Composition inspection method, composition inspection device, and composition manufacturing method

Abstract

Provided are: a composition inspection method in which a composition can be inspected in a non-contact manner and which is suited to in-line inspection of the composition in a short period of time; a composition inspection device; and a composition manufacturing method. A composition inspection method includes a step for continuously feeding a composition from a nozzle port of a nozzle and a step for acquiring a video or image of the composition continuously fed from the nozzle port, and also includes at least one step from among: a step A for calculating at least one from among the shape of the composition, the flow velocity of the composition, and the optical characteristics of the composition; and a step B for inspecting the composition by using a trained machine learning model that uses the video or image of the composition as input data.

Description

Method for inspecting a composition, apparatus for inspecting a composition, and method for manufacturing a composition

[0001] The present invention relates to a method for inspecting a composition, an apparatus for inspecting a composition, and a method for manufacturing a composition, and more particularly to a method for inspecting a composition continuously delivered from a nozzle, an apparatus for inspecting a composition, and a method for manufacturing a composition.

[0002] Currently, various compositions such as inks, photosensitive compositions, and magnetic pastes are manufactured. It is desirable that the composition exhibits little performance difference between manufacturing lots. For this reason, conventionally, when manufacturing compositions, attempts have been made to produce compositions that exhibit similar performance in each manufacturing lot to those in other manufacturing lots. However, it is not always possible to produce compositions that exhibit similar performance in all manufacturing lots. Therefore, for quality control, a sample of the composition is taken for inspection. When a sample of a composition is taken for inspection, the inspection of the composition is time-consuming. To solve this problem, for example, as a non-contact method for measuring physical properties, Patent Document 1 proposes a liquid property measurement system having an imaging device that images the liquid discharged from a nozzle, a first calculation unit that calculates characteristic quantities about the extension behavior of the liquid until it leaves the nozzle outlet based on the image captured by the imaging device, and a second calculation unit that calculates the liquid properties of the liquid based on the characteristic quantities calculated by the first calculation unit.

[0003] Japanese Patent Publication No. 2024-037509

[0004] In composition inspection, it is desirable to be able to inspect the composition non-contact, in-line, and in a short time without taking a sample of the composition. However, while Patent Document 1 can calculate liquid properties non-contact, it requires imaging the extension behavior of the liquid until it leaves the nozzle outlet, making it unsuitable for continuously measuring liquid properties. Thus, there is currently no method suitable for in-line, short-time composition inspection. The object of the present invention is to provide a method for inspecting a composition, an apparatus for inspecting a composition, and a method for manufacturing a composition that are suitable for non-contact, in-line, and short-time composition inspection.

[0005] The above objective can be achieved with the following configuration. Invention [1] is a method for inspecting a composition, comprising the steps of: continuously supplying a composition from the nozzle opening of a nozzle; acquiring a video or image of the composition being continuously supplied from the nozzle opening; and further comprising at least one of the following steps: step A, which calculates at least one of the shape of the composition, the flow velocity of the composition, and the optical properties of the composition; and step B, which inspects the composition using a trained machine learning model that takes the video or image of the composition as input data.

[0006] Invention [2] is a method for inspecting a composition according to Invention [1], wherein step A is to calculate at least two of the following: the shape of the composition, the flow rate of the composition, and the optical properties of the composition. Invention [3] is a method for inspecting a composition according to Invention [1] or [2], wherein step A is to calculate the shape of the composition, the flow rate of the composition, and the optical properties of the composition. Invention [4] is a method for inspecting a composition according to any one of Inventions [1] to [3], wherein the flow rate of the composition is 3 to 50 mL / min. Invention [5] is a method for inspecting a composition according to any one of Inventions [1] to [4], further comprising a step of inspecting the viscosity of the composition using at least one of the shape of the composition, the flow rate of the composition, and the optical properties of the composition calculated in step A.

[0007] Invention [6] is a method for inspecting a composition according to any one of Inventions [1] to [5], wherein in step B, a step is to create a reconstructed image of the composition as output data using a trained machine learning model that takes a video or image of the composition as input data, and a step is to create difference data between the input data and the output data, and to inspect the viscosity of the composition using the difference data. Invention [7] is a method for inspecting a composition according to any one of Inventions [1] to [6], wherein the composition has a solid content concentration of 50% by mass or more. Invention [8] is a method for inspecting a composition according to any one of Inventions [1] to [7], wherein the composition contains magnetic particles.

[0008] Invention [9] is a composition inspection device comprising an imaging unit that captures a video or image of a composition being continuously delivered from the nozzle opening of a nozzle, and an inspection unit that inspects the physical properties of the composition from the video or image of the composition obtained by the imaging unit, wherein the inspection unit calculates at least one of the shape of the composition, the flow velocity of the composition, and the optical properties of the composition from the video or image of the composition, or inspects the composition using a trained machine learning model that takes the video or image of the composition as input data. Invention

[10] is the composition inspection device according to Invention [9], further comprising a light source unit that irradiates light from two directions onto a composition being continuously delivered from the nozzle opening of a nozzle. Invention

[11] is a method for manufacturing a composition, comprising the steps of obtaining a composition containing magnetic particles and inspecting the composition using the composition inspection method described in any one of Inventions [1] to [8]. Invention

[12] is a method for manufacturing a composition, comprising the steps of obtaining a composition containing magnetic particles and inspecting the composition using the composition inspection device described in Invention [8] or [9].

[0009] According to the present invention, it is possible to provide a method for inspecting compositions, an apparatus for inspecting compositions, and a method for manufacturing compositions that are suitable for non-contact inspection of compositions, in-line inspection, and inspection in a short time.

[0010] This is a schematic diagram showing an example of an inspection apparatus according to an embodiment of the present invention. This is a schematic diagram showing an example of a light source unit of an example of an inspection apparatus according to an embodiment of the present invention. This is a schematic diagram showing an example of a binarized image obtained by an example of an inspection apparatus according to an embodiment of the present invention. This is a graph showing an example of the width result of a composition obtained by an example of an inspection apparatus according to an embodiment of the present invention. This is a graph for explaining the method for calculating the flow velocity of a composition according to an embodiment of the present invention. This is a flowchart showing an example of the method for calculating the flow velocity of a composition according to an example of an inspection apparatus according to an embodiment of the present invention. This is a schematic diagram showing an example of averaging processing by an example of an inspection apparatus according to an embodiment of the present invention. This is a schematic diagram showing an example of an averaged image obtained by an example of an inspection apparatus according to an embodiment of the present invention. This is a graph showing an example of the flow velocity result of a composition obtained by an example of an inspection apparatus according to an embodiment of the present invention. This is a flowchart showing an example of how to determine the optical properties of a composition according to an example of an inspection apparatus according to an embodiment of the present invention. This is a schematic diagram showing an example of a binarized image obtained by an example of an inspection apparatus according to an embodiment of the present invention. This is a schematic diagram showing an example of an image after averaging processing of a moving image obtained by an example of an inspection apparatus according to an embodiment of the present invention. This is a graph showing an example of the optical properties of a composition obtained by an example of an inspection apparatus according to an embodiment of the present invention. This is a graph showing an example of the optical properties result of a composition obtained by an example of an inspection apparatus according to an embodiment of the present invention. This is a flowchart illustrating an example of a method for inspecting a composition using a trained machine learning model, based on an example of an inspection apparatus according to an embodiment of the present invention. The graph shows the frequency distribution of the difference between input data and output data. This is a schematic diagram illustrating an example of a composition manufacturing apparatus according to an embodiment of the present invention.

[0011] The following describes in detail a method for inspecting a composition, an apparatus for inspecting a composition, and a method for manufacturing a composition of the present invention, based on preferred embodiments shown in the attached drawings. The figures described below are illustrative for illustrating the present invention and have been simplified for illustrative purposes. Therefore, the present invention is not limited to the figures shown below. In the following, "~" indicating a numerical range includes the numerical values ​​indicated on both sides. For example, ε is the numerical value ε α ~ numerical value ε β Therefore, the range of ε is the numerical value ε α and the numerical value ε βThis range includes ε α ≦ε≦ε β The orthogonality includes the generally accepted tolerance in the relevant art unless otherwise specified. The length includes the generally accepted tolerance in the relevant art unless otherwise specified. The density includes the generally accepted tolerance in the relevant art unless otherwise specified.

