- Email: [email protected]
Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP
Available online at www.sciencedirect.com
ScienceDirect International Journal of Pavement Research and Technology xxx (2018) xxx–xxx www.elsevier.com/locate/IJPRT
Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP Saman Esfandiarpour ⇑, Ahmed Shalaby Department of Civil Engineering, University of Manitoba, 15 Gilson Street, Winnipeg, Manitoba R3T 5V6, Canada Received 31 March 2017; received in revised form 13 March 2018; accepted 12 April 2018
Abstract MEPDG software can predict long term performance of the asphalt mixes based on asphalt input data. When laboratory measured data of an asphalt mix are not available, this software uses the predictive models to estimate the mix properties. AASHTO recommended calibrating these input models based on local materials and mixes. Previous studies showed that local calibration of dynamic modulus (E*) and creep compliance predictive model improved the reliability of the predictions. The purpose of this paper is to evaluate the impact of local calibration of E* and creep compliance models on long term performance of mixes containing reclaimed asphalt pavement (RAP), and to assess the sensitivity of the predicted distresses to RAP content. Three Levels of asphalt input data were considered; Level 1, calibrated Level 3, and Level 3. For calibrated Level 3, the predicted E* and creep compliance obtained from calibrated models were used as input data. The results showed that the Level 3 input data tend to overpredict the distress predictions of the asphalt mix compared to calibrated Level 3 or Level 1 asphalt input data. It was found that the calibrated Level 3 asphalt input data can be used for the design and analysis of mixes with comparable accuracy of Level 1 input. As conducting laboratory tests for individual mixes is expensive and time consuming, utilizing reliable calibrated models to predict E* and creep compliance can substantially reduce operating and testing expenses. Also, it was found that the predicted distresses are not sensitive to the RAP content. Ó 2018 Production and hosting by Elsevier B.V. on behalf of Chinese Society of Pavement Engineering. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction Mechanistic-Empirical Pavement Design Guide (MEPDG) software called Pavement ME Design uses the MEPDG to design and analyze flexible and rigid pavement structures. Different design traffic loadings, climate data and material properties are used as inputs into this software to predict future pavement distresses. The flexible pavement distresses included in this software are surface roughness,
⇑ Corresponding author.
E-mail addresses: [email protected] (S. Esfandiarpour), [email protected] (A. Shalaby). Peer review under responsibility of Chinese Society of Pavement Engineering.
total permanent deformation (rutting), asphalt layer permanent deformation, asphalt bottom-up fatigue cracking, asphalt top-down fatigue cracking, and asphalt thermal cracking. The distress prediction models require inputs to be defined by the user. The defined inputs include asphalt, base, subbase and subgrade materials characteristics, traffic loadings, and climate data for a pavement structure. Based upon the quality and quantity of the available material properties data, there are three levels of input options in the Pavement ME Design software [8]. Level 1 input option generally requires site specific material properties data which are obtained through laboratory or field testing. These data have the highest level of reliability, and are expected to provide the optimum design and analysis. Level 2 inputs have an intermediate level of
https://doi.org/10.1016/j.ijprt.2018.04.002 1996-6814/Ó 2018 Production and hosting by Elsevier B.V. on behalf of Chinese Society of Pavement Engineering. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
2
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Table 1 Required input asphalt mix and asphalt binder data for Level 1 and Level 3. Input data
Test measurement
Asphalt mix
Mix design (VMA %, Va %, and asphalt content percentages) Aggregate gradation Dynamic modulus Creep compliance (20 °C, 10 °C, and 0 °C) Indirect Tensile (10 °C)
Performance grade Complex shear modulus (G*) Phase angle (d)
Asphalt binder
Input alternatives
a) Globally calibrated model
Level 1 (laboratory-measured values)
Level 3 (default values)
b) ANN calibrated model
Fig. 1. Comparison of globally calibrated and ANN calibrated creep compliance model.
a) Globally calibrated NCHRP 1-37A
b) Nonlinear regression of NCHRP 1-37A
Fig. 2. Comparison of dynamic modulus model predictions.
reliability. These input data are generally obtained through limited laboratory or field testing, or estimated from the correlations with other measured properties. Level 3 inputs have the lowest level of reliability since the typical agency data or software default data are used [1]. Although Level 1 inputs for the AC mixes provide more reliable results than Level 2 and Level 3 inputs, the comprehensive laboratory
testing required to obtain them is time consuming and expensive. Table 1 shows the required input asphalt mix and asphalt binder data for Level 1 and Level 3. For Level 1 input data, all tests and volumetric properties are required except the aggregate gradation and performance grade. However, for Level 3, basic properties of a mix such as
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
3
Fig. 3. The hypothesis of this paper.
Table 2 Unbound materials properties. Properties
Subgrade
Subbase
Base
Materials Type Thickness (mm) Resilient modulus (MPa) Moisture content (%) Liquid limit Plasticity index Maximum dry density (Kg/m3 )
A-7-6 – 30 29 75 42 1410
Crushed lime stone C-base 300 120 8 11 0 2220
Crushed lime stone A-base 200 140 10.8 6 1 2170
Table 3 Binder complex shear modulus (G Þ and phase angle ðd). Mix design
Temperature (degree C)
Complex shear modulus G (Pa)
Phase angle d (degree)
Mix-0
15 35 58 64
3,000,000 116,000 3550 1700
57 71 82 85
Mix-15
15 35 58 64
3,640,000 130,000 3570 1680
58 71 83 85
Mix-50
15 35 64 70
7,580,000 331,000 3690 1640
44 63 81 84
mix design properties, aggregate gradation, and performance grade of asphalt binder should be entered to the software. Other laboratory properties of mixes are estimated based on models and default data.
When Level 1 inputs cannot be obtained, asphalt mix properties will be estimated using predictive inputs models. These predictive models were globally calibrated and were function of volumetric properties of asphalt mixes [8].
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
4
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Table 4 Volumetric properties and aggregate gradations of the mixes. Mix ID Aggregate gradation
Mix properties
19 mm 16 mm 12.5 mm 9.5 mm 4.75 mm 2.0 mm 425 mm 180 mm 75 mm AC % VMA % Va % VFA % Gmm RAP % Extracted binder PG Mix-0 100 Mix-15 100 Mix-50 100
98.6 98.5 98.8
90.4 91.8 94.2
80.2 80.2 83.3
64.4 61.2 66.3
50.1 48.1 50.7
25.8 28.0 27.7
10.0 8.0 9.6
4.5 3.7 6.7
6.0 5.9 5.2
15.2 13.7 12.2
4.2 3.8 3.6
72.7 72.3 70.6
2.445 0 2.437 15 2.516 50
58–28 58–28 64–16
AC = percent asphalt content; VMA = voids in mineral aggregates; Va = percent air voids; VFA = voids filled with asphalt binder; Gmm= maximum theoretical specific gravity.
