You are on page 1of 8

Construction and Building Materials 25 (2011) 26192626

Contents lists available at ScienceDirect

Construction and Building Materials


journal homepage: www.elsevier.com/locate/conbuildmat

Evaluating and comparing different methods and models for generating relaxation modulus master-curves for asphalt mixes
Lubinda F. Walubita a, Allex E. Alvarez b,, Geoffrey S. Simate c
a

TTI-Texas A&M University System, College Station, TX 77843, USA Department of Civil Engineering, University of Magdalena, Santa Marta, Magdalena, Colombia c Faculty of Engineering and the Built Environment, University of the Witwatersrand, Johannesburg, South Africa
b

a r t i c l e

i n f o

a b s t r a c t
Based on a master-curve, mixture properties can be predicted or interpolated at different temperatures and loading times of interest from a limited set of laboratory test data. This paper presents a comparative assessment of three methods used for generating the relaxation modulus (E(t)) master-curves of hot-mix asphalt (HMA). These methods were the Arrhenius, the WilliamsLandelFerry (WLF), and an optimization technique with the sum of square error (SSE) method. Experimental data (E(t) values) for different HMA mixtures were gathered by performing uniaxial loading (strain-controlled) relaxation modulus tests (RMT). The process for evaluating the three methods was based on using the same RMT laboratory data to generate E(t) master-curves and then comparing the best t functions. Corresponding results suggested that the SSE method generated the best t functions relative to the measured E(t) data. Unlike the Arrhenius and WLF models, the SSE is independent of external empirical material constants in its application and is universally applicable to any given material or HMA mix type. However, this study also demonstrated that both the Arrhenius and WLF methods can produce satisfactory (and in fact equivalent results) if appropriate constants, that are material or mix type specic, are used. Conclusively, the ndings of this paper suggested that satisfactory application of the Arrhenius and WLF methods should be with caution, in particular with respect to the material constants. Otherwise, the SSE method proved to be more accurate and would be preferred. Published by Elsevier Ltd.

Article history: Received 30 September 2010 Received in revised form 30 November 2010 Accepted 7 December 2010 Available online 30 December 2010 Keywords: Relaxation modulus Master-curve Material properties Uniaxial-direct loading Strained-controlled loading

1. Introduction A master-curve of a hot-mix asphalt (HMA) mixture property at any reference temperature (Tref) is dened as the relationship between that property and the reduced loading time, frequency, or strain rate. A master-curve is used primarily to model HMA material properties (e.g., stiffness or modulus) at any temperature of interest based on the timetemperature superposition principle and measurements at multiple test temperatures as a function of either loading time, frequency, or strain rate. With a master-curve, mixture properties and/or behavior can be predicted or interpolated at different temperatures and loading times of interest from a limited set of laboratory test data [1]. In generating these mastercurves, the timetemperature superposition principle (for thermorheologically simple materials) is often used with the assumption that similar material properties can be obtained at different temperatures and loading times for HMA mixtures that exhibit linear visco-elastic behavior [2]. Chehab et al. [3] showed that the

timetemperature superposition principle may still hold even when the material linear visco-elastic condition is violated. To construct a master-curve, laboratory test data measured at different temperatures must be shifted relative to the loading time (or frequency or strain rate) so that the various curves representing the response at different test temperatures can be aligned to form a single curve called the master-curve. The master-curve is often constructed using a selected reference temperature of interest to which all data must be shifted using shift factors. With the aforementioned background, the objective of the work contained in this paper was to evaluate the various methods and models that can be satisfactorily used for generating the relaxation modulus (E(t)) master-curves, as a function of time and at a reference temperature of 20 C. Three methods were evaluated in this paper and these are: (1) The Arrhenius method [46]. (2) The WilliamsLandelFerry (WLF) method [7]. (3) The sum of square error (SSE) method with an optimization technique and the solver function in the Excel spreadsheet. The methodology for evaluating these three methods was based on using the same laboratory test data to generate E(t)

Corresponding author. Tel.: +57 5 4301292; fax: +57 5 4303621.


E-mail addresses: lfwalubita@hotmail.com (L.F. Walubita), allexalvarez@yahoo. com (A.E. Alvarez), simateg@yahoo.com (G.S. Simate). 0950-0618/$ - see front matter Published by Elsevier Ltd. doi:10.1016/j.conbuildmat.2010.12.010

2620

L.F. Walubita et al. / Construction and Building Materials 25 (2011) 26192626

master-curves as a function of reduced time at a reference temperature of 20 C and then comparing the resulting curves through the coefcient of correlation (R2). In the paper, the relaxation modulus test (RMT) protocol is discussed rst, followed by the three methods of generating the E(t) master-curves. Results are then presented followed by a comparative discussion and synthesis of the results. A summary of ndings and recommendations is then presented to conclude the paper.

