Image Anal Stereol 2020;39:147-159 doi: 105566/ias.2318 Original Research Paper 147 QUANTIFICATION OF SEGREGATION IN PORTLAND CEMENT CON- CRETE BASED ON SPATIAL DISTRIBUTION OF AGGREGATE SIZE FRACTIONS MURAT OZEN, 1, MURAT GULER2 1Mersin University, Department of Civil Engineering, Mersin, Turkey; 2Middle East Technical University, De- partment of Civil Engineering, Ankara, Turkey e-mail: ozen.murat@mersin.edu.tr, gmurat@metu.edu.tr (Received December 27, 2019; revised July 16, 2020; accepted July 23, 2020) ABSTRACT Segregation is one of the quality standards that must be monitored during the fabrication and placement of Portland cement concrete. Segregation refers to separation of coarse aggregate from the cement paste, re- sulting in inhomogeneous mixture. This study introduces a digital imaging based technique to quantify the segregation of Portland cement concrete from 2D digital images of cut sections. In the previous studies, segregation was evaluated based on the existence of coarse aggregate fraction at different geometrical re- gions of a sample cross section without considering its distribution characteristics. However, it is shown that almost all particle fractions can form clusters and increase the degree of segregation, thus deteriorating the structural performance of concrete. In the proposed methodology, a segregation index is developed by based on the spatial distribution of different size fractions of coarse aggregate within a sample cross section. It is shown that degradation in mixture’s homogeneity is controlled by the combined effect of particle dis- tribution and their relative proportions in the mixture. Hence, a segregation index characterizing the mixture inhomogeneity is developed by considering not only spatial distribution of aggregate particles, but also their size fractions in the mixture. The proposed methodology can be successfully used as a quality control tool for monitoring the segregation level in hardened concrete samples. Keywords: Concrete, digital imaging, segregation, uniformity. INTRODUCTION The aggregate phase occupies up to 80% of concrete volume, leading to significant influence on both fresh and hardened properties of concrete. The size distribu- tion of aggregate particles is one of the key parameters of concrete design affecting the strength and workability properties of the mixtures. Apart from the initial design, maintaining the design size distribution during manufac- turing and placement is critical to achieve desired struc- tural performance of concrete. Segregation is one of the quality standards that must be monitored during the placement of concrete. It refers to separation of coarse aggregate from the cement paste, resulting in an inho- mogeneous mixture. Segregation can occur due to either settling or separation of coarse aggregate fraction in the mixture as a result of improper placing or vibration dur- ing the manufacturing process (Mindess et al., 2002). A highly segregated mixture will display decreased homo- geneity leading to greater variability in the strength properties because of inadequate internal structure (Na- varrete and Lopez, 2016). Because the reduced strength and durability are the consequences of segregation in Portland cement concrete, satisfying the quality require- ments of concrete by controlling segregation becomes critical for achieving adequate mechanical properties, hence higher structural performance (Ferraris et al., 2008; Mesbah et al., 2011). Although the conventional methods for measuring segregation consider only fresh properties of concrete, they don’t use standard tech- niques for monitoring the segregation properties in hard- ened concrete. Evaluation of segregation in the hardened phase is also important to confirm the test results for fresh concrete and can be used as a quality control tool when the test for fresh phase segregation is not per- formed (Navarrete and Lopez, 2016). Recently, digital imaging methods has found a great interest in studying micro-structural characteristics, par- ticle size distribution, size and shape characteristics and air void distribution of concrete (Yang et al., 2014; Nichols and Lange, 2006; Mora and Kwan, 2000; Ozen and Guler, 2014; Fernlund et al., 2007; Lee et al., 2007; He et al., 2016; Peterson et al., 2001). A main advantage of image processing relies on its ability to perform such analysis in a rapid, objective, efficient and cost effective OZEN M ET AL: Quantification of segregation in Portland Cement Concrete 148 manner with moderate level of hardware and software requirements. However, studies focusing on the analysis of segregation in hardened concrete are still limited and offer impractical methodologies. In a recent study by Solak et al. (2018), a segregation index was developed for lightweight concrete from the digital images of hor- izontal cut sections. Barbosa et al. (2011) proposed an index to evaluate distribution