[0012] [Inspection Apparatus for Composition] Figure 1 is a schematic diagram showing an example of an inspection apparatus according to an embodiment of the present invention. As shown in Figure 1, a tank 10 and a nozzle 12 are connected by piping 13, and a pump 14 and a valve 15 are provided on the piping 13 from the tank 10 side. A container 16 is also provided opposite the nozzle opening 12a of the nozzle 12. Composition Q is stored in the tank 10, and the pump 14 supplies composition Q to the nozzle 12, and the composition Q is continuously delivered from the nozzle opening 12a of the nozzle 12 and stored in the container 16. The pump 14 is the liquid delivery unit that continuously delivers composition Q from the nozzle opening 12a of the nozzle 12. The tank 10, nozzle 12, piping 13, pump 14 and valve 15 are used according to the composition Q. When composition Q is continuously delivered from the nozzle opening 12a of the nozzle 12, composition Q becomes a liquid column 17, and the inspection apparatus 20 inspects the composition Q in the liquid column 17 state. The inspection is performed not when composition Q is in droplet form, but after composition Q has been delivered and has become a liquid column 17.

[0013] The inspection device 20 will now be described in more detail. The inspection device 20 includes an imaging unit 21, an inspection unit 22, a control unit 24, a memory 25, and a display unit 26. The imaging unit 21 and the inspection unit 22 are connected, and the inspection unit 22 and the display unit 26 are connected. An input unit 27 is connected to the inspection device 20. Furthermore, the inspection device 20 includes a learning model creation unit 35.

[0014] The imaging unit 21 captures a video or image of composition Q as it is continuously delivered from the nozzle opening 12a of the nozzle 12. The imaging unit 21 obtains image data of the video or image of composition Q. The image data is represented by the pixel values ​​of each pixel representing the image. In the case of a color image, the pixel values ​​are represented by three signal values, for example, R (red), G (green), and B (blue). In the case of a binary image, the pixel values ​​are represented by image signal values ​​of 0 or 1. The image data of the video or image of composition Q obtained by the imaging unit 21 is output to the inspection unit 22. The imaging unit 21 captures the liquid column portion of composition Q. The imaging unit 21 is not particularly limited as long as it can obtain image data of the video or image of composition Q, but for example, it is configured to output a signal value proportional to the amount of light input. In this case, if the amount of light input to the imaging unit 21 is x, and the signal value output by the imaging unit 21 is y, then the relationship between the amount of light x and the signal value y is y = x γ It is known that it can be represented as follows. The imaging unit 21, for example, the above y = x γ The camera is configured such that the value of γ in the image sensor is 1. That is, the imaging unit 21 is configured such that it records a signal value that is linear with respect to the input light amount. The imaging unit 21 preferably uses a camera having a CMOS (Complementary Metal Oxide Semiconductor) image sensor, and preferably uses a telecentric lens. The imaging unit 21 is preferably capable of obtaining digital data as image data of a video or image of composition Q in order to facilitate image processing by the subsequent image processing unit 31.

[0015] Note that the imaging unit 21 is not particularly limited to being provided inside the inspection device 20 and may be provided outside the inspection device 20. In this case, the imaging unit 21 and the inspection unit 22 are connected by wire or wirelessly. Note that when the imaging unit 21 and the inspection unit 22 are separate bodies and physically separated from each other, the imaging unit 21 and the inspection unit 22 may be connected via a network such as the Internet. In the inspection device 20, the composition Q continuously fed from the nozzle opening 12a of the nozzle 12 may have a configuration having a light source unit (not shown) that irradiates light from one direction, but as shown in FIG. 2, it preferably has a light source unit 23 that irradiates light from two directions. By providing the light source unit 23 that irradiates light from two directions, the amount of light when the imaging unit 21 images can be increased compared to the case of irradiating light from one direction, and furthermore, the influence of ambient light depending on the shooting location can be reduced. An image including a specular reflection region and a diffuse reflection region can be obtained by the above-described light source unit 23, and the contrast of the moving image or image of the composition Q can be increased. Furthermore, by providing the light source unit 23 that irradiates light from two directions, the sampling range of the intensity of the diffused light can be defined as a minimum value within a region sandwiched between two maximum values caused by the specular reflection components of the light from two directions irradiated from the two light sources. Therefore, the sampling range of the intensity of the diffused light can be easily determined regardless of the size of the liquid column of the imaging target or the shooting distance. Further, it is more preferable that the light source unit 23 is provided with two light sources 23a and 23b at symmetric positions with respect to the imaging direction of the imaging unit 21. Thereby, an image including a specular reflection region, a diffuse reflection region, and a specular reflection region with the influence of ambient light further eliminated can be obtained. For the light sources 23a and 23b of the light source unit 23, for example, LEDs (Light Emitting Diodes) are used, but they are not particularly limited to LEDs.

[0016] The control unit 24 controls the operations of the respective components of the inspection device 20. The memory 25 stores data and the like obtained by the respective components of the inspection device 20 and is configured by, for example, a non-volatile memory capable of rewriting data.

[0017] The display unit 26 displays the inspection results. The display unit 26 can be, for example, a stationary liquid crystal display. Alternatively, the display screen of a mobile device such as a smartphone or tablet can be used. The display unit 26 is not particularly limited to being installed inside the inspection device 20, and may be installed outside the inspection device 20. In this case, the display unit 26 and the inspection unit 22 are connected by wire or wireless. If the display unit 26 and the inspection unit 22 are separate and physically separated from each other, the display unit 26 and the inspection unit 22 may be connected via a network such as the Internet. The input unit 27 is used for inputting instructions and data to operate the inspection device 20. For example, if the display unit 26 has a touch sensor function, a screen for input can be displayed on the display unit 26 and used as the input unit 27.

[0018] For example, the inspection device 20 may be configured by a computer in which each part shown in the inspection unit 22 functions when the control unit 24 executes a program stored in the memory 25, or may be a dedicated device or a dedicated terminal in which each part is configured by a dedicated circuit. For example, the inspection device 20 may be configured to be mounted on a mobile terminal such as a smartphone and a tablet terminal. Note that the inspection device 20 may be configured to be virtually provided on the cloud. In this case, the inspection device 20 is configured by a server so as to be executed on the cloud. In this configuration, for example, the inspection unit 22, the control unit 24, the memory 25, and the learning model creation unit 35 are virtually provided on the cloud. For example, the imaging unit 21 can use a camera provided in a mobile terminal such as a smartphone and a tablet terminal. Also, the display unit 26 can use a display (display screen) of a mobile terminal such as a smartphone and a tablet terminal. The mobile terminal can be used as a device for inputting data and the like to the inspection device 20 and displaying the determination result obtained by the inspection device 20. The imaging unit 21 and the display unit 26 are connected to the inspection unit 22 virtually provided on the cloud via a network such as the Internet. The above-described computer and dedicated device have, for example, a processor such as a CPU (Central Processing Unit). The processor may be constituted by one or more pieces of hardware, and the type of the hardware is not limited. For example, in addition to the CPU, the processor may be constituted by a programmable logic device such as an MPU (Micro Processing Unit), an FPGA (Field Programmable Gate Array), a dedicated circuit for executing specific processing such as an ASIC (Application Specific Integrated Circuit), a GPU (Graphic Processing Unit), or an NPU (Neural Processing Unit) and the like of hardware.

[0019] The inspection unit 22 includes, for example, a storage unit 30, an image processing unit 31, and an analysis unit 32. The storage unit 30 stores video or image data of composition Q acquired by the imaging unit 21. The storage unit 30, like the memory 25 described above, is composed of, for example, a non-volatile memory that allows data to be rewritten.

[0020] Preferably, the storage unit 30 has a conversion unit (not shown) that converts the video or image data of composition Q acquired by the imaging unit 21 into digital data when the image data is in analog data. The method of converting analog image data into digital data in the conversion unit is not particularly limited, and various known methods can be used.

