Machine learning device, inspection device, machine learning method and program

Abstract

To provide a machine learning device for more satisfactorily performing inspection of a membrane electrode assembly, an inspection device, a machine learning method and a program.SOLUTION: In an inspection system SYS, a machine learning device 100 includes: a data set acquisition unit 108 for acquiring foreign matter data 14 including at least one of area data 16, brightness data 18, and shape data 20 of a foreign material 42 grasped from an image of the foreign matter mixed in a membrane electrode assembly, and teacher data 40 including data showing the volume of the foreign matter; and a learning model generation unit 112 for generating a learning model 10 with the foreign matter data as an input and with the volume of the foreign matter as an output by executing supervised learning by using the teacher data acquired by the data set acquisition unit.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a machine learning device, an inspection device, a machine learning method, and a program.

Background Art

[0002] In recent years, in order to enable more people to access affordable, reliable, sustainable, and advanced energy, research and development on fuel cells that contribute to energy efficiency have been carried out. Japanese Patent Application Laid-Open No. 2021-135125 discloses an inspection method for a membrane electrode assembly. According to the disclosure, the inspection device determines the presence or absence of foreign matter in the membrane electrode assembly based on the amount of decrease in luminance in the X-ray transmission image.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Recently, there has been a long-felt need for a technique for better inspecting a membrane electrode assembly. [[ID=3,7]]

[0005] The present disclosure aims to solve the above-described problems.

Means for Solving the Problems

[0006] A first aspect of this disclosure is a machine learning device comprising: an acquisition unit that acquires training data including foreign object data, which includes at least one of area data indicating the area of ​​a foreign object mixed into a membrane electrode assembly, brightness data indicating the brightness of the foreign object as determined from the image, and shape data relating to the shape of the foreign object as determined from the image, and data indicating the volume of the foreign object; and a learning model generation unit that generates a learning model that takes the foreign object data as input and outputs the volume of the foreign object by performing supervised learning using the training data acquired by the acquisition unit.

[0007] A second aspect of the present disclosure is an inspection apparatus comprising: a foreign matter data acquisition unit that acquires foreign matter data including at least one of area data indicating the area of ​​a foreign matter found from an image of a foreign matter mixed into a membrane electrode assembly, brightness data indicating the brightness of the foreign matter found from the image, and shape data relating to the shape of the foreign matter found from the image; and an estimation unit that estimates the volume of the foreign matter using the foreign matter data acquired by the foreign matter data acquisition unit and the learning model generated by a machine learning apparatus according to the first aspect of the present disclosure.

[0008] A third aspect of this disclosure is a machine learning method performed by a computer, comprising: an acquisition step of acquiring training data including foreign object data, which includes at least one of area data indicating the area of ​​a foreign object mixed into a membrane electrode assembly, brightness data indicating the brightness of the foreign object as determined from the image, and shape data relating to the shape of the foreign object as determined from the image, and data indicating the volume of the foreign object; and a learning model generation step of generating a learning model that takes the foreign object data as input and outputs the volume of the foreign object by performing supervised learning using the training data acquired in the acquisition step.

[0009] A fourth aspect of this disclosure is a program for causing the computer to execute the machine learning method relating to the third aspect of this disclosure. [Effects of the Invention]

[0010] This disclosure provides a machine learning apparatus, an inspection apparatus, a machine learning method, and a program for better inspection of membrane electrode assemblies. [Brief explanation of the drawing]

[0011] [Figure 1] Figure 1 is a diagram showing the configuration of an inspection system according to one embodiment. [Figure 2] Figure 2A is a schematic diagram of foreign matter mixed into a membrane electrode assembly. Figure 2B is a schematic diagram of foreign matter mixed into a membrane electrode assembly. [Figure 3] Figure 3 shows the frequency distribution of the diameter ratio. [Figure 4] Figure 4 shows the frequency distribution of area. [Figure 5] Figure 5 is a flowchart of a machine learning method according to one embodiment. [Figure 6] Figure 6 is a flowchart of an inspection method according to one embodiment. [Modes for carrying out the invention]

[0012] A membrane electrode assembly (MEA) is a structure comprising a solid polymer membrane and a catalyst layer laminated on the solid polymer membrane. Membrane electrode assemblies are used, for example, in fuel cells. Foreign matter can be mixed into membrane electrode assemblies. For example, iron particles originating from the manufacturing equipment may be mixed into the membrane electrode assembly during the manufacturing process. Foreign matter may be mixed in, for example, between the solid polymer membrane and the catalyst layer.

