Image generation device, learning method for generator, generator, estimator and image confirmation system

Abstract

To provide an image generation device capable of generating an estimated image of a material surface having any friction force or friction coefficient.SOLUTION: An image generation device includes an acquisition unit that acquires a friction force and / or a friction coefficient input by a user, and a generator that estimates and generates an image representing the friction force and / or the friction coefficient. The image generation device 1 includes the acquisition unit 2, a generator 3, and an estimator 4. When a user inputs the friction force and / or the friction coefficient into the image generation device 1, the acquisition unit 2 acquires the friction force and / or the friction coefficient.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to an image generation device, a generator training method, a generator, an estimator, and an image confirmation system. [Background technology]

[0002] The ball-on-disk friction test is a known friction testing method for measuring the friction force and coefficient of friction under certain conditions when components slide against each other. In the ball-on-disk friction test, a ball is placed in contact with one side of a disk and rotated to generate friction between the ball and the disk, and the friction force and coefficient of friction are measured.

[0003] It is very important to observe the surface condition obtained after a friction test. In recent years, methods for estimating the friction coefficient from images obtained after a friction test have been reported. For example, Non-Patent Document 1 discloses a method in which a convolutional neural network (CNN) learns surface images of various materials after friction and the measured friction coefficients, and estimates the friction coefficient from the surface images. In addition, Non-Patent Document 2 discloses a method for diagnosing the friction coefficient from friction surface images using AI.

[0004] Incidentally, cycle consistent generative adversarial networks (CCGANs) are known as an AI algorithm that can learn features from prepared data and generate pseudo-data (Non-Patent Document 3). [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] H. Zhang, 2 others, “Friction from Reflectance: Deep Reflectance Codes for Predicting Physical Surface Properties from One-Shot In-Field Reflectance”, [online], May 25, 2016, Arxiv, [Retrieved May 13, 2020], Internet<URL:https: / / arxiv.org / abs / 1603.07998> Beginning of the form [Non-patent document 2] Motoyuki Murashima, "Emergence of new functionality through friction and wear estimation technology using deep learning and deformation surface control using AI", [online], Reiwa 2, Tribologist Vol. 67 No. 12 (2022) 830-837, [Retrieved May 13, 2024], Internet <URL:https: / / jstage.jst.go.jp / article / tribologist / 67 / 12 / 67_67.12_830 / _article / -char / ja / > [Non-patent document 3] X.Ding, 4 others, "Continuous Conditional Generative Adversarial Networks: Novel Empirical Losses and Label Input Mechanisms", [online], November 15, 2020, Arxiv, [searched on May 13, 2020], Internet<URL:https: / / arxiv.org / pdf / 2011.07466> Summary of the Invention [Problem to be solved by the invention]

[0006] However, there has been no known method for estimating an image of a surface of a material having a given friction force or coefficient of friction from information on the friction force or coefficient of friction of the surface of that material.

[0007] Therefore, an object of the present disclosure is to provide an image generation device capable of generating an estimated image of a material surface having any friction force or friction coefficient. Another object of the present disclosure is to provide a generator and estimator for use in the image generation device, as well as a training method for the generator. Another object of the present invention is to provide an image confirmation system capable of confirming the appropriateness of an estimated image obtained by the image generation device. Another object of the present invention is to provide an image generation device capable of generating an estimated image of a material surface having any physical properties. [Means for solving the problem]

[0008] The present disclosure provides a friction force sensor including: an acquisition unit that acquires a friction force and / or a friction coefficient input by a user; and a generator that estimates and generates an image representing the friction force and / or the friction coefficient.

[0009] The generator is preferably a generation AI that has learned training data consisting of a plurality of image data having different friction force information and / or friction coefficient information.

[0010] Preferably, the image generating device comprises an estimator for estimating the friction force and / or coefficient of friction of the image generated by the generator.

[0011] The estimator is an image recognition AI that is trained using information including a friction interface image obtained by a friction test and friction force and / or friction coefficient as training data, and it is preferable that the training data is information that is not used to train the generator.

[0012] The generator preferably includes a generation unit and an identification unit, and the generation unit is a generation unit of a generative adversarial network generated by competitive learning with the identification unit that identifies the authenticity of the image generated by the generation unit.

