Method for creating prediction model, method for predicting physical quantity of tire, and prediction device

Abstract

To provide a method for creating a prediction model capable of predicting a physical quantity between a tire and a substance in contact with the outer surface of the tire without requiring the experience or intuition of experts.SOLUTION: There is provided a method for creating a prediction model. This method comprises: a first step of determining at least one cross-section 18 so as to include a tire and a substance when the tire is in a first travel state; a second step of inputting first data 21 for identifying a first physical quantity 13 of the tire and / or the substance in the at least one cross-section 18; a third step of inputting a second physical quantity 22 of the tire or the substance in the first travel state; a fourth step of creating a dataset 23 that links the first data 21 and the second physical quantity 22 for the plurality of tires; and a fifth step of creating a prediction model that estimates the second physical quantity 22 from the first data 21 being the prediction target with the dataset 23 as teacher data.SELECTED DRAWING: Figure 7

Description

Technical Field

[0001] The present invention relates to a method for creating a prediction model, a method for predicting physical quantities of a tire, and a prediction device.

Background Art

[0002] In recent years, various simulation methods have been proposed that can easily define the fluid around a rotating tire (see, for example, Patent Document 1 below).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Generally, in simulations using a computer, when the area to be calculated and the number of physical quantities increase, a large amount of calculation time is required. Therefore, those areas and physical quantities are limited to a certain range, and phenomena between the tire and the substance are considered from those physical quantities. However, such limitations and considerations require the experience and intuition of a skilled person, so it is not easy to predict the above physical quantities.

[0005] The present invention has been devised in view of the above actual situation, and the main object is to provide a method for creating a prediction model that can predict physical quantities between a tire and a substance in contact with the outer surface of the tire without requiring the experience and intuition of a skilled person.

Means for Solving the Problems

[0006] The present invention is a method for creating a prediction model for predicting a physical quantity between a tire and a substance in contact with the outer surface of the tire, the method including: a first step of determining at least one cross section including the tire and the substance when the tire is in a first driving state; a second step of inputting, into a computer, first data for specifying a first physical quantity of the tire and / or the substance in the at least one cross section; a third step of inputting, into the computer, a second physical quantity of the tire or the substance in the first driving state; a fourth step of the computer creating a data set associating the first data and the second physical quantity for a plurality of tires; and a fifth step of the computer creating a prediction model for estimating the second physical quantity from the first data to be predicted, using the data set as teacher data.

Advantages of the Invention

[0007] By adopting the above steps, the method for creating a prediction model of the present invention can predict the physical quantity between a tire and a substance in contact with the outer surface of the tire without requiring the experience and intuition of a skilled person.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Mode for Carrying Out the Invention

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings. It should be understood that the drawings include exaggerated expressions and expressions different from the actual structural dimensional ratios in order to assist in understanding the content of the invention. Also, throughout each embodiment, the same or common elements are denoted by the same reference numerals, and redundant explanations are omitted. Furthermore, the specific configurations shown in the embodiments and the drawings are for understanding the content of the present invention, and the present invention is not limited to the specific configurations shown.

[0010] In the method for creating a prediction model of the present embodiment (hereinafter sometimes referred to as the "creation method"), a prediction model is created using a computer. This prediction model is for predicting (estimating) a physical quantity between a tire and a substance in contact with the outer surface of the tire. The prediction model of the present embodiment is used in a method for predicting a physical quantity of a tire (hereinafter sometimes referred to as the "prediction method").

[0011] FIG. 1 is a block diagram showing an example of a computer 1 in which a processing procedure of the method for creating a prediction model and a processing procedure of the method for predicting a physical quantity of a tire are executed.

[0012] [Tire physical quantity prediction device] The computer 1 of the present embodiment is configured as a creation device 1A for a prediction model (hereinafter sometimes referred to as the "creation device") and a prediction device 1B for a physical quantity of a tire (hereinafter sometimes referred to as the "prediction device").

[0013] The computer 1 of the present embodiment is provided with an input device 2 as an input device, an output device 3 as an output device, and an arithmetic processing device 4 for calculating a physical quantity of a tire and the like.

[0014] [Input device, output device, arithmetic processing device] For the input device 2, for example, a keyboard, a mouse, or the like is used. For the output device 3, for example, a display device, a printer, or the like is used. The arithmetic processing device 4 includes an arithmetic unit (CPU) 4A that performs various arithmetic operations, a storage unit 4B that stores data, programs, and the like, and a working memory 4C.

[0015] [Storage unit] The storage unit 4B 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 4B is provided with a data unit 5 and a program unit 6.

[0016] [Data unit] The data section 5 of the present embodiment is for storing data and the like necessary for executing the creation method and prediction method. The data section 5 of the present embodiment includes a first storage section 5A, a second storage section 5B, a third storage section 5C, a fourth storage section 5D, a fifth storage section 5E, and a sixth storage section 5F. Note that the data section 5 is not necessarily limited to such a form, and other storage sections or the like may be included as necessary, or a part of these storage sections may be omitted. Also, the details of the data input to these data sections 5 will be described later.

[0017] [Program section] The program section 6 is a program (application) necessary for executing the prediction method. The program section 6 is executed by the arithmetic section 4A.

[0018] The program section 6 of the present embodiment includes an output section 6A. Further, the program section 6 of the present embodiment includes a first input section 6B, a second input section 6C, a third input section 6D, a fourth input section 6E, a data set creation section 6F, a model creation section 6G, a determination section 6H, and an evaluation section 6J. Note that the program section 6 is not necessarily limited to such a form, and other programs may be included as necessary, or a part of these may be omitted. Also, the details of the functions of these program sections 6 will be described later.

[0019] [Tire - Vehicle] FIG. 2 is a perspective view showing an example of a vehicle 8 equipped with a tire 7. In the present embodiment, the tire 7 is exemplified as a pneumatic tire for a passenger car. However, the tire 7 is not limited to such a form, and for example, it may be a pneumatic tire for heavy loads, a pneumatic tire for a two - wheeler, etc., or it may be a non - pneumatic tire. Note that the structure of the tire 7 is not particularly limited, and for example, a structure similar to the conventional one may be adopted.

[0020] The vehicle 8 is not particularly limited as long as it can mount the tire 7 via the wheel rim 9. The vehicle 8 of the present embodiment includes a body 8A. The body 8A is provided with a fender 8b that covers the tire 7, a front bumper 8f, and a rear bumper 8r.

[0021] [Substance] The substance 10 is not particularly limited as long as it contacts the outer surface 7s of the tire 7. In the present embodiment, a case where the fluid 11 is included as the substance 10 is exemplified. In this case, in the creation method of the present embodiment, a prediction model for predicting the physical quantity acting between the tire 7 and the fluid 11 is created. The physical quantity (second physical quantity described later) estimated (predicted) by this prediction model includes, for example, a physical quantity related to fluid resistance.

[0022] The fluid 11 includes, for example, air 11a or water (not shown). In the present embodiment, a case where the fluid 11 is air 11a is exemplified. In this case, the physical quantity (second physical quantity) related to fluid resistance preferably includes an air resistance coefficient.

[0023] The substance 10 is not limited to the fluid 11, and may include, for example, road surface deposits (not shown). In this case, in the creation method of the present embodiment, a prediction model for predicting the physical quantity acting between the tire 7 and the road surface deposits is created. The physical quantity (second physical quantity described later) predicted by this prediction model includes, for example, a physical quantity related to at least one of the compression rate and pressure of the road surface deposits. The road surface deposits include, for example, snow or mud.

[0024] [Creation Method of Prediction Model (First Embodiment)] Next, the creation method of the present embodiment will be described. FIG. 3 is a flowchart showing an example of the processing procedure of the creation method of the prediction model. To implement the creation method of the present embodiment, the computer 1 (prediction model creation device 1A) shown in FIG. 1 is used.

[0025] [Determine the Cross-Section When the Tire is in the First Driving State (First Step)] In the creation method of this embodiment, first, when the tire 7 shown in FIG. 2 is in the first running state, at least one cross-section including the tire 7 and the substance 10 is determined (first step S1). The cross-section determined in the first step S1 is used for specifying the teacher data (first data described later) of the prediction model.

[0026] In the first step S1 of this embodiment, first, the first input unit 6B included in the program unit 6 shown in FIG. 1 is read into the work memory 4C. The first input unit 6B is a program for determining at least one cross-section including the tire 7 and the substance 10 when the tire 7 shown in FIG. 2 is in the first running state. By executing this first input unit 6B by the arithmetic unit 4A, the computer 1 can be made to function as means for determining at least one cross-section.

[0027] The first running state is appropriately set according to, for example, a physical quantity estimated by the prediction model (second physical quantity described later). For example, when the substance 10 shown in FIG. 2 is the fluid 11 (in this example, the air 11a), the first running state is specified in the state where the tire 7 is running while receiving fluid resistance or the like from the fluid 11. On the other hand, when the substance 10 is road surface deposits (not shown), the first running state is specified in the state where the tire 7 is running while receiving friction, pressing force, or the like from the road surface deposits.