Table 5 Creep compliance values used in different Levels. Mix type
Creep compliance, 1/Gpa
Globally calibrated model Mix-0
Mix-15
Mix-50 ANN calibrated model Mix-0
Mix-15
Mix-50 Laboratory-measured creep compliance Mix-0
Mix-15
Mix-50
Time (s) Temp. °C
1 (s)
2 (s)
5 (s)
10 (s)
20 (s)
50 (s)
100 (s)
20 10 0 20 10 0 20 10 0
0.0288 0.0424 0.0553 0.0284 0.0409 0.0528 0.0235 0.0349 0.0453
0.0318 0.0500 0.0729 0.0313 0.0481 0.0694 0.0253 0.0400 0.0564
0.0363 0.0621 0.1051 0.0356 0.0596 0.0997 0.0278 0.0478 0.0753
0.0401 0.0732 0.1385 0.0393 0.0702 0.1311 0.0299 0.0548 0.0936
0.0443 0.0863 0.1825 0.0433 0.0826 0.1723 0.0322 0.0627 0.1165
0.0505 0.1072 0.2629 0.0493 0.1024 0.2475 0.0355 0.0750 0.1555
0.0558 0.1264 0.3465 0.0543 0.1204 0.3255 0.0381 0.0859 0.1934
20 10 0 20 10 0 20 10 0
0.0498 0.0836 0.1336 0.0479 0.0756 0.1461 0.0488 0.0760 0.1576
0.0514 0.0884 0.1459 0.0494 0.0802 0.1640 0.0503 0.0818 0.1806
0.0552 0.0974 0.1697 0.0521 0.0889 0.2005 0.0533 0.0937 0.2289
0.0588 0.1066 0.1958 0.0552 0.0978 0.2425 0.0561 0.1073 0.2869
0.0642 0.1189 0.2316 0.0598 0.1091 0.3020 0.0591 0.1249 0.3704
0.0753 0.1435 0.304 0.0684 0.1307 0.4239 0.0646 0.1617 0.5461
0.0884 0.1714 0.3867 0.0778 0.1540 0.5646 0.0704 0.2048 0.7533
20 10 0 20 10 0 20 10 0
0.0511 0.0812 0.1710 0.0488 0.0760 0.1576 0.0386 0.0567 0.1262
0.0539 0.0853 0.1902 0.0503 0.0818 0.1806 0.0397 0.0600 0.1453
0.0582 0.0932 0.2291 0.0533 0.0937 0.2289 0.0416 0.0664 0.1772
0.0632 0.1020 0.2717 0.0561 0.1073 0.2869 0.0435 0.0737 0.2239
0.0694 0.1131 0.3314 0.0591 0.1249 0.3704 0.0453 0.0834 0.3009
0.0808 0.1352 0.4533 0.0646 0.1617 0.5461 0.0494 0.1059 0.4501
0.0934 0.1586 0.5904 0.0704 0.2048 0.7533 0.0552 0.1327 0.6420
Previous studies showed that there is a need to calibrate these predictive models [1,4,2,3,6]. For instance, Pavement ME Design software estimates the dynamic modulus (E*) and creep compliance values of the mixes using globally calibrated predictive models when Level 3 input data are selected. These predictive models are not necessarily reliable and are required to be locally calibrated. 2. Calibration of creep compliance predictive model The MEPDG program implements a regression model to estimate creep compliance when laboratory measured values are not available. The MEPDG creep compliance prediction model is based on asphalt binder and volumetric
properties of an asphalt mix. The creep compliance model was developed based on 714 data points from 32 mixes [7]. This model requires to be calibrated based on local materials and mixes. The artificial neural network (ANN) method has been using as an alternative to calibrate the mix properties and distress predictive models [11,2,9]. Recently, Esfandiarpour and Shalaby [4] conducted a local calibration of the creep compliance model based on 861 data points from 41 specimens representing 14 types of asphalt concrete (AC) mixes which were prepared and tested in the laboratory. The authors reported that ANN calibrated model can significantly improve the accuracy of the creep compliance predictions. In this paper, the developed calibrated model (ANN model) will be used to predict
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
5
Table 6 Dynamic modulus values used in different Levels. Mix
Globally calibrated model
Mix-0
Mix-15
Mix-50
Nonlinear regression of NCHRP 1-37A
Mix-0
Mix-15
Mix-50
Laboratory-measured E*
Mix-0
Mix-15
Mix-50
Dynamic modulus (E*), MPa Temp., °C
0.1 Hz
0.5 Hz
1 Hz
5 Hz
10 Hz
25 Hz
10 5 25 40 10 5 25 40 10 5 25 40
16,725 5667 970 301 16,616 5641 968 302 27,428 11,262 2003 560
19,665 7778 1544 490 19,533 7738 1541 490 30,473 14,422 3087 910
20,907 8804 1873 604 20,763 8757 1868 604 31,697 15,860 3680 1119
23,678 11,410 2867 979 23,510 11,343 2857 977 34,316 19,298 5389 1783
24,810 12,609 3407 1199 24,632 12,534 3395 1196 35,345 20,794 6274 2163
26,240 14,244 4092 1558 26,050 14,156 4070 1554 36,615 22,759 7396 2769
10 5 25 40 10 5 25 40 10 5 25 40
21,691 8582 1520 481 22,672 9270 1746 575 26,605 13,410 2679 769
24,960 12,021 2636 844 25,960 12,831 2969 990 28,807 17,038 4398 1347
26,225 13,617 3299 1079 27,228 14,470 3687 1255 29,615 18,575 5349 1710
28,805 17,409 5340 1900 29,807 18,339 5866 2166 31,196 21,948 8041 2916
29,763 19,023 6442 2405 30,762 19,976 7030 2718 31,762 23,286 9386 3621
30,895 21,088 8043 3245 31,890 22,065 8748 3629 32,417 24,927 11,467 4745
10 5 25 40 10 5 25 40 10 5 25 40
19,285 7609 1922 400 18,565 7540 2530 638 29,227 10,170 2486 786
23,299 10,899 3149 647 22,234 10,553 4052 981 33,806 13,858 4026 1223
24,807 12,424 3940 829 23,707 12,026 4944 1201 35,585 15,395 4875 1512
28,174 16,222 6232 1550 26,883 15,547 7485 2028 39,420 19,680 7688 2599
29,857 17,990 7438 1957 28,287 17,131 8937 2547 41,210 21,283 9085 3298
32,265 20,267 9457 3016 29,813 19,314 10,695 3479 42,866 23,827 10,954 4308
the creep compliance values for calibrated Level 3 input data. Fig. 1 compares the accuracy of predictions between globally calibrated model and ANN calibrated model [4]. 3. Calibration of dynamic modulus model The NCHRP 1-37A and NCHRP 1-40D models have been incorporated into the MEPDG program to estimate E* when Level 2 and Level 3 inputs for asphalt mix and asphalt binder are used in the design and analysis of pavement structures [1]. A total of 2750 data points from asphalt mixes containing unmodified and modified asphalt binders were used in developing the coefficients of the NCHRP 1-37A model. This model assumes a sigmoid function of inputs for the AC mix. This model was constructed based upon asphalt binder viscosity and asphalt mix volumetric properties [8]. NCHRP 1-40D model was developed based on 7400 data points from 346 mixes to predict the E* of asphalt
mixes. This model is a sigmoid function of volumetric properties and aggregate gradation similar to NCHRP 137A model, however, complex shear modulus (G*) and phase angle (d) of asphalt binder were used to characterize the asphalt binder instead of asphalt binder viscosity [12,10]. Esfandiarpour and Shalaby [5] analyzed the reliability of these models based on 1386 laboratory measured data points obtained from 51 specimens from 17 types of local AC mixes containing 0–50% RAP. The study showed that the nonlinear calibration of NCHRP 1-37A yielded the most accurate predicted dynamic modulus values. Fig. 2 compares the prediction accuracy between globally calibrated and nonlinear regression of NCHRP 1-37A E* model. The nonlinear regression of NCHRP 1-37A was selected as the calibration alternative to be used for calibrated Level 3 input data in this paper. The aim of this paper is to compare the differences in predicted asphalt distresses using Pavement ME Design
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
6
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Table 7 Predicted performance of Mix-0 for Level 1, Level 3, and calibrated Level 3. Distresses
Design criteria
Input alternatives Level 3
Calibrated model
Level 1
Terminal IRI
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
2.5 2.8 78.1 Fail 16
2.5 2.53 88.8 Fail 19
2.5 2.53 88.8 Fail 19
AC surface down cracking
Target (m/km) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
380 506.2 82.3 Fail 10
380 64 100 Pass >20
380 50.8 100 Pass >20
AC bottom up cracking
Target (%) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
15 11.6 95.86 Pass >20
15 1.6 100 Pass >20
15 1.7 100 Pass >20
AC thermal fracture
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
200 608.6 11 Fail <2
200 491 0.2 Fail <2
200 491 0.2 Fail <2
Permanent deformation (AC rutting only)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
12 6.8 100 Pass >20
12 0.84 100 Pass >20
12 0.9 100 Pass >20
Permanent deformation (total pavement rutting)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
19 19.9 81.9 Fail 17
19 12 100 Pass >20
19 11.8 100 Pass >20
software when calibrated E* and creep compliance prediction models were utilized over Level 3 E* and creep compliance prediction models. For this purpose, the locally calibrated models, previously developed by [4,5] are used as calibrated Level 3 asphalt input data [4,5]. The objectives of this paper are to evaluate the impact of locally and globally calibrated models on the performance of asphalt mixes using Pavement ME Design software, and to assess the sensitivity of the predicted distresses to mechanical mix properties as well as RAP content of asphalt mixes. Fig. 3 shows the hypothesis of this paper.
