Construction and Building Materials 30 (2012) 36–49

Contents lists available at SciVerse ScienceDirect

Construction and Building Materials

journal homepage: www.elsevier .com/locate /conbui ldmat

A micromechanical approach to investigate asphalt concrete rutting mechanisms

Erdem Coleri a,⇑, John T. Harvey a,1, Kai Yang b,2, John M. Boone b,3

a University of California Pavement Research Center, Department of Civil and Environmental Engineering, University of California, Davis, One Shields Avenue, Davis, CA 95616, USAb Department of Radiology, University of California, Davis Medical Center, 4860 Y Street, Suite 3100 Ellison Building, Sacramento, CA 95817, USA

a r t i c l e i n f o

Article history:Received 11 August 2011Received in revised form 27 October 2011Accepted 26 November 2011Available online 29 December 2011

Keywords:X-ray computed tomographyRuttingAccelerated pavement testingHeavy vehicle simulatorDensificationShear flowAsphalt concretePavementPolymer modifiedRubberized

0950-0618/$ - see front matter � 2011 Elsevier Ltd. Adoi:10.1016/j.conbuildmat.2011.11.041

⇑ Corresponding author. Tel.: +1 510 305 6246.E-mail addresses: ecoleri@ucdavis.edu (E. Co

(J.T. Harvey), kai.yang@ucdmc.ucdavis.edu (K. Y(J.M. Boone).

1 Tel.: +1 530 754 6409.2 Tel.: +1 916 734 0314.3 Tel.: +1 916 734 3158.

a b s t r a c t

In this study, a new approach was used to evaluate changes in asphalt concrete microstructure under full-scale accelerated pavement test loading with a Heavy Vehicle Simulator (HVS) of composite pavement.The approach compared X-ray computed tomography (CT) images taken before and after HVS rut testing.Results were used to identify the differences in the movement of aggregate and changes in air-void con-tent and distribution occurring during rutting accumulation of rubberized gap graded and polymer mod-ified dense graded mixes for two overlay thicknesses (64 and 114 mm). Although high air void content forthe sections constructed with rubberized gap graded mix were expected to cause more densificationrelated rutting and earlier failure related to this densification, the actual reason behind the earlier failurewas determined to be primarily greater shear flow to the sides of the wheelpath associated with the gapgradation and small aggregate size. Significant movement of aggregate was observed in the direction oftravel as well as to the side under the pushed wheel. Important differences in aggregate movement andair-void changes were also observed between different overlay thicknesses indicating the depth of the rutphenomenon, important information for the design of overlays on aged asphalt concrete as well.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

The general trend in truck tires used to haul heavy freight hasindicated the need for changes in tire technology to reduce rollingresistance and improve fuel economy, including the switch frombias-ply to radial tires beginning 45 years ago, and continuing withthe movement to wide-base single tires over the last 20 years. At thesame time, tire inflation pressures have generally been increasingfor the same reasons. The combined effects of each of these tire tech-nology changes and increasing inflation pressures and truck traffichave increased the risk of rutting of asphalt concrete (AC) and a needfor high performance pavement structures [1]. Therefore, new pave-ment design and construction specifications have been developed toreduce the risk of early failure of pavement structures [2]. Threeother trends affecting the design and construction of asphalt mixesare: (1) the change of focus from designing and constructing newpavements to rehabilitation and maintenance; (2) the increaseduse of modified asphalt binders and gap-, open- and stone-matrix

ll rights reserved.

leri), jtharvey@ucdavis.eduang), jmboone@ucdavis.edu

type aggregate gradations rather than dense-gradations; and (3) ofthe increasing use of specifications to reward or penalize the con-tractor for performance-related construction quality or perfor-mance [3]. Together, these trends require the use of designmethods that consider material properties, structural design (partic-ularly the use of relatively thin modified asphalt overlays on con-crete and aged existing asphalt layers), and the effects of trafficand climate to deliver the required future performance of the con-structed structure.

Mechanistic–empirical (ME) design methods are developed bycalibrating critical mechanistic responses of pavement structurescalculated using theoretical structural models with responses mea-sured in the field, and then correlating those responses with empir-ical performance data. The structural models are developed based ongenerally accepted theories such as linear elastic theory (LET) and fi-nite element (FE) theory, with specific constitutive relations thatconsider the effects of environmental conditions and traffic loads,and dimensions. The structural models use laboratory test resultsto characterize the damage in the materials (usually loss of stiffnessor permanent deformation) resulting from the pavement response(stress and/or strain). Structural models are calibrated based onmeasured pavement performance that is generally determined fromaccelerated pavement testing (APT) and/or less commonly field testresults [4]. APT is also commonly used to calibrate responses withobserved distresses because it offers controlled experiments withthe ability to comprehensively and cost-effectively characterize

E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49 37

the pavement and measure responses and distress development, orto relate material properties to full-scale performance [5]. However,ME design methods use greatly simplified models (such as relatingelastic strains to plastic strains) [4,6,7] that rely on correct concep-tual understanding of causal relationships between responses thatcan be measured in APT or in the field and distress development.There has been considerable theoretical, laboratory testing andmodeling work to relate material properties to performance of mod-ified binder mixes and different aggregate gradations. However, thedensification and shearing processes that cause rutting of asphaltmixes have not had experimental measurements of the movementof aggregate and changes in air-void content and distribution thatexplain rutting failure at a microstructure scale. The literature alsoindicates that greater attention has been paid to the mechanics ofthick asphalt layers as opposed to the thinner overlays on existingstiff layers that make up the majority of the applications in Californiaand many other locations.

Asphalt concrete is a composite material which consists of aggre-gates, air-voids and mastic, and each phase plays a role in develop-ment of total accumulated rutting. Asphalt concrete mixes act ascontinuum materials at low temperatures when the stiffness ofthe asphalt binder and the aggregate are close [8]. However, at hightemperatures, i.e. the critical conditions for rutting deformationaccumulation, stiffness of the aggregate can be two to three ordersof magnitude higher than the asphalt binder stiffness. As a result, as-phalt concrete pavement layers start to act as particulate structureswhere the contact points between the aggregates become moreimportant in terms of rutting deformation accumulation [9]. There-fore, permanent deformation accumulation mechanisms predictedby using continuum mechanics may not reflect the actual mecha-nism under the in situ truck traffic. This has been recognized inrecent years and has lead to increased use of multi-phase continuummechanics and discrete element method approaches [10–15]. Thesemethods consider the properties of the aggregate (shape, texture,gradation, modulus), of the asphalt binder (visco-elasticity) and dis-tribution of air voids in the asphalt concrete layers. However, asnoted, experimental measurements for validation of these modelsfor permanent deformation under full-scale loading and for thinoverlays has not been available to date.

