Comparing the Volumetric and Mechanical Properties of Laboratory and Field Specimens of Asphalt Concrete (2016)

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.

28 C H A P T E R 4 4.1 Description of Specimen Preparation Three possible scenarios for production of asphalt mixture specimens were considered in this project: (1) laboratory- mixed–laboratory-compacted specimens (LL) produced dur- ing the design process; (2) plant-mixed–laboratory-compacted specimens (PL) produced for volumetric acceptance and QC testing of plant-produced mix; and (3) plant-mixed–field- compacted specimens (PF), used in testing in situ pavement. The following sections detail the procedures followed to pre- pare the three specimen types. 4.1.1 Laboratory-Mixed–Laboratory- Compacted (LL) Specimens Figure 4-1 depicts the sample collection and fabrication process for LL specimens. The composition of LL specimens is detailed in the JMF (Appendix D) resulting from the labo- ratory design process (AASHTO R 35, “Standard Practice for Superpave Volumetric Design for Asphalt Mixtures”). The following steps were used to make the LL specimens: 1. Aggregates from each stockpile were sampled in accor- dance with ASTM D75, “Standard Practice for Sampling Aggregates.” [Figure 4-1(a) and (b)]; 2. Aggregates were oven dried at 110°C to constant mass; 3. Dry aggregates were separated into individual sieve sizes, [Figure 4-1(c) and (d)] (AASHTO T 27, “Standard Method of Test for Sieve Analysis of Fine and Coarse Aggregates”); 4. Aggregates were blended in accordance with the JMF, [Figure 4-1 (e)] (AASHTO R 35, “Standard Practice for Superpave Volumetric Design for Asphalt Mixtures”); 5. The aggregate blend was heated to production tempera- ture, (AASHTO R 35, “Standard Practice for Superpave Volumetric Design for Asphalt Mixtures”); 6. Liquid asphalt binder was mixed with the heated aggregate blend in accordance with the JMF, [Figure 4-1 (f) and (g)] (AASHTO R 35, “Standard Practice for Superpave Volu- metric Design for Asphalt Mixtures”); 7. The resulting mixture was put in an oven at the produc- tion temperature (which varied with asphalt binder per- formance grade) for short-term aging and volumetric stabilization in accordance with AASHTO R 30, “Stan- dard Practice for Mixture Conditioning of Hot Mix Asphalt (HMA)”; 8. Samples were prepared for compaction and testing size requirements in accordance with AASHTO R 47, “Standard Practice for Reducing Samples of Hot Mix Asphalt (HMA) to Testing Size”; and 9. The loose mixture was compacted into specimens using the Superpave gyratory compactor (SGC) to meet testing protocols. [Figure 4-1 (h), (i), and (j)] (AASHTO T 312, “Standard Method of Test for Preparing and Determin- ing the Density of Hot-Mix Asphalt (HMA) Specimens by Means of the Superpave Gyratory Compactor”). 4.1.2 Plant-Mixed–Laboratory-Compacted (PL) Specimens Figure 4-2 depicts the sample collection and fabrication process for PL specimens. The PL samples were composed of asphalt mixture collected from the truck in accordance with ASTM D979, “Standard Practice for Sampling Bituminous Paving Mixtures.” The mix constituents of each PL specimen are detailed in the mix JMF (Appendix D). The following steps were used to make the PL specimens: 1. Samples were fabricated by collecting loose mixture from the truck according to state protocol and ASTM D979, “Standard Practice for Sampling Bituminous Paving Mix- tures,” [Figure 4-2 (c) and (d)]; 2. Loose mixture was split into required weight size in accor- dance with AASHTO R 47, “Standard Practice for Reduc- ing Samples of Hot Mix Asphalt (HMA) to Testing Size,” [Figure 4-2 (e)]; Method

29 Figure 4-1. Laboratory-mixed–laboratory-compacted (LL) specimen fabrication. (a) Aggregate Stockpiles (b) Stockpile Sampling (c) Sieve Set (d) Fractionated Aggregate Samples (e) Blended Aggregate (f) Liquid AC mixing with Heated Aggregate Blend (g) LL Mixture (h) Reducing into specimen weight requirements (i) Superpave Gyratory Compactor (j) Completed LL Specimens Figure 4-2. Plant-mixed–laboratory-compacted (PL) specimen fabrication. (a) Asphalt Production Facility (b) Silo Loading Truck (c) Mixture Sampling (d) Mixture Sampling (e) Reducing mixture (f) Checking temperature (g) Superpave Gyratory Compactor (h) Completed PL Specimens