[0013] If a relatively large foreign object is mixed into a membrane electrode assembly, the assembly should be judged as defective. However, accurately determining the volume of a foreign object through non-destructive testing is not easy. If inspection standards are set too strictly to prevent defective products from being shipped as good products, good products may be judged as defective, leading to a decrease in yield.

[0014] Based on the above preliminary description, an embodiment will be described below.

[0015] Note that the computer program (computer software) in the following description is also referred to as a computer program product. A computer program product is not limited to a computer program recorded on a recording medium, and also includes a computer program transmitted, distributed, or downloaded via an information communication network such as the Internet.

[0016] (One Embodiment) FIG. 1 is a configuration diagram of an inspection system SYS according to an embodiment.

[0017] The inspection system SYS is a system that performs quality inspection of a membrane electrode assembly provided in a fuel cell or the like. The inspection system SYS includes a machine learning device 100 and an inspection device 200.

[0018] The machine learning device 100 is a device that executes machine learning to generate a learning model (trained model) 10. A more detailed description of the learning model 10 will be given later. The machine learning device 100 is an electronic device such as a computer, for example. The machine learning device 100 includes an operation unit 102, a storage unit 104, and an arithmetic unit 106.

[0019] The operation unit 102 includes an input device (not shown). The input device includes a pointing device, a keyboard, and the like. The pointing device includes, for example, a mouse, a track pad, etc., but is not limited thereto.

[0020] The storage unit 104 includes one or more memories. The one or more memories include a non-volatile memory. The non-volatile memory is a recording medium that non-temporarily stores programs (computer programs), tables, maps, etc. For example, ROM (Read Only Memory), flash memory, etc. are included in the non-volatile memory. The storage unit 104 (one or more memories) may further include a volatile memory. For example, RAM (Random Access Memory) is included in the volatile memory. Also, at least a part of the storage unit 104 may be realized by a recording medium such as a USB (Universal Serial Bus) memory, a memory card, an optical disk, etc.

[0021] As shown in FIG. 1, a learning model 10 is stored in the storage unit 104. The learning model 10 can be realized by, for example, a machine learning algorithm applicable to regression problems. The machine learning algorithm is a support vector machine (for example, a support vector regression model), a neural network (for example, a deep neural network), etc., but is not limited thereto. The learning model 10 outputs volume estimation data corresponding to the foreign object data 14 when the foreign object data 14 is input. The foreign object data 14 is data grasped from an image of a foreign object 42 (see also FIGS. 2A and 2B) mixed in the membrane electrode assembly. A more detailed description of the foreign object data 14 will be given later. The volume estimation data is data indicating the volume (estimated value of the volume) of the foreign object 42.

[0022] The arithmetic unit 106 includes a predetermined processing circuit (Processing circuitry) not shown. The processing circuit includes, for example, one or more processors such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit). The processing circuit may include a predetermined integrated circuit such as an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array).

[0023] The arithmetic unit 106 comprises a dataset acquisition unit (acquisition unit) 108, a training data selection unit 110, and a learning model generation unit 112. The dataset acquisition unit 108, the training data selection unit 110, and the learning model generation unit 112 are realized by the arithmetic unit 106 (processor) executing a program stored in the storage unit 104 (memory). At least a part of the dataset acquisition unit 108, the training data selection unit 110, and the learning model generation unit 112 may be realized by integrated circuits such as the ASIC and FPGA mentioned above.

[0024] The dataset acquisition unit 108 acquires the dataset 12. The dataset 12 includes foreign object data 14 for training and volume data 38 corresponding to the foreign object data 14.

[0025] As described above, the foreign object data 14 is data obtained from an image of the foreign object 42. The image includes, for example, an X-ray transmission image of the foreign object 42. The foreign object 42 (membrane electrode assembly), which is the subject of the image, is obtained, for example, by actually operating the manufacturing equipment described above. The foreign object data 14 includes, for example, at least one of area data 16, luminance data 18, and shape data 20. The area data 16 is data indicating the area (projected area) of the foreign object 42. The luminance data 18 is data indicating the luminance (luminance value) of the foreign object 42. The shape data 20 is data relating to the shape of the foreign object 42. The area data 16 and the luminance data 18 are obtained by applying a predetermined image analysis process to the image of the foreign object 42.

