Powder property prediction system, powder property prediction method, and powder property prediction program

Abstract

To provide technique enabling powder physical properties to be more easily evaluated.SOLUTION: An image of particles constituting powder is analyzed to acquire the particle size distribution of the particles, a parameter related to a shape of the particles, and a parameter (pixel variance value, surface feature amount) related to a surface state of the particles. The relationship between these parameters and the actually measured physical property values of the powder is learned by deep learning to create an AI prediction model. The parameter obtained from the image of the particle with unknown physical property value is input to the AI prediction model, and the physical property value of the particle is predicted.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a technique for predicting the physical properties of powder. [Background technology]

[0002] Various powders are used as materials for pharmaceuticals, medicines, food, ceramic products, metal products, cosmetics, electronic components, toner, raw materials for 3D printers, and various other products. It is important to evaluate the physical properties of powders, such as adhesion, flowability, and compressibility. Techniques and devices for evaluating the physical properties of powders are described, for example, in Non-Patent Document 1. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Journal of the Society of Powder Technology, Vol. 54, No. 2, 2017, pp. 90-96 Summary of the Invention [Problem to be solved by the invention]

[0004] While the technology described in Non-Patent Document 1 is effective, it requires dedicated measuring equipment and specialized skills and expertise for handling samples and measuring instruments. Furthermore, current technology requires powder samples in the 100 g range, but preparing samples in the gram range can be difficult for pharmaceuticals and other products. In light of this, the present invention aims to provide a technology that can more easily evaluate the physical properties of powders. [Means for solving the problem]

[0005] The present invention includes an image data acquisition unit that acquires image data of an image obtained by photographing particles that constitute a powder, an image analysis unit that calculates the particle size distribution of the powder, shape parameters that characterize the shapes of the particles, and surface parameters that indicate the uneven state of the particle surface structure of the particles, based on the image data, and a powder physical property value prediction unit that predicts the physical property values ​​of the powder based on the particle size distribution of the powder, the shape parameters, and the surface parameters. A plurality of shape types relating to particle structures and correction coefficients corresponding to the plurality of shape types are prepared in advance, and the predicted physical property values ​​are corrected using the correction coefficients corresponding to the shape types classified based on the image data. It is a powder property prediction system. The present invention is a powder property prediction system comprising: an image data acquisition unit that acquires image data of an image obtained by photographing particles that constitute a powder; an image analysis unit that calculates the particle size distribution of the powder, shape parameters that characterize the shape of the particles, and surface parameters that indicate the unevenness of the particle surface structure of the particles based on the image data; and a powder property value prediction unit that predicts the physical property values ​​of the powder based on the particle size distribution of the powder, the shape parameters, and the surface parameters, wherein the physical property value prediction unit has a physical property value prediction model that predicts the physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder, and a plurality of physical property value prediction models are prepared corresponding to each of a plurality of shape types that indicate the particle structure, and the physical property values ​​are predicted using one of the plurality of physical property value prediction models corresponding to one of the plurality of shape types.

[0006] In the present invention, the shape parameters include convexity, circularity, aspect ratio, and linearity, where the convexity is the ratio of the perimeter of a convex polygon that covers a particle to the perimeter of the particle itself, the circularity is a parameter that indicates how close the particle shape is to a circle, the aspect ratio is the ratio of the major axis to the minor axis when the particle shape is approximated as an ellipse, and the linearity is a parameter that indicates the degree of linearity of the particle.

[0007] In the present invention, the surface parameters include a pixel variance value and a surface feature amount, the pixel variance value is a parameter for evaluating the magnitude of the variance of pixel values, and the surface feature amount is a parameter for evaluating the degree of the characteristics of the particle surface.

[0008] In the present invention, the surface feature amount is calculated based on a change in pixel shading in a specific direction, the surface feature amount is determined from a plurality of surface feature amount candidates, the plurality of surface feature amount candidates being surface feature amounts in a plurality of different directions as the specific direction, and the largest of the plurality of surface feature amount candidates is adopted.

[0009] In the present invention, the physical property values ​​may include two or more selected from the powder dynamic friction angle, stress transmission rate, stress relaxation rate, compressibility, bulk density, internal friction angle, shear adhesion, and flow factor.

[0012] In the present invention, the shape types are classified based on the characteristics of the three-dimensional structure of the particles. In the present invention, the shape types include a thin rod shape whose longitudinal length is 20 times or more the diameter of the cross section approximated by a circle (Type 1), a rectangular parallelepiped shape whose ratio of the minimum side length to the maximum side length is less than 20 (Type 2), a thin plate shape whose thickness is 10 times or more the width in the shortest direction (Type 3), an aggregate of particles whose diameter is 1 / 30 or less of the diameter of the aggregate (Type 4), an aggregate of particles whose diameter is more than 1 / 30 of the diameter of the aggregate (Type 5), a Type 2 rectangular parallelepiped whose surface is bonded to a smaller rectangular parallelepiped (Type 6), a Type 2 rectangular parallelepiped whose surface is bonded to a smaller round three-dimensional object (Type 7), and a hexagonal thin plate shape that satisfies the conditions of Type 3.(Type 8), a mass consisting of many Type 2 shapes (Type 9), a thin, irregular polygonal shape with pentagons or more that meets the conditions of Type 3 (Type 10), a flat disk shape that meets the conditions of Type 3 (Type 11), a shape consisting of cubes whose diameter approximates a sphere and meets the conditions of Type 5 (Type 12), a long, thin polygonal shape with sharp corners (like a piece of broken wood) (Type 13), a flat oval shape that meets the conditions of Type 3 (Type 14), a gourd-shaped shape (Type 15), a shape consisting of a collection of thin, irregular particles that meets the conditions of Type 3 (Type 16), a shape consisting of a collection of thin, irregular particles with a hole in the center (Type 17), a shape consisting of many overlapping thin, irregular particles that meets the conditions of Type 3 (Type 18), a collection of thin, irregular particles with many sharp corners that meets the conditions of Type 3 The particle shape may be one or more selected from the following: particles with a rounded shape (Type 19), particles with many sharp corners (Type 20), hollow pipe-shaped particles (Type 21), cylindrical particles with a larger diameter than Type 1 (Type 22), particles in the shape of a triangular prism whose height is shorter than the length of the sides that make up the triangle (Type 23), particles with eight or more sides (Type 24), particles in the shape of a disk (donut shape) with a hole in the center (Type 25), spherical particles with multiple holes on their surface (Type 26), particles in which many particles with diameters smaller than the sphere are embedded on their surface (Type 27), particles in which many particles with diameters less than 1 / 50 of the sphere's diameter are attached to their surface (Type 28), and particles in which many particles with diameters 1 / 10 to 1 / 50 of the sphere's diameter are attached to their surface (Type 29).