Microstructures of AC laboratory samples have been analyzedby various researchers. Masad et al. [16,17] characterized the air-void distribution in AC samples using the Weibull model andconcluded that the method of laboratory specimen compactionsignificantly influences the air-void size distributions. Masad andButton [18] developed a procedure to quantify the distribution ofaggregate skeleton and air-voids by analyzing images of the inter-nal structure. You et al. [10,11] processed the AC X-ray computedtomography (CT) images to develop 2D and 3D discrete and finiteelement models to evaluate the effect of binder, aggregate andair-void interaction on AC performance.

Changes in material microstructure under laboratory testingwere also evaluated by various researchers. Braz et al. [12,19] ana-lyzed the propagation of cracks under diametral compression test-ing to determine the effect of AC air-void distributions on crackingperformance. Warr et al. [13] analyzed the 2D transitional androtational motion of granular particles under laboratory loadingby using the images collected by high-speed photography. Wanget al. [14] used photograph images taken before and after GeorgiaLoaded Wheel Test loading (GLWT) to determine the 2D perma-nent strain field of an asphalt concrete specimen. Wang et al.[20] also calculated the plastic strain field in asphalt concrete lab-oratory samples accumulated for specimens tested with the As-phalt Pavement Analyzer (APA). It was concluded that permanentdeformation of asphalt concrete is localized mainly in the soft mas-tics due to low mastic stiffness at high test temperatures (60 �C).Synolakis et al. [21] used X-ray CT images that were taken before

and after the application of a diametral load on a cylindrical labo-ratory sample to calculate the plastic displacement field.

In this study, changes in AC microstructure were determinedusing full-scale test sections and Heavy Vehicle Simulator (HVS)loading, and X-ray CT images taken before and after HVS testing.Asphalt concrete blocks sawn from four HVS test sections werescanned to determine the microstructure of the blocks before test-ing. 3D distributions of air-voids, location and shape parametersfor the aggregates were determined using these images. ScannedAC blocks were installed back into their original locations using afast-setting epoxy. HVS tests were conducted at high temperatures(50 �C at 50 mm depth) that are critical for rutting until the surfacefailure was observed. Deformed AC blocks were re-sawn to per-form X-ray CT scanning after HVS loading. Deformed and un-de-formed 3D air-void and aggregate distributions were comparedto determine the changes in air-void content distributions andaggregate positions. Changes in air-void content distributions wereused to identify the contribution of densification to total accumu-lated downward rut, while changes in aggregate positions wereused to determine the displacement field under full-scale wheelloading and explain the reason behind the earlier failure of rubber-ized gap graded AC mixes when compared to polymer modifieddense graded AC mixes. Changes in air-void distributions and dis-placement field can also be used to validate and/or calibrate micro-mechanical finite element or discrete element models.

This study is the first in situ investigation of micromechanicalchanges in AC layers under full-scale loading, and provides exper-imental data for continued development of multi-scale continuummechanics and discrete element method analysis approaches. Re-sults of the analyses will give important information about the rut-ting failure mechanisms of gap-graded rubberized and polymermodified dense graded asphalt mixes. The method used in thisstudy can be used to develop additional experimental data forother mix types and under different testing conditions.

2. Objectives

The main objectives of the work presented in this paper were:

� Use the X-ray CT imaging method to identify changes in micro-structure caused by the rutting process, namely aggregate dis-placements and changes in air-void content and distribution,under full-scale loading and for different thicknesses of overlayon very stiff underlying layers.� Use the experimental measurements to perform a first-level

assessment of the micromechanical phenomena to provideinsight into mix performance for mix design and developmentand understanding of laboratory tests, and information regard-ing the mechanics of rutting that can be used to improve micro-mechanical analysis approaches.

3. Heavy vehicle simulator testing

The HVS is a mobile load frame that uses a full-scale wheel (dual or single) totraffic the pavement test section. The trafficked test section is 8 m long, of which1 m on each end are used for turnaround of the wheel and are discarded.

Composite pavement HVS test sections were constructed at the AdvancedTransportation Infrastructure Research Center (ATIRC) facility at UC Davis consist-ing of two thicknesses of AC overlay on non-doweled jointed plain portland cementconcrete (PCC) slabs that were 178 mm (7 in.) thick, 3.7 m wide (12 ft) and 4.5 m(15 ft) long over a 150 mm (6 in.) aggregate base layer. The 6 m long HVS wheelpathincluded two joints and one slab.

The failure criterion was defined as an average maximum rut (defined as thesummation of the downward deformation and ‘‘humping’’ of material sheared tothe sides of the wheelpath) of 12.5 mm over the full monitored section. The pave-ment temperature at 50 mm depth was maintained at 50 �C ± 4 �C to assess ruttingpotential under typical pavement conditions.

38 E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49

3.1. Material and structure properties

Two asphalt mixture types were used in this study, a dense-graded mixturewith PG 64-28 polymer modified (PM) binder and a gap-graded mixture (RHMA-G) with asphalt rubber binder (PG 64-16). Both mixes were specified based on Cal-ifornia Department of Transportation (Caltrans) specifications [22,23] for both mixand binder. Nominal maximum aggregate sizes (NMASs) for PG64-28PM andRHMA-G mixtures were 19 mm and 12.5 mm, respectively. Table 1 shows theaggregate gradations, asphalt binder content and air-void content of the asphaltmixtures. The overlay thicknesses were 64 mm and 114 mm (2.5, 4.5 in.).

3.2. Experiment program

The HVS loading program for each section is summarized in Table 2. All traffick-ing was carried out at 8.7 km/h (5.4 mph) with a dual-wheel configuration with thecenterlines of the two tires spaced 360 mm apart, using radial truck tires inflated to690 kPa (100 psi), in a channelized (no wander), unidirectional loading mode inwhich the wheel travels one direction loaded and is lifted off the pavement when

Table 2Summary of HVS loading program.

Section Mix type As-built thickness (mm) Wheel lo

609HB PG64-28PM 116 4060

610HB PG64-28PM 72 4060

611HB RHMA-G 118 40

612HB RHMA-G 74 40

a 40 kN = 9000 lb.b 50 �C = 122 F.c SD: standard deviation.