[0021] The inspection unit 22 of the inspection device 20 calculates at least one of the following from a video or image of the composition: the shape of the composition, the flow velocity of the composition, and the optical properties of the composition, or inspects the composition using a trained machine learning model that takes a video or image of the composition as input data. The image processing unit 31 has, for example, a function to perform image processing in accordance with the calculation of the shape of the composition, the flow velocity of the composition, or the optical properties of the composition as described above, and a function to perform image processing in accordance with the inspection of the composition using the trained machine learning model. The analysis unit 32 has, for example, a function to calculate the shape of the composition, the flow velocity of the composition, or the optical properties of the composition as described above, and a function to perform analysis in accordance with the inspection of the composition using the trained machine learning model, on the image obtained by the image processing of the image processing unit 31. The calculation of the shape of the composition, the flow velocity of the composition, or the optical properties of the composition will be described below. The shape of the composition, the flow velocity of the composition, or the optical properties of the composition are calculated by the image processing unit 31 and the analysis unit 32 of the inspection unit 22. Hereinafter, unless otherwise specified, the image data is digital data that can be processed by the image processing unit 31. The various types of data processed by the analysis unit 32 are also digital data unless otherwise specified. For example, image data consists of R data representing red, G data representing green, and B data representing blue. R data, G data, and B data are collectively called RGB data.

[0022] (Shape of the Composition) Hereinafter, the calculation of the shape of the composition will be described by taking the width of the composition as an example of the shape of the composition. When calculating the shape of the composition, for example, the image processing unit 31 performs color tone correction processing, trimming processing, and binarization processing on the image data of the video or image of the composition Q in this order. The color tone correction processing is a process of multiplying coefficients for each of the R data, G data, and B data so that the average value of the RGB data of the entire image is equal in the image data, or when the image data is 8-bit gradation (256 gradations), the average value of the data representing each color of the RGB data becomes 128. The color tone correction is appropriately to suppress the color tone change depending on the environment at the time of shooting. The trimming processing is, for example, setting a trimming range in advance with respect to the angle of view of the imaging unit 21 that images the above-described video or image of the composition. Based on the above range, the image data is trimmed to obtain a trimmed image. The binarization processing is, for example, setting a binarization threshold value in advance based on the imaging conditions of the imaging unit 21, and performing binarization processing on the above-described trimmed image based on the threshold value to obtain, for example, a binarized image 40 of the nozzle and the composition shown in FIG. 3. The image data of the binarized image 40 is output to the analysis unit 32.

[0023] The analysis unit 32 identifies a nozzle portion 41 corresponding to the nozzle 12 shown in FIG. 1 and a nozzle opening portion 41a corresponding to the nozzle opening 12a in the binarized image 40 shown in FIG. Further, a composition portion Q corresponding to the composition Q shown in FIG. 1 １ (see FIG. 3) is identified, and a liquid column portion 42 (see FIG. 3) corresponding to the liquid column 17 is identified. For example, the nozzle opening portion 41a shown in FIG. 3 is used as a reference for the length of the composition portion Q １ In this case, the nozzle opening portion 41a is set at the position of 0 mm. The length of the portion corresponding to the width of the composition portion Q １ can be obtained from the number of pixels of the portion corresponding to the width of the composition portion Q １ Along the length of the composition portion Q, that is, along the y direction shown in FIG. 3, in the x direction orthogonal to the y direction, the composition portion Q １ １The length of the portion corresponding to the width is determined. In this way, the width of composition Q is obtained in the analysis unit 32. An example of the result for the width of the composition is shown in Figure 4. Figure 4 shows the results for compositions A to E from different lots when attempting to manufacture compositions with the same composition using a predetermined manufacturing procedure. Normally, in the manufacture of a composition, even if the same procedure is followed, the properties of the final composition may change due to differences in the lot of raw materials used, or the manufacturing environment such as humidity and temperature. Here, when a viscoelastic liquid flows out of a capillary tube, there is a Dicewell effect in which the liquid changes from the inner diameter of the tube at the outlet of the tube. Due to this Dicewell effect, it is known that the diameter increases as the viscosity increases. From this, the viscosity of composition Q can be inspected by the inspection device 20. Furthermore, it is also possible to evaluate the viscosity of a composition using the width of the composition as a shape of the composition.

[0024] When calculating the shape of a composition, image data of multiple images of composition Q at specified times may be acquired, or image data of multiple images of composition Q at specified times may be acquired from the image data of a moving image of composition Q. When calculating the shape of a composition, it is preferable to acquire image data of a moving image of composition Q from the viewpoint of robustness. The image data of the moving image is acquired at a preset frame rate. As an example of the shape of a composition, the width of the composition has been used in the explanation, but it is not particularly limited to this, and information quantifying the shape change that occurs depending on the viscosity of the composition may also be used. Examples of the shape of a composition may be the magnitude of the time variation of the width of the composition, the periodicity of the time variation of the width of the composition, or the frequency of disconnection of the leaked composition material.

[0025] (Flow velocity of the composition) Figure 5 is a graph illustrating the method for calculating the flow velocity of the composition according to an embodiment of the present invention. The signal profiles 45a to 45f shown in Figure 5 are obtained by averaging the signal values ​​of the liquid column portion in the width direction of the liquid column portion in the images at each time t to t+5. The width direction of the liquid column portion corresponds to the x direction in Figure 3. For the averaging described above, for example, the arithmetic mean is used. The vertical axis in Figure 5 shows the signal values ​​of the liquid column portion, and the horizontal axis shows the position of the liquid column. The rightward direction on the horizontal axis indicates the direction in which the composition is moving. Note that the signal values ​​of the liquid column portion in the image are the pixel values ​​of the pixels representing the liquid column portion in the image data. As shown in the signal profiles 45a to 45f, a characteristic signal peak P １ ~P ４ As time changes, the horizontal axis moves to the right. This can be used to determine the flow velocity of the composition. For example, when using the signal profiles 45a to 45f shown in Figure 5, first, the signal profile 45a at time t and the signal profile 45b at time t+1 are compared, and the signal profile 45a is shifted in the direction that increases the position value of the liquid column on the horizontal axis of Figure 5 to find the shift amount that best matches the signal profile 45b. This shift amount is the correlation distance. The correlation distance is the distance the composition has moved in the time interval from time t to time t+1. The shift amount can be expressed, for example, in terms of the number of pixels in the image. By converting the number of pixels representing the shift amount into the actual distance in the liquid column, the actual distance the composition has moved in the time interval from time t to time t+1 can be determined. For example, the above-mentioned agreement of signal profiles can be determined using known pattern matching methods such as template matching.

[0026] Next, the signal profile 45b at time t+1 and the signal profile 45c at time t+2 are compared, and the signal profile 45b is shifted in the direction that increases the position value of the liquid column on the horizontal axis of Figure 5 to find the shift amount that best matches the signal profile 45c. This shift amount is the correlation distance. The correlation distance is the distance the composition has moved in the time interval from time t+1 to time t+2. By converting the number of pixels representing the shift amount to the actual distance in the liquid column, the actual distance the composition has moved in the time interval from time t+1 to time t+2 is determined. In this way, the correlation distance is calculated between the data for all time points and the data for the next time point for each of those time points, and the average correlation distance is calculated by taking the arithmetic mean. The average time interval between the time for which the correlation distance was calculated and the time interval for the next time point is calculated. The average flow velocity of the composition is calculated by (average correlation distance) / (average time interval). The method for calculating the flow velocity of the composition will be explained in more detail below.

[0027] Figure 6 is a flowchart showing an example of a method for calculating the flow velocity of a composition according to an example of an inspection apparatus according to an embodiment of the present invention. Figure 7 is a schematic diagram showing an example of averaging processing according to an example of an inspection apparatus according to an embodiment of the present invention. Figure 8 is a schematic diagram showing an example of an averaged image according to an example of an inspection apparatus according to an embodiment of the present invention. When calculating the flow velocity of a composition, first, for example, as described above, the imaging unit 21 acquires video or still image data of composition Q (step S10). In the method for calculating the flow velocity of a composition, averaging processing is performed on the image at each time (step S12). Therefore, in step S10, image data of multiple images of composition Q at a specified time may be acquired, or image data of multiple images of composition Q at a specified time may be acquired from video image data of composition Q. In step S10, from the viewpoint of robustness, it is preferable to acquire video image data of composition Q. The image data acquired in step S10 is output to the image processing unit 31.