[0013] The present invention obtains image data of an image obtained by photographing particles constituting a powder, calculates the particle size distribution of the powder based on the image data, calculates shape parameters that characterize the shapes of the particles, and calculates surface parameters that indicate the unevenness of the particle surface structure of the particles, and predicts the physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder. A plurality of shape types relating to particle structures and correction coefficients corresponding to the plurality of shape types are prepared in advance, and the predicted physical property values ​​are corrected using the correction coefficients corresponding to the shape types classified based on the image data. It can also be understood as a method for predicting powder properties. The present invention acquires image data of an image obtained by photographing particles constituting a powder, and calculates, based on the image data, the particle size distribution of the powder, shape parameters characterizing the shape of the particles, and surface parameters indicating the unevenness of the particle surface structure of the particles. The method predicts physical property values ​​of the powder based on the particle size distribution, shape parameters, and surface parameters of the powder, and the prediction of the physical property values ​​utilizes a physical property prediction model that predicts the physical property values ​​of the powder based on the particle size distribution, shape parameters, and surface parameters of the powder. A plurality of physical property prediction models are prepared corresponding to each of a plurality of shape types indicating the particle structure, and the method can also be understood as a powder property prediction method in which the physical property values ​​are predicted using one of the plurality of physical property prediction models corresponding to one of the plurality of shape types.

[0014] The present invention is a program readable and executed by a computer, which causes the computer to acquire image data of an image obtained by photographing particles constituting a powder, calculate the particle size distribution of the powder based on the image data, calculate shape parameters that characterize the shapes of the particles, and calculate surface parameters that indicate the unevenness of the particle surface structure of the particles, and predict the physical property values ​​of the powder based on the particle size distribution of the powder, the shape parameters, and the surface parameters. A plurality of shape types relating to particle structures and correction coefficients corresponding to the plurality of shape types are prepared in advance, and the predicted physical property values ​​are corrected using the correction coefficients corresponding to the shape types classified based on the image data. It can also be understood as a powder property prediction program. The present invention can also be understood as a powder property prediction program, which is read and executed by a computer, and which causes the computer to acquire image data of an image obtained by photographing particles constituting a powder, and, based on the image data, calculate the particle size distribution of the powder, calculate shape parameters that characterize the shape of the particles, and calculate surface parameters that indicate the unevenness of the particle surface structure of the particles, and predict the physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder, and the physical property value prediction model that predicts the physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder is used in the prediction of the physical property values, and a plurality of physical property value prediction models are prepared corresponding to each of a plurality of shape types that indicate the particle structure, and the physical property value can be predicted using one of the plurality of physical property value prediction models corresponding to one of the plurality of shape types. [Effects of the Invention]

[0015] According to the present invention, a technique is provided that allows the physical properties of powder to be evaluated more simply and easily. [Brief explanation of the drawings]

[0016] [Figure 1] This is an image of a system for predicting the physical properties of powder. [Figure 2] FIG. 1 is a block diagram of a powder property prediction device. [Figure 3] FIG. 1 is a diagram illustrating definitions of convexity, circularity, linearity, and aspect ratio. [Figure 4] 1A and 1B are photographs substituting for drawings showing examples of SEM images of particles. [Figure 5] FIG. 10 is an explanatory diagram of pixel variance values. [Figure 6] 10A and 10B are photographs substituted for drawings showing an example of pixel variance values. [Figure 7] 1A and 1B are diagrams illustrating a method for calculating a surface feature amount. [Figure 8] FIG. 10 is a diagram illustrating normalization of pixel values. [Figure 9] 1A and 1B are photographs substituting for drawings showing examples of surface features. [Figure 10]1A, 1B, and 1C are diagrams illustrating a method for calculating surface feature amounts. [Figure 11] 1A to 1E are diagrams showing examples of particle shape types. [Figure 12] 10 is a flowchart showing a processing procedure. [Figure 13] 10 is a flowchart showing a processing procedure. DETAILED DESCRIPTION OF THE INVENTION

[0017] 1. First embodiment (overview) Figure 1 is an image of a system for predicting the physical properties of powder. This technology analyzes images of the particles that make up the powder to obtain the particle size distribution, shape parameters related to the particle shape, and surface parameters (pixel variance, surface feature values) related to the state of the particle surface. The relationship between these parameters and the measured physical properties of the powder is learned using deep learning, and an AI prediction model, which is a physical property prediction model, is created. The above parameters obtained from images of particles with unknown physical properties are input into this AI prediction model, and the physical properties are predicted.

[0018] (Hardware configuration) 2 is a block diagram of a powder property prediction apparatus 100 using the present invention. The powder property prediction apparatus 100 is configured by a PC (personal computer). The powder property prediction apparatus 100 can also be configured by a dedicated computer. It is also possible to configure the powder property prediction apparatus 100 using a data processing server.

[0019] The powder property prediction device 100 includes an image data acquisition unit 101, an image analysis unit 102, a property value prediction unit 103, a shape type determination unit 104, a correction coefficient acquisition unit 105, and an AI prediction model selection unit 106. These functional units are realized in software by installing application software on the PC. Some of these functional units can also be configured with dedicated hardware such as an FPGA. It is also possible to realize the functional units of Figure 2 in a data processing server and perform processing there.

[0020] The powder property prediction device 100 includes a data storage unit 107. The data storage unit 107 is configured using a semiconductor memory or a hard disk drive included in the PC used. The PC used includes an input / output interface and a user interface represented by a GUI. It is also possible to store data in an external storage device.

[0021] The image data acquisition unit 101 acquires image data of an enlarged image of a powder particle whose physical property value is to be predicted. In this example, the image data acquisition unit 101 receives image data of an SEM image of the particle.

[0022] The image analysis unit 102 analyzes the SEM image of the particles and calculates the particle size distribution and shape parameters (convexity, circularity, linearity, and aspect ratio shown in FIG. 3). The image analysis unit 102 also analyzes the SEM image of the particles and calculates the surface parameters (pixel dispersity and surface feature amount of the particle) described later.

[0023] The physical property value prediction unit 103 predicts the physical property values ​​of the powder composed of the particles using an AI prediction model based on the particle size distribution, shape parameters (convexity, circularity, linearity, aspect ratio), and surface parameters (pixel dispersity and surface feature values ​​of the particles) obtained from the SEM image of the particles. The particle size distribution, shape parameters, and surface parameters will be described later.

[0024] As physical property values of the powder, powder dynamic friction angle (angle), stress transmission rate (%), stress relaxation rate (%), compression rate (%), bulk density (kg / m 3 )), internal friction angle (angle), shear adhesion force (kPa), and flow factor (ffc) can be mentioned. The actual measurement of these physical property values is carried out by using a shear force measuring device according to the method specified in the JIS standard (JIS-Z8835). As the shear stress measuring device, for example, "Powder Layer Shear Force Measuring Device NS-S" manufactured by Nano Seeds Co., Ltd. is used. Based on these physical property values, the fluidity, frictional property, adhesiveness, compressibility, and filling property of the powder can be evaluated.

