Characterization of Various Plant-Produced Asphalt Concrete Mixtures Using Dynamic Modulus Test

1Military College of Engineering, National University of Sciences & Technology (NUST), Risalpur Campus, Risalpur 24080, Pakistan 2Department of Civil, Environmental, and Construction Engineering, University of Central Florida (UCF), 12800 Pegasus Drive No. 211, Orlando, FL 32816-2450, USA 3National Institute of Transportation, School of Civil and Environment Engineering, National University of Sciences & Technology, Islamabad 44000, Pakistan


Introduction
Hot-mix asphalt (HMA) consists of the optimum combination of two basic ingredients: aggregate and asphalt binder.In order to meet the diverse and often conflicting performance parameters, for example, resistance to fatigue, deformation, cracking, and moisture damage; durability; skid resistance; and workability and economy, the mix designer generally manipulates three variables, namely, aggregates, asphalt binder, and the ratio of asphalt binder to aggregates, and thus seeks to achieve the aforementioned performance requirements.In Pakistan, the past few years have seen an increase in the premature failure due to fatigue cracking and rutting in both newly constructed and rehabilitated asphalt concrete pavements.The phenomenon of premature failure of pavement structures is attributed to the current design procedures based on 1993 AASHTO design guide which are inherently empirical and incapable of providing adequate and reliable designs for heavy axle loads and tyre pressures in diversity of climatic regions and necessitates a more comprehensive design approach which incorporates both mechanistic and empirical aspects of design.Mechanistic-empirical pavement design guide (AASHTO Pavement ME) encompasses two parts: mechanistic (determine pavement responses) and empirical (distress prediction models/transfer functions).The success of the mechanisticempirical structural design approach or framework lies in the accurate material characterization for predicting realistic pavement responses and ultimate performance.However, the viscoelastic nature of HMA is a challenge to be considered for its accurate characterization by the material properties.Dynamic modulus | * | of HMA is one such material property Advances in Materials Science and Engineering which reflects the loading time and temperature dependency of HMA.Dynamic modulus is considered as the stiffness property of HMA which can partially characterize its viscoelastic nature.It is the measure of the HMA's resistance to deformation under sinusoidal loading and is given by the absolute value of the complex modulus [1,2].Dynamic modulus of HMA has gained attention of the researchers during the past decade especially after its selection as a design input parameter for material characterization of asphalt concrete in the AASHTO Pavement ME pavement design guide and a candidate for a simple performance test to complement Superpave mix design methodology.
Several studies have been conducted in order to gain an insight into the factors affecting | * |.Bonnaure et al. [3] determined the modulus of asphalt mixtures by the application of a sinusoidal load to trapezoidal specimens and reported that the amount of aggregate and percent of air voids had a significant effect on the stiffness of the mix.Another study reported that the mixtures containing stiffer binders resulted in higher values of | * |.Furthermore, sample preparation techniques did not affect the dynamic modulus test results [4].Kim et al. [5] reported that aggregate sources and gradation, within the North Carolina Department of Transportation, Superpave classification, did not seem to have a significant effect on dynamic modulus.This study also determined that the binder source, binder performance grade (PG), and asphalt content seemed to affect the dynamic modulus of asphalt mixtures.Flintsch et al. [6] concluded that mixes of the same type resulted in different measured | * | values because of different constituents, that is, aggregate type, asphalt content, and percentage RAP which showed that | * | was sensitive to mix constituents and properties.Cross et al. [7] evaluated the factors affecting | * | and observed that the testing on lowest temperature of −10 ∘ C caused significant frost buildup on the test frame, samples, and LVDTs making it a difficult and time consuming task to determine | * | below 0 ∘ C. Further, it was reported that the gyratory sample should be compacted to 6.0 ± 1.0% air voids in order to obtain a cored test specimen of required dimensions at 4.5 ± 1.0% air voids.PG grade in addition to test temperature and frequency was reported to have a significant effect on | * |.However, no significant effect of nominal maximum aggregate size (NMAS) or mix designation was observed.Tashman and Elangovan [8] developed a database of | * | values of seven different job mix formulae (JMF) mixes with aggregates of different types and sources typically used in