[0028] Next, the image processing unit 31 performs an averaging process on the signal values ​​(image data) of the images for the liquid column portions of multiple images of composition Q whose timestamps were identified in step S10, or for the liquid column portions of multiple images of composition Q whose timestamps were identified from the image data of a moving image of composition Q (step S12). In step S12, the liquid column portions of the images at each timestamp are identified, and an averaging process is performed on the images of the liquid column portions at each timestamp, averaging the signal values ​​(image data) of the liquid column portions in the width direction of the liquid column portions, as shown in Figure 7. This yields an averaged signal (data signal) of the liquid column portions, for example, the signal profile 46 shown in Figure 7. Step S12 yields an averaged signal (data signal) of the liquid column portions at each timestamp. More specifically, the averaging process in step S12 is a process of taking an arithmetic mean, for example, of the signal values ​​(image data) of the liquid column portions of the images in the width direction of the liquid column portions. The averaged signal (data signal) obtained by the averaging process, which is obtained by averaging the signal values ​​(image data) of the liquid column portions in the width direction of the liquid column portions, represents, for example, the signal profile 46. When N is the number of pixels in the liquid column portion of the image, the averaged signal (data signal) consists of 1 × N data points. The signal profile is represented by 1 × N data points. The signal profiles 45a to 45f shown in Figure 5 above were obtained by the same averaging process as signal profile 46.

[0029] Next, the image processing unit 31 creates an image using the averaged signal representing the signal profile 46 and performs a concatenation process to obtain the averaged image 50 shown in Figure 8 (step S14). The concatenation process described above is a process that creates an averaged image by arranging the signal profiles 45a to 45f from each time t to t+5 in chronological order. More specifically, the concatenation process arranges 1 × N data representing the signal profiles 45a to 45f in chronological order of the acquired images and creates an image to obtain the averaged image 50. From time t to time t ＭIf the number of acquired images up to this point is M, the averaged image 50 is represented by averaged image data with a pixel count of N × M (N > M). In the averaged image 50, the number of pixels on side 50a is N, and the number of pixels on side 50b is M. The averaged image 50 has a rectangular outline, for example, as shown in Figure 8. The averaged image 50 has four vertices: a first vertex 51a, a second vertex 51b, a third vertex 51c, and a fourth vertex 51d, and four edges. The edge 50a connecting the first vertex 51a and the second vertex 51b is the same as the edge connecting the third vertex 51c and the fourth vertex 51d. The edge 50b connecting the second vertex 51b and the third vertex 51c is the same as the edge connecting the first vertex 51a and the fourth vertex 51d.

[0030] In the averaged image 50, among the first vertex 51a, the second vertex 51b, the third vertex 51c, and the fourth vertex 51d, the direction D from the first vertex 51a toward the second vertex 51b. １ This is the direction of extension of the liquid column. The number of pixels from the first vertex 51a to the second vertex 51b is N. The side of the first vertex 51a is the side of the nozzle opening 12a of the nozzle 12 shown in Figure 7. The position on the edge 50a of the averaged image 50 corresponds to the distance from the nozzle opening of the nozzle, and the distance from the nozzle opening of the nozzle can be determined by the position on the edge 50a. Direction D from the second vertex 51b to the third vertex 51c. ２ The graph shows the time progression of image acquisition, where M is the number of pixels from the second vertex 51b to the third vertex 51c. The second vertex 51b is the position at the first time t when the image was acquired, and the third vertex 51c is the position at the last time t when the image was acquired. Ｍ This is the position. The position on edge 50b of the averaged image 50 corresponds to the time the image was acquired, and the time the image was acquired can be determined by the position on edge 50b. The averaged image 50 contains the signal profiles (data signals) of all acquired images.

[0031] Next, the image processing unit 31 performs correlation analysis using the averaged image data representing the averaged image 50 (see Figure 8) obtained by the concatenation process (step S16). In the correlation analysis in step S16, the displacement distance at which the correlation is maximized for each distance from the nozzle opening of the nozzle in the averaged image data of the averaged image is determined. The displacement distance is expressed in terms of the number of pixels in the averaged image. The correlation analysis will be explained in more detail below, but although the correlation analysis is performed on the averaged image data, the averaged image 50 shown in Figure 8 will be used for visualization purposes. In the correlation analysis, first, as shown in Figure 8, a rectangular region 52 is set in the averaged image 50, for example. The region 52 has side 52a in direction D. １ It extends and has a length of M １ Therefore, side 52b is in direction D. ２ It extends to and has a length of M ２ That is. M １ < M ２ The length of side 52a of region 52 is M. １ This is a predetermined length from the nozzle opening of the nozzle. The side 52b of region 52 is from time t to time t Ｍ－１ This is within the range of the last time t Ｍ It does not include.

[0032] Next, the vertices 53 of region 52 are positioned to coincide with, for example, point β on edge 50a of the averaged image 50, i.e., the first vertex 51a. The position of this region 52 is set to α １ Next, in the averaged image 50, the region 52 is located at position α. １ Identify the location where the correlation is maximized, other than the specified location. To identify the location where the correlation is maximized, known pattern matching methods such as template matching are used. For example, in region 52, vertex 53 is located in direction D from the first vertex 51a (point β). １ Distance δ, direction D ２ time t １ The best match was found at this position. As a result, region 52 is from time t to time t １ It is determined that a distance δ has been moved during the time interval up to time t. At this time, for example, the edge 52b of region 52 is １ ~Time t ＭThis is the range. Distance δ is the distance traveled where the correlation in region 52 is maximized. Distance δ is in direction D １ This is determined by counting the number of pixels. The number of pixels at distance δ is the number of pixels at which the correlation of region 52 is maximized, and the number of pixels at distance δ is output to the analysis unit 32.

[0033] Also, for example, by changing the distance of the nozzle from the nozzle opening, that is, point β on the side 50a of the averaged image 50 １ The vertices 53 of region 52 are aligned and placed on point β. １ The position of region 52 is α ２ Next, in the averaged image 50, the region 52 is located at position α. ２ Identify the location where the correlation is maximized, other than the specified location. To identify the location where the correlation is maximized, known pattern matching methods such as template matching, as described above, are used. For example, in region 52, vertex 53 is at point β １ From direction D １ Distance δ, direction D ２ time t １ The best match was found at this position. As a result, region 52 is from time t to time t １ It is determined that a distance δ has been moved during the time interval up to time t. At this time, for example, the edge 52b of region 52 is １ ~Time t Ｍ This is within the specified range. As described above, the number of pixels at distance δ, which is the distance traveled that maximizes the correlation of region 52, is determined. The number of pixels at distance δ is output to the analysis unit 32 as the number of pixels at which the correlation of region 52 maximizes. In this way, the position of region 52 on the edge 50a of the averaged image 50 is changed, and the location at which the correlation of region 52 maximizes is repeatedly identified for each distance from the nozzle opening of the nozzle, and the number of pixels at distance δ, which is the distance traveled that maximizes the correlation, is determined for each distance from the nozzle opening of the nozzle and output to the analysis unit 32.

[0034] Next, the analysis unit 32 determines the flow velocity of composition Q from the number of pixels that maximize the correlation obtained for each distance from the nozzle opening in the correlation analysis of step S16, the image resolution, and the frame rate of the moving image (step S18). The unit of image resolution is preferably ppi (pixels per inch). For example, the actual distance traveled in the liquid column portion is determined from the number of pixels that maximize the correlation obtained for each distance from the nozzle opening in the correlation analysis, and the image resolution. From this actual distance traveled in the liquid column portion and the actual time interval at each time obtained from the frame rate, the distance traveled per unit time, i.e., the flow velocity of composition Q, is obtained. When the flow velocity of the composition is Vf, Vf = (actual distance traveled in the liquid column portion obtained from distance δ) / (actual time interval). The flow velocity of the composition can be determined using this formula for the flow velocity Vf of the composition. As described above, the velocity of the composition is, for example, a characteristic signal peak P １ ~P ４ Instead of tracking and calculating the position of the liquid column at each time point (see Figure 5), the velocity of the composition is calculated by using two-dimensional information, i.e., an averaged image 50 (see Figure 8), which represents the signal value of the liquid column's position at each time point, to determine the average displacement at each liquid column's position. Therefore, the velocity of the composition can be calculated stably.

[0035] As described above, the flow velocity of the composition can be obtained for each distance from the nozzle opening by determining the number of pixels at distance δ, which is the distance at which the correlation is maximized. An example of the flow velocity results of the composition is shown in Figure 9. Figure 9 shows the results for compositions A to E from different lots when attempting to manufacture compositions with the same composition using a predetermined manufacturing procedure, similar to Figure 4 described above. Normally, in the manufacture of a composition, even when the same procedure is followed, the characteristics of the final composition may change due to differences in the lot of raw materials used, manufacturing environment such as humidity and temperature, etc. As shown in Figure 9, compositions A to E from different lots when attempting to manufacture compositions with the same composition using a predetermined manufacturing procedure may have different flow velocities. Here, it is known that when the viscosity of a composition is high, the flow velocity of the composition decreases. From this, the viscosity of the composition can be inspected using the inspection device 20. Furthermore, it is also possible to evaluate the viscosity of composition Q using the flow velocity of the composition.