[0028] The running state of the vehicle 8 (tire 7) changes according to, for example, the running conditions of the vehicle 8 (tire 7). Therefore, in order to uniquely specify the first running state, it is preferable that the running conditions of the vehicle 8 (tire 7) are specified in advance.

[0029] Examples of the running conditions include the internal pressure condition, load condition, and running speed of the tire 7. These running conditions can be appropriately set according to, for example, the category of the tire 7, the tire size, and the displacement of the vehicle 8.

[0030] The cross-section when in the first driving state (i.e., at least one cross-section including the tire 7 and the substance 10) is determined as appropriate. For example, in the first driving state when the vehicle 8 (tire 7) is actually driven, the cross-section may be specified. However, in such actual vehicle driving, for example, since the driving state changes moment by moment according to the conditions of the road surface 12 and the like, it is difficult to uniquely determine the cross-section when in the first driving state. Furthermore, actual vehicle driving requires the prototyping of the tire 7 and the preparation of the vehicle 8 and the like, and a lot of time and cost are required for the creation of the prediction model and the prediction of physical quantities.

[0031] In this embodiment, the cross-section when in the first driving state is determined by a simulation using the computer 1 (shown in FIG. 1). In such a simulation, since an arbitrary driving state can be calculated, it is possible to uniquely specify the first driving state. Furthermore, from the calculation results of the simulation, the tire and the substance in the first driving state can be easily specified (extracted). Therefore, the cross-section when in the first driving state can be uniquely determined. Also, in the simulation, the prototyping of the tire 7 and the preparation of the vehicle and the like are not required. Therefore, it is possible to implement the creation of the prediction model and the prediction of physical quantities in a short time and at low cost.

[0032] For the simulation to specify the cross-section when in the first driving state, for example, a tire model, a vehicle model, a road surface model, and a substance model are used. FIG. 4 is a conceptual diagram showing an example of the tire model 14, the vehicle model 15, the road surface model 16, and the substance model 17. In FIG. 4, the element G(i) of the substance model 17 is partially shown.

[0033] In the first step S1 of this embodiment, four tire models 14 (including two front tire models 14f, 14f and two rear tire models 14r, 14r) are defined. These tire models 14 are models of one type of tire 7 (shown in FIG. 1) among a plurality of types of tires 7 (not shown). Here, one type of tire means a tire in which the tread pattern, structure, etc. are set to be the same. Note that "the same" is assumed to allow variations such as manufacturing errors. Also, the vehicle model 15 is a model of the vehicle 8 shown in FIG. 2, and the road surface model 16 is a model of the road surface 12 shown in FIG. 2.

[0034] The tire model 14, the vehicle model 15, and the road surface model 16 can be defined, for example, by discretizing (dividing) the tire 7, the vehicle 8, and the road surface 12 shown in FIG. 2 using a finite number of elements (not shown) that can be handled by a numerical analysis method.

[0035] As the numerical analysis method, for example, the finite element method, the finite volume method, the difference method, or the boundary element method can be appropriately adopted. In this embodiment, the finite element method is adopted. Also, numerical data such as element numbers, node (not shown) numbers, node coordinate values, and material properties (such as density, Young's modulus, and / or attenuation coefficient, etc.) are defined for the elements (not shown).

[0036] The substance model 17 of this embodiment is a model of the substance 10 (in this example, air 11a) shown in FIG. 2. Such a substance model 17 can be defined, for example, by discretizing (dividing) a certain region (not shown) surrounding the tire 7, the vehicle 8, and the road surface 12 shown in FIG. 2 using a finite number of elements (not shown). The substance model 17 of this embodiment is defined as a rectangular parallelepiped, but is not limited to such a form.

[0037] For the elements G(i) (i = 1, 2, …) of the material model 17, for example, Euler meshes (Euler elements) are used. Physical quantities such as the flow velocity and pressure of the fluid 11 (in this example, air 11a) shown in FIG. 2 are assigned to these elements G(i). For the discretization of the elements of the material model 17, for example, the finite volume method is used. In the material model 17 of the present embodiment, a front wall surface 17f is formed in front (in the traveling direction) of the vehicle model 15.

[0038] The simulation using the above model can be appropriately performed based on a known procedure (for example, the procedure described in Japanese Patent Application Laid-Open No. 2020-185913 etc.) and the driving conditions of the first driving state (for example, internal pressure conditions, load conditions, driving speed, etc.).

[0039] In the present embodiment, first, based on the internal pressure condition (the first driving condition), a tire model 14 filled with internal pressure is calculated. Next, the tire model 14 after filling with internal pressure is mounted on the vehicle model 15. Next, based on the load condition (the first driving condition), the contact between the vehicle model 15 (tire model 14) and the road surface model 16 is calculated. Next, based on the driving speed V (the first driving condition), the vehicle model 15 traveling on the road surface model 16 is calculated.

[0040] For the deformation calculation of the tire model 14, the driving calculation of the vehicle model 15, etc., for example, commercially available application software for finite element analysis such as Abaqus manufactured by Dassault Systems, LS-DYNA manufactured by LSTC, or NASTRAN manufactured by MSC is used.

[0041] Next, an inflow F of a fluid (in this example, air) having a speed approximating the traveling speed V (first traveling condition) can be defined on the front wall surface 17f of the substance model 17. Thereby, a simulation of bringing the substance 10 (air 11a) into contact with the outer surface 7s of the tire 7 shown in FIG. 2 can be performed. In the simulation of the present embodiment, a state where the substance model 17 is in contact with the outer surface of the tire model 14 (vehicle model 15) traveling on the road surface model 16 can be calculated for each unit time (micro time) of the simulation. For such a simulation using the substance model 17, for example, commercially available application software for fluid analysis such as STAR-CD manufactured by CD-adapco or FLUNET of ANSYS is used.

[0042] In each element G(i) constituting the substance model 17, physical quantities and the like acting due to contact with the vehicle model 15, the tire model 14, etc. are calculated. When air is modeled as in the substance model 17 of the present embodiment, in each element G(i) of the substance model 17, physical quantities related to air (including, for example, the total pressure coefficient of air, pressure, shear stress, etc.) are calculated respectively.

[0043] Furthermore, in each element (not shown) constituting the tire model 14 and the vehicle model 15, physical quantities acting due to contact with the substance model 17 may be calculated. The physical quantities of the tire model 14 and the vehicle model 15 include, for example, shear stress, pressure, and the like.

[0044] In the above simulation, the traveling state of the tire model 14 (vehicle model 15) gradually approaches the first traveling state after the start of the traveling calculation. Therefore, in the present embodiment, among the plurality of unit times in which the above simulation is performed, the unit time when the traveling state of the tire model 14 becomes the first traveling state is specified. Then, at the specified unit time (first traveling state), at least one cross section 18 is determined to include the tire model 14 and the substance model 17 (tire and substance).

[0045] In addition, when the substance model 17 models water, snow, or mud, for example, modeling and simulation are performed based on known procedures (such as Japanese Patent Application Laid-Open No. 2022-040965). Thereby, the first running state of the tire in contact with water, snow, or mud is specified, and at least one cross-section (not shown) in the first running state can be determined.

[0046] The cross-section 18 in the first running state may be automatically determined, for example, by the first input unit 6B shown in FIG. 1, or may be determined by an operator or the like. When determined by an operator or the like, the cross-section 18 can be determined based on the tire model 14 and the substance model 17 (shown in FIG. 4) of the first running state output to the output device (for example, a display device) 3. Further, the position of the cross-section 18 can be specified, for example, by orthogonal coordinates of the x-axis, y-axis, and z-axis.

[0047] The cross-section 18 can be appropriately determined as long as it includes the tire model 14 and the substance model 17 (that is, the tire 7 and the substance 10 shown in FIG. 2). As described above, the cross-section 18 is used to specify the teacher data (first data) of the prediction model. This teacher data (first data) is for specifying the following first physical quantity (in this example, the total air pressure coefficient) of the tire 7 and / or the substance 10 shown in FIG. 2. Such a first physical quantity may be affected not only by the tire 7 and the substance 10 (air 11a), but also by the shape and structure of the vehicle 8 (including the fender 8b) on which the tire 7 is mounted. In order to consider such an influence, as shown in FIG. 4, a cross-section including the vehicle model 15 may be specified together with the tire model 14 and the substance model 17.