2. Relaxation modulus (E(t)) under uniaxial loading testing E(t) is one of the fundamental mechanical properties that can be used to model the visco-elastic behavior of HMA for various purposes including material property characterization, design, and performance prediction. The time-dependent E(t) obtained in the RMT is a function of time because of the visco-elastic nature of HMA. Under deformation, the stress builds up because of the HMAs elastic nature but then relaxes at constant-xed strain because of its ability to undergo viscous ow, a phenomenon generally known as stress relaxation. This relaxation is usually reected in the decrease of E(t) over time in a strain-controlled RMT. As demonstrated by Walubita et al. [8], the use of the Laplace (Euler) Gamma transformation functions in combination with elastic theory allows for the determination of the HMA complex dynamic shear modulus (G) and viscosity (g0 ) from the RMT data; thus also characterizing the shear properties of HMA [9,10]. The RMT conducted in this study used uniaxial static-direct loading in a strain-controlled mode and was considered to be non-destructive. The test involved applying a constant uniaxialstatic trapezoidal-shaped strain to a cylindrical HMA specimen either in tension or compression for a given time period and then releasing the strain for another given time period, thereby allowing the specimen to rest or relax (elastic recovery). The RMT loading conguration is shown in Fig. 1. The RMT loading parameters consisted of an input strain magnitude of 200 le (tension or compression), a ramp time of 0.6 s, a static loading time of 60 s, and a rest period of 600 s [10]. Based on the RMT loading conguration shown in Fig. 1, 200 le was equivalent to a displacement of 0.02 mm, and the 0.6 s ramp time was equivalent to a ramp speed of 2 mm/min [10]. The RMT was conducted in an environmental-temperature controlled chamber at temperatures of 10, 20, and 30 C starting with the lowest temperature so as to minimize any possible micro-damage. The temperatures were monitored and controlled to a tolerance of 0.5 C through a thermocouple probe attached inside a dummy HMA specimen also placed in the same environmentaltemperature chamber as the test specimens. The total RMT time

for each test temperature was approximately 25 min. HMA specimens were, however, preconditioned for at least 2 h. A 9.5 mm nominal maximum aggregate size (NMAS) densegraded HMA mix (denoted as Texas Type C mix) with 4.6% of PG 64-22 asphalt-binder and limestone aggregate was used and subjected to the RMT. Superpave gyratory compactor molded cylindrical HMA specimens were used with nal dimensions of 100 mm in diameter by 150 mm in height, with a target air void (AV) content of 7 0.5% [11]. From the RMT, the measurable data include time (s), load (kN), deformation (mm), and temperature (C). From the measured RMT data, E(t) is calculated as indicated in Eq. (1).

Et

Pt 1 106 Pt 5 103 Pt 200pr 2 pr 2 e pr 2

where E(t) is the time-dependent relaxation modulus (MPa); P(t) is the measured load (kN); r is the radius of the cylindrical HMA specimen (mm); and e is the applied input loading strain, which was 200 le in this study (both in tension and compression). Fig. 2 shows an example of the measured E(t) data in tension at three test temperatures of 10, 20, and 30 C on a loglog scale. From the data shown in Fig. 2, a master-curve of E(t) versus reduced time n (s) can then be generated in the form of a simple power function shown in Eq. (2) [10].

Et E1 nm n t aT

2 3

In Eqs. 2, 3, E1 is the E(t) value at n = 1.0 s (MPa); n and t are reduced and actual RMT time (s), respectively; and m is the stress relaxation rate (0 6 m < 1). The parameter aT is the temperature shift factor. For a reference temperature of 20 C, the value of aT(20C) is 1.0. Various methods for generating the E(t) master-curves are discussed and evaluated in subsequent sections of this paper. 3. Methods for generating the relaxation modulus (E(t)) mastercurves The various methods and models for generating the E(t) mastercurves are discussed and evaluated in this section. These methods include the Arrhenius method, the WLF method, and the SSE method. The major difference among these methods is basically in the computation of the temperature shift factors (aT, Eq. (3)). It should be emphasized here that the aT is the main driving force in the generation of any master-curve and is dened as the shift factor at a temperature of interest Ti relative to the reference temperature.

