Performance evaluation system, performance evaluation method, program, and trained model

Abstract

This performance evaluation system comprises an evaluation unit which, using a learned model machine-learned on the basis of a learning image of an ion-exchange resin captured for learning and an evaluation result of the appearance of the ion-exchange resin, evaluates the performance of an ion-exchange resin to be inspected from an inspection image in which the ion-exchange resin to be inspected is captured.

Description

[Technical Field]

[0001] The present invention relates to a performance evaluation system, a performance evaluation method, a program, and a trained model. This application claims priority based on Japanese Patent Application No. 2020-172521, filed in Japan on October 13, 2020, and Japanese Patent Application No. 2021-159660, filed in Japan on September 29, 2021, the contents of which are incorporated herein by reference. [Background technology]

[0002] Ultrapure water is used for a variety of purposes, including as wash water in the manufacturing processes of semiconductors, liquid crystal displays, wafers, precision components, and the like; as desalinated water produced in power plant condensate demineralizers; and as water for pharmaceutical manufacturing. Ultrapure water is produced by combining a wide variety of devices, including microfiltration (MF) membranes, ultrafiltration (UF) membranes, reverse osmosis (RO) membranes, ion exchange resins, continuous electrolytic pure water systems (EDI), ultraviolet water sterilization devices, and degassing devices. Ion exchange resins are used as the main component of these devices. Such ultrapure water production processes include systems that primarily use single-bed ion exchange resin towers, systems that use mixed-bed ion exchange resin towers, and systems that combine these. (See, for example, Patent Document 1.) [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2014-104413 Summary of the Invention [Problem to be solved by the invention]

[0004] For example, a vendor that manufactures and supplies ion exchange resins to customers periodically receives samples of the ion exchange resins used in ion exchange resin towers from their customers to evaluate their performance. One method of evaluating ion exchange resin performance is the appearance index. New ion exchange resins are perfect spheres, but as they deteriorate, they develop cracks and break. The appearance index is calculated by observing over 300 ion exchange resins one by one under a magnifying glass, counting them, and categorizing them into appearance categories such as perfect spheres, cracks, and broken spheres. The appearance index is an indicator of the degree of deterioration of the ion exchange resin and serves as a guide for replacing the ion exchange resin used in the customer's ion exchange resin towers.

[0005] However, in the above-mentioned performance evaluation of ion exchange resins, an analyst visually inspects each ion exchange resin particle using a magnifying glass and calculates the appearance index, which takes a lot of time even for an experienced analyst. Note that the magnifying glass can be a microscope, a camera with a magnifying function, or the like.

[0006] The present invention has been made in consideration of the above points, and an object of the present invention is to provide a performance evaluation system, a performance evaluation method, a program, and a trained model that can accurately evaluate the performance of an ion exchange resin in a short period of time. [Means for solving the problem]

[0007] The present invention has been made to solve the above-mentioned problems, and one aspect of the present invention is a performance evaluation system that includes an evaluation unit that evaluates the performance of an ion exchange resin for testing from an inspection image of the ion exchange resin for testing, using a trained model that has been machine-learned based on training images of the ion exchange resin taken for training purposes and evaluation results of the appearance of the ion exchange resin.

[0008] In addition, in the above-mentioned performance evaluation system, a classification according to the type of appearance of the ion exchange resin is pre-set as a result of the appearance evaluation, and the evaluation unit may evaluate the performance of the ion exchange resin for inspection using a trained model that has been machine-learned based on the training images of multiple ion exchange resins photographed for each type of appearance.

[0009] In addition, in the above-mentioned performance evaluation system, the type of appearance may be classified based on at least one of the presence or absence of cracks and the presence or absence of fractures in the appearance of the ion exchange resin, and the evaluation unit may calculate an appearance index indicating the state of the appearance of the ion exchange resin by evaluating the performance of the ion exchange resin for inspection.

[0010] In addition, in the above-mentioned performance evaluation system, the evaluation unit may input the inspection images, in which multiple images of the inspection ion exchange resins are taken, into the trained model, and use the trained model to analyze which of the appearance types each of the inspection ion exchange resins falls into, and calculate an appearance index indicating the appearance state of the inspection ion exchange resins contained in the inspection image based on the number of ion exchange resins classified into each appearance type.

[0011] In addition, in the above-mentioned performance evaluation system, the test image may be an image obtained by enlarging and photographing a plurality of the test ion exchange resins placed in a container with a magnifying camera, and when photographing with the magnifying camera, an image may be obtained by selecting an area where there is little overlap between the ion exchange resins to be photographed.

[0012] In addition, in the above-mentioned performance evaluation system, the inspection image may be an image obtained by adding a surfactant to a container containing a plurality of the inspection ion exchange resins and stirring the mixture before being photographed with a magnifying camera, and then photographing the image with the magnifying camera.

[0013] The performance evaluation system may further include a learning unit that performs machine learning based on the learning image of the ion exchange resin photographed for learning purposes and the evaluation results of the appearance of the ion exchange resin.

[0014] In addition, in the above-mentioned performance evaluation system, the training images may include images of the appearance of the training ion exchange resin photographed multiple times while changing at least one of the size, angle, and color tone, or multiple images generated by photographing an image of the appearance of the ion exchange resin and changing at least one of the size, angle, and color tone.

[0015] Another aspect of the present invention is a performance evaluation method for an ion exchange resin, comprising the steps of acquiring an inspection image of an ion exchange resin for inspection, and evaluating the performance of the ion exchange resin for inspection from the inspection image using a trained model that has been machine-learned based on the training image of the ion exchange resin taken for training purposes and the evaluation results of the appearance of the ion exchange resin.

[0016] Another aspect of the present invention is a program for causing a computer to execute the steps of acquiring an inspection image of an ion exchange resin for inspection, and evaluating the performance of the ion exchange resin for inspection from the inspection image using a trained model that has been machine-learned based on the training image of the ion exchange resin taken for training purposes and the evaluation results of the appearance of the ion exchange resin.

[0017] Another aspect of the present invention is a trained model for evaluating the performance of an ion exchange resin for inspection from an inspection image of the ion exchange resin for inspection, the trained model being machine-learned based on the training image of the ion exchange resin taken for training purposes and the evaluation results of the appearance of the ion exchange resin, and causing a computer to function to evaluate the performance of the ion exchange resin for inspection from the inspection image.

[0018] Another aspect of the present invention is a performance evaluation system comprising a water treatment facility and a performance evaluation device, wherein the water treatment facility comprises an ion exchange resin tower, an imaging unit that photographs the ion exchange resin in the ion exchange resin tower, and a communication unit that transmits the image photographed by the imaging unit as an inspection image, and the performance evaluation device comprises a communication unit that receives the inspection image, and an evaluation unit that evaluates the performance of the inspection ion exchange resin from the inspection image using a trained model that has been machine-learned based on a training image of the ion exchange resin photographed for training purposes and an evaluation result of the appearance of the ion exchange resin.

[0019] In one aspect of the present invention, in the performance evaluation system, the performance evaluation device may transmit information based on an evaluation result by the evaluation unit to the water treatment facility.

[0020] In another aspect of the present invention, the water treatment facility may generate order information for replacement ion exchange resin when replacement of the ion exchange resin is necessary based on information obtained from the performance evaluation device. [Effects of the Invention]

[0021] According to the present invention, the performance of an ion exchange resin can be evaluated accurately in a short time. [Brief explanation of the drawings]

[0022] [Figure 1] 3A to 3C are diagrams showing examples of different types of appearance of ion exchange resins according to an embodiment. [Figure 2] FIG. 1 is a diagram showing an overview of a method for evaluating the performance of an ion exchange resin according to an embodiment. [Figure 3] 1 is a system diagram showing an example of the configuration of a performance evaluation system 1 according to an embodiment. [Figure 4] FIG. 4 is a diagram showing an example of a GUI screen according to the embodiment. [Figure 5] FIG. 2 is a block diagram showing an example of the configuration of a camera 10b according to the embodiment. [Figure 6]FIG. 1 is a block diagram showing an example of the configuration of a performance evaluation device 30 according to an embodiment. [Figure 7] FIG. 2 is a diagram showing an example of criteria for determining the performance of an ion exchange resin according to an embodiment. [Figure 8] 4 is a flowchart showing an example of an ion exchange resin evaluation process according to the embodiment. [Figure 9] FIG. 1 is a block diagram showing an example of the configuration of a machine learning device 50 according to an embodiment. [Figure 10] FIG. 1 is an explanatory diagram illustrating an execution procedure in machine learning according to an embodiment. [Figure 11] FIG. 2 is an explanatory diagram illustrating a learning procedure in machine learning according to the embodiment. [Figure 12] 10A and 10B are diagrams showing analysis results by AI and visual analysis results according to an embodiment. [Figure 13] FIG. 10 is a comparison diagram of the analysis results by AI according to the embodiment and the analysis results by visual inspection. [Figure 14] FIG. 10 is a diagram showing an example in which a particle is mistakenly determined to be a cracked sphere before the addition of an anionic surfactant according to the embodiment. [Figure 15] FIG. 10 is a diagram showing an example in which the particles are erroneously determined to be crushed balls before the addition of an anionic surfactant according to the embodiment. [Figure 16] FIG. 10 is a diagram showing an example in which overlapping resin portions cannot be determined before the addition of an anionic surfactant according to the embodiment. [Figure 17] FIG. 10 is a diagram showing an example of an analysis result of an inspection image after adding an anionic surfactant according to the embodiment. [Figure 18] FIG. 10 is a system diagram showing another example of the configuration of a performance evaluation system according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0023] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Ion exchange resins remove impurities from water by adsorbing cations such as sodium, calcium, and magnesium, and anions such as chloride and carbonate, and then releasing their original ions to perform ion exchange. They are used to produce ultrapure water, which is used for a variety of purposes, including cleaning water used in the manufacturing processes of semiconductors, liquid crystals, wafers, and precision parts; desalted water produced in power plant condensate demineralizers; and pharmaceutical manufacturing water. Ultrapure water production processes include systems that primarily use single-bed ion exchange resin towers, systems that use mixed-bed ion exchange resin towers, and combinations of these.