[0048] In cross-section 18 of the present embodiment, in the plan view of the vehicle model 15 (in the present embodiment, the bottom view), the cross-section 18 is determined so as to include the region from the front end (front bumper 15f) to the rear end (rear bumper 15r) of the vehicle model 15. By determining such a cross-section 18, teaching data can be created that can take into account the influence on the first physical quantity described later for the entire range in the front-rear direction of the vehicle model 15. Therefore, the prediction accuracy of the physical quantity (the second physical quantity described later) by the prediction model is improved. Further, in the present embodiment, a planar cross-section 18 is determined, but it is not necessarily limited to such a mode. For example, an arc-shaped (curved surface-shaped) cross-section (not shown) or the like may be used.

[0049] The tire models 14f, 14r mounted on one side (in this example, the left side) in the left-right direction of the vehicle model 15 and the tire models 14f, 14r mounted on the other side (in this example, the right side) in the left-right direction are mounted symmetrically left and right. When it is considered that there is no substantial difference in the first physical quantity (in this example, the total pressure coefficient of air) between the tire models 14f, 14r on one side and the tire models 14f, 14r on the other side, the cross-section 18 may be determined so as to include only the tire models 14f, 14r on one side. Thereby, since the region of the cross-section 18 to be learned becomes smaller, an increase in the number of teaching data used for creating the prediction model can be suppressed. Note that the cross-section 18 is not necessarily limited to such a mode. For example, for further improvement in prediction accuracy, the cross-section 18 may be determined so as to include all (four) tire models 14 on the left and right.

[0050] In the present embodiment, a plurality of cross-sections 18 having different heights from the running surface 16s of the tire (road surface model 16) are determined. By determining such a plurality of cross-sections 18, the first physical quantity (in this example, the total pressure coefficient of air) can be specified at a plurality of positions having different heights from the running surface 16s. As a result, since the first physical quantity can be grasped three-dimensionally, by using it for the teaching data of the prediction model, it becomes possible to create a prediction model with improved prediction accuracy of the second physical quantity (in this example, the air resistance coefficient) described later.

[0051] In this embodiment, three cross-sections 18 are determined as a plurality of cross-sections 18. Note that it is not limited to such three cross-sections, and for example, four or more cross-sections (not shown) may be determined.

[0052] The three cross-sections 18 include a first cross-section 18a, a second cross-section 18b, and a third cross-section 18c. In the height direction (vertical direction) from the tread surface 16s, the interval C1 between the first cross-section 18a and the second cross-section 18b and the interval C2 between the second cross-section 18b and the third cross-section 18c are set to be the same, but it is not limited to such a mode. For example, the intervals C1 and C2 may be different from each other. The intervals C1 and C2 can be set as appropriate, and for example, they are set to 60 to 160 mm.

[0053] The first cross-section 18a of this embodiment is set near the rotation axis of the tire model 14. The height H1 from the tread surface 16s of the third cross-section 18c of this embodiment is set to 60 to 160 mm, for example. The second cross-section 18b of this embodiment is set between the first cross-section 18a and the third cross-section 18c. By these first cross-section 18a to third cross-section 18c, in the region from the tread surface 16s to the rotation axis, the first physical quantity described later can be specified three-dimensionally. In such a region, it includes many portions where the tire 7 shown in FIG. 2 is not covered by the fender 8b. Therefore, it is considered that the first physical quantity in this region has a greater influence on the second physical quantity described later than the first physical quantity in other regions (for example, the portion covered by the fender 8b).

[0054] The determined cross-sections 18 (in this example, the first cross-section 18a to the third cross-section 18c) are stored in the first storage unit 5A (computer 1) shown in FIG. 1. Further, the calculation results of the simulation are stored in the first storage unit 5A. The calculation results include physical quantities (in this example, including the total pressure coefficient of air, pressure, shear stress, etc.) calculated by each element G(i) constituting the substance model 17 shown in FIG. 4 in the first driving state. Further, the calculation results include physical quantities (shear stress and pressure) calculated by each element (not shown) constituting the tire model 14 in the first driving state.

[0055] [Input the first data for specifying the first physical quantity of the tire and / or the substance (second step)] Next, in the creation method of the present embodiment, first data for specifying the first physical quantity of the tire and / or the substance in at least one cross-section 18 shown in FIG. 4 is input to the computer 1 (shown in FIG. 1) (second step S2).

[0056] The first physical quantity of the present embodiment is preferably related to the physical quantity (second physical quantity described later) estimated by the prediction model. When the air resistance coefficient (second physical quantity) is estimated by the prediction model as in the present embodiment, the total pressure coefficient of air is specified as the first physical quantity. This total pressure coefficient is calculated by each element G(i) of the substance model 17 shown in FIG. 4 in the above-described simulation. Therefore, in the second step S2 of the present embodiment, the first data for specifying the first physical quantity is acquired from the calculation results of the simulation.

[0057] In the second step S2 of the present embodiment, first, the cross-sections 18 (in this example, the first cross-section 18a to the third cross-section 18c shown in FIG. 4) input to the first storage unit 5A shown in FIG. 1 and the calculation results of the simulation are read into the working memory 4C. The calculation results include physical quantities (in this example, the total pressure coefficient of air, etc.) calculated by each element (not shown) constituting the substance model 17 in the first driving state.

[0058] Furthermore, in the second step S2 of the present embodiment, the second input unit 6C included in the program unit 6 shown in FIG. 1 is read into the working memory 4C. The second input unit 6C is a program for inputting first data for specifying the first physical quantity of the tire and / or the substance in the determined cross-section 18 shown in FIG. 4. By executing this second input unit 6C by the arithmetic unit 4A, the computer 1 can function as a means for inputting the first data.

[0059] In the second step S2 of the present embodiment, first, among the plurality of elements G(i) constituting the substance model 17 shown in FIG. 4, the elements G(i) arranged in the determined cross-section 18 (in this example, the first cross-section 18a to the third cross-section 18c) are specified. Then, for the element G(i) specified in the cross-section 18, the physical quantity (in this example, the total pressure coefficient of air) calculated when in the first running state is specified as the first physical quantity.

[0060] Next, in the second step S2 of the present embodiment, the first physical quantity of each element G(i) specified in the cross-section 18 (in this example, the first cross-section 18a to the third cross-section 18c) is acquired as a scalar value, and the scalar value is converted into a value of 256 gradations of grayscale. Thereby, the first physical quantity can be specified as grayscale data (image data). The conversion from the first physical quantity (scalar value) to the value of 256 gradations can be appropriately performed, for example, based on the threshold values obtained by dividing the range between the minimum value and the maximum value of the first physical quantity for all the elements G(i) of the substance model 17 into 256 equal parts.

[0061] Next, in the second step S2 of the present embodiment, in the cross-section (in this example, the first cross-section 18a to the third cross-section 18c), the tire model 14 and the vehicle model 15 can be specified as grayscale data (image data). For specifying such image data, for example, the above application software, known image editing software, etc. are used.

[0062] Next, in the second step S2 of the present embodiment, in the first cross section 18a to the third cross section 18c, the grayscale data of the first physical quantity and the grayscale data of the tire model 14 and the vehicle model 15 are respectively superimposed. As a result, first data including three grayscale data (image data) capable of specifying the first physical quantity, the tire model 14, and the vehicle model 15 is acquired.

[0063] FIG. 5 is a diagram showing an example of the first data 21. The first data 21 of the present embodiment includes three image data 21a to 21c (grayscale data 20) specified by the determined first cross section 18a to the third cross section 18c. In FIG. 5, a cross-sectional view in bottom view is illustrated, but it is not necessarily limited to such a mode, and for example, a plan view may be used.

[0064] In the three image data 21a to 21c, in the first cross section 18a to the third cross section 18c, the first physical quantity 13 (in this example, the total air pressure coefficient) of the substance model 17 acting by contacting the outer surface 14s of the tire model 14 (including the vehicle model 15) is specified. This first physical quantity 13 is specified (expressed) by the shade of the grayscale, and the darker the grayscale color, the smaller the physical quantity.

[0065] The image data 21a to 21c of the cross section 18 included in the first data 21 is input as teacher data of the prediction model in the fifth step S6 described later. When a size (for example, the number of pixels) that can be input is specified in advance in this prediction model, it is preferable that the image data of the first data 21 is resized to match the specified size.

[0066] FIG. 6 is a diagram showing an example of first data 21 including resized image data 21a to 21c. In the present embodiment, the size of each of the image data 21a to 21c shown in FIG. 5 in the x-axis direction is enlarged so as to match the size that can be input to the prediction model. Further, the size of the image data 21a to 21c in the y-axis direction is reduced. As a result, as shown in FIG. 6, the first data 21 including the resized image data 21a to 21c can be acquired. For such resizing, known image editing software or the like is used. The first data 21 is stored in the second storage unit 5B (computer 1) shown in FIG. 1.

[0067] [Input of the second physical quantity (third step)] Next, in the creation method of the present embodiment, the second physical quantity of the tire or substance in the first driving state is input to the computer 1 (shown in FIG. 1) (third step S3). As described above, the second physical quantity of the present embodiment includes the air resistance coefficient (air resistance value) of the tire 7 (shown in FIG. 2).