Load

200

Microstrain

100 Rest Period 0 0 -100 200 400 600 800

Tension

1000

1200

1400

Compression
-200 -200

Load

Time (s)
Fig. 1. Relaxation modulus test loading conguration.

L.F. Walubita et al. / Construction and Building Materials 25 (2011) 26192626

2621

1.0E+04

where

E(t) (MPa)

1.0E+03

10 C 20 C

" # 5 103 Pt Log Et meas Log pr2  Log Et pred Log E1 mLog n Log E1 mLog t aTi 

1.0E+02

30 C

Based on Eq. (8), the predicted E(t)pred using t functions at each test temperature were simultaneously computed as follows:
1.0E+01 1 10 100

Log Et pred@10C Log E1 mLog

Time (s)
Fig. 2. Example of E(t) tension data measured at 10, 20, and 30 C.

aT 10C  t  Log E1 mLog t

Log Et pred@20C Log E1 mLog

aT 20C  t 

10

3.1. The Arrhenius method

Log Et pred@30C Log E1 mLog


Eq. (4) is the Arrhenius timetemperature superposition model for computing aT [4]:

aT 30C

11

And the SSE is then given as follows:

Log aT C a

1 1 T i T ref

SSE

X Log Etpred Log Et meas 2 0:00

12

where Ca is the material constant that is a function of the activation energy (DHa) and ideal gas constant (Rg) (i.e., Ca = DHa/ 2.303Rg); Ti is the test temperature of interest in degrees Kelvin (K = 273 + C); and Tref is the reference temperature (K) (e.g., Tref = 273 + 20 = 293 K in this study). Traditionally, a default Ca value of 13,631.28 for DHa 261,000 J/mol and Rg 8.314 J/(molK) is often used; herein denoted as the traditional Arrhenius method [8]. As discussed subsequently, other Ca values were also investigated in this study and resulted in generating the modied Arrhenius method. 3.2. The WilliamsLandelFerry (WLF) method The WLF timetemperature superposition model for computing aT is shown in Eq. (5) [7]:

with this SSE method, the fundamental concept is basically to generate an E(t) t function by iteratively changing the aTi, E1, and m values using the solver function subject to the primary condition that the SSE between E(t)meas and E(t)pred using a t function should be equal to zero. Consequently, the solver parameters were set as follows and are further demonstrated in Fig. 3 with an analysis example shown in Fig. 4:  Target: SSE is equal to (or close to) zero.  Parameters to iteratively change: aT10C, aT30C, E1, and m.  Constraints: aT10C P 0.0, aT20C = 1.0, aT30C P 0.0. As shown in Fig. 4, 100% iterative convergence with an SSE exactly equal to zero may not always be attained. However, the spreadsheet will always generate the best t as close as it can get. Note, however, that in this SSE method, model formulation and appropriate setting of the solver parameters is critical in achieving satisfactory results. 4. Results and analysis RMT data were analyzed as per Eq. (1), and master-curves were then comparatively generated using Eqs. (2)(12). The results presented and analyzed in this section include the E(t) master-curves and the associated analysis parameters for the RMT data in tension loading mode only. 4.1. Relaxation modulus (E(t)) master-curves The master-curves generated based on the Arrhenius, WLF, and SSE methods using the same RMT data given in Fig. 2 are shown in Figs. 59. The plots are for both the E(t) measured and E(t) predicted values (using a t function Eq. (8)) as a function of reduced time (n) on loglog scales. For most analysis applications however, what is generally shown is a loglog plot of only the E(t) measured values versus reduced time (n). Based on Figs. 59, the rank order of generating a good t, in particular for the E(t) measured values as a function of reduced time, is the SSE method followed by the Arrhenius method and last, the WLF method. This is not surprising because the SSE method uses only the laboratory determined RMT data, whereas the other two methods do introduce external empirically-determined material constants in the analyses. These empirical material constants