[0026] The shape data 20 includes at least one of the following: perimeter data 24, skewness data 26, kurtosis data 28, and diameter data 30. Perimeter data 24 is data relating to the perimeter of the foreign object 42. The perimeter of the foreign object 42 is, for example, the length of the contour line BL of the foreign object 42 in a plan view of the image, but it may also be the perimeter of the virtual ring VR described later, or the perimeter of the virtual quadrilateral VQ described later (see also Figures 2A and 2B). Skewness data 26 is data relating to the skewness of the foreign object 42. Kurtosis data 28 is data relating to the kurtosis of the foreign object 42. Diameter data 30 is data relating to the diameter of the virtual ring VR, which is a virtual ring that circumscribes (approximates) the foreign object 42 in a plan view, or the ferret diameter of the foreign object 42 in a plan view. The virtual ring VR is a virtual ellipse VE, but it may also be a virtual circle. The ferret diameter may be the maximum ferret diameter or the minimum ferret diameter. The diameter data 30 may indicate the ferret diameter of the foreign object 42 along any direction in a plan view of the image.

[0027] The skewness data 26, kurtosis data 28, and diameter data 30 are obtained by applying a predetermined image analysis process to the image of the foreign object 42, similar to the area data 16, brightness data 18, etc.

[0028] Figure 2A is a schematic diagram of foreign matter 42 mixed into the membrane electrode assembly. Figure 2B is a schematic diagram of foreign matter 42 mixed into the membrane electrode assembly. A virtual ellipse VE circumscribing the foreign matter 42 is shown in Figure 2A. In contrast, a virtual quadrilateral (virtual quadrilateral VQ) circumscribing the foreign matter 42 is shown in Figure 2B.

[0029] The diameter data 30 may include a first diameter data 32 and a second diameter data 34. The first diameter data 32 is data representing the major axis A1 of the virtual ellipse VE. In contrast, the second diameter data 34 is data representing the minor axis A2 of the virtual ellipse VE. The diameter data 30 may further include ratio data 36. The ratio data 36 is data representing the ratio (ellipticity) of the major axis A1 to the minor axis A2.

[0030] The first diameter data 32 may represent the first ferret diameter F1. In this case, the second diameter data 34 may represent the second ferret diameter F2. The first ferret diameter F1 is the ferret diameter of the foreign object 42 in the first direction D1. The first direction D1 is a predetermined direction in a plan view of the image. In contrast, the second ferret diameter F2 is the ferret diameter of the foreign object 42 in the second direction D2. The second direction D2 is a direction that intersects the first direction D1 in a plan view of the image. In this case, the ratio data 36 may represent the ratio (aspect ratio) of the first ferret diameter F1 and the second ferret diameter F2.

[0031] The second direction D2 is, for example, perpendicular to the first direction D1. In this case, the first ferret diameter F1 may correspond to the length of one side of a virtual rectangle VQ (Figure 2B). The second ferret diameter F2 may correspond to the length of the other side of the virtual rectangle VQ that is perpendicular to the aforementioned side. Note that the virtual rectangle VQ is not limited to a rectangle.

[0032] In the following explanation, the ellipticity or aspect ratio shown by the ratio data 36 mentioned above is also collectively referred to as the diameter ratio. In other words, the diameter ratio in the following explanation is the information shown by the ratio data 36.

[0033] The volume data 38 represents the volume of the foreign matter 42 mixed into the membrane electrode assembly. The volume data 38 is obtained by actually measuring the volume of the foreign matter 42, which is the subject of the image described above. Therefore, the volume data 38 can be obtained, for example, by experiment.

[0034] The dataset 12, which includes the foreign object data 14 and volume data 38 described above, is provided to the machine learning device 100, for example, by a user of the machine learning device 100. The dataset acquisition unit 108 can acquire multiple datasets 12. The dataset acquisition unit 108 may store the acquired datasets 12 in the storage unit 104.

[0035] The training data selection unit 110 selects training data 40 to be used for machine learning from among multiple datasets 12 acquired by the dataset acquisition unit 108. The training data selection unit 110 may select multiple training data 40 from among the multiple datasets 12. The selection criteria for training data 40 by the training data selection unit 110 can be determined in advance as appropriate. The training data selection unit 110 selects training data 40 based on, for example, the conditions (Condition 1, Condition 2) described below. The training data selection unit 110 may select a dataset 12 that satisfies at least one of Condition 1 and Condition 2 as training data 40.