(AADTT) of 1660. It was assumed that 50% of total trucks is in the design direction and 100% trucks of each direction is on the design lane. The design life of the pavement is assumed to be 20 years. Winnipeg was selected as the project environmental condition and its historical climate data were used in all analysis. The typical Manitoba structural pavement thicknesses were estimated based on subgrade strength and traffic level. In order to evaluate the asphalt mix performance, the thicknesses for all layers were kept unchanged. 4.1. Materials inputs
4. Design example Three mixes containing 0–50% RAP (Mix-0, Mix-15, and Mix-50) were selected to evaluate their performance as predicted by Pavement ME Design software. The number after each mix is indicating the percent of RAP in the mix. A section of Highway 1 located at west of the city of Winnipeg in Manitoba, Canada, was selected as the design example presented in this paper. This section is a four-lane divided expressway with annual average daily truck traffic
Typical properties of Manitoba unbound materials were used for all the analysis. Table 2 shows the subgrade, subbase, and base material properties and their layer thicknesses. 4.2. Asphalt binder and asphalt mix inputs The thickness of asphalt mix layer was 150 mm for all the mixes. Complex shear modulus (G*) and phase angle (d) for three mixes were measured using dynamic shear
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
7
Table 8 Predicted performance of Mix-15 for Level 1, Level 3, and calibrated Level 3. Distresses
Design criteria
Input alternatives Level 3
Calibrated model
Level 1
Terminal IRI
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
2.5 2.8 78.1 Fail 16
2.5 2.53 88.8 Fail 19
2.5 2.53 88.8 Fail 19
AC surface down cracking
Target (m/km) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
380 468.5 84.6 Fail 12
380 62.1 100 Pass >20
380 50.3 100 Pass >20
AC bottom up cracking
Target (%) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
15 6.2 99.9 Pass >20
15 1.62 100 Pass >20
15 1.6 100 Pass >20
AC thermal fracture
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
200 608.6 11 Fail <2
200 491 0.2 Fail <2
200 491 0.2 Fail <2
Permanent deformation (AC rutting only)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
12 6.8 100 Pass >20
12 0.8 100 Pass >20
12 0.9 100 Pass >20
Permanent deformation (total pavement rutting)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
19 19.9 82.9 Fail 17
19 11.9 100 Pass >20
19 11.8 100 Pass >20
rheometer (DSR). Binder performance grade inputs are shown in Table 3. The volumetric properties and aggregate gradations of these mixes are shown in Table 4. 4.3. Dynamic modulus and creep compliance input data To assess the impact of local and global calibration of E* and creep models, three Levels of asphalt input data were considered; Level 1, calibrated Level 3, and Level 3. For Level 1, the laboratory measured dynamic modulus and creep compliance values were used while for the Level 3, the globally calibrated models were used to estimate theses values. For calibrated Level 3, dynamic modulus and creep compliance values were predicted based on the calibrated models, developed by [4,5], and manually entered into Pavement ME Design software as a Level 1 analysis. The globally and locally calibrated creep compliance and dynamic modulus values as well as laboratorymeasured values used in the analysis of mixes are shown in Tables 5 and 6, respectively. Since the E* values of Mix-50 at 10 °C exceeded the maximum allowable of dynamic modulus input in this software, dynamic modulus at 10 °C was not used in Pavement ME Design software.
5. AC predicted distresses for different Levels For all the three mixes, the same subgrade, subbase and base properties and thicknesses were used to evaluate the impact of asphalt binder and mix properties in Level 1, Level 3, and calibrated Level 3. The reliability for each distress was selected to be 90% for 20 years pavement design life. The reliability is dependent on the standard error of distress prediction model. When a high reliability is selected for predicting a pavement distress, a higher threshold is used to compute the pavement distresses resulting in higher predicted distresses [1]. The distress criteria of predicted distresses were selected based on MEPDG manual of practice recommendation. Tables 7 through 9 summarized the comparison of distress prediction among Level 1, Level 3, and calibrated Level 3 input data. The limits of international roughness index (IRI) and AC thermal cracking were chosen as 2.5 m/km and 200 m/km, respectively. The maximum value of AC surface down cracking (longitudinal cracking) and AC bottom up cracking (alligator cracking) were selected as 380 m/ km and 15%, respectively. The limits for permanent deformation (AC only), and permanent deformation (total pave-
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
8
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Table 9 Predicted performance of Mix-50 for Level 1, Level 3, and calibrated Level 3. Distresses
Design criteria
Input alternatives Level 3
Calibrated model
Level 1
Terminal IRI
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
2.5 2.7 81.5 Fail 17
2.5 2.6 87.7 Fail 19
2.5 2.6 87.6 Fail 19.5
AC surface down cracking
Target (m/km) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
380 316.7 93.9 Pass >20
380 51.3 100 Pass >20
380 51.7 100 Pass >20
AC bottom up cracking
Target (%) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
15 3.1 100 Pass >20
15 1.8 100 Pass >20
15 1.8 100 Pass >20
AC thermal fracture
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
200 608.6 11 Fail <1
200 491 0.2 Fail <1
200 491 0.2 Fail <1
Permanent deformation (AC rutting only)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
12 5 100 Pass >20
12 1.3 100 Pass >20
12 1.4 100 Pass >20
Permanent deformation (total pavement rutting)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
19 17.7 96.5 Pass >20
19 12.9 100 Pass >20
19 13 100 Pass >20
ment rutting) were selected as 12 mm and 19 mm, respectively. As it can be seen from Tables 7–9, the calibrated Level 1 input data showed better-predicted performance when it was compared to Level 3 input data for all mixes regardless of RAP content. When Level 3 was used for Mix-0 and Mix-15, all the predicted distresses failed to meet the target value except AC permanent deformation. In contrast, when Level 1 or calibrated Level 3 was used as input data, only IRI and AC thermal cracking failed to meet the target value. AC thermal cracking failed to meet the target 200 m/ km for all mixes. This indicated that the binder type used in these mixes was not adequate for this location. The predicted life at 90% reliability for each distress was higher for Level 1 and calibrated Level 3 asphalt mix data compared to the Level 3 asphalt mix data, regardless of the distress prediction passed or failed to meet the target values. In other words, the amounts of predicted distresses were lower for Level 1 and calibrated Level 3 asphalt data compared to the Level 3 asphalt data. All the predicted distresses from each Level were normalized by dividing the predicted distress values obtained from Level 3 and calibrated Level 3 to corresponding
predicted distress values obtained from Level 1. Figs. 4–6 show a relative difference in predicted distresses of each mix for three Levels of input data. The predicted performance of Level 3 showed a higher normalized number in predicted rutting and fatigue distresses compared to calibrated Level 3. This indicates that Level 3 overpredicts the distress predictions of the asphalt mix. The predicted performance of calibrated Level 3 was found similar to Level 1 input data for all three mixes. This implies that utilizing calibrated models improves the reliability of the MEPDG distress predictions. Therefore, calibrated Level 3 asphalt input data can be used for the analysis and design of mixes with comparable accuracy of Level 1 input data. This can save substantial time and cost since it is time consuming and expensive to conduct laboratory asphalt testing for each mix. 6. Distress predictions of mixes containing RAP for calibrated Level 3 The predicted distresses of asphalt mixes for calibrated Level 3 were plotted to evaluate the sensitivity of the asphalt mix properties and RAP content. Fig. 7 shows
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Normalized distress prediction
Default Values (Level 3)
Calibrated Model
9
Level 1
11.00 10.00 9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 Terminal IRI
AC Surface AC Bottom Up Down Cracking Cracking
AC Thermal Fracture
AC Rutting
Total Pavement Rutting
Fig. 4. Long-term performance of Mix-0 for three Levels of input.
Normalized distress prediction
Default Values (Level 3)
Calibrated Model
Level 1
11.00 10.00 9.00
8.00 7.00 6.00 5.00 4.00 3.00 2.00
1.00 0.00 Terminal IRI
AC Surface AC Bottom Up AC Thermal Down Cracking Cracking Fracture
AC Rutting Total Pavement Rutting
Fig. 5. Long-term performance of Mix-15 for three Levels of input.
Normalized distress prediction
Default Values (Level 3)
Calibrated Model
Level 1
11.00 10.00 9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 Terminal IRI
AC Surface AC Bottom Up AC Thermal Down Cracking Cracking Fracture
AC Rutting Total Pavement Rutting
Fig. 6. Long-term performance of Mix-50 for three Levels of input.
the IRI performance of the mixes. The terminal IRI of all three mixes were slightly more than 2.5 at the end of 20 years pavement design life. It should be noted that MEPDG roughness prediction is affected by the predicted extent of cracking and other distresses. Fig. 8 shows the predicted alligator cracking of three mixes over pavement life. All the mixes successfully met
the alligator cracking limit. The predicted alligator cracking of the mixes did not show noticeable differences. Fig. 9 shows the predicted longitudinal cracking of three mixes over pavement life. Mix-50 showed slightly lower predicted longitudinal cracking in comparison to the other mixes. Predicted thermal cracking of the mixes are shown in Fig. 10. Fig. 10 illustrates that all three mixes failed to
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
10
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx 3
IRI (m/Km)
2.5 2 1.5
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
1 0.5 0 0
24
48
72
96
120
144
168
192
216
240
Pavement Age (month) Fig. 7. Predicted IRI and pavement life for mixes.
Alligator Cracking (%)
17 15 13 11 9
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
7 5 3 1 0
24
48
72
96
120
144
Pavement Age (month)
168
192
216
240
Fig. 8. Predicted alligator cracking and pavement life of mixes.
Longitudinal Cracking (m/Km)
430 380 330 280
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
230 180 130 80 30 0
24
48
72
96
120
144
168
192
216
240
Pavement Age (month) Fig. 9. Predicted longitudinal cracking and pavement life of mixes.
Thermal Cracking (m/Km)
600 500
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
400 300 200 100 0 0
24
48
72
96
120
144
168
192
216
240
Pavement Age (month) Fig. 10. Predicted thermal cracking and pavement life of mixes. Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
11
14
AC Rut (mm)
12 10 8
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
6 4 2 0 0
24
48
72
96
120
144
Pavement Age (month)
168
192
216
240
Total Rut (mm)
Fig. 11. Predicted AC rut and pavement life of mixes.
