Construction and Building Materials 54 (2014) 27–38

Contents lists available at ScienceDirect

Construction and Building Materials

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

Characterisation of micro-structural damage in asphalt mixtures usingimage analysis

0950-0618/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.conbuildmat.2013.12.047

⇑ Corresponding author. Tel.: +60 7 5531678.E-mail address: hnorhidayah@utm.my (N. Abdul Hassan).

Norhidayah Abdul Hassan a,⇑, Gordon D. Airey b, Mohd Rosli Hainin a

a Faculty of Civil Engineering and Construction Research Alliance, Universiti Teknologi Malaysia, Skudai 81310, Johor, Malaysiab Nottingham Transportation Engineering Centre (NTEC), University of Nottingham, University Park, Nottingham NG7 2RD, United Kingdom

h i g h l i g h t s

� X-ray CT Scanning and Digital Image Processing.� Validating the 2D analysis and image Slices Verification for Damage Comparison.� Extracting the damage area from X-ray CT images.� Microstructural Changes in Air Voids Properties and Analysis of the Damage for Specimen Subjected to Compression.� Quantification of crack formation and crack propagation and analysis of the Damage for Specimen Subjected to Fatigue.

a r t i c l e i n f o

Article history:Received 11 July 2013Received in revised form 29 November 2013Accepted 16 December 2013Available online 10 January 2014

Keywords:Image analysis techniqueX-ray computed tomography (CT)Microstructural damageAsphalt mixtures

a b s t r a c t

This paper is an attempt to provide information on the application of an image analysis technique to char-acterise the internal structural damage of asphalt mixtures captured on X-ray images. In this paper, twomodes of failures under uniaxial monotonic compression and indirect tensile fatigue were employed forillustrating the procedures undertaken. Air voids and crack properties as a result of the applied stressesand strains were analysed using two-dimensional image analysis and introduced as damage indicatorsfor characterising the micro-structural damage of asphalt mixtures. A set of procedures for extractingand verifying the damage area were also established by comparing the X-ray images before and afterthe loading application. The proposed damage parameters were showed to be useful for interpretingthe damage behaviour particularly the changes in air void properties and the characteristics of crack for-mation and crack propagation. In addition, it was also found that the damage parameters adopted aresubjected to failure type.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Researchers have been able to utilise X-ray CT images to mon-itor the evolution of internal failure in engineering materials suchas soils, mortar and metal and relate this damage to measuredstrain. Internal failure is measured in terms of internal displace-ment associated with the permanent deformation of the materialsand the formation of cracks within the materials [1–3]. The utilisa-tion of this advanced imaging technology along with image analy-sis techniques has also led to the investigation of asphalt mixturedamage particularly at the micro level. This technique requiresno specimen preparation prior to scanning and therefore the spec-imen is still intact for further mechanical testing after scanningwhich can be effectively used for damage investigation [4–6]. For

asphalt mixtures, X-ray technology has been extensively used tocharacterise air void distribution and microstructural damage evo-lution within asphalt mixture specimens. Air void plays a majorrole in asphalt mixture performance with their distribution beingvery significant in determining the overall mechanical responseof the mixture [7–9]. Under the loading action, the existing airvoids might coalesce resulted in micro-cracks that initiate at theinterface between the aggregates and the mastic. The micro-crackspropagate and grow under the deformation to become macro-cracks and lead to an increase in the air void content [10]. There-fore, it would be of considerable interest to analyse the damagebehaviour in asphalt mixtures by characterising the properties ofthe cracks once the specimens undergo deformation.

Works done by previous researchers in characterising asphaltmixture damage using image analysis were prepared with detailson the air voids distribution and air voids properties but not muchemphasis on the crack properties. Wang et al. [4,5,11] proposed

0

20

40

60

80

100

0.01 0.1 1 10 100

Perc

ent p

assi

ng (

%)

Sieve size (mm)

Fig. 1. Aggregate gradation curves of HRA 60/20.

28 N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38

damage parameters, namely specific damaged surface area (areaper unit volume), average spacing between the damaged surfacesand the average size of defect (average segment length in differentorientation). These parameters can be used to describe the damageand the strength of the interaction between two damaged surfaces.Tashman et al. [6] measured the changes in the voids propertiesthroughout compacted asphalt specimens to characterise the dam-age evolution before and after the deformation caused by triaxialcompression. They were able to relate changes in the voids proper-ties to the materials behaviour at different strain levels and confin-ing pressures. Song [12] quantified microstructure damage interms of the amount of voids and characteristics of voids such asvoids area, average voids perimeter as well as the average radiusof the voids and cracks and relates them to changes in stiffness.Khan [13] used voids parameters including voids fraction (area),number of voids and average voids size to characterise the damagein asphalt mixtures by conducting Monotonic Compression andTension Compression Fatigue Tests which corresponded to the in-crease in strain to failure. The author correlated the area of the de-fects (voids fraction) measured from the X-ray images to the strainand stiffness values obtained from the mechanical testing.

This paper is aimed at providing information through exampleson the establishment of image analysis procedures and parametersfor quantifying the air void and crack properties for damage char-acterisation including the verification method. In addition, theexamples provided is not only to quantify the changes in the airvoid properties after the damage but also to identify the damageconcentration area and the established crack parameters are usedto describe the severity of the damage. The authors’ motivationin writing this paper arises from our observation that, this subjectof image analysis is widely used in various engineering materialcharacterisation including asphalt mixture, but details of the pro-cedures involved is inadequately highlighted in the technical liter-ature and often for complex problem. Even though most of thecurrent studies involved with three-dimensional analysis and itseems crucial in getting a valid analysis to represent the actualsample, but the understanding of the basic two-dimensional mea-surements in damage analysis is also significant as it could providean example of parameter for a specific problem. This is importantfor beginners (who started to get involved in image analysis) tobetter understand the application of image analysis for microstruc-tural damage characterisation. Using the same concept, furtheranalysis on the damage could be extended for a complex problemin either three-dimensional or four-dimensional analysis [14,15].Hopefully the images selected and the accompanying discussionwill help to promote a clearer picture of the image analysis proce-dures for micro-structural damage investigation.