[0025] Hereinafter, each physical property value will be described. The powder dynamic friction angle is defined in JIS-Z8835, indicates the frictional characteristics of the powder layer during movement, and the smaller the value, the higher the fluidity and the less likely it is to adhere.

[0026] The stress transmission rate is the value obtained by dividing the shear vertical stress by the upper vertical stress, and represents the frictional property between the powder layer and the cell wall surface. The stress relaxation rate is the ratio (%) of the relaxation stress to the initial stress, and indicates the degree of non-return when deformed. The stress relaxation rate is one of the consolidation characteristics of the powder. The compression rate is an index for evaluating the ease of compression.

[0027] The internal friction angle is defined in JIS-Z8835 and indicates the degree of resistance to the shear force between particles. For example, the difficulty of powder avalanche is evaluated by the internal friction angle. The shear adhesion force (kPa) is an index for evaluating the adhesiveness of the powder.

[0028] The flow factor (ffc) is an index for evaluating the fluidity of the powder. The flow factor (ffc) represents the relationship between the uniaxial collapse stress fc and the maximum principal stress σ1, and is given by ffc = σ1 / fc. The fluidity is evaluated as follows: 1 < ffc < 2: very difficult to flow, 2 < ffc < 4: slightly difficult to flow, 4 < ffc < 10: easy to flow, ffc > 10: very easy to flow.

[0029] Generally, if the shear adhesion force (kPa) is large, the flow factor (ffc) is small, and if the shear adhesion force (kPa) is small, the flow factor (ffc) is large.

[0030] The stress relaxation rate, stress transmission rate, shear adhesion (kPa), and flow factor (ffc) are described in J. Soc. Powder Technol., Japan, 54, 90-96 (2017), Vol. 54, No. 2, pp. 90-96.

[0031] The shape type determination unit 104, the correction coefficient acquisition unit 105, and the AI ​​prediction model selection unit 106 are not used in the first embodiment. These functional units will be described later.

[0032] The data storage unit 107 stores data, operation programs, and data obtained as a result of processing and necessary data for operating the powder property prediction apparatus 100. The storage unit 107 stores a program for implementing the functional units of Fig. 2 on a PC, a method for calculating shape parameters of Fig. 3, an AI prediction model, a method for calculating pixel variance values ​​of Fig. 5, a method for calculating pixel feature amounts described in relation to Figs. 7 and 10, a method for normalizing pixel values ​​of Fig. 8, particle shape type data of Figs. 11(A) to (E), and operation programs for executing the processing procedures of Figs. 12 and 13.

[0033] (Explanation of parameters obtained from images) In this embodiment, particle size distribution, four shape parameters, and two surface parameters are obtained from SEM images of particles. According to the inventors' findings, these parameters significantly affect the behavior of the powder. Therefore, the relationship between these parameters and physical property values ​​is learned by AI using deep learning, and an AI prediction model is created.

[0034] First, we will explain the particle size distribution. The particle size distribution of the powder is calculated based on the SEM image of the target powder. The particle size distribution is calculated using a known image analysis method (dedicated software is also available). In this example, x10 ,x 50 ,x 90 The particle size distribution is defined by: where x 10 is the particle size at which the ratio of particles smaller than this is 10%. x 50 x is the particle size at which 50% of the particles are larger and 50% are smaller. 90 is the particle size at which 90% of particles are smaller than this value. The definition of particle size distribution is not limited to the above example, as long as it can evaluate the distribution of the sizes of particles that make up the powder.

[0035] Laser diffraction / scattering methods can also be used to measure particle size distribution. Other methods for measuring particle size distribution include specific gravity, liquid-phase sedimentation, light blocking, electrical detection, optical correlation, and chromatography.

[0036] Next, we will explain four types of shape parameters. The four types of shape parameters are convexity, circularity, linearity, and aspect ratio. The definitions are shown in Figure 3. These shape parameters are examples of parameters that characterize the shape of particles that make up powder. These shape parameters are obtained from image analysis of enlarged images of the particles.

[0037] The above image analysis is performed as follows. First, the SEM image of the particle is subjected to black and white binarization to obtain a binary image. Next, this binary image is subjected to contour extraction processing to extract the particle contour. From this particle contour, image analysis is performed to calculate the convexity, circularity, linearity, and aspect ratio defined in Figure 3.

[0038] Next, the surface parameters will be described. The surface parameters in this example are a pixel variance value and a surface feature amount. The pixel variance value and the surface feature amount indicate the uneven state of the structure of the particle surfaces that make up the powder.

[0039] Here, the particles in Figure 4(A) and (B) are almost the same shape and size. However, (A) has low fluidity, while (B) has high fluidity. This difference is due to the difference in surface condition. (A) has fine irregularities and a fine surface microstructure, while (B) does not. Convexity, circularity, linearity, and aspect ratio cannot effectively quantify this difference in the detailed surface structure. Therefore, we consider a method to detect (evaluate) the difference between Figure 4(A) and (B) from images.

[0040] Here, we introduce pixel variance and surface feature quantities as parameters for quantitatively evaluating the image fineness resulting from the structure of the particle surface. First, we explain pixel variance. Figure 5 shows the definition of pixel variance. Pixel variance is an index that indicates the degree of dispersion of pixels in an image.

[0041] In the case of an image of a particle with a smooth surface, such as that shown in Figure 4(B), the pixel variance value is small. This is because in an image of a smooth surface, the difference in pixel values ​​depending on the location is small, and the right-hand side of the definition equation in Figure 5 is small.

[0042] In contrast, in the case of an image of a fine surface condition such as that shown in Figure 4(A), the pixel variance value is large. This is because the pixel values ​​vary greatly depending on the location, and the right-hand side of the definition equation in Figure 5 becomes large.

[0043] Fig. 6(A) shows pixel variance values ​​for particles with relatively smooth surfaces, and Fig. 6(B) shows pixel variance values ​​for particles with relatively rough surfaces.

[0044] Next, we will explain surface feature quantities. If the surface of a particle has fine irregularities, the image will have a fine pattern. We will consider evaluating this fineness. Here, we consider the pixels that make up the image. Images with fine patterns have many fine changes in shade. Therefore, the gradation of the pixels is quantified as pixel values, and the degree of change depending on the location is measured.

[0045] For example, consider the 4x4 pixel images (A) and (B) shown in Figure 7. Here, the numbers for each pixel are pixel values. For example, pixel values ​​are expressed in five shades of gray.