30 3. The mixture was put in the oven and brought to com- paction temperature (typically in less than 45 minutes) (Figure 4-2 (f)); and 4. The mixture was compacted using the SGC in accordance with AASHTO T 312, “Standard Method of Test for Pre- paring and Determining the Density of Hot-Mix Asphalt (HMA) Specimens by Means of the Superpave Gyratory Compactor.” (Figure 4-2 (g) and (h)). In some cases, reheat- ing of the specimens was required to evaluate the effect of time delay in specimen fabrication. Additional 5-gallon buckets of loose mixture were sampled from the truck and stored at room temperature for 3 days. Mixture from the buckets was then reheated to compaction temperature (typically for 1 hour) and specimens were prepared. This reheating was done to model the asphalt absorption into the aggregate that typically occurs when samples are taken from a project, stored, and reheated before conducting QA testing. This is different from holding the sample at an ele- vated temperature to artificially age the mixture. Other than the short-term aging used to prepare the samples for speci- men fabrication, possible effects of long-term aging were not evaluated in this study. 4.1.3 Plant-Mixed–Field-Compacted (PF) Specimens Figure 4-3 depicts the construction and sample collection process for PF specimens. The PF samples consisted of cores collected after placement and compaction of the asphalt mixture. The cores were trimmed to ensure that only the mix- ture of interest was obtained (i.e., without the underlying layers). Each core was then trimmed to the required speci- men size for testing. 4.2 Volumetric Tests This section describes how the volumetric properties iden- tified in the test factorial were determined. 4.2.1 Aggregate Gradation The aggregate gradation was determined in accordance with AASHTO T 27, “Sieve Analysis of Fine and Coarse Aggregates.” The aggregate gradation represents the par- ticle size distribution of the aggregates in the mixtures. 4.2.2 Aggregate Bulk Specific Gravity (Gsb) and Absorption The blended aggregate specific gravity and water absorption were determined in accordance with AASHTO T 84, “Specific Gravity and Absorption of Fine Aggregate” and AASHTO T 85, “Specific Gravity and Absorption of Coarse Aggregate.” The bulk specific gravity represents the ratio of the mass in air of a unit volume of a material (including both permeable and impermeable voids) at a standard temperature to the mass in air of an equal volume of water at the same temperature. Equation 5 presents the mathematical computation for deter- mining aggregate bulk specific gravity (Gsb). (a) Truck Loading into Material Transfer Vehicle (b) Material Transfer Vehicle Moving Mixture to Paver (c) Pavement Mat Behind Paver (d) Roller Compacting Mat (e) Finished Pavement Surface (f) Roadway Core Figure 4-3. Plant-mixed–field-compacted (PF) specimen fabrication.

31 G Aggregate Oven Dry Weight Aggregate SSD weight Aggregate SubmergedWeight (5)sb water = −  γ When the total aggregate consists of separate fractions of coarse aggregate, fine aggregate, and mineral filler, all having different specific gravities, the bulk specific gravity for the aggregate blend is calculated. Equation 6 is used to calculate the specific gravity of an aggregate blend: (6) 1 2 1 1 2 2 G P P P P G P G P G blend n n n   = + + + + + + where Gblend = average specific gravity; P1, P2, . . . Pn = weight percentages of fraction 1, 2, . . . , n; and G1, G2, . . . Gn = specific gravity values for fraction 1, 2, . . . , n. Additionally, the blend absorption is computed using Equation 7: (7)1 1 2 2Absorption P A P A P Ablend n n= + + + where, Absorptionblend = average absorption; P1, P2, . . . Pn = weight percentages of fractions 1, 2, . . . , n; and A1, A2, . . . An = absorption percentages for fractions 1, 2, . . . , n. 4.2.3 Mixture Bulk Specific Gravity (Gmb) The mixture bulk specific gravity was determined in accor- dance with AASHTO T 166, “Bulk Specific Gravity of Com- pacted Asphalt Mixtures using Saturated Surface-Dry Specimens (SSD).” This parameter was used to determine weight per unit volume of the compacted mixture. It was very important to measure Gmb as accurately as possible, given that it is used to convert weight measurements to volumes. Any small errors in Gmb will be reflected in significant volume errors, which may be undetected. In addition, Gmb was required for volu- metric evaluation and determination of mixture density in accordance with AASHTO T 269, “Standard Method of Test for Percent Air Voids in Compacted Dense and Open Asphalt Mixtures.” Equation 8 presents the mathematical computa- tion for determining mixture bulk specific gravity (Gmb): G Specimen Oven Dry Weight Specimen SSD weight Specimen Submerged Weight (8)mb water = −  γ 4.2.4 Mixture Maximum Specific Gravity (Gmm) This parameter was measured experimentally using the test procedure described in AASHTO T 209, “Theoretical Maximum Specific Gravity and Density of Bituminous Pav- ing Mixtures.” The theoretical maximum specific gravity, or theoretical maximum density, is the density of an asphalt con- crete mix without air voids, or the highest possible density of the mix. The theoretical maximum specific gravity was used for calculating volumetric parameters. Equation 9 