2. Material and methods

2.1. Specimen preparation and testing

In this study, a gap graded mixture of Hot Rolled Asphalt (HRA 60/20) was se-lected in accordance to BS 954-1:2005 [16]. The HRA 60/20 contains 60 percentcoarse aggregate with a maximum aggregate size of 20 mm. The specimen wascompacted using gyratory compactor with a vertical pressure (0.6 MPa), angle ofgyration (1.25�) and gyration speed (30 rpm). Details of the mix design, specimendimension and test condition are shown in Fig. 1 and Table 1. For damage evalua-tion, a total of 16 specimens were damaged under the Uniaxial Monotonic Com-pression (UMC) and Indirect Tensile Fatigue (ITF) Tests with four replicates foreach condition as specified in the table. For the Uniaxial Monotonic Compressiontest, the specimens were conditioned at the target test temperature and the testwas performed at a constant strain rate of 0.1 mm/s. This strain rate was adoptedin such a way that the load was sustained for 100–200 s. This test length was cho-sen to ensure that the tested specimens reached the failure stage within a reason-able testing time, thereby reducing the effect of aging on the other queuedspecimens (under conditioning) waiting to be tested. For the Indirect Tensile Fati-gue Test, the repeated load acts along the vertical diameter and produces variousmagnitudes of vertical compressive stress and horizontal tensile stress along the

diameter of the specimen. The application of non-destructive imaging technique,X-ray CT has reduced the number of specimen for this investigation. The specimenswere X-rayed before and after the mechanical tests.

2.2. X-ray CT Scanning and Digital Image Processing

In this study, the X-ray CT was used to capture digital images of the internalstructure of asphalt mixtures. There are three main components of the X-ray CT ma-chine including the X-ray source, specimen position and the detector of the X-raysintensity after penetrating the specimen. The measured intensities are then con-verted to a map of internal structure distribution in grey scale which depends onthe density and size of the object. For 8-bit image, the grey scale consists of 256 lev-els. Brighter regions (close to white) correspond to the higher density materialswhile darker regions (close to black) correspond to the lower density materials.The captured images were then processed for quality enhancement before quanti-tatively analysed for measuring and interpreting the properties using an imageanalysis technique. Fig. 2 shows the different stages undertaken in the image pro-cessing technique for extracting the image of air voids within the asphalt mixturespecimen. Basically, the digital image processing involves a set of procedures thatapply various algorithms to enhance the image quality, threshold and extract theobject of interest within the captured digital images. In this study, these procedureswere undertaken using ImageJ, the public domain Java Image processing program,inspired by National Institute of Health (NIH) Image. Details of the proposed digitalimage processing can be found in Abdul Hassan Airey [17].

3. Proposed Microstructural Damage Analysis

The following procedure has been established for characterisingthe microstructural damage. It consists of details for verifying theimage slices (by X-ray) before and after deformation for damagecomparison made on 2D images and the proposed methods forextracting the damaged area (i.e. identifying increases in voidsand cracks after deformation). For damage analysis, two main cat-egories of damage parameters are specified based on differentdamage mechanisms, namely microstructural changes in voidproperties and the quantification of crack formation and crackpropagation.

3.1. Validating the 2D analysis

To better interpret the measurements made on the X-rayimages from 2D analysis, it is necessary to ascertain the image sliceinterval required to obtain an unbiased characterisation of thespecimens’ microstructural properties. The chosen interval wasconsidered practical with respect to obtaining optimal resultsand the time necessary to scan, as more images will need moretime to complete the CT scanning. An asphalt specimen wasX-rayed and analysed for air voids distribution at different sliceintervals (between 0.1 mm and 8 mm) as shown in Fig. 3a. It canbe seen that the curves of the air voids distribution analysed at0.1–2 mm are very close with minimal different in the air voidscontent along the height of the specimen. They also seem to convey

Table 1Details of specimens for Uniaxial Monotonic Compression (UMC) and Indirect Tensile Fatigue (ITF) Tests.

Mechanicaltest

Testing details Percent air voids(%)

Design bitumen content(%)

Bitumentype

Aggregatetype

Approximate dimensions(mm)

UMC - Test temperature, 25 �C 5.0 5.7 100/150 pen Granite Diameter – 100; height – 100- Constant strain rate, 0.1 mm s�1

ITF - Stress level, 100, 200 and300 kPa

5.0 5.7 100/150 pen Granite Diameter – 100; height – 40

-Loading rate, 40 pulses/min- Test temperature, 20 �C

Original X -ray image Image enhancement Image thresholding Binary image of air voids

Air voids

Fig. 2. Image processing technique.

(a) (b)

0

10

20

30

40

50

60

0 2 4 6 8 10 12

Spec

imen

's h

eigh

t (m

m)

Air voids content (%)

0.1 mm0.5 mm1 mm2 mm4 mm8 mm

4.70

4.75

4.80

4.85

4.90

4.95

5.00

5.05

1.10

1.15

1.20

1.25

1.30

1.35

1.40

1.45

1.50

1.55

0 2 4 6 8 10

Ave

rage

air

voi

ds c

onte

nt (

%)

Stan

dard

dev

iatio

n

Slice interval (mm)

SD Average voids

Fig. 3. (a) Analysis of air voids distribution and (b) standard deviation and average voids content at different slice intervals.

N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38 29

more detail of the air voids distribution compared to the curvesanalysed at 4 mm and 8 mm slice intervals. A statistical validation,specifically standard deviation, SD also provided a good indicationof the variation from the mean or average value calculated from anumber of images. Fig. 3b shows that fewer images (with higherslice intervals) will lead to a higher standard deviation of the aver-age values of the air voids content. Therefore, this study hasadopted a 1 mm slice interval for the scanning.

