Apparatus for creating training datasets, method for creating training datasets, method for generating trained models, and method for predicting tire performance.

Abstract

This invention provides a creation device capable of creating a training dataset that includes material information of tire components, which has a high correlation with tire performance data. [Solution] This is a device for creating a training dataset used in machine learning to estimate tire performance data. The device includes an acquisition unit that acquires first data 21 consisting of a cross-sectional image 21A in which tire components 13 are represented by the values ​​24 of pixels 23 in a meridional cross-section of the tire, and a replacement unit that generates second data 22 by replacing the values ​​24 of pixels 23 in the cross-sectional image 21A with component information 26 including physical property values ​​25 of the tire components 13 corresponding to the pixels 23.

Description

Technical Field

[0001] The present invention relates to an apparatus for creating a learning dataset, a creation method, a method for generating a learned model, and a method for predicting tire performance.

Background Art

[0002] Patent Document 1 below describes a method for recognizing a tire image. In this method, an image representing the contact surface of a tire is input into a machine learning model that has learned the relationship between the image representing the contact surface of the tire and the contour of the contact surface of the tire, an output is derived, and based on the derived output, the contour of the contact surface of the target tire is estimated.

[0003] For learning the machine learning model, a learning dataset is prepared. This learning dataset is a dataset consisting of a plurality of combinations of an RGB image representing the contact surface of a tire and correct answer data.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Generally, values for representing the intensity of light and color are set for the pixels constituting an image. Although these values are extracted as feature amounts for machine learning, the pixel values themselves have a very low correlation with tire performance data. Therefore, there has been room for further improvement in the machine learning dataset for estimating performance data.

[0006] This invention was devised in view of the above circumstances, and its main objective is to provide a creation device capable of creating a training dataset that includes material information of tire components that have a high correlation with tire performance data. [Means for solving the problem]

[0007] The present invention relates to a device for creating a training dataset used in machine learning to estimate tire performance data, and includes: an acquisition unit that acquires first data consisting of a cross-sectional image in which tire components are represented by pixel values ​​in a meridional cross-section of the tire; and a replacement unit that generates second data by replacing the pixel values ​​of the cross-sectional image with component information including physical property values ​​of the tire components corresponding to the pixels. [Effects of the Invention]

[0008] By adopting the above configuration, the tire performance prediction device of the present invention makes it possible to create a training dataset that includes information on tire components that have a high correlation with tire performance data. [Brief explanation of the drawing]

[0009] [Figure 1] This block diagram shows examples of a device for creating training datasets, a device for generating trained models, and a device for predicting tire performance. [Figure 2] This is a meridian cross-sectional view of a tire, showing an example of a tire. [Figure 3] This flowchart shows an example of the processing steps for creating a training dataset. [Figure 4] This figure shows an example of the first data set, consisting of cross-sectional images. [Figure 5] (a) is a conceptual diagram showing an example of pixels in the cross-sectional image (first data) of Figure 4, and (b) is a conceptual diagram showing an example of second data in which the pixel values ​​of the cross-sectional image of Figure 5(a) are replaced with member information including material properties. [Figure 6]This table shows an example of the physical properties of tire components. [Figure 7] This figure shows an example of a training dataset. [Figure 8] This flowchart shows an example of the processing steps for generating a pre-trained model. [Figure 9] This is a conceptual diagram showing an example of a machine learning model (a pre-trained model). [Figure 10] This flowchart shows an example of the processing procedure for predicting tire performance. [Figure 11] This figure shows an example of a training dataset for another embodiment of the present invention. [Figure 12] This figure shows an example of a training dataset for yet another embodiment of the present invention. [Modes for carrying out the invention]

[0010] Embodiments of the present invention will be described below with reference to the drawings. It should be understood that the drawings contain exaggerations and representations that differ from the actual dimensional ratios of the structures in order to aid in understanding the content of the invention. Furthermore, the same or common elements are denoted by the same reference numerals throughout each embodiment, and redundant explanations are omitted. Moreover, the specific configurations shown in the embodiments and drawings are for the purpose of understanding the content of the present invention, and the present invention is not limited to the specific configurations shown in the drawings.

[0011] [Device for creating training datasets] FIG. 1 is a block diagram showing an example of a learning data set creation device 1A, a learned model generation device 1B, and a tire performance prediction device 1C. The learning data set creation device (hereinafter sometimes referred to as the "creation device") 1A is a device for creating a learning data set used for machine learning to estimate tire performance data. The creation device 1A of the present embodiment is also configured as a device (hereinafter sometimes referred to as the "generation device") 1B for generating a learned model and a tire performance prediction device (hereinafter sometimes referred to as the "prediction device") 1C. Note that the creation device 1A, the generation device 1B, and the prediction device 1C may be configured separately.

[0012] The creation device 1A (generation device 1B and prediction device 1C) of the present embodiment is configured as a general-purpose computer 1. Examples of the computer 1 include a desktop personal computer, a laptop personal computer, a tablet, a smartphone, and a cloud server.

[0013] The creation device 1A (generation device 1B and prediction device 1C) of the present embodiment includes an input device 3, an output device 4, and an arithmetic processing device 5.

[0014] [Input section, output section, arithmetic processing device] The input device 3 is configured as an input device. Examples of this input device include a keyboard or a mouse. The output device 4 is configured as an output device. Examples of this output device include a display device or a printer. The arithmetic processing device 5 is for executing the creation of a learning data set, the generation of a learned model, and the prediction of the performance of a target tire. The arithmetic processing device 5 of the present embodiment includes one or a plurality of processors (arithmetic units) 5A that perform various arithmetic operations, a storage unit 5B in which data, programs, etc. are stored, and a working memory 5C.

[0015] [Processor] The processor 5A in this embodiment is configured as a central processing unit (CPU), but it is not particularly limited. For example, it may be composed of a microprocessor or other processing units. In this embodiment, an example is illustrated in which the method for creating a learning dataset, the method for generating a learned model, and the method for predicting the performance of a tire are executed by one processor 5A, but they may also be executed (parallel processing) by a plurality of processors.

[0016] [Storage unit] The storage unit 5B is, for example, a non-volatile information storage device composed of a magnetic disk, an optical disk, an SSD, or the like. The storage unit 5B is provided with a data unit 7 and a program unit 8.

[0017] [Data unit] The data unit 7 is for storing data and the like necessary for executing the method for creating a learning dataset (hereinafter sometimes referred to as the "creation method"). Further, the data unit 7 in this embodiment stores data and the like necessary for executing the method for generating a learned model (hereinafter sometimes referred to as the "generation method") and the method for predicting the performance of a tire (hereinafter sometimes referred to as the "prediction method").

[0018] The data unit 7 in this embodiment includes a first data storage unit 7A, a second data storage unit 7B, a performance data storage unit 7C, a learning dataset storage unit 7D, a model storage unit 7E, a target data storage unit 7F, and a target performance data storage unit 7G. Note that the data unit 7 is not necessarily limited to such a form, and may include a storage unit for storing other data as necessary, or some of these may be omitted. Also, the data stored in each data unit 7 will be described in each step of the creation method, generation method, and prediction method described later.

[0019] [Program unit] The program unit 8 of this embodiment is a program (application) necessary for executing the creation method, generation method, and prediction method. When such a program unit 8 is executed by the processor (arithmetic unit) 5A, the computer 1 (creation device 1A, generation device 1B, and prediction device 1C) can be made to function as a specific means. The program unit 8 of this embodiment includes an acquisition unit 8A, a replacement unit 8B, a performance data acquisition unit 8C, a dataset creation unit 8D, a preparation unit 8E, a learning unit 8F, an input unit 8G, a derivation unit 8H, and an evaluation unit 8J. Note that the program unit 8 is not limited to this form, and other programs may be included or some of these may be omitted as needed. Furthermore, the function of each program unit 8 will be explained in each step of the prediction method and generation method described later.