[0068] The air resistance coefficient can be appropriately acquired. The air resistance coefficient can be calculated, for example, by integrating the value obtained by adding the difference in the pressure of the air 11a before and after the tire 7 and the shear stress of the air 11a on the outer surface 7s of the tire 7 by the surface area of the tire 7 (the area of the outer surface 7s). The air pressure and shear stress required for such calculation are calculated by each element G(i) of the substance model 17 shown in FIG. 4 in the above-described simulation. Therefore, in the third step S3 of the present embodiment, the air resistance coefficient can be acquired using the calculation result of the simulation.

[0069] In the third step S3 of the present embodiment, first, the calculation result of the simulation input to the first storage unit 5A shown in FIG. 1 is read into the working memory 4C. This calculation result includes physical quantities (in this example, air pressure, shear stress, etc.) calculated by each element G(i) constituting the substance model 17 shown in FIG. 4 in the first driving state.

[0070] Furthermore, in the third step S3 of the present embodiment, the third input unit 6D included in the program unit 6 is read into the working memory 4C. The third input unit 6D is a program for inputting the second physical quantity of the tire 7 or the substance in the first running state. By executing this third input unit 6D by the arithmetic unit 4A, the computer 1 can be made to function as means for inputting the second physical quantity.

[0071] In the third step S3 of the present embodiment, the air resistance coefficient is calculated using the pressure and shear stress of the air in the first running state. Thereby, the air resistance coefficient (second physical quantity) of the tire in the first running state can be acquired.

[0072] In the present embodiment, it is preferable to calculate the air resistance coefficient for the tire models 14 (in this example, tire models 14f, 14r) included in the specified cross section 18 (in this example, the first cross section 18a to the third cross section 18c) of the first data 21 shown in FIGS. 5 and 6. When a plurality of tire models 14f, 14r are included in the cross section 18 as in the present embodiment, the air resistance coefficient of each may be calculated, or the average value of these air resistance coefficients may be calculated.

[0073] In the present embodiment, the air resistance coefficient is calculated for each of the tire models 14f, 14r mounted on one side (in this example, the left side) in the left - right direction of the vehicle model 15 included in the cross section 18. Then, the average value of the air resistance coefficients of these tire models 14f, 14r is specified as the second physical quantity of the tire 7 shown in FIG. 2. The second physical quantity is stored in the fourth storage unit 5D (computer 1) shown in FIG. 1.

[0074] [Create a data set associating the first data and the second physical quantity] Next, in the creation method of the present embodiment, the computer 1 (shown in FIG. 1) creates a data set associating the first data 21 (shown in FIG. 6) and the second physical quantity (not shown) (fourth step S4). The data set created in the fourth step S4 is used as teacher data for the prediction model.

[0075] In the fourth step S4 of the present embodiment, first, the first data 21 input to the second storage unit 5B shown in FIG. 1 and the second physical quantity input to the fourth storage unit 5D are read into the working memory 4C. Further, the dataset creation unit 6F included in the program unit 6 is read into the working memory 4C. The dataset creation unit 6F is a program for creating a dataset that associates the first data 21 (shown in FIG. 6) and the second physical quantity (not shown). By this dataset creation unit 6F being executed by the arithmetic unit 4A, the computer 1 can be made to function as means for creating a dataset.

[0076] FIG. 7 is a conceptual diagram showing an example of the dataset 23. In the present embodiment, the first data 21 and the second physical quantity 22 are acquired based on the calculation results of the above-described simulation for one type of tire 7 (shown in FIG. 2). By creating a dataset 23 in which such first data 21 and second physical quantity 22 are associated (linked) as teacher data, it is useful for learning a prediction model capable of estimating the second physical quantity from the first data to be predicted. The dataset 23 is stored in the fifth storage unit 5E (computer 1) shown in FIG. 1.

[0077] Determine whether datasets for multiple tires have been created Next, in the creation method of the present embodiment, it is determined whether datasets 23 for a plurality of tires (a plurality of types of tires) 7 (not shown) have been created (step S5). In the present embodiment, the determination as to whether datasets 23 for a plurality of tires 7 (shown in FIG. 7) have been created is performed by the computer 1 shown in FIG. 1, but is not necessarily limited to such a mode. For example, an operator or the like may determine whether datasets 23 for a plurality of tires 7 have been created.

[0078] In step S5 of the present embodiment, first, the determination unit 6H included in the program unit 6 shown in FIG. 1 is read into the working memory 4C. The determination unit 6H is a program for determining whether a data set 23 of a plurality of tires 7 (not shown) has been created. By this determination unit 6H being executed by the arithmetic unit 4A, the computer 1 can function as means for determining whether a data set 23 of a plurality of tires 7 has been created.

[0079] The number (number of types) of the plurality of tires 7 for which the data set 23 is created can be appropriately set according to the number of teacher data required for creating the prediction model. For example, when a plurality of teacher data are created for one tire 7, the number of tires 7 may be set to be small (for example, 1 to 8).

[0080] When it is determined that the data sets 23 of the plurality of tires 7 have been created (``Yes'' in step S5), a data set (teacher data) 23 for which a prediction model can be created has already been created. In this case, the next fifth step S6 is performed.

[0081] On the other hand, when it is determined that the data sets 23 of the plurality of tires 7 have not been created (``No'' in step S5), the first step S1 to step S5 are performed again. In the first step S1 to step S5 performed again, a data set 23 in which the first data 21 and the second physical quantity 22 shown in FIG. 7 are associated with each other is created for tires 7 of different types (different tread patterns, structures, etc.) from the tires 7 for which the data set 23 has already been created. Thereby, in the fourth step S4, data sets 23 of a plurality of tires 7 (that is, a plurality of types of tires 7) can be created.

[0082] [Create a prediction model using the data set as teacher data (fifth step)] Next, in the creation method of the present embodiment, the computer 1 (shown in FIG. 1) creates a prediction model using the data set 23 as teacher data (fifth step S6). The prediction model of the present embodiment is for estimating the second physical quantity 22 from the first data 21 of the prediction target. Note that the first data 21 of the prediction target is for specifying the first physical quantity (total air pressure coefficient) 13 of the tire and / or substance of the prediction target. The second physical quantity 22 estimated from the first data 21 of the prediction target is the second physical quantity (in this example, air resistance coefficient) of the tire or substance of the prediction target.

[0083] In the fifth step S6 of the present embodiment, first, the data set 23 of a plurality of tires 7 input to the fifth storage unit 5E shown in FIG. 1 (in FIG. 7, the data set 23 of one type of tire 7 is representatively shown) is read into the working memory 4C. Further, the model creation unit 6G included in the program unit 6 is read into the working memory 4C. The model creation unit 6G is a program for creating a prediction model using the data set 23 as teacher data. By the model creation unit 6G being executed by the arithmetic unit 4A, the computer 1 can be made to function as means for creating a prediction model.

[0084] The prediction model is not particularly limited as long as it can be created so as to be able to estimate the second physical quantity 22 from the first data 21 of the prediction target by using the data set 23 shown in FIG. 7 as teacher data. Such a prediction model is preferably configured using AI.

[0085] The prediction model of the present embodiment is constructed as a machine learning model that is learned so as to be able to estimate the second physical quantity 22 of the tire or substance of the prediction target from the first data 21 for specifying the first physical quantity 13 of the tire and / or substance of the prediction target. As such a machine learning model, a deep learning model equipped with a neural network is desirable. Thereby, the prediction model itself can extract the feature amount of the first data 21 and estimate the second physical quantity 22.

[0086] The prediction model of this embodiment preferably includes a convolutional neural network (CNN). As the convolutional neural network, for example, a known one can be adopted.

[0087] The convolutional neural network of this embodiment is capable of performing convolution operations on each of the R channel, G channel, and B channel of the RGB color model. Examples of such neural networks capable of performing convolution operations include VGG16 and VGG19. The prediction model (convolutional neural network) of this embodiment is created based on VGG16.

[0088] FIG. 8 is a conceptual diagram of VGG16. VGG16 is a convolutional neural network composed of 16 layers. This VGG16 is provided with an input layer 25, an output layer 26, a convolutional layer 27, a pooling layer 28, and a fully connected layer 29.

[0089] The input layer 25 is capable of inputting color image data (RGB color model) 32. In such an input layer 25, one piece of color image data 32 can be decomposed into an R channel, a G channel, and a B channel. The output layer 26 outputs determination results (probabilities) based on the color image data 32 for each of a predetermined number of classes (for example, 1000). In this case, the output layer 26 is composed of, for example, 1000 nodes (not shown).

[0090] In the convolutional layer 27, filter processing is performed on each of the R channel, G channel, and B channel into which the color image data 32 is decomposed. Thereby, a feature map is generated. The convolutional layer 27 is defined in each of a plurality of convolutional blocks 30.