3.2. Image Slices Verification for Damage Comparison

For damage characterisation, the compacted asphalt mixturespecimens were scanned before and after conducting the destruc-tive test to assess any changes to their microstructure. The de-formed specimens were noted to have reduced in height(uniaxial monotonic compression) or having a different diametralgeometry (indirect tensile fatigue). The deformed specimen wasthen X-rayed at the same slice interval as used for scanning before

the test, which produced fewer image slices and some aggregatedisplacements. This process can cause misinterpretation or mea-surement errors as the captured image slices (of the same speci-men) would technically be at a slightly different location afterundergoing deformation. Therefore, to ensure that the analysedX-ray image slices of the specimen are comparable before and aftertesting (which means the image slices are compared exactly or atthe closest possible to the initial location before testing), the greylevels of the image slices prior to and following testing are verified.By plotting the total number of pixels against the grey level ofimages taken before and after testing, a pair of curves are obtained,which if they match or can be superimposed on one anotherexactly, determines the extent to which the slices differ. The curvesthat overlap the most determine that the images from the closestlocation can be compared for measuring the changes in the micro-structural properties. In other words, if the difference in the totalpixel number for the specified grey level between the image slices(before and after test) is close to zero, it means they are from

(a) 5B (b) 6B (c) 5A

Fig. 4. X-ray images of asphalt mixture before the test (a) slice 5 (5B), (b) slice 6 (6B) and after the test (c) slice 5 (5A) (air voids highlighted in red). (For interpretation of thereferences to colour in this figure legend, the reader is referred to the web version of this article.)

(a) (b)

0

10000

20000

30000

40000

0 50 100 150 200 250

Tot

al n

umbe

r of

pix

els

Grey level

slice 5A slice 5B slice 6B

-25000

-15000

-5000

5000

15000

25000

0 50 100 150 200 250

Dif

fere

nce

in to

tal n

umbe

r of

pix

els

Grey level

difference between slice 5A and 6B

difference between slice 5A and 5B

Fig. 5. (a) The total number of pixels and (b) the difference in the total number of pixels for different grey level within the X-ray images.

30 N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38

exactly the same location or height. Examples of the images andplots described above are illustrated in Figs. 4 and 5. Fig. 4 showsthe image slices of an asphalt mixture specimen before (slice 5Band slice 6B) and after (slice 5A) the test. By comparing the grey le-vel distribution and the difference in total number of pixels in bothimages i.e. before the test and after the test, as shown in Fig. 5, it isclear that slices 6B and 5A are more comparable than slices 5B and5A. Therefore, for damage characterisation, the changes in themicrostructural properties of slices 6B and 5A would be more suit-able candidates for comparison.

3.3. Extracting the damage area

In order to characterise the connectivity between voids or thepreferred travel path of cracks, it is imperative to quantify the in-crease in voids area (or damaged area), Ai, caused as a result ofdeformation. This study proposes a procedure for differentiatingand extracting the image of the original air voids area, Ao (speci-men in an undamaged state) from the total final air voids content,At of an X-ray image slice of a compacted asphalt mixture specimenin its damaged state using digital image processing. This is ex-plained using Eq. (1). By differentiating and extracting the increasein voids area from the original void area, the characteristics andconcentration of the damaged areas within the material’s composi-tion can be analysed.At ¼ Ao þ Ai ð1Þ

The extraction procedure utilises a superimposition techniqueor so called ‘operations between images’ which involve a set ofmathematical operations namely, image logical operations. Forgetting good results from these operations, the basic restrictionis that the initial images have to be of the same dimensions anddata type. These operations are very useful for feature extractionwhen dealing with binary images. In common practice, there arethree basic functions of logic operators i.e. AND (logical intersec-tion), OR (logical sum) and NOT (negative of image). Other logicoperators can be implemented by combining these three basicfunctions including XOR (exclusive or) and AND(NOT) (logical dif-ference). For example, the ‘logical difference’ can be used to detectdifferences between two images. In this study, the above men-tioned operations are performed on two stacks of X-ray images(before and after deformation) using the ImageJ software wherethese operations are utilised for voids extraction on binary images.Details of these logical functions will not be discussed further inthis document with complete details contained in Gonzalez andWoods [18], but the transformations resulted from these logicaloperations will be demonstrated using a simple example as illus-trated in Fig. 6. The illustration describes the original void area(in yellow), A, damaged area (increase in voids), B, voids after dam-age (A + B) (green), C and overlap area between A and B (red), D.For the two overlapping images (between Fig. 6a and b), it canbe seen in Fig. 6c, that the position of the voids created by the dam-age deviates from the location of the initial voids. As a result of the

(A) AND (C) (overlap area)

Image of undamaged Image after damage (A) OR (C)

(A)

(C)

(C) AND[NOT(A)](damaged area)

(a) (c)(b)

(d) (f)

Both not well overlapped (deviates)

(B)

(D)

(D)

(A) XOR (C)

(e)

Fig. 6. Illustration of the different logical operations on image A (initial void) and image C (voids after damage) for (A – D).

N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38 31

added discrepancy in voids configuration, the extracted damagedarea in Fig. 6f is slightly higher than the actual damaged area be-cause the difference is occupied by the original void area.

However, based on a number of extraction procedures con-ducted in this study, the difference between the selected damagedareas (from the extraction procedure) to the actual increase in thevoids content was found to be very minimal. Moreover, the pur-pose of this extraction procedure is to obtain an overview graphicalrepresentation of the voids propagation caused by the damage,thereby enabling the concentration area of the damage formationto be identified. By assigning with different colours1, the initialvoids, voids caused by the damage and the overlap area betweenthem can be differentiated as shown in Fig. 7. It can be seen that,the damaged area is concentrated within the circled area.