[0036] (Condition 1) The diameter ratio shown by ratio data 36 is within a predetermined range.

[0037] Figure 3 shows the frequency distribution of the diameter ratio.

[0038] In the following explanation, the range mentioned above relating to condition 1 is also referred to as the tolerance ratio range RR. The tolerance ratio range RR can be determined in advance as appropriate, but it may also be determined based on the statistics of multiple datasets 12 acquired by the dataset acquisition unit 108. More specifically, the distribution (frequency distribution) of the frequency (probability of occurrence) of diameter ratios can be derived based on the multiple datasets 12. Based on this distribution, a sample group GR of diameter ratios can be extracted. This sample group GR consists of diameter ratios whose probability of occurrence is greater than or equal to a predetermined probability. The sample group GR includes the mode of the diameter ratios, Rmode. Based on the lower limit Rmin of the diameter ratios and the upper limit Rmax of the diameter ratios in this sample group GR, the tolerance ratio range RR can be determined. The predetermined probability mentioned above can be determined as appropriate, but for example, it is 99.7%. The distribution of diameter ratios may also be derived including the ellipticity of a virtual circle by deliberately defining the ellipticity of a virtual circle. The ellipticity of a virtual circle can be defined as, for example, 1.00 (1.000…).

[0039] (Condition 2) The area shown by the area data 16 must be within a predetermined range.

[0040] Figure 4 shows the frequency distribution of area.

[0041] In the following description, the range relating to condition 2 is also referred to as the allowable area range RA. The lower limit Amin of the allowable area range RA can be determined in advance as appropriate, but may also be determined based on the lower limit of the resolution of the inspection device 200 included in the inspection system SYS. The upper limit Amax of the allowable area range RA can be determined in advance as appropriate, but may also be determined based on the statistics of multiple data sets 12 acquired by the data set acquisition unit 108. For example, the upper limit Amax of the allowable area range RA may be determined based on the maximum area of ​​the foreign matter 42 identified based on the multiple data sets 12. In this case, the maximum area of ​​the foreign matter 42 may be the maximum area of ​​the foreign matter 42 within the sample group GR described above.

[0042] The learning model generation unit 112 generates a learning model 10 by performing supervised learning using the training data 40. The learning model generation unit 112 can perform supervised learning using multiple training data 40 selected by the training data selection unit 110. The learning model generation unit 112 performs machine learning with the goal of obtaining a learning model 10 that represents the relationship between foreign object data 14 included in the training data 40 and volume data 38 included in the training data 40. In this case, the foreign object data 14 can be used as input data that shows features for learning. The volume data 38 can also be used as a label corresponding to the input data. As a result, the learning model generation unit 112 generates a learning model 10 that takes foreign object data 14 as input and outputs volume estimation data (volume of foreign object 42).

[0043] The learning model generation unit 112 stores the generated learning model (trained model) 10 in the storage unit 104 described above. The learning model 10, which is the trained model stored in the storage unit 104, is used by the inspection device 200 for quality inspection of the membrane electrode assembly.

[0044] Figure 5 is a flowchart of a machine learning method according to one embodiment.

[0045] The machine learning device 100 described above is capable of executing the machine learning method shown in Figure 5. This machine learning method can be realized, for example, by having the processing unit 106 (processor) of the machine learning device 100 execute a program stored in the memory unit 104 of the machine learning device 100, which is a computer. As shown in Figure 5, the machine learning method includes a dataset acquisition step (acquisition step) S11, a training data selection step S12, and a learning model generation step S13.

[0046] In the dataset acquisition step S11, the dataset acquisition unit 108 acquires a dataset 12. The dataset 12 includes foreign object data 14 for training and volume data 38 corresponding to the foreign object data 14. The dataset acquisition unit 108 may acquire multiple datasets 12.

[0047] In the training data selection step S12, the training data selection unit 110 selects a dataset 12 to be used as training data 40 from among multiple datasets 12. The training data selection unit 110 may select multiple datasets 12 (multiple training data 40).