20 18 16 14 12 10 8 6 4 2 0
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value 0
24
48
72
96
120
144
168
192
216
240
Pavement Age (month) Fig. 12. Predicted total rut and pavement life of mixes.
1.E+05
Dynamic Modulus (MPa)
1.E+03
1.E+02
Mix-0 Mix-15 Mix-50
1.E+01
1.E+00 1.E-05
Creep compliance (1/GPa)
1.E+00
1.E+04
1.E-01
1.E-02
1.E-03 1.E-02
Mix-0 Mix-15 Mix-50 1.E-01
1.E+00
1.E+01
1.E+02
1.E+03
Reduced time (sec.) 1.E-03
1.E-01
1.E+01
1.E+03
1.E+05
Reduced Freq. (Hz)
Fig. 14. Creep compliance master curves of mixes obtained from ANN calibrated model.
Fig. 13. Dynamic modulus master curves of mixes obtained from nonlinear regression of NCHRP 1-37A.
meet the thermal cracking criterion. Mix-50 exceeded the thermal criterion before the first year of pavement life which is expected since the binder performance grade of Mix-50 was PG 64–16 while the binder performance grade of Mix-0 and Mix-15 was PG 58–28. Figs. 11 and 12 show the predicted AC rut and total rut for the mixes, respectively. All the mixes successfully met the distress rut limit. It was noted that the AC rut for all mixes was noticeably low. Moreover, the AC rut of
Table 10 Predicted distresses for RAP mixes for calibrated Level 3. Predicted distresses
Asphalt mixes Mix-0
Mix-15
Mix-50
Terminal IRI (m/km) AC surface down cracking (m/km) AC bottom up cracking (%) AC thermal fracture (m/km) AC rutting only (mm) Total pavement rutting (mm)
2.53 64.00 1.60 491.00 0.84 12.00
2.53 62.10 1.62 491.00 0.8 11.90
2.60 51.30 1.80 491.00 1.30 12.90
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
12
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Mix-50 was slightly higher than Mix-0 and Mix-15, which was not expected. When a type of distress occurs in a pavement, it will cause the deterioration of the asphalt pavement to accelerate. The possible interactions between distresses did not appear to fully consider in the analysis of the mixes. When the AC thermal cracking failed to meet the target value, it was expected to affect the other distresses, however, no corresponding impact or acceleration in prediction trends of other distresses can be observed. This indicates that all distresses were predicted independently and the effect of one distress was not considered in prediction of other distresses. 7. Sensitivity of predicted distresses to the mix properties and RAP content The master curves of dynamic modulus and creep compliance of RAP mixes were constructed to show the possible difference in asphalt mix properties, and to determine the sensitivity of distress predictions to mix properties inputs and RAP content. Figs. 13 and 14 show the master curves of dynamic modulus and creep compliance for RAP mixes obtained from calibrated models. Fig. 13 shows that the dynamic modulus of Mix-50 is higher than Mix-0 and Mix-15 at high temperatures (Fig. 13). In addition, this mix has much lower creep compliance at low temperature when it is compared to Mix-0 and Mix-15 as shown in Fig. 14. In contrast, the dynamic modulus and creep compliance values of Mix-0 and Mix-15 were found to be very similar. It is expected that Pavement ME Design software would predict the asphalt distresses of Mix-50 to be different from those of Mix-0 and Mix-15, since Mix-50 has different mechanical properties and is a much stiffer mix. However, the predicted distresses of all mixes were found to be very similar as shown in Table 10. This indicated that an increase in RAP content did not noticeably impact the predicted distresses, although the fundamental mechanical properties, dynamic modulus and creep compliance, were substantially different. Furthermore, the total predicted rutting was found slightly higher for high RAP mix compared to Mix-0 and Mix-15. This could be explained by the lack of sensitivity in the rutting model to RAP content, the low truck traffic volume and traffic load, and that the volumetric mix properties of three AC mixes are very similar. In the previous section, it was shown that the use of locally calibrated models provides reliable results which are comparable to results obtained from Level 1 input data. However, the similarity in the distress predictions of the mixes with different mechanical properties and RAP content implied that the distress predictive models must be also locally calibrated based on field performance of mixes containing RAP. It is expected that locally calibrated distress models would be more responsive to the impact of different asphalt mix properties and RAP content.
8. Conclusions This paper presents the long-term performance of three asphalt mixes; Mix-0, Mix-15, and Mix-50 for three Levels of asphalt input data using Pavement ME Design software. Level 3, Calibrated Level 3, and Level 1 were used as three Levels of asphalt input data. For Level 1 input data, the laboratory-measured E* and creep compliance values were used whereas the Level 3 input data used globally calibrated predictive E* and creep models to estimate these values. The calibrated dynamic modulus and creep compliance models, developed by [4,5], were used for calibrated Level 3 input data. Based on the results, the following conclusions were drawn: The predicted performance of Level 3 showed higher amounts of distresses compared to those of Level 1 and calibrated Level 3. The Level 3 noticeably overpredicted the distress predictions of the asphalt mix. The results of using calibrated Level 3 inputs were very similar to those of Level 1 Inputs. As would be expected, the use of calibrated Level 3 input data instead of Level 3 increased the accuracy of predicted distresses. The calibrated Level 3 asphalt input data can be used for the design and analysis of mixes with comparable accuracy of Level 1 input data. This can save a substantial amount of time and cost since it is time consuming and expensive to conduct laboratory asphalt testing for each mix. Although the mechanical properties of the Mix-0 with no RAP and Mix-50, containing high RAP, were noticeably different, Pavement ME Design software predicted very similar distresses. Increase in RAP amounts did not significantly change the predicted distresses of the RAP mixes. It was noted that all distresses were predicted independently. The possible interactions between predicted distresses were not modeled. For instance, the effect of excessive thermal cracking was not considered in the prediction of other distresses. It is recommended to locally calibrate the distress models (transfer functions) based on field performance of local mixes containing RAP. It is expected that the calibrated distress models would be more responsive to the impact of different asphalt mix properties and RAP content.
References [1] American Association of State Highway and Transportation Officials (AASHTO), Mechanistic-389 Empirical Pavement Design Guide: A Manual of Practice. Washington, D.C., 2015. [2] T.R. Clyne, X. Li, M.O. Marasteanu, E.L. Skok, Dynamic and Resilient Modulus of MN/DOT Asphalt Mixtures, University of Minnesota, Minneapolis, MN, 2009.
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx [3] S. El-Badawy, F. Bayomy, A. Awed, Performance of MEPDG dynamic modulus predictive models for asphalt concrete mixtures: local calibration for Idaho, J. Mater. Civ. Eng. 1412–1421 (2012). [4] S. Esfandiarpour, A. Shalaby, Local calibration of creep compliance models of asphalt concrete, Constr. Build. Mater. 132 (2017) 313–322. [5] S. Esfandiarpour, A. Shalaby, Alternatives for calibration of dynamic modulus prediction models of asphalt concrete, Int. J. Pavement Res. Technol. 10 (3) (May 2017). [6] K. Georgouli, A. Loizos, C. Plati, Calibration of dynamic modulus predictive model, Constr. Build. Mater. 102 (2016) 65–75. [7] National Cooperative Highway Research Program, Revision of Level 3 Thermal Fracture Predictive Models for the M-EPDG Version 1.0 and Final Level 1, 2, 3 Thermal Fracture Model Recalibration NCHRP 1-40D, Final Report, Transportation Research Board, Washington, D.C., 2006. [8] National Cooperative Highway Research Program, Guide for Mechanistic-Empirical Design of New and Rehabilitated Pavement
[9]
[10]
[11]
[12]
13
Structures NCHRP 1-37A Final Report, Transportation Research Board, Washington, D.C., 2004. D. Singh, M. Zaman, S. Commuri, Artificial neural network modeling for dynamic modulus of hot mix asphalt using aggregate shape properties, J. Mater. Civ. Eng. 25 (1) (2012) 54–62. M. Witczak, M. El-Basyouny, S. El-Badawy, Incorporation of the new, E* (2005) Predictive Model in the MEPDG NCHRP 1-40D Inter-Team Technical Report, Arizona State University, Tempe, AZ, 2006. Zeghal, M., Modeling the Creep Compliance of Asphalt Concrete Using the Artificial Neural Network Technique. GeoCongress. Characterization, Monitoring, and, Modeling of Geosystems. New Orleans, Louisiana, United States, 2008, pp. 910–916. J. Bari, M.W. Witczak, Development of a new revised version of the Witczak E* predictive model for hot mix asphalt mixtures, J. Assoc. Asphalt Paving Technol. 75 (2006) 381–424.