Yellow = existing voidGreen = voids due to damageRed = overlap area

Higher damaged area

3.4. Microstructural Changes in Air Voids Properties

A few parameters have been successfully used to characteriseair voids properties such as the voids content, voids number, aver-age voids size and voids shape. This method is well established inprevious research work [4,6,7,13,19]. In the study of microstruc-tural damage, changes in these parameters (by comparing theparameters before and after deformation), provide valuable infor-mation on the severity of the damage at different depths withinthe compacted asphalt mixture specimen. In other words, theyare able to reflect the accumulation of damage under differentloading conditions. Number of voids and average voids size havebeen found to be significant for the characterisation of air voidsproperties. This is based on an observation which showed thattwo replicate specimens, produced with the same target air voidscontent, can possibly have different distributions of air void num-ber and average air voids size throughout both specimens. Speci-mens having the same total air voids content could be a result ofcontributions from a lesser number of large air voids or highernumber of smaller-sized air voids. As a consequence, two replicatespecimens may fail differently under the same loading conditioneven though they would have been designed to achieve the sametarget air voids content. In comparison to smaller void sizes,

1 For interpretation of colour in Fig. 7, the reader is referred to the web version ofthis article.

samples with large voids are more susceptible to relatively earlierfailure because of the increased likelihood of larger strains beinginduced once the specimen is subjected to loading conditions.The scenario described may cause instability, where the existingvoids may potentially propagate throughout the specimen andinterlink (or coalesce with other voids) to cause severe damage.Wang et al. related the air voids content to the concept of effectivestress [11]. The concept agrees that the presence of more air voidswill decrease the mixture’s resistance against pavement damage.This is because air voids cannot transmit loading and thereforeweaken the material due to the reduced effective area availableto resist the applied forces (higher concentrated forces per area).

Shape factor is another important air voids property that is veryuseful for differentiating between original air voids and cracks in aspecimen after deformation [6]. Many earlier studies used shapeanalysis to evaluate aggregates’ characteristics in terms of the sur-face irregularities or angularity [20–22]. The most common shapefactors used in image analysis are circularity, roundness and aspectratio. These parameters measure the extent to which a particleshape differs or deviates from a perfect circle or sphere, whichare ideal reference shapes. Shape factor can be considered as ameasure of the compactness of a shape, where a circle is regardedas the most compact shape, such that more compact a shape, themore closely it resembles a circle. If the circularity, roundness

Fig. 7. Illustration of the voids area in compacted asphalt mixture specimen afterdeformation under Indirect Tensile Fatigue Test.

Peri

met

er

Length of major axis

Cir

cula

rity

Roundness Aggregate1.0 0.0

0.0

0 0.01 0.1 1 100.001

Particle length X-axis (mm)

0.01

0.1

1

10

0.001

SphereAspect ratio = 1

Asp

ect r

atio

> 1

Aspect ratio > 1

Fig. 8. Illustration of shape characteristics.

32 N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38

and aspect ratio values are close to 1.0, it indicates circular airvoids while cracks have very high aspect ratio and low values ofroundness and circularity (approaching 0.0), indicating an increas-ingly elongated shape.

In this study, the changes in the air void properties were quan-titatively analysed along the specimen’s height (vertical) by mea-suring its content, number, size and shape (aspect ratio;circularity and roundness) using Eqs. (2)–(6). Fig. 8 illustrates theparticles shape characteristics in accordance to the definitions ofcircularity, roundness and aspect ratio.

Air voids fraction ð%Þ ¼ Air voids area ðmm2ÞTotal image area ðmm2Þ � 100 ð2Þ

Average air voids size ¼ Total voids areaNumber of voids

ð3Þ

Circularity ¼ 4p� Area

Perimeter2 ð4Þ

Roundness ¼ 4� Area

p�Major axis2 ð5Þ

Aspect ratio ¼ Length of major axisLength of minor axis

ð6Þ

3.4.1. Analysis of the Damage for Specimen Subjected to CompressionThe distribution of the air voids properties for the compacted

asphalt mixture tested under compression at 25 �C are shown inFig. 9. The bar graphs compare the changes in the air voids proper-ties by averaging the properties along the height of the specimensat 5 mm intervals. An observation of the individually presented airvoids parameter distributions along the specimen’s height, led toseveral noteworthy observations. Significant changes in the prop-erties distribution along the specimen’s height were revealed inthree main regions referred to as bottom (0–30 mm or 0–0.3 rela-tive height), middle (30–70 mm or 0.3–0.7 relative height) and top(70–100 mm or 0.7–1.0 relative height). Relative height is the ratioof the referred specimen height (at any level) to the total height ofthe specimen. The total height of the specimen initially before thetest was approximately 100 mm, but became less after it was de-formed in the compression test. Therefore the relative height ismore appropriate for the air voids analysis to better describe theexact location of the damage. Explanations on the measured airvoids parameters (changes in air voids content and air voids shapeproperties) are significantly correlated in the following sub-sections with reference to Fig. 9.

3.4.1.1. Changes in the air voids content properties. The distributionsof total voids number, total voids area and the average voids sizewere compared as shown in Fig. 9(a–c). The total increase in theair voids area and average void size were found to be higher inthe middle to top region. This was in agreement with the minordilation observed to have occurred at the middle region of thetested specimen after it had been mechanically tested. The per-centage reduction in the total voids number is attributed to theair voids coalescing to form micro-cracks while the minor increaseat the middle region of the specimen is thought to be the result ofnew air voids formation under the loading. Before testing, it wasidentified through air voids distributions that undeformed speci-mens have higher voids content and voids sizes in their top andbottom regions than in the middle region. However, a higher voidscontent and voids sizes that initially exist within the specimenprior testing do not necessary mean a potential for damage initia-tion. This is because, although the initial air voids content (beforedeformation) in the few slices at the top and bottom regions washigher than in the middle region, the majority of the assessed dam-age following mechanical testing appeared to have occurred in themiddle and top regions.