[0048] In the learning model generation step S13, the learning model generation unit 112 performs machine learning using the multiple training data 40 selected in the training data selection step S12. As a result, the learning model generation unit 112 generates a learning model 10. The generated learning model (trained model) 10 is stored in the storage unit 104.

[0049] The inspection device 200 (Figure 1) is a device that inspects the quality of a membrane electrode assembly using a learning model 10 generated by the machine learning device 100. The inspection device 200 is an electronic device such as a computer. The inspection device 200 may also be a server device. The inspection device 200 comprises an operation unit 202, a storage unit 204, and a calculation unit 206.

[0050] The operation unit 202 includes an input device (not shown). This input device includes a pointing device, a keyboard, etc. The pointing device includes, but is not limited to, a mouse, a trackpad, etc.

[0051] The storage unit 204 includes one or more memory locations. These one or more memory locations include non-volatile memory. Non-volatile memory is a recording medium that stores programs, tables, maps, etc., on a non-temporary basis. Examples of non-volatile memory include ROM and flash memory. The storage unit 204 (one or more memory locations) may further include volatile memory. For example, RAM is included in volatile memory. Furthermore, at least a portion of the storage unit 204 may be implemented by recording media such as USB memory, memory cards, or optical discs.

[0052] The arithmetic unit 206 includes a predetermined processing circuit (not shown). This processing circuit includes, for example, one or more processors such as a CPU or GPU. This processing circuit may also include a predetermined integrated circuit such as an ASIC or FPGA.

[0053] The arithmetic unit 206 comprises a foreign object data acquisition unit 208, an estimation unit 210, and an inspection unit 212. The foreign object data acquisition unit 208, the estimation unit 210, and the inspection unit 212 are realized by the arithmetic unit 206 (processor) executing a program stored in the storage unit 204 (memory). If an integrated circuit such as an ASIC or FPGA is included in the arithmetic unit 206, at least a part of the foreign object data acquisition unit 208, the estimation unit 210, and the inspection unit 212 may be realized by the integrated circuit.

[0054] The foreign object data acquisition unit 208 acquires foreign object data 14 obtained from an image of the foreign object 42. The image includes, for example, an X-ray transmission image of the foreign object 42. The foreign object data 14 may, but is not limited to, be provided to the inspection device 200 by the user of the inspection device 200 via the operation unit 202. The foreign object data acquisition unit 208 may perform a predetermined image analysis process when an image (image data) of the foreign object 42 is provided to the inspection device 200. The predetermined image analysis process may include, for example, image recognition. As a result, the foreign object data acquisition unit 208 may acquire (generate) foreign object data 14 obtained from the image by performing the predetermined image analysis process.

[0055] The estimation unit 210 estimates the volume of the foreign object 42 using the foreign object data 14 acquired by the foreign object data acquisition unit 208 and the learning model 10 generated by the machine learning device 100. That is, the estimation unit 210 estimates the volume of the foreign object 42 based on the volume estimation data output from the learning model 10 in response to the foreign object data 14. The estimation unit 210 may, but is not limited to, use the learning model 10 stored in the storage unit 204 in advance to estimate the volume of the foreign object 42. The estimation unit 210 may also access the learning model 10 stored in an external memory (e.g., storage unit 104) of the inspection device 200 to estimate the volume of the foreign object 42. In that case, the inspection device 200 may be equipped with a communication module (not shown) that enables access to the learning model 10 stored in an external memory of the inspection device 200.

[0056] The inspection unit 212 performs a quality inspection of the membrane electrode assembly based on the estimation results from the estimation unit 210. For example, if the volume of foreign matter 42 estimated by the estimation unit 210 (learning model 10) is greater than or equal to a predetermined volume threshold, the inspection unit 212 may determine that the membrane electrode assembly containing the foreign matter 42 is a defective product. The inspection unit 212 may also control an notification device (not shown) to notify the user of the quality inspection results. For example, if the membrane electrode assembly is a defective product, the inspection unit 212 may control an alarm (not shown) to notify the user that the membrane electrode assembly is a defective product.

[0057] Figure 6 is a flowchart of an inspection method according to one embodiment.

[0058] The inspection device 200 described above is capable of performing the inspection method shown in Figure 6. This inspection method can be realized, for example, by having the processing unit 206 (processor) of the inspection device 200 execute a program stored in the memory unit 204 (memory) of the inspection device 200, which is a computer. As shown in Figure 6, the inspection method includes a foreign object data acquisition step S21, an estimation step S22, and an inspection step S23.