of aggregate particles in lightweight concrete. Fang and Labi (2007) developed an automated imaging methodology to identify the loca- tion of aggregate particles and mortar layer thickness for the study of segregation in self-consolidating concrete (SCC). In another study by Khayat et al. (2007), cylin- drical samples of SCC were cut vertically and each sec- tion was further horizontally divided into six sections and then the percentage of coarse aggregate areas from top to bottom sections were calculated to evaluate the segregation. Johnson et al. (2010) calculated the ratio of cement area between top and bottom sections from the digital images of horizontal cut sections to quantify seg- regation in SCC samples. Erdem (2014) used digital im- ages obtained from X-ray computed tomography to study segregation in SCC concrete samples by compu- ting the volume of coarse aggregate particles in the up- per and lower parts of the vertical and horizontal cut sec- tions. In these previous studies, segregation phenomena were considered by evaluating the percentage of coarse aggregate particles and cement paste at different sec- tions of concrete samples. However, the main problem in these methods is the lack of considering the spatial distribution of coarse aggregate fractions within the cross sections. Even if the coarse aggregates are homo- geneously distributed in the cement paste, larger or smaller coarse particles can form clusters within the sample cross section, thus affecting the structural perfor- mance of concrete. In this study, a digital imaging based technique is introduced to quantify the segregation of Portland cement concrete from 2D digital images of cut sections. In the proposed methodology, a segregation in- dex is developed by considering the spatial distribution of different size fractions of coarse aggregate within a sample cross section. It is believed that the proposed methodology can be successfully used as a quality con- trol tool for monitoring the segregation level in hardened concrete samples. MATERIAL AND METHODS MIXTURE PROPORTIONS This study was conducted using two different con- crete mix designs fabricated with natural crushed aggre- gates of Dmax=13 cm and 25 cm. The selected gradations can be found in Table 1. Standard 150 mm cubical con- crete samples with five replicates were prepared for each mix design combination. The purpose of using different maximum aggregate sizes was to evaluate the sensitivity of the proposed methodology. Table 1. Particle size distribution of aggregates used in the mix designs Sieve size (mm) Percent passing by weight Mix De- sign 1 (%) Mix De- sign 2 (%) 25.4 100.0 100.0 19.1 100.0 85.0 12.7 100.0 72.0 9.5 87.0 60.0 4.76 60.0 45.0 2.38 44.8 33.0 1.19 31.5 23.0 0.59 15.4 11.0 0.297 14.2 10.1 0.149 13.5 9.8 IMAGE ACQUISITION AND OPTIMUM THRESHOLD After the curing period, each cubical sample was cut into four equal pieces using a circular diamond saw. For each sample, this process produced a total of six cross sectional surfaces with three overlapping cut sections as presented in Fig. 1. However, only one of the overlapped cut surfaces was used to avoid duplicating the image analysis of the same cross section. After the cutting pro- cess, the cross sections were digitized using a desktop flatbed scanner at a resolution of 150 dpi in order to keep the file size moderate for further processing and analy- sis. Therefore, the pixels’ size in digitized images were 0.169 mm x 0.169 mm. The digital images were stored in gray scale format in which every pixel has a color depth ranging from 0 to 255, i.e., “0” indicates black and “255” as white. Analysis of cross sectional images starts with the implementation of gray scale segmentation. Threshold- ing is one of the commonly used image segmentation methods in which objects of interest are extracted from the background by selecting a threshold value (T), which varies between 0 and 255 for a gray scale image. Any pixel value lower than the threshold is classified as ob- ject pixel; otherwise, it belongs to the background of the image frame. In this study, an optimum threshold method previously developed by the authors was applied Image Anal Stereol 2020;39:147-159 149 to successfully detect aggregate particles within a cross sectional image (Ozen and Guler, 2014). Cut sections 150 mm 15 0 m m Fig. 1. Concrete sample cut sections The resultant binary images may need morphologi- cal processing due to the existence of noise speckles, particles touching the border of image frames and over- lapped particles. The number of erosion-dilation (open- ing) cycles with a 3x3 matrix of ones was applied to re- move these artifacts and separate overlapped particles in the binary images. The removal of the small particles does have any negative impact on the results, as the fo- cus of the study is the particles larger than 4.76 mm that cannot be removed during the erosion-dilation