[0013] The present disclosure also provides a learning method for performing machine learning on a generator, which estimates and generates a second image having second friction force information and / or second friction coefficient information from a first image having first friction force information and / or first friction coefficient information, The present invention provides a method for training a generator, which performs the machine learning using information including a friction interface image obtained by a friction test and the friction force and / or the friction coefficient as training data.

[0014] It is preferable that the generator includes a generation unit and a recognition unit, and that the machine learning is performed by competitive learning with the recognition unit that recognizes the authenticity of the image generated by the generation unit.

[0015] The present disclosure also provides a generator that estimates and generates an image representing a friction force and / or friction coefficient input by a user.

[0016] Preferably, the generator has been trained on training data consisting of a plurality of image data having different friction force information and / or friction coefficient information.

[0017] The generator is preferably a generation AI that is trained using friction interface images obtained by friction tests and information including friction force and / or friction coefficient as training data.

[0018] The present disclosure also provides an estimator for estimating the friction force and / or friction coefficient of the image generated by the generator.

[0019] The present disclosure also provides a method for generating an image using the generator; causing an estimator to estimate the friction force and / or friction coefficient of the image generated by the generator; and a step of comparing the friction force and / or friction coefficient input by the user with the friction force and / or friction coefficient estimated by the estimator to confirm the appropriateness of the image generated by the generator.

[0020] The present disclosure also provides an acquisition unit that acquires physical characteristics input by a user; and a generator that estimates and generates an image of a surface of a material having the physical property.

[0021] The physical properties may be a combination of one or more selected from the group consisting of a polarizing microscope image of the resin and mechanical strength, a friction interface image and friction coefficient, and a thermography image and electrothermal properties. [Effects of the Invention]

[0022] According to the image generating device of the present disclosure, it is possible to generate an estimated image of the surface of a material having any physical property such as any friction force or friction coefficient. [Brief explanation of the drawings]

[0023] [Figure 1] FIG. 1 is a block diagram illustrating an example of the configuration of an image generating device. [Figure 2] FIG. 1 is a front view showing one embodiment of a friction testing device. [Figure 3] This is a pseudo-SEM image generated by a friction test device. DETAILED DESCRIPTION OF THE INVENTION

[0024] [Image generation device] The image generating device of the present disclosure includes an acquisition unit that acquires a friction force and / or a friction coefficient input by a user, and a generator that estimates and generates an image representing the friction force and / or the friction coefficient.

[0025] An embodiment of an image generating device according to the present disclosure will be described with reference to Fig. 1. Note that the configurations shown in the drawings and the following description are merely examples, and the scope of the present disclosure is not limited to those shown in the drawings and the following description.

[0026] The image generating device 1 shown in FIG. 1 includes an acquisition unit 2, a generator 3, and an estimator 4. When a user inputs a friction force and / or a friction coefficient to the image generating device 1, the acquisition unit 1 acquires the friction force and / or the friction coefficient. The generator 3 then estimates and generates an image representing the friction force and / or the friction coefficient based on the friction force and / or the friction coefficient acquired by the acquisition unit 1. That is, the generator 3 can estimate and generate an image representing the friction force and / or the friction coefficient input by the user. In this way, the image generating device 1 can generate an estimated image of a material surface having any friction force or friction coefficient input by the user. Note that the estimator 4 is not used when estimating and generating the image.

[0027] The generator 3 is a generation AI that has learned training data consisting of multiple image data having different friction force information and / or friction coefficient information. The generator 3 may be, for example, a GAN (generative adversarial network), or may be a CCGAN, StarGAN, or StyleGAN. However, the generator 3 is not limited to these and may be other generative models such as a diffusion model, a variational autoencoder (VAE), or a U-net.

[0028] The estimator 4 is independent of the generator 3 and estimates whether the image generated by the generator 3 has the input friction force and / or friction coefficient. The estimator 4 is an image recognition AI trained using information including the friction interface image obtained by the friction test and the friction force and / or friction coefficient as training data. The training data used for training the estimator 4 is information that is not used for training the generator 3.