[0059] In the foreign object data acquisition step S21, the foreign object data acquisition unit 208 acquires foreign object data 14.

[0060] In estimation step S22, the estimation unit 210 uses the foreign matter data 14 acquired in foreign matter data acquisition step S21 and the learning model 10 generated by the machine learning device 100 (machine learning method shown in Figure 5) to estimate the volume of the foreign matter 42 mixed into the membrane electrode assembly.

[0061] In the inspection step (determination step) S23, the inspection unit 212 performs a quality inspection of the membrane electrode assembly. This quality inspection includes a process in which the inspection unit 212 determines the quality of the membrane electrode assembly based on the estimation results from the estimation step S22. Therefore, in the inspection step S23, the quality may be determined based on the volume of foreign matter 42.

[0062] According to this embodiment, the machine learning device 100 (machine learning method) and the inspection device 200 (inspection method) provide the effects described below, for example.

[0063] The machine learning device 100 comprises a dataset acquisition unit 108 and a learning model generation unit 112. The dataset acquisition unit 108 acquires training data 40. The learning model generation unit 112 generates a learning model 10 using the training data 40. The training data 40 includes foreign object data 14 and data indicating the volume of the foreign object 42. The foreign object data 14 is data obtained from an image of the foreign object 42 mixed into the membrane electrode assembly. The foreign object data 14 includes at least one of the following: area data 16 of the foreign object 42, brightness data 18 of the foreign object 42, and shape data 20 of the foreign object 42. The learning model 10 takes the foreign object data 14 as input and outputs the volume (estimated value) of the foreign object 42. As a result, the machine learning device 100 can provide the user with a learning model 10 that estimates the volume of the foreign object 42 based on data obtained from an image of the foreign object 42. The user can perform quality inspections of the membrane electrode assembly using the learning model 10. The image is, for example, an X-ray transmission image.

[0064] The shape data 20 includes at least one of the following: circumference data 24, skewness data 26, kurtosis data 28, and diameter data 30. The learning model generation unit 112 generates a learning model 10 that can accurately estimate the volume of the foreign object 42 according to its shape by performing machine learning using the shape data 20. Preferably, but not limited to, the shape data 20 includes all of the following: circumference data 24, skewness data 26, kurtosis data 28, and diameter data 30.

[0065] The diameter data 30 includes ratio data 36. The ratio data 36 represents, for example, the ratio of the diameter F1 of a first ferret to the diameter F2 of a second ferret. The ratio data 36 may also represent the ellipticity of a virtual ellipse VE (virtual circle). The learning model generation unit 112 generates a learning model 10 that can accurately estimate the volume of a foreign object 42 according to its shape by performing machine learning using the ratio data 36. The ellipticity of the virtual circle is defined, for example, as 1.00.

[0066] The machine learning device 100 further includes a training data selection unit 110. The training data selection unit 110 selects training data 40 to be used for supervised learning by the learning model generation unit 112 from among a plurality of datasets 12. The training data selection unit 110 selects the training data 40 based on, for example, at least one of the allowable ratio range RR and the allowable area range RA. In this way, the training data selection unit 110 suppresses the use of datasets 12 containing outliers by the learning model generation unit 112 for supervised learning. As a result, the learning model generation unit 112 can generate a learning model 10 that can accurately estimate the volume of a foreign object 42 according to its shape.

[0067] The inspection device 200 includes a foreign object data acquisition unit 208 and an estimation unit 210. The foreign object data acquisition unit 208 acquires foreign object data 14. The estimation unit 210 uses the foreign object data 14 and a learning model 10 to estimate the volume of the foreign object 42. As a result, the inspection device 200 can accurately determine the quality of the membrane electrode assembly based on the estimated volume.

[0068] One embodiment may be modified as shown below. Descriptions that overlap with the first embodiment will be omitted as appropriate below. Furthermore, among the configurations described below, those identical to those described in the first embodiment will be denoted by the same reference numerals as in the first embodiment.

[0069] (Variation 1) The inspection system SYS may further include a display device (not shown). This display device may include, for example, a display panel, display elements, etc., that form a display screen. The display device may appropriately display information stored by the storage unit 104 of the machine learning device 100 or the storage unit 204 of the inspection device 200. For example, the display device provided in the inspection device 200 may display the results of the quality inspection by the inspection unit 212. In other words, the display device provided in the inspection device 200 may function as the notification device described above. At least one of the machine learning device 100 and the inspection device 200 may also be equipped with a display device (display).

