Analysis of Powder Binder Separation through Multiscale Computed Tomography

In this study, X-ray computed tomography was used to analyze powder binder separation in TC4 green bodies. Firstly, for the scanned results of the whole green body, because of the relative low resolution (36 μm), the powder binder separation can only be analyzed by using gray value distribution. Then, local regions (areas near the gate and the central parts) were scanned by using a much higher resolution (2.3 μm). Both of the volume fraction of powder content and gray value distributions indicate that powder particles tend to accumulate in the central parts. Finally, based on the results tested by using submicron resolution (0.8 μm), the effects of the volume and morphology of the powder particles on the powder binder separation were analyzed.


Introduction
Powder injection molding (PIM) has an excellent ability to manufacture products with large quantities, low cost, complicated shape and excellent performance [1,2].Thus, these products have been used to many industrial and research fields such as for automobile and aerospace parts.Four crucial steps are involved in PIM: feedstock made by mixing binder and powder particles, mold filling, binder removing from green body, and sintering of the powder particles [3].The powder-binder separation phenomenon emerging in the mold filling is the major cause of nonhomogeneous composition distribution, which leads to the collapse and sagging of green body during debinding, as well as the geometrical error and non-uniform properties of the final sintered parts.
During mold filling, because of the physical property differences and pseudo-plastic behavior, the complex physical environments may lead to relative movement between the powder particles and binder.In the previous studies [4][5][6][7][8], much research was performed to analyze the powder-binder separation, and the characteristics of these methods were summarized in Table 1.It can be seen that an effective method for analyzing powder-binder separation of the green body nondestructively is urgently needed.
Recently, X-ray computed tomography has been widely applied to material science and engineering [9][10][11] and some researchers began to use this technology to analyze the powder-binder separation [12,13], because of its obvious advantages, including 3D and the non-destructive characterization of the spatial structure.In the previous studies, powder-binder separation was analyzed by using gray value distribution in the CT reconstructed data [14,15].However, due to the spatial resolution limit (40 µm), those results are indirect evidence to characterize this phenomenon, and the specific distribution of powder particles cannot be analyzed.In this study, a CT scan was performed by using a much higher spatial resolution to study powder-binder separation and further verifying the accuracy of previous studies.
Titanium and its alloy are famous due to their relative low density, high specific strength and corrosion resistance [16].Because of these advantages, this material has attracted attention for many years.However, due to high processing costs, the application of titanium alloy is limited.To overcome the drawbacks of the conventional processing techniques, PIM are increasingly used to produce titanium alloy parts [17,18].Thus, Ti6Al4V (TC4) powder was selected as the subject for this study.
The aim of this present study is a detailed quantification analysis of the powder-binder separation characteristics in the TC4 green bodies by using multiscale CT tests.

