2023 Volume 63 Issue 9 Pages 1487-1495
In recent years, three-dimensional (3D) measurements of actual coke particles have been conducted using a laser scanner in order to acquire knowledge about the gas/liquid permeability of blast furnaces. In order to ensure representativeness, a large number of coke particles need to be measured during actual operation. Therefore, we measured over 100 coke particles using a medical X-ray computed tomography (CT) scanner and obtained 3D shape information of each particle using image analysis. The validity of the proposed method was confirmed by comparing the analysis results with the actual measurement results from sieve separation. In this study, we mainly focused on the sphericity and flattening ratio as 3D shape indices. A 10 kg sample contains coke with a wide distribution of sizes and shapes, and the standard error in addition to the mean value should be considered when comparing samples with different production conditions. The results of the analysis targeting two samples with different manufacturing conditions showed that the sphericity was greatly affected by the impact of the transportation process and closely related surface breakage. Furthermore, the flattening ratio was greatly affected by the fissures formed during the carbonization process, which is closely related to the furnace temperature and volatile matter of the blended coal. This study shows that a medical X-ray CT scanner is a useful and practical tool for acquiring 3D shape of coke particles.
In ironmaking, coke not only serves as a reducing agent and heat source but also plays a role in ensuring gas/liquid permeability, and carburization in the blast furnace. Among these roles, achieving gas/liquid permeability is difficult for other reducing materials. In the case of coke, ensuring gas/liquid permeability requires it to be fed into the blast furnace separately from the sintered ore. In a blast furnace, a layered structure consisting of many coke particles is formed, and gas and molten iron pass through the voids in this packed bed. Therefore, the effect of the coke size and shape on the structure of the packed bed is a topic of great interest to ironmaking engineers.
Coke size has been the most debated topic with regards to the relationship between coke and packing structure. It has been reported that a large coke size results in a high void fraction of the packed bed1) and high gas permeability,2) so a large coke particle size is considered desirable; small coke particles are removed by sieving and large ones are fed into the blast furnace. Meanwhile, Chung et al.3) investigated the relationship between the mean coke-particle size and reducing agent ratio (RAR) of the blast furnace both analytically and experimentally and showed that a large coke size decreases the reaction rate and thermal efficiency and increases the RAR. While this result suggests that increasing the coke particle size may not improve the efficiency of blast-furnace operation, the size would still be recognized as important. Chung et al. also experimentally confirmed that coke with a large particle size has a high void fraction and good liquid permeability. Additionally, many researchers have employed the discrete element method (DEM), which assumes that coke particles are spheres that reproduce the packing structure of a blast furnace. The particle size was determined from realistic coke size distribution, and the results of such analyses have revealed the importance of coke size.4,5,6) Meanwhile, particle size is not the only shape feature that affects the coke packing structure. According to Ergun’s equation,7) which is used to calculate the pressure drop in blast furnaces, the shape factor along with the particle size contributes to the pressure drop, and it has been speculated that the coke shape affects the packing structure of the blast furnace. Ichida et al.1) conducted an experiment focusing on the relationship between sphericity and void fraction of coke layer, and their results showed that a higher coke sphericity led to a lower layer void fraction for particle sizes of 50 mm or less. Other researchers have been focusing on coke particle shape using DEM, such as by expressing coke particles by connecting several spheres.8) With the progress of measurement technologies in recent years, gas/liquid permeability in the coke-packed bed have been discussed based on the detailed measurement results of 3D coke shape. The group of Natsui et al. actively utilizes 3D polygonal data of coke particles obtained using a laser scanner. The packing structure resulting from dropping many coke particles was reproduced by calculating the movement, rotation, and contact of particles with realistic shapes using DEM. Additionally, particles whose shape was changed by the tumbler test were also simulated.9) In a further study, the flow of liquid9,10,11) and gas12,13) in the packing structure was also directly simulated by conducting computational fluid analysis.
Understanding the 3D shape of coke, as studied by Natsui et al., for example, is important for not only basic research but also actual blast-furnace operation, which could be conducted in a more stable manner by recording daily measurements and controlling the coke shape. However, in blast-furnace operations, it is difficult to utilize indices related to shapes other than the mean particle size of coke. The authors believe that a major reason for this is representativeness. Coke is produced in large quantities using large-scale furnaces (for example, the Nippon Steel Oita/Nagoya five-coke oven has a width of 0.45 m, height of 6.7 m, and depth of 16.6 m, with a coke production capacity of 1 million tons per year14)), and lumps with different properties, sizes, and shapes are formed in the width and height directions of oven. For example, in Japan, the drum index (DI), which is used as an index of coke strength, ensures representativeness by a 10 kg coke sample.15) Therefore, a method for quickly measuring a large number of coke particles is desirable in order to apply the 3D evaluation methods in steelworks. In blast furnace operations, the mean coke particle size is measured quickly in a large amount of particles using a sieve.