[0070] (Modification 2) A single electronic device (computer) may function as both a machine learning device 100 and a testing device 200. In this case, the machine learning device 100 and the testing device 200 may share hardware resources such as a processor and memory.

[0071] (Variation 3) The inspection system SYS may further include terminal devices (one or more terminal devices) not shown. The inspection system SYS may be equipped with multiple terminal devices. The terminal devices are implemented by electronic devices such as PCs (Personal Computers). The terminal devices may also be portable smart devices (such as tablet terminals). The terminal devices may remotely control at least one of the machine learning device 100 and the inspection device 200 via a network. In this case, the network may include the Internet.

[0072] (A combination of multiple variations) The aforementioned variations may be combined as appropriate, within the bounds of consistency.

[0073] According to the above-described embodiment and modifications, a machine learning device 100, an inspection device 200, a machine learning method, and a program are provided for better inspection of a membrane electrode assembly.

[0074] The following additional information is disclosed regarding the above embodiments.

[0075] (Note 1) The machine learning device according to this disclosure is a machine learning device (100) comprising: an acquisition unit (108) that acquires training data (40) including foreign object data (14) which includes at least one of area data (16) indicating the area of ​​a foreign object (42) mixed into a membrane electrode assembly as determined from an image of the foreign object, brightness data (18) indicating the brightness of the foreign object as determined from the image, and shape data (20) relating to the shape of the foreign object as determined from the image, and data indicating the volume of the foreign object; and a learning model generation unit (112) that generates a learning model (10) which takes the foreign object data as input and outputs the volume of the foreign object by performing supervised learning using the training data acquired by the acquisition unit. With such a configuration, a learning model that can accurately estimate the volume of the foreign object can be generated. By using such a learning model, the volume of the foreign object can be accurately estimated, so that the inspection of the membrane electrode assembly can be performed well.

[0076] (Note 2) The machine learning device described in Appendix 1 may also be a machine learning device in which the shape data includes at least one of the following: circumference data (24) relating to the circumference of the foreign object, skewness data (26) relating to the skewness of the foreign object, kurtosis data (28) relating to the kurtosis of the foreign object, and diameter data (30) relating to the diameter of a virtual circle approximating the foreign object, the diameter of a virtual ellipse (VE) approximating the foreign object, or the ferret diameter of the foreign object.

[0077] (Note 3) The machine learning device described in Appendix 2 may also be a machine learning device in which the diameter data includes the ratio of a first ferret diameter (F1), which is the ferret diameter in a first direction (D1), to a second ferret diameter (F2), which is the ferret diameter in a second direction (D2) intersecting the first direction, or ratio data (36) indicating the ellipticity of the virtual ellipse.

[0078] (Note 4) The machine learning device described in Appendix 3 may also be a machine learning device in which the learning model generation unit performs supervised learning using the foreign object data in which the ratio or ellipticity is within a predetermined range.

[0079] (Note 5) The machine learning device described in Appendix 1 may also be a machine learning device in which the learning model generation unit performs supervised learning using the foreign object data whose area is within a predetermined range.

[0080] (Note 6) The machine learning device described in Appendix 1 may also be a machine learning device in which the image is an X-ray transmission image.

[0081] (Note 7) The inspection apparatus according to this disclosure is an inspection apparatus (200) comprising: a foreign matter data acquisition unit (208) that acquires foreign matter data (14) including at least one of area data (16) indicating the area of ​​a foreign matter (42) mixed into a membrane electrode assembly as determined from an image of the foreign matter, brightness data (18) indicating the brightness of the foreign matter as determined from the image, and shape data (20) relating to the shape of the foreign matter as determined from the image; and an estimation unit (210) that estimates the volume of the foreign matter using the foreign matter data acquired by the foreign matter data acquisition unit and the learning model generated by the machine learning apparatus described in any one of Appendices 1 to 5. With such a configuration, the volume of the foreign matter can be accurately estimated, making it possible to inspect the membrane electrode assembly well.