[0029] When the generator 3 is a GAN, the generator 3 may include a generation unit and a discrimination unit (not shown). The training of the discrimination unit of the generator 3 when the generator 3 is a GAN will be specifically described. The discrimination unit is used when training the generation unit of the generator 3. Random noise and arbitrary friction force information and / or friction coefficient information are input to the generation unit, which generates an estimated image based on the random noise. The discrimination unit is trained to recognize that the obtained estimated image is a fake image (an image that is not a real image). Meanwhile, for a real image (e.g., a first image having first friction force information and / or first friction coefficient information), multiple different data groups (training data) are input to the discrimination unit, which is trained to recognize that the real image is a real image. In this way, the discrimination unit is trained to recognize whether an image of the surface of an arbitrary material is a fake image or a real image. Note that the generation unit is not trained when the discrimination unit is trained.

[0030] Next, the learning of the generation unit will be described. The generation unit is obtained by performing machine learning through competitive learning with the identification unit, which distinguishes between the authenticity of the image generated by the generation unit. As in the case of learning the identification unit, random noise and arbitrary friction force information and / or friction coefficient information are input to the generation unit, and a fake image based on the random noise is generated. The generation unit is trained so that this fake image approaches an image that the identification unit recognizes as a genuine image. In this way, the generation unit is trained by machine learning so that it can estimate and generate an image that allows the identification unit to recognize the fake image as a genuine image (for example, a second image having second friction force information and / or second friction coefficient information). Note that the identification unit is not trained when the generation unit is trained.

[0031] The learning data used for training the estimator 4 can be information including friction interface images obtained by friction tests and friction forces and / or friction coefficients. For machine learning by the generator 3 and the estimator 4, the image generation device requires a large amount of data (at least 1,000 or more) including friction interface images and friction forces and / or friction coefficients. In addition, in order for the generator 3 and the image generation device 1 to generate images that are closer to actual images, it is desirable to prepare image data having various values ​​of friction forces and / or friction coefficients for the actual images to be input to the generator 3 for training the generator 3.

[0032] The image data can be obtained, for example, by a friction testing method in which a first sliding member and a second sliding member are caused to slide relative to each other to measure the frictional force of the sliding region, and the frictional force is continuously measured while irradiating the sliding region with infrared light and visible light via a dichroic mirror and transmitted through the first sliding member. According to the friction testing method, the sliding region where the members are in contact with each other and sliding against each other can be continuously observed during the friction test, and multiple images of the friction coefficients and frictional forces can be easily obtained in large quantities.

[0033] The above-mentioned friction test method will be described below with reference to an embodiment. Fig. 2 shows a schematic diagram (front view) of one embodiment of a friction test device for carrying out the above-mentioned friction test method. The friction test device 11 shown in Fig. 2 includes a first sliding member 12, a second sliding member 13, a first fixing member 14, a second fixing member 15, a dichroic mirror 16, a microscopic radiation thermometer 17, a lens 18, a visible light source 19, a motor 20, and a friction force detection unit (not shown) that measures the friction force in the sliding region.

[0034] The first sliding member 12 is located on the dichroic mirror 16 side with respect to the second sliding member 13. The first sliding member 12 is plate-shaped (disk-shaped) and is supported by a first fixing member 14 that is connected to the center of the first sliding member 12 and fixes the first sliding member 12 so that it can rotate in a horizontal plane. The first fixing member 14 is column-shaped and can be rotated in a horizontal plane by a motor 20. By rotating the first fixing member 14, the first sliding member 12 can be rotated, and the first sliding member 12 can slide against the second sliding member 13.

[0035] The second sliding member 13 is a ball having a spherical surface, and is located vertically below the disk-shaped first sliding member 12, on the opposite side of the dichroic mirror 16 with respect to the first sliding member 13, and is fixed by a second fixing member 15 so as to be able to slide on the lower surface of the first sliding member 12. The dichroic mirror 16 and the second sliding member 13 are positioned above and below in the vertical direction. Note that the second sliding member 13 being a ball means that the sliding surface is spherical, and may be, for example, hemispherical.

