Study on Void Structure Reconstruction of Asphalt Mixture by X-Ray Computed Tomography and Otsu’s Method

)e purpose of this paper is to obtain the pore distribution of asphalt mixture accurately by nondestructive technology. Specimens prepared with four gradations of asphalt mixtures were scanned using X-ray computed tomography (CT) which was used to measure air void sizes at different depths within specimens. )e air void distributions of obtained CT images were analyzed using ring blocking segmentation combining Otsu’s method, which provided an accurate estimate of air voids in asphalt mixtures. )e image processing results showed that air void distribution was not uniform in the specimens; higher air void concentrations were found at the top and bottom of the specimen, and lower, in the rest of the sample depth. )e air void sizes of SUP13 and AC13 are mainly distributed between 0.15 to 0.2mm, while PA13 and SMA13 are 0.4 to 0.65mm and 0.4 to 0.7mm, respectively. It is believed that the CT pictures processed by the ring blocking segmentation combining Otsu’s method is feasible and rational to capture the air voids size and content of asphalt mixtures.+


Introduction
Performances of asphalt mixtures are dependent on the volume of aggregate, asphalt binder, as well as air voids. Specifically, the stiffness of the mixture is highly related to the air void content of the mixture. It is generally accepted that air voids play an important role in determining the resistance of asphalt mixtures to major pavement distresses including rutting, fatigue cracking, and low temperature cracking [1]. e research shows that stiffness and compressive strength increase with decreasing air void content, when the binder content is kept constant [2]. Air void content also influences the durability of asphalt mixes in terms of aging and stripping. For example, low air void content minimizes the aging of the asphalt binder films within the aggregate mass, the possibility of water penetrating into the mix, and stripping the asphalt binder off the aggregates [3]. It is also found that low air void content can lead to asphalt bleeding [4], while high air voids may result in early aging processes of the mixture, which make it prone to moisture damage, cracking, and pavement deterioration [5].
Air void content is a common parameter used to characterize the void structure of compacted asphalt mixtures [6]. It should be noted that the level of homogeneity of the distribution of air voids inside the asphalt course has significant impacts on the response of the material [7][8][9][10][11]. It is necessary to further study the characterization of air voids such as void's size including radius and area of air void, distribution, and content. However, it is difficult to obtain the air void characterization by using traditional techniques. erefore, the computer imaging technology and nondestructive techniques which can solve the problem described above and have the ability to characterize internal structure of materials [12][13][14] were applied in this study. First of all, CT technology is used to scan asphalt mixture specimens. e microstructure characteristics of asphalt mixture, such as voids and coarse aggregate, are studied using Otsu image processing method, MATLAB software, and image processing software, and then the virtual asphalt mixture specimen is reconstructed.

Objectives
e objective of this study is to investigate the air void distribution (i.e., air void content and air void size) using X-ray CT, Otsu image processing method, and MATLAB software. Image analysis software and computer algorithms were used to analyze the X-ray CT images and quantify the air void content, size distribution, and connectivity.

Test Materials and Sample Preparation
Specimens with 150 mm in diameter and 165 mm in height were fabricated using the superpave gyratory compactor (SGC). In order to obtain high definition CT picture, the specimens were further cored into 50 mm in diameter and 75 mm in height for test purposes. Four types of asphalt mixtures known as AC13, SMA13, PA13, and SUP13 were studied in details. e aggregate gradations for the four mixtures are shown in Table 1.

X-Ray CT Scanning and Image
Analysis Techniques 4.1. Image Scanning Analysis Study. e specimens were nondestructively scanned using an X-ray CT machine to generate a map representing the density at every point in the microstructure with the resolution of approximately 0.083 mm/pixel. Horizontal two-dimensional image slices were captured relative to the specimen's height from bottom to top at every 0.1 mm interval. 750 CT images were obtained using X-ray CT scanning for each specimen. e 8-bit grayscale images have a range of pixel intensities from 0 (pure black) to 255 (pure white). e geometry of the cylindrical specimen and an example of a captured X-ray CT image are shown in Figure 1. Two imaging software packages, MATLAB, and Image-Pro Plus, were used to process and analyze the CT images.