[0082] (Note 8) The machine learning method according to this disclosure is a machine learning method that is executed by a computer and comprises: an acquisition step (S11) of acquiring training data (40) including foreign object data (14) which includes at least one of area data (16) which indicates the area of ​​a foreign object (42) mixed into a membrane electrode assembly, brightness data (18) which indicates the brightness of the foreign object as determined from the image, and shape data (20) which relates to the shape of the foreign object as determined from the image, and data indicating the volume of the foreign object; and a learning model generation step (S13) which generates a learning model (10) which takes the foreign object data as input and outputs the volume of the foreign object by performing supervised learning using the training data acquired in the acquisition step.

[0083] (Note 9) The program relating to this disclosure is a program for causing the computer to execute the machine learning method described in Appendix 8.

[0084] Furthermore, this disclosure may take various forms, not limited to the disclosure described above, without departing from the gist of this disclosure. [Explanation of Symbols]

[0085] 10…Learning Model 14…Foreign object data 16…Area data 18…Brightness data 20…Shape data 24…Circumference data 26... Skewness data 28…Kuturonic data 30…Diameter data 36…Ratio data 40…Teacher data 42...Foreign object 100…Machine learning device 108...Dataset acquisition unit (acquisition unit) 112...Learning Model Generation Unit 200... Inspection device 208... Foreign object data acquisition unit 210…Estimation part F1...First ferret diameter F2... Second ferret diameter VE…Virtual Ellipse

Claims

1. An acquisition unit acquires training data which includes foreign object data that includes at least the shape data from area data indicating the area of ​​the foreign object mixed into a membrane electrode assembly, brightness data indicating the brightness of the foreign object as determined from the image, and shape data relating to the shape of the foreign object as determined from the image, and data indicating the volume of the foreign object. A learning model generation unit generates a learning model that takes the foreign object data as input and outputs the volume of the foreign object by performing supervised learning using the training data acquired by the acquisition unit, Equipped with, A machine learning device wherein the shape data includes circumference data relating to the circumference of the foreign object, skewness data relating to the skewness of the foreign object, kurtosis data relating to the kurtosis of the foreign object, and diameter data relating to the diameter of a virtual circle approximating the foreign object, the diameter of a virtual ellipse approximating the foreign object, or the ferret diameter of the foreign object.

2. A machine learning device according to claim 1, A machine learning device in which the diameter data includes the ratio of a first ferret diameter, which is the ferret diameter in a first direction, to a second ferret diameter, which is the ferret diameter in a second direction intersecting the first direction, or ratio data indicating the ellipticity of the virtual ellipse.

3. A machine learning apparatus according to claim 2, The learning model generation unit is a machine learning device that performs supervised learning using the foreign object data in which the ratio or ellipticity is within a predetermined range.

4. A machine learning device according to claim 1, The learning model generation unit is a machine learning device that performs supervised learning using the foreign object data, the area of ​​which is within a predetermined range.

5. A machine learning device according to claim 1, The image above is an X-ray transmission image from a machine learning device.

6. A foreign matter data acquisition unit acquires foreign matter data which includes at least the shape data, among area data indicating the area of ​​the foreign matter obtained from an image of the foreign matter mixed into the membrane electrode assembly, brightness data indicating the brightness of the foreign matter obtained from the image, and shape data relating to the shape of the foreign matter obtained from the image. An estimation unit that estimates the volume of the foreign object using the foreign object data acquired by the foreign object data acquisition unit and the learning model generated by the machine learning device according to any one of claims 1 to 5, An inspection device equipped with the following features.

7. An acquisition step to acquire training data including foreign object data which includes at least the shape data, among area data indicating the area of ​​the foreign object mixed into a membrane electrode assembly, brightness data indicating the brightness of the foreign object as determined from the image, and shape data relating to the shape of the foreign object as determined from the image, and data indicating the volume of the foreign object; A learning model generation step involves generating a learning model that takes the foreign object data as input and outputs the volume of the foreign object by performing supervised learning using the training data acquired in the acquisition step, It has, The shape data includes circumference data relating to the circumference of the foreign object, skewness data relating to the skewness of the foreign object, kurtosis data relating to the kurtosis of the foreign object, and diameter data relating to the diameter of a virtual circle approximating the foreign object, the diameter of a virtual ellipse approximating the foreign object, or the ferret diameter of the foreign object, and is performed by a computer.

8. A program for causing the computer to execute the machine learning method described in claim 7.

Images