[0020] [tire] Figure 2 is a meridian cross-sectional view of a tire, showing an example of a tire 11. In this embodiment, a pneumatic tire mounted on a passenger car is shown as an example of a tire 11, but the embodiment is not limited to this configuration. The tire 11 may be a tire of another category, such as a heavy-duty tire for trucks and buses.

[0021] The tire 11 of this embodiment includes a tread portion 11a, a pair of sidewall portions 11b, and a pair of bead portions 11c.

[0022] [Tire components] The tire 11 of this embodiment is composed of a plurality of tire components 13. The tire components 13 of this embodiment include a fiber member 14 and a rubber member 15.

[0023] [Fiber material] The fiber member 14 is formed by covering a fibrous material (not shown) with topping rubber. Therefore, the fiber member 14 is configured as a cord ply. The fiber member 14 in this embodiment includes a carcass ply 16A and belt plies 17A and 17B.

[0024] The carcass ply 16A constitutes the carcass 16 and extends between the pair of bead portions 11c. In this embodiment, the carcass 16 is composed of one carcass ply 16A, but the invention is not limited to this configuration. For example, the carcass 16 may be composed of two or more carcass plies (not shown).

[0025] The carcass ply 16A includes, for example, a main body portion 16a and a folded portion 16b. The main body portion 16a extends, for example, between two bead portions 11c. The folded portion 16b is connected to the main body portion 16a and is folded back around the bead core 12 from the inside to the outside in the tire axial direction. Between the main body portion 16a and the folded portion 16b, a bead apex rubber 15d is provided, extending radially outward from the bead core 12.

[0026] The carcass ply 16A consists of multiple carcass cords (fibrous material) arranged in parallel (not shown). The arrangement angle of these carcass cords (not shown) is set to, for example, 80 to 90 degrees with respect to the tire equator C. Organic fiber cords such as aramid and rayon may be used for the carcass cords.

[0027] [Belt layer] The belt plies 17A and 17B constitute the belt layer 17. In this embodiment, the belt layer 17 is composed of two belt plies 17A and 17B, but the embodiment is not limited to this configuration. For example, the belt layer 17 may be composed of one belt ply (not shown), or it may be composed of three or more belt plies (not shown).

[0028] Belt plies 17A and 17B consist of multiple belt cords (fibrous material) arranged in parallel (not shown). The arrangement angle of these belt cords (not shown) is set to, for example, 10 to 35 degrees with respect to the tire circumferential direction. Steel cords are used for the belt cords, but highly elastic organic fiber cords such as aramid or rayon may be used as needed.

[0029] The properties of the fiber members 14 (carcass ply 16A and belt ply 17A, 17B) can be specified by member information including physical properties. These physical properties include, for example, the modulus of elasticity of the fiber member 14. This modulus of elasticity is calculated by performing a tensile test in accordance with JIS-L1017's "Test Method for Chemical Fiber Tire Cords," under conditions of a gripping distance of 250 mm and a tensile speed of 300 ± 20 mm / min, and converting the slope of the tangent line at the point corresponding to a predetermined load (e.g., 2.0 cN / dtex) in the load-elongation curve to a value per tex. The member information also includes the arrangement angle of the fiber members (fiber material). Furthermore, the member information may include the type of fiber member (fiber material) 14, the cord diameter, and the cord structure. Since this member information (including physical properties) can specify the rigidity of the tire 11, it has a high correlation with the performance data of the tire 11, which is affected by its rigidity.

[0030] [Rubber material] The rubber component 15 is composed of a tread rubber 15a, a sidewall rubber 15b, a clinch rubber 15c, a bead apex rubber 15d, and an inner liner rubber 15e. However, the rubber component 15 is not limited to this configuration, and may include, for example, other rubber components.

[0031] The tread rubber 15a is positioned on the radially outer side of the belt layer 17 in the tread portion 11a. This tread rubber 15a is composed of a cap rubber 19 that forms the outer surface (contact surface) of the tread portion 11a and a base rubber 20 positioned radially inward of the cap rubber 19. The sidewall rubber 15b is positioned radially inward of the tread rubber 15a (in this example, on the axially outer side of the carcass 16 in the sidewall portion 11b).

[0032] The clinch rubber 15c is located radially inward of the sidewall rubber 15b (in this example, the bead portion 11c). The bead apex rubber 15d extends radially outward from the bead core 12. The inner liner rubber 15e is located on the inner surface of the carcass 16.

[0033] The characteristics of the rubber components 15 (tread rubber 15a (cap rubber 19 and base rubber 20), sidewall rubber 15b, clinch rubber 15c, bead apex rubber 15d, and inner liner rubber 15e) can be specified by component information including physical properties. These physical properties include, for example, the elastic modulus and loss tangent of the rubber components 15, but are not limited to this configuration; for example, tensile strength may also be included. Furthermore, the component information for the rubber components 15 may include the compounding, vulcanization conditions, and temperature history during vulcanization (relationship between rubber temperature and vulcanization time). Since this component information (physical properties) can specify the rigidity of the tire 11, it has a high correlation with the performance data of the tire 11, which is affected by its rigidity.

[0034] In this embodiment, the elastic modulus is exemplified as the complex elastic modulus, but is not limited to this form; for example, it may be the storage modulus or the storage modulus. These elastic moduli and loss tangents can be measured as appropriate. If the member information in this embodiment includes the complex elastic modulus and loss tangent, these complex elastic moduli and loss tangents can be measured using a viscoelastic spectrometer under the following conditions in accordance with the provisions of JIS-K6394. Initial distortion: 10% Amplitude: ±2% Frequency: 10Hz Deformation mode: Tension Temperature: 100℃ Viscoelastic spectrometer: GABO Corporation's "Iplexer®" (registered trademark)

[0035] By the way, in order to generate a machine learning model to predict the performance of tire 11, a training dataset is required. If this training dataset includes an image of tire 11 (not shown), then the pixels (i.e., picture elements) that make up that image have values ​​set to represent the intensity of light and color. Although these values ​​are extracted as machine learning features, the values ​​themselves have an extremely low correlation with tire performance data.

[0036] In the creation method of this embodiment (creation apparatus 1A shown in Figure 1), the pixel values ​​of the cross-sectional image of the tire 11 are replaced with material information including the physical properties of the tire components 13 (fiber member 14 and rubber member 15) corresponding to the pixels. As described above, since this material information (including physical properties) can identify the rigidity of the tire 11, it has a high correlation with the performance data of the tire 11, which is affected by its rigidity. In this way, by replacing the pixel values, which have a very low correlation with the performance data, with material information (physical properties) that have a high correlation with the performance data, it becomes possible to create a training dataset that includes that material information. By using such a training dataset in machine learning, a trained model capable of accurately estimating performance data can be generated.

[0037] [Method for creating a training dataset (first embodiment)] In the creation method of this embodiment, a training dataset is created. The training dataset of this embodiment is used for machine learning (i.e., generation of a trained model) to estimate the performance data of the tire 11 shown in Figure 2. The trained model of this embodiment is a model that has been trained to output the performance data of the tire 11 when the second data described later is input. To create such a trained model, it is effective to train a machine learning model using a training dataset that combines the second data and performance data obtained from multiple types of tires 11.

[0038] The second data set in the training dataset is obtained by replacing the pixel values ​​of the cross-sectional image of the tire 11 with material information including the physical properties of the tire component 13 corresponding to the pixels. Therefore, the training dataset including the second data set may contain material information of the tire component 13 that has a high correlation with the performance data of the tire 11.