cycles. Image Processing Toolbox of the MATLAB® environ- ment was used to calculate the optimum threshold for segmentation and perform and the erosion-dilation op- erations. Fig. 2 illustrates the original cross sectional im- age on the left, and its binary phase on the right after the implementation of the erosion-dilation cycles. PARTICLE SIZE DISTRIBUTION ANALYSIS To evaluate the segregation of concrete samples from 2D cross sectional images, sieve sizes to be used for calculating the size fraction of aggregate particles need to be determined. Traditionally, the gradation of aggregate samples is determined by mechanically siev- ing the material through a series of sieves arranged in the order of decreasing opening sizes. The size fractions of aggregate samples retained on each sieve are then converted into percent passing by weight of total aggre- gate and reported as aggregate gradation. The gradation from 2D digital images are, however, calculated based on percent passing by the number of particles rather than weight of aggregate that are smaller than a specific sieve size. Furthermore, the determination of gradation from digital images requires identification of the following parameters: a) appropriate sieve size definition (square or diagonal), b) optimal particle shape parameter to compare with sieve sizes, so that a gradation obtained from digital images will be as close as possible to the actual gradation obtained from the mechanical sieving (Ozen and Guler, 2014). Fig. 2. Gray scale cross sectional image and binary im- age after erosion-dilation cycles In the literature, there are a number of shape param- eters calculated from 2D digital images to evaluate sev- eral aspects of aggregates particles. The authors have previously shown that maximum Feret diameter is the most suitable shape parameter producing a particle size distribution close to that of mechanical sieving when computed based on diagonal sieve opening for the con- crete mixtures generated by the natural and crushed ag- gregates. These two aggregate sources are generally used for concrete production in the market and cover all possible variations in aggregate shapes for analysis. Nat- ural aggregates are mostly in rounded in shape, whereas crushed aggregates become more angular after produc- tion (Ozen and Guler, 2014). As illustrated in Fig. 3, the maximum Feret diameter represents the length of the major axis of an aggregate particle that connects two points on its boundary with the farthest distance between them. Image Processing Toolbox of the MATLAB® en- vironment was used to label individual aggregate parti- cles and calculate the maximum Feret diameter of each OZEN M ET AL: Quantification of segregation in Portland Cement Concrete 150 particle. Since crushed aggregates were used in this study, the sieve size of each aggregate particle was de- termined using the maximum Feret diameter in conjunc- tion with the diagonal sieve openings, as seen in Fig. 4. The authors previously showed that the proposed meth- odology can accurately estimate the sample gradation with only minor errors for small size aggregates. Based on this finding, it can be presumed that the gradation computed from several 2D cross sections of a sample can well approximate its 3D or actual gradation. Maximum Feret diameter Particle boundary Fig. 3. Maximum Feret diameter of particle Sieve size Fig. 4. Plan view of sieve opening with a particle cross section SEGREGATION ANALYSIS As described in the above section, segregation re- fers to the separation of coarse aggregates from the mor- tar in fresh concrete, yielding an inhomogeneous mix- ture (Mindess et al., 2002). Accordingly, coarse aggre- gate particles are expected to be uniformly distributed within the non-segregated mixtures. Based on the previ- ous studies, it was decided that the distribution of aggre- gate particles retaining on 4.76 mm sieve will be suffi- cient to evaluate the degree of segregation in concrete mixtures (Navarrete and Lopez, 2016; Khayat et al., 2007; Solak et al.; 2018, Barbosa et al., 2011). The pro- posed methodology herein is implemented in three steps as illustrated in Fig. 5. The developed algorithm in MATLAB® environment is able to complete all the com- putational steps in less than minute for the analysis of a sample. In the first step, the optimum threshold is calcu- lated to perform the gray scale thresholding. Then, the number of erosion-dilation cycles is applied to remove speckles and separate overlapping particles in the thresholded images. In the second step, aggregate parti- cles are labelled, the maximum Feret diameter of each aggregate particle is calculated and compared with the diagonal sieve openings to determine the percent pass- ing of aggregate particles. Next, fractional digital im- ages are generated displaying only the aggregate parti- cles of a specific sieve size equal to and larger than 4.76 mm. In the final step, segregation index of the specimen (SI) is calculated using Eq. (1-5). 