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.

27 C h a p t e r 4 The literature review and survey of state DOT and industry practices regarding the use of ignition furnaces conducted in Phase I were used to refine the list of factors that may affect the variability of asphalt content and aggregate gradation correc- tion factors for ignition furnaces. Based on this information, a series of studies were designed, conducted, and analyzed to evaluate different ignition furnaces for a number of factors to determine how these factors affect the results and what improvements could be made to minimize variability when different units are used. Three studies were conducted: • A sensitivity study at the NCAT laboratory, • A round-robin study, and • Troubleshooting of outliers from the round-robin study. A description of each experiment is presented in the fol- lowing sections. 4.1 Sensitivity Study at the NCAT Lab The purpose of the sensitivity study was to evaluate differ- ent factors to determine which of these factors may have the greatest influence on the ignition test results. In this study, the results of the sensitivity study were intended to mini- mize the variability of ignition furnace correction factors. The experimental plan involved the determination of dif- ferent sources of variation, each with the potential to affect the correction factor results. For this purpose, this study fol- lows a procedure recommended in ASTM E1169, “Standard Guide for Conducting Ruggedness Test.” The test procedure requires making systematic changes in each variable and then analyzing the effect of those changes on the test results for each test combination. The procedure recommends a statisti- cally designed experiment involving two levels of each factor (low and high) using a full factorial or a fractional factorial, also known as Plackett-Burman designs. These designs are efficient in evaluating the effect of changes in a variable factor when there is no interaction between the variables. The steps to conduct the study were: • Identify the materials to be included in the analysis; • Identify factors that are considered significant and their levels; • Display the treatment combination, which assigns factors and levels to run; and • Conduct the test and the statistical analysis to determine the effect of factors on the asphalt and aggregate correction factors. An explanation of materials, factors, and treatment com- binations used in this study is presented in the following sections. 4.1.1 Materials and Design Variables 4.1.1.1 Materials The selection of different aggregates based on their approx- imate correction factors was considered to be crucial in this study since it is well known that different aggregates can have correction factors of between 0% and 1%, and even sub- stantially higher for high-loss aggregates such as dolomites. Therefore, aggregates were selected as follows. Aggregate Type. It was decided that selecting the aggre- gate based on correction factors instead of the aggregate type was a better way to target the difference in aggregates. Four aggregates were selected based on their correction fac- tors: 0.0% to 0.5%, 0.0% to 0.5% with 1% lime, 0.5% to 1.0%, and 1.0% to 3.0%. Aggregate 2 corresponded to Aggregate 1 (correction factor of between 0.0% and 0.5%) but with 1% lime added. Based on the literature review conducted as part of Phase I, it was found that lime typically reduced the correc- tion factors. Therefore, its effect needed further investigation. An addition of 1% lime to reduce the stripping susceptibility of asphalt mixtures is a common practice of various agencies. By using this aggregate with 1% lime, additional information Experimental Plan Description

28 on the differences in correction factors in one mix design with and without lime was also explored. The Phase I survey also indicated that most agencies use aggregates with correction factors of less than 0.5% or between 0.5% and 1.0%. However, approximately 10% of the respon- dents indicated that the correction factor for at least one of the aggregates in their state was higher than 1%. With these four aggregates, four mix designs were devel- oped with a 12.5 mm NMAS and a PG 67-22 binder. It was believed that significant additional information would not be gained by using additional NMASs. 4.1.1.2 Design Variables The selection of test factors as variables in the sensitiv- ity study was based on their probable level of significance on the test results. These factors included furnace type, test temperature, sample mass, air flow, and asphalt content. In addition, for the Troxler furnace only, burning profile was selected as an additional variable. For the Thermolyne unit only, rate of change of temperature was selected as an addi- tional variable. A description of the variables to be considered and the reasoning behind their selection are presented in the next paragraphs. Ignition Furnace Type. As expected, results from the sur- vey conducted in Phase I of this study showed that the majority of the furnaces used (93.3%) included internal balances, and Thermolyne and Troxler were the most widely used brands. This study included the evaluation of three different furnaces: a convection unit (Thermolyne), an infrared unit (Troxler), and a