 to 800°F reduced the amount of aggregate mass loss for all mixes. This resulted in a lower correction factor for all mixes except Mix 2, which included lime. The correction factor for Mix 2 was negative, resulting in an increase in correction factor for the lower test temperature. 3. Air flow was found to be significant only for Mix 4. There- fore, differences in furnace exhaust configuration may be more critical for aggregates with higher correction factors. Surprisingly, the mass loss was more for the 30% damper opening than for the 100% damper opening. 4. For Mixes 1, 3, and 4, the Gilson unit yielded the lowest average correction factor, with average correction factors equal to -0.03%, 0.33%, and 1.24%. One possible reason could be the basket used during ignition. Instead of two baskets, the Gilson unit has only one basket, which may have resulted in the sample temperature being more uni- form, and as a result, the total decomposition of aggregate may be reduced. The Gilson unit requires that all weight measurements be taken outside the furnace, and the other units make the weight measurements automatically while samples are in the hot furnace. It is possible that weighing the sample outside the furnace may result in some differ- ence in measured mass loss when compared to measure- ments taken inside the furnace. As was presented in Table 27, although different variables were found to be significant for Mixes 1 and 2, the magnitude of the correction factor when compared to the other mixes is smaller. This suggests that, from a practical point of view, sharing correction factors between units for aggregates with low correction factors may not cause significant error. To further investigate the effect of test temperature, air flow, asphalt content, and sample mass by mix and furnace type, a similar statistical analysis to the one presented in Table 26 was conducted, but for this analysis, test results by mixture and furnace type were analyzed individually. Table 28 summarizes the p-value by variable, per each mixture and ignition unit. P-values that were found to be significant are shown in bold. The test results showed that test temperature was statistically

37 significant for 10 of 12 combinations, air flow for seven of 12 combinations, and asphalt content for eight of 12 com- binations. Sample mass was found significant only for one combination, Mix 2 for the Troxler furnace. In order to evaluate the statistical difference and practical difference of these results, Table 29 through Table 32 show the average correction factor by mix and furnace type at each level for the other four variables: temperature, air flow, asphalt content, and sample mass. Similar to the analysis of the results in Table 27, a difference in asphalt correction factor of more than 0.10 between levels of each variable was considered practically significant. The last two columns in Table 28. Significance of test parameters on asphalt content correction factors by mix and furnace type based on p-values – sensitivity study – Matrix 1. Material Ignition Unit Test Temperature (800°F vs. 1,000°F)* Air Flow (Damper 30% Open vs. Full Open) AC Content (Opt. − 1% vs. Opt. + 1%) Sample Mass (1,500 g vs. 2,000 g) Mix 1 Thermolyne 0.00 0.90 0.02 0.62 Troxler 0.01 0.00 0.10 0.58 Gilson 0.00 0.01 0.00 0.42 Mix 2 Thermolyne 0.00 0.06 0.00 0.55 Troxler 0.01 0.01 0.02 0.02 Gilson 0.00 0.51 0.01 0.17 Mix 3 Thermolyne 0.00 0.83 0.11 0.37 Troxler 0.26 0.00 0.01 0.09 Gilson 0.00 0.11 0.38 0.65 Mix 4 Thermolyne 0.00 0.37 0.71 0.69 Troxler 0.16 0.00 0.04 0.33 Gilson 0.00 0.03 0.94 0.29 Notes: Bold type indicates p-values that were found to be significant; Opt. = optimum. *For Troxler furnace default vs. Option 1. Ignition Unit Variable Levels Average Correction Factor (%) Statistically Significant (Y/N) Practically Significant (Y/N) Thermolyne Temperature 800°F Y Y 1,000°F Air flow Damper 30% N –Damper 100% AC content Opt. AC − 1% Y N Opt. AC + 1% Sample mass 1,500 g N – 2,000 g Troxler Temperature Option 1 Y N Default Air flow Damper 30% Y Y Damper 100% AC content Opt. AC − 1% Y N Opt. AC + 1% Sample mass 1,500 g N –2,000 g Gilson Temperature 800°F Y Y 1,000°F Air flow Damper 30% Y N Damper 100% AC content Opt. AC − 1% Y N Opt. AC + 1% Sample mass 1,500 g N N 2,000 g −0.02 0.15 0.07 0.06 0.11 0.01 0.06 0.08 0.07 0.11 0.20 0.04 0.15 0.09 0.11 0.13 −0.10 0.03 −0.08 0.01 0.02 −0.08 −0.04 −0.02 Note: Opt. = optimum. Table 29. Average asphalt content correction factors for Mix 1 by furnace type for variables at each level – Experiment 1.

Ignition Unit Variable Levels Average Correction Factor (%) Statistically Significant (Y/N) Practically Significant (Y/N) Thermolyne Temperature 800°F −0.31 Y N 1,000°F −0.21 Air flow Damper 30% −0.28 Y N Damper 100% −0.24 AC content Opt. AC − 1% −0.22 Y N Opt. AC + 1% −0.30 Sample mass 1,500 g −0.26 N –2,000 g −0.27 Troxler Temperature Option 1 −0.24 Y N Default −0.17 Air flow Damper 30% −0.18 Y N Damper 100% −0.25 AC content Opt. AC − 1% −0.18 Y N Opt. AC + 1% −0.24 Sample mass 1,500 g −0.18 Y N 2,000 g −0.24 Gilson Temperature 800°F −0.43 Y Y 1,000°F −0.29 Air flow Damper 30% −0.36 N – Damper 100% −0.35 AC content Opt. AC − 1% −0.33 Y N Opt. AC + 1% −0.38 Sample mass 1,500 g −0.37 N –2,000 g −0.35 Note: Opt. = optimum. Table 30. Average asphalt content correction factors for Mix 2 by furnace type for variables at each level – Experiment 1. Ignition Unit Variable Levels Average Correction Factor (%) Statistically Significant (Y/N) Practically Significant (Y/N) Thermolyne Temperature 800°F 0.23 Y Y 1,000°F 0.87 Air flow Damper 30% 0.55 N – Damper 100% 0.56 AC content Opt. AC − 1% 0.62 N – Opt. AC + 1% 0.47 Sample mass 1,500 g 0.51 N – 2,000 g 0.59 Troxler Temperature Option 1 0.79 N – Default 0.86 Air flow Damper 30% 1.00 Y Y Damper 100% 0.64 AC content Opt. AC − 1% 0.73 Y Y Opt. AC + 1% 0.91 Sample mass 1,500 g 0.77 N – 2,000 g 0.87 Gilson Temperature 800°F 0.03 Y Y 1,000°F 0.63 Air flow Damper 30% 0.38 N – Damper 100% 0.28 AC content Opt. AC − 1% 0.36 N – Opt. AC + 1% 0.30 Sample mass 1,500 g 0.32 N –2,000 g 0.35 Note: Opt. = optimum. Table 31. Average asphalt content correction factors for Mix 3 by furnace type for variables at each level – Experiment 1.

39 Table 29 through Table 32 indicate whether the results are statistically significant and practically significant. The analysis of the results for Mix 1 presented in Table 29 indicate the following: • Test temperature (burning profile for the Troxler unit) was found to be statistically significant for the three units, but practically significant only for the Thermolyne and Gilson units. The highest effect was found for the Thermolyne unit, with a difference in correction factor between high and low levels (D correction factor) of 0.17, followed by Gilson with a D correction factor equal to 0.13. • Air flow was statistically significant for Troxler and Gilson, but only the Troxler unit showed a practical difference of D correction factor equal to -0.16. (The negative sign indi- cates a lower correction factor with the damper fully open.) • Asphalt content was statistically significant for all units, but the differences were found to be equal to or less than 0.10 and were not considered practically significant. • Sample mass was not found to be statistically significant for any unit. From Table 30, the test results for Mix 2 indicate the following: • Test temperature was statistically significant for the three units, but practically significant only for the Gilson, with a D correction factor equal to -0.14. The negative sign in this case indicates that the correction factor is lower for higher test temperature. • Air flow was found to be statistically significant for the Thermolyne and Troxler units only, but the differences found for both units are not practically significant. • Asphalt content was statistically significant for all units, but none of these differences is practically significant. • Sample mass was found to be significant only for the Trox- ler unit, with a D correction factor equal to -0.06; there- fore, the difference is not practically significant. From Table 31, the test results for Mix 3 can be interpreted as follows: • Test temperature was statistically and practically signifi- cant for Thermolyne and Gilson only. Both units showed Ignition Unit Variable Levels Average Correction Factor (%) Statistically Significant (Y/N) Practically Significant (Y/N) Thermolyne Temperature 800°F 0.63 Y Y 1,000°F 3.51 Air flow Damper 30% 2.23 N – Damper 100% 1.91 AC content Opt. AC − 1% 2.10 N – Opt. AC + 1% 2.00 Sample mass 1,500 g 2.00 N – 2,000 g 2.14 Troxler Temperature Option 1 1.74 N –Default 2.07 Air flow Damper 30% 2.52 Y Y Damper 100% 1.29 AC content Opt. AC − 1% 1.64 Y Y Opt. AC + 1% 2.17 Sample mass 1,500 g 1.79 N – 2,000 g 2.02 Gilson Temperature 800°F 0.25 Y Y 1,000°F 2.22 Air flow Damper 30% 1.39 Y Y Damper 100% 1.08 AC content Opt. AC − 1% 1.24 N – Opt. AC + 1% 1.23 Sample mass 1,500 g 1.16 N – 2,000 g 1.31 Note: Opt. = optimum. Table 32. Average asphalt content correction factors for Mix 4 by furnace type for variables at each level – Experiment 1.