Otsu's Method.
In image segmentation, threshold segmentation is common and effective method especially when the objective and background grayscale are distinctively different. Otsu [16] presented the maximum class square error method (that is, Otsu's method) which was simple, adaptive, and widely used in threshold segmentation. Otsu's method was used to characterize the microstructure of asphalt mixtures [17].
In Otsu's algorithm, supposing the image has L gray levels (the gray value ranges from [1, 2 · · · L]) and considers the pixel of which gray value is i, so the total pixel numbers n i are obtained by Assuming the entire image is divided by gray value t into regions A (background) and region B (destination), the gray level of region A ranges from 1 to t, the gray level of region B ranges from t + 1 to L − 1, the class probabilities of A and B are estimated as equations (2) and (3), respectively.
us, the gray level of A and B can be calculated using the following equations: and the total gray value of image is expressed as e class variance expression between regions A and B is calculated as follows:

Advances in Materials Science and Engineering
According to equation (6), the largest t value is the optimum value needed for complete image segmentation. e optimized threshold t * is given by In a recursion algorithm, the class probabilities and class means are given in equations (8)(10), respectively.

Image Processing and Analysis Using Ring Blocking
Method and Otsu's Method. Image threshold processing plays an important role in image segmentation. According to the characterization of the grayscale which is increasing from the center to the edge of CT pictures for the same phase (there are three phases in asphalt mixtures, that is, aggregate, air void, and asphalt mastic), the CT pictures are segmented using ring block (shown in Figure 2) combining Otsu's method to characterize the microstructure of asphalt mixtures [18]. e CT pictures are segmented with 5 rings as shown in Figure 2.
Image processing is explained in following steps. Firstly, the image analysis process includes the image format, noise reduction, image enhancement, and a hole filling algorithm to recognize the hole region. Secondly, the preprocessed image is segmented using the ring blocking method. And there are 50% overlaps among the adjacent rings. irdly, the ring segmentation regions are processed using Otsu's method to calculate the threshold of air voids and the background. Finally, the ring segmentation regions are combined to obtain the whole segmentation image (Figure 3). e aggregate, air void, and asphalt mastic of each segmentation image are captured using Otsu's method as shown in Figure 3(c).
In order to illustrate the validity and accuracy of ring blocking segmentation combining Otsu's method, Bernsen's method [19] and traditional Otsu's method are also used to process the CT pictures. Otsu's method is the global threshold technique, and Bernsen's method is the local threshold technique. Figure 4 shows comparisons among the  Advances in Materials Science and Engineering original CT picture processed by Bernsen's method, conventional Otsu's method, and ring segmentation using Otsu's method, respectively. e variation in gray intensity of the CT image depends on the density of each material in the composite. Brighter regions correspond to denser objects, such as aggregates, and dark regions correspond to lower density objects, for instance, air voids. Figure 4(b) shows that Bernsen's method is hard to distinguish the fine aggregate and mastic inside the specimen. e conventional Otsu's method shown in Figure 4(c) can be used to capture fine aggregate structures. However, it cannot distinguish the denser structure of the fine aggregate and mastic. erefore, it can be seen that the ring segmentation combining Otsu's method is the most effective, compared with other two methods to process the CT picture.

Air Void Distribution.
e Image-Pro ® Plus software and MATLAB software are used to visualize the air void structure of the asphalt mixture samples. e air void structure includes the radii, areas, and air void content. e percent air voids for an image (%AV j ) and the total percent air voids of the sample (%AV) are computed as follows: where A vj is the area of air voids in image j, A j is the crosssectional area of image j, and N is the total number of images. e air void radii in image j, also called air void size in this study, are computed as follows: where M j is the number of air voids in each image. Several threshold values are used as input to the macro until the total percent air voids of the sample (equation (12)) matches the laboratory measurements. Based on the aforementioned information, the air void content and air void size in terms of air void radius and area can be calculated. Figure 5 shows that the air void distributes along with the height of the specimen based on the CT picture. Each CT picture's air voids content can be calculated according to the modified Otsu's method. Figure 6 shows that the air void size plotted against the void number in SUP13, AC13, SMA13, and PA13 specimens. e air void size is represented by the air void radius which can be obtained from the air void area. e air void radius is used to evaluate the magnitude of air voids. e air void sizes of SUP13 and AC13 mainly are distributed from 0.15 to 0.2 mm in radius, following by 0.1 to 0.15 mm and 0.2 to 0.25 mm. In terms of air void size greater than 0.3 mm, the number is 9 and 2 for SUP13 and AC13, respectively. For PA13 specimen, the air void sizes mainly are distributed between 0.4    Advances in Materials Science and Engineering Advances in Materials Science and Engineering in field sample air voids is roughly two times higher near the surface of the asphalt course than on the middle and bottom of the layer [21,22]. Figures 7-10 indicate that the air void content is depend on the air void areas which relate to the air void size (that is, air void radius). e graphs in Figures 7 and 8 are distinctively different than those in Figures 9 and 10. e reason for this is that the used large stones and SBS-modified binder in Figure 9 and high-viscosity binder in Figure 10. Both of these produce a more uniform distribution of air voids vertically, even though there is more variability in the air void content than is the case with Figures 7 and 8.