Fig. 1. Comparison of field and HVS rutting: (a) rutting unde

Table 1Aggregate gradation for asphalt mixtures.

Sieve size (mm) Percent passing (%)

Mix PG64-28PM Mix RHMA-G

25.4 100 10019 99 10012.5 87 989.5 74 844.75 50 342.36 36 211.18 27 140.6 21 100.3 16 70.15 10 50.075 6.2 3.5Asphalt binder contenta (%) 5.1 7.6Air void contentb (%) 9.9 13.6

a Design binder content (by dry weight of aggregate).b As-built average air void content.

returning. Channelized trafficking is used to simulate the tracking of radial tiresin the wheelpath once a small rut forms, and is more aggressive than fieldconditions in the initial stages of rutting. However, it should be noted for the fieldcase that the accumulated rutting at the initial stages will force the tires to travel ina more channelized pattern. Thus, at the failure point, surface permanent deforma-tion for the field case generally has a channelized pattern too. Fig. 1 compares theshapes for the accumulated surface rutting under HVS and field trafficking. It can beobserved that surface rutting for both field and HVS cases have a channelized shapealthough slightly larger channels were observed for the field case as a result of thewandered trafficking at the initial stages.

4. General procedure for X-ray CT image data collection

The general procedure followed for X-ray CT image data collection(before-after HVS testing) is illustrated with photographs in Fig. 2.

The steps followed for X-ray CT image data collection are de-scribed as follows:

� Preliminary tests were performed to determine the best adhesiveto glue sawn AC blocks back into their original locations. Becausesawing changes the boundary conditions around the AC blocks,adhesive type to create the best connectivity between the ACblocks and pavement was selected. The adhesive also needed tohave similar material properties as the original AC pavement toavoid non-uniformity around the blocks. Epoxy and asphalt mas-tic were evaluated using Hamburg Wheel Tracking Device(HWTD) results (Fig. 2a). Small AC blocks were sawn from theHWTD test samples and glued back with epoxy and mastic. Local-ized rutting failure was observed around the block that was gluedwith mastic (Fig. 3). On the other hand, more uniform ruttingdeformations were observed on the HWTD sample with epoxyglued block. Fast-setting epoxy was used in the HVS sectionsfor gluing the sawn AC blocks.� The 3D internal structures of the asphalt mixtures were acquired

by X-ray CT imaging at University of California Davis Medical

ada (kN) Temperatures at 50 mm Total repetitions

Averageb (�C) SDc (�C)

49.5 1.1 63,750136,250

49.8 1.0 64,000137,200

48.7 1.1 18,503

49.7 1.3 90,000

r HVS trafficking and (b) rutting under field trafficking.

(a)

Non-stick paper

(b) (c)

(d) (e) (f)

(g)

Extendedcut

(h) (i)

(j) (k) (l)

Fig. 2. The general procedure followed for CT image data collection (before–after HVS testing).

E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49 39

Center (UCDMC). The allowable maximum length of the AC blockdimension that can be scanned with a medical scanner was deter-mined by scanning laboratory samples with various dimensions.It was determined that largest dimension of the AC block shouldnot exceed 17 cm (6.7 in.) in order to acquire X-ray CT imageswith an acceptable level of quality. Square 17 cm AC blocks weresawn from HVS test sections to achieve the maximum volumethat could be scanned.� AC blocks were located so that one half of the block was in the

wheelpath while the other half was next to the wheelpath inorder to monitor the lateral shear flow. Four blocks were sawnfrom distributed locations (to avoid localized failures) on eachHVS test section (16 blocks from 4 HVS rutting sections) toget the complete deformation profile (Fig. 4). Locations weredifferent on sections with a thick AC layer to avoid cutting thecables of strain gauges at 50 mm depth.� Because tack coat was applied on the PCC layer to create bonding

between PCC and AC layers during construction, non-stick paperwas used to break the bond between PCC and AC layers (Fig. 2b)under the blocks. This facilitated the removal of sawn AC blockswhile avoiding any damage to the blocks during removal.� AC layers were constructed on top of PCC layer (Fig. 2c).� AC block locations were marked on AC layer and blocks were

sawn from those previously determined locations (Fig. 2d).� 3D internal structures of the asphalt mixtures were acquired by

X-ray CT imaging at UCDMC (Fig. 2e).� X-ray CT images were processed to obtain aggregate location

and shape information. In addition, distributions of air-voidsin AC blocks were determined (Fig. 2f).

� Extended cuts that were created during sawing of the blocksdue to the circular shape of the concrete saw were filled withan emulsion-sand mixture to avoid localized failures (Fig. 2g).� Scanned AC blocks were installed back into their original posi-

tion with fast-setting epoxy (Fig. 2h).� HVS test sections were prepared for testing. Boundaries for the

HVS wheelpaths were drawn on the sections by considering theAC block locations (Fig. 2i).� HVS sections were tested (Fig. 2j).� Rutted AC blocks were re-sawn after trafficking (Fig. 2k). In

order to perform sawing with high precision, blocks were firstsawn with a perimeter larger than the original block. Originalblocks were sawn with high precision using a stationary saw.� After test-AC blocks were re-scanned at UCDMC and images

were processed to acquire the new 3D aggregate and air-voiddistributions (Fig. 2l).

5. CT image acquisition and processing

The X-ray CT scanner at UCDMC reconstructs the 3D spatial dis-tribution of the attenuation coefficients of the scanned materials.Three dimensional images of the specimen are generated by com-bining a series of 2D images generated by the scanner. The qualityof the 3D image depends on the X-ray CT image resolution. In thisstudy, horizontal planar images were acquired at 1 mm intervalswhile the resolution cell for the other two dimensions was deter-mined to be 0.24 mm. At each sampling point within the sample vol-ume, each pixel from a CT image measures a value that is related tothe density and atomic number of the material at that particular

1 2 3 4 5 6 7 8 9 100

1000

2000

3000

-7.7-6.8

-6.0

-5.1

-5.1

-4.3

-3.4

-2.6

-2.6

-1.7 -0.9

Longitudinal Location #

Rep

etiti

ons

Block locationTireLoad direction(a)

1 2 3 4 5 6 7 8 9 100

1000

2000

3000

-13.6-12.0

-10.5

-9.0-7.5

-6.0-4.5

-3.0

-1.5

Longitudinal Location #

Rep

etiti

ons

Block locationTireLoad direction(b)

Fig. 3. HWTD test results: (a) for the AC block that was glued with epoxy and (b) forthe AC block that was glued with mastic.