40 approximately the same D correction factor, equal to 0.6. For both units, higher temperature resulted in a higher correction factor. • Air flow was found to be both practically and statistically significant for the Troxler unit, with a D correction factor equal to 0.36. A higher correction factor was found when the damper was partially open. • Asphalt content was found to be statistically significant for the Troxler unit only, with a D correction factor equal to 0.18. • Sample mass was not statistically significantly different for any unit. The test results for Mix 4 provided in Table 32 showed the following: • Test temperature was statistically significant for Thermo- lyne and Gilson only, similar to the findings for Mix 3. The D correction factors were 2.88 and 1.95 for the Thermolyne and Gilson units, respectively. The Thermolyne unit showed the highest D correction factor; for all cases, the higher the temperature, the higher the correction factor. • Air flow was found to be statistically significant for the Troxler and Gilson units, with a D correction factor equal to -1.23 for the Troxler and -0.31 for the Gilson. The nega- tive signs indicate that a higher correction factor was found when the damper was partially open. • Asphalt content was found to be statistically and practi- cally significant for only the Troxler unit, with a D correc- tion factor equal to 0.52. The samples with higher asphalt content produced higher correction factor values. • Sample mass was not statistically significant for any of the three units. Based on the evaluation of the different variables by aggre- gate type and furnace type presented in Table 29 through Table 32, the following observations can be made: 1. Test Temperature. For seven of 12 combinations shown in Table 29 through Table 32, test temperature was found to affect the correction factor significantly. For the convection units, Thermolyne and Gilson, the effect of temperature was more significant than for the Troxler unit, which is indicated by a lower D correction factor for the Troxler unit. It is important to keep in mind that for the Troxler unit, the temperature cannot be adjusted; it is indi- rectly selected by choosing a burning profile. For five of the seven cases where temperature was found to affect the asphalt correction factor, Aggregates 3 and 4 were used. Decreasing the test temperature (or changing the burning profile from default to Option 1 for the Troxler unit) decreases the aggregate mass loss for all mixtures. 2. Air Flow (Damper Opening). For four of 12 com- binations presented in Table 29 through Table 32, air flow affected the correction factor results significantly. The results indicate that a restriction in air flow may cause an increase in the asphalt correction factor. Air flow seems to have a higher impact for the Troxler unit (higher D correction fac- tors were found). It is surprising that partially closing the damper resulted in higher aggregate mass loss, but this was the general trend. 3. Asphalt Content. This factor was found to be sig- nificant in three of 12 combinations, with two of these com- binations for Mix 3 and one for Mix 4. The effect of asphalt content on the correction factor was also found to be more significant for the Troxler unit. 4. Sample Mass. This factor was found not to be significant. An additional evaluation was conducted to determine the effect that furnace type had on the total test duration and maximum furnace temperature recorded during the test. Fig- ure 27 through Figure 29 summarize the average test time by test temperature/burning profile for Thermolyne, Troxler, and Gilson units. From the figures, it can be observed that the average test times were significantly lower for the Troxler furnace for all mixtures. Test times for the Troxler unit ranged from 34 to 45 min for Option 1 and 32 to 47 min for the default burning profile. For the Thermolyne furnace, the average test time ranged from 59 to 77 min for testing con- ducted at 800°F (427°C) and from 53 to 146 min for testing conducted at 1,000°F (538°C). Finally, for the Gilson unit, average test time ranged from 118 to 152 min for testing con- ducted at 800°F (427°C) and 81 to 120 min for testing con- ducted at 1,000°F (538°C). For the Thermolyne unit, higher test times were needed for Mix 4 for testing conducted at 1,000°F (538°C) since the aggregate used for this mix con- tains dolomites, which makes it difficult for the weight to stabilize during testing, and the burning continues for longer periods of time. Since the Gilson unit has no internal balance, AASHTO T 308 standard procedure requires burning the asphalt mix specimen for at least 45 min. After this time, the sample is removed from the furnace and allowed to cool for approxi- mately 30 min, and then the mass is determined with an external balance. The sample is placed in the furnace again for at least 15 min after the furnace reaches the set tempera- ture. After this time, the sample is allowed to cool for an addi- tional 30 min and the mass is determined. This procedure is repeated until the change in measured mass of the specimen after ignition does not exceed 0.01% of the initial specimen mass. The results for the Gilson unit presented in Figure 29

41 0 20 40 60 80 100 120 140 160 Mix 1 Mix 2 Mix 3 Mix 4 Mix 1 Mix 2 Mix 3 Mix 4 800°F 1000°F A ve ra ge T es t Ti m e (m in ) Figure 27. Average test time by mix type and test temperature for Thermolyne unit. 0 20 40 60 80 100 120 140 160 Mix 1 Mix 2 Mix 3 Mix 4 Mix 1 Mix 2 Mix 3 Mix 4 Opon 1 Default A ve ra ge T es t T im e (m in ) Figure 28. Average test time by mix type and burning profile for Troxler unit. 0 20 40 60 80 100 120 140 160 Mix 1 Mix 2 Mix 3 Mix 4 Mix 1 Mix 2 Mix 3 Mix 4 800°F 1000°F A ve ra ge T es t T im e (m in ) Figure 29. Average test time by mix type and test temperature for Gilson unit.

42 only include the time the samples were inside the furnace, not the time the samples were allowed to cool. Figure 30 and Figure 31 show the average maximum test temperature by mixture type and test temperature for the Thermolyne and Troxler furnaces for testing conducted at 800°F (427°C) (Option 1 for the Troxler unit) and 1,000°F (427°C) (default for the Troxler unit). During the test, the temperatures were recorded by thermocouples mounted on the back wall of the furnace chambers. The Gilson furnace does not record the test temperatures. The average test tem- peratures were lower for the Troxler furnace for all mixtures. For tests conducted using Option 1 and default, the average maximum test temperatures for the Troxler unit ranged from 958 to 1,008°F (514 to 542°C) and 1,016 to 1,057°F (547 to 569°C), respectively. For tests conducted at 800 and 1,000°F (427 and 538°C), the average maximum test temperatures for the Thermolyne unit ranged from 1,006 to 1,043°F (541 to 562°C) and 1,069 to 1,088°F (576 to 587°C), respectively. 5.1.1.2 Factors Affecting Aggregate Gradations In order to determine what factors may affect the aggre- gate gradations of the different mixtures, an analysis similar to the one used for the asphalt content correction factor was conducted. Regression equations for each aggregate grada- tion were developed including factors A to E in Table 19. Table 33 through Table 36 summarize the probability values (p-values) for each mixture by sieve size. The same criteria used for correction factors were used, and a p-value ≤0.10 was considered a statistically significant difference. 960 980 1000 1020 1040 1060 1080 1100 Mix 1 Mix 2 Mix 3 Mix 4 Mix 1 Mix 2 Mix 3 Mix 4 F°0001F°008 A ve ra ge M ax .T es t Te m pe ra tu re (° F) Figure 30. Average maximum temperature by mix type and test temperature for Thermolyne unit. 900 920 940 960 980 1000 1020 1040 1060 1080 Mix 1 Mix 2 Mix 3 Mix 4 Mix 1 Mix 2 Mix 3 Mix 4 Opon 1 Default A ve ra ge M ax . T es t Te m pe ra tu re , ° F Figure 31. Average maximum temperature by mix type and test temperature for Troxler unit.