[0036] (Optical Properties of the Composition) Figure 10 is a flowchart showing an example of how to determine the optical properties of a composition using an example of an inspection apparatus according to an embodiment of the present invention. The gloss value of the composition will be used as an example to explain the optical properties of the composition. When calculating the optical properties of a composition, first, for example, as described above, the imaging unit 21 acquires video or still image data of composition Q (step S20). The acquisition of video or still image data of composition Q in step S20, and the image data acquired in step S20, are the same as in step S10 described above, so a detailed explanation will be omitted. In calculating the optical properties of a composition, an averaging process of the signal values ​​of the video (step S26) is performed, and from the viewpoint of robustness, it is preferable to acquire video image data of composition Q in step S20. When the optical property is the gloss value, it is preferable to acquire video image data of composition Q in step S20 because the specular reflection component information and diffuse reflection component information necessary for calculating the gloss value can be acquired in a single image. Furthermore, it is more preferable to irradiate the composition with light from two directions in step S20 and acquire video image data of composition Q. The image data acquired in step S20 is output to the image processing unit 31.

[0037] Next, the image processing unit 31 performs a first trimming process, a binarization process, and a second trimming process on the video or image data of composition Q acquired in step S20, in this order (step S22). Step S22 yields the binarized image 40a shown in Figure 11. The binarized image 40a is subjected to a second trimming process to obtain an image 47 of the liquid column portion. The first trimming process involves, for example, setting a trimming range in advance with respect to the field of view of the imaging unit 21 that captures the video or image of the composition described above. Based on the range described above, the image data is trimmed to obtain the first trimmed image. The binarization process involves, for example, setting a binarization threshold in advance based on the imaging conditions of the imaging unit 21, and performing a binarization process on the first trimmed image based on the threshold to obtain, for example, the binarized image 40a shown in Figure 11. The second trimming process involves, for example, setting a trimming range in advance for the binarized image 40a. Based on the above range, the image data is cropped to obtain image 47 of the liquid column portion. Image 47 of the liquid column portion is a binary image.

[0038] Next, the image processing unit 31 enlarges or reduces the image 47 of the liquid column portion to a preset pixel size (step S24). Step S24 yields an image of the liquid column portion with a preset pixel size (not shown). The enlargement or reduction of the image 47 in step S24 can be performed using a known method. Next, the image processing unit 31 performs an averaging process of the signal values ​​of the moving image on the image of the liquid column portion with a preset pixel size (step S26). Step S26 yields an image 48, shown in Figure 12, in which the signal values ​​of the moving image are averaged within the range of the liquid column image, for example, by arithmetic mean processing. Image 48 is the image after the averaging process of the moving image. The image data of image 48 (see Figure 12), in which the signal values ​​of the moving image have been averaged, is output to the analysis unit 32. The image data of image 48 shown in Figure 12 consists of the average value of the signal values ​​of the moving image. This average value is, for example, an arithmetic mean. The averaging process of the video signal values ​​in step S26 is more specifically a process of obtaining the average value of the video signal values ​​by calculating the integrated value of the signal values ​​of each pixel that make up the image of the liquid column, and then dividing the integrated value of the signal values ​​of each pixel by the frame rate of the video. The average value of the video signal values ​​is, for example, the arithmetic mean of the video signal values.

[0039] Next, the analysis unit 32 determines the ratio of the highest signal value to the lowest signal value in the signal values ​​of image 48 shown in Figure 12 (step S28). The ratio of the highest signal value to the lowest signal value corresponds to the gloss value among the optical properties of the composition. Image 48 shown in Figure 12 is represented, for example, by the signal profile 49 shown in Figure 13. The specular reflection region 48a of image 48 shown in Figure 12 corresponds to the peak 49a of the signal profile 49 shown in Figure 13. The diffuse reflection region 48b of image 48 shown in Figure 12 corresponds to the bottom 49b of the signal profile 49 shown in Figure 13. The specular reflection region 48c of image 48 shown in Figure 12 corresponds to the peak 49c of the signal profile 49 shown in Figure 13. Peaks 49a and 49c of the signal profile 49 are local maxima, and the bottom 49b is a local minimum. The signal profile 49 has two local maxima, and the local minimum is sandwiched between the two local maxima. This indicates that the diffuse reflection region 48b is sandwiched between two specular reflection regions 48a and 48c. Therefore, by identifying the bottom 49b of the signal profile 49, the position of the diffuse reflection region 48b can be easily determined regardless of the size of the liquid column being photographed or the shooting distance.

[0040] As described above, a gloss value is obtained as an optical property of composition Q. An example of the gloss value result among the optical properties of a composition is shown in Figure 14. Figure 14 shows the results for compositions A to E from different lots when attempting to manufacture compositions with the same composition using a predetermined manufacturing procedure, similar to Figure 4 described above. Normally, in the manufacture of a composition, even when the same procedure is followed, the properties of the final composition may change due to differences in the lot of raw materials used, the manufacturing environment such as humidity and temperature, etc. As shown in Figure 14, compositions A to E from different lots when attempting to manufacture compositions with the same composition using a predetermined manufacturing procedure may have different gloss values, that is, the optical properties of the compositions may differ. Here, among the optical properties, the gloss value is due to the surface irregularities of the composition, and the gloss value decreases when the surface irregularities of the composition are large. The surface irregularities of the composition depend, for example, on the dispersion state of the particles contained in the composition. Different gloss values ​​indicate that there are variations in the dispersion state of the particles contained in the composition. From this, the inspection device 20 can inspect the viscosity of the composition using the gloss value as an optical property. Furthermore, it is also possible to evaluate the viscosity of the composition using the optical properties of the composition, for example, the gloss value.

[0041] While the gloss value of a composition was used as an example to describe its optical properties, the explanation is not limited to this. Here, the intensity of light reflected in each direction changes depending on the roughness of the material or surface. By measuring specular reflection (the angle at which the reflected light is strongest) and other angles, differences in material properties such as the roughness of the material or surface can be detected. However, the more angles measured, the more accurately differences in material properties such as the roughness of the material or surface can be detected. For this reason, in addition to the gloss value of the composition, the intensity changes of reflected light at two or more locations may also be used as an optical property of the composition. By examining the magnitude of the intensity changes of reflected light at two or more locations, the degree of diffusion can be examined, the roughness of the material or surface can be inspected, and differences in material properties can be detected.

[0042] (Inspection of Composition) Next, we will describe the inspection of the composition other than the shape, flow rate, and optical properties of the composition described above. The image processing unit 31 has the function of performing image processing in accordance with the inspection of the composition using the trained machine learning model as described above. For example, the image processing unit 31 trims the video or image data of composition Q to the same size as the training data used to train the machine learning model. The analysis unit 32 has the function of performing analysis in accordance with the inspection of the composition using the trained machine learning model as described above. For example, an autoencoder (AE) can be used as the machine learning model. An autoencoder is a type of neural network and basically aims to output data that matches the input data. The autoencoder creates a reconstructed image from the input image that is as close as possible to the input image. In addition to an autoencoder (AE), a VAE (Variational Auto Encoder) or a GAN (Generative Adversarial Networks) can also be used as the machine learning model. A trained machine learning model is a machine learning model that has been trained using videos or images of compositions as training data. Specifically, these include trained autoencoders (AEs), trained video-encoders (VAs), and trained genetically engineered networks (GANs).

[0043] The training data is appropriately set according to the machine learning model. In the case of an autoencoder, for example, image data of a video or image of a composition having set physical properties is used as training data. The set physical properties are, for example, the viscosity of the composition. In this case, for example, image data of a video or image of a composition whose viscosity has been specified in advance is used as training data. Although not described in detail, VAEs and GANs can also be trained using image data of videos or images of compositions according to each machine learning model as training data. The same training data as the autoencoder described above can be used for VAEs and GANs. The trained machine learning model may be created outside the inspection device 20 and stored in the memory 25 from the input unit 27. Alternatively, the trained machine learning model may be created, for example, in the training model creation unit 35 described later and stored in the memory 25. When using the trained machine learning model, the analysis unit 32 reads the trained machine learning model from the memory 25.