Experimental Procedure
The powder used in this study has a particle size distribution of D 1 0 = 15 µm, D 50 = 29.1 µm and D 90 = 49.2 µm.SEM observation image and chemical information are shown in Figure 1 and Table 2, respectively.For the binder system shown in Table 3, this study used the same components and corresponding proportion as the research performed by Liu [19].The feedstock was prepared by mixing TC4 powder (AMC Powders, Beijing, China) and binder, and the solid loading used in this study was 62 vol%.The mixture was injected into the mold cavity by using an injection pressure of 90 MPa, an injection temperature of 160 • C, and a mold temperature of 45     It is known that a CT test with a higher spatial resolution obtains a smaller tested area and vice-versa.To solve this problem, multiscale tests can be performed on the same sample by using a different resolution.The advantages of this method are that high resolution tested results can give clear and accurate regional information, because the structure is constituted by more voxels.For example, if a structure's volume is about 512 µm 3 , for the 4 µm and 1 µm resolution tested results, the corresponding amount of voxels are 8 and 512, respectively.These data can help to speculate the structural characteristics in the whole sample tested by using a low resolution precisely.In this study, multiscale CT tests were performed on nanoVoxel-3000 equipment (Sanying Precision, Tianjin, China) which has two X-ray imaging paths.Firstly, a whole green body can be scanned by using the large view system then, for the interested areas, optical coupled with CCD system was used to perform the high resolution tests (2.3 µm and 0.8 µm).The scanning method and parameters is shown in Figure 2A and Table 4, respectively.Besides high spatial resolution, the density distinguishing ability of this system is 0.1%, thus, the subtle change of the powder contents can be distinguished precisely.Seen from Figure 2B, the regions near the gates and central part were recorded as Region 1, 2 and 3, respectively.The CT data was analyzed by Avizo 8.1 software (Thermo Fisher Scientific, Merignac, France).It is noted that every data in this paper represents a mean value of five samples.Flat panel Optical coupled with CCD-4×lens Optical coupled with CCD-10×lens For the X-ray CT tests, gray value of each voxel represents the corresponding matter's absorbing ability of X-ray.In Yang's study, the experiment results verified that matter's density is proportional to the gray value within certain range [14].Moreover, in this study, to verify the accuracy of the image segmentation (OSTU method), porosities of 5 porous sample prepared by different processing were obtained by Archimedes' principle and CT data, respectively.Seen from Table 5, the results tested by the two methods are similar.Thus, the accuracy of analyzing powder-binder separation by using  For the X-ray CT tests, gray value of each voxel represents the corresponding matter's absorbing ability of X-ray.In Yang's study, the experiment results verified that matter's density is proportional to the gray value within certain range [14].Moreover, in this study, to verify the accuracy of the image segmentation (OSTU method), porosities of 5 porous sample prepared by different processing were obtained by Archimedes' principle and CT data, respectively.Seen from Table 5, the results tested by the two methods are similar.Thus, the accuracy of analyzing powder-binder separation by using gray value distribution and image segmentation can be guaranteed.