[0039] The categories of the multiple types of tires 11 are not particularly limited. For example, from the viewpoint of improving the prediction accuracy of the trained model (machine learning model), the multiple types of tires 11 may be in the same category, or from the viewpoint of improving the versatility of the trained model (machine learning model), they may be in different categories. Furthermore, it is preferable that the multiple types of tires 11 differ from each other in at least one aspect, such as the tire components 13, the component information including the physical properties of the tire components 13, the category, and the tire size.

[0040] Figure 3 is a flowchart showing an example of the processing steps for creating a training dataset. Each step of the creation method in this embodiment is performed by one or more processors 5A included in the creation device 1A (computer 1) shown in Figure 1.

[0041] [Retrieve first data] In the manufacturing method of this embodiment, first, first data is acquired (step S1). The first data consists of a cross-sectional image in which the tire components 13 are represented by pixel values ​​in the meridian cross-section of the tire 11 shown in Figure 2. Figure 4 shows an example of the first data 21 consisting of the cross-sectional image 21A. Figure 4 shows the first data 21 of one tire 11 as representative.

[0042] In step S1 of this embodiment, first, the acquisition unit 8A included in the program unit 8 shown in Figure 1 is loaded into the working memory 5C. The acquisition unit 8A is a program for acquiring the first data 21 shown in Figure 4. When this acquisition unit 8A is executed by the processor 5A, the computer 1 (creation device 1A) can be made to function as a means for acquiring the first data 21.

[0043] In step S1 of this embodiment, first data 21 consisting of cross-sectional images 21A shown in Figure 4 is acquired for each of several types of tires 11. The cross-sectional images 21A (first data 21) can be acquired as appropriate. For example, the cross-sectional images 21A may be acquired by actually photographing the meridian cross-section of the tire 11 shown in Figure 2, or they may be acquired from the design data (CAD data) of the tire 11, or they may be hand-drawn images by an operator. Computed tomography (CT), which involves irradiating the tire 11 with X-rays and taking images, may be used to acquire the cross-sectional images. Alternatively, the cross-sectional images 21A may be acquired from a tire model (not shown) modeled based on the design data of the tire 11. This tire model can be modeled as appropriate based on, for example, a known method (for example, the method described in Japanese Patent Application Publication No. 2023-084539). In this embodiment, the cross-sectional images 21A can be acquired from a tire model. This allows cross-sectional images 21A (first data 21) to be acquired at low cost and in a short time without actually manufacturing multiple types of tires 11.

[0044] The cross-sectional image 21A (first data 21) can be configured as, for example, a color image, a grayscale image, or a black and white image. In this embodiment, the first data 21 is configured as a color image. Compared to grayscale and black and white images, such a color image has a larger range of values ​​that its pixels can hold, and can be easily replaced with member information including physical property values. However, the first data 21 is not limited to being configured as a color image. For example, if the range of values ​​for the member information is small, the first data 21 may be configured as a grayscale image or a black and white image.

[0045] The cross-sectional image 21A (first data 21) can be acquired as appropriate. In this embodiment, the cross-sectional image 21A can be acquired on a tire 11 (shown in Figure 2) that is not filled with internal pressure. This allows the cross-sectional image 21A to be acquired in a shorter time compared to, for example, when the tire is filled with internal pressure. Note that the cross-sectional image 21A is not limited to being acquired from a tire 11 that is not filled with internal pressure. For example, the cross-sectional image 21A may be acquired on a tire 11 that is mounted on a rim and filled with a predetermined internal pressure, or it may be acquired on a tire 11 that is subjected to a predetermined load. This stabilizes the shape of the tire 11, and furthermore, cross-sectional images 21A of multiple types of tires 11 can be acquired under the same conditions.

[0046] The rim (not shown in the illustration) is exemplified as a standard rim. A "standard rim" is the rim specified for each tire in the standards system, including the standard on which the tire 11 is based. Therefore, a standard rim is, for example, a "standard rim" for JATMA, a "Design Rim" for TRA, or a "Measuring Rim" for ETRTO.

[0047] The internal pressure shown is the standard internal pressure. "Standard internal pressure" refers to the air pressure specified for each tire in the standards system, including the standard on which the tire is based. Therefore, the standard internal pressure is, for example, the "maximum air pressure" for JATMA, the maximum value listed in the table "TIRE LOAD LIMITS AT VARIOUS COLD INFLATION PRESSURES" for TRA, and the "INFLATION PRESSURE" for ETRTO.

[0048] The load shown is an example of the standard load. The "standard load" is the load specified for each tire in the standards system, including the standard on which the tire 11 is based. Therefore, the standard load is, for example, the "maximum load capacity" for JATMA, the maximum value listed in the table "TIRE LOAD LIMITS AT VARIOUS COLD INFLATION PRESSURES" for TRA, and the "LOAD CAPACITY" for ETRTO.

[0049] The cross-sectional image 21A (first data 21) shows the tire components 13 (fiber member 14 and rubber member 15). Figure 5(a) is a conceptual diagram showing an example of a pixel 23 in the cross-sectional image 21A (first data 21) of Figure 4. In Figure 5(a), the pixels 23 of the cap rubber 19 and the base rubber 20 are shown representatively, and the values ​​24 of those pixels (hereinafter sometimes referred to as "pixel values") are shown representatively.

[0050] As shown in Figure 5(a), the pixels 23 representing the tire components 13 are assigned pixel values ​​24 to represent the intensity and color of light. In this embodiment, when the cross-sectional image 21A (first data 21) is configured as a color image, the pixel values ​​24 can be configured by a combination of the R channel value 24r, the G channel value 24g, and the B channel value 24b. In this embodiment, different pixel values ​​24 are assigned to the pixels 23 representing the cap rubber 19 and the pixels 23 representing the base rubber 20. As a result, the cap rubber 19 and the base rubber 20 are shown in different colors, making it easy to distinguish between them in the cross-sectional image 21A (first data 21). Furthermore, different pixel values ​​24 are assigned to the pixels 23 representing the carcass ply 16A, belt ply 17A, 17B, sidewall rubber 15b, clinch rubber 15c, bead apex rubber 15d, and inner liner rubber 15e shown in Figure 4. As a result, the tire component 13 can be distinguished in the cross-sectional image 21A (first data 21).

[0051] In this embodiment, as shown in Figure 4, cross-sectional images 21A (first data 21) on both sides in the tire axial direction relative to the tire equator C are shown, but this is not particularly limited, and for example, a cross-sectional image 21A on either one side in the tire axial direction may be shown. Also, as shown in Figure 4, the color of the tire component 13 is different on both sides in the tire axial direction relative to the tire equator C, but it may be the same color. The cross-sectional image 21A (first data 21) is input to the first data storage unit 7A shown in Figure 1.

[0052] [Generate second data] Next, in the creation method of this embodiment, second data is generated (step S2). The second data is obtained by replacing the value (pixel value) 24 of pixel 23 in the cross-sectional image 21A shown in Figure 5(a) with member information (not shown) that includes the physical property values ​​of the tire component 13 corresponding to pixel 23. In this embodiment, the second data is generated for cross-sectional images 21A (first data 21) obtained from each of multiple types of tires 11.

[0053] In step S2 of this embodiment, the first data 21 (cross-sectional image 21A) input to the first data storage unit 7A shown in Figure 1, and the replacement unit 8B included in the program unit 8 are loaded into the working memory 5C. The replacement unit 8B is a program for generating second data by replacing the value (pixel value) 24 of pixel 23 in the cross-sectional image 21A shown in Figure 5(a) with member information including the physical property values ​​of the tire component 13 corresponding to pixel 23. When this replacement unit 8B is executed by the processor 5A, the computer 1 (creation device 1A) can be made to function as a means for generating the second data.

[0054] In step S2 of this embodiment, first, component information including the physical properties of the tire component 13 is identified. The component information may be obtained, for example, based on a known procedure as described above, or it may be a design value.