[0044] The learning model creation unit 35 creates the trained machine learning model described above. The learning model creation unit 35 is connected to the memory 25. The learning model creation unit 35 may also be connected to the imaging unit 21 or the image processing unit 31. For example, the learning model creation unit 35 inputs video or image data of a composition whose viscosity has been predetermined as described above as training data into the machine learning model and trains it to create a trained machine learning model. Video or image data of a composition captured by the imaging unit 21 can be used as training data. Alternatively, the data obtained after image processing of video or image data of a composition captured by the imaging unit 21 by the image processing unit 31 can also be used as training data.

[0045] Furthermore, the trained machine learning model may be further trained or retrained. For further training or retraining, for example, data other than the video or image of the composition with a predetermined viscosity used as training data when training the machine learning model may be used as training data. In this case, for example, video or image of a composition with the same viscosity as the training data may be used as training data for further training or retraining. Such further training or retraining can improve the performance of the trained machine learning model and improve the accuracy of composition inspection. The learning model creation unit 35 may be configured to enable the above-mentioned further training or retraining. If the learning model creation unit 35 does not create a trained machine learning model, the learning model creation unit 35 is not necessarily required. The trained machine learning model can also be stored in the memory 25 in advance. The analysis unit 32 may read the trained machine learning model from the memory 25 and perform composition inspection.

[0046] Figure 15 is a flowchart illustrating an example of a method for inspecting a composition using a trained machine learning model according to an example of an inspection apparatus of an embodiment of the present invention. A trained autoencoder will be used as an example of a trained machine learning model. For the flow velocity of the composition, for example, as described above, the imaging unit 21 acquires video or image data of composition Q (step S30). The acquisition of video or image data of composition Q in step S30 is the same process as in step S10 described above, so a detailed explanation will be omitted. The image data acquired in step S30 is output to the image processing unit 31.

[0047] Next, the image processing unit 31 performs a trimming process on the video or image data of composition Q to create input data (step S32). In step S32, if the video image data of composition Q is used, each frame image, i.e., a still image, is trimmed to the same size as the training data used to train the machine learning model. This provides input image data of composition Q as input data. If the image data of an image of composition Q is a still image, it is trimmed directly to the same size as the training data used to train the machine learning model. This provides input image data of composition Q as input data. The input image data of composition Q created in step S32 is output to the image processing unit 31 as input data.

[0048] Next, the analysis unit 32 takes the input image data of composition Q created in step S32 as input data and inputs it to a trained machine learning model, for example, a trained autoencoder, to create a reconstructed image of the composition as output data (step S34). Furthermore, the analysis unit 32 calculates the difference between the input data and the output data and creates difference data (step S36). A difference image is obtained from the difference data created in step S36. The difference image is obtained by calculating the difference in pixel values ​​of corresponding pixels in the input image data (input data) and the image data of the reconstructed image (output data), and visualizing the difference in pixel values ​​of corresponding pixels. The difference image allows for the identification of the differences between the input data and the output data. The smaller the difference between the input data and the output data, the closer the input data is to the trained data.

[0049] Here, Figure 16 shows a graph illustrating the frequency distribution of the difference between the input data and the output data. In Figure 16, the vertical axis represents frequency, and the horizontal axis represents the difference value. The difference value on the horizontal axis indicates the amount of deviation of the output data from the input data. A larger difference value indicates a greater difference between the input data and the output data (reconstructed image), that is, that the input data is different from the training data. Figure 16 shows the difference between the input data and the training data. Figure 16 shows the results for compositions A to D from different lots when attempting to manufacture compositions with the same composition using a predetermined manufacturing procedure, similar to Figure 4 above. Normally, in the manufacture of a composition, even when the same procedure is followed, the characteristics of the final composition may change due to differences in the lot of raw materials used, manufacturing environment such as humidity and temperature, etc. As shown in Figure 16, compositions A to D from different lots when attempting to manufacture compositions with the same composition using a predetermined manufacturing procedure may differ from the output data (reconstructed image). Composition C uses the training data as input data. As shown in Figure 16, composition C has a small difference value from the output data (reconstructed image). This indicates that composition C has a small difference from the output data (reconstructed image). Composition D has a large difference value from the output data (reconstructed image). This indicates that composition D is significantly different from the output data (reconstructed image). Therefore, it is possible to inspect the composition based on the difference data. If, for example, image data of a composition having set physical properties is used as training data, the difference data reflects the set physical properties, so it is possible to use the difference data to check whether or not the composition has the set physical properties. Also, if, for example, image data of a composition with a predetermined viscosity is used as training data, the difference data reflects the viscosity, so it is possible to use the difference data to check the viscosity of the composition.

[0050] In the inspection device 20, the inspection unit 22 either calculates at least one of the following from a video or image of the composition: the shape of the composition, the flow velocity of the composition, and the optical properties of the composition, or inspects the composition using a trained machine learning model that takes a video or image of the composition as input data. Therefore, the inspection unit 22 only needs to have one of the following functions: the function to calculate at least one of the shape of the composition, the flow velocity of the composition, and the optical properties of the composition, or the function to inspect the composition using a trained machine learning model. Alternatively, the inspection unit 22 may have both the function to calculate at least one of the shape of the composition, the flow velocity of the composition, and the optical properties of the composition, and the function to inspect the composition using a trained machine learning model, and the ability to select either function. In this case, for example, the input unit 27 may be configured to allow selection of either the function to calculate at least one of the shape of the composition, the flow velocity of the composition, and the optical properties of the composition, or the function to inspect the composition using a trained machine learning model. Furthermore, since the inspection unit 22 has the function of calculating at least one of the following: the shape of the composition, the flow velocity of the composition, and the optical properties of the composition, it may calculate one, two, or three of these. As described above, the composition inspection device 20 inspects the liquid column portion of composition Q as it is continuously delivered from the nozzle opening 12a of the nozzle 12, and can inspect the composition non-contact and in-line as composition Q is stored in the container 16. Moreover, since the composition is not removed during inspection, the composition can be inspected in a short time.

[0051] [Method for Inspecting the Composition] The method for inspecting the composition will now be described. For example, the inspection device 20 shown in Figure 1 above is used for the method of inspecting the composition, but the method for inspecting the composition is not particularly limited to using the inspection device 20 shown in Figure 1. The method for inspecting the composition includes the steps of continuously supplying composition Q from the nozzle opening 12a of the nozzle 12, and acquiring a video or image of the composition Q being continuously supplied from the nozzle opening 12a, and further includes at least one of the steps of calculating at least one of the shape of the composition, the flow velocity of the composition, and the optical properties of the composition, and inspecting the composition using a trained machine learning model that takes the video or image of the composition as input data. In the step of continuously supplying composition Q from the nozzle opening 12a of the nozzle 12, for example, composition Q is stored in the tank 10 shown in Figure 1, and the valve 15 is opened and composition Q is supplied to the nozzle 12 by the pump 14. As a result, composition Q is continuously supplied from the nozzle opening 12a of the nozzle 12 to the container 16. When composition Q is continuously delivered from the nozzle opening 12a of nozzle 12, composition Q becomes a liquid column 17, and in this state, the liquid column portion of composition Q is inspected. In this method of inspecting the composition, the composition is inspected in the state of a liquid column 17, not in the state of droplets. In the next step of acquiring a video or image of composition Q being continuously delivered from the nozzle opening 12a, the imaging unit 21 shown in Figure 1 is used to image the liquid column portion of composition Q, thereby acquiring a video or image of composition Q. In the image acquisition step described above, it is preferable to irradiate composition Q with light from two directions using the light source unit 23 shown in Figure 2. This makes it possible to obtain an image with high contrast.

[0052] In step A, which calculates at least one of the composition shape, composition velocity, and composition optical properties, the composition shape is calculated as described above, and for example, the width of composition Q shown in Figure 4 is obtained as described above. For this reason, a detailed explanation of the calculation of the composition shape is omitted. The composition velocity can be calculated as shown in Figure 6 above, and for example, the flow velocity of composition Q shown in Figure 9 is obtained as described above. For this reason, a detailed explanation of the calculation of the composition velocity is omitted. The composition optical properties can be calculated as shown in Figure 10 above, and for example, the gloss value of composition Q shown in Figure 14 is obtained as described above. For this reason, a detailed explanation of the calculation of the composition optical properties is omitted. The composition inspection method may include a step of inspecting the viscosity of the composition using at least one of the composition shape, composition velocity, and composition optical properties calculated in step A. The viscosity of composition Q can be inspected by utilizing the fact that the diameter increases as the viscosity increases, as described above. The viscosity of composition Q can be tested by utilizing the fact that, as mentioned above, the flow velocity of a composition decreases when its viscosity is high. Regarding the optical properties of a composition, as mentioned above, the gloss value is due to surface irregularities; a larger irregularity results in a smaller gloss value. Different gloss values ​​indicate variations in the dispersion state of the particles contained in the composition. Therefore, the viscosity of a composition can be tested using the gloss value.