is used for determining mixture maximum specific gravity (Gmm): G Dry Weight Dry Weight Pycnometer Calibration Specimen, Pycnometer, and Water (9)mm water = + −  γ 4.2.5 Asphalt Binder Content (AC) The asphalt binder content of the mixtures was determined in accordance with AASHTO T 164, Method B, “Standard Method of Test for Quantitative Extraction of Asphalt Binder from Hot Mix Asphalt (HMA).” Method B describes the pro- cedure for quantitative extraction by use of a reflux apparatus. Solvent extraction was selected due to its higher repeatability and accuracy when compared to other extraction methods. Solvent extraction uses a chemical solvent (trichloroethylene [TCE]) to separate asphalt binder from the aggregate. The weight of the asphalt removed is determined and the asphalt binder content is computed. 4.3 Mechanical Tests This section describes how the mechanical properties iden- tified in the test factorial were determined. 4.3.1 Loaded-Wheel Test (LWT) This test was conducted according to AASHTO T 324, “Standard Method of Test for Hamburg Wheel-Track Test- ing of Compacted Hot Mix Asphalt (HMA).” This device was manufactured by PMW, Inc., of Salina, KS. The test applies a repetitive load on gyratory specimens compacted to 7 ± 1.0% air voids that have a diameter of 150 mm and a thickness of 40 mm. This test is considered a torture test that produces damage by rolling a 703-N stainless steel wheel across the surface of a compacted gyratory sample, submerged in 50°C water for 20,000 passes at 52 passes a minute. Four states (Arkansas, Ohio, Texas, and Utah) have implemented rut- ting performance criteria based on the Hamburg type wheel tracking test. Current research has shown that, for Louisiana, LWT-measured rut depths of 10 mm and 6 mm can be used

32 as performance targets for low and high traffic, respectively (Kim et al. 2015). Other states, such as Texas, allow up to 12.5 mm of rut depth after a minimum number of passes based on the performance grade of the binder. The rut depths at 1,000; 5,000; and 20,000 cycles were measured and used in the analysis. The stripping inflection point (SIP) was also deter- mined from this test and used in the analysis where applicable. A standard 50°C testing temperature was used for all mixtures studied in order to combine the mixture test results for meta- analysis. LTPPBind software was used to verify that the high temperature for the mixture was greater than 50°C. 4.3.2 Axial Dynamic Modulus (E*) This test was conducted in accordance with AASHTO T 342, “Standard Method of Test for Determining Dynamic Modu- lus of Hot-Mix Asphalt Concrete Mixtures,” by applying a uniaxial sinusoidal (i.e., haversine) compressive stress to an unconfined HMA cylindrical test specimen. The haversine compressive stress was applied on each sample to achieve a target vertical strain level of 100 microns in an unconfined test mode. The dynamic modulus is mathematically defined as the maximum (i.e., peak) dynamic stress (s0) divided by the peak recoverable axial strain (e0): (10) 0 0 E = σ ε Following the AASHTO T 342 testing protocol, samples were tested at temperatures of -10, 4.4, 20, 37.8, and 54.4°C and at loading frequencies of 0.1, 0.5, 1.0, 5, 10, and 25 Hz at each temperature for the development of master curves for use in pavement response and performance analysis. 4.3.3 IDT Dynamic Modulus (IDT E*) IDT dynamic modulus of the mixtures was measured accord- ing to the draft test procedure proposed by Kim et al. (2004), “Dynamic Modulus Testing of Asphalt Concrete in Indirect Tension Mode.” This test was conducted by applying a sinu- soidal compressive stress to the diametric axis of an uncon- fined cylindrical HMA test specimen. Dynamic modulus tests were conducted at temperatures of -10, 10, and 35°C and at loading frequencies 0.1, 0.5, 1.0, 5, and 10 Hz at each temperature for the development of master curves. The com- pressive stress was applied on each sample to achieve target strain levels (40–60 horizontal microstrain and <100 vertical microstrain) in the linear viscoelastic region. Equation 11 pres- ents the mathematical relationship between load and deforma- tion in the indirect tension-loading mode: 2 (11) 0 1 2 2 1 2 0 2 0 E P ad V U  = pi β γ − β γ γ − β where P0 = Peak-to-peak load, N; a = loading strip width, m; d = thickness of specimen, m; V0 = peak-to-peak vertical deformation, m; U0 = peak-to-peak horizontal deformation, m; and g1, g2, b1, and b2 = geometric constants. The geometric constants are functions of gauge length, specimen diameter, and loading strip width. A loading strip of 19.0 mm width is required when testing 150-mm-diameter specimens (AASHTO T 322/ASTM D4123). Table 4-1 presents the coefficients derived and used in this research. Samples were first compacted, using