Gilson unit with no internal balance. A small percentage of the laboratories reported the use of units with no internal scale, but the proposal specifically requested to evaluate units with external weighing; therefore, the Gilson unit with no internal balance was included. This unit was also a convection unit. Test Temperature. Two test temperatures were evalu- ated: 800 and 1,000°F (427 and 538°C). From the literature and past experience of the research team, it was clear that tem- perature affected the correction factors, but this effect had not been quantified. The standard procedure currently allows the use of a temperature of 900°F (482°C) if the measured asphalt correction factor exceeds 1%. A study conducted in Indiana (23) recommended conducting the test at 800°F (427°C) for high-mass-loss materials, and their results appeared to be promising. It was decided to use 800°F (427°C) and the stan- dard test temperature of 1,000°F (538°C) for this sensitivity study. For the Troxler unit, the temperature could not be manu- ally selected. The unit has the option to select between three profiles: the default profile recommended for most materials; Option 1, which is designed for aggregates such as dolomites (or any mixture with a large correction factor); and Option 2, which is recommended for some mixtures with higher asphalt content, such as stone matrix asphalt and special modifiers. The default option (recommended for most materials) was selected as the equivalent of 1,000°F used for convection units, and Option 1 (for aggregates with high mass loss, such as dolomite) was selected as the equivalent of 800°F used for convection units. Sample Mass. Two sample masses were evaluated for the four mixtures with a 12.5 mm NMAS, 1,500 and 2,000 g. Although the standard recommends a minimum mass based on the NMAS of the mixture, it allows up to 500 additional grams for this particular NMAS, which represents 25% more weight. There was no need to evaluate sample weights outside this requirement, but there was a need to verify that there were no differences in results obtained for the two sample masses selected. Air Flow. Although flue characteristics and configuration may affect the sample burns, their effect will most likely be reflected in the air flow through the furnace. Because of this, air flow was selected as one of the variables that might affect the correction factors. The equipment setup at NCAT had dampers installed in the duct systems. Since it was difficult to control the exact air flow rate, the levels for this variable were selected based on the opening of the damper: full open and 30% open. Two damper settings were selected that would provide two extremes of the air flow recommendations based on lift on the scales (Thermolyne recommendation). According to the manufacturer for the Thermolyne unit used by NCAT (1087 series), the recommended lift load was supposed to be -2.0 to -3.0 g at room temperature. Lift tests were conducted at different damper settings (% open) for the Thermolyne unit, and air flow was also mea- sured near the center of the exhaust for the Thermolyne and Troxler units. The lift tests were conducted with no sample in the furnace by measuring the lift on the scales when the fur- nace was in operation. The results are summarized in Table 18. Based on these results, running the test with the damper 100% open resulted in one extreme within the recommended range (-2.2 g). The other extreme was selected to be outside of the recommended range to represent air flow issues that may be reflected when a rapid drop in the readings (lift and air flow) is observed. Changing the damper settings from 40% to 30% open, as shown in Table 18, caused the lift load to drop 0.4 g; therefore, 30% open was selected as the other level for this variable. It can also be observed that when the damper setting was 30%, the air flow measured at the exhaust for both units also dropped considerably.

29 Asphalt Content. Two asphalt content levels were evalu- ated with each of the four aggregates: optimum asphalt con- tent minus 1% and optimum asphalt content plus 1%. Based on the literature, mixtures with higher binder content should exhibit greater mass changes since higher binder content should lead to a faster increase in the furnace temperature and higher temperatures as the binder ignites, likely resulting in changes in the correction factor (23). Rate of Temperature Increase. Another variable included in this study was the rate of temperature increase (termed “temperature rate” throughout) because a faster increase in the furnace temperature might affect the correction factors. Since it was only possible to control the temperature rate for the convection furnace (by disconnecting a heating element), an experiment was conducted with the convection unit only. Burning Profiles. This variable applied to the Troxler unit only. Since the Troxler furnace offers three different burning profiles as a function of the aggregate/mixture type, it was con- sidered necessary to evaluate the effect of changing the profile on the asphalt correction factor. Table 19 summarizes the factors that were evaluated in the sensitivity study and their corresponding levels. 