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
ScienceDirect International Journal of Pavement Research and Technology xxx (2018) xxx–xxx www.elsevier.com/locate/IJPRT
Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP Saman Esfandiarpour ⇑, Ahmed Shalaby Department of Civil Engineering, University of Manitoba, 15 Gilson Street, Winnipeg, Manitoba R3T 5V6, Canada Received 31 March 2017; received in revised form 13 March 2018; accepted 12 April 2018
Abstract MEPDG software can predict long term performance of the asphalt mixes based on asphalt input data. When laboratory measured data of an asphalt mix are not available, this software uses the predictive models to estimate the mix properties. AASHTO recommended calibrating these input models based on local materials and mixes. Previous studies showed that local calibration of dynamic modulus (E*) and creep compliance predictive model improved the reliability of the predictions. The purpose of this paper is to evaluate the impact of local calibration of E* and creep compliance models on long term performance of mixes containing reclaimed asphalt pavement (RAP), and to assess the sensitivity of the predicted distresses to RAP content. Three Levels of asphalt input data were considered; Level 1, calibrated Level 3, and Level 3. For calibrated Level 3, the predicted E* and creep compliance obtained from calibrated models were used as input data. The results showed that the Level 3 input data tend to overpredict the distress predictions of the asphalt mix compared to calibrated Level 3 or Level 1 asphalt input data. It was found that the calibrated Level 3 asphalt input data can be used for the design and analysis of mixes with comparable accuracy of Level 1 input. As conducting laboratory tests for individual mixes is expensive and time consuming, utilizing reliable calibrated models to predict E* and creep compliance can substantially reduce operating and testing expenses. Also, it was found that the predicted distresses are not sensitive to the RAP content. Ó 2018 Production and hosting by Elsevier B.V. on behalf of Chinese Society of Pavement Engineering. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction Mechanistic-Empirical Pavement Design Guide (MEPDG) software called Pavement ME Design uses the MEPDG to design and analyze flexible and rigid pavement structures. Different design traffic loadings, climate data and material properties are used as inputs into this software to predict future pavement distresses. The flexible pavement distresses included in this software are surface roughness,
⇑ Corresponding author.
E-mail addresses: [email protected] (S. Esfandiarpour), [email protected] (A. Shalaby). Peer review under responsibility of Chinese Society of Pavement Engineering.
total permanent deformation (rutting), asphalt layer permanent deformation, asphalt bottom-up fatigue cracking, asphalt top-down fatigue cracking, and asphalt thermal cracking. The distress prediction models require inputs to be defined by the user. The defined inputs include asphalt, base, subbase and subgrade materials characteristics, traffic loadings, and climate data for a pavement structure. Based upon the quality and quantity of the available material properties data, there are three levels of input options in the Pavement ME Design software [8]. Level 1 input option generally requires site specific material properties data which are obtained through laboratory or field testing. These data have the highest level of reliability, and are expected to provide the optimum design and analysis. Level 2 inputs have an intermediate level of
https://doi.org/10.1016/j.ijprt.2018.04.002 1996-6814/Ó 2018 Production and hosting by Elsevier B.V. on behalf of Chinese Society of Pavement Engineering. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
2
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Table 1 Required input asphalt mix and asphalt binder data for Level 1 and Level 3. Input data
Test measurement
Asphalt mix
Mix design (VMA %, Va %, and asphalt content percentages) Aggregate gradation Dynamic modulus Creep compliance (20 °C, 10 °C, and 0 °C) Indirect Tensile (10 °C)
Performance grade Complex shear modulus (G*) Phase angle (d)
Asphalt binder
Input alternatives
a) Globally calibrated model
Level 1 (laboratory-measured values)
Level 3 (default values)
b) ANN calibrated model
Fig. 1. Comparison of globally calibrated and ANN calibrated creep compliance model.
a) Globally calibrated NCHRP 1-37A
b) Nonlinear regression of NCHRP 1-37A
Fig. 2. Comparison of dynamic modulus model predictions.
reliability. These input data are generally obtained through limited laboratory or field testing, or estimated from the correlations with other measured properties. Level 3 inputs have the lowest level of reliability since the typical agency data or software default data are used [1]. Although Level 1 inputs for the AC mixes provide more reliable results than Level 2 and Level 3 inputs, the comprehensive laboratory
testing required to obtain them is time consuming and expensive. Table 1 shows the required input asphalt mix and asphalt binder data for Level 1 and Level 3. For Level 1 input data, all tests and volumetric properties are required except the aggregate gradation and performance grade. However, for Level 3, basic properties of a mix such as
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
3
Fig. 3. The hypothesis of this paper.
Table 2 Unbound materials properties. Properties
Subgrade
Subbase
Base
Materials Type Thickness (mm) Resilient modulus (MPa) Moisture content (%) Liquid limit Plasticity index Maximum dry density (Kg/m3 )
A-7-6 – 30 29 75 42 1410
Crushed lime stone C-base 300 120 8 11 0 2220
Crushed lime stone A-base 200 140 10.8 6 1 2170
Table 3 Binder complex shear modulus (G Þ and phase angle ðd). Mix design
Temperature (degree C)
Complex shear modulus G (Pa)
Phase angle d (degree)
Mix-0
15 35 58 64
3,000,000 116,000 3550 1700
57 71 82 85
Mix-15
15 35 58 64
3,640,000 130,000 3570 1680
58 71 83 85
Mix-50
15 35 64 70
7,580,000 331,000 3690 1640
44 63 81 84
mix design properties, aggregate gradation, and performance grade of asphalt binder should be entered to the software. Other laboratory properties of mixes are estimated based on models and default data.
When Level 1 inputs cannot be obtained, asphalt mix properties will be estimated using predictive inputs models. These predictive models were globally calibrated and were function of volumetric properties of asphalt mixes [8].
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
4
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Table 4 Volumetric properties and aggregate gradations of the mixes. Mix ID Aggregate gradation
Mix properties
19 mm 16 mm 12.5 mm 9.5 mm 4.75 mm 2.0 mm 425 mm 180 mm 75 mm AC % VMA % Va % VFA % Gmm RAP % Extracted binder PG Mix-0 100 Mix-15 100 Mix-50 100
98.6 98.5 98.8
90.4 91.8 94.2
80.2 80.2 83.3
64.4 61.2 66.3
50.1 48.1 50.7
25.8 28.0 27.7
10.0 8.0 9.6
4.5 3.7 6.7
6.0 5.9 5.2
15.2 13.7 12.2
4.2 3.8 3.6
72.7 72.3 70.6
2.445 0 2.437 15 2.516 50
58–28 58–28 64–16
AC = percent asphalt content; VMA = voids in mineral aggregates; Va = percent air voids; VFA = voids filled with asphalt binder; Gmm= maximum theoretical specific gravity.