a Superpave gyratory compactor, to a 75-mm height by 150-mm diameter and then cut to the test specimen dimensions of a 38-mm height by 150-mm diameter. Laboratory specimens were compacted to the same air void levels measured in PF cores immediately following construc- tion (~ 7 to 8%). Triplicates were tested for each specimen type. 4.4 Statistical Analyses Statistical Analysis Software (SAS) version 9.2 was used to determine the statistical significance of the comparison between specimen types. An analysis of variance (ANOVA) with a significance level of a = 0.05 was used to determine the statistical significance. Within ANOVA, individual pair-wise property comparisons (i.e., PL vs. LL, PL vs. PF, and LL vs. PF) were conducted using Duncan’s Multiple Comparison Test (MCT) (Freund and Wilson 1997). Triplicate specimens were Gauge Length, mm Loading Strip Width, mm Specimen Diameter, mm β1 β2 γ1 γ2 38.1 19.0 150 -0.0147 -0.0047 0.0043 0.0136 50.8 19.0 150 -0.0199 -0.0062 0.0054 0.0173 76.2 19.0 150 -0.0317 -0.0091 0.0069 0.0229 Table 4-1. IDT dynamic modulus geometric constants.

33 evaluated for each specimen type. Additionally, an analysis of covariance (ANCOVA) was conducted with the guidance of a statistician. The ANCOVA allows the individual process- based factors from the mixtures to be used in determining the main effects. 4.4.1 Statistical Basics Four main sources of variability contribute to the mea- sured overall variation defined in Freund and Wilson (1997). The first type is “inherent variation” (i.e., random variation due to the material itself that cannot be removed). The second type is “sampling and testing variation,” which includes vari- ability due to sampling technique, test procedure, operator, equipment, and calibration. The third type of variation is “within-batch variation,” or the variability observed between samples taken from the same batch. The fourth type of vari- ation is “batch-to-batch variation,” or the variability observed between batches (i.e., from one batch to another). The most widely used measure of variability in the asphalt pavement practice is the standard deviation (St Dev or s), defined as follows (Freund and Wilson 1997): 1 (12) 2 s x x n ∑( ) = − − where x = the individual values of the measured (or response) variable; x– = the sample mean (or sample average); and n = the sample size. The sample standard deviation (with n–1 degrees of free- dom) measures the square root of the sum of the squared devia- tions of the individual observations (or measurements) from the sample average. The variability in mechanical properties of an asphalt mixture is often expressed in terms of the coef- ficient of variation (COV). The COV is a normalized mea- sure of dispersion of a probability distribution. It is defined as follows: COV % 100 (13) s x ( ) = where s = the sample standard deviation and x– = the sample mean. The COV is dimensionless and measures variability rela- tive to the sample mean without considering the units used to define the sample mean and standard deviation. However, when the mean is close to zero, the COV becomes very sensi- tive to small changes in the mean. The COV is used to mea- sure the variability of test results when the standard deviation (testing error) increases in proportion to the magnitude of the result. 4.4.2 Analysis of Variance Statistical ANOVA is used to determine whether the means of response variables measured on two or more populations are statistically equivalent. The null hypothesis is that the popula- tion means (of the response variables) are statistically equiva- lent; the alternate hypothesis is that the population means are not statistically equivalent. Assuming that the response vari- ables are normally distributed and that the variances are sta- tistically equivalent for all populations, the test statistic MSTR/ MSE follows the non-central F distribution (see Equations 14 and 15 for the definitions of MSTR and MSE). When the null hypothesis is true, the non-centrality parameter is zero, causing the test statistic to follow the central F distribution. Therefore, large values of the test statistic (which result in small p-values) support the alternate hypothesis, while small values (which result in large p-values) support the null hypothesis. SAS ver- sion 9.2 was used to determine the statistical significance of the comparison of specimen types. MSTR df (14) 2 1n x xi i∑ ( )= − ÷ MSE 1 df (15)2 2n si i∑( )= − ÷ where MSTR = Mean square treatment; MSE = Mean standard error; x–i = the sample average for group (or population) i; x–– = the overall average of all observations taken; and df = degrees of freedom. If the null hypothesis is rejected, then the conclusion is that all population means are not statistically equivalent. If the means are concluded to be not statistically equivalent, the next step is to determine which of the population means are equivalent and which are different, at least on a pair-wise basis. Several multiple comparison tests are available for evaluating individual pairs evaluated under the ANOVA procedure. The Duncan multiple-range test was used in this study at a level of significance of 0.05. 4.4.3 Precision Limits ASTM C802, “Standard Practice for Conducting an Inter- laboratory Test Program to Determine the Precision of Test