[0055] Figure 6 is a table showing an example of the physical properties 25 of the tire component 13. In this table, the physical properties 25 of the tire component 13 of one type of tire 11 out of several types of tires 11 are shown as representative. It is preferable that such a table be created in which the physical properties 25 of the tire component 13 are obtained for each of several different types of tires 11.

[0056] Figure 6 shows the pixel values ​​24 of the tire component 13 in the cross-sectional image 21A shown in Figure 5(a). By correlating the pixel values ​​24 in the cross-sectional image 21A shown in Figure 5(a) with the pixel values ​​24 in the table shown in Figure 6, the component information 26, including the physical properties 25 of the tire component 13, can be uniquely identified.

[0057] Figure 6 shows the complex modulus of elasticity 25g and the loss tangent 25b as physical properties 25 (member information 26) of the rubber member 15. Note that the physical properties 25 are not limited to this form, and may include, for example, the above physical properties and vulcanization conditions. Similarly, the physical properties 25 (member information 26) of the fiber member 14 include the modulus of elasticity and the arrangement angle (both not shown), but are not limited to this form, and may include, for example, the cord diameter and the structure of the cord.

[0058] Next, in step S2 of this embodiment, the value (pixel value) 24 of pixel 23 in the cross-sectional image 21A shown in Figure 5(a) is replaced with member information 26 (shown in Figure 6) which includes the physical property value 25 of the tire component 13 corresponding to pixel 23.

[0059] In this embodiment, in the cross-sectional image 21A shown in Figures 4 and 5(a), the value 24 of pixel 23 representing the rubber member 15 is replaced with at least one (in this example, all) of the elastic modulus and loss tangent of the rubber member 15 corresponding to pixel 23. Furthermore, in this embodiment, the value 24 of pixel 23 representing the fiber member 14 is replaced with at least one (in this example, all) of the elastic modulus and arrangement angle of the fiber member 14 corresponding to pixel 23. As a result, a second data can be generated in which the pixel values ​​24 of pixels 23 representing the fiber member 14 and rubber member 15 in the cross-sectional image 21A (first data 21) are replaced with the physical property values ​​25 (member information 26) of the fiber member 14 and rubber member 15.

[0060] The substitution of pixel values ​​24 with member information 26 can be carried out as appropriate, for example, based on the cross-sectional images 21A shown in Figures 4 and 5(a) and the table showing physical property values ​​25 shown in Figure 6.

[0061] As described above, by matching the pixel values ​​24 of the cross-sectional image 21A shown in Figure 5(a) with the pixel values ​​24 in the table shown in Figure 6, the component information 26, including the physical properties 25 of the tire component 13, can be uniquely identified. This allows for easy replacement of the pixel values ​​24 of the cross-sectional image 21A for each tire component 13 with the component information 26, including the physical properties 25 of the tire component 13 corresponding to pixel 23. Such replacement can be performed using, for example, commercially available image editing software or the "+opencv" library for the well-known programming language "Python".

[0062] Figure 5(b) is a conceptual diagram showing an example of second data 22 in which the pixel values ​​24 of the cross-sectional image 21A in Figure 5(a) are replaced with member information 26 including material property values ​​25. As shown in Figure 5(b), the pixel values ​​24 of the cap rubber 19 and base rubber 20 shown in Figure 5(a) are replaced with the material property values ​​25 (complex modulus of elasticity 25g and loss tangent 25b) of the cap rubber 19 and base rubber 20 shown in the table in Figure 6.

[0063] In this embodiment, the value 24g of the G channel shown in Figure 5(a) is replaced with the complex modulus 25g as shown in Figure 5(b), and the value 24b of the B channel shown in Figure 5(a) is replaced with the loss tangent 25b as shown in Figure 5(b). When the loss tangent contains a decimal number, as in this embodiment, it can be replaced, for example, based on a fixed-point format. Also, if there is no physical property value to be replaced, such as the value 24r of the R channel shown in Figure 5(a), it may be replaced with "0" as shown in Figure 5(b), or the R channel may be deleted. Note that the replacement of member information 26 including physical property values ​​25 is not limited to these forms, and the value 24r of the R channel may be replaced with the complex modulus 25g or the loss tangent 25b, or with other physical property values. In addition, a new channel capable of storing other physical property values ​​25 may be added separately from the three channels (R channel, G channel, and B channel).

[0064] Figure 7 shows an example of a training dataset 28. In Figure 7, a representative training dataset 28 obtained from one type of tire 11 is shown. In the second data 22 of Figure 7, the pixels 23 and material properties 25 (material information 26) shown in Figure 5(b) are omitted.

[0065] In the second data set 22, similar to the first data set 21 (cross-sectional image 21A) shown in Figure 4, the position (coordinate values) of the tire components 13 can be identified by pixels 23 (shown in Figure 5(b)) that have component information 26 including physical property values ​​25. On the other hand, unlike the first data set 21 (cross-sectional image 21A), the second data set 22 does not have pixel values ​​24 (shown in Figure 5(a)) that represent light intensity and color, but instead has component information 26 (shown in Figure 5(b)) including physical property values ​​25. Therefore, the second data set 22 is constructed as a matrix in which component information 26 including physical property values ​​25 is arranged in pixels 23 corresponding to each position of the tire components 13, and thus differs from a general image like the first data set 21 shown in Figure 4. Thus, although the second data set (matrix) 22 differs from a general image, component information 26 including physical property values ​​25 is identified based on the same arrangement of pixels 23 as in an image, making it possible to input it into a machine learning model (trained model) as a pseudo-image.

[0066] As described above, the component information 26 (shown in Figure 5(b)), which includes the physical properties 25 of the tire component 13, can identify the rigidity of the tire 11, and therefore has a high correlation with the performance data of the tire 11, which is affected by its rigidity. In the creation method (creation device 1A) of this embodiment, the second data 22 shown in Figures 5(b) and 7 is generated, making it possible to create a training dataset 28 that includes component information 26 (including physical properties 25) which has a high correlation with the performance data. The second data 22 is input to the second data storage unit 7B shown in Figure 1.

[0067] [Obtain tire performance data] Next, in the manufacturing method of this embodiment, performance data 27 for the tire 11 shown in Figure 7 is acquired (step S3). In step S3 of this embodiment, performance data 27 is acquired for multiple types of tires 11.

[0068] In step S3 of this embodiment, first, the performance data acquisition unit 8C included in the program unit 8 shown in Figure 1 is loaded into the working memory 5C. The performance data acquisition unit 8C is a program for acquiring performance data 27 of the tire 11. When this performance data acquisition unit 8C is executed by the processor 5A, the computer 1 (creation device 1A) can be made to function as a means for acquiring the performance data 27.

[0069] The performance data 27 is not particularly limited as long as it relates to the performance of the tire 11. The performance data 27 in this embodiment includes at least one of the rolling resistance coefficient of the tire 11, the stiffness of the tire, and the air resistance value of the tire. Since these performance data 27 are affected by the stiffness of the tire 11, which can be identified by the material information 26 (including physical properties 25) of the tire component 13 shown in Figure 5(b), they have a high correlation with the second data 22 which includes the material information 26 (physical properties 25).

[0070] Performance data 27 can be obtained as appropriate using known procedures. The rolling resistance coefficient of tire 11 can be obtained, for example, by using a known rolling resistance tester. The stiffness of tire 11 can be obtained, for example, by measuring the longitudinal spring constant of tire 11. The air resistance value of tire 11 can be obtained, for example, by wind tunnel testing of tire 11. Furthermore, these performance data 27 may also be obtained by simulation using computer 1 (shown in Figure 1). It is preferable to use the tire model described above for such simulations.