Log aT

C 1 T i T ref C 2 T i T ref

where C1 and C2 are empirically-determined material constants; Ti is the selected temperature of interest (C or K); and Tref is the reference temperature (C or K). For this study, the temperature unit was degrees Celsius (C). Values of 19 and 92 for C1 and C2, respectively, are the most commonly reported Ci values that provide satisfactory results in the traditional WLF method [5,8]. Like for the Arrhenius method, various other Ci constants were also evaluated for the WLF method in this study to generate the modied WLF method. 3.3. The sum of square error (SSE) method The SSE method is basically a spreadsheet optimization technique based on minimizing the sum of square error (SSE) between the measured E(t)meas and the predicted E(t)pred estimated using a t function. In this method, a solver spreadsheet function is iteratively used to compute the best aTi, E1, and m values (Eq. (2)) to generate a t function master-curve that best ts the measured E(t)meas. For convenience and simplicity, the log format was used for both E(t) and time values. The SSE model formulations were as follows:

Log Etpred Log Et meas

2622

L.F. Walubita et al. / Construction and Building Materials 25 (2011) 26192626

Fig. 3. Example of the Solver function formulation.

Test Time t (s) 1.010 1.295 1.590 1.885 2.180 2.475 2.770 3.065 3.360 3.655 3.950 4.245 4.540 4.835 5.130 5.425 5.720 6.015 6.310 6.605 6.900 7.195 7.490 7.785 8.080 8.375

E(t) Measured (MPa) LOG(E(t)) 4926.940 3.693 3733.303 3.572 3551.459 3.550 3515.706 3.546 3299.663 3.518 3231.265 3.509 3200.198 3.505 3106.920 3.492 2971.702 3.473 2921.969 3.466 2909.543 3.464 2803.864 3.448 2789.858 3.446 2749.445 3.439 2682.626 3.429 2600.250 3.415 2536.524 3.404 2530.308 3.403 2475.910 3.394 2430.838 3.386 2432.391 3.386 2381.100 3.377 2350.015 3.371 2326.702 3.367 2286.292 3.359 2244.328 3.351

Reduced Time (s) LOG( ) -1.496 0.032 -1.388 0.041 -1.299 0.050 -1.225 0.060 -1.162 0.069 -1.106 0.078 -1.057 0.088 -1.014 0.097 -0.974 0.106 -0.937 0.116 -0.903 0.125 -0.872 0.134 -0.843 0.144 -0.816 0.153 -0.790 0.162 -0.766 0.172 -0.743 0.181 -0.721 0.190 -0.700 0.200 -0.680 0.209 -0.661 0.218 -0.643 0.228 -0.625 0.237 -0.609 0.246 -0.593 0.256 -0.577 0.265

E(t) Predicted (MPa) LOG(Fit Fun) Fit Function 3.662 4587.168 3.624 4205.654 3.593 3914.691 3.567 3688.722 3.545 3506.038 3.526 3353.986 3.508 3224.606 3.493 3112.596 3.479 3014.261 3.466 2926.934 3.455 2848.634 3.444 2777.852 3.434 2713.412 3.424 2654.389 3.415 2600.034 3.406 2549.742 3.398 2503.010 3.391 2459.422 3.384 2418.627 3.377 2380.327 3.370 2344.269 3.364 2310.233 3.358 2278.030 3.352 2247.494 3.346 2218.481 3.341 2190.863

Square (Error) (Log[Fit Fun]-Log [E(t)]) 2 0.00096 0.00268 0.00179 0.00044 0.00069 0.00026 0.00001 0.00000 0.00004 0.00000 0.00008 0.00002 0.00015 0.00023 0.00018 0.00007 0.00003 0.00015 0.00010 0.00008 0.00026 0.00017 0.00018 0.00023 0.00017 0.00011

SSE
Fig. 4. Example of E(t) analysis based on the SSE method.

0.32910

are obviously material/mix type dependent and their universal application should be used with caution. However, the modied Arrhenius and WLF results appear to be signicantly improved and comparable to the SSE method if different material constants (Ci) are used as shown in Figs. 6 and 8. This again reinforces the fact that these material constants are material/ mix type dependent, and thus different Ci values should be used for different types of HMA to generate proper E(t) master-curves. Otherwise, the SSE method, which does not involve external data or empirical material constants, should be used. 4.2. Comparison of analysis parameters Table 1 is a summary of the analysis and computed E(t) parameters used/determined when generating the master-curves shown in Figs. 59. The E1 and m values including the aTi and R2 are com-

parable among the three methods once the material constants for the Arrhenius and WLF methods (modied) were adjusted (e.g., as shown in Table 1 for the Type C HMA mix studied). This in fact indicates that even the two methods (modied Arrhenius and WLF) can be satisfactorily applied to generate E(t) master-curves provided that appropriate material constants are used. Based on Table 1 and considering only the new material constants (Ca = 12,630.73, C1 = 31, C2 = 214) for the modied Arrhenius and WLF methods, the statistics are as shown in Table 2. All the coefcient of variation (COV) values in Table 2 are lower than the commonly used 20% threshold, indicating that the three methods are not statistically different; provided that the appropriate material constants are used in the modied Arrhenius and WLF methods. However, the SSE method would be preferred because it yielded the best t function (see R2 value in Table 1) and it is also independent of the empirical material constants.