In this study, the changes in the total voids number and totalvoids area were identified as two key parameters when describingdifferent damage phenomena that occur within a specimen as a re-sult of deformation. According to previous studies, two mainmechanisms occur under permanent deformation namely, densifi-cation and shear deformation [12,23,24]. By comparing thechanges in the voids number and voids area analysed from X-rayimages, a number of observations can possibly be correlated tothe damage mechanisms associated with the deformation undercompression as summarised in Table 2. The association betweenthe damage mechanisms and the air voids properties is then fur-ther clarified in the subsequent sub-section with respect to thechanges in air voids shape properties.

3.4.1.2. Changes in the air voids shape properties. Fig. 9(d–f) showsthe percentage changes in the air voids shape properties i.e. the as-pect ratio, circularity and roundness. The figure indicates that theair voids became increasingly elongated suggesting that crackgrowth was initiated within the specimen, under the applied load.With the voids size increase, two adjacent air voids are likely tohave coalesced forming an air void with an aspect ratio higher thantheir pre-coalescing aspect ratios and leading to the initial forma-tion of micro-cracks (increase in length). This again confirms theprevious finding on the changes in the air voids content properties,which proposed that a decrease in voids number is indicative of thepresence of micro-cracks and macro-cracks in the mechanicallytested specimens and such may have contributed to the dilation

-100 -50 0 50 1000.05

0.15

0.25

0.35

0.45

0.55

0.65

0.75

0.85

0.95

Change in total void area (%)

Rel

ativ

e he

ight

-100 -50 0 500.05

0.15

0.25

0.35

0.45

0.55

0.65

0.75

0.85

0.95

Change in void number (%)

Rel

ativ

e he

ight

-50 0 50 100 1500.05

0.15

0.25

0.35

0.45

0.55

0.65

0.75

0.85

0.95

Change in average void size (%)

Rel

ativ

e he

ight

-10 0 10 20 300.05

0.15

0.25

0.35

0.45

0.55

0.65

0.75

0.85

0.95

Change in void aspect ratio (%)

Rel

ativ

e he

ight

-10 -5 0 5 0.05

0.15

0.25

0.35

0.45

0.55

0.65

0.75

0.85

0.95

Change in void circularity (%)

Rel

ativ

e he

ight

-15 -10 -5 0 5 0.05

0.15

0.25

0.35

0.45

0.55

0.65

0.75

0.85

0.95

Change in void roundness (%)

Rel

ativ

e he

ight

(a) (b) (c)

(d) (e) (f)Fig. 9. Changes in void properties (a) total area; (b) number; (c) average size; (d) aspect ratio; (e) circularity; (f) roundness for specimen subjected to compression loading.

Table 2Correlation between damage mechanisms and changes in air voids content properties.

Damagemechanism

Changes in total voidsnumber

Changes in total voidsarea

Damage description Observations from X-ray images analysis

Densification Decrease Decrease - Material’s hardening - Fig. 10 (a and b); bottom section- Air voids closure

Shear deformation Decrease Increase - Debonding at binder-aggregateinterface

- Fig. 10 (a and b); middle and topsections

- Microcracks initiate and coalesce

N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38 33

observed in the middle region of the specimen. Overall, eventhough there was evidence of increasing total air voids area andaverage voids size of the specimen after deformation, measure-ments of voids number and shape properties provide a differentway of describing air voids properties in terms of air voids forma-tion, growth, connectivity and propagation. The damage was deter-mined to have occurred as a result of small air voids coalescing toform micro-cracks, increasing the aspect ratio of the voids andthereby reducing the voids number and circularity value. This find-ing concurs with two failure types identified by Kim et al. [25] whoasserted that the damage in an asphalt mixture is the result of mi-cro-cracks that initiate in the interfacial transition zone (ITZ) be-tween bitumen and aggregates (adhesive failure) and within thebinder (cohesive failure) [25].

The results appear to conform the dilation phenomenon that oc-curred at the middle region of the specimen. The results clearlyshow that the middle region was more susceptible to damage than

the top and bottom regions, having experienced higher percent in-creases in the air voids area and average voids size at that particu-lar location. Fig. 10 shows X-ray images of the damage distributionat three main regions of an asphalt mixture specimen with relativeheights of 0.1 (bottom), 0.5 (middle) and 0.9 (top) with air voidshighlighted in red. The figure compares images of the specimen be-fore and after mechanically testing. It must be noted that theimages (before and after deformation) have been compared onthe basis of identical height ratios. Due to some movement of theaggregate particles during testing, there are minor differences be-tween sets of compared images. In general it can be observed thatthe images’ characteristics are consistent with the findings de-scribed earlier for the measured changes in the air voids properties.For example, severe damage was evident within the middle andtop regions of the control specimen with micro-cracks initiatingin the ITZ area following deformation. This phenomenon is evenmore prevalent with increasing aggregate sizes due to the manner

TTopp reggionn MMidddlee reegioon

IITZZ craackss

Botttomm rregiion

Fig. 10. Comparison of image slices of asphalt mixture specimen, before (top images) and after (bottom images) the deformation at 25 �C.

34 N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38

in which the micro-cracks tend to bridge between proximal aggre-gates. After being formed, the micro-cracks and air voids arethought to have propagated to form macro-cracks, therebydecreasing the specimen’s strength. From the observations madeon a number of specimens, it was determined that damage gov-erned by the changes in air voids structure occurs in a variety ofways. First the changes might contributed by the increase in thesize of the existing individual air voids and propagate to formcracks. Second, two separated air voids might coalesce and propa-gate to form cracks within the vicinity of the damage. Third, as aresult of the deformation, new air voids are also formed having po-tential for crack initiation. These observations concur with thosethat were previously recorded in Table 2. It is also noteworthy thatthey are not mutually exclusive and can act simultaneously to ini-tiate micro-cracks and macro-cracks under the loading.