[0091] The plurality of convolutional blocks 30 are composed of a first convolutional block 30A, a second convolutional block 30B, a third convolutional block 30C, a fourth convolutional block 30D, and a fifth convolutional block 30E. Two convolutional layers 27 are respectively defined in the first convolutional block 30A and the second convolutional block 30B. Three convolutional layers 27 are defined in the third convolutional block 30C, the fourth convolutional block 30D, and the fifth convolutional block 30E.

[0092] In the pooling layer 28, the size of the feature map generated by the convolutional layer 27 is reduced. The pooling layer 28 includes a first pooling layer 28A, a second pooling layer 28B, a third pooling layer 28C, a fourth pooling layer 28D, and a fifth pooling layer 28E.

[0093] The first pooling layer 28A is defined between the first convolutional block 30A and the second convolutional block 30B. The second pooling layer 28B is defined between the second convolutional block 30B and the third convolutional block 30C. The third pooling layer 28C is defined between the third convolutional block 30C and the fourth convolutional block 30D. The fourth pooling layer 28D is defined between the fourth convolutional block 30D and the fifth convolutional block 30E. The fifth pooling layer 28E is defined between the fifth convolutional block 30E and the fully connected layer 29.

[0094] The fully connected layer 29 combines the feature maps and outputs the determination result (probability) to the output layer 26. The fully connected layer 29 of the present embodiment includes a first fully connected layer 29A, a second fully connected layer 29B, and a third fully connected layer 29C. Among these first fully connected layer 29A to third fully connected layer 29C, the third fully connected layer 29C is configured as the output layer 26.

[0095] As described above, in the output layer 26 of VGG16, determination results (probabilities) for predetermined classes (e.g., 1000 classes) are output respectively. That is, numerical values (determination results) corresponding to the number of classes can be output. On the other hand, in the prediction model of this embodiment, it is sufficient that one second physical quantity (numerical value) can be output. Therefore, in VGG16, since the number of output numerical values is more than necessary, it is preferable to change the number of nodes in the output layer 26 (the third fully connected layer 29C) to be smaller (in this example, from 1000 to 1).

[0096] Also, when the number of output numerical values (the number of nodes) is changed to be smaller, adjustment of a plurality of convolutional blocks 30 (convolutional layers 27) and pooling layers 28 is required. In this embodiment, among the plurality of convolutional blocks 30, for example, the third convolutional block 30C, the fourth convolutional block 30D, and the fifth convolutional block 30E are deleted. Further, among the plurality of pooling layers 28, for example, the third pooling layer 28C, the fourth pooling layer 28D, and the fifth pooling layer 28E are omitted.

[0097] In this embodiment, a second pooling layer 28B is defined between the second convolutional block 30B and the first fully connected layer 29A. Since the number of data is different between the second pooling layer 28B and the first fully connected layer 29A in the VGG16 shown in FIG. 8, it is preferable to adjust the first fully connected layer 29A based on the number of data in the second pooling layer 28B.

[0098] FIG. 9 is a conceptual diagram showing an example of the prediction model 31 (convolutional neural network 24). The convolutional neural network 24 (prediction model 31) of this embodiment is provided with an input layer 25, an output layer 26, a convolutional layer 27, a pooling layer 28, and a fully connected layer 29, similar to the VGG16 shown in FIG. 8.

[0099] In the convolutional neural network 24 of this embodiment, unlike the VGG16 shown in FIG. 8, the convolutional block 30 in which the convolutional layer 27 is defined is limited to the first convolutional block 30A and the second convolutional block 30B. Also, in the convolutional neural network 24 of this embodiment, unlike the VGG16 shown in FIG. 8, the pooling layer 28 is limited to the first pooling layer 28A and the second pooling layer 28B. Further, the first fully-connected layer 29A is adjusted based on the number of data in the second pooling layer 28B.

[0100] In the convolutional neural network 24 of this embodiment, the number of nodes in the output layer 26 (the third fully-connected layer 29C) is set to 1. In such a convolutional neural network 24, by inputting one piece of color image data 32 into the input layer 25, one numerical value can be output.

[0101] In the fifth step S6 of this embodiment, a data set 23 of a plurality of tires (in FIG. 7, the data set 23 of one type of tire 7 is shown as a representative) is input as teacher data into the convolutional neural network 24. Thereby, the learning of the prediction model 31 is performed.

[0102] In this embodiment, first, one data set 23 of the tire shown in FIG. 7 is specified from among the data sets (teacher data) 23 of a plurality of tires. The three grayscale data 20 (image data 21a to 21c) included in the specified data set 23 are respectively assigned to the R channel, G channel, and B channel of the RGB color model. Thereby, one piece of color image data 32 (shown in FIG. 9) is pseudo-created. Next, the pseudo-created one piece of color image data 32 is input as the first data 21 into the input layer 25 of the convolutional neural network 24. Thereby, the second physical quantity 22 estimated by the convolutional neural network 24 is output from the output layer 26.

[0103] Next, an error between the estimated second physical quantity 22 (shown in FIG. 9) and the second physical quantity 22 (shown in FIG. 7) included in the data set (teacher data) 23 of the one tire is calculated. Then, various parameters (weight coefficients, biases, etc.) of the convolutional neural network 24 are updated so that this error is minimized.

[0104] By performing these series of steps (that is, steps from the pseudo creation of one piece of color image data 32 to the update of the parameters of the convolutional neural network 24) for the data sets 23 of a plurality of tires, the parameters can be optimized. Thereby, a prediction model 31 capable of estimating the second physical quantity 22 (shown in FIG. 7) of the tire to be predicted can be created by inputting the first data 21 (shown in FIG. 6) of the tire to be predicted. In such a prediction model 31, it is possible to predict the physical quantity between the tire and the substance without the need for the experience and intuition of an expert. The prediction model 31 is stored in the third storage unit 5C shown in FIG. 1.

[0105] In the creation method of the present embodiment, since the first data 21 (shown in FIG. 7) in a plurality of cross sections 18 (first cross section 18a to third cross section 18c) having different heights from the tread surface 16s of the tire shown in FIG. 4 is used, the first physical quantity 13 can be specified three-dimensionally. By inputting the first data 21 in which such a first physical quantity 13 is specified into the prediction model 31 (convolutional neural network 24) shown in FIG. 9, feature amounts for estimating the second physical quantity 22 can be effectively extracted. Therefore, the second physical quantity 22 can be accurately estimated.

[0106] Also, in the fifth step S6 of the present embodiment, three grayscale data (grayscale data specified by the first cross-section 18a to the third cross-section 18c) 20 shown in FIG. 7 are respectively assigned to the R channel, the G channel, and the B channel. Thereby, one piece of color image data 32 (shown in FIG. 9) is pseudo-created. By using such pseudo-created color image data 32, a prediction model 31 can be easily created based on the conventional VGG16 shown in FIG. 8. Thereby, the prediction model 31 (convolutional neural network 24) can be created in a short time.

[0107] Furthermore, in the present embodiment, regarding the data set 23 (shown in FIG. 7) used as teacher data, the first physical quantity (in this example, the total pressure coefficient of air) 13 specified by the first data 21 is related to the second physical quantity (in this example, the air resistance coefficient) 22 shown in FIG. 9. By inputting such a first physical quantity 13 into the prediction model 31 (convolutional neural network 24) shown in FIG. 9, features for estimating the second physical quantity 22 can be effectively extracted.

[0108] The prediction model 31 may include a function of visualizing features that contributed to the prediction of the second physical quantity 22 in the first data 21 of the tire to be predicted. In the present embodiment, in one piece of pseudo-created color image data 32 (shown in FIG. 9), a portion that contributed to the prediction of the second physical quantity 22 is output as a heatmap. Visualization of such features can be realized, for example, by implementing a known Grad-CAM in the prediction model 31.

[0109] FIG. 10 is a diagram showing an example of image data 33 in which features that contributed to the prediction of the second physical quantity 22 are visualized. FIG. 10(a) shows an example of image data 33 in which features are visualized based on the color image data 32 shown in FIG. 9. In FIG. 10, it is shown that the higher the brightness, the more it contributes to the prediction of the second physical quantity 22.

[0110] In the image data 33 shown in FIG. 10(a), the brightness is high on the side and behind the front tire model 14f, on the outside of the body of the vehicle model 15, and behind the rear tire model 14r. These parts are focused on the prediction model 31 (shown in FIG. 9) and contribute to the prediction of the second physical quantity 22. With such image data 33, the parts that are actually predicted to have a large impact on the second physical quantity 22 can be easily and visually grasped.