4.1.2 Experimental Designs for Sensitivity Study As mentioned previously, ASTM E1169 describes partial factorial Plackett-Burman designs that are commonly used in a sensitivity study. Since it was required to evaluate more than two levels per furnace type, a special design had to be developed. The experimental designs for this study included half of a fraction of a factorial. The ASTM standard procedure uses a design that is only 1⁄16th of a fraction of a factorial. As an example, a full factorial for a 27 design in the standard = 128 tests, so the standard only requires 8 tests (1⁄16th of 128) to be conducted with two replicates. Hence, the designs of the experiment developed for this study are much stronger than the design of the experiment in ASTM E1169. A total of three experiments were designed. The details of each experiment are presented in the next sections. 4.1.2.1 Experiment 1 This experiment was considered the main experiment and included five variables: furnace type, test temperature, air flow, sample mass, and asphalt content. The total number of tests conducted can be obtained as fol- lows: 4 materials (mixtures) × 3 furnaces × 2 air flow rates × 2 sample masses × 2 temperatures × 2 asphalt contents = 4 × 3 × 24 = 192 for a full factorial. A half factorial would require 96 tests = 4 × 3 × 24-1 = 96. Replicates were tested, resulting in a total of 192 tests. This is equivalent to conducting the experi- ment presented in Table 20 for each aggregate and furnace unit. The level setting for each factor is indicated by either -1 or 1 for low or high levels, respectively. Each combination was conducted with two replicates. 4.1.2.2 Experiment 2 A second experiment was designed to evaluate the tem- perature rate for the Thermolyne unit. For this experiment, four variables were included: temperature rate, sample mass, test temperature, and asphalt content. The total number of tests conducted can be obtained as follows: 4 materials (mixtures) × 1 furnace × 2 temperature rates × 2 sample masses × 2 test temperatures × 2 asphalt contents = 4 × 24 = 64 tests for a full factorial and 32 tests Damper Scale Lift Thermolyne Exhaust Speed (ft/min) Troxler Exhaust Speed (ft/min) % Open Load (g) Near Center Near Center 100 −2.2 155 200 90 −2.1 150 200 80 −2.1 145 200 70 −1.9 145 200 60 −1.9 145 190 50 −1.5 110 160 40 −1.2 92 100 30 −0.8 65 45 20 −0.5 33 15 10 0 13 10 0 0 10 10 Table 18. Lift test results and exhaust air flow speed at different damper settings. ID Factors Levels A Furnaces Thermolyne, Troxler, Gilson B Air flow 30% damper open, full open C Sample mass 1,500 g, 2,000 g D Test temperature 800 and 1,000°F (427 and 538°C) E AC content Optimum − 1%, optimum + 1% F Temperature rate (Thermolyne only) Slow rate, fast rate G Burning profile (Troxler only) Default, Option 1, Option 2 Table 19. Factors and levels for the asphalt content correction factors sensitivity study conducted on each mixture type.

30 for a half factorial. Since duplicate tests were conducted, this design results in 64 total tests for a half factorial. The experimental design conducted for each aggregate type is presented in Table 21. 4.1.2.3 Experiment 3 This experiment was conducted using only the Troxler infrared unit. The primary purpose of this experiment was to evaluate the effect of varying the burning profiles for the Troxler equipment. The variables for this experiment were burning profile, sample mass, and asphalt content. Since this experiment included one variable at three levels and only two variables at two levels, it was recommended to conduct a full factorial design. The total number of tests to conduct can be obtained as follows: 4 materials (mixtures) × 1 furnace × 3 profiles × 2 sample masses × 2 asphalt contents = 4 × 3 × 22 = 48 tests with 2 replicates = 96 tests. The experimental design conducted for each aggregate type and for each burning profile is presented in Table 22. The complete experimental design matrices are included in Appendix C. 4.1.3 Mix Designs and Sample Preparation Asphalt mixtures were designed using four different aggre- gate blends and one asphalt binder. As explained previously, the four aggregate blends were chosen based on their correction factors instead of aggregate type: 0.0% to 0.5%, 0.0% to 0.5% + 1% lime, 0.5% to 1.0%, and 1.0% to 3.0% for Aggregates 1 through 4, respectively. Table 23 shows the aggre gate blends used in this study. A target gradation was selected for all mixtures to ensure that all of the aggregates to be used for the different mix Combination Number Factor Air Flow Sample Mass Temperature AC Content 1 −1 −1 −1 −1 2 −1 −1 1 1 3 −1 1 −1 1 4 −1 1 1 −1 5 1 −1 −1 1 6 1 −1 1 −1 7 1 1 −1 −1 8 1 1 1 1 Note: −1 = low level, 1 = high level. Table 20. Sensitivity experimental plan conducted for each mixture and furnace type (Experiment 1). Combination Number Factor Temperature Rate Sample Mass Temperature AC Content 1 −1 −1 −1 −1 2 −1 −1 1 1 3 −1 1 −1 1 4 −1 1 1 −1 5 1 −1 −1 1 6 1 −1 1 −1 7 1 1 −1 −1 8 1 1 1 1 Note: −1 = low level, 1 = high level. Table 21. Sensitivity experimental plan conducted for each mixture (Experiment 2). Table 22. Sensitivity experimental plan conducted for each mixture and burning profile – Troxler unit (Experiment 3). Combination Number Factor Sample Mass Asphalt Content 1 −1 −1 2 −1 1 3 1 1 4 1 −1 Note: −1 = low level, 1 = high level.