Therefore, our research group focuses on medical X-ray computed tomography (CT). This imaging method is mainly used for non-destructive observation of the internal structure of materials. Microfocus X-ray CT has been used to observe the pore structure of coke, to investigate the effect of pore structure on strength or reactivity of coke,16,17,18) and to observe the carbonization process.19) Microfocus X-ray CT provides a resolution of several microns. By contrast, medical CT is mainly used for observing the inside of the human body, with a relatively wide field of view and short measurement time. Therefore, many coke particles can be measured simultaneously and quickly. The internal structure could not be analyzed in detail because the resolution is lower than that of microfocus X-ray CT, but we decided that it was sufficient for determining the shape of coke particles. In this study, we used a medical X-ray CT scanner to image 10 kg of coke particles produced at mills. The captured coke images were analyzed in order to separate individual particles and evaluate the 3D shape features. Furthermore, the 3D coke-particle shape index and actual shape were compared, and the difference in the 3D shape characteristics of the two types of samples manufactured under different conditions was investigated.
We used 10 kg of coke produced at two different steel mills and reduced for drum testing, as specified by Japanese standards.15) The differences for each coke sample are shown in Table 1. In each mill, particles smaller than 25 mm were excluded by sieving while preparing a 10 kg sample. The major difference between the manufacturing conditions of the two samples was the amount of volatile matter in the blended coal: the calculated weighted average volatile matter of blended coal in samples 1 and 2 was 25.8% and 30.5%, respectively. Additionally, the distance from the coke dry-quenching equipment (CDQ) to the points where samples 1 and 2 were taken was ≈7 m and ≈20 m, respectively, suggesting that Sample 1 underwent less impact during transportation. Furthermore, according to the image analysis results, samples 1 and 2 comprised 122 particles and 169 particles, respectively.
| Sample 1 | Sample 2 | |
|---|---|---|
| Steel mill | Mill A | Mill B |
| Mass [kg] | 10 | 10 |
| Volatile matter of the blended coal [%] | 25.8 | 30.5 |
| Distance from the CDQ to the sampling point [m] | 7 | 20 |
| Number of coke particles (Image analysis) [-] | 122 | 169 |
Paper boxes (W240 × H225 × L320 mm) were manually packed with 3–4 kg of coke particles, and 10 kg of coke samples were divided into three batches and imaged using medical X-ray CT equipment (TSX-201 Aquilion LB, Toshiba Medical Systems Corp, Tokyo.). The imaging area (FOV) was ϕ256 mm (512 × 512 pixel; 1 pixel: 0.5 mm), the slice pitch was 0.5 mm, the tube voltage was 120 kV, the tube current was 400 mA, and the irradiation time was 0.5 s. The imaged coke particles were subjected to sieving for comparison with the medical CT data, and the particle-size distribution was measured in the laboratory. Sieve mesh sizes were 25, 38, 50, 75, 100, and 125 mm, respectively.
2.3. Image AnalysisIn this method, many coke particles were scanned simultaneously. Figure 1(a) shows a cross-sectional image of the coke particle packing layer scanned by the medical CT. As circled, there are points where the coke particles came into contact with each other. Image processing that involves simple binarization results in many particles becoming continuous and thus prevents the evaluation of the characteristics of individual particles. Therefore, the particles were divided using the following procedure. First, the original image was converted into a 256-gradation image with upper and lower density limits of 0 and 2 g/cm3. Next, a binarization process was performed in order to separate the solid portion and spatial portion based on the brightness value of the image. The threshold value for binarization was searched for by trial and error while checking the images and minimizing the contact between particles, while at the same time, ensuring that a single particle is not divided. Ultimately, a luminance value of 67 was set, which corresponds to an apparent density of 0.52 g/cm3. Next, erosion processing of eight-voxel for solid portion was applied. Adjacent particles were separated in this process. For the solid parts, a labeling process was conducted in order to uniquely identify the eroded particles. At this stage, each particle was separated, but the surface was scraped as a result of erosion processing. Therefore, the dilation processing was carried out until all the particle portions of the original image were covered; this was achieved using 40-voxel processing (Fig. 1(b)). The areas attributed by dilation contained both a single coke particle and space. Therefore, a logical AND of the portion discriminated as a particle by the binarization process and the areas attributed by dilation was taken. At this time, the threshold for dividing the solid particles and spaces was set to 10, which corresponds to an apparent density of 0.08 g/cm3 (Fig. 1(c)). In this analysis, two luminance values of 67 and 10 were used as thresholds in order to separate the solid part from the spatial part. The former threshold value was set so that the contact between the particles was small, which is advantageous for the separation of the particles, but the surface of the original particle was discriminated as a partial space. Therefore, in order to ultimately identify the shape of the coke particles, we used the latter threshold, which more reliably discriminates the position of the surface. The ImageJ20) (version 1.50 g) plug-in was used for image processing.