34 Methods for Construction Materials,” defines single-operator precision (also known as repeatability) as “an estimate of the difference that may be expected between duplicate measure- ments made on the same material in the same laboratory by the same operator using the same apparatus within a time span of a few days.” On the other hand, multi-laboratory precision (also known as reproducibility) is “an estimate of the difference that may be expected between measurements made on the same material in two different laboratories.” (ASTM C802, pg. 3). VMA and voids filled with asphalt are calculated properties whose precision depends on the measurement precision of the aggregate bulk specific gravity and aggregate effective specific gravity. Similarly, air voids (AV) is a calculated property whose precision depends on the measured bulk specific gravity of the compacted mixture (Gmb) and the maximum theoretical specific gravity of the mixture (Gmm). ASTM D4460, “Stan- dard Practice for Calculating Precision Limits Where Values Are Calculated from Other Test Methods,” presents methods to estimate precision limits for properties that are calculated. If a property involves the addition or subtraction of test results from two other standards, the standard deviation on which precision limits should be set is calculated from Equation 16: (16)2 2x y x yσ = σ + σ± where sx±y = standard deviation for determining precision limits of a test result for a new standard based on either an addition or subtraction of test results from two other standards; sx = standard deviation from precision statement of one of the standards on which new standard is based; and sy = standard deviation from precision statement of other standard on which new standard is based. If a property involves the multiplication of test results from two other standards, the standard deviation on which preci- sion limits should be set is calculated from Equation 17: (17)2 2 2 2y xxy x yσ = σ + σ where sxy = standard deviation for determining precision limits of a test result for a new standard based on the products of two other test results from two other standards; y– = mean of average value of Y variable; and x– = mean of average of X variable. If a property involves the division of test results from two other standards such as air voids, the standard deviation on which precision limits should be set is calculated from Equation 18: (18) 2 2 2 2 4 y x y x y x y σ = σ + σ where sx/y = standard deviation for determining precision limits of a test result for a new standard based on the quo- tient of two other test results from two other stan- dards; and all other terms as previously defined. 4.4.4 Descriptive Statistics and Data Quality The mean, standard deviation, and coefficient of varia- tion were determined for each data set (i.e., mixture) gener- ated from the experimental plan. Three replicates within each specimen type for each property were measured and, given that split samples were obtained, replicates were assumed to be from the same population. ASTM C670, “Standard Practice for Preparing Precision and Bias Statements for Test Methods for Construction Materials,” defines the acceptable differ- ence between two test results (d2s) as the difference between two individual test results that would be equaled or exceeded in only one case in 20 under normal and correct operation of the method. This d2s value is computed by multiplying the appropriate standard deviation by 2√2 (equal to 2.8). In cases where more than two test results are available, the standard deviation is multiplied by a multiplier correspond- ing to the number of test results, given in Table 1 of ASTM C670 (reproduced herein as Table 4-2. An example data qual- ity evaluation for mixture maximum specific gravity data is shown in Table 4-3. The standard deviation reported from the experiment for three replicates is not directly comparable to the standard No. of Test Results Multiplier of (1s) or (1s%) for Maximum Acceptable Range^ 2 2.8 3 3.3 4 3.6 5 3.9 6 4.0 7 4.2 8 4.3 9 4.4 10 4.5 ^ Values were obtained from Table A7 of “Order Statistics and Their Use in Testing and Estimation,” Vol 1, by Leon Harter, Aerospace Research Laboratories, United States Air Force Table 4-2. ASTM C670 maximum acceptable range.