[0111] The size of the image data 33 in FIG. 10(a) is adjusted to the specified size of the prediction model 31 (shown in FIG. 9), similar to the image data 21a to 21c shown in FIG. 6. Therefore, since it does not match the size of the cross section 18 (the first cross section 18a to the third cross section 18c) shown in FIG. 4, it does not match the scale of the actual tire 7 and vehicle 8 shown in FIG. 2. Thus, in order to match the scale of the actual tire 7 and vehicle 8, the image data 33 may be resized to match the size of the cross section 18 shown in FIG. 4.

[0112] FIG. 10(b) shows an example of the resized image data 33. In the present embodiment, the size of the image data 33 shown in FIG. 10(a) in the x-axis direction is reduced and the size in the y-axis direction is enlarged so as to match the size of the cross section 18 shown in FIG. 4. For such resizing, known image editing software or the like is used. Thus, by resizing the image data 33, it is possible to match the scale of the actual tire 7 and vehicle 8. Therefore, the parts that have a large impact on the second physical quantity 22 can be grasped more accurately.

[0113] [Method for Creating Prediction Model (Second Embodiment)] In the previous embodiments, the first physical quantity 13 of the first cross-section 18a to the third cross-section 18c shown in FIG. 4 was acquired as a scalar value, and thus the first data 21 including the three grayscale data 20 shown in FIG. 7 was specified. However, the present invention is not limited to such a mode. For example, the first physical quantity (not shown) of one cross-section 18 (for example, the first cross-section 18a) may be acquired as a three-dimensional vector value, and the first data 21 (data set 23) including three grayscale data (not shown) composed of the vector values of each dimension may be specified.

[0114] By acquiring the first physical quantity as a three-dimensional vector value, for example, the air flow can be three-dimensionally specified. Three grayscale data converted from each of such three-dimensional vector values are used as the input of the prediction model 31 (convolutional neural network 24) shown in FIG. 9, so that the feature amount for estimating the second physical quantity 22 can be effectively extracted. Therefore, in this embodiment, similar to the previous embodiments, the second physical quantity 22 can be accurately estimated.

[0115] [Method for creating prediction model (third embodiment)] In the previous embodiments, the first physical quantity 13 of each element G(i) in the cross-section 18 (the first cross-section 18a to the third cross-section 18c) shown in FIG. 4 was converted into the image data 21a to 21c (grayscale data 20) shown in FIG. 7. However, the present invention is not limited to such a mode. Due to the limitation of the number of gradations, the first physical quantity 13 may not be accurately specified in the image data 21a to 21c. Therefore, for example, the conversion into the image data 21a to 21c may be omitted, and the first physical quantity (not shown) of each element G(i) in the cross-section 18 shown in FIG. 4 may be used as it is. In this case, in this embodiment, for each cross-section 18 (the first cross-section 18a to the third cross-section 18c), a two-dimensional matrix (not shown) in which the first physical quantities of each element G(i) are arranged may be specified.

[0116] In the two-dimensional matrix (not shown) of this embodiment, there is no limitation on the number of gradations. From such two-dimensional matrices of each cross section 18 (the first cross section 18a to the third cross section 18c), the color image data 32 shown in FIG. 9 is pseudo-created and used as the input to the prediction model 31, so that the feature amount for estimating the second physical quantity 22 can be effectively extracted. Therefore, in this embodiment, the second physical quantity 22 can be accurately estimated.

[0117] In this embodiment, it is preferable that the input layer 25 of the prediction model 31 is adjusted so that the color image data 32 pseudo-created from the two-dimensional matrix (not shown) can be input.

[0118] [Method for Creating Prediction Model (Fourth Embodiment)] As shown in FIG. 4, in the previous embodiments, a plurality of cross sections 18 having different heights from the tread surface 16s (road surface model 16) of the tire were determined, but the present invention is not limited to such a mode. FIG. 11 is a conceptual diagram showing a cross section of another embodiment of the present invention.

[0119] In this embodiment, a plurality of cross sections 18 having different positions in the front-rear direction of the vehicle 8 (vehicle model 15) are determined. With such a plurality of cross sections 18, the first physical quantity (in this example, the total pressure coefficient of air) can be specified at a plurality of positions in the front-rear direction of the vehicle model 15. As a result, since the first physical quantity can be three-dimensionally grasped in the front-rear direction of the vehicle, by using it as the teacher data of the prediction model, it becomes possible to create a prediction model with improved prediction accuracy of the second physical quantity (in this example, the air resistance coefficient).

[0120] [Method for Predicting Physical Quantity of Tire] Next, the prediction method of this embodiment will be described. FIG. 12 is a flowchart showing an example of the processing procedure of the prediction method of the physical quantity of the tire. To implement the prediction method of this embodiment, the computer 1 (prediction device 1B) shown in FIG. 1 is used.

[0121] [Determine the Cross Section When the Tire to Be Predicted Is in the First Driving State] In the prediction method of this embodiment, first, when the tire to be predicted is in the first driving state, at least one cross section including the tire and the substance is determined (step S11).

[0122] In step S11 of this embodiment, first, the first input unit 6B included in the program unit 6 shown in FIG. 1 is read into the working memory 4C. The first input unit 6B is a program for determining at least one cross section including the tire 7 and the substance 10 when the tire to be predicted is in the first driving state. By executing this first input unit 6B by the arithmetic unit 4A, the computer 1 can function as means for determining the cross section of the tire to be predicted.

[0123] Step S11 of this embodiment determines the cross section when the tire to be predicted is in the first driving state based on the same procedure as the first step S1 (shown in FIG. 3) of the creation method of the previous embodiments. In step S11 of this embodiment, at least one cross section 18 (in this example, the first cross section 18a to the third cross section 18c) when the tire to be predicted is in the first driving state is determined by simulation using the tire model 14 and the like shown in FIG. 4.

[0124] The details of the three cross sections 18 (the first cross section 18a, the second cross section 18b, and the third cross section 18c) are as described in the creation method of the previous embodiments. The determined cross sections 18 (the first cross section 18a to the third cross section 18c) and the calculation results of the simulation are stored in the first storage unit 5A (computer 1) shown in FIG. 1.

[0125] [Input the first data for specifying the first physical quantity of the tire and / or substance to be predicted] Next, in the prediction method of this embodiment, first data for specifying the first physical quantity of the tire and / or substance to be predicted at at least one cross section 18 is input to the computer 1 (shown in FIG. 1) (step S12). In step S12 of this embodiment, first data for specifying the first physical quantity is acquired from the calculation results of the simulation in the same manner as the creation method of the previous embodiments.

[0126] In step S12 of the present embodiment, first, the cross-sections 18 (in this example, the first cross-section 18a to the third cross-section 18c) input to the first storage unit 5A shown in FIG. 1 and the calculation results of the simulation are read into the working memory 4C. Further, the second input unit 6C included in the program unit 6 is read into the working memory 4C. The second input unit 6C is a program for inputting the first data for specifying the first physical quantity of the tire and / or substance to be predicted in the determined cross-section 18. By the second input unit 6C being executed by the arithmetic unit 4A, the computer 1 can function as means for inputting the first data.

[0127] In step S12 of the present embodiment, similar to the second step S2 (shown in FIG. 3) of the creation method of the previous embodiments, from the calculation results of the simulation, as shown in FIG. 6, the first data 21 for specifying the first physical quantity 13 is acquired. In this first data 21, for the tire to be predicted, three grayscale data 20 (image data 21a to 21c) in which the first physical quantity 13, the tire model 14, and the vehicle model 15 can be specified are included. The first data 21 is stored in the second storage unit 5B (computer 1) shown in FIG. 1.

[0128] [Input a prediction model capable of estimating the second physical quantity from the first data] Next, in the prediction method of the present embodiment, a prediction model 31 (shown in FIG. 9) that is machine-learned so as to be able to estimate the second physical quantity 22 of the tire or substance to be predicted in the first driving state from the first data 21 is input to the computer 1 (shown in FIG. 1) (step S13).

[0129] In step S13 of the present embodiment, first, the fourth input unit 6E included in the program unit 6 shown in FIG. 1 is read into the working memory 4C. The fourth input unit 6E is a program for inputting a prediction model 31 that has been machine-learned so as to be able to estimate the second physical quantity 22 of a tire or substance in the first driving state from the first data 21. By executing this fourth input unit 6E by the arithmetic unit 4A, the computer 1 can function as means for inputting the prediction model 31.

[0130] In step S13, when the prediction model 31 shown in FIG. 9 has already been created, the prediction model 31 is input into the third storage unit 5C by the fourth input unit 6E (computer 1) shown in FIG. 1. On the other hand, when the prediction model 31 has not been created, the fourth input unit 6E creates the prediction model 31 according to the processing procedure of the creation method shown in FIG. 3, and the created prediction model 31 is input into the third storage unit 5C.