𝐹𝐹𝐹𝐹𝐼𝐼𝑖𝑖𝑖𝑖 = 𝑚𝑚𝑚𝑚𝑥𝑥�𝑅𝑅𝐼𝐼𝑖𝑖𝑖𝑖 ,𝐶𝐶𝐼𝐼𝑖𝑖𝑖𝑖� (1) where FSI = fractional segregation index calculated from the maximum of row segregation index (RI) and column segregation index (CI). i = cross section number from 1 to 3; j = fraction number from 1 to n; n = number of sieve sizes equal to and larger than 4.76 mm. For in- stance, n = 2 for mix design 1 because only 4.76 mm and 9.5 mm sizes are used in the analysis. To calculate RI and CI from fractional images, first the number of particle pixels in each fractional image are calculated. If all the aggregate particles are spatially dis- tributed at equal distances within the cross section, it should be expected that a single size fraction must also be equally spaced; as a result of this, the pixels belong- ing to aggregates are located at spatially equal distances in the rows and columns of the fractional image. The methodology used to detect the spatial distances is shown in Fig. 6. The straight line (OA) represents the pixel distributions expected for a homogeneous mixture in which aggregates are distributed at nearly equal dis- tances in a fractional image, and can be computed by cu- mulatively summing the number of pixels in each row and column by taking the upper left corner as the origin of the image. When a certain level of inhomogeneity exists in the cross section, the calculated sum of the pix- els will start to deviate from the straight line and form a curve such as (OMA) in Fig. 6. For an actual concrete cross section, the calculated number of pixels in the row- wise and column-wise directions will be similar to ones as shown in Fig 7 and 8. The amount of deviation from the straight line depends on the level of inhomogeneity of the cross section, and the calculated curve can cross the straight line depending on the characteristic of the spatial distribution of aggregate particles. Image Anal Stereol 2020;39:147-159 151 Calculate optimum threshold (T) Diagonal sieve openings Binary image Cross section (i) Compute maximum feret diameter for each aggregate Generate fractional images (1 to n) Compute Cij and Rij Compute FSIij=max(Cij,Rij) Compute cross sectional segregation indexes (CSI1,CSI2,CSI3) Compute specimen segregation index (SI) Sieve size of each aggregate particle Eliminate particles passing from 4.76 mm Compute Pi1 to Pin Gray Scale Image Erosion-Dilation Fig. 5. Proposed algorithm to determine segregation of concrete samples To calculate the segregation indices in the row-wise and column-wise directions, the percent ratio is simply calculated between the gray region underneath the straight line (OA) and the shaded (triangle) region, which represents the maximum deviation for a segre- gated mixture, using the following relations; 𝑅𝑅𝐼𝐼𝑖𝑖𝑖𝑖 = 100𝐴𝐴𝑟𝑟/(𝑁𝑁𝑟𝑟𝑁𝑁𝑓𝑓𝑟𝑟/2) (2) 𝐶𝐶𝐼𝐼𝑖𝑖𝑖𝑖 = 100𝐴𝐴𝑐𝑐/(𝑁𝑁𝑐𝑐𝑁𝑁𝑓𝑓𝑟𝑟/2) (3) where Ar and Ac = area between actual and theoreti- cal curves (see Fig. 7 and 8); Nfr = total number of pixels within aggregate particles of fractional image; and Nr, Nc = number of rows and columns in the fractional im- age, respectively. In Eq. (2) and (3), the denominators represent hypo- thetically the maximum areas that can occur between ac- tual and theoretical curves when maximum segregation level is reached in the mixture. Theoretically, RI and CI values range between 0 and 100. Ideally, if the particles are equally spaced by their spatial coordinates, the curves for the theoretical and actual pixel distributions will coincide, leading RI and CI values equal to zero as Ar and Ac are approaching to zero. On the other hand, as the distribution of aggregate particles deviate from the idealized case, the actual pixel distribution graphs will start shifting from the straight (theoretical) line; as a re- sult, RI and CI values will increase. After determining the fractional segregation indices, the following equa- tions are used to calculate the cross sectional segregation indices (CSIs) and the overall segregation index of the specimen (SI). 𝐶𝐶𝐹𝐹𝐼𝐼𝑖𝑖 = �𝑃𝑃𝑖𝑖𝑖𝑖 𝑛𝑛 𝑖𝑖=1 𝐹𝐹𝐹𝐹𝐼𝐼𝑖𝑖𝑖𝑖 (4) 𝐹𝐹𝐼𝐼 = max(𝐶𝐶𝐹𝐹𝐼𝐼1,𝐶𝐶𝐹𝐹𝐼𝐼2,𝐶𝐶𝐹𝐹𝐼𝐼3) (5) where Pi1 to Pin are the percent normalized retaining for each size fraction calculated from the digital image analysis and account for the relative proportions of sieve sizes in the mixture and calculated based on percent pix- els of particles retaining on each sieve size (i.e. Pi1+Pi2+ ... +Pin = 100%). For instance, for the first cross sectional image of Fig. 9, total number of particle pixels retaining on 4.76 mm and 9.5 mm sieves is 192,502 (see Fig. 10); therefore, P11 = 23% and P12 = 77% (see Table 2). OZEN M ET AL: Quantification of segregation in Portland Cement Concrete 152 Nr or Nc Straight line (OA) for idealized (homogeneous) cross section A O Curve (OMA) for segregated cross section M Nfr Area: Ar or Ac Area of triangle: Nr Nfr/2 Fig. 6. Method to calculate segregation index