Washington state.Statistical analysis of the results revealed that use of different JMF mixes affected the dynamic modulus.However, it was also observed that the difference in | * | due to different JMF was more significant at high temperatures and low frequencies.It was also observed that variation in the aggregate percent passing number 200 sieve did not have any effect on the dynamic modulus.Mohammad et al. [9] documented the effect of aggregate size on | * | as a result of the study conducted for characterization of Louisiana asphalt mixtures.It was observed that NMAS had a significant effect on | * |, as the larger aggregate size combined with recycled asphalt (RAP) yielded higher values of | * | at high temperatures.Another study developed a dynamic modulus catalog for NJDOT for implementation of AASHTO Pavement ME structural design approach which was achieved by characterizing twenty-one different typical plant-produced HMA mixes [10].Bonaquist [11] conducted dynamic modulus testing and reported that, for the same aggregate source mixes, dynamic modulus values were not much different from one another.Further, the aggregate sources with higher dynamic modulus values had the higher limiting minimum modulus values also when compared to other aggregate sources.Limiting minimum modulus values represents the stiffness of aggregate.A study compared the dynamic modulus of field extracted core with laboratory fabricated specimens and concluded that at 4 ∘ C dynamic modulus can be compared [12].Laboratory evaluated dynamic modulus was related to field response and results indicated that laboratory obtained dynamic modulus was inversely proportional to the field measured pavement response of the asphalt longitudinal strains [13].A number of recent comparative research studies characterized different HMA mixes based upon dynamic modulus test and constructed master curves in order to meet the practitioner needs [14][15][16].
Kim et al. [16] investigated the effect of additive (LEAD-CAP) on warm-mix asphalt using dynamic modulus, resilient modulus, and in-door accelerated pavement test (APT).The results suggested that additive is capable of producing mixture at relatively low temperature up to 30 ∘ C in comparison to conventional mixture and can be compared reasonably well with controlled hot-mix asphalt.Various researches determined the dynamic modulus of plant-produced or laboratory prepared mixes and validated the predictive ability of Witczak 1-37A model to a reasonable conformity [17][18][19].El-Badawy et al. [20] in ensuing research study also calibrated the predictive equations for Idaho state mixtures.Cho et al. [21] evaluated the dynamic modulus of mixtures used in Korea and developed the | * | predictive equation for Korean AASHTO Pavement ME and concluded that developed predictive equations were well correlated with measured values.A study was carried out in Saudi Arabia for implementation of AASHTOWare Pavement design and evaluated two models, namely, NCHRP 1-37A and 1-40D, and concluded that NCHRP 1-37A | * | model showed accurate and unbiased results [22].A recent study evaluated the viscoelastic properties of a performance grade binder modified with different percentages of a wax-based warmmix asphalt (WMA) using dynamic modulus test.This study concluded that the Hirsch model (2003) [23] provided better approximations of the | * | values than the Witczak model [24].In consequence of requisition of skilled personnel, time, and cost effect associated with the dynamic modulus test, researchers have been attempting for several years by using various modeling techniques to develop the prediction equations which can predict the | * | values directly from the mix properties.The last few decades have seen the development of these predictive equations as listed in Table 1.
The most recent models listed in  1 indicate dynamic modulus, a key material property of HMA that better reflects the viscoelastic nature of asphalt concrete.Its significance has further been increased as a result of its selection as a material characterization design input parameter in the AASHTO Pavement ME and thus asphalt concrete mixtures need to be characterized in terms of | * | at regional levels [20][21][22].However, to the knowledge of authors, no such research study characterized locally used mixtures in Pakistan.Hence, this research study aimed to determine the dynamic modulus of plant-produced asphalt mixtures in laboratory and results obtained were employed to two-level factorial design for determination of factor affecting | * |.This research study characterizes and develops the master curves for different asphalt concrete mixes (for wearing and base course mixes), investigates the factors affecting dynamic modulus (stiffness parameter) in order to compare and rank selected asphalt concrete mixes, and validates the laboratory observed dynamic modulus values with the two dynamic modulus prediction models, namely, Witczak and Hirsch models.