40 E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49

point. These measured values are then converted to grayscale values[24]. The gray scale intensities for the aggregates, air voids and mas-tic range from�1000 to 3095, but these values were rescaled to therange from 0 to 255. Because aggregates are denser than mastic andair voids, they occupy the higher portion of the intensity scale. Lowdensity air voids are at the lowest intensity portion while masticintensity is between aggregate and air void intensities.

5.1. Development of 3D air-void distributions

Simpleware software [24] was used to create masks for the airvoid domain based on measured intensity groups. Air void percent-ages for each asphalt mixture sample were determined by using thestandard CoreLok method [25] (Table 1). The total volume of eachspecimen was calculated for the complete intensity range, 0–255.The upper limit for the air void intensity range was determined bytrial and error to match the measured air void contents. The averageair-void intensity range for the PG64-28PM blocks was determinedto be 0–119 with a calculated average air void content of 9.9% whilethe average range for RHMA-G blocks was 0–102 with a calculatedaverage air void content of 13.6%. After the thresholds for the air voiddomain were determined, masks with specific colors were assignedto clearly visualize the distribution of air void domain in the total

block volume. Fig. 5 illustrates the air void distributions for Block1 (see Fig. 4) of all HVS test sections.

Developed 3D images were further processed to obtain the distri-bution of air voids with depth. First, AC blocks were divided into2.93 mm thick volumes in the vertical direction. Volumetric air voidcontent for each volume was calculated with Simpleware software[24] to determine the distribution of air voids within the depth ofthe AC blocks. Fig. 6a shows the distribution of air-voids with depthfor the thick asphalt sections, also showing the construction lift. Itcan be observed that air-void content in the middle depth of each liftis lower than the bottom and top of the lift, most likely the result ofgreater temperatures being retained longer in the middle of the liftand less interference with aggregate rotation from stiff horizontalboundaries. It can also be seen that the bottom lift is better com-pacted than the top lift. This maybe a result of the reheating andadditional compaction of the bottom lift provided during the com-paction of the top lift and/or the high confinement provided by thestiff PCC layer during the compaction of the bottom lift.

Fig. 6b shows the distribution of air void contents for the thin(64 mm) AC blocks. Distributions appeared to be similar to the dis-tributions obtained for the bottom lift of thick sections except forhigher air void contents observed at the top and bottom of thinAC blocks. Air void content distributions for the thick and thinAC blocks also show that due to the larger NMAS for the PG64-28PM mix, air void contents at the bottom of each lift appear tobe relatively larger than the RHMA-G blocks. However, in the mid-dle depth of each lift, PG64-28PM blocks have significantly lowerair void contents.

5.2. Development of 3D aggregate distributions

Imaris Version 7.2 software was used in this study to createmasks for the aggregate domain based on measured intensitygroups. Intensity ranges for the aggregates were determined basedon the X-ray CT images. Surfaces and volumes for aggregates werecreated by Imaris in order to visualize the range of interest of a vol-ume object. Fig. 7 illustrates the procedure for the creation ofaggregate volumes and surfaces, which was as follows:

� Prepare X-ray CT image stacks using the software ImageJ Ver-sion 1.43 (Fig. 7a).� Import image stacks to Imaris and create 3D images from the

stack of 2D images (Fig. 7b).� Determine the intensity range that will create the aggregate

volumes that are close to the actual aggregates.� Create 3D aggregate volumes for before and after trafficking-AC

blocks (Fig. 7c).� Split discrete aggregates that appear unified in the image.

5.2.1. Particle tracking algorithm used for displacement fielddevelopment

The particle tracking algorithms of Imaris were used to deter-mine the displacement field in AC blocks. Aggregate and spottracking options of Imaris were evaluated to decide on the bestmethod for the identification of displacement field. The aggregatetracking option basically determines the coordinates for the vol-ume center of each aggregate for before and after trafficking. Basedon the measured coordinates, vectors were created between beforeand after trafficking coordinates. These vectors basically show thepermanent displacement field in the AC blocks.

The ‘‘spot object’’ feature of Imaris was also used for the devel-opment of the displacement fields. Adjacent voxels, which is a vol-ume element representing a value on a regular grid in 3D space,with close intensity values were combined to form spots in before-and after-trafficking images. Similar to aggregate tracking, coordi-nates for each spot in before- and after-trafficking images were

AC BlockWheel path

(a) PCC joint PCC joint

1 2

3

4

AC Block

Wheel path

(b)PCC joint PCC joint

1

2

3 4

Fig. 4. Locations for the AC blocks (scaled – dimensions in cm): (a) sections 609HB and 611HB with thick (114 mm) AC layer and (b) sections 610HB and 612HB with thin(64 mm) AC layer.

E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49 41

determined and used for displacement vector development. Themajor advantage of using spots for tracking is due to large numberof vectors that can be created in AC block volumes. With thismethod, the displacement field can be more effectively producedto visualize the general shear and densification related flow underHVS trafficking. It is also not necessary to assign geometrical shapeparameters for the shape identification in the spot object option,unlike the aggregate tracking option, which reduces the probabilityof having false-tracking. Fig. 8 shows the differences between thetwo algorithms. It can be observed that the displacement fieldand the plastic shear flow can be clearly visualized when spots ob-ject is used. On the other hand, the general displacement trendcannot be clearly captured when the aggregate tracking option isused due to the limited number of aggregates in the AC block.

Because RHMA-G mixes have higher air void contents and smal-ler aggregates than the PG64-28PM mixes, the quality of X-ray CTimages taken before HVS testing appear to be higher than the qual-ity of the images for the PG64-28PM mixes. However, since higherrut depths were observed for the sections with RHMA-G mix, areduction in image quality is observed under the HVS tire due todensification which created particle identification problems aroundthat area. However, this effect was only evident at the top 1 cm un-der the HVS tire and was mostly avoided by using the spot trackingoption. It can be observed from Fig. 8b that the number of vectorsright under the HVS tire is less than the number of vectors at otherlocations because most of the spots with same intensity groupscannot be identified in the after testing X-ray CT image.