from 2D cross section image 0 10000 20000 30000 40000 50000 60000 0 100 200 300 400 500 600 700 800 900 Theoretical Actaual Nr Nfr Rows Observed Theoretical Pi xe ls Ar Fig. 7. Theoretical and actual distributions of particle pixels in fractional image along row-wise direction 0 10000 20000 30000 40000 50000 60000 0 100 200 300 400 500 600 700 800 900 Theoritical Actual Nc Nfr Columns Observed Theoretical Pi xe ls Ac Fig. 8. Theoretical and actual distributions of particle pixels in fractional image along column-wise direction Image Anal Stereol 2020;39:147-159 153 RESULTS AND DISCUSSION To carry out segregation analyses, first the optimum threshold algorithm was implemented to convert gray scale cross sectional images into binary images. The op- timum threshold was calculated separately for the three cross sections of the test samples. The next step was to calculate the maximum Feret diameters of the aggre- gates particles from the binary images produced after thresholding. The maximum Feret diameters were then compared with the diagonal sieve openings, so that the size distribution of all size fractions can be determined. In order to perform such analysis, the maximum Feret diameters calculated in pixels were converted into actual dimensions, i.e., unit of millimeters, using the known resolutions of the images and then compared with the diagonal sieve openings. Fractional images were then generated displaying only aggregates retaining on each sieve size equal to and larger than 4.76 mm. Fig. 9 shows binary cross sectional images of a specimen produced for mix design 1. For brevity, the fractional images and pixel distribution curves will be presented for one of the cross sectional images. Fig. 10 illustrates the fractional images of the first cross section in Fig. 9 together with the theoretical and actual distri- bution of particle pixels through the rows and columns of images. As it can be seen, the actual pixel distribu- tions of particles retaining on 4.76 mm sieve almost co- incide with the theoretical distributions. However, for 9.5 mm sieve fraction, the actual distribution deviated from the straight line because of the fact that the parti- cles are not equally distributed in the fractional image as can be seen from Fig. 10, instead they seem to concen- trate on the upper and lower parts of the cross section. Cross Section I Cross Section II Cross Section III Fig. 9. Thresholded images for specimen of mix design 1 9.5 mm 4.76 mm 0 43,659 0 859 Theoretical Row Column 0 148,843 0 859 Theoretical Row Column Nc=Nr= Nc=Nr=859 Nfr= 43,659 Nfr= 148,843 Theor ti l Theoreti l l Fig.10. Fractional images of the first cut section for specimen of mix design 1 OZEN M ET AL: Quantification of segregation in Portland Cement Concrete 154 In Table 2, the calculated values of RI, CI, FSI, CSI and SI are listed for the three cross sectional images of this specimen shown in Fig. 9. In general, particles re- taining on 4.76 mm sieve results in the lowest FSI val- ues. On the other hand, the highest FSI values were ob- tained for 9.5 mm sieve size. After multiplying by the normalized percent retaining values (P1i), while the low- est cross sectional segregation index (CSI) was obtained for the first cut surface, the highest CSI was obtained for the third cut surface. As a result, the segregation index (SI) of this specimen was found to be 11.4%, which is the maximum of the computed CSI values. Table 2. FSI, CSI and SI values for specimen of mix design 1 Sieve (mm) Cross Section I Cross Section II Cross Section III RI CI FSI P1j (%) RI CI FSI P2j (%) RI CI FSI P3j (%) 9.5 9.9 14.8 14.8 23 12.4 9.3 12.4 29 8.4 19.1 19.1 43 4.76 4.7 4.5 4.7 77 5.8 4.3 5.8 71 5.6 4.2 5.6 57 CSI (%) 7.0 7.7 11.4 SI (%) 11.4 Fig. 11 shows binary cross sectional images of a specimen prepared for mix design 2. Fig. 12 illustrates the fractional images as well as theoretical and actual distributions of particle pixels for the second cross sec- tional image in Fig. 11. The calculated segregation in- dexes FSI, CSI and SI for these cross sections can be found in Table 3. A slight deviation can be observed for 4.76 mm in row-wise direction and for 12.7 mm in col- umn-wise direction. From the fractional images shown in Fig. 12, it can be seen that the homogeneity level in row-wise direction is slightly better than in column-wise direction for 4.76 mm; conversely it is slightly better in column-wise direction than in row-wise direction for 12.7 mm. On the other hand, the distribution curves con- siderably deviated from the theoretical lines in row-wise direction for 9.5 mm, 12.7 mm. and 19.1 mm. For the first and the third cut sections in Fig 11, particles retain- ing on 4.76 mm sieve results in the lowest FSI values (see Table 3). For the second cut section, particles re- taining on 12.7 mm has the lowest FSI value while the highest values were obtained for 19.1 mm sieve size. It can be noticed that