43 Factors 1/2” 3/8” #4 #8 #16 #30 #50 #100 #200 Furnace (Thermolyne, Troxler, Gilson) 0.06 0.09 0.00 0.02 0.05 0.15 0.71 0.99 0.08 Test temp. (800°F vs. 1,000°F)* 0.61 0.16 0.47 0.16 0.32 0.29 0.33 0.38 0.64 Air flow (damper 30% open vs. full open) 0.11 0.02 0.08 0.84 0.69 0.89 0.45 0.36 0.95 AC (Opt. − 1% vs. Opt. + 1%) 0.59 0.77 0.87 0.57 0.71 0.84 0.75 0.58 0.20 Sample mass (1,500 g vs. 2,000 g) 0.38 0.52 0.22 0.84 0.95 0.82 0.52 0.46 0.26 Notes: Bold type indicates p-values that were found to be significant; Opt. = optimum. *For Troxler furnace default vs. Option 1. Table 33. Significance of sensitivity test factors on gradation for Mix 1 based on p-values (Experiment 1). Factors 1/2” 3/8” #4 #8 #16 #30 #50 #100 #200 Furnace (Thermolyne, Troxler, Gilson) 0.17 0.07 0.07 0.08 0.22 0.30 0.14 0.07 0.0 Test temp. (800°F vs. 1,000°F)* 0.25 0.21 0.20 0.78 0.84 0.87 0.41 0.11 0.00 Air flow (damper 30% open vs. full open) 0.95 0.91 0.92 0.41 0.29 0.25 0.24 0.19 0.31 AC (Opt. − 1% vs. Opt. + 1%) 0.00 0.00 0.00 0.12 0.98 0.83 0.75 0.17 0.00 Sample mass (1,500 g vs. 2,000 g) 0.06 0.09 0.48 0.00 0.00 0.00 0.00 0.01 0.00 Notes: Bold type indicates p-values that were found to be significant; Opt. = optimum. *For Troxler furnace default vs. Option 1. Table 34. Significance of sensitivity test factors on gradation for Mix 2 based on p-values (Experiment 1). Factors 1/2” 3/8” #4 #8 #16 #30 #50 #100 #200 Furnace (Thermolyne, Troxler, Gilson) 0.01 0.51 0.27 0.12 0.12 0.13 0.19 0.50 0.38 Test temp. (800°F vs. 1,000°F)* 0.36 0.08 0.06 0.30 0.33 0.32 0.26 0.26 0.11 Air flow (damper 30% open vs. full open) 0.36 0.87 0.18 0.16 0.18 0.21 0.22 0.30 0.27 AC (Opt. − 1% vs. Opt. + 1%) 0.03 0.39 0.06 0.37 0.38 0.47 0.94 0.14 0.02 Sample mass (1,500 g vs. 2,000 g) 0.91 0.81 0.92 0.54 0.57 0.59 0.71 0.70 0.44 Notes: Bold type indicates p-values that were found to be significant; Opt. = optimum. *For Troxler furnace default vs. Option 1. Table 35. Significance of sensitivity test factors on gradation for Mix 3 based on p-values (Experiment 1). Factors 1/2” 3/8” #4 #8 #16 #30 #50 #100 #200 Furnace (Thermolyne, Troxler, Gilson) 0.79 0.88 0.75 0.97 0.80 0.70 0.82 0.24 0.80 Test temp. (800°F vs. 1,000°F)* 0.48 0.69 0.39 0.06 0.34 0.34 0.24 0.08 0.00 Air flow (damper 30% open vs. full open) 0.07 0.04 0.68 0.04 0.40 0.53 0.50 0.55 0.70 AC (Opt. − 1% vs. Opt. + 1%) 0.74 0.43 0.96 0.19 0.90 0.99 0.80 0.70 0.86 Sample mass (1,500 vs. 2,000 g) 0.00 0.00 0.07 0.00 0.00 0.00 0.00 0.00 0.00 Notes: Bold type indicates p-values that were found to be significant; Opt. = optimum. *For Troxler furnace default vs. Option 1. Table 36. Significance of sensitivity test factors on gradation for Mix 4 based on p-values (Experiment 1).

44 Based on the analysis results shown in Table 33 through Table 36, the following observations were made regarding the test results: For Mix 1. • Furnace type was found to be significant for six of nine sieve sizes. For sieve sizes equal to or larger than a #8 sieve, the difference of the average percentage passing (D% pass- ing) for the three furnaces was equal to or less than 2.4%. For sieves smaller than a #8 sieve and larger than #200, the D% passing was equal to or less than 0.98%. For #200 sieves, the D% passing was less than 0.27%. • Air flow was found to be significant only for two of nine sieve sizes (3⁄8″ and #4). For both of these sieves, the D% passing was equal to or less than 2.2%. For Mix 2. • Furnace type was found to be significant for five of nine sieve sizes. For sieve sizes equal to or larger than a #8 sieve, the D% passing for the three furnaces was equal to or less than 2.6%. For sieves smaller than a #8 sieve and larger than #200, the D% passing was equal to or less than 0.4%. The D% passing for a #200 sieve was 0.4%. • Test temperature was found to be significant only for #200 sieves. For this case, the D% passing (difference between low and high temperature) was 0.2%. • Asphalt content was significant for four out of nine sieve sizes. For sieves equal to or larger than #4, the D% passing was equal to or less than 4.7%. For #200 sieves, the D% pass- ing was 0.2%. • Sample mass was significant for eight out of nine sieve sizes. For sieves equal to or larger than #8, the D% passing was equal to or less than 1.5%. For sieves smaller than a #8 sieve and larger than #200, the D% passing was equal to or less than 0.8%. The D% passing for #200 sieves was 0.3%. For Mix 3. • Furnace type was significant only for ½″ sieves. For this case, the D% passing was 1.4%. • Test temperature was significant only for 3⁄8″ and #4 sieves. For these combinations, the D% passing was 0.8% and 0.6%, respectively. • Asphalt content was significant for three of nine sieve sizes. For ½″ and #4 sieves, D% passing was 0.8% and 0.7%, respectively. For #200 sieves, D% passing was 0.5%. For Mix 4 • Test temperature was significant for three of nine sieve sizes. For #8, #100, and #200 sieves, D% passing was 0.6%, 0.4%, and 0.5%, respectively. • Air flow was found to be significant only for three of nine sieve sizes. For ½″ and 3⁄8″ sieves, D% passing was 0.9% and 1.4%, respectively. For #8 sieves, D% passing was 0.6%. • Sample mass was significant for all of the sieve sizes. For sieves equal to or larger than #8, the D% passing was equal to or less than 2.2%. For sieves smaller than #8 and larger than #200, the D% passing was equal to or less than 2.4%. For #200 sieves, D% passing was 0.6%. Table 37 presents a summary of the number of times and corresponding percentage of times that each variable was found to be statistically significant for each mix for the total number of combinations presented in Table 33 through Table 36. Based on the evaluation of the different variables by mix and furnace type, the following observations can be made regarding aggregate gradations after ignition: • Furnace type and temperature seem to affect aggregate gra- dation, but the differences found in the D% passing were all less than or equal to 2.6% for sieve sizes equal to or larger than #8 sieves, 0.98% for sieves smaller than #8 sieves and larger than #200 sieves, and 0.5% for #200 sieves. • Test temperature appears to affect aggregate gradation, but only for 16% of the different combinations. This finding is somehow different from the findings for asphalt correction factors where temperature was found to have a significant impact for the different mixtures. • Air flow and AC content were found to be significant for 14% and 19% of the different combinations, respectively, but the differences found in the D% passing were rela- tively small. • Sample mass was found to be significant for 47% of the dif- ferent test combinations. All the sieve sizes of Mix 4 seem Variable Mix 1 Mix 2 Mix 3 Mix 4 Number of Combinations Factor Significant Percentage of Combinations Factor Significant Furnace 6/9 5/9 1/9 0/9 12/36 33 Temperature 0/9 1/9 2/9 3/9 6/36 16 Air flow 2/9 0/9 0/9 3/9 5/36 14 AC content 0/9 4/9 3/9 0/9 7/36 19 Sample mass 0/9 8/9 0/9 9/9 17/36 47 Table 37. Summary of statistically significant test factors on aggregate gradations.