[0053] Step B, which involves inspecting a composition using a trained machine learning model that takes a video or image of the composition as input data, uses, for example, an autoencoder (AE) as described above in the machine learning model. In Step B, which involves inspecting a composition using a trained machine learning model, the composition can be inspected as shown in Figure 15 above, and for example, the frequency distribution of the difference between the input data and the output data (reconstructed image) shown in Figure 16 above can be obtained. For this reason, a detailed explanation of the composition inspection using the machine learning model is omitted. The composition inspection method may include, in Step B, a step of creating a reconstructed image of the composition as output data using a trained machine learning model that takes a video or image of the composition as input data, and a step of creating difference data between the input data and the output data, and inspecting the viscosity of the composition using the difference data. The steps of creating a reconstructed image of the composition as output data and creating difference data between the input data and the output data are as described above, so a detailed explanation is omitted. As described above, if, for example, image data of a composition having set physical properties is used as training data, the difference data reflects the set physical properties, so it is possible to use the difference data to check whether or not the composition has the set physical properties. Also, as described above, if, for example, image data of a composition whose viscosity has been predetermined is used as training data, the difference data reflects the viscosity, so it is possible to use the difference data to check the viscosity of the composition. The method for inspecting the composition inspects the liquid column portion of composition Q as it is continuously delivered from the nozzle opening 12a of the nozzle 12, as described above, and can inspect the composition non-contact and in-line as composition Q is stored in the container 16. Moreover, since the composition is not removed during inspection, the composition can be inspected in a short time.

[0054] In the method for testing the composition, the flow rate of the composition is preferably 3 to 50 mL / min, and more preferably 5 to 15 mL / min. The flow rate of the composition is adjusted, for example, by the flow rate of the pump 14 shown in Figure 1. Furthermore, the shape of the nozzle opening 12a from which the composition Q is discharged is preferably circular. When the shape of the nozzle opening 12a is circular, the liquid column becomes cylindrical. Here, when obtaining specular and diffuse reflection components from a circular sample in a normal one-way imaging, because it is circular, the circular sample itself is oriented in various directions with respect to the incident light, and the incident light is reflected at various angles because its orientation with respect to the incident light changes. The specular and diffuse reflections of the incident light may be close together or overlap. For this reason, when the liquid column is cylindrical, the intensities of the specular and diffuse reflections reflected at different angles can be obtained in a single imaging. In the case of a flat sample, when acquiring specular and diffuse reflection components from a flat sample using normal one-way imaging, the orientation of the flat sample itself with respect to the incident light is the same regardless of location, and the specular reflection and diffuse reflection of the incident light have different reflection angles. Therefore, when the liquid column is flat, it is necessary to change the camera position and take two images to obtain the intensity of the specular and diffuse reflections that are reflected at different angles. Thus, by making the shape of the nozzle opening 12a from which the composition Q is discharged circular, the liquid column becomes cylindrical, and the imaging location can be reduced to one place, which is preferable. When the shape of the nozzle opening 12a is circular, the diameter is not particularly limited, but is appropriately determined according to the viscosity of the composition, etc., for example, it is 2 to 4 mm.

[0055] [Composition Manufacturing Apparatus] Figure 17 is a schematic diagram showing an example of a composition manufacturing apparatus according to an embodiment of the present invention. In Figure 17, the same reference numerals are used for components identical to those in the inspection apparatus 20 shown in Figure 1, and their detailed descriptions are omitted. The composition manufacturing apparatus 60 shown in Figure 17 differs from the tank 10 shown in Figure 1 in that a raw material supply unit 61 is provided in the tank 10, and the tank 10 has a mixer (not shown) for mixing the raw materials of the composition. The mixer for mixing the raw materials of the composition is provided, for example, inside the tank 10. The configuration of the mixer is not particularly limited as long as it can mix the raw materials of the composition, and known mixers can be used as appropriate.

[0056] The raw material supply unit 61 includes, for example, a first raw material tank 62a and a second raw material tank 62b. The first raw material tank 62a and tank 10 are connected by piping 63a. A pump 64a and a valve 65a are provided on piping 63a in this order from the first raw material tank 62a side. The raw materials for the composition are stored in the first raw material tank 62a. The second raw material tank 62b and tank 10 are connected by piping 63b. A pump 64b and a valve 65b are provided on piping 63b in this order from the second raw material tank 62b side. The raw materials for the composition are stored in the second raw material tank 62b.

[0057] The raw materials for the composition stored in the first raw material tank 62a and the second raw material tank 62b may be in powder form or in liquid form obtained by mixing the raw materials for the composition with a solvent, and the form of the raw materials for the composition is not particularly limited. The composition is obtained, for example, by mixing the raw materials from the first raw material tank 62a and the raw materials from the second raw material tank 62b in tank 10 using a mixer (not shown). The raw materials from the first raw material tank 62a are, for example, magnetic particles, and the raw materials from the second raw material tank 62b are, for example, polymerizable compounds (e.g., epoxy compounds). In this case, the resulting composition is a magnetic paste. The configuration includes a first raw material tank 62a and a second raw material tank 62b, but the number of raw material tanks is not particularly limited and is appropriately determined according to the composition of the composition. Alternatively, one type of raw material itself may be used as the composition. In this case, at least one raw material tank for storing the raw materials for the composition is sufficient. The pumps, valves, piping, and raw material tanks mentioned above can be those known in the relevant technical field or otherwise as appropriate. The composition manufacturing apparatus 60 mixes the raw materials for the composition from the first raw material tank 62a and the raw materials for the composition from the second raw material tank 62b in the tank 10 using a mixer (not shown). The composition Q is then supplied from the tank 10 to the nozzle 12 by a pump 14, and the composition Q is continuously delivered from the nozzle opening 12a of the nozzle 12 to the container 16. As described above, when the composition Q is continuously delivered from the nozzle opening 12a of the nozzle 12, the composition Q forms a liquid column 17, and the inspection device 20 inspects the composition Q in the liquid column portion. The inspection of the composition by the inspection device 20 is as described above, so a detailed explanation is omitted.

[0058] [Method for Manufacturing the Composition] The method for manufacturing the composition will now be described. For example, the composition manufacturing apparatus 60 shown in Figure 17 above is used for manufacturing the composition, but the method for manufacturing the composition is not particularly limited to using the composition manufacturing apparatus 60 shown in Figure 17. The method for manufacturing the composition includes the steps of mixing raw materials to obtain a composition and inspecting the composition using the composition inspection method described above. More specifically, the method for manufacturing the composition includes the steps of obtaining a composition containing magnetic particles and inspecting the composition using the composition inspection apparatus described above. Furthermore, the method for manufacturing the composition includes the steps of obtaining a composition containing magnetic particles and inspecting the composition using the composition inspection method described above. The step of obtaining a composition containing magnetic particles described above includes, for example, storing magnetic particles as raw materials for the composition in a first raw material tank 62a and storing polymerizable compounds (e.g., epoxy compounds) as raw materials for the composition in a second raw material tank 62b. Valve 65a is opened to transfer magnetic particles from the first raw material tank 62a to tank 10 using pump 64a, and valve 65b is opened to transfer polymerizable compound from the second raw material tank 62b to tank 10 using pump 64b. In tank 10, the magnetic particles and polymerizable compound are mixed using a mixer (not shown) to obtain a magnetic paste as a composition. The magnetic paste is inspected using the composition inspection device or composition inspection method described above. Note that the inspection of the composition using the composition inspection device or composition inspection method is as described above, so a detailed explanation is omitted. The magnetic paste is used, for example, as an inductor, and viscosity control is very important for embedding it in through-holes of an inductor substrate, but as described above, the inspection device and inspection method can inspect viscosity. For this reason, the inspection device and inspection method are also suitable for controlling the viscosity of the magnetic paste.