Image analysis procedure for multiple particles. (Online version in color.)
The parameters for evaluating the 3D shape characteristics of coke particles were calculated for each particle. In order to evaluate the 3D coke size, the three axes of coke particles were fitted to an ellipsoid (long diameter: d1; middle diameter: d2; and short diameter: d3). The “3D ImageJ Suite”,21) a plug-in for the ImageJ, was used for ellipsoid fitting. Figure 2(a) shows a schematic diagram of the coke particles fitted with an ellipsoid. Here, the calculated sieving diameter shown in Eq. (1) was defined and used as an index for evaluating the size of the coke particles, which is thought to be consistent with the particle size obtained by sieving in the experiment.
| (1) |
| (2) |
| (3) |
| (4) |
| (5) |
Indices related to 3D shape used in this analysis. (Online version in color.)
Figure 3 shows the results of the particle-size distribution calculated using image analysis and that was actually sieved, respectively. Table 2 shows the mean particle size. The two were compared in order to evaluate the validity of the image analysis method. Samples 1 and 2 had an actual sieved mean particle size of 56.0 and 48.8 mm, respectively. The larger coke particle size of Sample 1 is thought to be due to the low volatile matter of the blended coal, and the contraction ratio after re-solidification in the carbonization process was low, so the amount of fissures was low, as shown in previous studies by Nomura et al.23) and Jenkins et al.24) The image analysis results showed that the mean sieving diameters of samples 1 and 2 were 57.2 mm and 51.4 mm, respectively, which is the same order as the actual measurement results, and the analytical results reproduced the measured values well. As for the particle-size distribution, the analytical values were close to the measured values. These results confirmed the validity of the image analysis method. In further detail, the mean particle sizes in both samples 1 and 2 were large as a result of image analysis, and the weight of the smallest particle size category (25–38 mm) was small. This indicates that the sieving diameter calculated from this analysis is slightly larger than the actual mesh size, presumably because the actual sieving process was not strictly reproduced, such as by approximating an uneven particle as a perfect ellipsoid.
Size distribution of coke particles. (Online version in color.)
| Sample 1 | Sample 2 | |
|---|---|---|
| Sieving | 56.0 mm | 48.8 mm |
| Image analysis (present method) | 57.2 mm | 51.4 mm |
Figure 4 shows the relationship between sieving diameter and sphericity, as obtained using image analysis. It was confirmed that a larger sieving diameter resulted in a lower sphericity. This test employed actual coke, and after being discharged from the coke oven, it passed through a CDQ system and was transported by a conveyor belt and sampled. The large coke lumps might separate into several small lumps owing to the impact during transport. For example, a large, elongated lump may crack upon impact and separate into several approximately cubic lumps. This type of breakage is presumed to be due to volume breakage, as described by Arima.25) In other words, those that were cracked by the impact had small particle sizes and approached a shape similar to that of a sphere, whereas those that remained uncracked even with impact had a large particle size and a shape very different from that of a sphere, resulting in the relationship between sieving diameter and sphericity shown in Fig. 4. The test conducted by Ichida et al.1) showed that the shape of the coke particles approached a sphere as the particle size increased, which was inconsistent with the results of the present test, in which a larger particle size resulted in a lower sphericity. However, the experiment performed by Ichida et al. covered a wide range of target particle sizes and investigated the relationship, including the range of small particle sizes from 4 mm to 25 mm. Focusing on the large-particle-size range, Ichida et al. also reported that the average sphericity of coke particles in the particle size range of 25–35 mm was higher than that of the 35–50 mm and 50–75 mm ranges. Therefore, for the particle size range up to 25 mm or more, it is thought that the measurement results by Ichida et al. did not contradict the results of image analysis in the present study. Focusing on the different sphericities of samples 1 and 2, it is apparent that the sphericity of sample 1 tended to be low, but this is not clear from Fig. 4 alone. Therefore, Fig. 5 shows the mean sphericities with standard error values. It can be seen from Fig. 5 that the sphericity of Sample 1 is lower on average than that of Sample 2, and when factoring in the standard error, the sphericity of Sample 1 is quantitatively evaluated to be lower. As shown in Fig. 4, the coke size and shape vary widely, indicating a wide range of sphericities, so data that include the standard error values should be given, as shown in Fig. 5. The importance of ensuring representativeness by measuring many particles was represented by this method. The reason the sphericity of Sample 1 is lower than that of Sample 2 will be discussed later.