[0024] Ion exchange resins can be broadly categorized by their structure into "gel type" and "porous type," each of which includes cation exchange resins and anion exchange resins. Gel type resins have a larger ion exchange capacity per volume than porous types, making them advantageous for producing ultrapure water, but they have the disadvantage of lower cycle strength than porous types. Furthermore, gel type resins generally have a smaller specific surface area than porous type resins, so while they present no problem in adsorbing ordinary inorganic ions (such as chloride ions), they are disadvantageous for adsorbing high molecular weight substances.

[0025] New ion exchange resins are perfect spheres, but as they deteriorate, they crack or break. As ion exchange resins deteriorate, their ion exchange performance decreases, and they need to be replaced. An appearance index, which indicates the state of appearance, is used as an indicator of the degree of deterioration of ion exchange resins. The appearance index is a value calculated based on the state of the appearance of the ion exchange resin.

[0026] For example, based on the appearance of cracks or fractures that occur in ion exchange resins due to deterioration, they are classified into appearance types (categories), such as cracked spheres (cracked), crushed spheres (crushed), and perfect spheres (no cracks or fractures). Figure 1 shows an example of the appearance types of ion exchange resins. The illustrated example shows an example of the appearance photographs of "gel-type" ion exchange resins classified into three appearance types: perfect spheres, cracked spheres, and crushed spheres, based on the differences in appearance due to deterioration.

[0027] For example, in the past, for one sample, an analyst would visually inspect each of 300 pieces of ion exchange resin using a magnifying glass (such as a microscope) and classify them into perfect spheres, cracked spheres, and broken spheres. Then, they would calculate an appearance index, which is an indicator of the degree of deterioration of the ion exchange resin, using the following formula 1 to evaluate its performance.

[0028] Appearance index (%) = (300 - total number of crushed balls) / 300 × 100 (Equation 1)

[0029] However, since it takes a long time for an analyst to visually observe 300 ion exchange resin particles one by one using a magnifying glass, classify them into appearance types such as perfect spheres, cracked spheres, and broken spheres, and then count and calculate the appearance index, a method for evaluating the performance of ion exchange resins more quickly and accurately is desired. Therefore, the ion exchange resin performance evaluation system according to this embodiment is configured to use AI (artificial intelligence) to automatically calculate the appearance index from photographed images of the appearance of the ion exchange resin. The ion exchange resin performance evaluation system according to this embodiment will be described in detail below.

[0030] [Outline of the performance evaluation system] FIG. 2 is a diagram illustrating an overview of a method for evaluating the performance of ion exchange resins according to this embodiment. In this embodiment, a large number of pairs (training datasets) of photographed images of the appearance of ion exchange resins and appearance evaluation results of the ion exchange resins are prepared as training data, and machine learning is performed based on these photographed images and appearance evaluation results of the large number of training ion exchange resins. The appearance evaluation results are the results of previous classification of ion exchange resins into appearance types such as perfect spheres, cracked spheres, and broken spheres by an analyst visually observing them using a microscope. For example, in an example comparing the AI ​​described below with an analyst (visual inspection), previously photographed data of approximately 2,000 perfect spheres, approximately 200 cracked spheres, and approximately 200 broken spheres was prepared as training data, and machine learning was performed.

[0031] By inputting a photographed image of the appearance of the ion exchange resin to be inspected into this machine-learned model (AI: Artificial Intelligence), the performance of the ion exchange resin shown in the photographed image can be evaluated. Specifically, each ion exchange resin shown in the photographed image of the ion exchange resin to be inspected is classified into one of perfect spheres, cracked spheres, or broken spheres, and an appearance index is calculated.

[0032] [Configuration of Performance Evaluation System 1] 3 is a system diagram showing an example of the configuration of a performance evaluation system 1 according to this embodiment. The performance evaluation system 1 includes a microscope camera 10 and a performance evaluation device 30.

[0033] The microscope camera 10 has a camera 10b attached to a microscope 10a. The camera 10b is, for example, a digital camera. The camera 10b captures an optical image magnified by the microscope 10a, converts it into electronic data, and transmits the converted electronic data (captured image) to the performance evaluation device 30. For example, the microscope camera 10 and the performance evaluation device 30 are connected via a USB (Universal Serial Bus). The microscope camera 10 and the performance evaluation device 30 may be connected by other wired or wireless connection methods, not limited to USB. The microscope camera 10 may have a configuration in which the microscope 10a and the camera 10b are integrated (a configuration in which they cannot be removed), or may be an electron microscope.

[0034] The microscope camera 10 photographs an ion exchange resin sample for inspection contained in a petri dish 20 and transmits the photographed image to the performance evaluation device 30. At this time, the operator photographs a large number (e.g., 300 or more) of ion exchange resins for one sample contained in the petri dish 20, and photographs 10 images so that different parts of the sample are photographed while shifting the petri dish 20. Note that the operator selects and photographs parts with minimal overlap, because if adjacent ion exchange resins overlap, they may be judged as scratches (e.g., cracked balls). Note that, hereinafter, the photographed image of the ion exchange resin for inspection is referred to as the "inspection image."

[0035] The performance evaluation device 30 is a so-called desktop computer to which a monitor 30a and a keyboard 30b are connected as external devices (peripheral devices). One or both of the monitor 30a and the keyboard 30b may be built into the performance evaluation device 30. For example, the performance evaluation device 30 is not limited to a desktop computer, and may be a tablet computer, a notebook computer, or the like.

[0036] The performance evaluation device 30 acquires and stores the inspection images transmitted from the microscope camera 10. The performance evaluation device 30 then evaluates the performance of the ion exchange resin to be inspected from the inspection images using a trained model that has been machine-learned based on the images of the ion exchange resin taken for training purposes and the evaluation results of the appearance of the ion exchange resin. Note that, hereinafter, the images of the ion exchange resin to be inspected are referred to as "training images." This trained model has been machine-learned based on multiple training images of ion exchange resins taken for each type of appearance of the ion exchange resin (e.g., perfect sphere, cracked sphere, crushed sphere). For example, the performance evaluation device 30 uses the trained model to calculate an appearance index from the inspection image to evaluate the performance of the ion exchange resin to be inspected.

[0037] The operator operates the keyboard 30b while viewing the GUI screen displayed on the monitor 30a of the performance evaluation device 30, thereby analyzing and evaluating the performance of the ion exchange resin shown in the inspection image.

[0038] FIG. 4 is a diagram showing an example of a GUI screen according to this embodiment. The illustrated GUI screen G10 is an image displayed when an inspection image captured in the performance evaluation device 30 is analyzed using AI. A screen area 101 displays options for an ion exchange resin sample to be inspected. When a sample selection button 102 is operated, the inspection image (original image) of the selected sample is displayed in a screen area 103. There are ten inspection images taken for one sample. By performing an operation to select each of No. 1 to No. 10 in the screen area 103, the inspection image displayed in the screen area 103 can be switched to the selected inspection image.

[0039] When the analysis start button 104 is pressed, an analysis is performed using AI on the test image displayed in the screen area 103, and an analysis image is displayed in the screen area 105. This analysis image displays the results of classifying each of the ion exchange resins shown in the test image into perfect spheres, cracked spheres, or broken spheres using a trained model. In the illustrated example, the test image No. 1 of Sample 1 is displayed in the screen area 103, and an analysis image obtained by analyzing the test image using AI is displayed in the screen area 105. Here, perfect spheres are indicated by a solid frame, cracked spheres by a dashed frame, and broken spheres by a double-line frame. Note that the analysis image only needs to display the different appearance types of the ion exchange resins in a distinguishable manner, and the display format can be determined arbitrarily. For example, the different appearance types of the ion exchange resins may be distinguished by different frame colors or different frame shapes.

[0040] Additionally, the screen area 106 displays the analysis results, including the number of perfect spheres, cracked spheres, and broken spheres (43, 1, 2), their percentages (93.5%, 2.2%, 4.3%), the total number of each (46), and the appearance index value (95.7%). The appearance index value is calculated using the following formula 2. Note that this evaluation number (total number) is not particularly limited.

[0041] Appearance index (%) = (total - total number of crushed balls) / total × 100 (Equation 2) Here, "total" is the total number of complete spheres, cracked spheres, and broken spheres. Equation 2 is simply a generalization of Equation 1, where "300" is used to represent the total, but the basic calculation method is the same.