[0071] The performance data 27 may include at least one of the following: the rolling resistance coefficient of the tire 11, the stiffness of the tire 11, and the air resistance value of the tire 11; all of these may be obtained; or other performance data may be obtained. In this embodiment, the rolling resistance coefficient 27A of the tire 11 is obtained as performance data 27. Figure 7 shows the case where the rolling resistance coefficient 27A is "6.0". The performance data 27 is input to the performance data storage unit 7C shown in Figure 1.

[0072] [Create a training dataset] Next, in the creation method of this embodiment, a training dataset 28 (shown in Figure 7) is created (step S4). In step S4 of this embodiment, the training dataset 28 is created by combining the second data 22 and the performance data 27.

[0073] In step S4 of this embodiment, first, the second data 22 input to the second data storage unit 7B shown in Figure 1 and the performance data 27 input to the performance data storage unit 7C are loaded into the working memory 5C. Furthermore, the dataset creation unit 8D included in the program unit 8 is loaded into the working memory 5C. The dataset creation unit 8D is a program for creating a training dataset 28. When this dataset creation unit 8D is executed by the processor 5A, the computer 1 (creation device 1A) can be made to function as a means for creating the training dataset 28.

[0074] In step S4 of this embodiment, the second data 22 and performance data 27 are combined for each of the multiple types of tires 11. This creates multiple training datasets 28. The training datasets 28 are input to the training dataset storage unit 7D shown in Figure 1.

[0075] In this embodiment, in the method for generating a trained model described later, a training dataset 28 created based on multiple types of tires 11 is used to train the machine learning model. This makes it possible to generate a trained model that can output performance data 27 (rolling resistance coefficient 27A in this example) of a target tire when a second data 22 of the target tire is input.

[0076] The training dataset 28 of this embodiment includes second data 22 containing material information 26 (physical properties 25) of the tire components 13, which has a high correlation with the performance data 27. Such second data 22 allows for the effective extraction of features necessary for estimating the performance data 27. Therefore, the creation method (creation device 1A) of this embodiment makes it possible to create a training dataset 28 capable of generating a trained model that can accurately estimate the performance data 27.

[0077] [Method for generating a pre-trained model (First Embodiment)] Next, an example of the processing procedure for generating a pre-trained model is shown. This generation method generates a pre-trained model for estimating tire performance data 27. The pre-trained model in this embodiment is a machine learning model that, when second data 22 is input for the target tire (the tire to be evaluated 11), outputs the performance data 27 of the target tire. To generate such a pre-trained model, the training dataset 28 (shown in Figure 7) created by the creation method shown in Figure 3 is used.

[0078] Figure 8 is a flowchart showing an example of the processing steps for generating a trained model. Each step of the generation method in this embodiment is performed by one or more processors 5A included in the generation device 1B (computer 1) shown in Figure 1.

[0079] [Prepare the training dataset] In the generation method of this embodiment, first, the training dataset 28 shown in Figure 7 is prepared (step S5). The training dataset 28 is a combination of the second data 22 and performance data 27 of the tire 11.

[0080] In step S5 of this embodiment, first, the preparation unit 8E included in the program unit 8 shown in Figure 1 is loaded into the working memory 5C. The preparation unit 8E is a program for preparing the training dataset 28. When this preparation unit 8E is executed by the processor 5A, the computer 1 (generator 1B) can be made to function as a means for acquiring the training dataset 28.

[0081] In step S5 of this embodiment, a training dataset 28 (shown in Figure 7) is created based on the creation method shown in Figure 3. As described above, in this embodiment, a training dataset 28 is created for multiple types of tires 11. If the training dataset 28 has already been created before the generation method is performed, step S5 may be omitted. The training dataset 28 is input to the training dataset storage unit 7D shown in Figure 1.

[0082] [Train the machine learning model] Next, in the generation method of this embodiment, the machine learning model is trained using the training dataset 28 (step S6). In this embodiment, the machine learning model is trained so that when the second data 22 of the target tire is input, the performance data 27 of the target tire (in this example, the rolling resistance coefficient 27A) is output.

[0083] In step S6 of this embodiment, first, the training dataset 28 (shown in Figure 7) input to the training dataset storage unit 7D shown in Figure 1, and the learning unit 8F included in the program unit 8 are loaded into the working memory 5C. The learning unit 8F is a program that uses the training dataset 28 to train the machine learning model. When this learning unit 8F is executed by the processor 5A, the computer 1 (generator 1B) can be made to function as a means for training the machine learning model.

[0084] The machine learning model is not particularly limited, as long as it can be trained to output performance data 27 (shown in Figure 7) of the target tire when second data 22 (shown in Figure 7) of the target tire is input. In this embodiment, it is desirable to use a deep learning model equipped with a neural network as the machine learning model. This makes it possible for the machine learning model itself to extract features from the input second data 22 and estimate the performance data 27. Such a machine learning model can be appropriately constructed based on known methods such as multiple regression, ridge regression, random forest, Gaussian process regression, CNN, and GNN.

[0085] As described above, although the second data (matrix) 22 input to the machine learning model is different from a typical image, component information 26 including material properties 25 is identified based on the arrangement of pixels 23 similar to the image shown in Figure 5(b). Therefore, the second data 22 can be treated as a pseudo-image. In this case, the machine learning model (deep learning model) is preferably composed of a convolutional neural network (CNN) capable of extracting image features.

[0086] The convolutional neural network (CNN) may be a publicly known one, or a customized version of such a publicly known one. In this embodiment, a customized version of VGG16, one of the publicly known CNNs, may be used as the machine learning model (trained model). Figure 9 is a conceptual diagram showing an example of machine learning model 34A (trained model 34B).

[0087] The machine learning model 34A (trained model 34B) includes an input layer 35, an output layer 36, a convolutional layer 37, a pooling layer 38, and a fully connected layer 39.

[0088] The input layer 35 is capable of receiving the second data 22. Unlike VGG16, the output layer 36 (third fully connected layer 39C) has only one node. As a result, the machine learning model 34A of this embodiment can output a single performance data value 27 (in this example, the tire rolling resistance coefficient 27A) when the second data 22 is input to the input layer 35.

[0089] In the convolutional layer 37, filtering is performed on the second data 22. This generates a feature map. The convolutional layer 37 is defined in each of the multiple convolutional blocks 40. These convolutional blocks 40 consist of a first convolutional block 40A and a second convolutional block 40B. Two convolutional layers 37 are defined in the first convolutional block 40A and the second convolutional block 40B, respectively.

[0090] In the pooling layer 38, the size of the feature map generated in the convolutional layer 37 is reduced. The pooling layer 38 consists of a first pooling layer 38A and a second pooling layer 38B. The first pooling layer 38A is defined between the first convolutional block 40A and the second convolutional block 40B. The second pooling layer 38B is defined between the second convolutional block 40B and the fully connected layer 39.

[0091] The fully connected layer 39 combines feature maps and outputs performance data 27 (in this example, the rolling resistance coefficient of the tire) to the output layer 36. The fully connected layer 39 in this embodiment includes a first fully connected layer 39A, a second fully connected layer 39B, and a third fully connected layer 39C. Of these first to third fully connected layers 39A to 39C, the third fully connected layer 39C is configured as the output layer 36.

[0092] In step S6 of this embodiment, first, the second data 22 included in one of the training datasets 28 (not shown) shown in Figure 7 is input to the machine learning model 34A. This allows the machine learning model 34A to output performance data 27 (in this example, the tire rolling resistance coefficient 27A) estimated by the machine learning model 34A. Next, in step S6 of this embodiment, the error between the output performance data (estimated data) 27 and the performance data (ground truth data) 27 included in the training dataset 28 shown in Figure 7 is calculated. Then, various parameters of the machine learning model 34A (e.g., weighting coefficients, biases, etc.) are updated so that this error is minimized. In this way, the machine learning model 34A can be optimized by performing a series of steps from inputting the training dataset 28 to updating the parameters for each of the multiple training datasets 28.