Results and Discussion
Firstly, powder-binder separation in the whole green body can be analyzed by using the results tested by 36 µm resolution.Figure 3A shows the reconstructed 3D rendering.Under this tested condition, it can be seen that powder particles and binder cannot be clearly distinguished, thus, the quantitative calculation of the feature structure by using threshold segmentation method cannot be used.Due to the resolution limitation, the previous study used gray value distribution to analyze powder-binder separation in the whole green body [14,15].In this study, the average gray value of each slice image along Z direction was obtained.As shown by the curve in Figure 3B, the gray value increases from region 1 to region 2, then decreases from region 2 to region 3.Because the gray value is proportional to the matter density, it can be seen that powder particles and binder accumulate in the central parts and the areas near the gate, respectively, and the powder-binder separation mainly occurs in these three regions.However, the powder-binder separation analysis by using gray value distribution is only based on the physical inferences and this method lacks direct observation results to prove its accuracy.Next, the results tested by a much higher resolution are analyzed to solve this problem.However, the powder-binder separation analysis by using gray value distribution is only based on the physical inferences and this method lacks direct observation results to prove its accuracy.Next, the results tested by a much higher resolution are analyzed to solve this problem.
Figure 4A shows the reconstructed slice images tested by 2.3 µm resolution and this scanned area is in the middle of the green body.It can be seen that powder particles and binder can be clearly distinguished, thus volume fraction of the corresponding structure can be calculated by the way of OSTU segmentation method.The extracted powder particles are shown in Figure 4B and the calculated volume is 600 3 voxels.Seen from Figure 4B, along Z direction, the average volume fraction and gray value of each slice were calculated and the results are shown in Figure 4C,D, respectively.It can be seen that the distributional characteristics of the curves are similar in the two graphs.The accuracy of the analysis method by using gray value distribution was verified by the direct observation results.Seen from Figure 4C,D, the volume fraction of powder particle and gray value of each slice in Region 2 are both higher than those of Region 1 and 3.This means that powder content tends to accumulate in the central part.In the previous studies, this phenomenon was explained by the velocity difference between the powder particles and binder during injection process.In the gate areas, shear stress for the feedstock can be very high due to the maximum shear stress usually occurs near the gate.Because the feedstock exhibits pseudo-plastic behavior, viscosity of the feedstock decreases near the gate, which may induce the powder-binder separation.Moreover, note that the ability to maintain the original velocity is related to the density of material, therefore, the velocity of powder particles was faster than that of the binder near the gates.The binder with slower velocity was surely in arrears of the powder particles, which led to more powder particles appearing in the central parts [14].Thus, for the three regions, the different physical effects such as velocity and shearing rate lead to the different powder-binder separation phenomenon.
Seen from Figure 4C the variation range of the powder content is about 62 ± 3 vol%.According to the sintered quality of these samples, ±3% is a reasonable variation range.
To detect the more subtle phenomenon of the powder-binder separation during mold filling, the Seen from Figure 4C,D, the volume fraction of powder particle and gray value of each slice in Region 2 are both higher than those of Region 1 and 3.This means that powder content tends to accumulate in the central part.In the previous studies, this phenomenon was explained by the velocity difference between the powder particles and binder during injection process.In the gate areas, shear stress for the feedstock can be very high due to the maximum shear stress usually occurs near the gate.Because the feedstock exhibits pseudo-plastic behavior, viscosity of the feedstock decreases near the gate, which may induce the powder-binder separation.Moreover, note that the ability to maintain the original velocity is related to the density of material, therefore, the velocity of powder particles was faster than that of the binder near the gates.The binder with slower velocity was surely in arrears of the powder particles, which led to more powder particles appearing in the central parts [14].Thus, for the three regions, the different physical effects such as velocity and shearing rate lead to the different powder-binder separation phenomenon.
Seen from Figure 4C the variation range of the powder content is about 62 ± 3 vol%.According to the sintered quality of these samples, ±3% is a reasonable variation range.
To detect the more subtle phenomenon of the powder-binder separation during mold filling, the tested results of 0.8 µm resolution were analyzed.Seen from Figure 5A,B, there are many pores in the powder particles and the feature sizes can be obtained by using the measurement tool in visual analysis software.It can be seen that CT test with high resolution can also be regarded as a powerful method to evaluate the quality of powder particles.
accumulate in the central part.In the previous studies, this phenomenon was explained by the 130 velocity difference between the powder particles and binder during injection process.In the gate 131 areas, shear stress for the feedstock can be very high due to the maximum shear stress usually occurs 132 near the gate.Because the feedstock exhibits pseudo-plastic behavior, viscosity of the feedstock 133 decreases near the gate, which may induce the powder-binder separation.Moreover, note that the 134 ability to maintain the original velocity is related to the density of material, therefore, the velocity of 135 powder particles was faster than that of the binder near the gates.The binder with slower velocity 136 was surely in arrears of the powder particles, which led to more powder particles appearing in the 137 central parts [14].Thus, for the three regions, the different physical effects such as velocity and With the resolution increasing, the micro-structure details become much more clear and the accurate morphology analysis can be performed.By the way of OSTU segmentation method, Figure 5C shows the stack state of powder particles and the size of the calculated volume is also 600 3 voxels.
Firstly, the volume size of each powder particle in the three regions was calculated and the statistical results are shown in Table 6.It can be seen that a larger number of powder particles appears in the central part and the average volumes of the three regions are similar.These data can explain that why the powder content is higher in Region 2. By comparing the standard variance of the three regions in Table 6, the size of the powder particles in the central part is revealed to be more uneven.The volume distributions of powder particles in the three regions are shown in Figure 5D.The numbers of powder particle with volume larger than 5000 μm 3 are similar and there are more relatively small ones in Region 2.
Based on the computational principle of 3D morphology, the sphericity degree of the powder particles were obtained and Table 6 shows the statistic results.Since this parameter of the standard sphericity is obviously 1, it can be seen that much more spherical particles appeared in the central part than in the gate area.With the resolution increasing, the micro-structure details become much more clear and the accurate morphology analysis can be performed.By the way of OSTU segmentation method, Figure 5C shows the stack state of powder particles and the size of the calculated volume is also 600 3 voxels.Firstly, the volume size of each powder particle in the three regions was calculated and the statistical results are shown in Table 6.It can be seen that a larger number of powder particles appears in the central part and the average volumes of the three regions are similar.These data can explain that why the powder content is higher in Region 2. By comparing the standard variance of the three regions in Table 6, the size of the powder particles in the central part is revealed to be more uneven.The volume distributions of powder particles in the three regions are shown in Figure 5D.The numbers of powder particle with volume larger than 5000 µm 3 are similar and there are more relatively small ones in Region 2.