[0042] [Configuration of camera 10b] Next, the configuration of the camera 10b according to this embodiment will be described in detail. Fig. 5 is a block diagram showing an example of the configuration of a camera 10b according to this embodiment. In this figure, components corresponding to those in Fig. 3 are assigned the same reference numerals. The camera 10b can be attached to the microscope 10a via an adapter. The camera 10b includes, for example, a communication unit 11, an imaging unit 12, a storage unit 13, and a control unit 15.

[0043] The communication unit 11 is configured to include a digital input / output port such as a USB (Universal Serial Bus). For example, the communication unit 11 is communicatively connected to the performance evaluation device 30 and transmits images captured by the camera 10b to the performance evaluation device 30. Note that the communication unit 11 may transmit the captured images to the performance evaluation device 30 using a method other than USB. For example, the communication unit 11 may be configured to include a video output terminal such as HDMI (registered trademark) and a communication device compatible with communication standards such as wireless LAN, wired LAN, and Bluetooth (registered trademark) instead of or in addition to USB.

[0044] The imaging unit 12 is configured to include an imaging element, an optical lens provided in front of the imaging surface of the imaging element, and the like. Under the control of the control unit 15, the imaging unit 12 captures an optical image magnified by the microscope 10a through the optical lens. Under the control of the control unit 15, the imaging unit 12 also performs image processing on the captured image and stores it in the storage unit 13 as a captured image. Here, the captured image is, for example, an image for inspection.

[0045] The memory unit 13 includes, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), an EEPROM (Electrically Erasable Programmable Read-Only Memory), a ROM (Read-Only Memory), a RAM (Random Access Memory), etc., and stores various information, images, programs, etc. used by the camera 10b for processing.

[0046] The operation unit 14 receives user operations on the camera 10b. For example, the operation unit 14 includes operation buttons such as a shutter button. The operation unit 14 may also include operation buttons, operation dials, operation switches, etc. that receive operations for setting the shooting mode and shooting conditions. Furthermore, a performance evaluation device 30 that is communicatively connected to the camera 10b may receive some or all of the user operations on the camera 10b, and the performance evaluation device 30 may control the camera 10b, such as taking pictures, in accordance with the user operations.

[0047] The control unit 15 includes a CPU (Central Processing Unit) and executes various programs stored in the storage unit 13 to control each unit of the camera 10b. For example, the control unit 15 instructs the imaging unit 12 to capture an image based on a user's operation on the operation unit 14. The control unit 15 also stores in the storage unit 13 a captured image based on an image captured by the imaging unit 12 in response to the capture instruction. The control unit 15 also transmits the captured image based on the image captured by the imaging unit 12 to the performance evaluation device 30 via the communication unit 11. Here, the captured image is, for example, an inspection image, as described above. The captured image is stored (saved) in the storage unit 13 as an image file associated with a file name determined according to a predetermined rule, information on the date and time of capture (timestamp), and the like, and is also transmitted to the performance evaluation device 30.

[0048] [Configuration of performance evaluation device 30] Next, the configuration of the performance evaluation device 30 according to this embodiment will be described in detail. Fig. 6 is a block diagram showing an example of the configuration of a performance evaluation device 30 according to this embodiment. In this figure, components corresponding to those in Fig. 3 are assigned the same reference numerals. The performance evaluation device 30 shown in the figure is configured to include a communication unit 31, a video output unit 32, USB connectors 33 and 34, a storage unit 35, and a control unit 36.

[0049] The communication unit 31 is configured to include, for example, multiple Ethernet (registered trademark) ports, wireless communication ports such as Wi-Fi (registered trademark) and mobile phone lines, and communicates (sends or receives) with external devices via a communication network (such as the Internet) based on control by the control unit 36.

[0050] The video output unit 32 is configured to include an external monitor output terminal for outputting a video signal to an external display device (monitor, projector, etc.). The external monitor output terminal is an HDMI (registered trademark) terminal, a DVI terminal, a D-SUB terminal, a Display Port terminal, etc. For example, the video output unit 32 is connected to the monitor 30a and outputs a video signal to the monitor 30a.

[0051] The monitor 30a is a display device having a display for displaying information such as images, text, etc. For example, the monitor 30a includes a liquid crystal display panel, an organic EL (ElectroLuminescence) display panel, or the like.

[0052] The USB connectors 33 and 34 are connection terminals for connecting to USB-compatible external devices. For example, the USB connector 33 is connected to the keyboard 30b and acquires output signals in response to operations on the keyboard 30b. The USB connector 34 is connected to the camera 10b of the microscope camera 10 and acquires inspection images (image files) and the like transmitted from the camera 10b.

[0053] The storage unit 35 includes, for example, an HDD, SSD, EEPROM, ROM, RAM, etc., and stores various information, images, programs, etc. used for processing by the performance evaluation device 30. Note that the storage unit 35 is not limited to being built into the performance evaluation device 30, but may be an external storage device connected via a digital input / output port such as a USB. For example, the storage unit 35 includes an inspection image storage unit 351, a learning model storage unit 352, and an evaluation data storage unit 353.

[0054] The test image storage unit 351 stores the test image (image file) transmitted from the camera 10b. For example, the test image storage unit 351 stores the image file of the test image transmitted from the camera 10b in association with a sample number (e.g., sample 1, 2, . . .) and a photographing number (e.g., No. 1 to No. 10). The sample number and the photographing number may be associated in the performance evaluation device 30 based on a user operation, or may be automatically associated in the performance evaluation device 30 according to a predetermined rule.

[0055] The learning model storage unit 352 stores a trained model for evaluating ion exchange resins shown in an inspection image. As described above, this trained model is machine-learned based on a plurality of training images of ion exchange resins taken for each type of ion exchange resin appearance (e.g., perfect sphere, cracked sphere, broken sphere). By inputting an inspection image into the trained model, each of the ion exchange resins shown in the inspection image is classified into one of perfect sphere, cracked sphere, and broken sphere. Details of the configuration and process for generating a trained model by machine learning will be described later.

[0056] The evaluation data storage unit 353 stores the evaluation results obtained by evaluating the inspection image using the trained model. The evaluation results include the number and percentage of perfect spheres, cracked spheres, and broken spheres contained in the ion exchange resin included in the inspection image, the total number of each type, and a calculated appearance index. The evaluation results are stored in association with the file name of the inspection image.

[0057] The control unit 36 ​​is configured to include a CPU and the like, and executes various programs stored in the storage unit 35 to control each unit of the performance evaluation device 30. For example, the control unit 36 ​​includes an inspection image acquisition unit 361, an evaluation unit 362, and an output control unit 363 as functional components realized by executing the various programs stored in the storage unit 35.

[0058] The test image acquisition unit 361 acquires the test image (image file) transmitted from the camera 10b via the communication unit 31 and stores it in the test image storage unit 351. For example, the test image acquisition unit 361 associates the image file of the test image with a sample number (e.g., sample 1, 2, . . .) and a shooting number (e.g., No. 1 to No. 10), and stores the image file in the test image storage unit 351. As described above, the test image acquisition unit 361 may associate the sample number and the shooting number based on a user operation (e.g., an operation on the keyboard 30b), or may automatically associate them according to a predetermined rule.

[0059] The evaluation unit 362 evaluates the performance of the ion exchange resin to be inspected from an inspection image of the ion exchange resin to be inspected, using a trained model that has been machine-learned based on training images of the ion exchange resin photographed for training purposes and the evaluation results of the appearance of the ion exchange resin. The appearance evaluation results are the results of evaluating whether the ion exchange resin is a perfect sphere, a cracked sphere, or a broken sphere, which are classified based on the presence or absence of cracks and fractures in the appearance of the ion exchange resin. For example, the trained model is trained based on training images of multiple ion exchange resins photographed for each type of appearance of the ion exchange resin (e.g., perfect sphere, cracked sphere, broken sphere).

[0060] The evaluation unit 362 inputs inspection images of multiple ion exchange resins for inspection into the trained model, and then uses the trained model to analyze which appearance type (perfect sphere, cracked sphere, or crushed sphere) each of the ion exchange resins for inspection falls into.The evaluation unit 362 then calculates an appearance index based on the number of ion exchange resins for each appearance type. Specifically, the evaluation unit 362 calculates the appearance index using the above-mentioned Equation 2.

[0061] The output control unit 363 controls the display of the GUI screen G10 displayed on the monitor 30a when analyzing the performance of the ion exchange resin. For example, the output control unit 363 displays the GUI screen G10 on the monitor 30a, and also displays the test image selected in response to a user operation and the evaluation results by the evaluation unit 362 (such as an analysis image obtained by analyzing the test image and the analysis results). The output control unit 363 also stores the evaluation results (analysis results) by the evaluation unit 362 in the evaluation data storage unit 353.