[0093] In step S6 of this embodiment, the machine learning model 34A is optimized using multiple training datasets 28 (not shown). As a result, when the second data 22 of the target tire is input, a trained model 34B is generated that can output the performance data 27 of the target tire (in this example, the rolling resistance coefficient 27A). With such a trained model 34B, it becomes possible to estimate the performance data 27 of the target tire without requiring the experience or intuition of an expert. The trained model 34B is input to the model storage unit 7E (shown in Figure 1).

[0094] As described above, the performance of the tire 11 shown in Figure 2 (performance data 27) is affected by the stiffness of the tire 11, which can be identified by the material information 26 (physical properties 25) of the tire components 13 shown in Figures 5(b) and 6. For this reason, the performance data 27 has a high correlation with the second data 22 (shown in Figures 5(b) and 7), which includes the material information 26 (physical properties 25). When this second data 22 is input into the machine learning model 34A shown in Figure 9, features for estimating the performance data 27 can be effectively extracted. Therefore, the generation method (generation device 1B) of this embodiment makes it possible to generate a trained model 34B (shown in Figure 7) that can accurately estimate the performance data 27.

[0095] Furthermore, in this embodiment, the trained model 34B is generated based on the conventional model (VGG16). This allows the trained model 34B to be created in a short amount of time.

[0096] [Method for predicting tire performance (first embodiment)] Next, an example of the processing procedure for a tire performance prediction method is shown. In this prediction method, the second data 22 of the target tire (tire 11 to be evaluated) is input to a trained model 34B, and the trained model 34B outputs performance data 27 of the target tire. Such a trained model 34B can be generated based on the generation method shown in Figure 8.

[0097] Figure 10 is a flowchart showing an example of the processing procedure for a tire performance prediction method. Each step of the prediction method in this embodiment is performed by one or more processors 5A included in the prediction device 1C (computer 1) shown in Figure 1.

[0098] [Enter the second data for the target tire] In the prediction method of this embodiment, first, the second data 22 of the target tire is input to the trained model 34B shown in Figure 9 (step S7). This trained model 34B is machine-trained to output performance data 27 of the tire 11 (in this example, the tire rolling resistance coefficient 27A) when the second data 22 is input.

[0099] In step S7 of this embodiment, the trained model 34B (shown in Figure 9) input to the model storage unit 7E shown in Figure 1, and the input unit 8G included in the program unit 8 are loaded into the working memory 5C. The input unit 8G is a program for inputting the second data 22 of the target tire to the trained model 34B. When this input unit 8G is executed by the processor 5A, the computer 1 (prediction device 1C) can be made to function as a means for inputting the second data 22 to the trained model 34B.

[0100] In step S7 of this embodiment, first, first data 21 (shown in Figure 4), consisting of a cross-sectional image 21A of the target tire, is acquired. This first data 21 (cross-sectional image 21A) is acquired based on the same method as in step S1 of the creation method shown in Figure 3, and the tire components 13 of the target tire are represented by pixel values ​​24 (shown in Figure 5(a)). The acquired first data 21 of the target tire is input to the target data storage unit 7F (shown in Figure 1).

[0101] Next, in step S7 of this embodiment, second data 22 (shown in Figure 7) of the target tire is generated. This second data 22 is generated based on the same method as in step S2 of the creation method shown in Figure 3. Therefore, the pixel value 24 (shown in Figure 5(a)) of pixel 23 in the cross-sectional image 21A is replaced with member information 26 (shown in Figure 5(b)) which includes the physical property value 25 of the tire component 13 corresponding to pixel 23. Since such second data 22 includes member information 26 (physical property value 25) that can identify the rigidity, etc., of the target tire, it has a high correlation with the performance data 27 of the target tire, which is affected by its rigidity, etc. The acquired second data 22 of the target tire is input to the target data storage unit 7F (shown in Figure 1).

[0102] Next, in step S7 of this embodiment, as shown in Figure 9, the second data 22 of the target tire is input to the trained model 34B. This makes it possible to output performance data 27 of the target tire (in this example, the tire's rolling resistance coefficient 27A) from the trained model 34B.

[0103] [Output performance data for the target tire] Next, the prediction method of this embodiment outputs performance data 27 of the target tire from the trained model 34B (step S8).

[0104] In step S8 of this embodiment, first, the derivation unit 8H included in the program unit 8 shown in Figure 1 is loaded into the working memory 5C. The derivation unit 8H is a program for outputting performance data 27 of the target tire from the trained model 34B shown in Figure 9. When this derivation unit 8H is executed by the processor 5A, the computer 1 (prediction device 1C) can be made to function as a means for outputting the performance data 27 of the target tire.

[0105] In this embodiment, in step S7, as shown in Figure 9, the second data 22 of the target tire is input to the trained model 34B. As a result, in step S8, the trained model 34B can output performance data 27 of the target tire (in this example, the tire's rolling resistance coefficient 27A). Therefore, the prediction method of this embodiment makes it possible to estimate the performance data 27 of the target tire without requiring the experience or intuition of an expert.

[0106] As shown in Figure 5(b), the second data set 22 includes component information 26 (physical property values ​​25) that can identify the stiffness of the target tire, and therefore has a high correlation with the performance data 27 of the target tire (shown in Figure 9), which is affected by its stiffness. By inputting such second data set 22 into the machine learning model 34A, it becomes possible to estimate the performance data 27 with high accuracy.

[0107] The performance data 27 (in this example, the tire rolling resistance coefficient 27A) output from the trained model 34B is input to the target performance data storage unit 7G shown in Figure 1. Furthermore, in step S8, the performance data 27 output from the trained model 34B may be output (displayed) to an output device 4, for example, a display device. This allows operators and others to understand the performance data 27 and to evaluate the performance of the target tire.

[0108] [Evaluate the performance data of the target tires] Next, in the prediction method of this embodiment, the quality of the performance data 27 (shown in Figure 9) of the target tire is evaluated (step S9). The quality of the performance data 27 may be evaluated by the prediction device 1C (computer 1) shown in Figure 1, or by an operator or the like.

[0109] In step S9 of this embodiment, first, the performance data 27 (in this example, the tire rolling resistance coefficient 27A) input to the target performance data storage unit 7G shown in Figure 1, and the evaluation unit 8J included in the program unit 8 are loaded into the working memory 5C. This evaluation unit 8J is a program for determining whether the performance data 27 of the target tire is good or not based on the output performance data 27. When this evaluation unit 8J is executed by the processor 5A, the computer 1 (prediction device 1C) can be made to function as a means for evaluating the performance data 27.

[0110] The quality of the performance data 27 can be evaluated as appropriate. If the performance data 27 in this embodiment is the rolling resistance coefficient 27A of the tire, for example, if the rolling resistance coefficient of that tire is below a predetermined threshold, it can be judged that the rolling resistance performance is excellent and good. The threshold can be set as appropriate, for example, according to the performance required of the target tire (rolling resistance performance, etc.).

[0111] If the performance data 27 of the target tire is determined to be good (Yes in step S9), the target tire is manufactured based on the material information 26 (including physical properties 25) of the tire components 13 of the target tire shown in Figure 6 (step S10). On the other hand, if the performance data 27 of the target tire is determined to be poor (No in step S9), at least a part of the material information 26 (including physical properties 25) of the tire components 13 of the target tire is changed (step S11). Then, steps S7 to S9 are performed again. As in this embodiment, if the rolling resistance coefficient 27A of the tire is output, in step S11, the loss tangent of the cap rubber 19 and the base rubber 20 is changed to be small so that the rolling resistance is reduced. Such changes may be made by an operator or by a computer 1 (prediction device 1C) based on known optimization methods, etc. This ensures that a tire 11 with desired performance (e.g., rolling resistance performance) can be reliably designed and manufactured.