6. Test results

6.1. HVS test results

The rutting performance of the four sections is shown in Fig. 9(average maximum rut, defined as the summation of the down-

ward deformation and ‘‘humping’’ of material sheared to the sidesof the wheelpath averaged across 13 measurements, transverseprofiles at 0.5 m intervals along the 6 m wheelpath). The wheelload for the two sections with the PG64-28PM mix (609HB and610HB) was increased from 40 kN to 60 kN at around 64,000 rep-etitions (Table 2). However, increasing the load did not have anysignificant effect on the rutting accumulation rate. Earlier failurewas observed for the thick RHMA-G section (611HB).

The average maximum rut depth is approximately two timesgreater than the downward rut depth for both of the PG64-28PMsections, and nearly three times greater for both of the RHMA-Gsections, indicating that shearing of material to the side of thewheelpath is as important a contributor or more important thanthe downward rut.

Profilometer measurements were also taken on the AC blocksat many load intervals. Fig. 10 shows the block surface deforma-tion at the end of HVS testing. It can be observed that part ofhump (right next to HVS tire) and rut (under HVS tire) createdby HVS loading were captured by the profilometer on the ACblocks to be able to compare the shear flow and densificationwith X-ray CT imaging.

6.2. Changes in 3D air-void distributions with HVS loading

Changes in air void content distributions with HVS traffickingwere investigated by comparing the X-ray CT images taken beforeand after testing. The purpose was to determine the contributionof densification (air void reduction) to total downward surface rutalong the depth of the AC layer. Fig. 11 shows the changes in air voidcontents after HVS trafficking for the AC block of Section 609HB –PG64-28PM with 114 mm thickness. In order to determine thechange in air void content, air void content distributions, similarto Fig. 6, for both before and after trafficking images were calculatedfor hump (right next to HVS tire) and wheelpath locations.

Fig. 5. Distributions of air voids in AC blocks (colored volumes are air-voids): (a) Block 1 – Section 609HB – PG64-28PM – 114 mm from two perspectives, (b) Block 1 –Section 611HB – RHMA-G – 114 mm from two perspectives, (c) Block 1 – Section 610HB – PG64-28PM – 64 mm from two perspectives, and (d) Block 1 – Section 612HB –RHMA-G – 64 mm from two perspectives. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

42 E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49

Before-trafficking air void content distributions were subtractedfrom the after trafficking distributions to calculate the change inair void content distributions. The areas under change in air voidcontent curves for the unit volume thickness (2.93 mm) were calcu-lated to determine the distribution of densification related ruttingalong the thickness of the AC layers.

Fig. 12 shows the calculated distributions for rutting caused bydensification for Block 1 of all HVS test sections. It can be observedthat sections with RHMA-G mix (611HB and 612HB) experiencedmore uniform densification related rutting, which was a result ofthe uniform constructed air void content distributions (see Fig. 6).For sections constructed with PG64-28PM mix, high air void con-tents at the bottom of each lift caused large densification relatedrutting. In order to determine the contribution of densification re-lated rutting to total downward rut, rut depths in Fig. 12 weresummed up and compared with the measured downward rut onAC blocks (Table 3). In addition, the contribution of densificationat the bottom 8.7 mm of each lift to total downward rut was calcu-lated. Results show that although average air void contents ofRHMA-G were 37% higher than PG 64-28PM, the contribution ofair void reduction to total downward rutting was close for all testsections. This result suggests that earlier failure of sections withRHMA-G mix was actually a result of the shear related deformation.

Displacement fields for each section will be analyzed to investigatethe reasons behind the faster shear failure for RHMA-G mixes.

Results further indicated that 43% of densification for the thick-PG64-28PM section accumulated at the bottom 8.7 mm of top andbottom construction lifts while this number increased to 67% forthe thin-PG64-28PM section. This result suggests that localizedcompaction problems related to large NMAS can cause large densi-fications around these poorly compacted locations.

6.3. Changes in 3D aggregate distributions with HVS loading – particletracking

The spot tracking option of Imaris was used to determine thedisplacement field in the AC blocks. Spots that were furthest fromthe HVS load (at the bottom of the block) were assumed to be sta-tionary after HVS trafficking. These spots were used as referencepoints to match before and after trafficking images. It should benoted that no significant differences between the displacementfields were observed when different spots (furthest from the HVSload) were used as reference. This result suggested that move-ments of the spots that are furthest from the HVS load were negli-gible and they can be used as reference points for particle tracking.

AC surface

PCC surface(609HB)

PCC surface(611HB)

(a)0

20

40

60

80

100

120

0% 10% 20% 30% 40%Airvoid Content

Dep

th (m

m)

609HB - PG64-28PM611HB - RHMA-GLift

PCC surface(610HB)

PCC surface(612HB)

AC surface

(b)0

20

40

60

80

100

120

0% 10% 20% 30% 40%Airvoid Content

Dep

th (m

m)

610HB - PG64-28PM612HB - RHMA-G

Fig. 6. The distribution of air void contents with depth for Block 1 of all HVS test sections: (a) Section 609HB – PG64-28PM – 114 mm and Section 611HB – RHMA-G –114 mm and (b) Section 610HB – PG64-28PM – 64 mm and Section 612HB – RHMA-G – 64 mm.

(a)

HVS Load

(b)HVS Load

(c)HVS Load

Fig. 7. The procedure followed for aggregate domain creation: (a) unprocessed 2D X-ray CT image used for the development of image stacks (left: before HVS trafficking –right: after HVS trafficking), (b) 3D images developed from X-ray CT image stacks (left: before HVS trafficking – right: after HVS trafficking), and (c) 3D aggregate volumes(left: before HVS trafficking – right: after HVS trafficking). Note: Direction of HVS traffic is out of the page.

E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49 43

HVS Load(a) HVS Load(b)

Fig. 8. Comparison of spot and aggregate tracking options for Block 1 – Section 609HB – PG64-28PM – 114 mm: (a) aggregate tracking and (b) spot tracking. Note: Directionof HVS traffic is out of the page.

0

2

4

6

8

10

12

14

16

0 50 100 150 200 250

Repetitions (Thousand)

Max

imum

Tot

al R

ut (m

m)

609HB: PG 64-28 PM, Hac=114mm610HB: PG 64-28 PM, Hac=64mm611HB: RHMA-G, Hac=114mm612HB: RHMA-G, Hac=64mm

40kN (50ºC) 60kN (50ºC) (for only sections 609HB and 610HB)

Fig. 9. Comparison of average maximum rut.