[0062] The performance (level of deterioration) of the ion exchange resin may be determined using the evaluation categories shown in FIG. 7 based on the appearance index calculated by the evaluation unit 362. FIG. 7 is a diagram showing an example of criteria for determining the performance (level of deterioration) of the ion exchange resin according to this embodiment. In the example of the criteria shown, the performance (level of deterioration) of the ion exchange resin is classified into five evaluation categories. When the appearance index is 95 to 100%, the resin is classified into "Evaluation Category 1," which is determined to have almost no broken particles. When the appearance index is 80 to 94%, the resin is classified into "Evaluation Category 2," which is determined to have a small amount of broken particles. When the appearance index is 60 to 79%, the resin is classified into "Evaluation Category 3," which is determined to have broken particles. When the appearance index is 40 to 59%, the resin is classified into "Evaluation Category 4," which is determined to have a large amount of broken particles. When the appearance index is less than 40%, the resin is classified into "Evaluation Category 5," which is determined to have a significant amount of broken particles.

[0063] For example, the ion exchange resin may be replaced when the evaluation category is 4 or 5, or when the evaluation category is 3 to 5. The evaluation category in which the ion exchange resin is replaced can be determined arbitrarily. The criteria for determining the performance (degree of deterioration) of the ion exchange resin shown in FIG. 7 are only an example, and the range of the appearance index for each evaluation category can be determined arbitrarily. The number of evaluation categories is not limited to five, and the evaluation may be classified into any number of categories.

[0064] [Ion exchange resin evaluation process] Next, the operation of the ion exchange resin evaluation process in which the control unit 36 ​​of the performance evaluation device 30 evaluates the ion exchange resin from the test image will be described. FIG. 8 is a flowchart showing an example of the ion exchange resin evaluation process according to this embodiment.

[0065] (Step S101) The operator uses the microscope camera 10 to capture images of the ion exchange resin to be inspected (inspection images). Specifically, the operator captures 10 images of one sample while shifting the petri dish 20 so that different parts of the sample are captured. The microscope camera 10 transmits the captured inspection images to the performance evaluation device 30. Then, the process proceeds to step S103.

[0066] (Step S103) When the control unit 36 ​​acquires an inspection image (image file) from the microscope camera 10, the control unit 36 ​​stores the acquired inspection image (image file) in the inspection image storage unit 351. Then, the process proceeds to step S105.

[0067] (Step S105) The control unit 36 ​​uses the trained model stored in the trained model storage unit 352 to analyze the appearance type of the ion exchange resin shown in the inspection image. Specifically, the control unit 36 ​​reads out the inspection image from the inspection image storage unit 351 in response to an operator's operation on the GUI screen G10 (see FIG. 4), and inputs the image into the trained model stored in the trained model storage unit 352. The control unit 36 ​​classifies the ion exchange resin for inspection by appearance type (perfect sphere, cracked sphere, broken sphere) based on the analysis results output from the trained model, and outputs the number of pieces for each appearance type. Then, the process proceeds to step S107.

[0068] (Step S107) Based on the analysis results of the ion exchange resin for inspection (the number of pieces for each appearance type), the control unit 36 ​​calculates the appearance index using the above-mentioned Equation 2. Then, the process proceeds to step S109.

[0069] (Step S109) The control unit 36 ​​outputs the analysis results (evaluation results). For example, the control unit 36 ​​displays an analysis image obtained by analyzing the inspection image and the analysis results (such as the number of complete spheres, cracked spheres, and broken spheres, and the calculated appearance index) as the analysis results (evaluation results) on the monitor 30a (GUI screen G10 (see FIG. 4)). The control unit 36 ​​also stores the analysis results (evaluation results) in the evaluation data storage unit 353.

[0070] As a result, the performance evaluation system 1 can photograph the ion exchange resin to be inspected and input the photograph into the performance evaluation device 30, thereby enabling the performance of the ion exchange resin to be evaluated accurately in a short time.

[0071] [Machine learning device configuration] Next, the configuration of the machine learning device 50 that generates a trained model used in the ion exchange resin evaluation process in the performance evaluation system 1 will be described. 9 is a block diagram showing an example of the configuration of a machine learning device 50 according to this embodiment. The machine learning device 50 includes a communication unit 510, a learning data setting unit 520, a learning data storage unit 530, a learning unit 540, and an output unit 550. The configuration of the machine learning device 50 may be included in the performance evaluation device 30.

[0072] The communication unit 510 is configured to include, for example, multiple Ethernet (registered trademark) ports, multiple digital input / output ports such as USB, wireless communication ports such as Wi-Fi (registered trademark) or mobile phone lines, etc., and communicates with other devices and terminals via a communication network.

[0073] The training data setting unit 520 acquires information necessary to generate a trained model. For example, the training data setting unit 520 may acquire training images of each ion exchange resin of each appearance type that have been previously captured from the camera 10b, or may acquire the images from the performance evaluation device 30, another device, or a storage medium such as an optical disk or a memory card. The training data setting unit 520 associates the acquired training images with the appearance evaluation results (appearance types) of the ion exchange resins shown in the training images as a training data set and stores them in the training data storage unit 530. Hereinafter, the appearance evaluation results (appearance types) will be referred to as "evaluation values."

[0074] The learning unit 540 performs machine learning using a training dataset in which training images of ion exchange resins are associated with evaluation values. Specifically, the learning unit 540 reads the training dataset from the training data storage unit 530. The learning unit 540 performs machine learning using the read training dataset to generate a trained model. The output unit 550 transmits the trained model to the performance evaluation device 30 via the communication unit 510. As a result, the trained model is stored in the training model storage unit 352 of the performance evaluation device 30 and becomes available for use. Note that the trained model may be input to the performance evaluation device 30 via a storage medium such as an optical disc or a memory card instead of the communication unit 510. Furthermore, the trained model stored in the performance evaluation device 30 may be updated as new training datasets are input to the machine learning device 50 and machine learning progresses.

[0075] [Details of ion exchange resin evaluation process] Below, we will explain in detail the ion exchange resin evaluation process using the trained model. The learning unit 540 sets the pixel values ​​of the learning images as input variables to be input to the input layer of a learning CNN (convolutional neural network), and sets the evaluation values ​​of the learning images as output variables to be output from the output layer. The learning unit 540 performs machine learning using a learning dataset of the learning images and evaluation values. Furthermore, the evaluation unit 362 of the performance evaluation device 30 inputs pixel values ​​of an image of an ion exchange resin for inspection (inspection image) to the input layer of the trained CNN, and obtains an evaluation value from the output layer.

[0076] (About CNN) FIG. 10 is an explanatory diagram illustrating the execution procedure of machine learning according to this embodiment. In this diagram, the CNN is composed of I+1 layers L0 to L1. Layer L0 is called the input layer, layers L1 to L(I-1) are called intermediate or hidden layers, and layer L1 is called the output layer. I is determined by the structure of the CNN, and can be, for example, 3 or 4.

[0077] In CNN, an input image is input to the input layer L0. The input image is represented by a pixel matrix D11, where the vertical and horizontal positions of the input image are the matrix positions. Each element of the pixel matrix D11 contains the R (red), G (green), and B (blue) subpixel values ​​as the subpixel values ​​of the pixel corresponding to the matrix position. The first hidden layer L1 is a layer where convolution processing (also called filtering processing) and pooling processing are performed.

[0078] (convolution processing) An example of the convolution process in the intermediate layer L1 will be described. The convolution process is a process of applying a filter to the original image and outputting a feature map. Specifically, the input pixel values ​​are divided into an R subpixel matrix D121, a B subpixel matrix D122, and a G subpixel matrix D123. For each of the subpixel matrices D121, D122, and D123 (each also referred to as a "subpixel matrix D12"), a first pixel value is calculated by multiplying each element of the submatrix by an element of an s-by-t-column convolution matrix CM1 (also called a kernel) and adding the resulting multiplied results. A second pixel value is calculated by multiplying each of the first pixel values ​​calculated in each subpixel matrix D12 by a weighting coefficient and adding the resulting multiplied results. The second pixel value is set for each element of the convolved image matrix D131 as a matrix element corresponding to the position of the submatrix. The position of the submatrix in each subpixel matrix D12 is shifted element by element (subpixel) to calculate the second pixel value at each position, and all of the matrix elements of the convolved image matrix D131 are calculated.

[0079] 10 shows an example of a 3-row, 3-column convolution matrix CM1, where the first pixel value of the convolved pixel value D1311 is calculated for the 3-row, 3-column submatrix from the second to fourth rows and the second to fourth columns of each subpixel matrix D12. A weighting coefficient is calculated and added to the first pixel value of each subpixel matrix D121, D122, and D123, thereby calculating the second pixel value as the matrix element in the second row and second column of the convolved image matrix D131. Similarly, the second pixel value of the matrix element in the third row and second column of the convolved image matrix D131 is calculated from the submatrix from the third row to the fifth row and the second to fourth columns. Similarly, convolved image matrices D132, . . . are calculated using other weighting coefficients or other convolution matrices.

[0080] (pooling process) An example of the pooling process in the hidden layer L1 will be described below. The pooling process is a process for reducing an image while retaining its features. Specifically, for each region PM of u rows and v columns, the convolved image matrix D131 calculates a representative value of the matrix elements in the region. The representative value is, for example, the maximum value. The representative value is set for each element of the CNN image matrix D141 as a matrix element corresponding to the position of the region. By shifting the regions in the convolved image matrix D131 for each region PM, a representative value at each position is calculated, and all matrix elements of the convolved image matrix D131 are calculated.