Objective and Scope.
The objective of this study is to develop master curves for various plant-produced asphalt concrete mixtures and rank them based on dynamic modulus.The study incorporates four wearing mixes of nominal maximum aggregate size (NMAS) of 19 mm and three base mixes of NMAS of 37.5 mm procured from different highway construction sites of Pakistan.The binder and aggregate type used for testing were penetration grade 60/70 and limestone aggregate, each from two different sources.The study variables for different selected mixtures are presented in Table 2.

Selection of Material.
Seven plant-produced asphalt concrete mixtures were procured from different highway construction projects keeping in view the desired variability and availability of the plant-produced HMA in Pakistan.
Representative samples of all seven mixes were collected from dump trucks at plant site, following the proper sampling techniques [26].The mixtures used in the study were designed using Marshall method (optimum bitumen content determination) by the contractors of the selected projects pursuant to specifications of National Highway Authority (NHA) [27].Job mix formulae and gradations of the selected asphalt mixtures are presented in Table 3.

Specimens Preparation.
Representative plant-produced asphalt mix samples were subjected to testing for maximum theoretical specific gravity, asphalt content, and gradation of the aggregate.Then these samples were reheated to compaction temperature of 135 ∘ C in the laboratory and triplicate specimens for each test temperature were fabricated in accordance with ASTM 3496-99 [28] using Superpave gyratory compactor with target voids in total mix (VTM) of 4% ± 1% (after coring and/or cutting).The binders on these plant-produced mixes underwent short-term aging during their production stage; however, these binders were extracted/tested.The variation among triplicates specimens was in specified range as specified in AASHTO TP62-07 [1].Each gyratory compacted specimen was carefully sawed up to required dimension of 100 mm diameter and 150 mm height to comply with the dynamic modulus test specimen requirements as per AASHTO TP 62-07 [1].

Laboratory Testing.
The | * | test yields phase angle and | * | value.The phase angle () is the angle by which induced axial strain lags behind the applied compressive stress.Figure 1 illustrates the sinusoidal stress and the resulting strain defined by the angular velocity which in turn are related to the loading frequency and time which implies that the phase lag represents the time dependency of HMA.AASHTO Pavement ME has three levels of input: level 1 includes determination of | * | in laboratory (material input), whereas levels 2 and 3 encompass determination of | * | by the use of predictive equations.Dynamic modulus has also been adopted by AASHTO as a provisional standard in AASHTO designation TP62-07 [1].The dynamic modulus test method is a widely used laboratory test that requires application of a compressive axial stress to a cylindrical specimen of HMA and the recoverable strain is calculated from axial deformations measured at two, three, or four locations (as  required) by using linear variable differential transformers (LVDTs).Dynamic modulus is then calculated as the ratio of stress magnitude to average strain magnitude.The dynamic modulus test was conducted at four temperatures and six frequencies for each mixture using asphalt mixture performance tester (AMPT).Specimens were conditioned for the required equilibrium in order to achieve the desired test temperatures before conducting the test.The test equipment consists of environmental chamber, which controls temperature ranging from 4 to 60 ∘ C, and confined pressure system which provides pressure up to 210 kPa.Three replicate specimens were tested at each test temperature with COV ranging from 4% to 21%.The tests were performed at the loading frequencies of 25, 10, 5, 1, 0.5, and 0.1 Hz and the temperatures used for the testing were 4.4 ∘ C, 21.1 ∘ C, 37.8 ∘ C, and 54.4 ∘ C. The laboratory results were further cast off to develop master curves and statistical analysis.