[0112] [Method for creating a training dataset (Second embodiment)] In previous embodiments, as shown in Figure 7, the rolling resistance coefficient 27A of the tire 11, the stiffness of the tire, and the air resistance value of the tire have been exemplified as performance data 27 of the training dataset 28, but the embodiment is not limited to this form. The performance data 27 may also be distribution data of physical quantities acting on the outer surface of the tread portion 11a shown in Figure 2 (hereinafter sometimes referred to as "outer surface distribution data"). Furthermore, the performance data 27 may also be distribution data of physical quantities acting inside the tread, which is radially inward from the outer surface (hereinafter sometimes referred to as "internal distribution data"). Since these outer surface distribution data and internal distribution data are affected by the stiffness of the tire 11, etc., which are identified by the member information 26 (physical property values ​​25) of the tire component 13 shown in Figures 5(b) and 6, they have a high correlation with the second data 22 which includes the member information 26 (physical property values ​​25).

[0113] In this embodiment of the creation method, at least one of the external distribution data (not shown) and internal distribution data (not shown) is acquired in step S3 shown in Figure 3.

[0114] Figure 11 shows an example of a training dataset from another embodiment of the present invention. In Figure 11, a training dataset 28 obtained from one type of tire 11 is shown as representative. In the second data 22 of Figure 11, the pixels 23 and physical property values ​​25 (material information 26) shown in Figure 5(b) are omitted.

[0115] As shown in Figure 11, the outer surface distribution data 27B is used to identify the physical quantities acting on the outer surface of the tread portion 11a when the tread portion 11a shown in Figure 2 is in contact with the road surface (not shown). These physical quantities are not particularly limited as long as they act on the outer surface of the tread portion 11a of the tire 11 that is in contact with the road surface (not shown). In this embodiment, an example of the physical quantity is the contact pressure acting on the outer surface. However, the physical quantity is not limited to contact pressure and may be, for example, shear force, slip amount, and wear energy.

[0116] The surface distribution data 27B can be constructed as a contour map represented by color information or grayscale information corresponding to the number of grayscale levels, depending on the magnitude of the physical quantity. In such surface distribution data 27B, the physical quantity (in this example, ground pressure) can be color-coded according to its magnitude.

[0117] The outer surface distribution data 27B is preferably determined by mounting a tire 11 (shown in Figure 2) on a rim (not shown) and filling it with a predetermined internal pressure, and then bringing it into contact with the road surface (not shown) based on a predetermined camber angle and load, and driving it under predetermined driving conditions. This allows the outer surface distribution data 27B to be determined under the same conditions. Details of the rim, internal pressure, and load are as described above. The camber angle is set appropriately according to the conditions under which the tire 11 is mounted on the vehicle, and can be set to, for example, 0 degrees. Driving conditions include, for example, driving speed.

[0118] The outer surface distribution data 27B may be measured, for example, using a known wear energy measuring device (not shown) or a stress measuring machine (not shown), or it may be determined by performing a simulation using computer 1 (shown in Figure 1). In the simulation, the outer surface distribution data 27B can be calculated by bringing a tire model (not shown), which models the tire 11, into contact with a road surface model (not shown), which models the road surface (not shown), and then driving the vehicle. Such a driving simulation can be appropriately performed based on known methods (for example, the method described in Japanese Patent Application Publication No. 2023-084539).

[0119] Figure 12 shows an example of a training dataset from another embodiment of the present invention. In Figure 12, a training dataset 28 obtained from one type of tire 11 is shown as representative. In the second data 22 of Figure 11, the pixels 23 and physical property values ​​25 (material information 26) shown in Figure 5(b) are omitted.

[0120] As shown in Figure 12, the internal distribution data 27C is for identifying physical quantities acting inside the tread. In this embodiment, the internal distribution data 27C can identify physical quantities acting inside the tire, including the tread, in the meridional cross-section of the tire 11. Such physical quantities are not particularly limited as long as they act inside the tire, including the tread. The physical quantities in this embodiment include strain and / or stress acting inside the tread. While von Mises stress is used as the physical quantity in this embodiment, it is not particularly limited, and various physical quantities can be used depending on the performance to be evaluated.

[0121] The internal distribution data 27C can be constructed as a contour plot represented by color information or grayscale information corresponding to the number of gradations, depending on the magnitude of the physical quantity. In such internal distribution data 27C, the physical quantity (in this example, von Mises stress) can be color-coded according to its magnitude.

[0122] The internal distribution data 27C is preferably identified under predetermined driving conditions when the tire 11, which is mounted on a rim and filled with a predetermined internal pressure, is in contact with the road surface (not shown) at a predetermined camber angle and is subjected to a predetermined load. This allows the internal distribution data 27C to be identified under the same conditions. Details of the rim, internal pressure, camber angle, load, etc., are as described above.

[0123] The internal distribution data 27C may be identified, for example, based on the calculation results of a simulation using computer 1 (shown in Figure 1). In the simulation, the internal distribution data 27C can be calculated by bringing a tire model (not shown), which models the tire 11, into contact with a road surface model (not shown), which models the road surface (not shown). Such a contact simulation can be appropriately carried out based on known methods (for example, the method described in Japanese Patent Application Publication No. 2023-084539).

[0124] In this embodiment, in step S4 shown in Figure 3, the second data 22 is combined with at least one of the external distribution data 27B (shown in Figure 11) and the internal distribution data 27C (shown in Figure 12). This allows for the creation of a training dataset 28 in which the second data 22 and the external distribution data 27B are combined, as shown in Figure 11, or a training dataset 28 in which the second data 22 and the internal distribution data 27C are combined, as shown in Figure 12. The training dataset 28 includes the second data 22 (shown in Figures 5(b) and 7) which contains material information 26 (physical properties 25) of the tire components 13 that have a high correlation with the performance data 27 (in this example, the external distribution data 27B and the internal distribution data 27C). Such second data 22 allows for the effective extraction of features for estimating the performance data 27. Therefore, the creation method (creation device 1A) of this embodiment makes it possible to create a training dataset 28 that can generate a trained model 34B (shown in Figure 9) capable of accurately estimating performance data 27.

[0125] [Method for generating a pre-trained model (Second Embodiment)] Next, a generation method for another embodiment of the present invention will be described. In this embodiment, a trained model 34B (shown in Figure 9) is generated. This trained model 34B is machine-trained to output performance data (in this example, at least one of the outer surface distribution data 27B shown in Figure 11 and the inner distribution data 27C shown in Figure 12) when second data 22 (shown in Figure 7) is input for the target tire (tire 11 to be evaluated).

[0126] The machine learning model 34A (trained model 34B) may appropriately employ the known models described above. When the performance data 27 is configured as a contour plot, as in the external distribution data 27B (shown in Figure 11) and internal distribution data 27C (shown in Figure 12) of this embodiment, the color information or grayscale information of the physical quantities can be treated, for example, as class labels for semantic segmentation. Therefore, the machine learning model 34A (trained model 34B) may employ a model (e.g., U-NET) that can extract features from the second data 22 and classify those features into class labels.

[0127] In this embodiment, the generation method involves step S5, which involves preparing a training dataset 28 (not shown), and step S6, which involves training a machine learning model 34A (shown in Figure 9) using the training dataset 28. As shown in Figures 11 and 12, the training dataset 28 in this embodiment is a combination of second data 22 of the tire 11 and performance data 27 (in this example, at least one of the outer surface distribution data 27B and the inner distribution data 27C). As a result, when the second data 22 of the target tire is input, a trained model 34B (shown in Figure 9) capable of outputting performance data of the target tire (in this example, at least one of the outer surface distribution data 27B and the inner distribution data 27C) is generated.