3.5. Quantification of crack formation and crack propagation

This investigation is intended to apply the concept of fractalanalysis to study the formation and propagation of cracks in themixtures. For decades this concept has been used in many

3×3 11×1

Fig. 11. Example of a crack image analyse

disciplines including medicine, geology and material sciences. Forasphalt mixtures, it has been used to measure the geometry ofirregular shapes and the surface roughness of aggregate particles,which are both important factors with potential to affect themechanical properties of asphalt mixtures [26,27]. In cement andconcrete research, fractal analysis has been widely adopted forcrack analysis [28–31]. Outputs of interest derived from fractalanalysis, containing crack information include fractal dimension,crack tortuosity, crack branching factor and crack density.

Fractal dimension, D, is an application of fractal analysis tocharacterise the shape of a crack’s boundary or contour. It is anon-integer number that measures the degree of fractal boundaryfragmentation or irregularity over multiple scales. The fractaldimension of cracks was determined using the ‘Box CountingMethod’ under the application of FracLac in ImageJ developed byKarperien [32]. The approach measures the lengths or distancesbetween points on the border of a shape using sets of square boxesor grids. Fig. 11 displays images showing the grids or arrays ofboxes of different sizes (e) overlaying the crack images. The boxesthat contained the pixels of the cracks image were counted (N) forthe plot of ln(box count, N) versus ln(box size, e), which is used to

1 24×24

d with different grid sizes (in pixels).

Binary image of cracks ln(box count, N) vs ln(box size, ε) with D=1.311

Fig. 12. Example plots obtained from the fractal dimension analysis for a binary image.

Notch direction

Main crack: Lm

Projected crack: Lp

Crack branching: Lb

Crack tortuosity: Lm/ Lp

Branching factor: Lb/ Lm

Fig. 13. Illustration of the crack descriptions [28].

Fig. 14. Conversion of binary image

N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38 35

obtain the fractal dimension, D, that is the slope of the logarithmicregression line as shown in Fig. 12. The fractal dimension is calcu-lated using Eq. (7). Higher fractal dimensions denote a wider crackor larger crack area. If the measured crack areas are the same, dif-ferent crack distributions (or propagation patterns) will producedifferent fractal dimensions.

D ¼ � LogðNðeÞÞLogðeÞ ð7Þ

Besides the use of fractal dimension which takes into accountthe area of the cracks, other crack parameters, namely crack tortu-osity, crack branching factor and crack density, characterise thecrack networks based on the crack length. It is necessary to exam-ine these parameters in order to understand crack propagationwithin a material’s structure due to deformation. Crack density(expressed in mm/mm2) can be calculated as the ratio of the totalcrack length, ‘, to the total image area, A, under consideration(refer to Eq. (8)) [33,34].

Crack density ¼PN

i¼1liA

ð8Þ

of cracks into cracks skeleton.

40

(a)

0 mmm10

10m

0mm

mm

((b)

10

10m

0mm

mm

Fig. 15. Image of specimen cross section for compacted asphalt mixture (a) before and (b) after ITFT.

(a) 10 mm

36 N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38

Fig. 13 illustrates the crack descriptions as in Song et al. [28].According to Song et al. [28] and Zhang and Nagy [35], crack tortu-osity is the ratio of the main crack length (true length), Lm, to theprojected length or linear distance between its end points, Lp.The crack branching factor is the ratio of the total length of allthe branches of the crack, Lb, to the main crack length [28,35].The higher the ratio of both aforementioned properties, the largerthe area over which the cracks can be said to be propagating orspreading. The measurement of the crack lengths is enabled bythe conversion of the binary image of the crack to crack skeletonsas shown in Fig. 14.

0

50

100

150

200

250

300

350

0 10 20 30 40 50 60 70 80 90 100 110

Mea

sure

d da

mag

e ar

ea (

mm

2 )

Box width (mm)

(b)

Higher damage area

Small increase in the damage area

Fixed height

20 mm

30 mm

40 mm

Box width Increase at interval 10 mm

Fig. 16. (a) Illustration of the selected area for different box widths and (b) plot ofdamage area vs box width.

3.5.1. Analysis of the Damage for Specimens Subjected to FatigueThe aforementioned crack properties were adopted to evaluate

the fatigue behaviour of the asphalt mixture. The specimens weretested using the Indirect Tensile Fatigue Test (ITFT) in controlledstress mode. Following the ITFT, the accumulated damage in thespecimens were then further characterised by analysing the result-ing cracks properties. This was necessary to ascertain the severityof the damage caused by the fatigue testing. The damage character-isation was achieved using X-ray images of the specimens, scannedafter they reached the failure stage (at 9 mm vertical deformation).Fig. 15 compares the X-ray CT images of the specimens’ cross sec-tions before and after the ITFT using an image grid size of 100 mm2.The image slice shown is the one closest to the surface of the crosssection. It can be seen from the images that the main cracks, alsoknown as diametric cracks, are formed in the middle area of thecross section. The remaining portions of the cross-sectional areaof the slices show very small increases in air voids area, thoughtto be possibly caused by specimen disintegration at failure. In thisstudy, cracks were analysed in the enclosed highlighted region toensure that only the main cracks initiated due to fatigue failurewere taken into consideration. This was done to reduce variationsin the measured parameters which would have been affected bythe changes in air voids distribution. The areas under considerationwere chosen as the 40 mm wide region located at the middle of thespecimens’ cross sections. The main cracks caused by the loadingaction were identified to be concentrated within this area.