[0081] For example, FIG. 10 shows an example of a 2-row, 2-column region PM, where the maximum value of the second pixel values ​​in the 2-row, 2-column region from the third row to the fourth row and the third column to the fourth column of the convolved image matrix D131 is calculated as a representative value. This representative value is set to the matrix element in the second row, the second column of the CNN image matrix D141. Similarly, a representative value of the matrix element in the third row, the second column of the CNN image matrix D141 is calculated from the submatrix from the fifth row to the sixth row and the second column to the fourth column. Similarly, a CNN image matrix D142,... is calculated from the convolved image matrix D132,...

[0082] The matrix elements (N elements) of the CNN image matrices D141, D142, ... are arranged in a predetermined order to generate a vector x. In Fig. 10, the elements xn (n = 1, 2, 3, ... N) of the vector x are represented by N nodes.

[0083] The hidden layer Li represents the ith hidden layer (i = 2 to I-1). The i-th hidden layer node outputs vector z(i) as the value of vector u(i) input to function f(u(i)). Vector u(i) is the vector obtained by multiplying vector z(i-1) output from the (i-1)-th hidden layer node by weight matrix W(i) from the left and adding vector b(i). Function f(u(i)) is the activation function, and vector b(i) is the bias. Vector u(0) is vector x.

[0084] The node of the output layer L4 is z(I-1), and its output is M elements ym (m=1, 2, M). In other words, the output layer LI of the CNN outputs a vector y (=(y1, y2, y3, yM)) with ym as an element. As a result, when pixel values ​​of an input image are input as input variables, the CNN outputs a vector y as an output variable. The vector y represents an evaluation value.

[0085] FIG. 11 is an explanatory diagram illustrating a learning procedure in machine learning according to this embodiment. This figure is an explanatory diagram of the case where the CNN in FIG. 10 performs machine learning. For the pixel values ​​of the images in the training dataset, the vector x output from the first hidden layer is defined as vector X. The vector representing the determined class of the training dataset is defined as vector Y.

[0086] An initial value is set for the weight matrix W(i). When an input image is input to the input layer, and as a result, a vector X is input to the second hidden layer, a vector y(X) corresponding to vector X is output from the output layer. The error E between vector y(X) and vector Y is calculated using a loss function. The gradient ΔEi of the i-th layer is calculated using the output zi from each layer and the error signal δi. The error signal δi is calculated using the error signal δi-1. Note that this process of transmitting the error signal from the output layer to the input layer is also called backpropagation. The weight matrix W(i) is updated based on the gradient ΔEi. Similarly, in the first hidden layer, the convolution matrix CM or the weight coefficients are updated.

[0087] (Setting the trained model) The learning unit 540 sets the number of layers, the number of nodes in each layer, the node connection method between each layer, the activation function, the error function, the gradient descent algorithm, the pooling region, the kernel, the weighting coefficient, and the weighting matrix for the CNN. For example, the learning unit 540 sets the number of layers to three (I=3). As the number of nodes in each layer (also referred to as the "number of nodes"), the learning unit 540 sets the number of elements of vector x (number of nodes N) to 800, the number of nodes in the second hidden layer (i=2) to 500, and the number of nodes in the output layer (i=3) to 10. However, the present invention is not limited to this, and the total number may be four or more layers, and a different value may be set for the number of nodes.

[0088] The learning unit 540 sets 20 convolution matrices CM with 5 rows and 5 columns and regions PM with 2 rows and 2 columns. However, the present invention is not limited to this, and a different number of matrices or columns or a different number of convolution matrices CM may be set. Furthermore, a different number of matrices or columns of regions PM may be set. The training unit 540 may perform more convolution processing or pooling processing.

[0089] The training unit 540 sets full connections as the connections for each layer of the neural network. However, the present invention is not limited to this, and the connections for some or all layers may be set to non-full connections. The training unit 540 sets sigmoid functions as the activation functions for all layers. However, the present invention is not limited to this, and the activation functions for each layer may be other activation functions such as step functions, linear combinations, soft sine functions, soft plus functions, ramp functions, truncated power functions, polynomials, absolute values, radial basis functions, wavelets, maxout functions, etc. Furthermore, the activation functions of some layers may be different from those of other layers.

[0090] The learning unit 540 sets squared loss (mean squared error) as the error function. However, the present invention is not limited to this, and the error function may also be cross entropy, τ-quantile loss, Huber loss, or ε-sensitivity loss (ε-tolerant error function). Furthermore, the learning unit 540 sets SGD (stochastic gradient descent) as the algorithm (gradient descent algorithm) for calculating the gradient. However, the present invention is not limited to this, and Momentum (inertia term) SDG, AdaGrad, RMSprop, AdaDelta, Adam (adaptive moment estimation), etc. may also be used as the gradient descent algorithm.

[0091] The learning unit 540 is not limited to a convolutional neural network (CNN), and may set other neural networks such as a perceptron neural network, a recurrent neural network (RNN), a residual neural network (ResNet), etc. Furthermore, the learning unit 540 may set a trained model of supervised learning, such as a decision tree, a regression tree, a random forest, a gradient boosting tree, a linear regression, a logistic regression, or an SVM (support vector machine), to some or all of the models.

[0092] (Variations of learning) In the above embodiment, the machine learning may be supervised learning other than neural network. For example, the learning unit 540 may perform supervised machine learning using not only neural network but also decision tree, regression tree, random forest, gradient boosting tree, linear regression, logistic regression, SVM (support vector machine), etc.

[0093] In the above embodiment, the machine learning may be unsupervised learning. For example, the learning unit 540 may perform unsupervised learning by inputting a large number of learning images of ion exchange resins and performing regression or classification.

[0094] In the above embodiment, the machine learning may be machine learning using reinforcement learning. For example, the learning unit 540 may perform reinforcement learning using a Q value (Q-learning) as the reinforcement learning, or may perform reinforcement learning using the Sarsa or Monte Carlo method.

[0095] [Verification of AI evaluation results] Next, the results of verifying the accuracy of the evaluation results obtained by the ion exchange resin evaluation process using AI according to this embodiment will be described. In this verification, the AI ​​used Python as the programming language and Pytorch as the machine learning library, and deep learning was performed in the machine learning device 50. Furthermore, OpenCV (Open Source Computer Vision Library), a general library for computer vision, was used as the image processing library. The learning images used in the machine learning were photographs of the appearance of ion exchange resins taken in the past by analysts (approximately 2,000 perfect spheres, approximately 200 cracked spheres, and approximately 200 broken spheres).

[0096] Furthermore, as inspection images for verification, 12 samples of ion exchange resin were prepared, which were intentionally selected by the analyst, such as individuals that seemed difficult to classify into any of the evaluation categories (see Figure 7), very average individuals, and individuals with different appearance indices, and 10 photographs were taken with the microscope camera 10 for each sample so that the total number of ion exchange resins (total count) would be 300 or more. Individuals that are difficult to classify into categories are individuals that beginners would likely mistake, such as individuals with a wrinkled appearance, individuals with only a few cracks that should be judged as cracked, and individuals that look like half-moons but are not perfect spheres.

[0097] The performance evaluation device 30 analyzed the verification test images taken by the microscope camera 10 using the trained model that underwent the above-mentioned deep learning, and calculated the appearance index using the above-mentioned formula 2. The results of this AI analysis were then compared with the results of visual analysis by an analyst to verify the accuracy of the ion exchange resin evaluation results using AI. Note that the results of the visual analysis by the analyst were not the verification test images, but rather were obtained by visually checking 300 optical images of ion exchange resins magnified by a microscope and classifying them into perfect spheres, cracked spheres, or broken spheres, and calculating the appearance index using the above-mentioned formula 1.

[0098] Figure 12 shows the results of analysis by AI and the results of analysis by an analyst's visual inspection. In this figure, the results of analysis by the analyst's visual inspection and the results of analysis by AI are shown side by side for each of samples 1 to 12. The analysis results include the total amount of ion exchange resin (total resin count), the appearance index, and the percentage of cracked balls. The appearance index values ​​for samples 2 to 6 and 8 to 10 were identical between the results of analysis by the analyst's visual inspection and the results of analysis by AI. Furthermore, the appearance index values ​​for samples 1, 7, 11, and 12 differed between the results of analysis by the analyst's visual inspection and the results of analysis by AI. However, these differences were within the range of the evaluation categories shown in Figure 7, and therefore the differences were deemed to be at an acceptable level. In other words, the accuracy of the evaluation results obtained by the ion exchange resin evaluation process using AI is not a problem.

[0099] Samples 7 and 12 have more crack balls than the other samples. Even if cracks are present, the crack balls do not affect the resin's performance or cause problems with the equipment (such as pressure loss), so they are not included in the calculation of the appearance index. However, the crack balls are a sign of the final stage of fracture, and can be considered a sign that the appearance index will deteriorate in the future. For this reason, in the analysis results shown, a warning is displayed if the crack balls exceed 3%.

[0100] Figure 13 summarizes a comparison between the results of analysis by AI and those obtained by human visual inspection. The accuracy of the results of analysis by human visual inspection is naturally good, but only if the analyst is an expert. Furthermore, because the judgment is made by humans, there is some variability even among experts. Furthermore, in visual analysis, it takes approximately 13 minutes for an expert to calculate the appearance index per sample, while it takes approximately 25 minutes for a beginner. On the other hand, the accuracy of the results of analysis by AI is similarly good and consistent regardless of who is operating the system, since the analysis is performed by AI. Furthermore, the calculation time for the appearance index per sample by AI analysis is approximately 6 minutes, which is only half the time required for visual analysis. Of these 6 minutes, only approximately 1 minute is spent on the AI ​​analysis, and the remaining time is spent almost entirely on photographing (taking 10 photographs).