[0046] Here, the surface feature amount is calculated as follows. First, consider the sum Vd of the differences in pixel values ​​in the vertical direction. Here, we focus on the pixels in the vertical direction (columns). In the image in Figure 7, there are four vertical columns. For each of these columns, the values ​​of the pixels adjacent above and below are compared and the difference is calculated. The above differences for all columns are then accumulated. This accumulated value is called Vd. In the case of (A), the accumulated value of the differences between adjacent pixels in the first column from the left is 3, the accumulated value of the differences between adjacent pixels in the second column is 3, the accumulated value of the differences between adjacent pixels in the third column is 4, and the accumulated value of the differences between adjacent pixels in the fourth column is 4. Therefore, Vd = 3 + 3 + 4 + 4 = 14.

[0047] Using the same concept, consider the sum Hd of the differences in pixel values ​​in the horizontal direction. In the case of (A) above, the differences in pixel values ​​between adjacent pixels arranged horizontally (a row of pixels in four rows) are summed. This sum is taken as Hd. In the case of (A), there is only a difference between adjacent pixels in the horizontal direction in the fourth row, so Hd=1.

[0048] Here, the number of pixels in the vertical direction is Nv and the number of pixels in the horizontal direction is Nh, and the surface feature amount is defined as follows: Surface feature value = (Vd / Nh) + (Hd / Nv)

[0049] In the case of Figure 7(A), (Vd / Nh) is (3+3+4+4) / 4=14 / 4, and (Hd / Nh) is 1 / 4. Therefore, the surface feature value is 14 / 4+1 / 4=15 / 4=3.75.

[0050] In the case of Figure 7(B), (Vd / Nh) is (1+1+1) / 4, and (Hd / Nv) is 1 / 4. Therefore, the surface feature amount is 3 / 4+1 / 4=1.

[0051] The actual calculation of surface feature values ​​is performed as follows. Here, we will explain the case where pixel values ​​have 256 gradations. Furthermore, the smallest unit pixel is used. It is also possible to consider multiple pixels, such as 2x2 pixels, as a single pixel. In this case, the amount of calculation can be reduced, but the accuracy will decrease.

[0052] First, an SEM image of the particle is obtained and an approximate ellipse of the particle is set. Then, the image is rotated so that the major axis of this approximate ellipse is horizontal, and the direction of the major axis of the approximate ellipse is set to horizontal and the direction of the minor axis is set to vertical. Then, the Vd is calculated for the pixels related to the particle arranged in the vertical direction, and the Hd is calculated for the pixels related to the particle arranged in the horizontal direction.

[0053] In this case, in order to reduce the influence of brightness, a standard value X is used as the pixel value used in the calculation. The standard value X is X=255×(xx min ) / (x max -x min ) where x is the pixel value of the target pixel, and x max is the maximum pixel value in the target image, and x min is the minimum pixel value. Figure 7 shows an image of the above normalization. In the case of 512 gradations, X=511×(xx min ) / (x max -x min ) is used. The use of normalized pixel values ​​is the same in other embodiments.

[0054] 9(A) shows the surface feature amount for a particle with a relatively smooth surface, and FIG. 9(B) shows the surface feature amount for a particle with a relatively rough surface.

[0055] (Example of processing: Pre-processing) First, an AI prediction model is created by deep learning the relationship between the particle size distribution, shape parameters shown in Figure 3, pixel variance value, and surface feature values, and the physical properties of powder particles.

[0056] Deep learning is performed as follows: First, an SEM image of the sample powder particles is obtained, and image analysis data (particle size, convexity, circularity, linearity, aspect ratio, pixel dispersion, and surface features) of the particles is calculated through image analysis. The physical properties of this powder are then measured through a shear test.

[0057] This is performed on as many types of powder as possible, and the AI ​​learning model is trained to learn the correspondence between the image analysis data for each powder and the measured shear test data. Random forest is used as the AI ​​learning method. This technology utilizes the strong correlation between image analysis data and measured physical property values. This correlation is trained by machine learning into the AI ​​learning model, and an AI prediction model is created that predicts the physical property values ​​of the powder from the image analysis data. The created AI prediction model is stored in data storage unit 107. Then, when operating the system, image analysis data from SEM images of the sample powder is input into the AI ​​prediction model to obtain predicted physical property values.

[0058] (Example of processing: Prediction of physical properties) First, image data of the SEM image of the powder whose physical properties are to be predicted is acquired. Next, the acquired SEM image data is input into an AI prediction model to predict the physical properties of the powder (see Figure 1).

[0059] The results are shown in Table 1 below. Here, parameters (1) to (4) are the convexity, circularity, linearity, and aspect ratio defined in Figure 3. Parameter (5) is the pixel variance value. Parameter (6) is the surface feature value. The experimental values ​​are actual measurements obtained using a measuring device (shear test device).

[0060] The physical properties shown in Table 1 are powder kinetic friction angle, stress transmission rate, stress relaxation rate, compressibility, bulk density, internal friction angle, shear adhesion, and flow factor.

[0061] [Table 1]

[0062] As shown in Table 1, by adding parameters (5) and (6) in addition to parameters (1) to (4), the accuracy of the predicted value is improved compared to when only parameters (1) to (4) are used.

[0063] 2. Second embodiment Below, we will explain examples of surface features for achieving even higher prediction accuracy. Figures 10(A), (B), and (C) show examples of pixel values ​​(pixel values ​​are in five levels from 1 to 5) for an image consisting of 5 × 5 pixels.

[0064] The surface feature values ​​defined in the first embodiment are the same in Figures 10(A), (B), and (C). However, the state of change in pixel values ​​differs between Figures 10(A), (B), and (C). That is, Figure 10(A) shows a sloped but smooth surface, Figure 10(B) shows a surface with one convex portion, and Figure 10(C) shows a surface with two convex portions.

[0065] In order to quantitatively evaluate the difference in the surface state described above, the following is considered as a second surface feature. Here, an increase / decrease coefficient δ is introduced. The increase / decrease coefficient δ is defined as follows:

[0066] First, when viewed in a specific direction, the number of poles where the number of pixels related to the particle changes from an increase to a decrease is set as the increase / decrease coefficient. For example, when viewed in a horizontal arrangement, (1) If the pixel values ​​are in the order 1, 2, 3, 4, 5, there is no pole and δ=0. (2) If the pixel values ​​are in the order 1, 2, 3, 2, 1, then δ=1. (3) If the pixel values ​​are in the order 1, 2, 1, 2, 1, then δ=2.

[0067] Here, the increase / decrease coefficient for vertical alignment is denoted as δv, and the increase / decrease coefficient for horizontal alignment is denoted as δh. In the case of (1), δh=0, in the case of (2), δh=1, and in the case of (3), δh=2.

[0068] δv and δh are the number of peaks of the unevenness. In a specific range, a larger number of peaks means that there are finer unevennesses. In this sense, the coefficient of increase / decrease δ can be considered an index (parameter) that indicates the "amount of unevenness on the surface."