45 to be affected by sample mass. This contrasts with the find- ings for the asphalt correction factors since this particular variable did not affect the asphalt correction factor results. For the gradation analysis, the trend indicates that the higher the sample mass, the lower the percentage passing any particular size. • Although the results show that some of the variables affected the aggregate gradation test results when a p-value of 0.10 was used, in general, the differences in percent- ages passing were typically not practically different when compared to the permitted sieving difference shown in Table 3 (AASHTO T 308). This observation is illustrated in Table 38. This table presents the average gradation of all the combinations used with each mixture. These results are compared to the blank sample (aggregate only) gradations that were determined for each mixture (per AASHTO T 308). The average sample gradations are shown to be very close to the blank sample gradations, Mix # Sieve Size JMF Gradation Blank Sample Gradation (% Passing) Average Gradation (% Passing) Aggregate Correction Factor (%) Mix 1 1/2” 96.0 3/8” 85.2 #4 61.7 #8 48.7 #16 37.1 #30 27.5 #50 17.9 #100 10.1 #200 6.2 Mix 2 1/2” 96.0 3/8” 85.2 #4 61.7 #8 48.9 #16 37.5 #30 28.0 #50 18.5 #100 10.9 #200 7.0 Mix 3 1/2” 95.2 3/8” 88.7 #4 69.8 #8 46.7 #16 33.1 #30 24.5 #50 14.5 #100 6.6 #200 5.0 98.2 89.6 66.3 48.8 37.9 28.2 18.3 10.7 6.5 98.3 87.2 63.3 48.7 38.5 28.9 18.9 11.1 6.9 95.6 86.7 69.6 47.2 33.6 24.4 12.9 6.0 4.3 95.8 88.5 66.1 49.9 39.6 30.0 20.1 12.0 7.3 95.9 85.2 61.1 49.8 39.3 29.7 20.2 12.4 8.1 94.6 86.6 69.2 45.1 31.3 23 13.4 6.9 4.5 −2.4 −1.1 −0.2 1.1 1.7 1.8 1.8 1.3 0.8 −2.4 −2.0 −2.2 1.1 0.8 0.8 1.3 1.3 1.2 −1.0 −0.1 −0.4 −2.1 −2.3 −1.4 0.5 0.9 0.2 Mix 4 1/2” 94.0 3/8” 83.1 #4 56.3 #8 38.2 #16 24.5 #30 15.8 #50 10.1 #100 6.7 #200 5.7 92.7 79.3 55.2 38.4 24.3 15.9 10.3 6.9 5.4 93.6 81.7 56.1 38.2 23.6 15.1 9.9 5.5 3.0 0.9 2.4 0.9 −0.2 −0.7 −0.8 −0.4 −1.4 −2.4 JMF = job mix formula. Table 38. Average aggregate gradation from sensitivity Experiment #1 versus blank specimen gradation.

46 with the exception of #200 sieves for Mixes 1, 2, and 4. Using the average gradation and the blank sample grada- tions, the approximate aggregate correction factors were determined for each sieve size. It is important to keep in mind that, for this evaluation, all the combinations used for each mixture were included. 5.1.2 Experiment 2 – Thermolyne Furnace Only A second experiment was conducted with the Thermolyne unit to evaluate whether the temperature rate affects the test results. In order to conduct this evaluation, one of the heating elements inside the unit was disconnected. For this experi- ment, four variables were included: temperature rate, sample mass, test temperature, and asphalt content. Table 39 presents the factors with their corresponding p-values; p-values that are less than or equal to 0.10 are shown in bold. This table summarizes the results by aggregate type. Table 40 presents the average asphalt correction factors for temperature rate at each level; the last two columns in this table specify whether the results are statistically significant and practically significant. Based on these results, tempera- ture rate seems to have a statistically significant effect only on Mixes 1 and 2. For Mix 1, the average correction factor dif- ference between levels is only -0.05%. For Mix 2, this differ- ence is equal to -0.10%. Therefore, the differences are not large enough to be considered practically significant. In gen- eral, it seems that even when one heating element may not be working properly, the unit still seems to function, and the asphalt correction factors are not greatly affected. The aver- age test time for the test conducted under a slow temperature rate condition was approximately 7 min longer than the test under normal rate conditions. Temperature was found to be the most significant vari- able affecting the test results, which confirms the results from Experiment 1. Asphalt content and sample mass did not sig- nificantly affect the asphalt correction factor results. 5.1.3 Experiment 3 – Troxler Furnace Only The third experiment was conducted to evaluate the effect of burning profiles on the asphalt content correction factors. Changing the burning profile in the Troxler equipment is the equivalent of adjusting the test temperature for the Thermo- lyne unit. For this experiment, three variables were included: burning profile, sample mass, and asphalt content. Table 41 presents Mix # Temperature Rate Test Temperature (°F) Asphalt Content (%) Sample Mass (Grams) Mix 1 0.03 0.00 0.74 0.11 Mix 2 0.00 0.00 0.16 0.49 Mix 3 0.99 0.00 0.95 0.25 Mix 4 0.90 0.00 0.14 0.15 Note: Bold type indicates p-values that are less than or equal to 0.10. Table 39. Significance of asphalt content correction factor test parameters based on p-values – sensitivity study – Experiment 2 by mix. Mix # Levels Average Correction Factor (%) Statistically Significant (Y/N) Practically Significant (Y/N) Mix 1 Slow rate Y N Normal rate Mix 2 Slow rate Y N Normal rate Mix 3 Slow rate N – Normal rate Mix 4 Slow rate N – Normal rate −0.02 −0.07 −0.34 −0.24 0.53 0.53 1.91 1.87 Table 40. Average asphalt content correction factor by mix for temperature rate at each level.

47 the factors with their corresponding p-values; p-values that are less than or equal to 0.10 are shown in bold. Table 42 summarizes the average asphalt correction factors for each significant variable at each level of analysis by mixture. The information presented in this table shows that, for all cases, the results that were found statistically significant were also found practically significant. From the results presented in Table 42, burning profile was found to have a significant effect on the asphalt correction factor, as follows: • For Mixes 1, 3, and 4, Option 2 yielded the highest asphalt correction factor. This option is recommended by the manufacturer for “rich mixes” and is considered a more aggressive profile. • For Mixes 1, 3, and 4, Option 1 resulted in the lowest asphalt correction factor, with values of -0.04%, 0.64%, and 1.27%, respectively. • The change in asphalt correction factor caused by chang- ing the burning profiles was more pronounced for Mix 4. In order to better understand the differences in furnace temperatures for the Troxler unit, the average maximum test temperature results for the three burning profile test combi- nations were calculated based on the test results. For default, Option 1, and Option 2, the average maximum temperatures Mix # Burning Profile Asphalt Content (%) Sample Mass (Grams) Mix 1 0.00 0.11 0.71 Mix 2 0.04 0.22 0.25 Mix 3 0.00 0.00 0.58 Mix 4 0.00 0.01 0.10 Note: Bold type indicates p-values that are less than or equal to 0.10. Table 41. Significance of asphalt content correction factor test parameters based on p-values – sensitivity study – Experiment 3 by mix. Mix # Variable Levels Average Correction Factor (%) Statistically Significant (Y/N) Practically Significant (Y/N) Mix 1 Burning profile Default Y Y Option 1 Option 2 AC content Opt. − 1% N — Opt. + 1% Sample mass 1,500 g N — 2,000 g Mix 2 Burning profile Default N —Option 1 Option 2 AC content Opt. − 1% N — Opt. + 1% Sample mass 1,500 g N — 2,000 g Mix 3 Burning profile Default Y Y Option 1 Option 2 AC content Opt. − 1% Y Y Opt. + 1% Sample mass 1,500 g N — 2,000 g Mix 4 Burning profile Default Y Y Option 1 Option 2 AC content Opt. − 1% Y Y Opt. + 1% Sample mass 1,500 g N — 2,000 g 0.11 −0.04 0.17 0.09 0.06 0.07 0.08 −0.20 −0.24 −0.15 −0.18 −0.21 −0.18 −0.21 0.70 0.64 0.92 0.68 0.83 0.74 0.77 1.59 1.27 2.15 1.52 1.82 1.55 1.68 Note: Opt. = optimum. Table 42. Average asphalt content correction factors by mix for variables at each level – Experiment 3.