[0101] As described above, in the performance evaluation system 1 according to this embodiment, the evaluation unit 362 evaluates the performance of the ion exchange resin under test from an inspection image of the ion exchange resin under test, using a trained model that has been machine-learned based on training images of the ion exchange resin taken for training purposes and the evaluation results of the appearance of the ion exchange resin.

[0102] This allows the performance evaluation system 1 to accurately evaluate the performance (degree of deterioration) of ion exchange resins in a short time from captured images of the ion exchange resins. In addition, because the performance evaluation system 1 evaluates the performance of ion exchange resins using AI, it does not require an expert analyst and can obtain good evaluation results without variation regardless of who operates the system.

[0103] For example, classifications according to the types of appearance of the ion exchange resins are set in advance as the appearance evaluation results. Then, the evaluation unit 362 evaluates the performance of the ion exchange resins to be inspected using a trained model that has been machine-learned based on multiple training images of ion exchange resins captured for each type of appearance.

[0104] This allows the performance evaluation system 1 to accurately evaluate the performance (deterioration level) of the ion exchange resin based on the appearance of the ion exchange resin in a short time.

[0105] Here, the type of appearance is classified based on the presence or absence of cracks and fractures in the appearance of the ion exchange resin (for example, classified into perfect spheres, cracked spheres, and fractured spheres).

[0106] This allows the performance evaluation system 1 to accurately evaluate the performance (deterioration level) of the ion exchange resin in a short time based on the appearance of the ion exchange resin, such as the presence or absence of cracks and fractures.

[0107] Specifically, the evaluation unit 362 inputs inspection images of multiple ion exchange resins for inspection into the trained model, and then analyzes which appearance type each of the ion exchange resins for inspection falls into using the trained model.The evaluation unit 362 then calculates an appearance index that indicates the appearance state of the ion exchange resins for inspection contained in the inspection images, based on the number of ion exchange resins for each classified appearance type (e.g., perfect sphere, cracked sphere, crushed sphere).

[0108] As a result, the performance evaluation system 1 calculates an appearance index based on the number of ion exchange resins for each appearance type (e.g., perfect spheres, cracked spheres, crushed spheres) classified by AI, and therefore the performance (degree of deterioration) of the ion exchange resin can be judged using the same criteria as when an analyst visually observes and analyzes using a microscope.

[0109] The inspection image is an image of a plurality of ion exchange resins for inspection placed in a petri dish (an example of a container) that is magnified and photographed by the microscope camera 10. When this inspection image is photographed by the microscope camera 10, a portion of the ion exchange resins to be photographed that has little overlap with each other is selected and photographed.

[0110] This allows the performance evaluation system 1 to accurately classify the types of appearance of the ion exchange resins to be inspected.

[0111] Furthermore, in the performance evaluation system 1 according to this embodiment, the learning unit 540 performs machine learning based on learning images of ion exchange resins taken for learning purposes and the evaluation results of the appearances of the ion exchange resins.

[0112] As a result, the performance evaluation system 1 can use AI to accurately evaluate the performance (deterioration level) of the ion exchange resin from the captured image of the ion exchange resin in a short time.

[0113] The training images include multiple images of the appearance of the training ion exchange resin captured multiple times while changing at least one of the size, angle, and color tone, or multiple images of the appearance of the ion exchange resin captured multiple times while changing at least one of the size, angle, and color tone. That is, the training images include, for example, images of the appearance of the ion exchange resin captured multiple times while changing at least one of the size, angle, and color tone. Another example of the training images is a plurality of images of the appearance of the ion exchange resin captured by performing at least one image processing operation on the captured images, such as changing the size or angle (e.g., the angle of the rotation direction) of the ion exchange resin in the image, inverting the image, color correcting the image, or extracting (cropping) a portion of the image. In this way, the training images may be captured multiple times while changing the shooting conditions, or may be generated by later changing the conditions through image processing of the captured images, or a combination of these.

[0114] This increases the robustness of the performance evaluation system 1, enabling it to accurately classify the type of appearance of the ion exchange resin under test that is captured in the test image.

[0115] As mentioned above, if there is overlap between the ion exchange resins for inspection contained in the Petri dish 20 when capturing an inspection image, the overlapping portion may be erroneously determined to be a scratch (e.g., a cracked ball), etc. For example, if the ion exchange resin is floating in the water layer, the floating ion exchange resin and the settled ion exchange resin may be photographed overlapping each other.

[0116] When air, oil, or hydrophobic solids adhere to ion exchange resins, the ion exchange resins may float in water or may be repelled by water, causing adjacent ion exchange resins to stick together, resulting in overlapping of the ion exchange resins. For example, organic matter may be trapped by the functional groups of the ion exchange resins, causing adsorption or aggregation of the ion exchange resins, leading to overlapping of the resins. In particular, anionic ion exchange resins are prone to trap organic matter with carboxyl groups, which are abundant in nature, and this effect is particularly pronounced.

[0117] Surfactants have the effect of solubilizing air, oil, or hydrophobic solids by adsorbing to their interfaces. Therefore, the use of surfactants makes floating ion exchange resins more likely to settle in water and reduces repulsion from water, thereby reducing overlapping of ion exchange resins. This reduces erroneous judgments due to overlapping areas of ion exchange resins, improving the accuracy of performance evaluation of ion exchange resins. Among surfactants, anionic surfactants are preferred from the viewpoint of improving the accuracy of performance evaluation of ion exchange resins.

[0118] The following describes the results of an experiment that confirmed the effect of adding a surfactant. In this experiment, anionic surfactants were used. The anionic surfactants used were polyoxyethylene (3) lauryl ether sodium sulfate, which is an alkyl ether sulfate ester (AES), and polyoxyethylene (2) lauryl ether sodium sulfate. It should be noted that commercially available synthetic detergents containing anionic surfactants are also effective.

[0119] (Experimental Method) 1. Place an appropriate amount of ion exchange resin into a petri dish 20. 2. The ion exchange resin contained in the petri dish 20 is photographed with the microscope camera 10, and the photographed inspection image is analyzed with the performance evaluation system 1 to evaluate the appearance of the ion exchange resin. 3. Add an anionic surfactant dropwise to the same Petri dish 20 containing the ion exchange resin and stir thoroughly. 4. As in "2." above, the ion exchange resin contained in the petri dish 20 is photographed with the microscope camera 10, and the photographed inspection image is analyzed with the performance evaluation system 1 to evaluate the appearance of the ion exchange resin. 5. Compare the evaluation results of "2." above with "4." above.

[0120] (Evaluation results) The results of the appearance evaluation before and after the addition of the anionic surfactant will be described with reference to Figures 14 to 17. Figures 14 to 17 are analytical images showing the results of photographing ion exchange resin contained in Petri dish 20 and classifying it into perfect spheres, cracked spheres, or broken spheres, and correspond to the analytical image displayed in screen area 105 in Figure 4.

[0121] First, we will explain the results of the evaluation of the appearance before adding the anionic surfactant in "2." of the above experimental method. In the evaluation of the appearance before adding the anionic surfactant, incorrect judgments occurred in some areas where the ion exchange resins overlapped.

[0122] Figure 14 shows an example of a misidentification of a cracked sphere before the addition of an anionic surfactant. This example shows an example of the analysis results of an inspection image before the addition of an anionic surfactant, in which ion exchange resin is identified as a cracked sphere at three locations indicated by symbols R1, R2, and R3. The ion exchange resin at the location indicated by symbol R1 is correctly identified as a cracked sphere, but at the two locations indicated by symbols R2 and R3, it is misidentified as a cracked sphere due to overlapping of the ion exchange resin.

[0123] Figure 15 shows an example of a misidentification of broken spheres before the addition of an anionic surfactant. This example shows an example of the analysis results of the test image before the addition of an anionic surfactant, and ion exchange resin was identified as broken spheres at three locations indicated by symbols R4, R5, and R6. The ion exchange resin at the location indicated by symbol R4 was correctly identified as broken spheres and cracked spheres, but at the two locations indicated by symbols R5 and R6, it was misidentified as broken spheres due to overlapping ion exchange resin particles.

[0124] 16 shows an example in which overlapping resin portions cannot be determined before the addition of an anionic surfactant. The example shown in this figure shows an example of the analysis results of an inspection image before the addition of an anionic surfactant, and in the range indicated by the symbol R7 (the range indicated by the solid circle), there is a resin whose appearance cannot be determined due to overlapping of multiple ion exchange resins.

[0125] Next, the results of the evaluation of appearance after adding the anionic surfactant in "4." above will be described. Figure 17 shows an example of the analysis results of the inspection image after adding an anionic surfactant. In this example, adding an anionic surfactant reduced the overlapping areas of the ion exchange resin, allowing for accurate assessment of the appearance. In this example, the ion exchange resin was correctly identified as a cracked sphere at three locations indicated by the symbols R8, R9, and R10. Furthermore, the ion exchange resin without cracks or fractures was correctly identified as a perfect sphere.