35 deviation reported by the corresponding AASHTO standard test method. The standard deviation reported by the AASHTO method is calculated for the entire population from a large number of replicates (e.g., nGmm = 626, nAC = 308, nAV = 654). Thus, it should not be expected that the standard deviation of the data set with n=3 would match that of the population. However, the standard deviations calculated for the three rep- licates were often lower than the ones reported by AASHTO, which indicates good control of the experiment. 4.4.5 Individual Mixture Analyses SAS version 9.2 was used to determine the statistical sig- nificance of the comparison of specimen types. A t-test with a significance level of a = 0.05 was used for comparing the means when only two groupings (i.e., PL vs. LL only) were available. However, most of the comparisons had more than two groupings (i.e., PL vs. LL vs. PF). For these comparisons, an ANOVA with a significance level of a = 0.05 was used to determine the statistical significance. Within ANOVA, indi- vidual pair-wise comparisons (i.e., PL vs. LL, PL vs. PF, and LL vs. PF) were conducted using Duncan’s MCT. 4.5 Delta Analyses The term delta, D, is used to identify the difference between the mean values of two specimen types (LL, PL, and PF) of any given parameter (e.g., AV, Gmm, Rut Depth, Modulus). For example, Equation 19 represents the mathematical relation- ship for calculation of the delta of rut depth between LL and PL specimens within a mixture: (19), - , ,Mean MeanRut Depth PL LL Rut Depth PL Rut Depth LL∆ = − Once the D values for each mixture were determined, addi- tional analyses were conducted to determine which factors had the greatest effect on the differences between specimen types. Meta-analysis was conducted to evaluate the effects of process-based based factors on the magnitude of the differ- ences among specimen types. Specifically, the ANCOVA was conducted with the guidance of a statistician. The ANCOVA allows the individual process-based factors from the mix- tures to be used in determining the main effects. This dif- fers from the original analysis developed in the experimental factorial, which could not be used due to inability to collect the entire factorial. The original factorial required categori- cal evaluation of the process-based factors (i.e., high and low). In the ANCOVA, the numerical values associated with each process-based factor were incorporated into the analy- sis (e.g., absorption = 1.7%). The analysis was conducted for the differences of properties measured among LL, PL, and PF specimens of the evaluated mixtures. For the meta-analysis, all plant-produced-laboratory-compacted specimens were designated as PL. The meta-data considered whether or not the sample was reheated. Table 4-4 presents an example of the format of the data input for the asphalt binder content. As shown in the table, each mixture evaluated was treated as a replicate in each property comparison (i.e., LL-PF, LL-PL, and PL-PF). This means that each comparison has 11 observa- tions with 10 degrees of freedom available for the evaluation. Each specimen comparison was performed individually to determine which factors had a statistically significant effect on the considered property (i.e., volumetric and mechanical). The level of significance used in the analysis was a = 0.05. Table 4-5 presents an example of the ANCOVA for the dif- ference in asphalt binder content among specimen types. A p-value less than 0.05 indicates a statistically significant rela- tionship. As shown in the table, the use of baghouse fines return had a statistically significant effect on the difference between laboratory-prepared specimens as compared to plant- produced specimens. This is as expected, especially if baghouse fines are not used during laboratory mixture design. The effect of aggregate absorption was marginal for the LL comparisons. There were no statistically significant process-based factors for the PL-PF comparisons. This seems reasonable for AC content because both PL and PF specimens are processed through the plant. 4.6 Pavement Performance Prediction AASHTOWare Pavement ME Design software was used to evaluate the effects of the measured mechanical properties (i.e., E*) for the three specimen types (LL, PL, PF) on the Test Results Computed AASHTO T 209 limit Gmm data n Max - Min St dev (from T 209) Multiplier (from ASTM C670) Acceptable range 2.508 2.514 2.524 3 0.016 0.0051 3.3 0.017* *Acceptable Range = 0.0051 x 3.3 = 0.017 Table 4-3. Example of data quality criterion applied to mixture maximum specific gravity data.