48 were 1,010, 965, and 1,050°F (543, 518, and 566°C), respectively. Asphalt content was found to have a significant effect on the asphalt correction factor for Mixes 3 and 4. For both aggre- gates, the higher the asphalt content, the higher the correc- tion factor. This trend was also found for Experiment 1. 5.2 Round-Robin Study The results of the RRS were collected from each laboratory, and the test results for the asphalt content correction factors were analyzed in accordance with ASTM E691. Although the primary purpose of the ASTM standard is to develop a preci- sion statement for a test method, one important part of the procedure is the calculation of consistency of the test results using two statistics: h and k values for the determination of data that may be suspected outliers. The h statistic is an indicator of how one laboratory’s average for a material compares with the average of the other laboratories. The k statistic is an indicator of how one laboratory’s variability for a given set of replicate samples compares with that of all the other laboratories. The h values can be positive or negative, with 0 represent- ing a laboratory average equal to the average of the labora- tory averages. The k values are always positive, with a value of 1 representing the average within-laboratory variability. A k value of greater than 1 represents higher within-laboratory variability when compared to the rest of the laboratories combined, and a k value of less than 1 indicates less within- laboratory variability compared to all the rest of the labo- ratories combined. ASTM E691 recommends critical k and h values with a 0.5% significance level. Following the ASTM procedure, the within-laboratory and between-laboratory standard deviations were calculated for the asphalt correction factor for each material using Equations 3 through 8. For each set of laboratory test results, k and h statis- tics were calculated using Equations 9 and 10, respectively. ∑= Equation 31x x pp ∑ ( )= −1 Equation 421s d px p d x x Equation 5= − ∑= Equation 621s s pr p = − Equation 72 2 2s s s nL x r = + Equation 82 2s s sR L r h d sx Equation 9= k s sr Equation 10= where x– – = average of the laboratory averages, x– = the average of the test results for each laboratory, p = number of laboratories in the RRS, d = deviation for each laboratory, s = standard deviation of the test results for each laboratory, n = number of test results for each laboratory, sx– = standard deviation of laboratory averages, sr = repeatability standard deviation, sL = between-laboratory standard deviation, h = between-laboratory consistency statistic, k = within-laboratory consistency statistic, and sR = reproducibility standard deviation. The k and h statistics for each material are compared with critical values from ASTM E691, which are a function of the number of laboratories and number of replicates. 5.2.1 Asphalt Content Correction Factor Data Analysis to Identify Outliers The test results from the RRS study included four mix- tures and 23 laboratories with 28 ignition units. Three tests were conducted for each mixture. Figure 32 through Figure 35 present the asphalt content correction factor test results for Mixes 1 through 4 for each laboratory. The correction fac- tors were calculated as the difference between the actual and measured asphalt binder content for each specimen. The cor- rection factors for each mixture reported represent the aver- age of the differences between the measured asphalt content and the actual asphalt content. The minimum and maximum correction factors are also reported in these figures. In these figures, the information was grouped by unit type. A total of 17 Thermolyne (TH), three Gilson (GS), and eight Troxler (TX) units were included in the RRS. Within each unit group, the information was also grouped by age of the unit to iden- tify any possible trend (age from left to right in the figures). The complete information about individual test results, lab- oratories, and units is presented in Appendix E. Information includes furnace ticket results (Thermolyne and Troxler) and external weight results (all units). Results presented in Fig- ure 32 through Figure 35 correspond to furnace ticket results for Thermolyne and Troxler units and external weight results for Gilson units. It is important to point out that results from tickets and external weight were significantly different for some labs.

49 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 4- TX La b 21 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler A sp ha lt C on te nt C F Mix 1 Average=0.11 Min. =-0.66 Max.=0.62 Figure 32. Asphalt content average correction factors from RRS results – Mix 1. -0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 4- TX La b 21 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler A sp ha lt C on te nt C F Mix 2 Average=-0.23 Min. =-0.67 Max.=0.18 Figure 33. Asphalt content average correction factors from RRS results – Mix 2.

50 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 4- TX La b 21 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler A sp ha lt C on te nt C F Mix 3 Average=0.92 Min. =0.55 Max.=1.51 Figure 34. Asphalt content average correction factors from RRS results – Mix 3. -2.0 -1.0 0.0 1.0 2.0 3.0 4.0 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 4- TX La b 21 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler A sp ha lt C on te nt C F Mix 4 Average=1.25 Min. =-1.57 Max.=3.58 Figure 35. Asphalt content average correction factors from RRS results – Mix 4.

51 Figure 36 shows the average test time by mixture type for Thermolyne (17 units), Troxler (eight units), and Gilson (three units). Similar to the findings from the sensitivity study, the average test times were lower for the Troxler furnaces for all mixtures. From Figure 32 though Figure 35, the average correction factors for Mixes 1 through 4 were 0.11%, -0.23%, 0.92%, and 1.25%, including all of the laboratory test results. Initial observations from these figures indicate that no trend was identified for furnace type or age of the units. However, some possible outliers were identified as follows. For Mix 1 Lab 4-TX and Lab 21-TX average correction factor test results were -0.91 and 0.62, which are clearly significantly lower and higher, respectively, than the average correction factor of 0.11. For Mix 2 Similar to the trend found for Mix 1, Lab 4-TX and Lab 21-TX correction factors were significantly different from the average correction factor of -0.23. The correction fac- tors found were -0.67 and 0.18 for Lab 4-TX and Lab 21-TX, respectively. For Mix 3 Lab 17-TH, Lab 21-TX, and Lab 23-TX correction factors were 1.37, 1.51, and 1.44, which are higher than the average correction factor of 0.92. For Mix 4 Lab 21-TX and Lab 23-TX test results were also signifi- cantly different from the average of 1.25. For these two lab- oratories, the correction factors were -1.57 and 3.58. The standard deviations for both laboratories were also signifi- cantly high, 0.59 and 0.55, respectively. Based on these results, the extreme outliers identified for Mixes 1, 2, and 4 (for Lab 4-TX, Lab 21-TX, and Lab 23-TX) were eliminated. Although the outlier identified for Mix 3 yielded higher than average results, the differences with the average were not as evident as the differences for the other mixtures. Therefore, it was decided to keep these test results for further analysis. This preliminary elimination of outliers was considered necessary to ensure that these test results would not mask other potential outliers. Following the procedure recommended in ASTM E691, the RRS test results were analyzed to determine any inconsistent data that could be used as additional outliers for the trouble- shooting study. These calculations include the parameters presented in Equations 2 through 9: within-laboratory and between-laboratory standard deviations and also the k and h statistics. The within-laboratory k statistic values by mixture type are presented in Figure 37 through Figure 40. The critical k values recommended by ASTM E691 are 2.23 for Mixes 1, 2, and 4 and 2.22 for Mix 3. These critical values were exceeded only once for Mix 2, by Lab 16-GS, and once for Mix 3, by Lab 17-TH. The k values found were 3.5 and 2.45, respectively. The between-laboratory h statistic values by mixture are presented in Figure 41 through Figure 44. The critical h val- ues recommended by ASTM E691 are 2.62 for Mixes 1, 2, and 4 and 2.63 for Mix 3. These values were exceeded only 0 10 20 30 40 50 60 70 80 Mix 1 Mix 2 Mix 3 Mix 4 Mix 1 Mix 2 Mix 3 Mix 4 Mix 1 Mix 2 Mix 3 Mix 4 Thermolyne Troxler Gilson A ve ra ge T es t T im e (m in ) Figure 36. Average test time by mix type for Thermolyne, Troxler, and Gilson furnaces.

52 0 0.5 1 1.5 2 2.5 3 3.5 4 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler k St a s cs Mix 2 Crical value =2.23 Figure 38. Within-laboratory k values by laboratory and furnace type – Mix 2 (without extreme outliers). Figure 37. Within-laboratory k values by laboratory and furnace type – Mix 1 (without extreme outliers). 0 0.5 1 1.5 2 2.5 3 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler k St a s cs Mix 1 Crical value =2.23

53 0 0.5 1 1.5 2 2.5 3 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 4- TX La b 21 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler k St a s cs Mix 3 Crical value =2.22 Figure 39. Within-laboratory h values by laboratory and furnace type – Mix 3. 0 0.5 1 1.5 2 2.5 3 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 4- TX La b 22 -T X La b 12 -T X Thermolyne Gilson Troxler k St a s cs Mix 4 Crical value =2.23 Figure 40. Within-laboratory k values by laboratory and furnace type – Mix 4 (without extreme outliers).