[0059] The compositions described above for the method of inspecting the composition, the apparatus for inspecting the composition, and the method for manufacturing the composition preferably contain particles. The particles may be inorganic compounds, organic compounds, or organic-inorganic composite materials. Examples of inorganic compounds include magnetic particles, inorganic pigments, high-refractive-index inorganic particles, and low-refractive-index inorganic particles. Examples of organic compounds include organic pigments and polymer particles. Examples of organic-inorganic composite materials include inorganic compound particles coated with organic compounds and organic compound particles coated with inorganic compounds.

[0060] The type of magnetic particles used as raw materials for the above composition is not particularly limited, and the magnetic particles usually contain metal atoms. In the magnetic particles, the metal atoms may be included as alloys containing metal elements, metal oxides, metal nitrides, or metal carbides.

[0061] The above metal atoms are not particularly limited, but it is preferable to include at least one metal atom selected from the group consisting of Fe, Mn, Co, Zn, and Ni in terms of superior magnetic permeability. The content of at least one metal atom selected from the group consisting of Fe, Mn, Co, Zn, and Ni (or the total content if multiple types are included) is preferably 50% by mass or more, more preferably 60% by mass or more, and even more preferably 70% by mass or more, based on the total mass of metal atoms in the magnetic particles. The upper limit of the above content is not particularly limited, for example, it is 100% by mass or less, preferably 98% by mass or less, and more preferably 95% by mass or less.

[0062] The magnetic particles may contain materials other than Fe, Mn, Co, Zn, and Ni. Specific examples include Al, Si, S, Sc, Ti, V, Cu, Y, Mo, Rh, Pd, Ag, Sn, Sb, Te, Ba, Ta, W, Re, Au, Bi, La, Ce, Pr, Nd, P, Sr, Zr, Mn, Cr, Nb, Pb, Ca, B, C, N, and O. The shape of the magnetic particles is not particularly limited and may be plate-shaped, elliptical, spherical, or amorphous.

[0063] As magnetic particles, ferrite particles are preferred. In addition to Fe, which constitutes iron oxide, the ferrite particles preferably contain at least one metal atom selected from the group consisting of Mn, Co, Zn, and Ni, and preferably contain at least one metal atom selected from the group consisting of Mn and Co.

[0064] The type of polymerizable compound used as a raw material for the above composition is not particularly limited. Polymerizable compounds are compounds having polymerizable groups (thermally polymerizable compounds or photopolymerizable compounds), and examples include compounds containing a group with an ethylenically unsaturated bond (hereinafter also simply referred to as "ethylenically unsaturated group"), and compounds having an epoxy group and / or an oxetanyl group. The composition preferably contains a compound having an epoxy group and / or an oxetanyl group as a polymerizable compound. Among these, polyfunctional epoxy compounds (compounds having multiple epoxy groups) are preferred.

[0065] The above composition may contain a resin. Examples of resins include (meth)acrylic resins, epoxy resins, ene-thiol resins, polycarbonate resins, polyether resins, polyarylate resins, polysulfone resins, polyethersulfone resins, polyphenylene resins, polyarylene etherphosphine oxide resins, polyimide resins, polyamide-imide resins, polyolefin resins, cyclic olefin resins, polyester resins, styrene resins, and phenoxy resins.

[0066] One preferred embodiment of the resin is a resin that functions as a dispersant for dispersing magnetic particles in a composition (hereinafter also referred to as "dispersion resin"). Examples of dispersion resins include resins having repeating units including graft chains. Here, a graft chain refers to the unit from the base of the main chain (the atom bonded to the main chain at the branching group) to the end of the branching group. The graft chain preferably contains a polymer structure, and examples of such polymer structures include poly(meth)acrylate structures (e.g., poly(meth)acrylic structures), polyester structures, polyurethane structures, polyurea structures, polyamide structures, and polyether structures.

[0067] In addition to the components described above, the above composition may also contain solvents, reactive diluents, vibration modifiers, curing agents, curing accelerators, sensitizers, co-sensitizers, plasticizers, diluents, oil-sensing agents, fillers, surfactants, adhesion aids (e.g., silane coupling agents), and rubber components.

[0068] If the composition contains a solvent, the solid content concentration is preferably 50% by mass or more, more preferably 70% by mass or more, and the upper limit is 99% by mass. If the composition is a magnetic paste, the solid content consists of components other than the solvent. If the composition is a magnetic paste and contains a solvent, the solid content concentration is preferably 80% by mass or more, more preferably 90% by mass or more. The upper limit of the solid content concentration can be 98% by mass or 99% by mass. Furthermore, if the composition is a magnetic paste, it does not have to contain a solvent. If the magnetic paste does not contain a solvent, it is preferable to include a reactive diluent. For example, if the composition contains a liquid epoxy resin, the liquid epoxy resin is also included in the solid content.

[0069] The present invention is basically configured as described above. Although the method for inspecting the composition, the apparatus for inspecting the composition, and the method for manufacturing the composition of the present invention have been described in detail above, the present invention is not limited to the embodiments described above, and various improvements or modifications may be made without departing from the spirit of the present invention.

[0070] 10 Tank 12 Nozzle 12a Nozzle opening 13, 63a, 63b Piping 14, 64a, 64b Pump 15, 65a, 65b Valve 16 Container 17 Liquid column 20 Inspection device 21 Imaging unit 22 Inspection unit 23 Light source unit 23a, 23b Light source 24 Control unit 25 Memory 26 Display unit 27 Input unit 30 Storage unit 31 Image processing unit 32 Analysis unit 35 Learning model creation unit 41 Nozzle unit 41a Nozzle opening unit 42 Liquid column unit 45a, 45b, 45c, 45d, 45e, 45f Signal profile 46, 49 Signal profile 47, 48 Image 48a, 48c Specular reflection region 48b Diffuse reflection region 49a, 49c Peak 49b Bottom 50 Averaging image 50a, 50b, 52a, 52b Edge 51a First vertex 51b Second vertex 51c Third vertex 51d Fourth vertex 52 Region 53 Vertex 60 Manufacturing equipment 61 Raw material supply unit 62a First raw material tank 62b Second raw material tank D １ , D ２ Direction P １ , P ２ , P ３ , P ４ Signal peak Q Composition Q １ Composition part β, β １ point

Claims

1. A method for inspecting a composition, comprising the steps of: continuously supplying a composition from the nozzle opening of a nozzle; acquiring a video or image of the composition being continuously supplied from the nozzle opening; and further comprising at least one of the following steps: step A, which calculates at least one of the shape of the composition, the flow velocity of the composition, and the optical properties of the composition; and step B, which inspects the composition using a trained machine learning model that takes the video or image of the composition as input data.

2. The method for inspecting a composition according to claim 1, wherein step A is to calculate at least two of the following: the shape of the composition, the flow velocity of the composition, and the optical properties of the composition.

3. The method for inspecting a composition according to claim 1, wherein step A is used to calculate the shape of the composition, the flow velocity of the composition, and the optical properties of the composition.

4. The method for testing the composition according to claim 1, wherein the flow rate of the composition is 3 to 50 mL / min.

5. A method for inspecting a composition according to claim 1, comprising the step of inspecting the viscosity of the composition using at least one of the shape of the composition, the flow velocity of the composition, and the optical properties of the composition calculated in step A.

6. A method for inspecting a composition according to claim 1, further comprising: in step B, a step of creating a reconstructed image of the composition as output data using the trained machine learning model which takes a video or image of the composition as input data; and a step of creating difference data between the input data and the output data, and inspecting the viscosity of the composition using the difference data.

7. The method for testing the composition according to claim 1, wherein the composition has a solid content concentration of 50% by mass or more.

8. A method for inspecting the composition according to claim 1, wherein the composition comprises magnetic particles.

9. A composition inspection device comprising: an imaging unit that captures a video or image of a composition being continuously delivered from the nozzle opening of a nozzle; and an inspection unit that inspects the physical properties of the composition from the video or image of the composition obtained by the imaging unit, wherein the inspection unit calculates at least one of the following from the video or image of the composition: the shape of the composition, the flow velocity of the composition, and the optical properties of the composition, or inspects the composition using a trained machine learning model that takes the video or image of the composition as input data.

10. An inspection apparatus for a composition according to claim 9, comprising a light source unit that irradiates light from two directions onto the composition being continuously delivered from the nozzle opening of the nozzle.

11. A method for producing a composition, comprising the steps of: obtaining a composition containing magnetic particles; and inspecting the composition using the composition inspection method described in any one of claims 1 to 8.

12. A method for producing a composition, comprising the steps of: obtaining a composition containing magnetic particles; and inspecting the composition using the composition inspection apparatus described in claim 9 or 10.

Images