[0069] Here, the second surface feature amount is defined as follows, where Nv is the number of pixels of the particle in the vertical direction and Nh is the number of pixels of the particle in the horizontal direction.

[0070] Second surface feature value = (Σδv / Nh) + (Σδh / Nv) If Σ is the vertical direction, it indicates that all δv in the vertical direction is integrated, and if it is the horizontal direction, it indicates that all δh in the horizontal direction is integrated.

[0071] The second surface feature amount in the case of FIG. 10(A) is as follows. First, δv in the first to fifth vertical columns is all 0. Therefore, (Σδv / Nh) = 0. Also, δh in the first to fifth horizontal rows is all 0. Therefore, (Σδh / Nv) = 0. Therefore, the surface feature amount of the second image in FIG. 10(A) is 0.

[0072] The second surface feature amount in the case of FIG. 10(B) is as follows. First, the δv in the first to fifth vertical columns is all 0. Therefore, (Σδv / Nh)=0. Also, the δh in the first to fifth horizontal rows is all 1. Therefore, (Σδh / Nv)=(1+1+1+1+1) / 5=1. Therefore, the surface feature amount of the second image in FIG. 10(A) is 1.

[0073] The second surface feature amount in the case of FIG. 10(C) is as follows. First, the δv in the first to fifth vertical columns is all 0. Therefore, (Σδv / Nh)=0. Also, the δh in the first to fifth horizontal rows is all 2. Therefore, (Σδh / Nv)=(2+2+2+2+2) / 5=2. Therefore, the surface feature amount of the second image in FIG. 10(A) is 2.

[0074] The second surface feature value can be regarded as a parameter that indicates the degree of unevenness in the change in pixel value. The second surface feature value can quantify the unevenness of the particle surface.

[0075] 3. Third embodiment In the second embodiment, a viewpoint of quantifying the number of unevenness poles is introduced as a surface feature amount. In this embodiment, a viewpoint of quantitatively grasping the degree of steepness of unevenness poles is further introduced as a surface feature amount.

[0076] First, we focus on the pole of the convex / concave pattern (the part where the pixel value changes from increasing to decreasing) when viewing the pixel array in a specific direction (for example, horizontally). This pole is the apex of the convex part. Next, we calculate the slope of the slope before and after this pole.

[0077] For example, if we focus on a certain pole, let's say the pixel values ​​change from 3, 2, 1, 3, 5, 4, 2, 3. In this case, we measure the change in pixel value using the bottoms of the concave parts before and after the convex part as the start and end points. Here, the slope can be calculated from (change in pixel value / number of pixels).

[0078] In this example, the part where the pixel values ​​are lined up as 1, 3, 5, 4, 2 is the convex part (pole), and the slope of the front slope part is (5-1) / 2 = 2, and the slope of the back slope part is (5-2) / 2 = 1.5. Here, the slope of the convex part is the average value of the front and back slopes, which is 1.75.

[0079] In specific calculations, the sum of the polar gradients when focusing on a specific column of pixels that make up the particle in the vertical direction is Gv, and the sum of the polar gradients when focusing on a specific column of pixels that make up the particle in the horizontal direction is Gh. Furthermore, the third surface feature is defined as follows, where Nv is the number of pixels in the vertical direction and Nh is the number of pixels in the horizontal direction. Third surface feature value = (ΣGv / Nh) + (ΣGh / Nv) If Σ is the vertical direction, it indicates that all Gv values ​​in the vertical direction are integrated, and if Σ is the horizontal direction, it indicates that all Gh values ​​in the horizontal direction are integrated.

[0080] The more sharp the poles (protrusions), the larger the third surface feature value. In this way, a surface feature value that quantitatively evaluates the sharpness of fine protrusions is obtained.

[0081] Generally, the more small and sharp protrusions there are, the worse the particle flow becomes. Therefore, by introducing an index based on this viewpoint, it is possible to improve the accuracy of prediction of physical properties related to particle flow.

[0082] 4. Fourth Embodiment The number of feature points (density) can also be used as a surface feature. Feature points are extracted as points in an image where brightness or shade changes sharply. Feature points are used for matching between multiple images. Several algorithms have been developed for extracting feature values ​​from images, and these can be used.

[0083] In order to reduce the influence of differences in image brightness, the image to be used for feature point extraction is one that has been recreated using the pixel volume of each pixel normalized as described in the first embodiment.

[0084] Here, if the number of extracted feature points is f, the surface feature amount is defined as (f / total number of pixels of the particle). In the case of the particle in Figure 4(A), the surface has a fine structure, and the number of feature points in the image is relatively large. In contrast, in the case of the particle in Figure 4(B), the surface is relatively smooth, and the number of feature points is relatively small.

[0085] 5. Fifth Embodiment As mentioned above, there are various methods for calculating the surface feature values ​​of particles. When calculating the surface feature values ​​using these methods, rotating an enlarged image of a particle may result in a change in the surface feature values. Below, we will explain a method for calculating surface feature values ​​that addresses this issue.

[0086] First, an enlarged image of the particle (e.g., an SEM image) is acquired. This is the same as in other embodiments. Next, the image is rotated by an angle θ, and the surface feature amount is calculated using the method for calculating the surface feature amount disclosed in this specification. This surface feature amount is designated as q(θ).

[0087] By changing the value of θ, q(θi) is calculated at various angles θi (which may include θi = 0). The multiple calculated q(θi) values ​​are compared to find the θi = θmax at which the value is maximum. q(θmax) is then used as the surface feature of the particle.

[0088] For example, for a certain particle, θi is set from 10° to 180° in increments of 10° to calculate a total of 18 surface feature amounts. In this case, the largest surface feature amount among them is adopted as the surface feature amount of the particle.

[0089] This method solves the problem that the surface parameters, which indicate the unevenness of the particle surface structure, change depending on the orientation (rotation) of the particle. In theory, rotation-invariant feature quantities can be obtained. This embodiment can be applied to any of the surface feature quantities described in this specification.

[0090] By using two or more of the surface feature amounts described in the first to fifth embodiments for learning the AI ​​prediction model, it is possible to expect prediction of powder properties (particularly properties related to flowability) with even higher accuracy.

[0091] 6. Sixth Embodiment As can be seen from Table 1, the predicted values ​​of shear adhesion and flow factor by the AI ​​prediction model differ greatly from the measured values. Furthermore, this difference is not uniform. According to the analysis by the inventors, the difference between the predicted and measured values ​​of physical properties tends to depend on the particle shape. This is not limited to shear adhesion and flow factor, but is also seen in other physical properties.