54 -3 -2 -1 0 1 2 3 4 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler h St a s cs Mix 1 Crical value =2.62 Crical value =-2.62 Figure 41. Between-laboratory h values by laboratory and furnace type – Mix 1 (without extreme outliers). -3 -2 -1 0 1 2 3 4 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler h St a s cs Mix 2 Crical value =2.62 Crical value =-2.62 Figure 42. Between-laboratory h values by laboratory and furnace type – Mix 2 (without extreme outliers).

55 -3 -2 -1 0 1 2 3 4 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 4- TX La b 21 -T X La b 22 -T X La b 12 -T X La b 23 -T X Thermolyne Gilson Troxler h St a s cs Mix 3 Crical value =2.63 Crical value =-2.63 Figure 43. Between-laboratory h values by laboratory and furnace type – Mix 3 (without extreme outliers). -4 -3 -2 -1 0 1 2 3 4 La b 1- TH La b 2- TH La b 3- TH La b 4- TH La b 5- TH La b 6- TH La b 7- TH La b 8- TH La b 9- TH La b 10 -T H La b 11 -T H La b 12 -T H La b 13 -T H La b 14 -T H La b 15 -T H La b 16 -T H La b 17 -T H La b 16 -G S La b 18 -G S La b 19 -G S La b 10 -T X La b 20 -T X La b 17 -T X La b 4- TX La b 22 -T X La b 12 -T X Thermolyne Gilson Troxler h St a s cs Mix 4 Crical value =2.62 Crical value =-2.62 Figure 44. Between-laboratory h values by laboratory and furnace type – Mix 4 (without extreme outliers).

56 once for Mix 4, by Lab 16-GS. For Mix 3 and Lab 21-TX, the h value was 2.45; this number closely approached the critical value but did not exceed it. From the analysis of k and h statistics, the laboratories that had additional outliers are presented in Table 43. The values shown in bold are the values that exceeded the critical k and h values. In addition to the outliers selected based on their signif- icantly higher average correction factors and their k and h statistics, one additional laboratory was selected for further evaluation. Lab 19-GS yielded significantly lower correction factors for Mixes 3 and 4; the average correction factors for Mixes 3 and 4 were 0.55% and 0.46%, respectively, and the correction factor averages from all were 0.92% and 1.25% for Mixes 3 and 4, respectively. Since one of the goals of this study was to find ways to minimize differences in correction factor, it was considered important to investigate what was different with this unit that yielded these low test results. The laboratories selected as outliers based on their average asphalt correction factors and their h and k value statistics were: • Lab 4-TX, • Lab 16-GS, • Lab 17-TH, • Lab 19-GS, • Lab 21-TX, and • Lab 23-TX. These laboratories were visited by NCAT personnel to troubleshoot possible causes of test result differences. This information is presented in Section 5.3. 5.2.2 Precision Statistics for RRS Test Results Although it was not the main purpose of the RRS, precision statements by mixture type per ASTM E691 were calculated. In order to perform this analysis, the procedure recommended in the ASTM standard was conducted, including all test results and recalculating the h and k parameters presented in Equations 9 and 10. The results of this analysis by mixture type are presented in Appendix F. From the analysis of k and h statistics, the labo- ratories that were found to be outliers are presented in Table 44. Table 45 summarizes the average asphalt content, aver- age correction factor, and within- and between-laboratory standard deviations for each mix. This table also includes the acceptable range of two test results from the same labora- tory and from different laboratories. These ranges were cal- culated in accordance with ASTM E691. From this table, it can be observed that for Mixes 1 and 2, within- and between- laboratory standard deviations are relatively similar: 0.089 and 0.074 for within-laboratory, and 0.131 and 0.111 for between-laboratory. These numbers are close to the values recommended in AASHTO T 308: 0.069 and 0.117, respec- tively. As the correction factors increased for Mixes 3 and 4, the standard deviations also increased. The within-laboratory standard deviations for Mixes 3 and 4 were 0.112 and 0.178, and the between-laboratory standard deviations were 0.264 and 0.403. This suggests that although multiple precision statements are not ideal from a practical standpoint, differ- ent precision statements may be needed for aggregates with higher correction factors. Mixes 1 and 2 used the same mix design, and the only dif- ference was that Mix 2 included 1% lime. The results pre- sented in Table 45 show that the addition of lime caused a significant variation in correction factor, from 0.12% with no lime to -0.23% with 1% lime, resulting in a difference of 0.35. As mentioned earlier, one of the concerns with the accu- racy of the ignition test is the effect of lime on the measured correction factor. Table 45 shows that the within-laboratory and between-laboratory statistics are very similar for Mixes 1 and 2. The primary difference between these two mixtures is that lime was used in Mix 2. The data show that Mix 2 actually had slightly less variability in test results than Mix 1. Hence, if Table 43. Outliers based on k and h statistic analysis (without extreme outliers). Lab Mix # Critical k Values k Value Critical h Values h Value Lab16-GS Mix 2, Mix 4 2.23 for Mixes 2 and 4 and 2.22 for Mix 3 3.5, 1.20 2.62 for Mixes 2 and 4 and 2.63 for Mix 3 −0.66, 3.21 Lab 17-TH Mix 3 2.4 1.88 Note: Bold type indicates values that exceeded the critical k and h values. Table 44. Outliers based on k and h statistic analysis. Mix Lab k Value h Value 1 Lab 4-TX Lab 21-TX 3.40 3.20 2 Lab 4-TX Lab 21-TX 4.10 0.40 3 Lab 17-TX 2.32 4 Lab 16-GS Lab 21-TX Lab 23-TX 4.60 3.10 1.10 Critical values 2.22 3.90 1.70 −2.80 2.66 1.69 −4.10 4.30 3.00 2.59 Note: Bold type indicates values that exceed the critical values.

57 lime is accurately added to a mixture at a consistent rate, then the ignition test should be satisfactory for use, at least for low- mass-loss aggregates. No testing was performed with lime in high-mass-loss aggregates, so it is not certain what the effect of that might be. If for some reason the amount of lime varies during construction, this will affect the correction factor and possibly result in an error in asphalt content measurement. It is interesting to note that a limited study conducted by Prowell and Youtcheff (17) reported that the addition of 1% lime changed the asphalt correction factor of the mixture used in their study by 0.37. The authors showed that the mix- ture with no lime had an asphalt correction factor of 0.64% and the one with 1% lime had an asphalt correction factor of 0.27%. As was explained in Section 2.3.7, the lime appears to react with the SO2 formed from the combustion of organic sulfur to generate calcium sulfate. 5.2.3 Comparison of Asphalt Content Correction Factors Using Test Results for Labs with Two Different Furnaces As mentioned in Section 4.2, one criterion for the selection of RRS participants was the availability of two furnaces at the same laboratory. Five of the laboratories selected had two units of different brands. These brands were Thermolyne, Troxler, and Gilson. As part of this study, it was important to assess the test results from different units at the same loca- tion. The instructions to the RRS participants requested that, to minimize variability, the ignition tests be conducted by the same technician. Some state agencies have their own test procedures for asphalt content determination by ignition, and although the procedures are mostly based on AASHTO T 308, they have differences. For example, Florida test procedure FM-5-563 requires the determination of an asphalt correction factor for every mix design, but it does not specifically mention that each ignition furnace should have its own unique asphalt cor- rection factor. Table 46 summarizes the tests results for the laboratories with multiple units by mixture type. The information pre- sented includes unit brand, average correction factor, and stan- dard deviation. These results do not include the test results that were identified as outliers; therefore, some results are missing from Lab 4-TX, Lab 16-GS, and Lab 17-TH. Overall, the test results showed that, for most cases, the correction factors are very different, with larger differences for Mixes 3 and 4. Mix # Actual Asphalt Content (%) Measured Asphalt Content (%) Correction Factor (%) sr sR r R 1 5.2 5.32 0.089 0.131 0.250 0.367 2 5.2 4.97 0.074 0.111 0.207 0.311 3 6.2 7.08 0.112 0.264 0.314 0.740 4 6.1 7.31 0.12 −0.23 0.90 1.21 0.178 0.403 0.499 1.128 AASHTO T 308 0.069 0.117 0.196 0.330 Notes: sr = repeatability standard deviation, sR = reproducibility standard deviation, r = repeatability acceptable range of two test results, and R = reproducibility acceptable range of two test results. Table 45. Precision statement for asphalt binder content by mixture type. Table 46. Average correction factors and standard deviations for laboratories with two furnaces. Lab # Unit Mix 1 Mix 2 Mix 3 Mix 4 Average Correction Factor (%) Standard Deviation Average Correction Factor (%) Standard Deviation Average Correction Factor (%) Standard Deviation Average Correction Factor (%) Standard Deviation Lab 4 TH 0.036 −0.24 0.071 0.62 0.084 0.64 0.185 TX – – – – 0.78 0.154 1.66 0.161 Lab 10 TH 0.104 −0.23 0.029 0.65 0.066 1.06 0.072 TX 0.038 −0.16 0.020 0.91 0.125 1.46 0.067 Lab 12 TH 0.026 −0.05 0.025 0.90 0.067 0.99 0.155 TX 0.023 −0.22 0.038 0.64 0.061 0.88 0.045 Lab 16 TH 0.051 −0.26 0.006 0.76 0.041 1.20 0.275 GS 0.163 – – 0.75 0.143 – – Lab 17 TH 0.119 −0.42 0.051 – – 1.80 0.198 TX 0.08 0.09 0.16 0.34 0.05 0.10 −0.01 0.34 0.13 0.067 −0.19 0.015 0.88 0.099 1.51 0.068