31 designs would have approximately the same gradation. Once the aggregates were collected, each aggregate type was oven dried and then batched to achieve the desired blend. Design verifications were performed to determine the optimum asphalt content according to AASHTO T 312. All mixtures were 12.5 mm NMAS mixtures using a PG 67-22. The target air void content for the mixtures was 4.0%. Table 24 shows the design gradations and optimum asphalt content deter- mined for each aggregate blend. Also shown is the target blended gradation for the four mixtures. It was not critical that the gradation closely follow the target gradation, but it was used as a general guide. After the optimum asphalt content was determined for each mixture, individual aggregate samples were batched and mixed with the required amount of asphalt binder. “Butter” mixtures at the design asphalt binder content were mixed and discarded prior to preparing any test sample. 4.2 Round-Robin Study at Various Labs The second study conducted as part of the experimen- tal plan was an RRS designed to look at differences between equipment and also look for equipment that provided a cor- rection factor that was significantly different from that pro- vided by other equipment. From the results of the survey conducted as part of Phase I, the research team identified and selected 25 labo- ratories to participate in this study. A total of 19 labora- tories from different DOT agencies and six laboratories from different contractors were selected. Five main factors were considered in the selection of these laboratories: location of the laboratory, brand of the unit, age of the unit, more than one brand available, and agency type (DOT/ contractor). Aggregate/ Mixture Aggregate Description Source Expected Correction Factor Range (%) #1 Limestone and granite Calera, AL – Vulcan Materials 0.0–0.5 #2 Limestone and granite with 1% lime Calera, AL – Vulcan Materials 0.0–0.5 #3 Limestone Barbeau, MI – Payne and Dolan 0.5–1.0 #4 Dolomite Delphi, IN – USA Aggregates 1.0–3.0 Table 23. Aggregate types used for mix designs. Sieve Size, mm Target Gradation Mix #1 Mix #2 Mix #3 Mix #4 % Passing 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 95.0 96.0 96.0 95.2 94.0 85.0 85.2 85.2 88.7 83.1 65.0 61.7 61.7 69.8 56.3 45.0 48.7 48.9 46.7 38.2 — 37.1 37.5 33.1 24.5 25.0 27.5 28.0 24.5 15.8 — 17.9 18.5 14.5 10.1 4.75 2.36 1.18 50.0 37.5 25.0 19.0 12.5 9.5 0.6 0.3 0.15 — 10.1 10.9 6.6 6.7 0.075 [#200] 6.0 6.2 7.0 5.0 5.7 Optimum asphalt content — 5.2 5.2 6.2 6.1 Table 24. Design aggregate structure for mixtures used for NCHRP Project 9-56.