[0092] 11(A) to 11(E) show examples of shape types categorized by particle shape. The shape types are: a thin rod-like shape with a longitudinal length 20 times or more than the diameter of the cross section approximated by a circle (Type 1); a rectangular parallelepiped with the ratio of the minimum side length to the maximum side length less than 20 times; a thin plate-like shape with a ratio of thickness to width in the shortest direction 10 times or more (Type 3); an aggregate of particles with a particle diameter of 1 / 30 or less of the aggregate diameter (Type 4); an aggregate of particles with a particle diameter exceeding 1 / 30 of the aggregate diameter (Type 5); a Type 2 rectangular parallelepiped with smaller rectangular parallelepipeds bonded to the surface (Type 6); a Type 2 rectangular parallelepiped with smaller round three-dimensional objects bonded to the surface (Type 7); and a hexagonal thin plate-like shape that meets the conditions of Type 3.(Type 8), a mass consisting of many Type 2 shapes (Type 9), a thin, irregular polygonal shape with pentagons or more that meets the conditions of Type 3 (Type 10), a flat disk shape that meets the conditions of Type 3 (Type 11), a shape consisting of a collection of cubes whose diameter approximates a sphere and meets the conditions of Type 5 (Type 12), a long, thin polygonal shape with sharp corners (like a piece of broken wood) (Type 13), a flat oval shape that meets the conditions of Type 3 (Type 14), a gourd-shaped shape (Type 15), a shape consisting of a collection of thin, irregular particles that meets the conditions of Type 3 (Type 16), a shape consisting of a collection of thin, irregular particles with a hole in the center (Type 17), a shape consisting of many overlapping thin, irregular particles that meets the conditions of Type 3 (Type 18), a thin, irregular, plate-shaped shape with many sharp corners that meets the conditions of Type 3 Examples of such particles include a collection of particles with sharp corners (Type 19), particles with many sharp corners (Type 20), hollow pipe-shaped particles (Type 21), cylindrical particles with a larger diameter than Type 1 (Type 22), triangular prism-shaped particles whose height is shorter than the length of the sides that make up the triangle (Type 23), polyhedral particles with eight or more sides (Type 24), disk-shaped (donut-shaped) particles with a hole in the center (Type 25), spherical particles with multiple holes on their surface (Type 26), spherical particles with many particles embedded on their surface that are relatively small in diameter compared to the sphere (Type 27), spherical particles with many particles attached to their surface that are less than 1 / 50 the diameter of the sphere (Type 28), and spherical particles with many particles attached to their surface that are 1 / 10 to 1 / 50 the diameter of the sphere (Type 29).

[0093] The differences in shape types shown in Figures 11(A) to (E) are primarily related to the three-dimensional structure. Shape parameters, pixel variance values, and surface feature values ​​are information obtained from two-dimensional images, and there are limitations to their ability to determine the characteristics of three-dimensional shapes. This is thought to be one of the reasons for the discrepancy between the predicted and measured values ​​mentioned above.

[0094] Below, we will explain the relationship between the above shape types and the shape parameters and surface parameters in Figure 3. Type 1 shape type can be evaluated using the linearity and aspect ratio in the shape parameters of Figure 3, but when quantified using these parameters, the differences between Types 3, 9, and 16 may not be as apparent. Furthermore, the characteristics of Type 1 shape type cannot be accurately evaluated using the surface parameters in Figures 5, 7, and 10.

[0095] The characteristics of Types 2, 3, 8, 10, 11, 13, 14, and 15 cannot be accurately evaluated using shape and surface parameters. Types 4 to 7 can be evaluated to some extent using surface parameters. However, the differences in their structures cannot be properly evaluated using surface parameters. For example, the difference between Types 6 and 7 cannot be distinguished using surface parameters. This tendency is also true for Types 9, 12, 16, 17, 18, and 19.

[0096] For the above reasons, in this embodiment, classification by shape type is introduced to improve the accuracy of prediction of physical properties by shape parameters and surface parameters.

[0097] As mentioned above, the difference between predicted and measured values ​​tends to depend on the particle shape. That is, it has been found that the deviation between the predicted and measured values ​​of physical properties of powders composed of particles classified as Type 2 shows a specific tendency, and the deviation between the predicted and measured values ​​of physical properties of powders composed of particles classified as Type 4 shows another specific tendency.

[0098] Therefore, correction coefficients are calculated in advance according to the particle shape type as shown in Figures 11(A) to (E), and the physical property values ​​predicted by the AI ​​prediction model are corrected using these correction coefficients. Examples of correction coefficients are shown in Table 2 below. Table 2 shows an example of calculating the correction coefficient for each parameter using the "experimental value (actual measured value) / predicted value" formula.

[0099] [Table 2]

[0100] In this case, correction coefficients for each physical property value, such as those shown in Table 2, are prepared in advance for each particle shape type and stored in the data storage unit 107. Then, the physical property values ​​predicted by the AI ​​prediction model are corrected using the correction coefficients according to the particle shape type to obtain corrected predicted values. For particles of similar shape types, by using the same correction coefficients, predicted values ​​close to the actual measured values ​​can be obtained, although the accuracy is not perfect.

[0101] Although it takes time and effort, by classifying the shape types in detail and calculating the correction coefficients shown in Table 2 for each shape type, it is possible to predict the physical properties of powders with higher accuracy.

[0102] An example of the process is shown in Figure 12. The program that executes the process in Figure 12 is stored in an appropriate storage area or storage medium, and is executed by the PC that constitutes the powder property prediction device in Figure 2. The same applies to the process in Figure 13.

[0103] As a pre-processing step, an AI prediction model is created in advance, as in the first embodiment. In addition, correction coefficients corresponding to the particle shape types exemplified in Figures 11(A) to 11(E) are prepared (see, for example, Table 2).

[0104] When the process starts, first, an SEM image of particles constituting the powder is acquired (step S201). Next, the acquired SEM image is analyzed to calculate the particle size distribution, shape parameters (convexity, circularity, aspect ratio, linearity), pixel variance, and surface feature amount (step S202). This process is performed by the image analysis unit 102 in FIG. 2.

[0105] Next, the particle size distribution, shape parameters (convexity, circularity, aspect ratio, linearity), pixel variance, and surface feature values ​​obtained in step S202 are input into a pre-prepared AI prediction model to predict the physical property values ​​of the powder (step S203). This process is performed in the physical property value prediction unit 103.

[0106] 11(A) to 11(E) is determined based on the SEM image of the particle acquired in step S201 (step S204). This process is performed by the shape type determination unit 104 in Fig. 2. This determination is performed using image recognition technology.

[0107] For example, images of shape types are stored in advance, and these are compared with the particle image acquired in step S201. From this comparison, it is determined which shape type is most similar to the particle image. This process is performed using AI image judgment technology, which is used in crime prevention, automatic driving of vehicles (autonomous driving), accident prevention technology, product image recognition, etc.