[0036] 2 is a ball-on-disk friction test device in which the first sliding member 12 is a disk and the second sliding member 13 is a ball, but the first sliding member 12 may be a ball and the second sliding member 13 may be a disk. Furthermore, the friction test device 11 is not limited to a ball-on-disk friction test device, but may be a friction test device that measures friction force based on other principles, such as a ring-on-disk (ring-on-plate) friction test, a pin-on-disk friction test, a ball-on-plate friction test, a pin-on-plate friction test, or a fretting friction test. In the case of such other friction test devices, the sliding mechanism can be appropriately designed using a known or commonly used method based on the other principles.

[0037] The dichroic mirror 16 is located on the opposite side of the first sliding member 12 from the second sliding member 13. In FIG. 2 , the dichroic mirror 16 is located vertically above the first sliding member 12, more specifically, vertically above the sliding region. The dichroic mirror transmits either infrared light or visible light and reflects the other. That is, the dichroic mirror transmits at least some wavelengths of infrared light and reflects at least some wavelengths of visible light, or transmits at least some wavelengths of visible light and reflects at least some wavelengths of infrared light. In the former case (i.e., the dichroic mirror transmits infrared light, as shown in FIG. 2 ), infrared light L1 emitted from the sliding region of the first sliding member 12 passes through the dichroic mirror 16 and enters the microscopic radiation thermometer 17. Visible light L2 emitted from the visible light source 19 is reflected by the dichroic mirror 16 and irradiates the first sliding member 12. In the latter case (i.e., when the dichroic mirror reflects infrared light), the positions of the microscopic radiation thermometer 17 and the lens 18 shown in FIG. 2 are reversed. That is, the microscopic radiation thermometer is at position 18, and the lens is at position 17. In this case, infrared light emitted from the sliding region of the first sliding member 12 is reflected by the dichroic mirror and enters the microscopic radiation thermometer. Visible light emitted from the visible light source passes through the dichroic mirror and is irradiated onto the first sliding member 12. Here, the first sliding member 12 transmits infrared light and visible light. Therefore, the visible light irradiated onto the first sliding member 12 passes through the first sliding member 12 and irradiates the sliding region between the first sliding member 12 and the second sliding member 13. In this way, visible light is irradiated onto the sliding region via the dichroic mirror, and the first sliding member 12 transmits the visible light reflected by the surface of the second sliding member 13 and the infrared light L1 emitted from the sliding region.

[0038] The dichroic mirror 16, the microscopic radiation thermometer 17, and the visible light source 19 are preferably designed to be positioned so that the infrared light L1 and the visible light L2 are coaxially irradiated onto and incident on the sliding area. This coaxial arrangement allows for more accurate measurement of the temperature of the observation area. The microscopic radiation thermometer 17, the dichroic mirror 16, and the sliding area are positioned linearly in this order. When the dichroic mirror transmits visible light, the visible light source, the dichroic mirror, and the sliding area are preferably positioned linearly in this order.

[0039] The visible light irradiated onto the sliding region is reflected and again enters the dichroic mirror 16. In addition, the infrared light irradiated from the sliding region also enters the dichroic mirror 16. At this time, the infrared light and visible light are transmitted through or reflected by the dichroic mirror 16, respectively, and the infrared light L1 enters the microscopic radiation thermometer 17, and the visible light L2 enters the lens 18. In this way, while the first sliding member 12 and the second sliding member 13 are sliding, the sliding region can be continuously observed and the temperature of the sliding region can be measured at the same time. In addition, the friction force in the sliding region can be measured at the same time by the friction force detection unit. The visible light source 19 and the lens 18 are positioned coaxially with the visible light irradiated by the visible light source 19.

[0040] 2, lens 18 may be a camera, or lens 18 and visible light source 19 may be integrated. For example, a visible camera / coaxial visible light source can be used as lens 18 and visible light source 19. In addition, in FIG. 2, it is preferable that the visible light emitted by visible light source 19 and the visible light incident on lens 18 are coaxial, but the visible light emitted by visible light source 19 may indirectly illuminate the sliding region.

[0041] Examples of the lens 18 include an objective lens and a lens of an optical device capable of capturing images, such as a camera lens or a video camera lens. The friction test device may be equipped with an electronic device having a detection unit capable of detecting light, such as an imaging device such as a camera or a video camera. The electronic device is installed so as to include the lens 18 as a part thereof, or in a location where reflected light that has passed through the lens 18 can be detected.