[0126] According to these evaluation results, in the analysis of the appearance of ion exchange resins before the addition of anionic surfactants, overlapping resin particles were sometimes mistakenly identified as cracked or broken particles, or were not identified at all. As a result, the calculated number of cracked or broken particles was increased, or the accurate number of ion exchange resin particles was not calculated, and the appearance index was not accurately calculated. On the other hand, after the addition of anionic surfactants, the overlapping resin particles decreased, so the number of perfect spheres, cracked spheres, and broken spheres was accurately calculated, and the appearance index was accurately calculated, improving the accuracy of the performance evaluation of ion exchange resins.

[0127] In this way, before the inspection image is captured by the microscope camera 10 (an example of a magnifying camera), a surfactant (e.g., an anionic surfactant) is added to a petri dish 20 (an example of a container) containing multiple ion exchange resins for inspection, the mixture is stirred, and then the image is captured by the microscope camera 10, thereby improving the accuracy of the performance evaluation of the ion exchange resins.

[0128] Although anionic surfactants were used in the above experiments, surfactants other than anionic surfactants may also be used, although anionic surfactants are preferred in terms of their effectiveness in reducing overlapping of ion exchange resin particles and their ease of availability.

[0129] Although an example of adding an anionic surfactant has been described here, other surfactants may also be used, although the use of an anionic surfactant is preferred in terms of reducing overlap of ion exchange resin particles and ease of availability.

[0130] One embodiment of the present invention has been described in detail above with reference to the drawings, but the specific configuration is not limited to that described above, and various design changes and the like are possible within the scope that does not deviate from the gist of the present invention.

[0131] In the above embodiment, the appearance index of a sample is calculated by using the number of perfect spheres and the number of broken spheres in the sample, and calculating the ratio of the number of perfect spheres to the total number of perfect spheres and broken spheres as the appearance index. However, this is not limited to this. For example, the appearance index may be calculated by calculating the ratio of the number of perfect spheres to the total number of perfect spheres, broken spheres, and cracked spheres.

[0132] In the above embodiment, the ion exchange resins are classified into three appearance types, namely, perfect spheres, cracked spheres, and broken spheres, based on their appearance. However, the classification by appearance type may be limited to two types (e.g., perfect spheres and broken spheres). For example, the appearance types may be classified based on whether or not the ion exchange resins are broken (e.g., classified into perfect spheres and broken spheres). In this case, the ratio of the number of perfect spheres to the total number of perfect spheres and broken spheres may be calculated as the appearance index. The appearance types may also be classified based on whether or not the ion exchange resins have cracks (e.g., classified into perfect spheres and cracked spheres). In this case, the ratio of the number of perfect spheres to the total number of perfect spheres and cracked spheres may be calculated as the appearance index. In other words, the appearance types may be classified based on at least one of the presence or absence of cracks and the presence or absence of breakage in the appearance of the ion exchange resin. The classification by appearance type may also be limited to four or more types. For example, when gel-type ion exchange resins deteriorate, their appearance may rarely become wrinkled. This wrinkled state may be added to the appearance types and machine learning may be performed. Also, appearance types indicating appearance states other than those described above may be added.

[0133] Furthermore, in the above embodiment, the camera 10b is not limited to a dedicated camera such as a digital camera, but may be an electronic device that has a camera function as part of its functions, such as a smartphone. Furthermore, the camera 10b and the performance evaluation device 30 do not need to be connected for communication, and the image (such as an inspection image) captured by the camera 10b may be transferred to the performance evaluation device 30 via a storage medium such as an optical disk or a memory card.

[0134] Furthermore, the camera 10b and the performance evaluation device 30 may be integrated into one device. Furthermore, the microscope camera 10 and the performance evaluation device 30 may be integrated into one device.

[0135] In the above embodiment, the ion exchange resin sample placed in a petri dish was photographed or observed with the microscope camera 10, but a container other than a petri dish may be used. However, it is desirable that the container have a shape that makes it difficult for the ion exchange resin to overlap (for example, a shape with as large a flat area as possible).

[0136] In the above embodiment, an example has been described in which an evaluation process for ion exchange resins is performed using a trained model generated using training images of ion exchange resins and evaluation results (evaluation values) of their appearances as training data, but the present invention is not limited to this. For example, the evaluation process for ion exchange resins may be performed using a data table in which training images of ion exchange resins are associated with evaluation results (evaluation values) of their appearances, or the evaluation process for ion exchange resins may be performed using a program that embodies an algorithm regarding the relationship between training images of ion exchange resins and evaluation results (evaluation values) of their appearances.

[0137] The camera 10b, the performance evaluation device 30, or the machine learning device 50 included in the performance evaluation system 1 described above may each have an internal computer system. A program for implementing the functions of each component of the camera 10b, the performance evaluation device 30, or the machine learning device 50 may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be loaded into a computer system and executed to perform processing in each component of the camera 10b, the performance evaluation device 30, or the machine learning device 50. Here, "loading a program recorded on a recording medium into a computer system and executing it" includes installing the program into a computer system. The term "computer system" here includes hardware such as an OS and peripheral devices. The term "computer system" may also include multiple computer devices connected via a network, including communication lines such as the Internet, a WAN, a LAN, or a dedicated line. The term "computer-readable recording medium" refers to portable media such as a flexible disk, a magneto-optical disk, a ROM, or a CD-ROM, or a storage device such as a hard disk built into a computer system. In this way, the recording medium storing the program may be a non-transitory recording medium such as a CD-ROM.

[0138] The recording medium also includes internal or external recording media accessible from a distribution server for distributing the program. The program may be divided into multiple parts, downloaded at different times, and then combined by the components of the camera 10b, the performance evaluation device 30, or the machine learning device 50. Alternatively, each divided program may be distributed by a different distribution server. Furthermore, the term "computer-readable recording medium" also includes a storage medium that stores a program for a certain period of time, such as volatile memory (RAM) within a computer system that serves as a server or client when a program is transmitted over a network. The program may also be a storage medium for implementing part of the above-described functions. Furthermore, the program may be a so-called differential file (differential program) that can realize the above-described functions in combination with a program already stored in the computer system.

[0139] Furthermore, some or all of the camera 10b, the performance evaluation device 30, or the machine learning device 50 may be realized as an integrated circuit such as an LSI (Large Scale Integration). Furthermore, each component in the camera 10b, the performance evaluation device 30, or the machine learning device 50 of this embodiment may be individually implemented as a processor, or some or all of them may be integrated into a processor. The integrated circuit implementation method is not limited to LSI, and may be implemented using a dedicated circuit or a general-purpose processor. Furthermore, if an integrated circuit implementation technology that replaces LSI emerges due to advances in semiconductor technology, an integrated circuit based on that technology may be used.

[0140] <Other Examples> In the above embodiment, an example has been described in which an ion exchange resin sample for testing sent from a facility equipped with an ion exchange resin tower is placed in a petri dish 20 and photographed with the microscope camera 10 to evaluate the performance of the ion exchange resin. However, a magnified image of the ion exchange resin in the ion exchange resin tower may also be transmitted to the performance evaluation device 30. In this case, the performance evaluation device 30 may be located in a facility equipped with the ion exchange resin tower, or in a facility of a vendor that manufactures ion exchange resins and provides them to customers. In this embodiment, a system will be described in which a magnified image of the ion exchange resin in the ion exchange resin tower is transmitted to the performance evaluation device 30.

[0141] FIG. 18 is a system diagram showing an example of the configuration of a performance evaluation system 1A according to this embodiment. The performance evaluation system 1A includes a performance evaluation device 30 and a water treatment facility 100. The basic configuration of the performance evaluation device 30 is similar to that shown in FIG. 6. The water treatment facility 100 is a facility that produces ultrapure water used for various purposes, such as cleaning water used in the manufacturing processes of semiconductors, liquid crystals, wafers, precision parts, etc., desalinated water produced in power plant condensate demineralizers, and water for pharmaceutical production. The water treatment facility 100 includes an ion-exchange resin tower 101, a communication unit 110, an imaging unit 120, a memory unit 130, and a control unit 140.

[0142] The communication unit 110 includes a communication device compatible with communication standards such as wireless LAN and mobile communication, and communicates (transmits or receives) with the performance evaluation device 30 and other devices via the communication network NW. The imaging unit 120 includes an imaging element and an optical lens provided in front of the imaging surface of the imaging element. The optical lens is a magnifying glass for magnifying and photographing the ion exchange resin. The imaging unit 120 photographs the ion exchange resin in the ion exchange resin tower 101 as a test sample. For example, a transparent window is provided in a part of the bottom or side of the ion exchange resin tower 101, and the imaging unit 120 photographs a large number of ion exchange resins present in the locations corresponding to the window as test samples. At this time, multiple captured images may be acquired by automatically or manually changing the angle of view of the imaging unit 120. The storage unit 130 includes, for example, an HDD, SSD, EEPROM, ROM, RAM, etc., and stores various information, images, programs, etc.