[0108] It is also possible to have this determination performed by a human and the result accepted. In this case, the shape type determination unit 104 functions as a shape type acceptance unit. In this embodiment, for example, prepared shape models are displayed on the display of the PC to be used. Then, from among the displayed models, the user selects a shape model that he or she determines to be suitable for the target particle. This selected shape model is accepted by the shape type acceptance unit.

[0109] Next, a correction coefficient according to the determined shape type is acquired (step S205). This process is performed by the correction coefficient acquisition unit 105. Next, using the correction coefficient selected in step S205, the predicted value is corrected as shown in Table 2 (step S206).

[0110] (others) The shape correspondence type can be set more precisely. If the number of shape types is small, the accuracy of prediction decreases. However, taking into consideration the cost of preparing data in advance, it is also possible to adopt a mode in which multiple shape types exemplified in FIGS. 11(A) to (E) are adopted. This is also true for the seventh embodiment.

[0111] 7. Seventh Embodiment In the sixth embodiment, when focusing on a single particle, suppose there is a particle that has two different shape types mixed together. Alternatively, suppose the target powder contains particles of two different shape types. In this case, the physical properties are predicted using the weighted average of the correction coefficients for the two shape types.

[0112] For example, when focusing on a single particle, suppose type 4 structure and type 5 structure are observed. Suppose the ratio is 7:3. In this case, a correction coefficient is used that is a weighted average of the type 1 correction coefficient and the type 2 correction coefficient in a ratio of 7:3. For example, if the correction coefficient for type 4 of a certain physical property is α and the correction coefficient for type 5 is β, then the correction coefficient in this case is (7 / 10)α + (3 / 10)β.

[0113] For example, suppose there is a powder in which 60% of the particles have a type 5 structure and 40% have a type 6 structure. In this case, a correction coefficient is used that is a weighted average of the type 5 correction coefficient and the type 6 correction coefficient in a ratio of 6:4. For example, if the correction coefficient for type 1 of a certain physical property is α and the correction coefficient for type 2 is β, the correction coefficient in this case is (6 / 10)α + (4 / 10)β. It is also possible to use three or more shape types.

[0114] 8. Eighth Embodiment As mentioned above, the deviation of predicted values ​​from actual measured values ​​tends to vary depending on the type of particle shape that makes up the powder. This is thought to be due to the fact that AI deep learning is unable to fully handle the diverse shape types.

[0115] Generally, in AI deep learning, the learning effect is higher when the learning content is narrowed (obviously, the more samples there are during learning, the better). Therefore, in this embodiment, an AI prediction model is prepared for each shape type of powder particles. That is, an AI prediction model is prepared by performing deep learning on particles of shape type 1, an AI prediction model is prepared by performing deep learning on particles of shape type 2, and so on. Note that, for example, the same AI prediction model may be used for shape types 2 and 3. The prepared AI prediction models are stored in the data storage unit 107. That is, the processing method differs from that of the sixth embodiment in that, before inputting the particle size distribution, shape parameters, pixel variance, and surface feature values ​​of the powder into the AI ​​prediction model and predicting the physical properties of the powder, the shape type of the target particles is first determined, an AI prediction model corresponding to that shape type is selected, and the particle size distribution, shape parameters, pixel variance, and surface feature values ​​of the powder are input into the selected AI prediction model to predict the physical properties of the powder.

[0116] An example of the process is shown in Fig. 13. First, an SEM image of particles constituting the powder is acquired (step S401). Next, the acquired SEM image is analyzed to calculate the particle size distribution, shape parameters (convexity, circularity, aspect ratio, linearity), pixel variance, and surface feature amount (step S402). This process is performed by the image analysis unit 102 in Fig. 2.

[0117] Next, based on the SEM image of the particle acquired in step S201, the shape type of the particle is determined as shown in Figures 11(A) to 11(E) (step S403). This process is performed by the shape type determination unit 104 in Figure 2. This determination is performed using image recognition technology. Alternatively, a human may determine the shape type and the result may be accepted.

[0118] Next, an AI prediction model corresponding to the shape type selected in the determination of step S403 is selected (step S404). This process is performed by the AI ​​prediction model selection unit 106.

[0119] Next, the particle size distribution, shape parameters (convexity, circularity, aspect ratio, linearity), pixel variance, and surface feature values ​​obtained in step S402 are input to the selected AI prediction model, and the physical property values ​​of the powder are predicted (step S405). This process is performed in the physical property value prediction unit 103.

[0120] 9. Superiority The physical properties of powders can be predicted from image data. As long as an image can be obtained, only a small amount of powder sample is required. Furthermore, compared to devices used to actually analyze powders, devices used to obtain images (e.g., SEM devices) are common and easy to handle. This makes it easier to evaluate the physical properties of powders.

[0121] 10. Other Embodiments As shape parameters of powder, uniformity, space filling, sphericity, circularity, irregularity, and surface roughness can also be used. Details of these parameters are described on pages 31 to 39 of the Powder Technology Handbook (Asakura Publishing, published February 20, 2014, ISBN978-4-254-25267-5).

[0122] In this specification, an SEM image is shown as an example of a particle image, but any particle image obtained by other methods may be used as long as it is an image from which convexity, circularity, linearity, aspect ratio, pixel dispersity, and surface feature values ​​can be calculated by image analysis.

[0123] For example, a system can be created in which the configuration shown in FIG. 2 is placed in a data processing server, image data of powder particles is sent to the server, and the physical property values ​​of the powder are predicted.

[0124] The physical properties of the powder to be predicted are the powder kinetic friction angle (angle), stress transfer rate (%), stress relaxation rate (%), compressibility (%), bulk density (kg / m 3 ), internal friction angle (angle), shear adhesion (kPa), and flow factor (ffc) may be used in combination with two or more of them.

Claims

1. an image data acquisition unit that acquires image data of an image obtained by photographing particles that constitute the powder; an image analysis unit that calculates a particle size distribution of the powder, shape parameters that characterize the shapes of the particles, and surface parameters that indicate the unevenness of the particle surface structure of the particles, based on the image data; a powder physical property value prediction unit that predicts physical property values ​​of the powder based on the particle size distribution of the powder, the shape parameters, and the surface parameters; Equipped with In advance, Multiple shape types for particle structure; correction coefficients corresponding to the plurality of shape types; are available, A powder property prediction system in which the predicted property values ​​are corrected using the correction coefficients corresponding to the shape types classified based on the image data.