L.F. Walubita et al. / Construction and Building Materials 25 (2011) 26192626

2623

1.0E+04

1.0E+04

E(t) (MPa)

10C 20C 30C

E(t) (MPa)

1.0E+03

1.0E+03

1.0E+02

1.0E+02

E(t) = 1250.8t -0.3117 R 2 = 0.997

1.0E+01

1.0E+01 0 1 100 10000

100

10000

Reduced Time (s)

Reduced Time (s)

(a) E(t) Predicted Fit function

(b) E(t) Measured

Fig. 5. Tension E(t) master-curve based on the traditional Arrhenius method with Ca = 13631.28.

1.0E+04

1.0E+04

E(t) (MPa)

10C 20C 30C

E(t) (MPa)
100 10000

1.0E+03

1.0E+03

1.0E+02

-0.3345 1.0E+02 E(t) = 1345t R2 = 0.9977

1.0E+01

1.0E+01

100

10000

Reduced Time (s)

Reduced Time (s)

(a) E(t) Predicted Fit function

(b) E(t) Measured

Fig. 6. Tension E(t) master-curve based on the modied Arrhenius method with Ca = 12630.73.

1.0E+04

1.0E+04

E(t) (MPa)

E(t) (MPa)

1.0E+03

1.0E+02
E(t) = 938.81t -0.2394 R2 = 0.9895

1.0E+02

10C 20C 30C

1.0E+01 0.001

0.1

10

1000

100000

1.0E+00 0.001 0.01

0.1

10

100

1000 10000

Reduced Time (s)

Reduced Time (s)

(a) E(t) Predicted Fit function

(b) E(t) Measured

Fig. 7. Tension E(t) master-curve based on the traditional WLF method with C1 = 19 and C2 = 92.

4.3. Other examples To further investigate the effects of the material constants in the modied Arrhenius and WLF methods, one additional 12.5 mm NMAS dense-graded Superpave HMA mix was evaluated [10]. The mix-design characteristics for this Superpave mix were 5.6% PG 6422 asphalt-binder plus gravel aggregate plus 1% hydrated lime. The results for the analysis and computed RMT parameters in tension are listed in Tables 35. The results also include Type C and Superpave HMA mixes aged for 6 months at 60 C in an environmental temperature-controlled room. With the exception of the aT10C shift factor results in Table 4, the COV values generally indicate that the results do not differ

signicantly among the three methods; albeit that the SSE method yielded the best R2 value in all cases. As evident from the R2 values, the different material constants used for the modied Arrhenius and WLF methods for different mix types yielded the best t functions for generating the E(t) master-curves. Thus, with appropriate material constants, both the modied Arrhenius and WLF methods can be satisfactorily used to generate an E(t) master-curve for any given HMA mix, just like the SSE method. In summary, it is clear from these results that using the same empirical material constants in the Arrhenius or WLF model may not always give the best t E(t) master-curves for different types of HMA; instead different material constants should be explored.

2624

L.F. Walubita et al. / Construction and Building Materials 25 (2011) 26192626

1.0E+04 1.0E+04

E(t) (MPa)

10C

E(t) (MPa)

1.0E+03

1.0E+02
E(t)= 1343.2t -0.3371 R2 = 0.9979

1.0E+02

20C 30C

1.0E+01 0.01

100

10000

1.0E+00 0.01

0.1

10

100

1000

10000

Reduced Time (s)

Reduced Time (s)

(a) E(t) Predicted Fit function

(b) E(t) Measured

Fig. 8. Tension E(t) master-curve based on the modied WLF method with C1 = 31 and C2 = 214.

1.0E+04 1.0E+04 1.0E+03


10C 20C

E(t) (MPa)

1.0E+02

30C

E(t) (MPa)

1.0E+03

1.0E+02

E(t) = 1374t -0.3491 R2 = 0.9987

1.0E+01 0.01

100

10000

1.0E+01 0.01

0.1

10

100

1000

10000

Reduced Time (s)

Reduced Time (s)

(a) E(t) Predicted Fit function


Fig. 9. Tension E(t) master-curve based on the SSE method.