36 predicted performance for four pavement structures. Three structures representing typical pavements used in Louisiana were used for three traffic levels (low, medium, and high). The fourth pavement structure, adopted from a research study conducted in North Carolina and published by Under- wood et al. (2011), represented an actual pavement in service in North Carolina. Figure 4-4 depicts the pavement struc- tures evaluated in this study. The layer of interest is the HMA layer. The mechanistic-empirical analysis was conducted by altering the material properties of the HMA layer, based on the results of the experimental program; Level 1 analysis was used. All other layer properties were kept constant. 4.6.1 Design Inputs A pavement structure was designed as a new flexible pave- ment with a service life of 20 years; given that results were compared relatively, default calibration factors were used in the analysis. The national default value available in Pavement ME Design for the initial international roughness index (IRI) Comparison Data Process-Based Factors Specimen Comparison Mixture ID Replicate Δ, AC - % Baghouse Reheat Aggregate Absorpon - % Aggregate Hardness - % Stockpile Moisture Content - % Lab-mixed–Lab compacted— Plant-mixed–Plant compacted Mix1 1 -0.25 No No 1.7 38 4.8 Mix2 2 0.00 Yes Yes 1.6 50 3.3 Mix3 3 0.00 Yes Yes 2.1 14 5.0 Mix4 4 -0.30 No Yes 0.8 17 3.5 Mix5 5 0.06 Yes No 1.2 22 5.0 Mix6 6 0.29 Yes No 0.7 22 5.4 Mix7 7 0.04 Yes No 1.3 18 5.4 Mix8 8 0.20 Yes No 0.5 15 4.5 Mix9 9 0.30 Yes Yes 0.5 15 4.0 Mix10 10 -0.20 Yes No 2.8 37 2.8 Lab-mixed–Lab compacted— Plant-mixed–Plant compacted Mix1 1 -0.11 No No 1.7 38 4.8 Mix2 2 0.30 Yes Yes 1.6 50 3.3 Mix3 3 0.10 Yes Yes 2.1 14 5.0 Mix4 4 -0.70 No Yes 0.8 17 3.5 Mix5 5 0.11 Yes No 1.2 22 5.0 Mix6 6 0.07 Yes No 0.7 22 5.4 Mix7 7 -0.03 Yes No 1.3 18 5.4 Mix8 8 0.20 Yes No 0.5 15 4.5 Mix9 9 0.40 Yes Yes 0.5 15 4.0 Mix10 10 -0.10 Yes No 2.8 37 2.8 Plant-mixed–Plant compacted—Plant- mixed–Field- compacted Mix1 1 -0.14 No No 1.7 38 4.8 Mix2 2 -0.30 Yes Yes 1.6 50 3.3 Mix3 3 -0.10 Yes Yes 2.1 14 5.0 Mix4 4 0.40 No Yes 0.8 17 3.5 Mix5 5 -0.05 Yes No 1.2 22 5.0 Mix6 6 0.22 Yes No 0.7 22 5.4 Mix7 7 0.07 Yes No 1.3 18 5.4 Mix8 8 0.00 Yes No 0.5 15 4.5 Mix9 9 0.10 Yes Yes 0.5 15 4.0 Mix10 10 -0.10 Yes No 2.8 37 2.8 Table 4-4. ANCOVA data set example.