Table 5 Creep compliance values used in different Levels. Mix type
Creep compliance, 1/Gpa
Globally calibrated model Mix-0
Mix-15
Mix-50 ANN calibrated model Mix-0
Mix-15
Mix-50 Laboratory-measured creep compliance Mix-0
Mix-15
Mix-50
Time (s) Temp. °C
1 (s)
2 (s)
5 (s)
10 (s)
20 (s)
50 (s)
100 (s)
20 10 0 20 10 0 20 10 0
0.0288 0.0424 0.0553 0.0284 0.0409 0.0528 0.0235 0.0349 0.0453
0.0318 0.0500 0.0729 0.0313 0.0481 0.0694 0.0253 0.0400 0.0564
0.0363 0.0621 0.1051 0.0356 0.0596 0.0997 0.0278 0.0478 0.0753
0.0401 0.0732 0.1385 0.0393 0.0702 0.1311 0.0299 0.0548 0.0936
0.0443 0.0863 0.1825 0.0433 0.0826 0.1723 0.0322 0.0627 0.1165
0.0505 0.1072 0.2629 0.0493 0.1024 0.2475 0.0355 0.0750 0.1555
0.0558 0.1264 0.3465 0.0543 0.1204 0.3255 0.0381 0.0859 0.1934
20 10 0 20 10 0 20 10 0
0.0498 0.0836 0.1336 0.0479 0.0756 0.1461 0.0488 0.0760 0.1576
0.0514 0.0884 0.1459 0.0494 0.0802 0.1640 0.0503 0.0818 0.1806
0.0552 0.0974 0.1697 0.0521 0.0889 0.2005 0.0533 0.0937 0.2289
0.0588 0.1066 0.1958 0.0552 0.0978 0.2425 0.0561 0.1073 0.2869
0.0642 0.1189 0.2316 0.0598 0.1091 0.3020 0.0591 0.1249 0.3704
0.0753 0.1435 0.304 0.0684 0.1307 0.4239 0.0646 0.1617 0.5461
0.0884 0.1714 0.3867 0.0778 0.1540 0.5646 0.0704 0.2048 0.7533
20 10 0 20 10 0 20 10 0
0.0511 0.0812 0.1710 0.0488 0.0760 0.1576 0.0386 0.0567 0.1262
0.0539 0.0853 0.1902 0.0503 0.0818 0.1806 0.0397 0.0600 0.1453
0.0582 0.0932 0.2291 0.0533 0.0937 0.2289 0.0416 0.0664 0.1772
0.0632 0.1020 0.2717 0.0561 0.1073 0.2869 0.0435 0.0737 0.2239
0.0694 0.1131 0.3314 0.0591 0.1249 0.3704 0.0453 0.0834 0.3009
0.0808 0.1352 0.4533 0.0646 0.1617 0.5461 0.0494 0.1059 0.4501
0.0934 0.1586 0.5904 0.0704 0.2048 0.7533 0.0552 0.1327 0.6420
Previous studies showed that there is a need to calibrate these predictive models [1,4,2,3,6]. For instance, Pavement ME Design software estimates the dynamic modulus (E*) and creep compliance values of the mixes using globally calibrated predictive models when Level 3 input data are selected. These predictive models are not necessarily reliable and are required to be locally calibrated. 2. Calibration of creep compliance predictive model The MEPDG program implements a regression model to estimate creep compliance when laboratory measured values are not available. The MEPDG creep compliance prediction model is based on asphalt binder and volumetric
properties of an asphalt mix. The creep compliance model was developed based on 714 data points from 32 mixes [7]. This model requires to be calibrated based on local materials and mixes. The artificial neural network (ANN) method has been using as an alternative to calibrate the mix properties and distress predictive models [11,2,9]. Recently, Esfandiarpour and Shalaby [4] conducted a local calibration of the creep compliance model based on 861 data points from 41 specimens representing 14 types of asphalt concrete (AC) mixes which were prepared and tested in the laboratory. The authors reported that ANN calibrated model can significantly improve the accuracy of the creep compliance predictions. In this paper, the developed calibrated model (ANN model) will be used to predict
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
5
Table 6 Dynamic modulus values used in different Levels. Mix
Globally calibrated model
Mix-0
Mix-15
Mix-50
Nonlinear regression of NCHRP 1-37A
Mix-0
Mix-15
Mix-50
Laboratory-measured E*
Mix-0
Mix-15
Mix-50
Dynamic modulus (E*), MPa Temp., °C
0.1 Hz
0.5 Hz
1 Hz
5 Hz
10 Hz
25 Hz
10 5 25 40 10 5 25 40 10 5 25 40
16,725 5667 970 301 16,616 5641 968 302 27,428 11,262 2003 560
19,665 7778 1544 490 19,533 7738 1541 490 30,473 14,422 3087 910
20,907 8804 1873 604 20,763 8757 1868 604 31,697 15,860 3680 1119
23,678 11,410 2867 979 23,510 11,343 2857 977 34,316 19,298 5389 1783
24,810 12,609 3407 1199 24,632 12,534 3395 1196 35,345 20,794 6274 2163
26,240 14,244 4092 1558 26,050 14,156 4070 1554 36,615 22,759 7396 2769
10 5 25 40 10 5 25 40 10 5 25 40
21,691 8582 1520 481 22,672 9270 1746 575 26,605 13,410 2679 769
24,960 12,021 2636 844 25,960 12,831 2969 990 28,807 17,038 4398 1347
26,225 13,617 3299 1079 27,228 14,470 3687 1255 29,615 18,575 5349 1710
28,805 17,409 5340 1900 29,807 18,339 5866 2166 31,196 21,948 8041 2916
29,763 19,023 6442 2405 30,762 19,976 7030 2718 31,762 23,286 9386 3621
30,895 21,088 8043 3245 31,890 22,065 8748 3629 32,417 24,927 11,467 4745
10 5 25 40 10 5 25 40 10 5 25 40
19,285 7609 1922 400 18,565 7540 2530 638 29,227 10,170 2486 786
23,299 10,899 3149 647 22,234 10,553 4052 981 33,806 13,858 4026 1223
24,807 12,424 3940 829 23,707 12,026 4944 1201 35,585 15,395 4875 1512
28,174 16,222 6232 1550 26,883 15,547 7485 2028 39,420 19,680 7688 2599
29,857 17,990 7438 1957 28,287 17,131 8937 2547 41,210 21,283 9085 3298
32,265 20,267 9457 3016 29,813 19,314 10,695 3479 42,866 23,827 10,954 4308
the creep compliance values for calibrated Level 3 input data. Fig. 1 compares the accuracy of predictions between globally calibrated model and ANN calibrated model [4]. 3. Calibration of dynamic modulus model The NCHRP 1-37A and NCHRP 1-40D models have been incorporated into the MEPDG program to estimate E* when Level 2 and Level 3 inputs for asphalt mix and asphalt binder are used in the design and analysis of pavement structures [1]. A total of 2750 data points from asphalt mixes containing unmodified and modified asphalt binders were used in developing the coefficients of the NCHRP 1-37A model. This model assumes a sigmoid function of inputs for the AC mix. This model was constructed based upon asphalt binder viscosity and asphalt mix volumetric properties [8]. NCHRP 1-40D model was developed based on 7400 data points from 346 mixes to predict the E* of asphalt
mixes. This model is a sigmoid function of volumetric properties and aggregate gradation similar to NCHRP 137A model, however, complex shear modulus (G*) and phase angle (d) of asphalt binder were used to characterize the asphalt binder instead of asphalt binder viscosity [12,10]. Esfandiarpour and Shalaby [5] analyzed the reliability of these models based on 1386 laboratory measured data points obtained from 51 specimens from 17 types of local AC mixes containing 0–50% RAP. The study showed that the nonlinear calibration of NCHRP 1-37A yielded the most accurate predicted dynamic modulus values. Fig. 2 compares the prediction accuracy between globally calibrated and nonlinear regression of NCHRP 1-37A E* model. The nonlinear regression of NCHRP 1-37A was selected as the calibration alternative to be used for calibrated Level 3 input data in this paper. The aim of this paper is to compare the differences in predicted asphalt distresses using Pavement ME Design
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
6
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Table 7 Predicted performance of Mix-0 for Level 1, Level 3, and calibrated Level 3. Distresses
Design criteria
Input alternatives Level 3
Calibrated model
Level 1
Terminal IRI
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
2.5 2.8 78.1 Fail 16
2.5 2.53 88.8 Fail 19
2.5 2.53 88.8 Fail 19
AC surface down cracking
Target (m/km) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
380 506.2 82.3 Fail 10
380 64 100 Pass >20
380 50.8 100 Pass >20
AC bottom up cracking
Target (%) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
15 11.6 95.86 Pass >20
15 1.6 100 Pass >20
15 1.7 100 Pass >20
AC thermal fracture
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
200 608.6 11 Fail <2
200 491 0.2 Fail <2
200 491 0.2 Fail <2
Permanent deformation (AC rutting only)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
12 6.8 100 Pass >20
12 0.84 100 Pass >20
12 0.9 100 Pass >20
Permanent deformation (total pavement rutting)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
19 19.9 81.9 Fail 17
19 12 100 Pass >20
19 11.8 100 Pass >20
software when calibrated E* and creep compliance prediction models were utilized over Level 3 E* and creep compliance prediction models. For this purpose, the locally calibrated models, previously developed by [4,5] are used as calibrated Level 3 asphalt input data [4,5]. The objectives of this paper are to evaluate the impact of locally and globally calibrated models on the performance of asphalt mixes using Pavement ME Design software, and to assess the sensitivity of the predicted distresses to mechanical mix properties as well as RAP content of asphalt mixes. Fig. 3 shows the hypothesis of this paper.
(AADTT) of 1660. It was assumed that 50% of total trucks is in the design direction and 100% trucks of each direction is on the design lane. The design life of the pavement is assumed to be 20 years. Winnipeg was selected as the project environmental condition and its historical climate data were used in all analysis. The typical Manitoba structural pavement thicknesses were estimated based on subgrade strength and traffic level. In order to evaluate the asphalt mix performance, the thicknesses for all layers were kept unchanged. 4.1. Materials inputs
4. Design example Three mixes containing 0–50% RAP (Mix-0, Mix-15, and Mix-50) were selected to evaluate their performance as predicted by Pavement ME Design software. The number after each mix is indicating the percent of RAP in the mix. A section of Highway 1 located at west of the city of Winnipeg in Manitoba, Canada, was selected as the design example presented in this paper. This section is a four-lane divided expressway with annual average daily truck traffic
Typical properties of Manitoba unbound materials were used for all the analysis. Table 2 shows the subgrade, subbase, and base material properties and their layer thicknesses. 4.2. Asphalt binder and asphalt mix inputs The thickness of asphalt mix layer was 150 mm for all the mixes. Complex shear modulus (G*) and phase angle (d) for three mixes were measured using dynamic shear
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
7
Table 8 Predicted performance of Mix-15 for Level 1, Level 3, and calibrated Level 3. Distresses
Design criteria
Input alternatives Level 3
Calibrated model
Level 1
Terminal IRI
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
2.5 2.8 78.1 Fail 16
2.5 2.53 88.8 Fail 19
2.5 2.53 88.8 Fail 19
AC surface down cracking
Target (m/km) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
380 468.5 84.6 Fail 12
380 62.1 100 Pass >20
380 50.3 100 Pass >20
AC bottom up cracking
Target (%) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
15 6.2 99.9 Pass >20
15 1.62 100 Pass >20
15 1.6 100 Pass >20
AC thermal fracture
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
200 608.6 11 Fail <2
200 491 0.2 Fail <2
200 491 0.2 Fail <2
Permanent deformation (AC rutting only)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
12 6.8 100 Pass >20
12 0.8 100 Pass >20
12 0.9 100 Pass >20
Permanent deformation (total pavement rutting)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
19 19.9 82.9 Fail 17
19 11.9 100 Pass >20
19 11.8 100 Pass >20
rheometer (DSR). Binder performance grade inputs are shown in Table 3. The volumetric properties and aggregate gradations of these mixes are shown in Table 4. 4.3. Dynamic modulus and creep compliance input data To assess the impact of local and global calibration of E* and creep models, three Levels of asphalt input data were considered; Level 1, calibrated Level 3, and Level 3. For Level 1, the laboratory measured dynamic modulus and creep compliance values were used while for the Level 3, the globally calibrated models were used to estimate theses values. For calibrated Level 3, dynamic modulus and creep compliance values were predicted based on the calibrated models, developed by [4,5], and manually entered into Pavement ME Design software as a Level 1 analysis. The globally and locally calibrated creep compliance and dynamic modulus values as well as laboratorymeasured values used in the analysis of mixes are shown in Tables 5 and 6, respectively. Since the E* values of Mix-50 at 10 °C exceeded the maximum allowable of dynamic modulus input in this software, dynamic modulus at 10 °C was not used in Pavement ME Design software.