(b) E(t) Measured

Table 1 List of analysis and computed RMT parameters for the type C HMA mix. Parameter The Arrhenius method Ca = 13,631.28 E1 (MPa) m aT10C aT20C aT30C R2 Comment 1, 251 0.3117 44.048 1.000 0.029 0.9970 Traditional Ca = 12,630.73 1345 0.3345 33.363 1.000 0.038 0.9977 Modied The WLF method C1 = 19, C2 = 92 939 0.239 207.526 1.000 0.0137 0.9895 Traditional C1 = 31, C2 = 214 1343 0.337 33.797 1.000 0.0405 0.9979 Modied 1374 0.349 31.62 1.000 0.0489 0.9987 The SSE method

Table 3 Analysis and computed RMT parameters for the Superpave HMA mix. Parameter The modied Arrhenius method Ca = 10,361.08 1084 0.4656 17.7641 1.000 0.0681 0.9747 The modied WLF method C1 = 42.56, C2 = 344.86 1059 0.4565 18.6692 1.000 0.0631 0.9752 1085 0.4314 15.3336 1.000 0.0492 0.9857 The SSE method Statistics

Mean 1076 0.4512 17.2556 1.0000 0.0601 0.9785

COV (%) 1.37 3.93 10.00 0.00 16.29 0.63

E1 m aT10C aT20C aT30C R2

Table 2 Statistics of analysis and computed RMT parameters for the type C HMA mix. Parameter E1 (MPa) m aT10C aT20C aT30C R2 Mean 1354 0.3402 32.9267 1.0000 0.0425 0.9981 Standard deviation 17 0.0078 1.1522 0.0000 0.0057 0.0005 Coefcient of variation (COV) (%) 1.28 2.28 3.50 0.00 13.45 0.05

5. Discussion and synthesis of results Among the three methods evaluated in this paper, the results indicated that the spreadsheet optimization technique with the

SSE method gave the best t function for generating the E(t) master-curves of HMA, followed by the Arrhenius method and lastly, the WLF method. The SSE method uses only the laboratory RMT data and does not incorporate externally introduced empirical constants in its analysis. As shown in this paper, the empirically determined constants in the Arrhenius and WLF methods are material/ mix type dependent and thus, may not be satisfactorily applicable to all materials or HMA mixes. Therefore, these models may not always generate the best t functions for the E(t) master-curves for HMA mixes. Additionally, the Ca constant which is a function of activation energy DH and gas constant Rg suggests that the Arrhenius model was not initially developed for HMA mixes. However, if appropriate material constants are used, satisfactory results can be obtained with either the modied Arrhenius or the WLF method. As demonstrated in this paper, these methods yielded almost similar results when varying material constants

L.F. Walubita et al. / Construction and Building Materials 25 (2011) 26192626

2625

were included. Thus, caution should always be exercised with the Arrhenius and WLF methods as the existing material constants in

Table 4 Analysis and computed RMT parameters for the type C HMA mix 6 months aged at 60 C. Parameter The modied Arrhenius method Ca = 11,311.29 6600 0.2473 23.1279 1.000 0.0532 0.9892 The modied WLF method C1 = 31, C2 = 214 6140 0.2264 33.0832 1.000 0.0413 0.9859 7445 0.2662 12.7973 1.000 0.0474 0.9944 The SSE method Statistics

Mean 6728 0.2466 23.0028 1.0000 0.0473 0.9898

COV (%) 9.84 8.07 44.10 0.00 12.58 0.43

E1 m aT10C aT20C aT30C R2

Table 5 Analysis and computed RMT parameters for the Superpave HMA mix 6 months aged at 60 C. Parameter The modied Arrhenius method Ca = 11,585.88 3749 0.3421 24.9605 1.000 0.04954 0.9908 The modied WLF method C1 = 22.71, C2 = 164.84 3792 0.3365 23.8764 1.000 0.04190 0.9926 3591 0.3345 29.3000 1.000 0.0502 0.9956 The SSE method Statistics