58 It is also interesting to note that higher correction factors for a specific mix and unit do not always translate to higher standard deviations when compared to lower correction factor test results for the same mix but a different unit. 5.3 Troubleshooting Outliers from Round-Robin Study NCAT personnel visited the six laboratories identified as out- liers to explore the possible causes of the differences in their test results. The laboratories were contacted, and visits were scheduled during December 2015 and January 2016. A checklist was developed to assess unit installation setup and operation and maintenance procedures. This checklist is included in Appendix G. During each laboratory visit, NCAT staff conducted additional ignition tests on samples that were prepared at the NCAT laboratory. Mixes 3 and 4 were cho- sen to conduct further investigation. The asphalt content test results for each mixture as measured by the six laboratories are provided in Table 47. The table also shows the average cor- rection factors by mixture for all laboratories. Descriptions of the activities conducted during each visit are presented in the next sections. 5.3.1 Lab 4 Visit NCAT personnel visited Lab 4 from December 2 through 4, 2015. This laboratory contained both Thermolyne and Trox- ler ignition furnaces, shown in Figure 45 and Figure 46, respectively. Table 47. Results for RRS laboratories visited by NCAT personnel. Test Lab Furnace Mix # Optimum Asphalt Content (%) Average Asphalt Content (%) Average Correction Factor (%) Standard Deviation Lab 21 Troxler 1 5.2 5.82 0.159 2 5.2 5.38 0.081 3 6.2 7.71 0.220 4 6.1 4.53 0.594 Lab 17 Thermolyne 1 5.2 5.54 0.119 2 5.2 4.78 0.051 3 6.2 7.57 0.273 4 6.1 7.90 0.198 Troxler 1 5.2 5.33 0.067 2 5.2 5.01 0.015 3 6.2 7.08 0.099 4 6.1 7.61 0.068 Lab 19 Gilson 1 5.2 5.26 0.112 2 5.2 4.90 0.083 3 6.2 6.75 0.110 4 6.1 6.56 0.142 Lab 16 Thermolyne 1 5.2 5.30 0.051 2 5.2 4.94 0.006 3 6.2 6.96 0.040 4 6.1 7.30 0.275 Gilson 1 5.2 5.19 0.163 2 5.2 4.91 0.312 3 6.2 6.95 0.143 4 6.1 8.80 0.198 Lab 4 Thermolyne 1 5.2 5.28 0.036 2 5.2 4.96 0.071 3 6.2 6.82 0.084 4 6.1 6.74 0.185 Troxler 1 5.2 4.54 0.620 2 5.2 4.53 0.775 3 6.2 6.98 0.154 4 6.1 7.76 0.161 Lab 23 Troxler 1 5.2 5.51 0.150 2 5.2 5.15 0.026 3 6.2 7.64 0.208 4 6.1 9.68 0.548 Average all round-robin participants 1 5.2 5.32 0.097 2 5.2 4.97 0.086 3 6.2 7.08 0.197 4 6.1 7.31 0.62 0.18 1.51 −1.57 0.34 −0.42 1.37 1.80 0.13 −0.19 0.88 1.51 0.06 −0.30 0.55 0.46 0.10 −0.26 0.76 1.20 −0.01 −0.29 0.75 2.70 0.08 −0.24 0.62 0.64 −0.66 −0.67 0.78 1.66 0.31 −0.05 1.44 3.58 0.12 −0.23 0.88 1.21 0.345

59 For the Troxler furnace, the results from the RRS for Mixes 1 and 2 were significantly different from the average correction factors for all laboratories (as presented in Table 48). The unit was inspected, and the checklist was reviewed. A review of the original tickets from the RRS results revealed that two of the tests for Mix 1 and one of the results for Mix 2 only ran for 15 min. It was also found that the results from weighing the samples outside the furnace were significantly different from the results reported from the tickets. These results are presented in Table 48. As shown in this table, the difference in the results for Mix 1 was close to 1.0, and for Mix 2 the difference was close to 0.40. When the results obtained from weighing the samples on scales outside the furnace were compared with the average results from the RRS, the results were more in line with the results from the other laboratories. It is likely that something was not reported correctly on the tickets, indicating a test error, which seems to explain the differences seen between the labo- ratory test results and results from other laboratories. 5.3.2 Lab 23 Visit NCAT personnel visited Lab 23 on December 7 and 8, 2015. The Troxler furnace used in this laboratory can be seen in Figure 47. As explained in Section 5.2.1, the average cor- rection factor results for Mix 4 in this laboratory were sig- nificantly higher than the average correction factor from all of the laboratories, and hence, this laboratory was deemed an outlier. The results for Mix 3 were also higher, but did not exceed the critical values. During the laboratory visit, samples Figure 45. Thermolyne ignition furnace in Lab 4. Figure 46. Troxler ignition furnace in Lab 4. Table 48. Lab 4 visit results for Troxler. Mix # Average Lab Asphalt Content from RRS (%) DOT Troxler Average Asphalt Content from RRS (%) DOT Troxler Std. Dev. From RRS Asphalt Content Average (Manually Weighed on Scales Outside the Furnace) Asphalt Content Std. Dev. (Manually Weighed on Scales Outside the Furnace) 1 5.32 4.54 0.620 5.52 0.62 2 4.97 4.53 0.775 4.92 0.13

60 from Mixes 3 and 4 were tested by NCAT to determine pos- sible causes of the high results. Two replicates of each mixture were first tested without any changes to verify the results from the RRS. Upon investiga- tion, it was determined that the factory-programmed default burn profiles for this furnace were different compared to other Troxler furnaces. Most notable was the fact that the idle temperature was 1,562°F (850°C) instead of 1,346°F (730°C). For the Troxler unit in this laboratory, the Option 1 burn profile was found to be exactly the same as the default burn profile. This would cause the samples tested using Option 1 to yield much higher results compared to other laboratories. The typical default and Option 1 burning profile settings compared to the setting found in the Lab 23 furnace can be found in Appendix H. Figure 47. Troxler ignition furnace in Lab 23. Table 49. Laboratory visit results at Lab 23. Mix # Average Lab AC Content from RRS (%) Lab 23 Average AC Content from RRS (%) Lab 23 Std. Dev. from RRS Average AC Content Measured During Lab Visit (%) AC Content Std. Dev. During Lab Visit (Based on Two Tests) Original Profile New Profile Original Profile New Profile 3 7.08 7.64 0.208 7.76 6.85 0.23 0.27 4 7.31 9.68 0.548 9.93 7.06 0.23 0.07 Note: Bold type indicates new measured AC content results. After verifying the results of the RRS, new profiles were created to match the typical burn profiles used in other Troxler furnaces. Samples of each mixture were then tested using these profiles. Table 49 shows the results from NCAT’s testing during the laboratory visit. After creating the new profiles, the measured AC content results decreased signifi- cantly. The new measured AC content results are shown in bold. For Mixes 3 and 4, the average AC content measured before any changes to the burning profiles were 7.76% and 9.93%; after the burning profiles were changed, the average AC content measured was 6.85% and 7.06%, respectively. The use of incorrect burning profiles was found to be the cause of the higher asphalt content test results for the Troxler furnace in this laboratory. 5.3.3 Lab 17 Visit NCAT personnel visited Lab 17 on December 16 and 17, 2015, to perform tests on both ignition furnaces it used to con- duct the RRS testing. The Thermolyne and Troxler furnaces can be seen in Figure 48. Figure 48. Troxler and Thermolyne ignition furnaces in Lab 17.