2. an image data acquisition unit that acquires image data of an image obtained by photographing particles that constitute the powder; an image analysis unit that calculates a particle size distribution of the powder, shape parameters that characterize the shapes of the particles, and surface parameters that indicate the unevenness of the particle surface structure of the particles, based on the image data; a powder physical property value prediction unit that predicts physical property values ​​of the powder based on the particle size distribution of the powder, the shape parameters, and the surface parameters; Equipped with the physical property value prediction unit has a physical property value prediction model that predicts physical property values ​​of the powder based on the particle size distribution of the powder, the shape parameters, and the surface parameters; a plurality of the physical property prediction models are prepared corresponding to a plurality of shape types that indicate the structure of the particle, a powder property prediction system for predicting the physical property values ​​using one of the plurality of physical property prediction models corresponding to one of the plurality of shape types;

3. The shape parameters include convexity, circularity, aspect ratio, and linearity; The convexity is the ratio of the perimeter of a convex polygon covering a particle to the perimeter of the particle, The circularity is a parameter indicating how close the particle shape is to a circle, The aspect ratio is the ratio of the major axis to the minor axis when the particle shape is approximated as an ellipse, 3. The powder property prediction system according to claim 1, wherein the linearity is a parameter indicating the degree of linearity of particles.

4. the surface parameters are pixel variance values ​​and surface feature quantities; the pixel variance value is a parameter for evaluating the magnitude of variance of pixel values, 4. The powder property prediction system according to claim 1, wherein the surface feature amount is a parameter for evaluating the degree of the feature of the particle surface.

5. The surface feature amount is It is calculated based on the change in pixel density in a specific direction, the surface feature amount is determined from a plurality of surface feature amount candidates; the plurality of surface feature value candidates are surface feature values ​​in a plurality of different directions as the specific direction, The powder property prediction system according to claim 4 , wherein the largest one of the plurality of surface feature candidates is adopted.

6. The powder property prediction system according to any one of claims 1 to 5, wherein the physical property values ​​include two or more selected from a powder kinetic friction angle, a stress transmission rate, a stress relaxation rate, a compressibility rate, a bulk density, an internal friction angle, a shear adhesion force, and a flow factor.

7. 3. The powder property prediction system according to claim 1, wherein the shape types are classified based on characteristics of the three-dimensional structure of the particles.

8. The shape types include: It is a thin rod-like shape with a longitudinal length 20 times or more than the diameter of the cross section approximated by a circle (Type 1). It is a rectangular parallelepiped with the ratio of the minimum side length to the maximum side length being less than 20 (Type 2). It is a thin plate with a ratio of thickness to width in the short direction of 10 or more (Type 3). It is an aggregate of particles, and the diameter of the particle is 1 / 30 or less of the diameter of the aggregate (Type 4). It is an aggregate of particles, and the diameter of the particle is more than 1 / 30 of the diameter of the aggregate (Type 5). A smaller rectangular parallelepiped is attached to the surface of a Type 2 rectangular parallelepiped (Type 6). Type 2: A smaller round, three-dimensional object is attached to the surface of the rectangular parallelepiped (Type 7). Hexagonal thin plate (Type 8) that meets the requirements of Type 3, A mass of many Type 2 shapes (Type 9), Thin, irregular polygonal shapes with five or more sides that meet the requirements of Type 3 (Type 10), A flat disk shape (Type 11) that satisfies the conditions of Type 3, A shape consisting of a collection of cubes whose diameters approximate a sphere and satisfy the conditions of Type 5 (Type 12). Long, polygonal shapes with sharp corners (like broken wood) (Type 13), A flat oval shape (Type 14) that meets the requirements of Type 3; Gourd-shaped (Type 15), Type 16 is a collection of thin plate-like particles that meet the conditions of Type 3. A collection of thin, irregular particles with a hole in the center (Type 17), A shape consisting of many overlapping thin, irregular particles that meet the conditions of Type 3 (Type 18), Type 19: a collection of thin, sharp particles that meet the requirements of Type 3; Particles with many sharp corners (Type 20), Hollow pipe-shaped particles (Type 21), Cylindrical particles with a larger diameter than Type 1 (Type 22), Particles in the shape of a triangular prism, the height of which is shorter than the length of the sides that make up the triangle (Type 23), Polyhedral particles with eight or more sides (Type 24), Disk-shaped (donut-shaped) particles with a hole in the center (Type 25), Spherical particles with multiple holes on the surface (Type 26), Particles with a spherical shape and many particles with a relatively small diameter compared to the sphere embedded on the surface (Type 27), Spherical particles with numerous particles attached to their surface, each with a diameter less than 1 / 50 of the diameter of the sphere (Type 28); 3. The powder property prediction system according to claim 1 or 2, wherein the system includes one or more particles selected from the group consisting of spherical particles having a large number of particles attached to their surface, each having a diameter 1 / 10 to 1 / 50 of the diameter of the sphere (Type 29).

9. Image data of the image obtained by photographing the particles that make up the powder is acquired, calculating a particle size distribution of the powder, calculating shape parameters that characterize the shapes of the particles, and calculating surface parameters that indicate the unevenness of the particle surface structure of the particles based on the image data; predicting physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder; In advance, Multiple shape types for particle structure; correction coefficients corresponding to the plurality of shape types; are available, The powder property prediction method includes correcting the predicted property value using the correction coefficient corresponding to the shape type classified based on the image data.

10. Image data of the image obtained by photographing the particles that make up the powder is acquired, calculating a particle size distribution of the powder, calculating shape parameters that characterize the shapes of the particles, and calculating surface parameters that indicate the unevenness of the particle surface structure of the particles based on the image data; predicting physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder; In predicting the physical property values, a physical property prediction model is used that predicts the physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder, a plurality of the physical property prediction models are prepared corresponding to a plurality of shape types that indicate the structure of the particle, A powder property prediction method in which the physical property is predicted using one of the plurality of physical property prediction models corresponding to one of the plurality of shape types.

11. A program to be read and executed by a computer, To the computer Acquiring image data of an image obtained by photographing particles that make up the powder; calculating a particle size distribution of the powder, a shape parameter characterizing the shape of the particles, and a surface parameter indicating the uneven state of the particle surface structure of the particles, based on the image data; predicting physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder; In advance, Multiple shape types for particle structure; correction coefficients corresponding to the plurality of shape types; are available, A powder property prediction program in which the predicted property values ​​are corrected using the correction coefficients corresponding to the shape types classified based on the image data.

12. A program to be read and executed by a computer, To the computer Acquiring image data of an image obtained by photographing particles that make up the powder; calculating a particle size distribution of the powder, a shape parameter characterizing the shape of the particles, and a surface parameter indicating the uneven state of the particle surface structure of the particles, based on the image data; predicting physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder; In predicting the physical property values, a physical property prediction model is used that predicts the physical property values ​​of the powder based on the particle size distribution, the shape parameters, and the surface parameters of the powder, a plurality of the physical property prediction models are prepared corresponding to a plurality of shape types that indicate the structure of the particle, A powder property prediction program that predicts the physical property values ​​using one of the plurality of physical property prediction models corresponding to one of the plurality of shape types.

Images