these models were developed based on different materials and may not be readily applicable to HMA mixes. Nonetheless, it should be emphasized here that generation of satisfactory master-curves is also largely dependent on the RMT test data consistency and the RMT raw data reduction process. RMT data consistency is a function of many variables including both the set-up and parameters of the RMT, machine noises/vibrations, specimen AV uniformity and homogeneity, and specimen fabrication (dimensions and parallelism of end-surfaces). Raw data reduction process includes truncation of the often distorted rst series of the captured data to account for possible machine overshoot and test stabilization when the RMT begins. For the same reasons, it may also be necessary at times to truncate the very last set of the captured data as the RMT approaches termination. How much data to truncate is often a subjective decision that depends on the RMT set-up, frequency of data capturing/recording, and the operator and/or the person analyzing the testing data. In this study, truncating the rst two data sets in every RMT proved satisfactory [10]. However, up to ve data sets was still considered reasonably acceptable. Fig. 10 shows the effects of truncating and discarding the rst series of the two RMT data sets on the E(t) master-curves. The improvement in Fig. 10b is evident, even in the R2 value.

Mean 3711 0.3377 26.0456 1.0000 0.0472 0.9930

COV (%) 2.85 1.17 11.02 0.00 9.77 0.24

6. Summary of ndings and recommendations In this paper, three methods (the Arrhenius, WLF, and SSE) were comparatively evaluated for generating relaxation modulus (E(t)) master-curves of HMA as a function of time and at a reference temperature of 20 C. The uniaxial-direct loading RMT was conducted in strain-controlled mode with a strain input load of 200 le

E1 m aT10C aT20C aT30C R2

Time (s) 0.0025 1.2950 1.5900 1.8850 2.1800 2.4750 2.7700 3.0650 3.3600 3.6550 3.9500 4.2450

E(t) 10 C (MPa) 4926.9397 3733.3031 3551.4590 3515.7065 3299.6632 3231.2647 3200.1977 3106.9202 2971.7021 2921.9694 2909.5426 2803.8637

E(t) 20 C (MPa) 1739.9812 1258.9372 1116.7240 1067.7641 1087.9705 990.0520 982.2814 899.9066 990.8299 944.2026 796.5488 863.3812

E(t) 30 C (MPa) 973.3455 385.9253 362.3757 343.9396 328.9343 316.3725 305.6292 296.2858 288.0493 280.7074 274.1017 268.1111

1.0E+04

E(t) (MPa)

1.0E+03

1.0E+02

E(t) = 1377.4t -0.3493 R2 = 0.997

1.0E+01 0.01

100

10000

Reduced Time (s)

(b) Full RMT Data Set


1.0E+04

Time (s) 1.5900 1.8850 2.1800 2.4750 2.7700 3.0650 3.3600 3.6550 3.9500 4.2450

E(t) 10 C (MPa) 3551.4590 3515.7065 3299.6632 3231.2647 3200.1977 3106.9202 2971.7021 2921.9694 2909.5426 2803.8637

E(t) 20 C (MPa) 1116.7240 1067.7641 1087.9705 990.0520 982.2814 899.9066 990.8299 944.2026 796.5488 863.3812

E(t) 30 C (MPa) 362.3757 343.9396 328.9343 316.3725 305.6292 296.2858 288.0493 280.7074 274.1017 268.1111

E(t) (MPa)

1.0E+03

1.0E+02

E(t) = 1375.1t -0.3492 R2 = 0.998

1.0E+01 0.01

100

10000

Reduced Time (s)

(a) First Two RMT Data Sets Truncated


Fig. 10. Effects of truncating the rst two series of the RMT data set (tension loading mode).