61 The results from the Thermolyne furnace during the RRS were higher than the mean AC content from the other laboratories for Mix 3. During the laboratory visit, several tests were conducted using samples from Mixes 3 and 4. However, the results from the RRS could not be replicated. The NCAT test results at the time of the visit were lower than the RRS results from this laboratory and more in line with the mean values for all laboratories. These results can be seen in Table 50. It was believed that for this laboratory and unit, the high test results were due to an error dur- ing the RRS testing. Since the results during the laboratory visit were similar to the average RRS results and different from the RRS results for this laboratory, it was believed that some error had occurred during the initial testing, but there appeared to be no issues with the equipment or the equipment setup. Although the test results for the Troxler unit at this labora- tory were not identified as outliers, the tests for Mix 4 showed asphalt content approximately 0.3% higher than that for the average for all laboratories. Several tests were conducted with Mixes 3 and 4 during the NCAT visit. Testing performed dur- ing the laboratory visit showed that the results for Mix 3 were similar to the results from the RRS and the average results from all laboratories. The results for Mix 4 conducted by NCAT were almost identical to the RRS average for all labo- ratories. The results are shown in Table 51. It is not clear if the Lab 17 results were high for Mix 4 or if this was simply ran- dom variation, but the NCAT results were almost identical to the average of all laboratories for Mix 4. It was concluded that there was no problem with the Troxler equipment or with the equipment setup. 5.3.4 Lab 16 Visit NCAT personnel visited Lab 16 on January 6 and 7, 2016. This laboratory used two furnaces for RRS testing. Figure 49 shows the Gilson furnace, and Figure 50 shows the Thermolyne ignition furnace. Testing was conducted using both furnaces during the laboratory visit. When programming the Gilson for a test, several tempera- tures must be set, with the most important being the “pre- heat” temperature and the “burnout” temperature. It is recommended that users consult the manufacturer’s manual before conducting any test. Figure 51 shows the listed settings from the Gilson furnace manual. Table 50. Thermolyne laboratory visit results at Lab 17. Mix # Average Lab- Measured AC Content from RRS (%) Lab 17 Thermolyne Average Measured AC Content from RRS (%) Lab 17 Std. Dev. AC Content from RRS Average AC Content Measured During Lab Visit (%) AC Content Std. Dev. During Lab Visit 3 7.08 7.57 0.273 7.19 0.01 4 7.31 7.90 0.198 7.43 0.37 Table 51. Troxler laboratory visit results at Lab 17. Mix # Average Lab- Measured AC Content from RRS (%) Lab 17 Troxler Average Measured AC Content from RRS (%) Std. Dev. Lab AC Content from RRS Average AC Content Measured During Lab Visit (%) AC Content Std. Dev. During Lab Visit 3 7.08 7.08 0.24 6.95 0.10 4 7.31 7.61 0.33 7.30 0.23 Figure 49. Gilson ignition furnaces in Lab 16.

62 Figure 50. Thermolyne ignition furnace in Lab 16. Figure 51. Settings for Gilson ignition furnace (29). Table 52. Gilson laboratory visit results at Lab 16. Mix # Measured Average Lab AC Content from RRS (%) Lab 16 Gilson Measured Average AC Content from RRS (%) Lab 16 Gilson Std. Dev. from RRS Measured AC Content Average During Lab Visit (%) AC Content Std. Dev. During Lab Visit Average AC Content Measured During Lab Visit (%) AC Content Std. Dev. During Lab Visit Furnace Set to Correct Burnout Temp Furnace Set to Incorrect Burnout Temp 3 7.08 6.95 0.143 6.98 0.02 NA NA 4 7.31 8.80 0.198 7.00 NA* 8.81 0.22 *Only one sample tested at this condition. Since these values are entered into the furnace in the order listed, it was suspected that during the RRS, the technician had only changed the preheat temperature to 482°C, and left the burnout temperature set to 1,000°F (538°C) when testing Mix 4. This would most likely be the cause of the high results from Mix 4. This was confirmed by testing a sample of Mix 4 using a preheat temperature of 900°F (482°C) and a burnout temperature of 1,000°F (538°C). A sample was then tested using 900°F (482°C) for both temperatures to confirm that this would yield lower results. It was only possible to run one replicate at the correct temperatures due to time constraints and the lengthy nature of testing using the Gilson furnace. However, it was believed that the reason for the high results from the RRS had been explained. The tests conducted for Mix 3 were similar to the results obtained from the RRS. Table 52 shows the results of the Gilson testing. This furnace is not typically used for ignition tests at this laboratory, so the technicians had very little experience with the machine. They typically use the Thermolyne for asphalt testing and use the Gilson for cleaning binder containers. One of the two elements in the Gilson furnace was not operational. The staff noted that it had been like this for some time but still seemed to work properly. It seems that the furnace still burned the sample completely. Figure 52 shows the inside of the Gilson furnace with the top element not functioning. Although the results from the Thermolyne furnace were not identified as outliers, tests were conducted during the laboratory visit. In Table 53, it can be seen that these results were consistent with the RRS results. 5.3.5 Lab 19 Visit Lab 19 was visited on January 12 and 13, 2016. Figure 53 shows the Gilson ignition furnace used in this laboratory. During the RRS, this furnace yielded lower than average results for Mixes 3 and 4. Tests were conducted with these two mixtures during the laboratory visit. The input param- eter for the afterburner temperature was set to 1,000°F

63 (538°C) instead of the recommended 1,562°F (850°C) (see Figure 51). Therefore, one test was conducted on Mix 3 with the afterburner temperature still set to 1,000°F (538°C) in order to determine whether the result was similar to the RRS. The resulting asphalt content for this sample was 6.76%, which was almost identical to the RRS average for Lab 19 of 6.75%. The following step was to increase the afterburner tem- perature to 1,562°F (850°C) to determine whether this would have an effect on the results. The average AC content results did increase after this change was made. The results are shown in Table 54. The test conducted during the laboratory visit Figure 52. Inside of Gilson furnace in Lab 16. Figure 53. Gilson ignition furnace in Lab 19. Mix # Average Lab AC Content from RRS (%) Lab 16 Thermolyne Average AC Content from RRS (%) Lab 16 Thermolyne Std. Dev. from RRS Average AC Content Measured During Lab Visit (%) AC Content Std. Dev. During Lab Visit 3 7.08 6.96 0.04 6.99 0.05 4 7.31 7.30 0.28 7.29 0.07 Table 53. Thermolyne laboratory visit results at Lab 16. Table 54. Thermolyne laboratory visit results at Lab 19. Mix # Average Lab- Measured AC Content from RRS (%) Lab 19 Thermolyne Average Measured AC Content from RRS (%) Lab 19 Std. Dev. AC Content from RRS Average AC Content Measured During Lab Visit (%) AC Content Std. Dev. During Lab Visit (Afterburner at 850°C) Afterburner at 538°C Afterburner at 850°C 3 7.08 6.75 0.110 6.76* 6.95 0.05 4 7.31 6.56 0.142 – 6.78 0.07 *Based on one test only.

64 Figure 54. Troxler ignition furnace in Lab 21. Table 55. Lab visit results at Lab 21. Mix # Average Lab AC Content from RRS (%) Lab 21 Thermolyne Average AC Content from RRS (%) Lab 21 Std. Dev. AC Content from RRS Average AC Content Measured During Lab Visit (%) AC Content Std. Dev. During Lab Visit 3 7.08 7.71 0.220 7.12 0.08 4 7.31 4.53* 0.594 7.55 0.33 *Based on two tests only due to furnace malfunction. after correcting the afterburner temperature increased the correction factors for both mixtures by approximately 0.20. These results may explain to some extent the reason for the low correction factors obtained during the RRS. 5.3.6 Lab 21 Visit NCAT personnel visited Lab 21 on January 20, 2016. After the results of the RRS had been submitted, this laboratory decided to replace the Troxler furnace that was used during the RRS. The main reason for this was that rain water had consistently leaked in through the exhaust vent and caused damage to the existing furnace, causing it to periodically mal- function. This problem more than likely explains the variabil- ity in the RRS results from this laboratory. The new Troxler ignition furnace that was used at the time of the laboratory visit is shown in Figure 54. Several tests were conducted on Mixes 3 and 4 using the new furnace. The results are shown in Table 55. With the new unit, the results are more in line with the mean from the RRS results.