Figure 2 .
Figure 2. A, Sketch map of multiscale CT tests.B, Side view of the green body, the yellow and red circle shows the corresponding interested areas scanned by high resolution.

Figure 2 .
Figure 2. (A) Sketch map of multiscale CT tests.(B) Side view of the green body, the yellow and red circle shows the corresponding interested areas scanned by high resolution.

Figure 3 .
Figure 3. A, 3D rendering of the whole green body.B, Gray value distribution along the corresponding direction.

Figure
Figure 4A shows the reconstructed slice images tested by 2.3 µ m resolution and this scanned area is in the middle of the green body.It can be seen that powder particles and binder can be clearly distinguished, thus volume fraction of the corresponding structure can be calculated by the way of OSTU segmentation method.The extracted powder particles are shown in Figure 4B and the calculated volume is 600 3 voxels.Seen from Figure 4B, along Z direction, the average volume fraction and gray value of each slice were calculated and the results are shown in Figure 4C,D, respectively.

Figure 3 .
Figure 3. (A) 3D rendering of the whole green body.(B) Gray value distribution along the corresponding direction.

Figure 3 .
Figure 3. A, 3D rendering of the whole green body.B, Gray value distribution along the

117
It can be seen that the distributional characteristics of the curves are similar in the two graphs.The 118 accuracy of the analysis method by using gray value distribution was verified by the direct 119 observation results.120 Metals 2018, 8, x FOR PEER REVIEW 6 of 9

Figure 4 .
Figure 4. A, Reconstructed slice image tested by 2.3 µ m resolution.B, The extracted powder particles in the calculated volume.C, Volume fraction distribution of the powder particles along Z direction.D, Gray value distribution of the slice images along Z direction.Every data point in the two figures is the mean value of 5 samples, and the relative differences calculated by each sample's data and the corresponding mean value were less than 10.6% (reasonable value for X-ray tests) and 4.3%, respectively.

Figure 4 .
Figure 4. (A) Reconstructed slice image tested by 2.3 µm resolution.(B) The extracted powder particles in the calculated volume.(C) Volume fraction distribution of the powder particles along Z direction.(D) Gray value distribution of the slice images along Z direction.Every data point in the two figures is the mean value of 5 samples, and the relative differences calculated by each sample's data and the corresponding mean value were less than 10.6% (reasonable value for X-ray tests) and 4.3%, respectively.

138 9 Figure 5 .
Figure 5. A, Reconstructed slice image tested by 0.8 µ m resolution.B, Defect measurement of the powder particles.C, Packing state of the powder particles.D, Statistic of the powder particle volume distribution in the three local regions.Every data point in the figure is the mean value of 5 samples, and the relative differences calculated by each sample's data and the corresponding mean value were less than 4.7%.

Figure 5 .
Figure 5. (A) Reconstructed slice image tested by 0.8 µm resolution.(B) Defect measurement of the powder particles.(C) Packing state of the powder particles.(D) Statistic of the powder particle volume distribution in the three local regions.Every data point in the figure is the mean value of 5 samples, and the relative differences calculated by each sample's data and the corresponding mean value were less than 4.7%.

Table 1 .
Analysis methods of powder-binder separation in the previous studies.

Table 2 .
Chemical elements of the TC4 powder.

Table 2 .
Chemical elements of the TC4 powder.

Table 3 .
Constituents of the binder.

Table 4 .
Scanning parameters of multiscale CT tests.

Table 4 .
Scanning parameters of multiscale CT tests.

Table 5 .
Porosity tested by Archimedes' principle and CT data.