Relationship between sphericity and sieving diameter of individual particles measured using image analysis. (Online version in color.)
Mean sphericity for each sample. Error bars represent ±1σ of standard error. (Online version in color.)
Figure 6 shows the results of plotting the flattening ratios 1 and 2 of each particle in samples 1 and 2 in order to evaluate the shape of the coke particles. Both samples 1 and 2 show a wide range of flattening ratios, indicating that a 10 kg sample contains various forms of coke. Figure 7 shows the outlines of four representative samples in order to confirm the relationship between flattening ratio and shape. Figure 7(a) shows that both flattening ratios 1 and 2 are relatively low and that the shape is close to that of a cube. Figure 7(b) shows that the flattening ratio is high and flattening ratio 2 is low; this particle shape is a rod. Figure 7(c) shows that flattening ratio 1 is low and flattening ratio 2 is high, corresponding to a plate-like particle shape. Figure 7(d) shows that flattening ratios 1 and 2 are relatively high, and many coke particles that were partially chipped owing to fissures were observed. Figure 7(e) shows a plot of the overall flattening ratio of sample 1 and four representative samples. Focusing on the difference between samples 1 and 2 in Fig. 6, it can be seen that both flattening ratios 1 and 2 tended to be high in Sample 2, which contains a high content of blended coal volatile matter. Therefore, in order to evaluate more quantitatively, the mean values and standard errors of flattening ratios 1 and 2 are shown in Fig. 8. It can be inferred from Fig. 8 that the flattening ratios 1 and 2 of Sample 2 are high even after considering variation. The flattening ratios of Sample 2 are suspected to be high because many partially chipped coke particles exist, as shown in Fig. 7(d), and because Sample 2 had high volatile matter and contraction ratio values, resulting in the formation of many fissures, and coke particles with more complex shapes were more likely to form. Jenkins et al.24) investigated fissure formation in coke both experimentally and theoretically and showed that a high volatile matter value of coking coal resulted in narrower intervals between fissures and coke with a high aspect ratio. The aspect ratio described in the study by Jenkins et al. can be viewed as an index that corresponds to the flattening ratio in the present study. Therefore, the present analysis results are consistent with the findings of Jenkins et al., and it is inferred that Sample 2 had a high flattening ratio owing to the high volatile matter content of the blended coal.
Relationship between flattening ratios 1 and 2 of individual particles measured using image analysis. (Online version in color.)
Four representative coke samples with different shapes owing to different flattening ratios. (Online version in color.)
Mean values for flattening ratios 1 and 2 for each sample. Error bars represent ±1σ of standard error. (Online version in color.)