even though particles retaining on 19.1 mm resulted in the highest FSI values for all the cut sections analyzed, these particles do not still dominate the overall segregation index as they constitute the smallest proportion in the mixture; thereby resulting in the lowest Pij values. For the same test sample, the larg- est CSI value was obtained from the first cross section, hence the sample segregation index (SI) is reported as 16.3%. Cross Section I Cross Section II Cross Section III Fig. 11. Thresholded images for specimen of mix design 2 Image Anal Stereol 2020;39:147-159 155 0 72,687 0 875 Theoretical Row Column 4.76 mm 12.7 mm 19.1 mm 0 38,757 0 875 Theoretical Row Column 9.5 mm 0 50,679 0 875 Theoretical Row Column 0 110,342 0 875 Theoretical Row Column Nc=Nr=875 Nc=Nr=875 Nc=Nr=875 Nc=Nr=875 Nfr= 38,757 Nfr= 72,687 Nfr= 50,769 Nfr= 110,342 Theoretical l Theoretical l Theoretical l Theoretical l Fig. 12. Fractional images of the second cut section for specimen of mix design 2 Table 3. FSI, CSI and SI values for specimen of mix design 2 Sieve (mm) Cross Section I Cross Section II Cross Section III RI CI FSI P1j (%) RI CI FSI P2j (%) RI CI FSI P3j (%) 19.1 20.7 34.0 34.0 13 36.1 20.1 36.1 14 30.4 32.3 32.3 14 12.7 14.1 12.4 14.1 30 12.0 6.4 12.0 26 9.8 26.8 26.8 21 9.5 8.9 26.4 26.4 17 14.4 8.7 14.4 19 10.8 9.0 10.8 23 4.76 7.2 7.8 7.8 40 6.7 12.8 12.8 41 8.1 4.5 8.1 42 CSI (%) 16.3 16.2 16.0 SI (%) 16.3 OZEN M ET AL: Quantification of segregation in Portland Cement Concrete 156 To verify the sensitivity of the proposed method, ar- tificial images were generated by algorithmically redis- tributing aggregate particles in the cross section images of mix design 1 and 2. During this process, overlapping or connected particles were avoided. As many as 5 arti- ficial sections were generated and plotted together with their original cross sections as shown in Fig. 13 and 14 by changing the spatial location of aggregate fractions used in the calculation of segregation index. For each mix design, the first sections (I1) were so produced that the section’s homogeneity level is better than their orig- inal ones, and the other sections were simply produced by progressively dislocating each size fraction to in- crease the segregation level. Generated artificial images for cross section III of mix design 1 are shown in Fig. 13, and results of the corresponding CSI values are given in Table 4. The cal- culated percent fractions (P3j) for 4.76 mm and 9.5 mm sieves were found to be 57% and 43%, respectively (see Table 2). It can be observed that section I2 produces a slightly higher CSI index of 9.9% due to agglomeration of 9.5 mm particles at nearly center of the section. The section I3, on the other hand, shows even higher segre- gation with a CSI of 20.4% by the distribution of the coarse fraction on the boundary of the cross section. When the segregation indexes of I2 and I3 are carefully compared, it can be noticed that particles retaining on 4.76 mm sieve generates better homogeneity in I2 as compared to both 4.76 mm and 9.5 mm sizes in I3, hence resulting in a significantly higher CSI value than I2. When only the 9.5 mm fraction is separated from the rest of the particles, CSI value was found to be 27.0%. For the last image, however, the largest CSI value of 28.6% was obtained when 4.76 mm is separated from the other fractions, and the largest and the smallest fractions are accumulated at the bottom of the section. The analysis of the generated artificial images indicate that the level of segregations identified by visual inspection agrees well with the calculated cross sectional index values. I1 I2 I5I3 Original I4 Fig. 13. Imaginary images generated for cross section 3 of mix design 1 Table 4. Segregation index for the generated images in Fig. 13 Sieve (mm) P3j (%) FSI (%) Original I1 I2 I3 I4 I5 9.5 43 19.1 10.0 17.4 22.7 44.9 45.9 4.76 57 5.6 4.7 4.3 18.6 13.5 15.5 CSI (%) 11.4 7.0 9.9 20.4 27.0 28.6 Image Anal Stereol 2020;39:147-159 157 A similar strategy was also used to generate the ar- tificial images for cross section II of mix design 2 as shown in Fig. 14. PI values of the size fractions were ranged from 14% to 41%, as can be seen in Table 5. A careful inspection of the original cross section reveals the grouping of coarse size fraction nearly at the left side of the section, resulting in a higher segregation level. It can be observed that section I1 with a CSI of 15.8% has slightly more homogenous aggregate distribution as compared to the original section with a CSI of 16.2%. Section I2 resulted in a higher segregation level than section I1 with a CSI of 17.9%. When the largest size fraction (19.1 mm) is agglomerated within the center of the cross section in I3, the calculated CSI value becomes significantly higher than the CSI of the original cross section. Segregation index of the cross section was fur- ther increased in I4 by locating some largest