-15

-10

-5

0

5

-800 -600 -400 -200 0 200

Transverse Distance (mm)

Surf

ace

Def

orm

atio

n (m

m)

609HB - PG 64-28 PM - Hac=114mm610HB - PG 64-28 PM - Hac=64mm611HB - RHMA-G - Hac=114mm612HB - RHMA-G - Hac=64mmLeft edge of blockRight edge of block

Fig. 10. Final surface profile on AC blocks.

44 E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49

Fig. 13 shows the distribution of displacement vectors for one ofthe outer tire edge blocks for all HVS test sections. Two shear flowpatterns can be observed for thick sections (609HB and 611HB),one flow to the hump and the other to the opposite direction underthe tire. For thin sections, only shear flow to the hump part was ob-served. It should be noted that shear flow patterns under wanderedtraffic might be different than the patterns observed for channeli-zed HVS trafficking. In addition, larger shear forces appeared tocause more upward movement for the sections constructed withRHMA-G mix (611HB and 612HB). This movement caused largerhumps which consequently increased the maximum deformationfor these sections. This result can be accepted to be the major causeof earlier failure for these sections (see Fig. 9). In order to quanti-tatively evaluate the differences between the shear flow pattern

Fig. 11. The changes in air void contents after HVS trafficking for the AC block of Sectitrafficking and (b) after trafficking. Note: Direction of HVS traffic is out of the page, wheellegend, the reader is referred to the web version of this article.)

at the humps for the two different mix types of this study, Y (ver-tical) component of the vectors at the top 30 mm of the humpswere given as distributions in Fig. 14. It can be observed that dis-tributions are similar for the same mix types. In addition, for thesections constructed with the RHMA-G mix, most part of the Ycomponent distribution lies in the negative part which suggestssignificant upward particle movement at the humps.

In order to determine the pattern of the displacement field inthe direction of traffic (Z direction), top-view for the AC block dis-placement field is given in Fig. 15. A significant shear flow patterncan be observed in the direction of traffic (Z direction) for sectionsconstructed with RHMA-G mix. In addition, larger displacementvectors in the direction of traffic were observed at the hump partof the thin-RHMA-G section (612HB).

on 609HB – PG64-28PM 114 mm thick (colored volumes are air-voids): (a) beforepath on left half of images. (For interpretation of the references to color in this figure

0

20

40

60

80

100

120

-0.75 -0.5 -0.25 0 0.25

Dep

th (m

m)

HumpRutLiftSurface Rut

AC surface

(a)

Wheelpath

Rutting caused by densification (mm)

Rutting caused by densification (mm)Rutting caused by densification (mm)

Rutting caused by densification (mm)

0

20

40

60

80

100

120

-0.75 -0.5 -0.25 0 0.25

Dep

th (m

m)

HumpRutLiftSurface Rut

AC surface

(b)

Wheelpath

AC surface

(c)0

20

40

60

80

100

120

-0.75 -0.5 -0.25 0 0.25

Dep

th (m

m)

HumpRutSurface RutWheelpath

0

20

40

60

80

100

120

-0.75 -0.5 -0.25 0 0.25

Dep

th (m

m)

HumpRutSurface Rut

AC surface

(d)

Wheelpath

Fig. 12. Calculated distributions of rutting caused by densification for block 1 of all HVS test sections: (a) Section 609HB – PG64-28PM – 114 mm, (b) Section 611HB – RHMA-G – 114 mm, (c) Section 610HB – PG64-28PM – 64 mm, and (d) Section 612HB – RHMA-G – 64 mm.

Table 3Contribution of densification related rutting to overall downward rut in the wheelpath.

Section Thickness (mm) Mix type Downward rut (mm) Densification rut (mm) Densification lifta (mm)

609HB 114 PG 64-28PM 13.42 5.19 2.25610HB 64 PG 64-28PM 5.37 2.11 1.41611HB 114 RHMA-G 9.22 4.07 0.94612HB 64 RHMA-G 6.02 1.89 0.62

a Densification lift: total densification related rutting at the bottom 8.7 mm of each lift.

E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49 45

Large shear flow in the direction of traffic can be accepted to beanother contributor to earlier failure of sections constructed withRHMA-G mix (611HB and 612HB). The movement in the direction

of traffic can be related to the fact that the HVS wheel is pushed, asopposed to being a traction wheel. The movement in the Z direc-tion would likely be in the opposite direction for a traction wheel.

HVS load(a) HVS load

(b)

HVS load(c)

HVS load(d)

Fig. 13. Distribution of displacement vectors for one of the outer tire edge blocks of all HVS test sections (front view): (a) Section 609HB – PG64-28PM – 114 mm, (b) Section611HB – RHMA-G-114 mm, (c) Section 610HB-PG64-28PM-64 mm, and (d) Section 612HB-RHMA-G-64 mm. Note: Direction of HVS traffic is out of the page.

0.10

0.15

0.20

0.25

0.30

0.35 609HB - PG 64-28 PM - Hac=114mm611HB- RHMA-G- Hac=114mm610HB- PG 64-28 PM - Hac=64mm612HB- RHMA-G Hac=64mm

Prob

abili

ty d

ensi

ty Y

X

Z

(0,0)ORIGIN

Direction of travel

46 E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49

This suggests that the passing of traction and pushed wheels on atruck would produce potentially important shear reversals in thewheelpaths and to some degree in the humps. The differing perfor-mance of the two mixes in the Z direction is a result of the differ-ences between the aggregate shape, gradation and size of two mixtypes. In addition, high air void content for the RHMA-G mix mighthave induced the excessive shear flow. In other words, althoughhigh air void content for the sections constructed with RHMA-Gmix was expected to cause larger densification related ruttingand earlier failure related to this densification, the actual reasonbehind the earlier failure was determined to be the gap gradationand small aggregate size resulting in greater shear flow.

-15 -10 -5 0 5 10 150.00

0.05

Displacement in Y direction (mm)

Fig. 14. Distribution of Y (vertical) component of the displacement vectors at thetop 30 mm of the humps. Note: Negative displacement in Y direction indicatesmovement in the upper direction.