[0131] [Input the first data of the tire to be predicted into the prediction model and output the second physical quantity] Next, in the prediction method of the present embodiment, as shown in FIG. 9, the computer 1 (shown in FIG. 1) inputs the first data 21 (shown in FIG. 6) of the tire to be predicted into the prediction model 31 and outputs the second physical quantity 22 of the tire to be predicted (step S14).

[0132] In step S14 of the present embodiment, first, the first data 21 of the tire to be predicted input to the second storage unit 5B shown in FIG. 1 and the prediction model 31 input to the third storage unit 5C are read into the working memory 4C. Further, the output unit 6A included in the program unit 6 is read into the working memory 4C. The output unit 6A is a program for inputting the first data 21 of the tire to be predicted into the prediction model 31 and outputting the second physical quantity 22 of the tire to be predicted. By executing this output unit 6A by the arithmetic unit 4A, the computer 1 can function as means for outputting the second physical quantity 22.

[0133] In step S14 of this embodiment, first, the three grayscale data 20 (image data 21a to 21c) of the first data 21 of the tire to be predicted shown in FIG. 6 are respectively assigned to the R channel, G channel, and B channel of the RGB color model. As a result, as shown in FIG. 9, a single color image data 32 is pseudo-created. Next, the single color image data 32 of the tire to be predicted is input to the input layer 25 of the prediction model 31. As a result, the second physical quantity 22 estimated by the prediction model 31 based on the first data 21 is output from the output layer 26.

[0134] Furthermore, in step S14 of this embodiment, the computer 1 (shown in FIG. 1) may include a step of visualizing features that contributed to the prediction of the second physical quantity by the prediction model in the first data 21 of the tire to be predicted, as shown in FIG. 10. Visualization of such features can be performed, for example, by a prediction model 31 implemented with a known Grad-CAM.

[0135] In step S14 of this embodiment, the output second physical quantity of the tire to be predicted and the image data 33 with visualized features are stored in the sixth storage unit 5F shown in FIG. 1. Furthermore, in step S14, the estimated second physical quantity 22 (shown in FIG. 9) and the image data 33 with visualized features (shown in FIG. 10) are output from an output device 3 (shown in FIG. 1) such as a display device or a printer.

[0136] As described above, in the prediction method of this embodiment, as shown in FIG. 9, by inputting the first data 21 (color image data 32) of the tire to be predicted into the prediction model 31, the second physical quantity 22 of the tire to be predicted can be estimated. Therefore, in the prediction method of this embodiment, it is possible to predict the physical quantity between the tire and the substance without the need for the experience and intuition of an expert.

[0137] Furthermore, in the prediction method of the present embodiment, first data (shown in FIG. 6) at a plurality of cross-sections 18 (first cross-section 18a to third cross-section 18c) having different heights from the tread surface 16s of the tire shown in FIG. 4 is input into the prediction model 31 shown in FIG. 9. Thereby, the feature amount for estimating the second physical quantity 22 can be effectively extracted. Therefore, the second physical quantity 22 can be accurately estimated.

[0138] [Evaluate the second physical quantity of the tire to be predicted] Next, in the prediction method of the present embodiment, it is evaluated whether the second physical quantity 22 (shown in FIG. 9) of the tire to be predicted is good or not (step S15). The second physical quantity may be evaluated by the computer 1 shown in FIG. 1, or may be evaluated by an operator or the like.

[0139] In step S15 of the present embodiment, first, the second physical quantity 22 (shown in FIG. 9) of the tire to be predicted, which is input to the sixth storage unit 5F shown in FIG. 1, is read into the working memory 4C. Further, the evaluation unit 6J included in the program unit 6 is read into the working memory 4C. The evaluation unit 6J is a program for determining whether the second physical quantity 22 of the tire to be predicted is good or not. By executing this evaluation unit 6J by the arithmetic unit 4A, the computer 1 can be made to function as means for evaluating the second physical quantity 22.

[0140] Whether the second physical quantity 22 is good or not can be appropriately evaluated. In the present embodiment, when the second physical quantity (in this example, the air resistance coefficient) 22 is less than a predetermined threshold value, it is determined to be good. The threshold value can be appropriately set according to, for example, the air resistance required for the tire to be predicted, the fuel consumption performance, and the like.

[0141] When it is determined that the second physical quantity 22 of the tire to be predicted is good (Yes in step S15), the tire is manufactured based on the design factors of the tire to be predicted (step S16). On the other hand, when it is determined that the second physical quantity 22 of the tire to be predicted is not good (No in step S15), at least one of the design factors of the tire to be predicted is changed (step S17), and steps S11 to S15 are performed again. In step S17, the design factors and the like may be changed based on the image data 33 (shown in FIG. 10) in which the features contributing to the prediction of the second physical quantity 22 by the prediction model are visualized. Thereby, with the prediction method of the present embodiment, it becomes possible to surely design and manufacture the tire 7 with less air resistance.

[0142] As described above, the particularly preferred embodiments of the present invention have been described in detail. However, the present invention is not limited to the illustrated embodiments and can be implemented in various forms.

Example

[0143] According to the processing procedure shown in FIG. 3, a prediction model for predicting the physical quantity between the tire and the substance in contact with the outer surface of the tire was created (Example).

[0144] In the example, first, when the tire is in the first running state, at least one cross section including the tire and the substance was determined (first step). In this first step, based on the result of the computer simulation, as shown in FIG. 4, three cross sections with different heights from the running surface of the tire were determined.

[0145] Next, in the example, the first data for specifying the first physical quantity of the tire and / or the substance in the determined cross section was input to the computer (second step). In this second step, based on the calculation result of the above simulation, as shown in FIG. 6, the image data (grayscale data) capable of specifying the first physical quantity in each of the three cross sections was input as the first data.

[0146] Next, in the embodiment, the second physical quantity of the tire or substance in the first driving state was input into the computer (third step). The second physical quantity was calculated using the calculation result of the simulation.

[0147] Next, in the embodiment, as shown in FIG. 7, a data set linking the first data and the second physical quantity was created (fourth step). In the fourth step, a data set of a plurality of tires was created using the first data and the second physical quantity obtained by performing the first to third steps on a plurality of tires.

[0148] Next, in the embodiment, using the data set of a plurality of tires as teacher data, as shown in FIG. 9, a prediction model for estimating the second physical quantity from the first data to be predicted was created (fifth step). In this fifth step, first, for each of the plurality of tires, three grayscale data were respectively assigned to the R channel, G channel, and B channel of the RGB color model, and thus a single color image data was pseudo-created. Next, by inputting this color image data into the input layer of the convolutional neural network, the estimated second physical quantity was output from the output layer. Then, various parameters of the convolutional neural network were optimized so that the error between the estimated second physical quantity and the second physical quantity included in the tire data set (teacher data) was minimized. Thereby, a prediction model was created. The detailed specifications such as the number of tire types are as follows.

[0149] Number of tire types: 8 Tire size: 215 / 55R18 Rim size: 7.5J×18 First driving state: Driving speed: 140 km / h Internal pressure: 250 kPa Substance: Air First physical quantity: Total pressure coefficient of air Second physical quantity: Air resistance coefficient

[0150] Next, in order to evaluate the prediction accuracy of the prediction model, the second physical quantity was estimated from the first data included in the data sets of a plurality of tires used to create the prediction model. Then, for a plurality of tires, the estimated value of the second physical quantity (air resistance coefficient) was compared with the measured value of the second physical quantity.

[0151] FIG. 13 is a graph showing the relationship between the estimated value and the measured value of the air resistance coefficient in the examples. As shown in FIG. 13, in the examples, it is shown that there is a correlation between the estimated value and the measured value of the air resistance coefficient. Therefore, the examples were able to predict the physical quantity between the tire and the substance in contact with the outer surface of the tire without requiring the experience and intuition of a skilled person.

[0152] In the examples, in the first data of the tire to be predicted, a step of visualizing the features contributing to the prediction of the second physical quantity by the prediction model was performed. As shown in FIG. 10, it is shown that the side and the rear of the front tire model 14f, the outside of the body of the vehicle model 15, and the rear of the rear tire model 14r are being focused on by the prediction model. In these parts, it is considered that the influence on the second physical quantity is actually large.

[0153] In order to determine whether the influence on the second physical quantity is actually large in the parts focused on by the prediction model, the ground planes of the first physical quantity (total air pressure coefficient) of the front and rear tires were calculated using the results of the above-described simulation. Further, for the front tire, the streamline of the air and the contour line of the air resistance coefficient were calculated.

[0154] FIGS. 14(a) and (b) are diagrams each showing an isosurface of the total air pressure coefficient for different types of front tires. In FIG. 14, the side surface of the front tire model and the side surface of the body of the vehicle model are shown. Further, the isosurface was set based on the maximum value of the total pressure coefficient.

[0155] As shown in FIG. 14, the total pressure coefficient is large on the side and the back of the front tire model 14f and on the outer side of the body of the vehicle model 15. Since it is considered that the air resistance coefficient also increases in these portions, they coincide with the portions that contribute to the prediction of the second physical quantity.