[0042] The material of the first sliding member 12 can be any known or commonly used material that is transmissive to infrared rays and visible light, and may be any of metal, resin, and rubber. Specific examples of the material of the first sliding member 12 include CaF, ZnSe, diamond, ZnS, and chalcogenide glass. Furthermore, the surface of these materials may be coated with a material such as chrome plating or a diamond-like carbon film, as long as the transmission of infrared rays and visible light is not hindered.

[0043] The material of the second sliding member 13 may be selected based on the object for which the friction force is to be measured. For example, crystalline resins such as POM (polyacetal), PE (polyethylene), PP (polypropylene), PA (polyamide), PET (polyethylene terephthalate), PBT (polybutylene terephthalate), PPS (polyphenylene sulfide), PEEK (polyether ether ketone), PEK (polyether ketone), LCP (liquid crystal polymer), and PTFE (polytetrafluoroethylene), PVC (polyvinyl chloride), Examples include resins such as PS (polystyrene), PMMA (polymethyl methacrylate), ABS (acrylonitrile butadiene styrene), PC (polycarbonate), m-PPE (modified polyphenylene ether), PES (polyethersulfone), PSU (polysulfone), PEI (polyetherimide), PAI (polyamideimide), and other amorphous resins; typical metals such as steels such as SS400 and SUJ2, aluminum, nickel, and copper; and glasses such as borosilicate glass, quartz glass, and sapphire glass.

[0044] Dichroic mirror 16 can be any known or commonly used mirror, such as a mirror made of a substrate that transmits visible light or infrared light, such as germanium, CaF2, diamond, KBr, BaF2, silicon, sapphire, ZnSe, ZnS, or chalcogenide glass, and is coated with various coatings to reflect specific wavelengths. Depending on the arrangement of the microscopic radiation thermometer and the lens / camera, the dichroic mirror can be either one that transmits infrared light and reflects visible light, or one that transmits visible light and reflects infrared light.

[0045] The friction test device may be equipped with a load application mechanism. An example of the load application mechanism is a mechanism that applies a load to the first sliding member 12 side of the second sliding member 22. The load in the friction test is not particularly limited, and can be appropriately selected depending on the friction force to be measured. In addition, the sliding speed (for example, the rotation speed of the disk in a ball-on-disk friction test) is also not particularly limited, and can be appropriately set.

[0046] In the above-mentioned friction test method, the sliding between the first sliding member and the second sliding member may be performed in the presence of a lubricant or a lubricant for initial break-in. Furthermore, the above-mentioned friction test method can be performed regardless of the friction environment, such as dry friction, in a lubricating oil, or in a vacuum.

[0047] The image data can be obtained by a friction testing method including the steps of: generating friction by pressing a pressure member against a first surface of an electron-transparent film and moving the pressure member along the surface direction of the electron-transparent film; irradiating an electron beam onto a second surface of the electron-transparent film opposite the first surface; and detecting electrons or electromagnetic waves generated at the interface between the electron-transparent film and the pressure member when the electron beam that has passed through the electron-transparent film is irradiated onto the interface, thereby acquiring an image of the region of the electron-transparent film where the friction is occurring. Note that the electron beam irradiation step may be performed after the friction generation step. According to the friction testing method, the sliding region where members are in contact and sliding with each other can be continuously observed during a friction test, and multiple images of friction coefficients and friction forces can be easily acquired in large quantities.

[0048] In the friction test method, the thickness of the electron-transparent film can be 1 to 200 nm. In the friction test method, the pressing member can have a curved surface, and the curved surface can be pressed against the electron-transparent film. In the friction test method, the electron-transparent film can be formed of Si3N4, SiO2, SiC, Si, DLC, amorphous carbon, graphene, or graphene oxide. The friction test method can further include preparing a container having an opening blocked by the electron-transparent film and a sealed internal space, with a first surface of the electron-transparent film facing the internal space. The friction-generating step can be configured to apply a lubricant to the first surface of the electron-transparent film, and generate friction with the pressing member in the area where the lubricant is applied. For details of the friction test method, see Japanese Patent Application Laid-Open No. 2020-180925.

[0049] In this manner, the generator 3 and the estimator 4 can be trained.