5. AC predicted distresses for different Levels For all the three mixes, the same subgrade, subbase and base properties and thicknesses were used to evaluate the impact of asphalt binder and mix properties in Level 1, Level 3, and calibrated Level 3. The reliability for each distress was selected to be 90% for 20 years pavement design life. The reliability is dependent on the standard error of distress prediction model. When a high reliability is selected for predicting a pavement distress, a higher threshold is used to compute the pavement distresses resulting in higher predicted distresses [1]. The distress criteria of predicted distresses were selected based on MEPDG manual of practice recommendation. Tables 7 through 9 summarized the comparison of distress prediction among Level 1, Level 3, and calibrated Level 3 input data. The limits of international roughness index (IRI) and AC thermal cracking were chosen as 2.5 m/km and 200 m/km, respectively. The maximum value of AC surface down cracking (longitudinal cracking) and AC bottom up cracking (alligator cracking) were selected as 380 m/ km and 15%, respectively. The limits for permanent deformation (AC only), and permanent deformation (total pave-
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
8
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Table 9 Predicted performance of Mix-50 for Level 1, Level 3, and calibrated Level 3. Distresses
Design criteria
Input alternatives Level 3
Calibrated model
Level 1
Terminal IRI
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
2.5 2.7 81.5 Fail 17
2.5 2.6 87.7 Fail 19
2.5 2.6 87.6 Fail 19.5
AC surface down cracking
Target (m/km) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
380 316.7 93.9 Pass >20
380 51.3 100 Pass >20
380 51.7 100 Pass >20
AC bottom up cracking
Target (%) Predicted (%) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
15 3.1 100 Pass >20
15 1.8 100 Pass >20
15 1.8 100 Pass >20
AC thermal fracture
Target (m/km) Predicted (m/km) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
200 608.6 11 Fail <1
200 491 0.2 Fail <1
200 491 0.2 Fail <1
Permanent deformation (AC rutting only)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
12 5 100 Pass >20
12 1.3 100 Pass >20
12 1.4 100 Pass >20
Permanent deformation (total pavement rutting)
Target (mm) Predicted (mm) Reliability predicted (%) Acceptance Predicted life at 90% reliability (yrs)
19 17.7 96.5 Pass >20
19 12.9 100 Pass >20
19 13 100 Pass >20
ment rutting) were selected as 12 mm and 19 mm, respectively. As it can be seen from Tables 7–9, the calibrated Level 1 input data showed better-predicted performance when it was compared to Level 3 input data for all mixes regardless of RAP content. When Level 3 was used for Mix-0 and Mix-15, all the predicted distresses failed to meet the target value except AC permanent deformation. In contrast, when Level 1 or calibrated Level 3 was used as input data, only IRI and AC thermal cracking failed to meet the target value. AC thermal cracking failed to meet the target 200 m/ km for all mixes. This indicated that the binder type used in these mixes was not adequate for this location. The predicted life at 90% reliability for each distress was higher for Level 1 and calibrated Level 3 asphalt mix data compared to the Level 3 asphalt mix data, regardless of the distress prediction passed or failed to meet the target values. In other words, the amounts of predicted distresses were lower for Level 1 and calibrated Level 3 asphalt data compared to the Level 3 asphalt data. All the predicted distresses from each Level were normalized by dividing the predicted distress values obtained from Level 3 and calibrated Level 3 to corresponding
predicted distress values obtained from Level 1. Figs. 4–6 show a relative difference in predicted distresses of each mix for three Levels of input data. The predicted performance of Level 3 showed a higher normalized number in predicted rutting and fatigue distresses compared to calibrated Level 3. This indicates that Level 3 overpredicts the distress predictions of the asphalt mix. The predicted performance of calibrated Level 3 was found similar to Level 1 input data for all three mixes. This implies that utilizing calibrated models improves the reliability of the MEPDG distress predictions. Therefore, calibrated Level 3 asphalt input data can be used for the analysis and design of mixes with comparable accuracy of Level 1 input data. This can save substantial time and cost since it is time consuming and expensive to conduct laboratory asphalt testing for each mix. 6. Distress predictions of mixes containing RAP for calibrated Level 3 The predicted distresses of asphalt mixes for calibrated Level 3 were plotted to evaluate the sensitivity of the asphalt mix properties and RAP content. Fig. 7 shows
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Normalized distress prediction
Default Values (Level 3)
Calibrated Model
9
Level 1
11.00 10.00 9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 Terminal IRI
AC Surface AC Bottom Up Down Cracking Cracking
AC Thermal Fracture
AC Rutting
Total Pavement Rutting
Fig. 4. Long-term performance of Mix-0 for three Levels of input.
Normalized distress prediction
Default Values (Level 3)
Calibrated Model
Level 1
11.00 10.00 9.00
8.00 7.00 6.00 5.00 4.00 3.00 2.00
1.00 0.00 Terminal IRI
AC Surface AC Bottom Up AC Thermal Down Cracking Cracking Fracture
AC Rutting Total Pavement Rutting
Fig. 5. Long-term performance of Mix-15 for three Levels of input.
Normalized distress prediction
Default Values (Level 3)
Calibrated Model
Level 1
11.00 10.00 9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 Terminal IRI
AC Surface AC Bottom Up AC Thermal Down Cracking Cracking Fracture
AC Rutting Total Pavement Rutting
Fig. 6. Long-term performance of Mix-50 for three Levels of input.
the IRI performance of the mixes. The terminal IRI of all three mixes were slightly more than 2.5 at the end of 20 years pavement design life. It should be noted that MEPDG roughness prediction is affected by the predicted extent of cracking and other distresses. Fig. 8 shows the predicted alligator cracking of three mixes over pavement life. All the mixes successfully met
the alligator cracking limit. The predicted alligator cracking of the mixes did not show noticeable differences. Fig. 9 shows the predicted longitudinal cracking of three mixes over pavement life. Mix-50 showed slightly lower predicted longitudinal cracking in comparison to the other mixes. Predicted thermal cracking of the mixes are shown in Fig. 10. Fig. 10 illustrates that all three mixes failed to
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
10
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx 3
IRI (m/Km)
2.5 2 1.5
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
1 0.5 0 0
24
48
72
96
120
144
168
192
216
240
Pavement Age (month) Fig. 7. Predicted IRI and pavement life for mixes.
Alligator Cracking (%)
17 15 13 11 9
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
7 5 3 1 0
24
48
72
96
120
144
Pavement Age (month)
168
192
216
240
Fig. 8. Predicted alligator cracking and pavement life of mixes.
Longitudinal Cracking (m/Km)
430 380 330 280
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
230 180 130 80 30 0
24
48
72
96
120
144
168
192
216
240
Pavement Age (month) Fig. 9. Predicted longitudinal cracking and pavement life of mixes.
Thermal Cracking (m/Km)
600 500
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
400 300 200 100 0 0
24
48
72
96
120
144
168
192
216
240
Pavement Age (month) Fig. 10. Predicted thermal cracking and pavement life of mixes. Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
11
14
AC Rut (mm)
12 10 8
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value
6 4 2 0 0
24
48
72
96
120
144
Pavement Age (month)
168
192
216
240
Total Rut (mm)
Fig. 11. Predicted AC rut and pavement life of mixes.
20 18 16 14 12 10 8 6 4 2 0
Mix-0-0 Mix-0-15 Mix-0-50 Threshold Value 0
24
48
72
96
120
144
168
192
216
240
Pavement Age (month) Fig. 12. Predicted total rut and pavement life of mixes.