2626

L.F. Walubita et al. / Construction and Building Materials 25 (2011) 26192626

(equivalent to 0.02 mm vertical displacement) at three test temperatures (10, 20, and 30 C). The ramp speed was 2 mm/min and the static sitting-loading time was 60 s, with a rest period of 600 s between the tension and compression loading cycles. Based on the RMT tension data analyzed in this paper, the SSE was found to be the best method for generating a t function for the E(t) master-curves, followed by the Arrhenius method and lastly, the WLF method. Compared to the other two methods, the major advantage of the SSE method is that it revolves around the actual RMT data without introducing any external constants in its application when generating a master-curve. Additionally, this spreadsheet optimization technique with the SSE method is universally applicable to any data set and any material/HMA mix type. In contrast, the traditional Arrhenius and WLF methods use externally and already existing empirically determined constants that are material/mix type dependent, and may thus often give unsatisfactory results. Conclusively, the ndings of this paper suggests that satisfactory application of the Arrhenius and WLF methods should be with caution, especially with respect to the material constants. However, the analysis conducted demonstrated that both the modied Arrhenius and WLF methods can produce satisfactory and in fact, equivalent results if appropriate material constants are used in the models. Between these two methods, the modied Arrhenius would be recommended over the modied WLF on the basis that it yielded a better E(t) t function. Otherwise, the SSE method would be most preferred. However, the success of the SSE method is largely dependent on the appropriateness of the model formulation process and setting of the solver parameters to achieve satisfactory results. Nonetheless, generation of satisfactory E(t) master-curves is also dependent on many other factors including the RMT data consistency and the RMT raw data reduction process. Variables such as the specimen AV uniformity and homogeneity, specimen fabrication (dimensions and parallelism of end-surfaces), RMT parameters and set-up, and machine noises/vibrations signicantly affect the RMT data consistency and efforts should always be taken to minimize these effects. Truncation of the often distorted rst and/or last series of the RMT captured test data is one of the critical and subjective aspects of the RMT raw data reduction process. How much data to truncate is a subjective decision and to a large extent depends on the RMT set-up and the frequency of RMT data capturing/recording. For this study, truncating up to the rst ve series of the captured data was found to be satisfactory. As a concluding remark, while only two mixes were used as demonstrative examples in this paper, a similar methodological approach can be applied to other HMA mixes or materials. In particular, the data and results presented herein can be used as a reference benchmark for selecting the appropriate methods and models for generating E(t) master-curves for HMA mixes. Furthermore, the study provides researchers and the scientic community with information on some of the key laboratory and RMT variables

(such as data consistency, and raw data reduction process) that inuence the accuracy and satisfactory generation of E(t) mastercurves. Thus, when generating E(t) master-curves, particularly with RMTs conducted in strain-controlled loading mode, such inuencing factors as discussed in this paper need to be taken into account as potential sources of errors in the nal results. Disclaimer The contents of this paper reect the views of the authors who are solely responsible for the facts and accuracy of the data presented herein. The contents do not necessarily reect the ofcial views or policies of any agency. This paper does not constitute a standard, specication, nor is it intended for design, construction, bidding, contracting, or permit purposes. Trade names were used solely for information and not for product endorsement. Acknowledgements The authors are thankful to all those who provided support (nancial or technical) in the course of this research work. In particular, the laboratory assistance provided by the following people is gratefully acknowledged: Vivekram Umashankar, Jeff Perry, Rick Canatella, Lee Gustavus, and Tony Barbosa. References
[1] Roque R, Birgisson B, Kim B, Cui Z. Use of binder rheology to predict the cracking performance of SBS modied mixtures. Guidelines for use of modiers in Superpave mixtures. vol. 2. Florida Department of Transportation, Florida; 2004. [2] Medani TO, Huurman M, Molenaar AAA. On the computation of master curves for bituminous mixes. In: 3rd EuroBitumen Congress, Vienna, Austria; 2004. [3] Chehab GR, Kim YR, Schapery RA, Witczak MW, Bonaquist R. Time temperature superposition for asphalt concrete mixtures with growing damage in tension state. In: Annual meeting of the asphalt paving technology. Colorado, USA: AAPT, Colorado Springs; 2002. [4] Francken L, Clauwaert C. Characterization and structural assessment of bound materials for exible road structures. In: 6th international conference on the structural design of asphalt pavements. Ann Arbor: University of Michigan; 1988. p. 13044 [5] Lytton R, Uzan J, Fernando EG, Roque R, Hiltmen D, Stoffels S. Development and validation of performance prediction models and specications for asphalt binders and paving mixtures. Report No. SHRP-A-357, Washington, DC; 1993. [6] Jacobs MMJ. Crack growth in asphaltic mixes. Netherlands: Delft University of Technology; 1995. [7] Williams ML, Landel RF, Ferry JD. Temperature dependence of relaxation mechanisms in amorphous polymers and other glass-forming liquids. J Appl Phys 1955;77:37016. [8] Walubita LF, Epps Martin A, Jung HS, Glover CJ, Park ES. Application of calibrated mechanistic fatigue analysis with aging effects. Report FHWA/TX05/0-4468-3. College Station, TX: Texas Transportation Institute-Texas A&M University; 2006. [9] Shames IH, Dym CL. Energy and nite element methods in structural mechanics; 1985. [10] Walubita LF. Comparison of fatigue analysis approaches for predicting fatigue lives of hot-mix asphalt concrete (HMAC) mixtures. College Station, TX. PhD dissertation, Texas A&M University; 2006. [11] TxDOT. TxDOT online manuals. TxDOT; 2007.

You might also like