particles at the top, one size smaller to the bottom and the interme- diate sizes in the middle of the cross section. For this section, the calculated segregation level was found to be 29.7%. The highest segregation of 35.3% was achieved by locating the larger fractions to the bottom of the cross section in section I5. I1 I2 I3 I4 I5 Original Fig. 14. Imaginary images generated for cross section 2 of mix design 2 Table 5. Segregation index for generated images shown in Fig. 14 Sieve (mm) P2j (%) FSI (%) Original I1 I2 I3 I4 I5 19.1 14 36.1 39.7 35.6 37.6 57.9 55.2 12.7 26 12.0 12.3 11.1 35.0 35.8 46.2 9.5 19 14.4 25.4 31.1 16.6 35.0 17.0 4.76 41 12.8 5.3 10.0 19.4 13.7 30.0 CSI (%) 16.2 15.8 17.9 25.5 29.7 35.3 OZEN M ET AL: Quantification of segregation in Portland Cement Concrete 158 It should be noticed that the computed CSI value for I5 of Fig. 13 (28.6%) is lower than for I5 of Fig. 14 (35.3%). Even though these two sections are not di- rectly comparable, due to having different aggregate fractions, the larger CSI value for I5 of Fig. 14 can be justified by carefully inspecting the segregation levels introduced by the 12.7 mm and 19.1 mm aggregate frac- tions. While these two factions do not exist in the mix- ture of Fig. 13., a FSI of 55.2% from 19.1 mm and 46.2% from 12.7 mm are added to the segregation level of the mixture by their relative percentages in Fig. 14. Besides, there is also an appreciable amount of fractional segre- gation that can be observed from the 4.76 mm fraction in Fig. 14 with a FSI of 30.0%. The analysis of the example mix designs shows that a qualitative rating of mixture segregation level is possi- ble. The calculated segregation levels for the artificial images of mix design 1 range from 7.0% to 28.6%, while 11.4% was found for the original design cross section (see Table 4). The aggregate distributions in these sec- tions indicate that sections I1 and I2 can be among the possibly generated segregation levels in the field with CSI values of 7.0% and 9.9%, respectively. On the other hand, sections I3 through I5 seem unlikely to be ob- served in the field, unless special effort is given to obtain such a high segregation level in the mixture. In Table 2, the CSI values for the three cut sections of mix design 1 ranged between 7.0% and 11.4% with a relatively ho- mogenous distribution for the first cut section. It should be noticed that the first cut section (CSI = 7.0%) has dis- tribution characteristics nearly equal to the artificially generated cross section I1 based on the comparison of their CSI values. However, a close inspection of the third section reveals that the mix design 1 was, in fact, pro- duced with a moderately segregated mixture, resulting in an overall segregation level of (SI) of 11.4%. For the mix design 2, the artificial images were gen- erated for the second cut section, for which the calcu- lated CSI values ranged between 16.0% and 16.3% as given in Table 3. The lowest CSI value for the artificially generated mixture was found to be 15.8% for section I1, as seen in Table 5. However, the SI value for the second cut section of this mixture is slightly higher (16.2%) than for I1 and very close to I2 indicating that the second cut section distribution characteristics somewhere between I1 and I2. However, the segregation levels for the re- maining sections I3, I4 and I5 seem significantly more heterogeneous than the original cross (cut) section, hence unlikely to appear in the field. The benefit of the proposed method relies on its simplicity to quantitatively determine the mixture segre- gation level based on 2D cut sections that are easily ob- tained using a document scanner. The method can also be applied for cylindrical core samples of concrete taken in the field. In this case, because only one vertical cut section can be taken for image analysis due to size limi- tations, the results for both gradation and segregation analyses can be statistically improved by increasing the number of core samples; at least three samples to be equivalent to one cubic sample used in this study. Using the described procedure, the determination of mixture segregation index can be easily performed for a hard- ened concrete sample to evaluate its segregation without any personal judgement or subjective rating based on visual inspection methods. In this way, the calculated SI index can be used as a quality control criterion for hard- ened concrete after the placement process. As indicated earlier, excessive segregation in the mixture can dramat- ically reduce the strength of hardened concrete by poor load transfer between mortar and the aggregate skeleton at micro-structure level. Since segregation is character- ized at micro-structural level in the proposed method, i.e., spatial distribution of size fractions, further investi- gations can be conducted to establish a possible correla- tion between segregation level and strength