The sphericity and flattening ratio are discussed in depth here. Focusing on the difference in 3D shape between samples 1 and 2, it can be seen that Sample 1 had a low sphericity and flattening ratio. Comparing a perfect sphere and ellipsoid based on a geometric definition, the ellipsoid has a low sphericity and high flattening ratio. Therefore, Sample 2 is closer to a sphere in terms of sphericity, and Sample 1 is closer to a sphere in terms of flattening ratio. This result appears to be contradictory at first glance, but this is because although both the sphericity and flattening ratio of coke quantify the features of 3D complexity, they have different meanings. The experimental results by Natsui et al.9) can serve as a reference for a more detailed explanation. Natsui et al. conducted a test for measuring the coke particle surface using a laser scanner and showed that when coke particles were subjected to rotational impact using a tumbler tester, the surface breakage progressed, and the sphericity evaluated in two dimensions increased. The sphericity evaluated by Natsui et al. was 2D, but the same argument is thought to be applicable to three dimensions. The 3D sphericity is calculated from the volume and surface area, so if, for example, surface breakage occurs and the corners are removed, the surface area decreases, and the sphericity increases. Meanwhile, it can be intuitively predicted that the change in the flattening ratio is small when the corner is taken in this way. We focus again on the research by Jenkins et al.24) in order to better understand the flattening ratio. Jenkins et al. predicted the interval between fissures formed in the carbonization process using carbonization temperature and volatile matter as factors and estimated the aspect ratio of coke size. They then compared the analytical results and experimental results, showing that the theoretical model could explain the changes in the aspect ratio owing to changes in the carbonization temperature and volatile matter. This suggests that the fissures formed during the carbonization process of coke had a large effect on the aspect ratio of coke particles, whereas the effect of surface breakage, which was not included in the model by Jenkins et al., was relatively small. Summarizing the aforementioned results, it is thought that the 3D sphericity reflects that the coke is impacted and surface breakage occurs, and the shape is rounded and smooth. Meanwhile, the flattening ratio evaluates the rough shape, and it largely reflects the interval of fissures formed during the carbonization process. Sample 1 had a lower volatile matter formulation than Sample 2, therefore, it is thought that the fissures were small, the finished coke was large, and the flattening ratio was low. On the other hand, the sphericity of Sample 1 was low because the impact during transportation was small, and the surface was not scraped. Compared with that for Sample 2, the mill that actually manufactured Sample 1 had a shorter transportation distance until the drum sample was taken.
Figure 9 and Table 3 show the characteristic examples in which sphericity and flattening ratio are indices of different shapes. Figures 9(a)–9(c) show three particles picked from Sample 2. Figure 9(d) shows picked samples with the relationship between sieving diameter and sphericity. Figure 9(e) shows picked particles with the entire plot range of flattening ratio. Table 3 also shows the sieving diameter, sphericity, and flattening ratio of the particles in Figs. 9(a)–9(c). It can be seen from Fig. 9(d) that the sieving diameter and sphericity of the three particles are similar. Meanwhile, Fig. 9(e) shows that the three particles differ in their flattening ratios 1 and 2, and Figs. 9(a)–9(c) show that they actually differ greatly in shape. In other words, the sieving diameters and sphericities of these three particles are similar but the flattening ratio differs, and the shape is also vastly different. From the above results, it is difficult to determine the characteristics of the 3D shape of coke particles by measuring only the sphericity or flattening ratio; for example, it is considered to be appropriate to use multiple indices to investigate the state of voids in the packed structure of particles and gas/liquid permeability.
Three coke samples with similar sphericities and sieving diameters but different flattening ratios. (Online version in color.)
| Sieving diameter [mm] | Sphericity [-] | Flattening ratio 1 [-] | Flattening ratio 2 [-] | |
|---|---|---|---|---|
| Particle 1 | 52.9 | 0.629 | 0.483 | 0.412 |
| Particle 2 | 52.6 | 0.638 | 0.629 | 0.093 |
| Particle 3 | 49.5 | 0.646 | 0.152 | 0.538 |
Image processing was conducted on 3D images taken using a medical X-ray CT scanner, and the particle size and shape of over 100 coke particles contained in 10 kg for a drum test were measured. A comparison of the mean particle size and particle-size distribution between the actual sieving and image analysis results showed good agreement, confirming the validity of the proposed method. The sphericity and flattening ratio of the measured coke showed a wide distribution, and the standard error also needed to be considered for quantitative evaluation, so we were able to confirm the usefulness of the medical CT method, which enables rapid and wide-field measurement. The sphericity was higher with a smaller coke particle size, which was thought to be due to the cracking that occurred during the transportation process. A comparison of coke produced at two different locations showed that the sphericity was high when the transportation distance from the coke discharge to sampling was long and the impacts were large, suggesting that the flattening ratio became high when the volatile matter content was high. Both sphericity and flattening ratio are evaluation indices of the 3D coke shape. However, although the sphericity is thought to be largely affected by the surface breakage represented by the impact during transportation, the flattening ratio is thought to be largely affected by the fissures that form during the carbonization process. Additionally, in order to confirm examples in which the flattening ratio is significantly different even if the particle size and sphericity are similar, and to evaluate the 3D coke shape, it is desirable to determine not only a single shape index but also a composite index.
d1: Long diameter (mm)
d2: Middle diameter (mm)
d3: Short diameter (mm)
S: Surface area (mm2)
Seqv: Equivalent surface area (mm2)
V: Volume (mm3)