Identification of the 40 mm crack envelope was establishedthrough an assessment of the damaged specimen at 10 mmintervals from the central 10 mm width of the specimen, which

progressively confirmed damage increasing to a maximum, witha crack envelope 40 mm wide (refer to Fig. 16a). Beyond 40 mm,the damage observed was only marginal and would have onlyintroduced variations on the measured properties of the sampleas previously mentioned. Fig. 16b shows the plot of damage areaagainst the box width, where it can be seen that, as the box widthincreased up to 40 mm, the damage measured within the box area

Table 3Results of crack analysis before and after the fatigue test.

Crack properties Test stress level (kPa)

100 200 300

Before After Before After Before After

Fractal dimension 1.390 1.428 1.215 1.310 1.095 1.394Crack density (mm/mm2) 0.009 0.063 0.017 0.064 0.014 0.074Crack tortuosity (mm/mm) – 1.417 – 1.392 – 1.394Crack branching factor (mm/mm) – 0.212 – 0.158 – 0.208

N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38 37

increased significantly indicating a very large damage area. Wherethe box width exceeds 40 mm, the plot is almost horizontal indi-cating very little increase in the measured damage area. Stacks ofimages were analysed and the results of crack properties throughthe thickness of the specimen were measured at different stresslevels as summarised in Table 3.

In this study several replicates of different specimens weretested at different stress levels so it was more useful to comparethe increase in the crack properties before and after the test to as-sess the severity of the resulting cracks. This was done to avoidsome existing elongated air voids within the specimen (before test-ing) being detected as micro-cracks during the analysis of the spec-imen after it was tested. Comparing the values before and after thefatigue test, it can be concluded that the values of fractal dimen-sion and crack density after the test are higher, indicating damagegrowth or increase in air voids size (area) and the development ofcracks. It can be seen that the increase in the values became higheras the test stress level increased. Consequently an increase in crackdensity corresponds to more cracks propagating within the areaand forming crack networks under the action of loading. In addi-tion, larger air voids could potentially grow into cracks and spreadall over the area, resulting in an increase in the crack density. An-other observation is that the percentage increase in crack densityvalue is highest at the lowest test stress level (100 kPa) indicatingthe initiation of major crack networks within the specimen. On theother hand, at a higher test stress level, the cracks coalesced toform larger individual cracks (with less branches).

The crack tortuosity and the crack branching factor are param-eters that can be used as instruments to determine the fracturetopography. Basically, the greater the value of crack tortuosityand crack branching factor of a material, the greater the material’scapability to resist crack growth. This is the case because the mate-rial is able to provide tortuous crack paths as observed for speci-mens tested at low stress level to prevent sudden catastrophiccrack growth which is inhibited by ‘crack blunting’ (refer toFig. 17). This phenomenon refers to the increase of plastic deforma-tion prior to crack propagation. By this means, cracks can bearrested and diverted, averting sudden failure of the material.

Crack blunting

Less tortuosity

Higher tortuosity

Main crack deviates

Sharp crack branch

Crack continues to propagate

Crack tip zone

Fig. 17. Illustration of crack tortuosity in crack growth.

However at higher stress level, the specimen is found to experiencea spontaneous fatigue failure where the energy of the propagatingcrack exceeds the cohesive strength of the material at the crack tipzone and the area become least resistance to catastrophic crackgrowth. This can be observed in all the specimens where a largenumber of cracks are present in the interfacial transition zonearound the larger aggregate particles and mastic. This is believedto have been caused by insufficient bonding between the aggre-gates and the mastic. The main cracks are shown to have propa-gated through the binder and fines. This shows that cracktortuosity is an important parameter affecting the fracture tough-ness of the material because it has the potential to prevent thelocalised stress intensity in the vicinity of the crack tip from reach-ing a critical level where catastrophic crack propagation occurs.

4. Conclusions

Based on the damage analysis provided, a number of parame-ters have been recommended to quantify the damage within as-phalt mixtures describing the type, severity and extent ofdamage. The changes in the air voids content and shape propertiescan be successfully used to describe the damage associated withthe deformation under compression. Whereas, the fractal dimen-sion, crack density, crack tortuosity and crack branching factorare good indicators for characterising the fracture topography ofcrack growth. The results demonstrated that more informationon the damage can be revealed by comparing the aforementionedproperties extracted from the images. In addition, the methodsproposed for the validation of two-dimensional analysis and imageslices and also for extracting the damage area, are found to be veryuseful and reliable.

Acknowledgements

The authors wish to acknowledge the support of the Universityof Nottingham, United Kingdom and special thanks go to thenumerous laboratory personnel from the Nottingham Transporta-tion Engineering Centre, University of Nottingham who assistedin conducting this study. The authors also acknowledge the PhDscholarship support contributed by the Ministry of Higher Educa-tion, Malaysia and Universiti Teknologi Malaysia.

References

[1] Shi B, Murakami Y, Wu Z, Chen J, Inyang H. Monitoring of internal failureevolution in soils using computerisation X-ray tomography. J Eng Geol1999;54(3–4):321–8.

[2] Landis E, Keane D. X-ray micro-tomography for fracture studies in cement-based materials. Int Soc Optical Eng 1999;3772:105–13.

[3] Maire E, Babout L, Buffiere JY, Fougeres R. Recent results on 3Dcharacterization of microstructure and damage of metal matrix compositesand a metallic foam using X-ray tomography. J Mater Sci Eng 2001;A319–321:216–9.

[4] Wang LB, Frost JD, Shasidhar N. Microstructure study of westrack mixture fromX-ray tomography images. Transportation Research Record 1767,Transportation Research Board, Washington, DC; 2001. p. 85–94.

38 N. Abdul Hassan et al. / Construction and Building Materials 54 (2014) 27–38

[5] Wang LB, Frost JD, Voyiadjis GZ, Harman T. Quantification of damageparameters using X-ray tomography images. Mech Mater 2003;35:777–90.