properties of hardened concrete. CONCLUSION The main objective of this paper is to evaluate seg- regation of Portland cement concrete samples using two- dimensional cross sectional images. Unlike the previous works on mixture segregation, this study considered the spatial distribution of coarse aggregate fractions within the concrete cross sections. However, by only looking at the distribution of coarse aggregate may not always help obtain the actual segregation level of mixture. Even if the coarse aggregates are equally distributed in the ce- ment paste, the small size fractions can still form clusters and result in segregation at relatively small scale. Based on the analyses of segregation levels for the test mix- tures, the following conclusions can be drawn: - Degree of segregation in hardened Portland ce- ment concrete can be successfully identified from two-dimensional digital images of cut sec- tions. - Segregation level depends highly on the per- centage of large size aggregates in the mixture. In general, the potential of generating segre- gated mixtures is more likely for coarser mix- tures than for fine graded mixtures. Image Anal Stereol 2020;39:147-159 159 - Degradation in mixture’s homogeneity is con- trolled by the combined effect of particle distri- bution and their relative proportions in the mix- ture. Hence, any measure of segregation used must account for not only spatial distribution of aggregate particles, but also their size fractions in the mixture. - Proposed segregation index is determined with- out any use of personal judgment or subjective rating procedure; therefore, it can be used as a quality control parameter for hardened concrete samples to evaluate the degree of homogeneity achieved after production. - The outcome of the study can be extended fur- ther to investigate a possible correlation be- tween the developed index and the strength properties of concrete. REFERENCES Barbosa FS, Beaucour AL, Farage MC, Ortola S (2011). Image processing applied to the analysis of segregation in lightweight aggregate concretes. Constr Build Mater 25(8):3375-81. Erdem S (2014). X-ray computed tomography and fractal analysis for the evaluation of segregation resistance, strength response and accelerated corrosion behavior of self-compacting lightweight concrete. Constr Build Mater 61:10-7. Fang C, Labi S (2007). Image-processing technology to evaluate static segregation resistance of hardened self- consolidating concrete. Transp Res Record 2020:1-9. Fernlund JM, Zimmerman RW, Kragic D (2007). Influ- ence of volume/mass on grain-size curves and conver- sion of image-analysis size to sieve size. Eng Geol 90(3-4):124-37. Ferraris CF, Koehler E, Amziane S (2008). Report on measurements of workability and rheology of fresh concrete. ACI Committee 238, Report 238.1R-08. He, H, Courard, L, Pirard, E, Michel, F (2016). Shape analysis of fine aggregates used for concrete. Image Anal Stereol 35(3): 159-66. Johnson D, Johnson G, Robertson IN (2010). Quantifying segregation in self-consolidating concrete through im- age analysis. Department of Civil and Environmental Engineering, University of Hawaii, Research Report UHM/CEE/10-14. Khayat KH, Vanhove Y, Pavate TV, Jolicoeur C (2007). Multi-electrode conductivity method to evaluate static stability of flowable and self-consolidating concrete. ACI Mater J 104(4):424-33. Lee JR, Smith ML, Smith LN (2007). A new approach to the three-dimensional quantification of angularity us- ing image analysis of the size and form of coarse ag- gregates. Eng Geol 91(2-4):254-64. Mesbah HA, Yahia A, Khayat KH (2011). Electrical con- ductivity method to assess static stability of self-con- solidating concrete. Cement Concrete Res 41(5):451- 58. Mindess S, Young JF, Darwin D (1981). Concrete. Pren- tice-Hall, NJ. Mora CF, Kwan AK (2000). Sphericity, shape factor, and convexity measurement of coarse aggregate for con- crete using digital image processing. Cement Concrete Res 30(3):351-58. Navarrete I, Lopez M (2016). Estimating the segregation of concrete based on mixture design and vibratory en- ergy. Constr Build Mater 122:384-90. Nichols AB, Lange DA (2006). 3D surface image analysis for fracture modeling of cement-based materials. Ce- ment Concrete Res 36(6):1098-107. Ozen M, Guler M (2014). Assessment of optimum thresh- old and particle shape parameter for the image analysis of aggregate size distribution of concrete sections. Opt Laser Eng 53:122-32. Peterson K, Swartz R, Sutter L, Van Dam T (2001). Hard- ened concrete air void analysis with a flatbed scanner. Transp Res Record 1775:36-43. Solak AM, Tenza-Abril AJ, Baeza Brotons F (2018). Im- age analysis applications for the study of segregation in lightweight concretes. Int J Comp Meth and Exp Meas 6(4):835-46. Yang XW, Man HG, Tian RL (2010). Deformation meas- urement of concrete by white light digital image anal- ysis in frequency domain. Appl Mech Mater 29:496- 01.