Air Void Distribution Reconstruction.
In order to construct the three-dimension distribution of air voids, the commercially available software, MATLAB program, is used to reconstruct the processed CT pictures. Remove the particles with the particle size less than 1.18 mm, and then reconstruct the four grading specimen model, as shown in Figure 11. e three-dimension air voids microstructure of asphalt mixtures are shown in Figures 12-15 for AC13, SUP13, PA13, and SMA13, respectively. e two-dimension CT images processed by ring blocking and Otsu's method are collected and converted into three-dimension image of the original sample. e images are thresholded to identify mixture constituents. resholding is the technique applied to characterize the components of the mixture. It converts the representative gray scale of a component into a value.
is value is the only value associated with an element, and so it can be identified in the mixture. e gray value is assigned to air voids. e reconstruction of air voids of asphalt mixtures can be simply carried out as in the following steps. Firstly, the horizontal slices of 1.0 mm thickness with an overlap of 0.2 mm were captured. Secondly, these slices were stacked together in the computer to form the actual three-dimension porous microstructure. irdly, the three-dimension structure of air voids is rotated to project in transverse section and vertical section.
It appears in Figures 12 and 13 that the distributions of the air void content of SUP13 and AC13 with respect to depth coincide with those in Figures 7 and 8; a higher air void content was observed at the sample top and bottom, while lower and more uniform air void content were observed in the middle section. Figures 14 and 15 show different trends compared with SUP13 and AC13, and in general, the air void content is higher at the top of the sample and lower in the rest of the sample depth. And the air void's distribution seen from Figures 14 and 15 are in line with Figures 9 and 10, respectively. It indicates that the CT pictures processed by the ring blocking segmentation and Otsu's method are feasible and rational to capture the air voids size and content of asphalt mixtures.

Conclusions and Discussion
e X-ray CT was used to scan the specimens of AC13, SUP13, PA13, and SMA13. e CT pictures were processed by ring blocking segmentation and calculated by Otsu's method. Based on preprocesses, segmentation, and identification, the air voids of asphalt mixtures were calculated and reconstructed with Image-Pro ® Plus software and MATLAB program. e key findings are summarized as follows: (1) e air voids of AC13, SUP13, PA13, and SMA13 specimens can be distinct to capture using the ring blocking segmentation combining Otsu's method to process the CT pictures. (2) According to the analysis of method processed, it is believed that the ring blocking segmentation combining Otsu's method is feasible and rational to identify the air voids of asphalt mixtures in comparison with Bernsen's method and the conventional Otsu's method. (4) Image analysis reveals that air void's distribution is not uniform through the specimens, namely, higher air voids concentrations were found at the specimen's top and bottom and lower air voids in between. (5) It is a significant progress of asphalt mixtures' research to conduct three-dimensional reconstruction of the air voids. e air voids of SUP13, AC13, SMA13, and PA13 were reconstructed and displayed. e three-dimensional air voids distribution of AC13, SUP13, PA13, and SMA13 are in accordance with the analysis of air void size distribution along with the height ratio of specimen. (6) e used large stones and SBS-modified binder in SMA13 and high-viscosity binder in PA13 produce a more uniform distribution of air voids vertically, even though there is more variability in the air void content than is the case with AC13 and SUP13.

Data Availability
e Fish code data and Excel data of experiment used to support the findings of this study were supplied by the National Natural Science Foundation of China and China Postdoctoral Science Foundation under license and so cannot be made freely available. Requests for access to these data should be made to Jiantong Zhang, tongtong01@ 126.com.

Conflicts of Interest
e authors declare that they have no conflicts of interest.