[0050] Furthermore, the estimator 4 can estimate the friction force and / or friction coefficient of the image generated by the generator 3. Furthermore, the image generation device 1 may include an estimator other than the estimator 4 that estimates the friction force and / or friction coefficient of the image generated by the generator 3. Such an estimator can compare the friction force and / or friction coefficient input by the user with the friction force and / or friction coefficient estimated by the estimator to confirm the appropriateness of the image generated by the generator 3. That is, the present disclosure can provide an image confirmation system that includes the steps of causing the generator 3 to generate an image, causing the estimator to estimate the friction force and / or friction coefficient of the image generated by the generator 3, and comparing the friction force and / or friction coefficient input by the user with the friction force and / or friction coefficient estimated by the estimator to confirm the appropriateness of the image generated by the generator 3.

[0051] Furthermore, while the image generating device 1 has been described as being capable of estimating and generating an image representing a friction force and / or a friction coefficient based on the friction force and / or a friction coefficient input by a user, it is also possible to provide an image generating device that can input other physical properties instead of the friction force or the friction coefficient and generate an image having the physical properties. An example of such an image generating device includes an acquisition unit that acquires the physical properties input by a user and a generator that estimates and generates an image of a material surface having the physical properties. Examples of the physical properties include a polarizing microscope image and mechanical strength of a resin, a friction interface image and friction coefficient in a lubricant or grease, and a thermographic image and electrothermal properties.

[0052] Each aspect disclosed in this specification can be combined with any other feature disclosed in this specification. Each configuration and combination thereof in each embodiment is an example, and addition, omission, substitution, and other modifications of configurations are possible as appropriate within the scope of the present disclosure. Furthermore, each invention according to this disclosure is not limited by the embodiments or the following examples, but is limited only by the scope of the claims. [Example]

[0053] Example 1 First, a generator and an estimator related to the CCGAN were trained. Specifically, an estimator, which is an image recognition AI, was trained using information including a friction interface image and friction force and / or friction coefficient obtained by the friction testing method described in JP 2020-180925 A as training data. Then, a generator was trained using the estimator. Images were generated using the CCGAN after training of the generator and estimator as an image generation device. A list of the obtained images is shown in Figure 3.

[0054] Figure 3 shows a total of 100 images, 10 × 10 in size. The bottom five rows of images are pseudo-SEM images with friction forces of 2 to 25 mN, but SEM images in this friction force range can also be obtained through actual friction tests. On the other hand, images in the friction force range of 0 to 2 mN are extremely difficult to obtain through actual friction tests because the friction force is extremely low and the friction force in this range elapses in an extremely short time during friction tests. However, by using the image generation device described above, it is possible to estimate and obtain images of such friction forces that are difficult to obtain in practice, as well as images of material surfaces with friction forces less than 0 mN.

[0055] Variations of the invention according to the present disclosure are described below. [Appendix 1] An acquisition unit that acquires a friction force and / or a friction coefficient input by a user; a generator that estimates and generates an image representing the friction force and / or the friction coefficient. [Appendix 2] The image generation device described in Appendix 1, wherein the generator is a generation AI that has learned training data consisting of multiple image data having different friction force information and / or friction coefficient information. [Appendix 3] The image generating device according to appendix 1 or 2, further comprising an estimator that estimates the friction force and / or friction coefficient of the image generated by the generator. [Appendix 4] The image generation device described in any one of Appendices 1 to 3, wherein the estimator is an image recognition AI trained using information including a friction interface image obtained by a friction test and friction force and / or friction coefficient as training data, and the training data is information that is not used for training the generator. [Appendix 5] The image generating device described in Appendix 4, wherein the generator comprises a generation unit and an identification unit, and the generation unit is a generation unit of a generative adversarial network generated by competitive learning with the identification unit that identifies the authenticity of the image generated by the generation unit. [Appendix 6] A learning method for performing machine learning of a generator, which estimates and generates a second image having second friction force information and / or second friction coefficient information from a first image having first friction force information and / or first friction coefficient information, A method for training a generator, in which the machine learning is performed using information including a friction interface image obtained by a friction test and the friction force and / or the friction coefficient as training data. [Appendix 7] A method for training a generator as described in Appendix 6, wherein the generator comprises a generation unit and a recognition unit, and performs the machine learning by competitive learning with the recognition unit, which recognizes the authenticity of the image generated by the generation unit. [Appendix 8] A generator that estimates and generates images representing the friction force and / or friction coefficient input by the user. [Appendix 9] The generator according to Appendix 8, which has been trained on training data consisting of a plurality of image data having different friction force information and / or friction coefficient information. [Appendix 10] A generator according to appendix 8 or 9, which is a generating AI trained using friction interface images obtained by friction tests and information including friction force and / or friction coefficient as training data. [Supplementary Note 11] An estimator that estimates the friction force and / or friction coefficient of an image generated by the generator according to any one of Supplementary Notes 8 to 10. [Appendix 12] A step of generating an image using the generator according to any one of Appendices 8 to 10; causing an estimator to estimate the friction force and / or friction coefficient of the image generated by the generator; and comparing the friction force and / or friction coefficient input by the user with the friction force and / or friction coefficient estimated by the estimator to confirm the appropriateness of the image generated by the generator. [Appendix 13] An acquisition unit that acquires physical properties input by a user; and a generator that estimates and generates an image of a surface of a material having the physical property. [Appendix 14] The image generating device described in Appendix 13, wherein the physical properties are a combination of one or more selected from the group consisting of a polarizing microscope image of the resin and its mechanical strength, a friction interface image and its friction coefficient, and a thermography image and its electrothermal properties. [Explanation of symbols]