[0143] The control unit 140 transmits the photographed image (inspection image) of the inspection ion exchange resin in the ion exchange resin tower 101 photographed by the imaging unit 120 to the performance evaluation device 30 via the communication unit 110. The communication unit 31 of the performance evaluation device 30 receives the photographed image (inspection image) of the inspection ion exchange resin transmitted from the water treatment facility 100. The evaluation unit 362 of the performance evaluation device 30 evaluates the performance of the inspection ion exchange resin from the received inspection image. Specifically, as described above, the performance of the inspection ion exchange resin from the inspection image is evaluated using a trained model that has been machine-learned based on the training image of the ion exchange resin photographed for training purposes and the evaluation results of the appearance of the ion exchange resin.

[0144] Furthermore, the performance evaluation device 30 may transmit information indicating the evaluation results and analysis results of the test images evaluated by the evaluation unit 362 to the water treatment facility 100. In this way, the control unit 140 acquires information indicating the evaluation results and analysis results of the test images evaluated by the evaluation unit 362 of the performance evaluation device 30 from the performance evaluation device 30. For example, the control unit 140 may display information based on the evaluation results and analysis results acquired from the performance evaluation device 30 (e.g., information such as the appearance index value, evaluation category, and need for replacement) on a display unit (not shown) within the water treatment facility 100, or may transmit the information to a terminal used by a facility manager of the water treatment facility 100, for example.

[0145] Furthermore, when replacement of the ion exchange resin is necessary based on the evaluation results of the inspection image acquired from the performance evaluation device 30, the control unit 140 may place an order for replacement ion exchange resin with the vendor that provides the ion exchange resin. For example, the control unit 140 may generate order information such as the type, quantity, and delivery destination of the replacement ion exchange resin, and transmit it to the vendor's service department via the communication unit 110. At this time, the control unit 140 may transmit the order information after confirming that the ordering department has approved it.

[0146] In this way, the performance evaluation system 1A can determine the performance (degree of deterioration) of the ion exchange resin in the ion exchange resin tower 101 based on the test image taken in the ion exchange resin tower 101 of the water treatment facility 100, which is convenient because it eliminates the need to send samples for testing and saves time.

[0147] In addition, the machine learning device 50 may acquire the test images taken by the water treatment facility 100 and the evaluation results obtained by the performance evaluation device 30 for the test images, and use the acquired test images as learning images to perform further machine learning on the trained model used for evaluation by the evaluation unit 362 based on the learning images and the evaluation results.

[0148] This allows the ion exchange resin used in the ion exchange resin tower 101 to be continuously evaluated and the evaluation results to be updated at any time, thereby enabling the ion exchange resin to be evaluated more accurately. [Explanation of symbols]

[0149] 1, 1A performance evaluation system, 10 microscope camera, 10a microscope, 10b camera, 11 communication unit, 12 imaging unit, 13 memory unit, 14 operation unit, 15 control unit, 20 petri dish, 30 performance evaluation device, 30a monitor, 30b keyboard, 31 communication unit, 32 video output unit, 33, 34 USB connector, 35 memory unit, 36 control unit, 351 inspection image memory unit, 352 learning model memory unit, 353 evaluation data memory unit, 361 inspection image acquisition unit, 362 evaluation unit, 363 output control unit, 100 water treatment equipment, 101 ion exchange resin tower, 110 communication unit, 120 imaging unit, 130 memory unit, 140 control unit

Claims

1. an evaluation unit that evaluates the performance of an ion exchange resin for inspection from an inspection image of the ion exchange resin for inspection, using a trained model that has been machine-learned based on a training image of the ion exchange resin taken for training purposes and an evaluation result of the appearance of the ion exchange resin; Equipped with a classification according to the type of appearance of the ion exchange resin is preset as a result of the appearance evaluation; The evaluation unit The test images of the ion exchange resins for inspection, each of which has been photographed, are input to the trained model, which has been machine-learned based on the training images of the ion exchange resins photographed for each appearance type, and the trained model is used to analyze which of the appearance types each of the ion exchange resins for inspection falls into; and an appearance index indicating the appearance state of the ion exchange resins for inspection included in the test images is calculated based on the number of ion exchange resins classified for each appearance type, thereby evaluating the performance of the ion exchange resins for inspection. Performance evaluation system.

2. The type of appearance is classified based on at least one of the presence or absence of cracks and the presence or absence of fractures in the appearance of the ion exchange resin. The performance evaluation system according to claim 1 .

3. The inspection image is an image of a plurality of the inspection ion exchange resins placed in a container, enlarged and photographed by a magnifying camera, and when photographing with the magnifying camera, a portion of the ion exchange resins to be photographed that has little overlap with each other is selected and photographed.

3. The performance evaluation system according to claim 1 or 2.

4. The inspection image is an image taken with a magnifying camera after adding a surfactant to a container containing the plurality of inspection ion exchange resins and stirring the mixture before the inspection image is taken with the magnifying camera. The performance evaluation system according to any one of claims 1 to 3.

5. a learning unit that performs machine learning based on the learning image of the ion exchange resin photographed for learning and an evaluation result of the appearance of the ion exchange resin; The performance evaluation system according to claim 1 , comprising:

6. The learning images include images of the appearance of the learning ion exchange resin photographed multiple times while changing at least one of the size, angle, and color tone, or multiple images of the appearance of the ion exchange resin photographed while changing at least one of the size, angle, and color tone. The performance evaluation system according to any one of claims 1 to 5.

7. A method for evaluating the performance of an ion exchange resin, comprising: acquiring an inspection image of the inspection ion exchange resin; a step of evaluating the performance of the ion exchange resin for inspection from the inspection image using a trained model that has been machine-learned based on a training image of the ion exchange resin taken for training and an evaluation result of the appearance of the ion exchange resin; and a classification according to the type of appearance of the ion exchange resin is preset as a result of the appearance evaluation; In the step of evaluating the performance of the test ion exchange resin, The test images of the ion exchange resins for inspection, each of which has been photographed, are input to the trained model, which has been machine-learned based on the training images of the ion exchange resins photographed for each appearance type, and the trained model is used to analyze which of the appearance types each of the ion exchange resins for inspection falls into; and an appearance index indicating the appearance state of the ion exchange resins for inspection included in the test images is calculated based on the number of ion exchange resins classified for each appearance type, thereby evaluating the performance of the ion exchange resins for inspection. Performance evaluation method.

8. On the computer, acquiring an inspection image of the inspection ion exchange resin; a step of evaluating the performance of the ion exchange resin for inspection from the inspection image using a trained model that has been machine-learned based on a training image of the ion exchange resin taken for training and an evaluation result of the appearance of the ion exchange resin; A program for executing a classification according to the type of appearance of the ion exchange resin is preset as a result of the appearance evaluation; In the step of evaluating the performance of the test ion exchange resin, The test images of the ion exchange resins for inspection, each of which has been photographed, are input to the trained model, which has been machine-learned based on the training images of the ion exchange resins photographed for each appearance type, and the trained model is used to analyze which of the appearance types each of the ion exchange resins for inspection falls into; and an appearance index indicating the appearance state of the ion exchange resins for inspection included in the test images is calculated based on the number of ion exchange resins classified for each appearance type, thereby evaluating the performance of the ion exchange resins for inspection. program.

9. A trained model for evaluating the performance of an ion exchange resin for inspection from a plurality of inspection images of the ion exchange resin for inspection, As a result of the appearance evaluation, classification according to the type of appearance of the ion exchange resin is preset, machine learning is performed based on learning images of a plurality of ion exchange resins photographed for each appearance type for learning purposes and the evaluation results of the appearances of the ion exchange resins; when the test images are input, an analysis is performed to determine which of the appearance types each of the test ion exchange resins falls into; and an appearance index indicating the appearance state of the test ion exchange resins included in the test images is calculated based on the number of ion exchange resins classified for each appearance type, thereby evaluating the performance of the test ion exchange resins. A trained model that makes a computer work.

10. A performance evaluation system including a water treatment facility and a performance evaluation device, The water treatment facility comprises: an ion exchange resin tower; an imaging unit that captures an image of the ion exchange resin in the ion exchange resin tower; a transmitting unit that transmits the image captured by the imaging unit as an inspection image; Equipped with The performance evaluation device a receiving unit that receives the inspection image; an evaluation unit that evaluates the performance of the ion exchange resin for inspection from the inspection image using a trained model that has been machine-learned based on a training image of the ion exchange resin taken for training and an evaluation result of the appearance of the ion exchange resin; Equipped with a classification according to the type of appearance of the ion exchange resin is preset as a result of the appearance evaluation; The evaluation unit The test images of the ion exchange resins for inspection, each of which has been photographed, are input to the trained model, which has been machine-learned based on the training images of the ion exchange resins photographed for each appearance type, and the trained model is used to analyze which of the appearance types each of the ion exchange resins for inspection falls into; and an appearance index indicating the appearance state of the ion exchange resins for inspection included in the test images is calculated based on the number of ion exchange resins classified for each appearance type, thereby evaluating the performance of the ion exchange resins for inspection. Performance evaluation system.

11. The performance evaluation device transmitting information based on the evaluation result by the evaluation unit to the water treatment facility; The performance evaluation system according to claim 10.

12. The water treatment facility comprises: generating order information for a replacement ion exchange resin based on the information acquired from the performance evaluation device if replacement of the ion exchange resin is necessary; The performance evaluation system according to claim 11.

Images