1.E+05
Dynamic Modulus (MPa)
1.E+03
1.E+02
Mix-0 Mix-15 Mix-50
1.E+01
1.E+00 1.E-05
Creep compliance (1/GPa)
1.E+00
1.E+04
1.E-01
1.E-02
1.E-03 1.E-02
Mix-0 Mix-15 Mix-50 1.E-01
1.E+00
1.E+01
1.E+02
1.E+03
Reduced time (sec.) 1.E-03
1.E-01
1.E+01
1.E+03
1.E+05
Reduced Freq. (Hz)
Fig. 14. Creep compliance master curves of mixes obtained from ANN calibrated model.
Fig. 13. Dynamic modulus master curves of mixes obtained from nonlinear regression of NCHRP 1-37A.
meet the thermal cracking criterion. Mix-50 exceeded the thermal criterion before the first year of pavement life which is expected since the binder performance grade of Mix-50 was PG 64–16 while the binder performance grade of Mix-0 and Mix-15 was PG 58–28. Figs. 11 and 12 show the predicted AC rut and total rut for the mixes, respectively. All the mixes successfully met the distress rut limit. It was noted that the AC rut for all mixes was noticeably low. Moreover, the AC rut of
Table 10 Predicted distresses for RAP mixes for calibrated Level 3. Predicted distresses
Asphalt mixes Mix-0
Mix-15
Mix-50
Terminal IRI (m/km) AC surface down cracking (m/km) AC bottom up cracking (%) AC thermal fracture (m/km) AC rutting only (mm) Total pavement rutting (mm)
2.53 64.00 1.60 491.00 0.84 12.00
2.53 62.10 1.62 491.00 0.8 11.90
2.60 51.30 1.80 491.00 1.30 12.90
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
12
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx
Mix-50 was slightly higher than Mix-0 and Mix-15, which was not expected. When a type of distress occurs in a pavement, it will cause the deterioration of the asphalt pavement to accelerate. The possible interactions between distresses did not appear to fully consider in the analysis of the mixes. When the AC thermal cracking failed to meet the target value, it was expected to affect the other distresses, however, no corresponding impact or acceleration in prediction trends of other distresses can be observed. This indicates that all distresses were predicted independently and the effect of one distress was not considered in prediction of other distresses. 7. Sensitivity of predicted distresses to the mix properties and RAP content The master curves of dynamic modulus and creep compliance of RAP mixes were constructed to show the possible difference in asphalt mix properties, and to determine the sensitivity of distress predictions to mix properties inputs and RAP content. Figs. 13 and 14 show the master curves of dynamic modulus and creep compliance for RAP mixes obtained from calibrated models. Fig. 13 shows that the dynamic modulus of Mix-50 is higher than Mix-0 and Mix-15 at high temperatures (Fig. 13). In addition, this mix has much lower creep compliance at low temperature when it is compared to Mix-0 and Mix-15 as shown in Fig. 14. In contrast, the dynamic modulus and creep compliance values of Mix-0 and Mix-15 were found to be very similar. It is expected that Pavement ME Design software would predict the asphalt distresses of Mix-50 to be different from those of Mix-0 and Mix-15, since Mix-50 has different mechanical properties and is a much stiffer mix. However, the predicted distresses of all mixes were found to be very similar as shown in Table 10. This indicated that an increase in RAP content did not noticeably impact the predicted distresses, although the fundamental mechanical properties, dynamic modulus and creep compliance, were substantially different. Furthermore, the total predicted rutting was found slightly higher for high RAP mix compared to Mix-0 and Mix-15. This could be explained by the lack of sensitivity in the rutting model to RAP content, the low truck traffic volume and traffic load, and that the volumetric mix properties of three AC mixes are very similar. In the previous section, it was shown that the use of locally calibrated models provides reliable results which are comparable to results obtained from Level 1 input data. However, the similarity in the distress predictions of the mixes with different mechanical properties and RAP content implied that the distress predictive models must be also locally calibrated based on field performance of mixes containing RAP. It is expected that locally calibrated distress models would be more responsive to the impact of different asphalt mix properties and RAP content.
8. Conclusions This paper presents the long-term performance of three asphalt mixes; Mix-0, Mix-15, and Mix-50 for three Levels of asphalt input data using Pavement ME Design software. Level 3, Calibrated Level 3, and Level 1 were used as three Levels of asphalt input data. For Level 1 input data, the laboratory-measured E* and creep compliance values were used whereas the Level 3 input data used globally calibrated predictive E* and creep models to estimate these values. The calibrated dynamic modulus and creep compliance models, developed by [4,5], were used for calibrated Level 3 input data. Based on the results, the following conclusions were drawn: The predicted performance of Level 3 showed higher amounts of distresses compared to those of Level 1 and calibrated Level 3. The Level 3 noticeably overpredicted the distress predictions of the asphalt mix. The results of using calibrated Level 3 inputs were very similar to those of Level 1 Inputs. As would be expected, the use of calibrated Level 3 input data instead of Level 3 increased the accuracy of predicted distresses. The calibrated Level 3 asphalt input data can be used for the design and analysis of mixes with comparable accuracy of Level 1 input data. This can save a substantial amount of time and cost since it is time consuming and expensive to conduct laboratory asphalt testing for each mix. Although the mechanical properties of the Mix-0 with no RAP and Mix-50, containing high RAP, were noticeably different, Pavement ME Design software predicted very similar distresses. Increase in RAP amounts did not significantly change the predicted distresses of the RAP mixes. It was noted that all distresses were predicted independently. The possible interactions between predicted distresses were not modeled. For instance, the effect of excessive thermal cracking was not considered in the prediction of other distresses. It is recommended to locally calibrate the distress models (transfer functions) based on field performance of local mixes containing RAP. It is expected that the calibrated distress models would be more responsive to the impact of different asphalt mix properties and RAP content.
References [1] American Association of State Highway and Transportation Officials (AASHTO), Mechanistic-389 Empirical Pavement Design Guide: A Manual of Practice. Washington, D.C., 2015. [2] T.R. Clyne, X. Li, M.O. Marasteanu, E.L. Skok, Dynamic and Resilient Modulus of MN/DOT Asphalt Mixtures, University of Minnesota, Minneapolis, MN, 2009.
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002
S. Esfandiarpour, A. Shalaby / International Journal of Pavement Research and Technology xxx (2018) xxx–xxx [3] S. El-Badawy, F. Bayomy, A. Awed, Performance of MEPDG dynamic modulus predictive models for asphalt concrete mixtures: local calibration for Idaho, J. Mater. Civ. Eng. 1412–1421 (2012). [4] S. Esfandiarpour, A. Shalaby, Local calibration of creep compliance models of asphalt concrete, Constr. Build. Mater. 132 (2017) 313–322. [5] S. Esfandiarpour, A. Shalaby, Alternatives for calibration of dynamic modulus prediction models of asphalt concrete, Int. J. Pavement Res. Technol. 10 (3) (May 2017). [6] K. Georgouli, A. Loizos, C. Plati, Calibration of dynamic modulus predictive model, Constr. Build. Mater. 102 (2016) 65–75. [7] National Cooperative Highway Research Program, Revision of Level 3 Thermal Fracture Predictive Models for the M-EPDG Version 1.0 and Final Level 1, 2, 3 Thermal Fracture Model Recalibration NCHRP 1-40D, Final Report, Transportation Research Board, Washington, D.C., 2006. [8] National Cooperative Highway Research Program, Guide for Mechanistic-Empirical Design of New and Rehabilitated Pavement
[9]
[10]
[11]
[12]
13
Structures NCHRP 1-37A Final Report, Transportation Research Board, Washington, D.C., 2004. D. Singh, M. Zaman, S. Commuri, Artificial neural network modeling for dynamic modulus of hot mix asphalt using aggregate shape properties, J. Mater. Civ. Eng. 25 (1) (2012) 54–62. M. Witczak, M. El-Basyouny, S. El-Badawy, Incorporation of the new, E* (2005) Predictive Model in the MEPDG NCHRP 1-40D Inter-Team Technical Report, Arizona State University, Tempe, AZ, 2006. Zeghal, M., Modeling the Creep Compliance of Asphalt Concrete Using the Artificial Neural Network Technique. GeoCongress. Characterization, Monitoring, and, Modeling of Geosystems. New Orleans, Louisiana, United States, 2008, pp. 910–916. J. Bari, M.W. Witczak, Development of a new revised version of the Witczak E* predictive model for hot mix asphalt mixtures, J. Assoc. Asphalt Paving Technol. 75 (2006) 381–424.
Please cite this article in press as: S. Esfandiarpour, A. Shalaby, Effect of local calibration of dynamic modulus and creep compliance models on predicted performance of asphalt mixes containing RAP, Int. J. Pavement Res. Technol. (2018), https://doi.org/10.1016/j.ijprt.2018.04.002