[0156] FIGS. 15(a) and 15(b) are diagrams showing the streamline of air and the contour line of the air resistance coefficient, respectively, for different types of front tires. In FIG. 15, the bottom surfaces of the front tire model and the body of the vehicle model are shown.

[0157] In FIG. 15, in the region surrounded by the two-dot chain line behind the front tire model 14f, there are portions where air vortices and air flows concentrate. Since it is considered that the air resistance coefficient increases in these regions, they coincide with the portions that contribute to the prediction of the second physical quantity.

[0158] As described above, it was confirmed that in the portions that contribute to the prediction of the second physical quantity, the influence on the second physical quantity is actually large. Therefore, in the embodiment, it was confirmed that by visualizing the features that contribute to the prediction of the second physical quantity, the portions where the influence on the second physical quantity is actually large can be easily grasped. Therefore, in the embodiment, it was confirmed that it is useful for the development of tires and vehicles capable of reducing the second physical quantity.

[0159] [Appendix] The present invention includes the following aspects.

[0160] [Invention 1] A method for creating a prediction model for predicting a physical quantity between a tire and a substance in contact with an outer surface of the tire, comprising: a first step of determining at least one cross section so as to include the tire and the substance when the tire is in a first running state; a second step of inputting, into a computer, first data for specifying a first physical quantity of the tire and / or the substance in the at least one cross section; A third step of inputting, into the computer, a second physical quantity of the tire or the substance in the first driving state; A fourth step of the computer creating a data set in which the first data and the second physical quantity are associated, for a plurality of tires; A fifth step of the computer creating a prediction model for estimating the second physical quantity from the first data to be predicted, using the data set as teacher data, A method for creating a prediction model. [Invention 2] The method for creating a prediction model according to Invention 1, wherein the at least one cross section is a plurality of cross sections having different heights from the tread surface of the tire. [Invention 3] The method for creating a prediction model according to Invention 1 or 2, wherein the substance includes a fluid containing air or water. [Invention 4] The method for creating a prediction model according to Invention 1 or 2, wherein the substance includes a road surface deposit containing snow or mud. [Invention 5] The method for creating a prediction model according to any one of Inventions 1 to 4, wherein the first data includes image data. [Invention 6] The plurality of cross sections are three cross sections, Each of the data sets includes three image data, The three image data are each grayscale data, The fifth step includes a step of respectively assigning the three grayscale data to an R channel, a G channel, and a B channel of an RGB color model. The method for creating a prediction model according to any one of Inventions 2 to 5. [Invention 7] The prediction model according to Invention 6 includes a convolutional neural network capable of performing a convolutional operation on each of the R channel, the G channel, and the B channel. The method for creating a prediction model according to Invention 6. [Invention 8] The second step of the method for creating a prediction model according to any one of inventions 1 to 7 of the present invention includes a step of obtaining the first data from the calculation result of the simulation of bringing the substance into contact with the outer surface of the tire. [Invention 9] The substance is a fluid containing air or water, The second physical quantity includes a physical quantity related to fluid resistance, and the method for creating a prediction model according to any one of inventions 1 to 8 of the present invention. [Invention 10] The second physical quantity includes an air resistance coefficient, The first physical quantity includes a total pressure coefficient of air, and the method for creating a prediction model according to invention 9 of the present invention. [Invention 11] The substance is a road surface deposit containing snow or mud, The second physical quantity includes a physical quantity related to at least one of compressibility and pressure, and the method for creating a prediction model according to any one of inventions 1 to 8 of the present invention. [Invention 12] A method for predicting a physical quantity between a tire and a substance in contact with the outer surface of the tire, When the tire to be predicted is in a first driving state, a step of determining at least one cross section so as to include the tire and the substance, A step of inputting, to a computer, first data for specifying a first physical quantity of the tire to be predicted and / or the substance in the at least one cross section, A step of inputting, to the computer, a prediction model machine-learned so as to be able to estimate a second physical quantity of the tire or the substance in the first driving state from the first data, A step in which the computer inputs the first data of the tire to be predicted into the prediction model and outputs the second physical quantity of the tire to be predicted, A method for predicting a physical quantity of a tire. [Invention 13] The method for predicting a physical quantity of a tire according to the twelfth aspect of the present invention, further comprising a step of visualizing, by the computer, features contributing to the prediction of the second physical quantity in the first data of the tire to be predicted. [The fourteenth aspect of the present invention] An apparatus for predicting a physical quantity between a tire and a substance in contact with an outer surface of the tire, a first storage unit that stores at least one cross section determined to include the tire and the substance when the tire to be predicted is in a first running state; a second storage unit that stores first data for specifying a first physical quantity of the tire to be predicted and / or the substance in the at least one cross section; a third storage unit that stores a prediction model learned by machine learning so as to be able to estimate a second physical quantity of the tire or the substance in the first running state from the first data; an output unit that inputs the first data of the tire to be predicted into the prediction model and outputs the second physical quantity of the tire to be predicted. An apparatus for predicting a physical quantity of a tire.

Explanation of reference numerals

[0161] 13 First physical quantity 18 Cross section 21 First data 22 Second physical quantity 23 Dataset

Claims

1. A method for creating a prediction model for predicting a physical quantity between a tire and a substance in contact with an outer surface of the tire, comprising: a first step of determining at least one cross-section so as to include the tire and the substance when the tire is in a first running state; a second step of inputting, into a computer, first data for specifying a first physical quantity of the tire and / or the substance in the at least one cross-section; a third step of inputting, into the computer, a second physical quantity of the tire or the substance in the first running state; a fourth step of the computer creating a data set associating the first data and the second physical quantity for a plurality of tires; a fifth step of the computer creating a prediction model for estimating the second physical quantity from the first data to be predicted, using the data set as teaching data. A method for creating a prediction model.

2. The method for creating a prediction model according to claim 1, wherein the at least one cross-section is a plurality of cross-sections having different heights from a tread surface of the tire.

3. The method for creating a prediction model according to claim 1, wherein the substance includes a fluid containing air or water.

4. The method for creating a prediction model according to claim 1, wherein the substance includes a road surface deposit containing snow or mud.

5. The method for creating a prediction model according to claim 1, wherein the first data includes image data.

6. The plurality of cross-sections are three cross-sections, each of the data sets includes three image data, the three image data are each grayscale data, and the fifth step includes a step of respectively assigning the three grayscale data to an R channel, a G channel, and a B channel of an RGB color model. The method for creating a prediction model according to claim 2.

7. The method for creating a prediction model according to claim 6, wherein the prediction model includes a convolutional neural network capable of performing a convolutional operation on each of the R channel, the G channel, and the B channel.

8. The method for creating a prediction model according to claim 1, wherein the second step includes a step of obtaining the first data from a calculation result of a simulation in which the substance is brought into contact with an outer surface of the tire.

9. The substance is a fluid containing air or water, The method for creating a prediction model according to claim 1, wherein the second physical quantity includes a physical quantity related to fluid resistance.

10. The second physical quantity includes an air resistance coefficient, The method for creating a prediction model according to claim 9, wherein the first physical quantity includes a total pressure coefficient of air.

11. The substance is a road surface deposit including snow or mud, The method for creating a prediction model according to claim 1, wherein the second physical quantity includes a physical quantity related to at least one of compressibility and pressure.

12. A method for predicting a physical quantity between a tire and a substance in contact with an outer surface of the tire, When the tire to be predicted is in a first driving state, determining at least one cross section including the tire and the substance; Inputting, into a computer, first data for specifying a first physical quantity of the tire to be predicted and / or the substance in the at least one cross section; Inputting, into the computer, a prediction model learned by machine learning so as to be able to estimate a second physical quantity of the tire or the substance in the first driving state from the first data; The computer includes: inputting the first data of the tire to be predicted into the prediction model and outputting the second physical quantity of the tire to be predicted. A method for predicting a physical quantity of a tire.

13. The method for predicting a physical quantity of a tire according to claim 12, further including a step of the computer visualizing, in the first data of the tire to be predicted, features contributing to the prediction of the second physical quantity by the prediction model.

14. An apparatus for predicting a physical quantity between a tire and a substance in contact with an outer surface of the tire, A first storage unit that stores at least one cross section determined to include the tire and the substance when the tire to be predicted is in a first driving state; A second storage unit that stores first data for specifying a first physical quantity of the tire to be predicted and / or the substance in the at least one cross section; A third storage unit that stores a prediction model learned by machine learning so as to be able to estimate a second physical quantity of the tire or the substance in the first driving state from the first data; An output unit that inputs the first data of the tire to be predicted into the prediction model and outputs the second physical quantity of the tire to be predicted. An apparatus for predicting a physical quantity of a tire.

Images