Results and Discussion
The dynamic modulus exhibits stress-strain relationship under compressive sinusoidal loading.As expected, for all the tested asphalt concrete mixtures, the dynamic modulus values decreased with an increase in temperature and decrease in the loading frequency.It was observed that LNLP and WUP wearing course mixtures have relatively higher dynamic modulus values at low and high temperature variations, respectively, whereas JPU base course mix has relatively highest dynamic modulus values at both low and high temperatures for the range of loading frequencies (Table 4).
The average test results/typical isothermal and isochronal curves for dynamic modulus at four different temperatures and six different loading frequencies for wearing and base mixtures tested in the study have been presented in Figures 2(a) and 2(b), respectively.It is evident from these plots that dynamic modulus decreases with increase in temperature and increases with increase in frequency.Also, sensitivity analysis reveals that, for a given loading frequency, an increase in temperature (from 21.1 to 37.8 ∘ C) translated into 57% and 55% drop in | * | values on average for wearing and base course mixes, respectively, whereas, for a given temperature, an increase in loading frequency (from 0.1 to 25 Hz), 68% and 79% of variation in | * | values on average, was attributed to wearing and base course mixes, respectively.

Master Curves for Dynamic Modulus.
The test results obtained were used for development of master curve for subsequent use in pavement structural response and design process.The test results for the replicate specimens were averaged for each test temperature and master curves for the average values of | * | at a reference temperature of 21 ∘ C for each mix were constructed using the time-temperature superposition principle using Microsoft5 solver sheet.Also, keeping in view the testing limitation and inability of the asphalt mixture performance tester (AMPT) to conduct test at −10 ∘ C, an abbreviated approach developed in NCHRP Project 9-29 was used for construction of the stress-dependent master curves [29].
The general form of sigmoidal function used to develop a master curve is given as follows [30]: where log(| * |) is log of dynamic modulus. is minimum modulus value.  is reduced frequency, that is, /(), where  is actual frequency and () is the required shift factor at reference temperature. is span of modulus value (range of lowest and highest dynamic modulus values).,  are shape parameters of sigmoidal function.
The master curves wearing and base mixes are presented in Figure 3 which illustrates that they tend to converge at higher frequencies; however they have considerable degree of separation at lower frequencies.Higher frequency is analogous to lower temperature and vice versa in terms of temperature.It can be inferred that | * | of LNLP wearing mix (Figure 3(a)) and JPU base mix (Figure 3(b)) are relatively better performing for given tested mixtures which is attributed to the mixtures packing arrangement with the relatively lower design asphalt content than other mixtures, thus exhibiting higher stiffness values (see Table 3).

Evaluation of Dynamic Modulus Prediction Equations.
As part of this study two prediction equations were evaluated for regional applicability in Pakistan.As there was no information available with reference to the default dynamic modulus values for commonly used mixtures in Pakistan, it was necessary to characterize the commonly used mixtures accordingly.The inputs parameters for both the equations were primarily obtained from the JMF of the mixtures.However, certain other information like viscosity and shear modulus of binder were obtained by conducting laboratory tests on the extracted binders.

Witczak Model. The current Witczak model used in AASHTO Pavement ME is improved version of the previous
Witczak and Fonseca model in order to use a wide range of AC mixtures.The Witczak model used 205 mixtures prepared with both modified and unmodified binders having a temperature range from 0 to 54.4 ∘ C and the loading frequency ranging from 0.1 to 25 Hz.A total of 39 different aggregate types were made part of this database and the mixtures tested included both kneading and gyratory compacted specimens [31].Input parameters of Witczak 1-37A viscosity based model (2) include bitumen viscosity, effective asphalt content, aggregate gradations, air voids, and loading frequency: where  * is dynamic modulus of mix, 10 For evaluation purposes, values of dynamic modulus were predicted using the Witczak model at the conditions corresponding to each measured value of | * |. | * | values were predicted using the volumetric details and aggregate gradations obtained from the JMF of the mixtures tested: loading frequency (0.1 to 25 Hz); % passing number 200 sieve (4.2 to 5.2%); cumulative % retained on number 4 sieve (35 to 49%); cumulative % retained on 3/8 in.sieve (43 to 69%); cumulative % retained on 3/4 in.sieve (67 to 100%); and air voids % by volume (3.5 to 4.5%).Results indicated that the Witczak model mostly underpredicts the values for dynamic modulus for HMA mixes of Pakistan.Figure 4(a) illustrates the comparison of measured and predicted dynamic modulus values grouped on the basis of temperature.It can be inferred from the figure that the values are distributed along the line of equality for lower temperatures.However, with an increase in temperature, the model tends to underpredict the values for dynamic modulus of the mixtures selected for the study.This implies that the said model is more relevant and significant at lower test temperatures as compared to higher temperature conditions for the tested mixtures in the study.