6.4. Investigation of the correlation between predicted elastic andmeasured plastic displacement fields

In order to determine the possible correlations between mea-sured plastic and calculated elastic displacement fields, LET simu-lations were performed for thick and thin sections constructedwith RHMA-G mix (611HB and 612HB). Stiffness for the asphaltlayer was determined from laboratory shear frequency sweep teststhat were conducted at 50 �C. Fig. 16 shows the predicted displace-ment fields for both thick and thin RHMA-G sections. Fig. 16a and bshows the predicted elastic displacement distributions in X direc-tion while Fig. 16c and d shows the displacement distributions inY direction. Because the LET simulation results were comparedwith the measured plastic deformations on the AC blocks, elasticdisplacement outputs only in the AC block boundaries were givenin Fig. 16. The large ratio of displacement in Y direction to the dis-placements in X direction for all analysis points (ranging from 7 to1000) suggests that movement in X direction is considerably smal-ler than the movement in Y direction. However, measured plasticdisplacement fields for the thick and thin RHMA-G sections givenin Fig. 13b and d show that significantly large movements in Xdirection created a shear flow which consequently controlled theaccumulation of permanent surface deformation. Fig. 16c and dalso shows that no upward movements (displacement in negativeY direction) occur with HVS loading which also conflicts with themeasured in situ plastic displacement fields (Fig. 13b and d).

These results suggest that there are significant differences be-tween the measured plastic and predicted elastic displacement

fields. Incorporating the particulate structure of the AC layer (airvoids, binder and aggregate phases) to the model will be expectedto improve the predictive capability of the model. In addition,adaptation of a viscoelastic model for the binder phase will resultin more realistic displacement fields that maybe similar to themeasured plastic displacements.

7. Conclusions

In this study, a new approach was proposed to evaluate thechanges in asphalt concrete microstructure with HVS loading bycomparing the X-ray CT images taken before and after HVS testing.Results were used to identify the differences between rutting accu-mulation mechanisms of RHMA-G and PG64-28PM mixes for twothicknesses.

This study will be the first in situ investigation of micromechan-ical changes in AC layers under a full-scale accelerated pavementtest.

Results of this study can be summarized as follows.

Wheel pathHump

(a)Wheel pathHump

(b)

Wheel pathHump(c) HumpWheel path(d)

Fig. 15. Distribution of displacement vectors for Block 1 of all HVS test sections (top view): (a) Section 609HB – PG64-28PM – 114 mm, (b) Section 611HB – RHMA-G –114 mm, (c) Section 610HB – PG64-28PM – 64 mm, and (d) Section 612HB – RHMA-G – 64 mm. Note: Direction of HVS traffic is from top to bottom.

-0.0191

-0.0158

-0.0125-0.0093

-0.0060

-0.0028

0.00050.0038

0.0038

0.0070

0.0070

20 40 60 80 100 120 140 160 180

HVS LOAD

X (mm)

60

50

40

30

20

10

0

Y (m

m)

(a) - direction + direction

-0.0351

-0.0292

-0.0233

-0.0173

-0.0114

-0.0055 0.0005

0.0005

0.00640.0123

20 40 60 80 100 120 140 160 180

HVS LOAD

X (mm)

100

80

60

40

20

0

Y (m

m)

(b) - direction + direction

0.22330.2373 0.2513

0.2653

0.2793

0.2933 0.3073

0.3213

20 40 60 80 100 120 140 160 180

HVS LOAD

X (mm)

60

50

40

30

20

10

0

Y (m

m)

(c)

+ di

rect

ion

-dire

ctio

n

0.2260

0.2491

0.2722

0.2953

0.3184

0.3415 0.3646

0.3877

0.4107

20 40 60 80 100 120 140 160 180

HVS LOAD

X (mm)

100

80

60

40

20

0

Y (m

m)

(d)

+ di

rect

ion

-dire

ctio

n

Fig. 16. Elastic displacement fields calculated from LET simulations: (a) distribution of displacement in X direction for thin RHMA-G section, (b) distribution of displacementin X direction for thick RHMA-G section, (c) distribution of displacement in Y direction for thin RHMA-G section, and (d) distribution of displacement in Y direction for thickRHMA-G section.

E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49 47

48 E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49

1. Sections constructed with RHMA-G mixes had more uni-form air void content distributions due to the smaller NMAS.Thus, these sections experienced more uniform densifica-tion related rutting while sections constructed with PG64-28PM mixes had more localized densifications.

2. Forty-three percent of densification for the thick-PG64-28PM section accumulated at the bottom 8.7 mm of topand bottom construction lifts while this number increasedto 67% for the thin-PG64-28PM section. This result suggeststhat localized compaction problems related to large NMASfor the PG64-28PM mix can cause large densificationsaround these poorly compacted locations.

3. Two shear flow patterns were observed for thick sections(609HB and 611HB), one flow to the hump and the otherto the opposite direction under the tire. For thin sections,only shear flow to the hump at the sides of the wheelpathwas observed. It should be noted that shear flow patternsunder wandered traffic might be different than the patternsobserved for channelized HVS trafficking.

4. Larger shear forces appeared to cause more upward move-ment for the sections constructed with RHMA-G mix(611HB and 612HB). This movement caused larger humpswhich consequently increased the maximum deformationfor these sections. This result can be accepted to be themajor cause of earlier failure for these sections.

5. A significant shear flow pattern in the direction of traffic (Zdirection) was observed for sections constructed withRHMA-G mix. In addition, larger displacement vectors inthe direction of traffic were observed at the hump of thethin-RHMA-G section. The movement in the direction oftraffic can be related to the fact that the HVS wheel ispushed, as opposed to being a traction wheel. This greatershear flow in the direction of traffic can be accepted to beanother contributor to earlier failure of sections constructedwith RHMA-G mix.

6. Although high air void content for the sections constructedwith RHMA-G mix was expected to cause larger densifica-tion related rutting and earlier failure related to this densi-fication, the actual reason behind the earlier failure wasdetermined to be the gap gradation and small aggregate sizewhich caused greater shear flow.

7. Significant differences between the measured plastic andLET predicted elastic displacement fields were observed.

8. Future study

Although the results obtained in this study provide new insight,further study is required to extend the analysis. The methodproposed in this study should also be used to evaluate the ruttingperformance and failure mechanism of other mix types under differ-ent testing conditions. The effects of aggregate shape on rutting per-formance should be determined by using the aggregate shapeparameters that can be obtained from X-ray CT images. Becauseshear related deformation was determined to be the major contrib-utor to the total accumulated rut, a micromechanical finite elementmodel should be developed for a laboratory shear test to evaluatethe effects of aggregate shape and binder properties on shear move-ment. This model can be used to describe the differences betweenthe internal microstructural changes for different mix types underthe same loading and testing conditions.