37 was used in the analysis. However, values consistent with the Louisiana Pavement Management System (PMS) failure lim- its were used for terminal IRI and total permanent deforma- tion. Louisiana PMS uses index values to describe pavement distress limits. In order to use these limits in Pavement ME Design, the index values were converted to the appropriate units. Louisiana Department of Transportation and Develop- ment (LADOTD) provided conversion equations for IRI and rutting as well as trigger values for rehabilitation. The values used in this study are given in Table 4-6. The national default reliability level of 90% for interstate and primary routes was used in the analysis. In addition, analyses were conducted at a reliability level of 50%, which more closely models typical pavement distresses. 4.6.2 Traffic Average annual daily traffic (AADT) values for multiple traffic classifications, as well as truck factors and distribu- tion for vehicle classes 1 to 13, were provided by LADOTD. Given that Pavement ME Design only supports truck classes 4 to 13, vehicle classes 1 to 3 were not considered, and the LADOTD vehicle class distributions were adjusted to consider only classes 4 to 13. Monthly distribution data were obtained from previous research (Ishak et al. 2009). The national default values from LTPP data for hourly distribution and growth factor were used. Table 4-6 shows the average daily truck traf- fic (ADTT) values associated with the traffic levels evaluated in this study. 4.6.3 Climate Climatic data were obtained from Pavement ME Design climate database for the city of Baton Rouge, LA (NCHRP Project 1-37A). One hundred and sixteen months of data were available for the selected location. The average water table depth was assumed to be 2.1 m. The water table depth determined via Equation 20 estimates the water table based a) High Traffic b) Medium Traffic c) Low Traffic d) Underwood-Low (Underwood et al., 2011) Figure 4-4. Typical pavement designs considered in the performance analysis. Comparison Process-Based Factor F Value p-value Design (LL) - Construction (PF) Baghouse 15.77 0.0165 Reheat 0.07 0.8111 Absorption 7.46 0.0524 Hardness 0.42 0.5538 Moisture 2.81 0.1689 Design (LL) - Production (PL) Baghouse 60.41 0.0015 Reheat 4.52 0.1006 Absorption 8.96 0.0402 Hardness 1.62 0.2719 Moisture 0.06 0.8148 Production (PL) - Construction (PF) Baghouse 3.23 0.1466 Reheat 2.66 0.1784 Absorption 0.57 0.4940 Hardness 0.54 0.5028 Moisture 0.70 0.4499 Table 4-5. Results of the ANCOVA.

38 on surface elevations in the Gulf Coast regions in the United States (Williams and Williamson 1989). The elevation was determined from Pavement ME Design climate database: Water Table Altitude Land-surface altitude .8978 (20)= 4.6.4 Asphalt Mixture Layer Properties Dynamic modulus values were determined from laboratory testing (Level 1 inputs). IDT dynamic modulus testing was conducted using triplicate samples. Air void contents of the samples were controlled between 7% and 8%. The COV of the test results was less than 20% for all test temperatures and frequencies. For performance evaluation using moduli deter- mined in the indirect testing mode, 54°C moduli values were extrapolated from the constructed master curves developed from laboratory testing due to the temperature constraints for dynamic modulus determined in the indirect mode of loading. This extrapolation approach is based on the work by Bonaquist and Christensen (2005). A reduced temperature range could be used to create master curves similar to that of full experimental testing (Guercio et al. 2005). 4.6.5 Base and Subgrade Properties Resilient modulus (MR) values for crushed limestone and clayey subgrade were collected from previous projects (Mohammad et al. 2008) and were used in the analysis of the various pavement structures. These values were kept con- stant for all four pavement structures. 4.7 Development of Specification Recommendations A recommended practice that addressed the cause and magnitude of variability within and among the three speci- men types (i.e., LL, PL, and PF) was developed from the data collected in the experimental program. Volumetric data were evaluated and tolerance values were proposed. Additionally, mechanical data were evaluated and conver- sion factors for estimating the values of the three specimen types were proposed. Distress Traffic Level (ADTT) High (14,554) Medium (1,992) Low (816) IRI (mm/km) 1973 (125 in/mile) 3175 (200 in/mile) 3969 (250 in/mile) Rut Depth (mm) 9.6 (3/8 in) 14.2 (9/16 in) 14.2 (9/16 in) Table 4-6. Louisiana PMS failure triggers.