[0056] 1. Image generation device 2 Acquisition part 3 generator 4 Estimator 11 Friction test equipment 12 First sliding member 13 Second sliding member 14 First fixing member 15 Second fixing member 16 Dichroic mirror 17 Microscopic radiation thermometer 18 Lenses 19 Visible light source 20 motors

Claims

1. an acquisition unit that acquires a friction force and / or a friction coefficient input by a user; a generator that estimates and generates an image representing the friction force and / or the friction coefficient.

2. The image generating device according to claim 1 , wherein the generator is a generating AI that has learned training data consisting of a plurality of image data having different friction force information and / or friction coefficient information.

3. The image generating device according to claim 1 or 2, further comprising an estimator for estimating a friction force and / or a friction coefficient of the image generated by the generator.

4. 3. The image generation device according to claim 1, wherein the estimator is an image recognition AI trained using information including a friction interface image obtained by a friction test and the friction force and / or the friction coefficient as training data, and the training data is information that is not used for training the generator.

5. The image generating device according to claim 4, wherein the generator comprises a generation unit and an identification unit, and the generation unit is a generation unit of an adversarial generative network generated by competitive learning with the identification unit that identifies the authenticity of the image generated by the generation unit.

6. A learning method for performing machine learning of a generator, which estimates and generates a second image having second friction force information and / or second friction coefficient information from a first image having first friction force information and / or first friction coefficient information, A generator learning method in which the machine learning is performed using information including a friction interface image obtained by a friction test and friction force and / or friction coefficient as learning data.

7. The generator training method according to claim 6 , wherein the generator comprises a generation unit and a classification unit, and performs the machine learning by competitive learning with the classification unit that classifies the authenticity of an image generated by the generation unit.

8. A generator that estimates and generates an image representing a friction force and / or friction coefficient input by a user.

9. The generator according to claim 8 , which has been trained with training data consisting of a plurality of image data having different friction force information and / or friction coefficient information.

10. The generator according to claim 8 or 9, which is a generating AI trained using a friction interface image obtained by a friction test and information including friction force and / or friction coefficient as training data.

11. An estimator for estimating a friction force and / or a friction coefficient of an image generated by a generator according to claim 8 or 9.

12. causing a generator according to claim 8 or 9 to generate an image; causing an estimator to estimate the friction force and / or friction coefficient of the image generated by the generator; and comparing the friction force and / or friction coefficient input by the user with the friction force and / or friction coefficient estimated by the estimator to confirm the appropriateness of the image generated by the generator.

13. an acquisition unit that acquires physical characteristics input by a user; and a generator that estimates and generates an image of a surface of a material having the physical property.

14. The image generating device according to claim 13 , wherein the physical properties are a combination of one or more selected from the group consisting of a polarizing microscope image of a resin and its mechanical strength, a friction interface image and its friction coefficient, and a thermography image and its electrothermal properties.

Images