32 Participating Lab Description Agency/Contractor Furnace Brand Furnace Age(Years) DOT Alabama Thermolyne Arizona Thermolyne Arkansas Troxler Florida Thermolyne Indiana Thermolyne Kansas Gilson Louisiana Thermolyne Massachusetts Troxler Minnesota Thermolyne Mississippi Thermolyne Missouri Thermolyne Troxler Montana Troxler New Mexico Gilson South Carolina Thermolyne Ohio Thermolyne, Gilson Tennessee Thermolyne, Troxler Virginia Thermolyne Washington Thermolyne Contractor/research The Miller Group Troxler Pike Industries Thermolyne, Troxler NCAT Troxler, Thermolyne Staker Parson (Utah) Thermolyne Mid-South Paving (Alabama) Thermolyne 15+ 1 15 10 8 17 20 2 5 20 7 7 13 20 6 20 5 2 8 10 2 15 15 15 1 15 8 15 Number of mixtures Four mixtures with Aggregates 1, 2, 3, and 4 at their optimum asphalt content Number of samples sent to each laboratory per unit Four asphalt mix samples per mix type (run three per mix, plus one extra per mix in case rerun needed) Total number of specimens 4 samples × 4 mixtures × (28 units) = 448 samples; 1,500 g each Test AC content determination per AASHTO T 308; sieve analysisper AASHTO T 30 Table 25. Round-robin experimental plan. 4.2.1 Round-Robin Experimental Design Commonly, the primary reason for conducting round- robin testing is to develop a precision statement. However, in this study, the results were used to identify correction factor outliers that needed to be investigated to determine causes for these specific test results being outliers. The RRS used ASTM E691-15, “Standard Practice for Conducting an Inter-laboratory Study to Determine the Precision of a Test Method,” as a guideline to collect and analyze the data. The same four aggregates used for the sensitivity study were also used for this RRS. A total of 464 samples were prepared and sent to the different laboratories selected, calculated as: 4 samples × 4 mixtures × (28 units) = 448 samples; 1,500 g each. Each laboratory was required to conduct the following tests: (1) asphalt content determination per AASHTO T 308, and (2) sieve analysis per AASHTO T 30. Each laboratory received a total of 16 specimens per fur- nace; four samples per mixture were provided to test three replicates. (One additional sample was provided per mixture in case it was needed due to problems during testing.) All of the specimens were prepared at their optimum binder con- tent, and the sample size for each specimen was 1,500 g. Two of the laboratories selected were not able to complete the tests as requested; therefore, the final group of participants included 23 laboratories. Table 25 provides information about the laboratories, igni- tion furnace brands, and approximate ages of the furnaces. Five of these laboratories (Missouri, Ohio, Tennessee, Pike Industries, and NCAT) had multiple furnaces and were able to conduct the testing on both units. This resulted in a total 28 furnaces (18 laboratories with one furnace and five lab- oratories with two furnaces). Additional information in

33 Table 25 includes total number of mixtures, samples sent to each laboratory per unit, number of specimens, and tests that were conducted by each laboratory. Similar to the procedure followed for the sensitivity study, individual specimens were batched at a mass that yielded a sample size of approximately 1,500 g after mixing with the asphalt binder. Each specimen was prepared at its optimum asphalt content. Once the samples were prepared, they were immediately placed in wax-lined boxes for shipping to the RRS laborato- ries. A wax-lined box containing a mixture is shown in Fig- ure 25. Figure 26 displays how the individual mix boxes were packaged to send to each laboratory. Two technicians com- pleted all of the mixing in order to minimize user variabil- ity. One technician mixed all of the samples from Mixtures 1 and 3, while the other technician mixed all of the samples from Mix tures 2 and 4. Each participant laboratory was asked to follow the proce- dure set forth by AASHTO T 308. They were asked to have one technician perform ignition testing using the furnace specified in the survey portion of the study to determine the asphalt con- tent for three replicates for each mixture. The fourth replicate for each mixture could be tested if any unforeseen problems arose during the testing of the first three replicates. Samples from Mixes 1 through 3 were to be tested at 1,000°F (538°C) for laboratories with Thermolyne and Gilson furnaces and the default profile for Troxler furnaces. Samples from Mix 4 were to be tested at 900°F (482°C) for Thermolyne and Gilson furnaces and the Option 1 profile for Troxler furnaces. The samples were to then be tested according to AASHTO T 30 to determine the mechanical analysis of extracted aggregate. Appendix D includes the instructions sent to each laboratory to conduct the tests. 4.3 Troubleshooting Correction Factors That Differ Significantly from Other Correction Factors in Round-Robin Study From the RRS, outliers were identified, and two people were sent to the laboratory to conduct additional testing to determine the reason for the differences in correction factors. Every effort was made to find laboratories that had issues with their correction factors. The research team looked at the significant factors iden- tified from the sensitivity study to help isolate the cause(s) of the differences in the correction factor for the laboratory being evaluated. When possible, adjustments were made in the equipment setup in an attempt to make the correc- tion factors fall in line with correction factors measured with other equipment in other laboratories. It was not pos- sible in all cases to adjust different types of equipment so that they had equal correction factors, but for the same brand of equipment it was hoped that it would be possible to make adjustments in the equipment setup to make the correction factors similar. It was also hoped that the differ- ences could be reduced in correction factors for different equipment types.Figure 25. Mixture boxes for RRS. Figure 26. Several mixtures packaged for sending to RRS lab.