Most of the current ME rutting performance prediction proce-dures rely on the assumption that distribution of the elastic strainvalues for an asphalt pavement layer calculated by using contin-uum mechanics should be highly correlated with the distributionof the in situ permanent strains. The validity of this assumptionfor different mix and structure types can be checked by comparing

the predicted and measured displacement and strain fields. Thiscomparison will help to understand the deficiencies in currentmethods that have been used for performance simulation.

Particle movement and air void distribution data obtained in thisstudy can also be used to calibrate or validate discrete element mod-els (DEMs). A DEM model simulating the in situ construction com-paction for different mix types can be developed. This DEM modelcan be calibrated using the construction air void content distribu-tions. Based on the calibrated DEM model, the effects of various com-paction variables, temperature, aggregate shape, NMAS, binder type,etc., on the distribution of air voids can be determined. Results canbe used to develop new construction and design specifications.DEM models for simulation of rutting performance can be calibratedbased on the particle movement and air void reduction data.Calibrated models can be used to identify the changes in rutting fail-ure mechanisms when different asphalt layer thicknesses are used.With this method, optimization of design thicknesses for relativelysoft overlays used for reflective cracking that will give adequaterutting performance can be determined.

Acknowledgments

This paper describes research activities that were requested andsponsored by the California Department of Transportation (Cal-trans), Division of Research and Innovation. The test section con-struction and HVS testing were funded by SHRP II Project R21.Caltrans and SHRP II sponsorship is gratefully acknowledged. Thecontents of this paper reflect the views of the authors and do notreflect the official views or policies of the State of California, theFederal Highway Administration or the SHRP II program.

References

[1] Al-Qadi I, Elseifi M. New generation of wide-base tires: impact on truckingoperations, environment, and pavements. J Transport Res Rec 2007;2008:100–9.

[2] White T, Haddock J, Hand A, Fang H. Contributions of pavement structurallayers to rutting of hot mix asphalt pavements. NCHRP Rep No. 468, NationalCooperative Highway Research Program, Transport Res Board, Washington(DC): National Academy Press; 2002.

[3] Monismith CL. WesTrack: performance-related specifications for hot-mixasphalt construction. NCHRP Rep No. 455, National Cooperative HighwayResearch Program Transport Res Board, Washington (DC): National AcademyPress; 2002.

[4] Ullidtz P, Harvey JT, Tsai BW, Monismith CL. Calibration of incremental-recursive flexible damage models in CalME using HVS experiments. CaliforniaDepartment of Transportation Rep No. UCPRC-RR-2005-06; 2006.

[5] Brown SF. Accelerated pavement testing in highway engineering. Proc InstCivil Eng Transport 2004;157(3):173–80.

[6] Mechanistic Empirical Pavement Design Guide (MEPDG). Part 3, designanalysis, Chapter 3, design of new and reconstructed flexible pavements; 2004.

[7] Sousa JB, Deacon JA, Weissman SL, Leahy RB, Harvey JT, Paulsen G, et al.Permanent deformation response of asphalt aggregate mixes. SHRP Rep No. A-415, Strategic Highway Research Program, Washington (DC); 1994. p. 437.

[8] Weissman SL, Harvey JT, Sackman JL, Long F. Selection of laboratory testspecimen dimension for permanent deformation of asphalt concretepavements. J Transport Res Rec 1999;1681:113–20.

[9] Weissman SL. The mechanics of permanent deformation in asphalt-aggregatemixtures: a guide to laboratory test selection. Report for SymplecticEngineering Corp.; December 1997.

[10] You Z, Adhikari S, Dai Q. Three-dimensional discrete element models forasphalt mixtures. J Eng Mech – ASCE 2008;134:12.

[11] You Z, Adhikari S, Kutay ME. Dynamic modulus simulation of the asphaltconcrete using the X-ray computed tomography images. Mater Struct2009;42(5):617–30.

[12] Braz D, Da Motta LMG, Lopes RT. Computed tomography in the fatigue testanalysis of an asphaltic mixtures. Appl Radiat Isot 1999;50(4):661–71.

[13] Warr S, Jacques GTH, Huntley JM. Tracking the translational and rotationalmotion of granular particles: use of high speed photography and imageprocessing. Powder Technol 1994;81:41–56.

[14] Wang LB, Frost JD, Lai JS. Non-invasive measurement of permanent strain fieldresulting from rutting in asphalt concrete. J Transport Res Rec1999;1687:85–94.

[15] Yin HM, Buttlar WG, Paulino GH, Di Benedetto H. Assessment of existingmicromechanical models for asphalt mastics considering viscoelastic effects. JRoad Mater Pavement Des 2008;9:31–57.

E. Coleri et al. / Construction and Building Materials 30 (2012) 36–49 49

[16] Masad E, Muhunthan B, Shashidhar N, Herman T. Internal structurecharacterization of asphalt concrete using image analysis. J Comput Civ Eng1999;13(2):88–95.

[17] Masad E, Jandhyala VK, Dasgupta N, Somadavan N, Shashidhar N.Characterization of air void distribution in asphalt mixes using x-raycomputed tomography. J Mater Civ Eng 2002;14(2):122–9.

[18] Masad E, Button J. Implications of experimental measurements and analysis ofthe internal structure of hot-mix asphalt. J Transport Res Rec2004;1891:212–20.

[19] Braz D, Lopes RT, Da Motta LMG. Analysis of the percentage voids of test andfield specimens using computerized tomography. Nucl Instrum Methods PhysRes 1999;A-422(1):942–8.

[20] Wang Y, Wang L, Harman TP, Li Q. Noninvasive measurement of three-dimensional permanent strains in asphalt concrete with X-ray tomographyimaging. J Transport Res Rec 2007;2005:95–103.

[21] Synolakis CE, Zhou Z, Leahy RM. Determination of internal deformation field inasphalt cores using x-ray computer tomography. J Transport Res Rec1996;1526:135–41.

[22] California Department of Transportation (Caltrans). Highway design manual2006 update chapter 630 flexible pavement. Caltrans, Sacramento (CA); 2006.

[23] California Department of Transportation (Caltrans). Use of rubberized asphaltconcrete for new construction. Memorandum, Caltrans, Sacramento (CA);2006.

[24] Simpleware. ScanIP and +ScanFE software developed by Simpleware Ltd,Innovation Centre, Rennes Drive. Exeter EX4 4RN, UK; 2010. <http://www.simpleware.com>.

[25] American Association of State Highways Transportation Officials. Bulk specificgravity and density of compacted asphalt mixtures using automatic vacuumsealing method. AASHTO T-331; 2009.