[0128] As described above, the performance data 27 (in this example, at least one of the external distribution data 27B and the internal distribution data 27C) is affected by the stiffness of the tire 11, which can be identified by the material information 26 (physical property values ​​25) of the tire components 13 shown in Figures 5(b) and 6. For this reason, the performance data 27 has a high correlation with the second data 22 (shown in Figures 5(b) and 7). When such second data 22 is input into the machine learning model 34A (shown in Figure 9), features for estimating the performance data 27 (in this example, at least one of the external distribution data 27B and the internal distribution data 27C) can be effectively extracted. Therefore, the generation method (generation device 1B) of this embodiment makes it possible to generate a trained model 34B (shown in Figure 9) that can accurately estimate the performance data 27.

[0129] [Method for predicting tire performance (second embodiment)] Next, a prediction method according to another embodiment of the present invention will be described. In this prediction method, as shown in Figure 9, second data 22 of the target tire (tire 11 to be evaluated) is input to a trained model 34B, and performance data 27 of the target tire (at least one of the external distribution data 27B shown in Figure 11 and the internal distribution data 27C shown in Figure 12) is output from the trained model 34B.

[0130] In this embodiment of the prediction method, step S7 is performed to input second data 22 (shown in Figure 7) of the target tire into a trained model 34B, and step S8 is performed to output performance data 27 of the target tire from the trained model 34B. As shown in Figure 5(b), the second data 22 includes material information 26 (physical property values ​​25) that can identify the stiffness, etc., of the target tire. For this reason, the second data 22 has a high correlation with the performance data 27 of the target tire (in this example, at least one of the external distribution data 27B shown in Figure 11 and the internal distribution data 27C shown in Figure 12), which is affected by the stiffness, etc. When such second data 22 is input into the machine learning model 34A, it becomes possible to estimate the performance data 27 with high accuracy.

[0131] In this embodiment of the prediction method, step S9 is performed to evaluate the quality of the performance data 27 of the target tire (in this example, at least one of the external distribution data 27B shown in Figure 11 and the internal distribution data 27C shown in Figure 12). If the performance data 27 of the target tire is determined to be good (Yes in step S9), the target tire is manufactured based on the material information 26 (including physical properties 25) of the tire components 13 of the target tire shown in Figure 6 (step S10). On the other hand, if the performance data 27 of the target tire is determined to be poor (No in step S9), at least a part of the material information 26 (including physical properties 25) of the tire components 13 of the target tire is changed (step S11). Then, steps S7 to S9 are performed again. This ensures that a tire 11 with the desired performance can be reliably designed and manufactured.

[0132] Although particularly preferred embodiments of the present invention have been described in detail above, the present invention is not limited to the illustrated embodiments and can be implemented in various modified forms.

[0133] [Note] The present invention includes the following embodiments.

[0134] [Invention 1] A device for creating training datasets used in machine learning to estimate tire performance data, An acquisition unit acquires first data consisting of a cross-sectional image in which the tire components are represented by pixel values ​​in the meridian cross-section of the tire, Includes a replacement unit that generates second data by replacing the pixel values ​​of the cross-sectional image with member information including the physical property values ​​of the tire component corresponding to the pixels, A device for creating training datasets. [2nd Invention] The tire component includes a rubber member, The apparatus for creating a training dataset according to the present invention 1, wherein the substitution unit substitutes the value of a pixel representing the rubber member with at least one of the elastic modulus and loss tangent of the rubber member corresponding to the pixel. [Invention 3] The tire component includes a fiber member, The apparatus for creating a learning dataset according to the present invention 1 or 2, wherein the substitution unit substitutes the pixel values ​​representing the fiber member with at least one of the elastic modulus and arrangement angle of the fiber member corresponding to the pixel. [4th Invention] A performance data acquisition unit that acquires the aforementioned performance data, A device for creating a training dataset according to any one of the present invention 1 to 3, further comprising: a dataset creation unit that combines the second data and the performance data to create the training dataset. [5th ​​Invention] A method for creating a training dataset used in machine learning to estimate tire performance data, The steps include: acquiring first data consisting of a cross-sectional image in which the tire components are represented by pixel values ​​in the meridional cross-section of the tire; The process includes the step of obtaining second data by replacing the pixel values ​​of the cross-sectional image with member information including the physical property values ​​of the tire component corresponding to the pixels. How to create a training dataset. [Invention 6] A method for generating a trained model for estimating tire performance data, The steps include: preparing a training dataset in which the second data of the tire and the performance data are combined; The process includes the step of training a machine learning model using the aforementioned training dataset so that when second data for the target tire is input, performance data for the target tire is output. The second data of the training dataset and the second data of the target tire are obtained by replacing the pixel values ​​constituting the cross-sectional image of the tire in the meridional cross-section with member information including the physical property values ​​of the tire components corresponding to the pixels. Method for generating a pre-trained model. [7th Invention] A method for predicting tire performance, which is performed by one or more processors, The process involves inputting the second data of the target tire into a trained model that outputs performance data of the aforementioned tire, and The step includes outputting performance data of the target tire from the trained model, The second data input to the trained model and the second data of the target tire are obtained by replacing the pixel values ​​constituting the cross-sectional image of the tire in the meridian cross-section with material information including the physical properties of the tire components corresponding to the pixels. Methods for predicting tire performance. [Explanation of symbols]

[0135] 13 Tire components 21 Data 1 21A Cross-sectional image 22 Second Data 23 pixels 25 Physical Properties 26. Component Information

Claims

1. A device for creating training datasets used in machine learning to estimate tire performance data, An acquisition unit acquires first data consisting of a cross-sectional image in which the tire components are represented by pixel values ​​in the meridional cross-section of the tire, Includes a replacement unit that generates second data by replacing the pixel values ​​of the cross-sectional image with member information including the physical property values ​​of the tire component corresponding to the pixels, A device for creating training datasets.

2. The tire component includes a rubber member, The apparatus for creating a training dataset according to claim 1, wherein the substitution unit substitutes the value of a pixel representing the rubber member with at least one of the elastic modulus and loss tangent of the rubber member corresponding to the pixel.

3. The tire component includes a fiber member, The apparatus for creating a learning dataset according to claim 1, wherein the substitution unit substitutes the pixel values ​​representing the fiber member with at least one of the elastic modulus and arrangement angle of the fiber member corresponding to the pixel.

4. A performance data acquisition unit that acquires the aforementioned performance data, A device for creating a training dataset according to any one of claims 1 to 3, further comprising: a dataset creation unit that combines the second data and the performance data to create the training dataset.

5. A method for creating a training dataset used in machine learning to estimate tire performance data, The steps include: acquiring first data consisting of a cross-sectional image in which the tire components are represented by pixel values ​​in the meridional cross-section of the tire; The process includes the step of obtaining second data by replacing the pixel values ​​of the cross-sectional image with member information including the physical property values ​​of the tire component corresponding to the pixels. How to create a training dataset.

6. A method for generating a trained model for estimating tire performance data, The steps include: preparing a training dataset in which the second data of the tire and the performance data are combined; The process includes the step of training a machine learning model using the aforementioned training dataset so that when second data for the target tire is input, performance data for the target tire is output. The second data of the training dataset and the second data of the target tire are obtained by replacing the pixel values ​​constituting the cross-sectional image of the tire in the meridional cross-section with member information including the physical properties of the tire components corresponding to the pixels. Method for generating a pre-trained model.

7. A method for predicting tire performance, which is performed by one or more processors, The process involves inputting the second data of the target tire into a trained model that outputs performance data of the aforementioned tire, and The step includes outputting performance data of the target tire from the trained model, The second data input to the trained model and the second data of the target tire are obtained by replacing the pixel values ​​constituting the cross-sectional image of the tire in the meridian cross-section with material information including the physical properties of the tire components corresponding to the pixels. Methods for predicting tire performance.

Images