[6] Tashman L, Masad E, Little D, Lytton R. Damage evolution in triaxialcompression tests of HMA at high temperatures. Assoc Asphalt PavingTechnol 2004;73:53–87.

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

[8] Pellinen TK, Song J, Xiao S. Chracterisation of hot mix asphalt with varying airvoids content using triaxial shear strength test. In: Proc. of the 8th conferenceon asphalt pavements for southern, Africa; 2004.

[9] Airey GD, Hunter AE, Collop AC. The effect of asphalt mixture gradation andcompaction energy on aggregate degradation. Constr Build Mater2008;22:972–80.

[10] Tashman L, Wang LB, Thyagarajan S. Microstructure characterisation formodelling HMA behaviour using imaging technology. Int J Road MaterPavement Des 2007;8(2):207–38.

[11] Wang LB, Paul HS, Harman T, D’Angelo J. Characterization of aggregates andasphalt concrete using X-ray tomography. Assoc Asphalt Paving Technol2004;73:467–500.

[12] Song I. Damage analysis in asphalt concrete mixtures based on parameterrelationships. Ph.D. Thesis. Texas A & M University, US; 2004.

[13] Khan R. Quantification of microstructural damage in asphalt. PhD. Thesis,University of Nottingham, Nottingham; 2009.

[14] You T, Abu Al-Rub R, Darabi M, Masad E, Little D. Three-dimensionalmicrostructural modeling of asphalt concrete using a unified viscoelastic–viscoplastic–viscodamage model. Constr Build Mater 2010;28:531–48.

[15] Bhasin A, Izadi A, Bedgaker S. Three dimensional distribution of the mastic inasphalt composites. Constr Build Mater 2011;25:4079–87.

[16] British Standards Institution. Hot rolled asphalt for roads and other paved -Part 1: Specification for constituent materials and asphalt mixtures. BS 594–1:2005, BSI, London; 2005.

[17] Abdul Hassan N, Airey GD. X-ray computed tomography and digital imageprocessing for thresholding asphalt mixture constituents. In: Proc. of the 14thREAAA international conference, Kuala Lumpur, Malaysia; 2013. p. 799–812.

[18] Gonzalez RC, Woods RE. Digital image processing. 3rd ed. UnitedStates: Prentice Hall; 2008.

[19] Masad E, Muhunthan B, Shashidhar N, Harman T. Internal structurecharacterization of asphalt concrete using image analysis. J Comput Civ Eng(Special Issue on Image Processing) 1999;13(2):88–95.

[20] Masad E, Little DN, Tashman L, Saadeh S, Al-Rousan T, Sukhwani R. Evaluationof aggregate characteristics affecting HMA concrete performance (Report No.ICAR 203-1). Texas A&M University System College Station, Texas: TexasTransportation Institute; 2003.

[21] Saadeh S, Masad E, Garboczi E, Harman T. Aggregate shape analysis using X-ray Computed Tomography. In: Proc. of the 11th symposium of theinternational center for aggregate research, Austin; 2003.

[22] Garboczi EJ. Three-dimensional mathematical analysis of particle shape usingX-ray tomography and spherical harmonics: application to aggregates used inconcrete. Cem Concr Res 2002;32:1621–38.

[23] Roberts FL, Kandhal S, Brown ER, Lee DY, Kennedy TW. Hot mix asphaltmaterials, mixture design, and construction, 2nd ed., NAPA Research andEducation Foundation 585; 1996.

[24] Garba R. Permanent deformation properties of asphalt concrete mixtures. PhD.Thesis, Norwegian University of Science and Technology, Trondheim, Norway;2002.

[25] Kim YR, Lee HJ, Kim Y, Little DN. Mechanistic evaluation of fatigue damagegrowth and healing of asphalt concrete: Laboratory and field experiments. In:Proc. of the 8th international conference on asphalt pavements, internationalsociety for asphalt pavements, University of Washington, Seattle, Washington;1997. p. 1089–1107.

[26] Arasan S, Yener E, Hattatoglu F, Akbulut S, Hinislioglu S. The relationshipbetween the fractal dimension and mechanical properties of asphalt concrete.Int J Civ Struct Eng 2010;1(2):165–70.

[27] Geng M, Dou Y, Li X. Study of micro-structure of asphalt mixture based onfractal theory. Mechanic Automation and Control Engineering (MACE), In: 2ndInternational Conference 2011. p. 2582–2584.

[28] Song JM, Lui TS, Chang YL, Chen LH. Compositional effects on themicrostructure and vibration fracture properties of Sn–Zn–Bi Alloys. J AlloysCompd 2005;403:191–6.

[29] Mihashi H, Ahmed SFU, Mizukami T, Nishiwaki T. Quantification of crackformation using image analysis and its relationship with permeability. Int JRestoration Build Monuments 2006;12(4):335–48.

[30] Erdem S. Impact load-induced microstructural damage of concrete made withunconventional aggregates. PhD. Thesis, University of Nottingham,Nottingham; 2011.

[31] Sun H, Ding J. The comparison of fractal dimensions of cracks on reinforcedconcrete beam. Adv Mater Res 2011;291–294:1126–30.

[32] Karperien A. FracLac for ImageJ, version 2.5. <http://rsb.info.nih.gov/ij/plugins/fraclac/FLHelp/Introduction.htm, 1999–2012.

[33] Akcaoglu R, Tokyay M, Celik T. Assessing the ITZ microcracking via scanningelectron microscope and its effect on the failure behavior of concrete. CemConcr Res 2005;35(2):358–63.

[34] Glinicki MA, Litorowicz A. Crack system evaluation in concrete elements atmesoscale. Bull Polish Acad Sci, Tech Sci 2006;54(4):371–9.

[35] Zhang T, Nagy G. Surface tortuosity and its application to analysing cracks inconcrete. In: Proc. of the 17th International conference on pattern recognition,Cambridge, England, UK; 2004. p. 851–854.