Hirsch Model.
Christensen et al. [23] developed Hirsch dynamic modulus prediction equation based on law of composite mixtures.The model was developed using database of various mixtures to determine dynamic modulus using binder shear modulus, that is,  * , and the volumetric properties of the mix, that is, voids in mineral aggregate (VMA) and voids filled with asphalt (VFA).The underprediction of dynamic modulus values is attributed to the variation in binder and volumetric (JMF) properties used for preparation of plant-produced mixtures compared to that used for the development of Witczak  and Hirsch models.The parameter values obtained from the density-void analysis on the test sample could be more accurate than JMF parameters.Also, these models were calibrated using datasets of conditions quite different from Pakistan which are quarry dependent: variation in aggregate mineralogical composition (limestone, etc.) and gradation (percent passing different sieves) and variation in binder properties and its manufacturing process and industry.Also, the underprediction of dynamic modulus values is in agreement with similar past studies conducted on the local material for calibration of these models [32,33].

Conclusions
Performance evaluation of seven plant-produced HMA mixtures was carried out using dynamic modulus testing and factors influencing the dynamic modulus of HMA were evaluated.It was observed that the test temperature and loading frequency are significant factors affecting dynamic modulus for both wearing and base course mixtures; however, VMA is marginally significant in case of base course mixtures, only.Furthermore, there was significant difference between the dynamic modulus of different mixtures based on different aggregate size, gradation, and aggregate source.At low (cold) temperatures, the parameter of viscous (or elastic) properties of the mixtures (phase angle) decreases which indicates more elastic behavior of material with high modulus values and is attributed to the dependency of binder on asphalt concrete response at lower temperatures.However, at high (warm) temperatures, the effect of aggregate skeleton/interlock starts to overpower the viscous binder effect causing the phase angle to decline.As part of this study the dynamic modulus prediction models, namely, Witczak and Hirsch, were also evaluated for their potential regional applicability.Results indicated that both models mostly underpredict the value of dynamic modulus for the selected conditions/mixtures.Nevertheless, these models could be used for evaluation purposes and for the design of low and medium traffic volumes pending future investigation of the revised prediction
of the Hirsch model by comparison of the measured | * | and the predicted | * | values.Hirsch model predictions were made using the input data, that is, VMA and VFA obtained from the JMF of each selected mix.Comparison of the results indicated that Hirsch model consistently underpredicts the dynamic modulus values for HMA mixes of Pakistan regardless of the test temperature, the frequency, or the mix selected as shown in Figure 4(b).

Figure 4 :
Figure 4: Measured versus predicted plot of dynamic modulus.
2003and Bonaquist (hereafter referred to as Hirsch model), have been reported to be reasonably accurate and were used to evaluate understudy HMA mixtures in this research.The findings of studies mentioned in Table

Table 2 :
Project and mixtures designation-wearing and base course mixtures.

Table 3 :
Job mix formulae for selected asphalt mixtures.
The Hirsch model used 18 different mixtures and five different binder types to fabricate both Marshall and Superpave designed specimens.The tests were conducted at temperature ranging from 4 to 38 ∘ C and loading frequencies of 0.1 and 5 Hz.The database used for the development of model had air voids ranging from 5.6 to 11.2%; VMA ranging from 13.7 to 21.65%; and VFA of 38.7 to 68%.However, the input parameters range used for validation purpose in this study is given in parentheses against each parameter below.Hirsch model is presented as follows: