Tire performance value estimation method, learning model training method, and tire performance value estimation device

Abstract

To estimate the performance value of a tire without using a ground surface image of a tire to which a load is actually or virtually applied.SOLUTION: A method for estimating the performance value of a tire using a computer has: a pre-processing step ST10 of acquiring tire information of a target tire for which the performance value is estimated; and an estimation step ST20 of outputting a performance value corresponding to the tire information by using a learned model. The tire information includes the specifications of the target tire, the load applied to the target tire, and pattern information related to a tread pattern of the target tire.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a method for estimating tire performance values, a learning method for a learning model used in the estimation method, and an apparatus for estimating tire performance values. [Background technology]

[0002] In the development of tires, there is a demand for a technique for predicting tire performance values. Patent Document 1 discloses a method for predicting tire performance values based on a contact patch image representing the tire contact patch shape. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2021-195038 Summary of the Invention [Problem to be solved by the invention]

[0004] In the case of the invention disclosed in Patent Document 1, in order to obtain an image of the contact patch, it is necessary to conduct a test in which a load is actually applied to the tire and the tire is photographed, or to conduct a simulation of what happens when a load is applied to the tire. An object of the present invention is to provide an estimation method that makes it possible to estimate tire performance values without using a tire contact patch image to which an actual or virtual load is applied, a learning method for a learning model used in the estimation method, and an estimation device that estimates tire performance values. [Means for solving the problem]

[0005] The present invention is a method for estimating tire performance values by a computer, comprising: a preprocessing step of acquiring tire information of a target tire for which the performance value is to be estimated; and an estimation step of outputting a performance value corresponding to the tire information using a trained model, wherein the tire information includes specifications of the target tire, a load to be applied to the target tire, and pattern information regarding the tread pattern of the target tire.

[0006] The present invention is a method for training a learning model for estimating tire performance values by a computer, comprising: a preparatory step of acquiring, for each of a plurality of tires, correspondence data that matches data based on tire information of the tires with the performance values; and a learning step of machine-learning a learning model using the correspondence data so that the learning model outputs the performance values when the tire information is input, wherein the tire information includes specifications of the tire, a load applied to the tire, and pattern information regarding the tread pattern of the tire.

[0007] The present invention is an apparatus for estimating tire performance values, comprising: a pre-processing unit that acquires tire information of a target tire for which the performance value is to be estimated; and an estimation unit that outputs a performance value corresponding to the tire information using a trained model, wherein the tire information includes specifications of the target tire, a load applied to the target tire, and pattern information regarding the tread pattern of the target tire. [Effects of the Invention]

[0008] According to the tire performance value estimation method and estimation device of the present invention, it is possible to estimate the tire performance value without using an image of the tire contact patch to which an actual or virtual load is applied. According to the learning method for a learning model of the present invention, a learning model (trained model) used in the estimation method and estimation device is obtained. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 2 is a block diagram of an estimation device and a learning device. [Figure 2] FIG. 2 is an explanatory diagram showing an example of a tread pattern. [Figure 3] FIG. 10 is a flowchart showing a method for estimating cornering power. [Figure 4] FIG. 4 is an explanatory diagram of a method for estimating cornering power. [Figure 5] FIG. 1 is a flow diagram showing a learning method for a learning model. [Figure 6] FIG. 2 is an explanatory diagram of primary image parameters and secondary image parameters. [Figure 7] FIG. 10 is an explanatory diagram of correspondence data. [Figure 8] FIG. 10 is an explanatory diagram of processing of pattern information. [Figure 9] FIG. 1 is a cross-sectional view schematically showing a tread portion of a tire. DETAILED DESCRIPTION OF THE INVENTION

[0010] [Details of the embodiment of the present invention] The present invention will now be described based on preferred embodiments with reference to the drawings.

[0011] The method for estimating a tire performance value of the present invention is a method for estimating a tire performance value by a computer. The estimation method uses a trained model generated by machine learning. There are various types of performance value, but the performance value in this embodiment is cornering power. In the following description, a tire for which cornering power is to be estimated may be referred to as a "target tire."

[0012] The learning model training method of the present invention is a method for training a learning model used in the estimation method. That is, the learning model training method of the present invention is a method for generating a learning model used to estimate a performance value (cornering power) of a tire (target tire) by training it using a computer. The computer generates the trained model by training the learning model using corresponding data, which will be described later and serve as training data.

[0013] As the training data, data based on tire information of the tire is used. In the following description, the tire for which the tire information serving as training data is used may be referred to as a "reference tire." The tire information used in the learning method includes the specifications of the tire (reference tire), the load applied to the tire (reference tire), and pattern information related to the tread pattern of the tire (reference tire).

[0014] Tests have been conducted on a large number of reference tires, and the characteristic value (cornering power) of each reference tire has been measured when a predetermined load is applied to it. As a result, the applied load, the characteristic value (cornering power) at that time, and the specifications of the reference tire have been obtained for each reference tire. Furthermore, pattern information regarding the tread pattern of each reference tire has been obtained. This information is associated with each reference tire. That is, data based on the specifications, the load, and the pattern information of the reference tire are associated with the characteristic value (cornering power) of the reference tire, and the corresponding data is acquired for each of a large number of reference tires. This data serves as the teacher data.

[0015] As will be described later, the estimation method for estimating cornering power uses tire information of the target tire. The tire information includes the specifications of the target tire, the load applied to the target tire, and pattern information related to the tread pattern of the target tire. The tread pattern of the target tire is the tread pattern when no load is applied to the target tire. Furthermore, the load included in the tire information of the target tire is not the load actually applied to the target tire, but an assumed load to be applied to the target tire. In other words, the estimation method determines the characteristic value (cornering power) of the target tire when an assumed load is applied to the target tire.

[0016] [Findings that form the basis of the present invention] The tread pattern of a tire affects various aspects of the tire's performance. For example, cornering power (hereinafter sometimes referred to as "CP") is known as an index of handling stability performance. Cornering power is generally measured using a drum testing machine or a flat belt testing machine. Cornering power is an index that indicates the force of a tire when a vehicle corners. The tire force described above is significantly affected by the size and shape of the blocks (land portions) in the tire tread, in addition to the physical properties of the rubber used in the tire. In other words, cornering power is affected by the way grooves are provided in the tire tread, i.e., the tread pattern.

[0017] Specifically, when the tread pattern of a tire changes, the tire's rigidity changes, and as a result, the cornering power of the tire changes. For example, as shown in the left diagram of Fig. 9, when the rubber blocks 92 surrounded by grooves 91 become smaller, the tire's rigidity decreases. In contrast, as shown in the right diagram of Fig. 9, when the rubber blocks 92 become larger, the tire's rigidity increases. Fig. 9 is a cross-sectional view schematically showing the tread portion of a tire.

[0018] The inventors have focused on this point and completed the invention relating to a method (estimation device) for estimating tire performance values, as described below, and a method (learning device) for learning a learning model used in the estimation method. The inventors have realized that cornering power can be estimated by using pattern information about the tread pattern as input to a computer. The estimated cornering power can be used, for example, in simulations for tire development.

[0019] Hereinafter, preferred embodiments of a method (estimation device) for estimating a tire performance value and a method (learning device) for learning a learning model will be described.

[0020] [Device configuration] FIG. 1 is a block diagram of an estimation device 10 and a learning device 50. The estimation device 10 is a device that executes estimation (estimation method) of the performance value (CP) of a tire. The learning device 50 is a device that executes learning (learning method) of a learning model used in the estimation method.

[0021] The estimation device 10 includes a computer 11. The computer 11 includes a processor (arithmetic processing device) 12 and a storage unit 13 such as a semiconductor memory including RAM, ROM, etc. or a hard disk. A computer program 15 for executing a method for estimating tire performance values is installed in the computer 11. The processor 12 reads out and executes the computer program 15 from the semiconductor memory (storage unit 13). This allows the computer 11 to function as the estimation device 10 that executes the method for estimating tire performance values.

[0022] The estimation device 10 may be configured with one or more computers. When the estimation device 10 is configured with multiple computers, the multiple computers can communicate information with each other and cooperate to execute various processes.

[0023] The processor 12 executes the computer program 15, and the estimation device 10 has various functional units. As the functional units, the estimation device 10 has a preprocessing unit 21 and an estimation unit 22. The pre-processing unit 21 acquires tire information of the target tire. The tire information includes specifications of the target tire, the load applied to the target tire, and pattern information related to the tread pattern of the target tire. The tire information is input through a data input unit (input interface) 14 included in the estimation device 10, and the pre-processing unit 21 acquires the tire information.

[0024] The specifications of the target tire are values indicating the dimensions and usage conditions of the target tire, and include, for example, tire size and air pressure. The specifications of the target tire are configured, for example, as listed data (specification data).

[0025] The load applied to the target tire is a load applied in the vertical direction to the target tire with its rotation axis horizontal. The load is one of the test conditions that affect the cornering power of the tire. Note that the tire information may include test conditions other than the load. The load to be applied to the target tire is input to the computer 11 together with the above-mentioned specifications, for example.

[0026] The pattern information of the target tire is information based on image information of the tread pattern of the target tire. In this embodiment, the pattern information is information based on image information for two pitches of the tread pattern.

[0027] Fig. 2 is an explanatory diagram showing an example of a tread pattern P. The tread pattern P is a pattern on the tread surface of a tire (target tire, reference tire), and is configured by unit patterns repeatedly arranged along the circumferential direction of the tire. The unit pattern is defined by grooves Q formed in the tread, and is a pattern within a single range that is not repeated along the circumferential direction. In the tread pattern P shown in FIG. 2, one area surrounded by a thick line is a "unit pattern (1 unit pattern)." A unit pattern corresponds to one pitch of the tread pattern. Therefore, two pitches of the tread pattern are two consecutive unit patterns.

[0028] The pattern information of the target tire is input to the computer 11 as image information. The pre-processing unit 21 acquires pattern information and processes it as data. The pre-processing unit 21 acquires pattern feature quantities of the target tire based on the pattern information. The acquisition means (acquisition method) will be described later, but image parameters (brightness values, multiplication values of brightness values) of the target tire are determined based on the pattern information, and the pattern feature quantities are determined by a calculation (convolutional neural network) using the image parameters as input. The pattern feature quantities are used as input data for estimating cornering power.

[0029] The estimation unit 22 outputs the cornering power (performance value) corresponding to the tire information acquired by the preprocessing unit 21 using the trained model 16. This trained model 16 is a mathematical model (computational model formula) that has been machine-learned to output the cornering power of a target tire when tire information of the target tire is input to the computer 11. In the present embodiment, the trained model 16 for cornering power (CP) is a model based on a neural network.

[0030] [Learning device 50 that executes the learning method for the learning model] As described above, the learning device 50 is a device that executes the learning method of the learning model (the neural network) used in the estimation method.

[0031] The learning device 50 (see FIG. 1) is configured with a computer 61. The computer 61 has a processor (arithmetic processing unit) 62 and a storage unit 63 such as a semiconductor memory including RAM, ROM, etc. or a hard disk. A computer program 65 for executing a learning method for a learning model is installed in the computer 61. The processor 62 reads and executes the computer program 65 from the semiconductor memory (storage unit 63). This allows the computer 61 to function as a learning device 50 that executes the learning method for a learning model.

[0032] The learning device 50 may be configured with one or more computers. When the learning device 50 is configured with multiple computers, the multiple computers can communicate information with each other and cooperate to execute various processes.

[0033] The processor 62 executes the computer program 65, and the learning device 50 has various functional units. As the functional units, the learning device 50 has a preparation processing unit 71 and a learning unit 72. As described above, tests have been carried out in advance on a large number of reference tires, and the cornering power and tire information of each reference tire have been obtained.

[0034] Therefore, the preparation processing unit 71 acquires, for each of a plurality of reference tires, correspondence data that associates data based on the tire information of the reference tire with the cornering power of the reference tire. The tire information of the reference tire includes the specifications of the reference tire, the load applied to the reference tire (load in the test), and pattern information (image information) related to the tread pattern of the reference tire. The tire information is input via a data input unit (input interface) 64 of the learning device 50, and the preparation processing unit 71 acquires the tire information.

[0035] The specifications of the reference tire are values indicating the dimensions and usage conditions of the reference tire, and include, for example, tire size and air pressure. The specifications of the reference tire are configured as, for example, listed data (specification data).

[0036] The load applied to the reference tire is a load applied in the vertical direction to the reference tire with its rotation axis horizontal. The load is one of the test conditions that affect the cornering power of the tire. Note that the tire information may include test conditions other than the load. The load to be applied to the reference tire is input to the computer 61 together with the above-mentioned specifications, for example.

[0037] The pattern information of the reference tire is information based on image information of the tread pattern of the reference tire. In this embodiment, the pattern information is information based on image information of two pitches of the tread pattern.

[0038] The pattern information of the reference tire is input to the computer 61 as image information. The preparation processing unit 71 acquires the pattern information and processes it as data. The preparation processing unit 71 acquires the pattern feature quantities of the reference tire based on the pattern information. The acquisition means (acquisition method) is the same as the acquisition means (acquisition method) for the pattern information of the target tire that is the target for CP estimation, and therefore a description thereof will be omitted.

[0039] As described above, the preparation processing unit 71 acquires the specifications of the reference tire, the load applied to the reference tire, and data based on the pattern information of the reference tire (pattern feature amount) as data based on the tire information of the reference tire. The preparation processing unit 71 acquires, for each of the multiple reference tires, correspondence data that associates the data based on the tire information of the reference tire with the cornering power (performance value) of the reference tire.

[0040] The learning unit 72 uses the correspondence data as training data to train a learning model so that when tire information is input to the computer, the learning model outputs the cornering power of the target tire. The mathematical model generated by this machine learning becomes the trained model 16 for the CP.

[0041] [Method for estimating cornering power] The estimation method executed by the estimation device 10 will now be described. Fig. 3 is a flow diagram showing a method for estimating cornering power. The estimation method includes a preprocessing step (ST10) and an estimation step (ST20), and cornering power is output through the estimation step (ST30). Fig. 4 is an explanatory diagram of the cornering power estimation method.

[0042] In the pre-processing step (ST10), tire information of a target tire for which cornering power is to be estimated is acquired. The pre-processing step is performed by a pre-processing unit 21 (see FIG. 1). In the estimation step (ST20), the cornering power corresponding to the tire information acquired in the preprocessing step is output using the trained model 16. The estimation step is performed by the estimation unit 22 (see FIG. 1).

[0043] As described above, the tire information includes the specifications of the target tire, the load applied to the target tire, and pattern information (image information) relating to the tread pattern of the target tire. The specifications of the target tire are configured as, for example, a list of data. This data is input into the computer 11, whereby the specifications of the target tire are acquired. The load applied to the target tire is acquired by inputting the above-mentioned specifications into the computer 11.

[0044] In this embodiment, the pattern information is information based on image information for two pitches of the tread pattern of the target tire. In the pre-processing step (ST10), the pattern information is processed by the pre-processing unit 21. The data processing will be described below.

[0045] The pre-processing step (ST10) includes a first process (ST11) and a second process (ST12). In the first process (ST11), image parameters of the target tire are determined based on pattern information of the target tire. In the second process (ST12), pattern features of the target tire are determined by a calculation using the image parameters as input. This calculation is based on a convolutional neural network (CNN), as will be explained later.

[0046] Then, the pattern features, the specifications of the target tire, and the load are input to the computer 11, and in an estimation step (ST20), the cornering power (CP) is output by the trained model (neural network) 16 for CP (ST30).

[0047] The first treatment (ST11) and the second treatment (ST12) carried out in the pretreatment step (ST10) will be further described. The pre-processing unit 21 has an image processing function. In the first process (ST11), a primary image parameter D2 is acquired for each of a plurality of regions obtained by dividing the tread pattern based on the pattern information D1 (see FIG. 4) of the target tire, which is image information. In the case of this embodiment, a "brightness value" for each of the divided regions is acquired as the primary image parameter D2. In the case of 8 bits, the brightness value of the image is expressed as a value ranging from 0 (zero) to 255. The minimum is zero and the maximum is 255.

[0048] In the example shown in Fig. 4, for ease of explanation, the image information for two pitches of the tread pattern is divided into four in the tread width direction (vertical direction of the image) and four in the tread circumferential direction (horizontal direction of the image), for a total of 16 divisions. The number of divisions is arbitrary, and is preferably set to be greater than the number shown in the figure. The brightness value of each of the 16 divided regions is obtained by image processing. Values such as "1," "3," and "5" for the primary image parameter D20 shown in Figure 4 indicate the brightness value of each region. Note that in Figure 4, image information for two pitches of the tread pattern is divided, and the brightness value for each divided region is shown.

[0049] Furthermore, in the first process (ST11), a secondary image parameter (calculated value of brightness value) D3 for each region is obtained by a first calculation of each primary image parameter (brightness value) D2 for each region and the load applied to the target tire. In the example shown in FIG. 4, the load is "5 kN (kilonewtons)." The first calculation is "multiplication," and the secondary image parameter D3 for each region is the multiplied value of the brightness value for each region.

[0050] In the second process (ST12), a pattern feature D4 of the target tire is determined by a second calculation using the secondary image parameters (calculated values of brightness values) as input. In this embodiment, the second calculation is a calculation performed using a trained model for tread patterns. The trained model for tread patterns 66 is a mathematical model that has been machine-trained to output a pattern feature D4 when the secondary image parameters (multiplied values of brightness values) are input. Specifically, the second calculation is a calculation using a convolutional neural network (CNN). For example, the second calculation is a convolutional neural network (VGG-16) with a depth of 16 layers.

[0051] As described above, in the pre-processing step (ST10), the pattern feature value D4 is acquired based on the pattern information D1 related to the tread pattern of the target tire. In the estimation step (ST20), the pattern feature value D4 is used as input data for estimating the cornering power. In the estimation method of this embodiment, it is possible to estimate cornering power as a performance value of a target tire by inputting tire information including the specifications of the target tire, the load to be applied, and pattern information D1 related to the tread pattern into the computer 11, without using an image of the contact patch of the tire to which a load is actually or virtually applied.

[0052] [Regarding pattern information D1] The pattern information D1 used in the pre-processing step (ST10) will be further explained. As shown in FIG. 4, the pattern information D1 is an image of a portion (two pitches) of the tread pattern, and is information on a rectangular image with a first side in the tread width direction and a second side in the tread circumferential direction.

[0053] As will be explained later in the learning method of the learning model, in processing the pattern information D1, the estimation device 10 (and the learning device 50) sets a "reference rectangular frame F." In Fig. 4, the reference rectangular frame F has a contour shape of an area including a part (two pitches) of the tread pattern and the blackened portion.

[0054] A rectangular image (pattern information D1) equivalent to two pitches of the tread pattern is set at the original scale without being enlarged or reduced within the reference rectangular frame F. The rest of the reference rectangular frame F other than the rectangular image equivalent to two pitches of the tread pattern is painted black. When calculating the brightness value as the primary image parameter D2, the primary image parameters (brightness values) of the portions (blackened portions) other than the rectangular image that is part of the tread pattern are invalidated in the reference rectangular frame F. In other words, as a process of invalidation, the brightness values of the blackened portions in the reference rectangular frame F are set to zero.

[0055] [Learning model learning method] In the estimation method, the pattern feature D4, specifications, and load of the target tire are input to the computer 11, and in the estimation step (ST20 in Figure 3), the cornering power is output by the trained model 16 for CP (neural network, see Figure 4). This trained model 16 for CP is generated by a learning method executed by a learning device 50 (FIG. 1). The trained model 16 for CP generated by the learning device 50 is provided to the estimation device 10 and stored in the storage unit 13. The estimation device 10 uses the trained model 16 to perform estimation processing. The learning method will be described below.

[0056] 5 is a flow diagram showing the learning method, which includes a preparation step (ST110) and a learning step (ST120). For the preparation step (ST110), information about a large number of reference tires is acquired in advance, as shown in Fig. 7. That is, tire information about each reference tire and information about the cornering power of each reference tire are acquired in advance. In Fig. 7, information about one reference tire is written as one corresponding data.

[0057] The tire information of the reference tire includes the specifications of the reference tire, the load applied to the reference tire, and pattern information related to the tread pattern of the reference tire. In Fig. 7, correspondence data is shown that associates the specifications and load of the reference tire with the cornering power (CP), and each of the correspondence data includes pattern information (not shown).

[0058] The specifications of the reference tire and the load applied to the reference tire are configured as, for example, a list of data, as shown in Fig. 7. In the preparation step (ST110), this data is input to the computer 61, whereby the specifications of the reference tire and the load applied to the reference tire are acquired.

[0059] The pattern information of the reference tire is processed to obtain the pattern feature quantities of the reference tire. That is, image parameters of the reference tire are obtained based on the pattern information of the reference tire, and the pattern feature quantities of the reference tire are obtained by a calculation (convolutional neural network) using the image parameters as input.

[0060] Specifically, in a preparation step (ST110), image parameters (brightness values) of the reference tire are determined based on the pattern information of the reference tire. This process is the same as the first process ST11 of the estimation method (FIGS. 3 and 4). That is, as shown in FIG. 4, the tread pattern (two pitches) is divided into a plurality of regions, and primary image parameters (brightness values) D2 for each region are multiplied by the load applied to the reference tire to determine secondary image parameters (multiplied values of brightness values) D3 for each region.

[0061] Then, the pattern feature quantity of the reference tire is calculated by a calculation (convolutional neural network) using the calculated secondary image parameters (multiplied values of brightness values) as input. This process is the same as the second process ST12 of the estimation method (FIGS. 3 and 4).

[0062] Note that there are cases where CPs are obtained by applying multiple loads to the same reference tire having the same tread pattern. For example, there are cases where CPs are obtained for the same reference tire when a load of 5 kN (kilonewtons) is applied and when a load of 8 kN (kilonewtons) is applied.

[0063] In this case, as shown in Fig. 6, a secondary image parameter D3-1 is obtained by multiplying each primary image parameter (brightness value) D2 for each region by "5", and a secondary image parameter D3-2 is obtained by multiplying each primary image parameter (brightness value) D2 for each region by "8", and these are used together. In other words, when multiple types of loads are applied to the same reference tire and CPs are obtained, different secondary image parameters (D3-1, D3-2) are obtained for each of the multiple types of loads. In this way, by multiplying the same primary image parameter (brightness value) D2 by different load values of "5" and "8", different pattern features are obtained even if the pattern information (two pitches of the tread pattern) is the same. In other words, if the load is different, the resulting pattern information will not be treated as the same.

[0064] For each reference tire (even if the reference tire is the same, for each reference tire with a different load), data based on tire information is configured from the pattern feature amount of the reference tire, the specifications of the reference tire, and the load applied to the reference tire. Then, for each reference tire, correspondence data is obtained that associates the data based on the tire information of the reference tire with the cornering power (CP) of the reference tire.

[0065] As described above, in the preparation step (ST110), the pattern feature values, specifications, and load of the reference tire are acquired as data based on the tire information of the reference tire. Then, as training data, correspondence data that associates the data based on the tire information of the reference tire with the cornering power (performance value) of the reference tire is acquired for each of the multiple reference tires. This preparation step (ST110) is performed by the preparation processing unit 71 (see FIG. 1).

[0066] In the learning step (ST120 in FIG. 5), machine learning is performed using the correspondence data. When the pattern feature, tire specifications, and load are input to the computer, a learning model is machine-learned to output cornering power. This learning model ultimately becomes the trained model 16 for CP.

[0067] As described above, in the learning step (ST120), the correspondence data acquired as described above is used as training data to train a learning model so that cornering power is output when tire information is input to the computer. The learning step is performed by the learning unit 72 (see FIG. 1). The machine learning of the learning model in this embodiment is machine learning using a neural network, and a known algorithm is adopted as the learning algorithm.

[0068] [Processing of pattern information] As described above, in the preparation step (ST110), pattern feature quantities of a reference tire are acquired based on the pattern information of the reference tire. As shown in FIG. 8, the pattern information D1 has various image sizes depending on the reference tire. Therefore, in the preparation step, a process is executed to set (fit) the pattern information into a common reference rectangular frame. This process will be described below. FIG. 8 is an explanatory diagram of the pattern information processing. FIG. 8 shows pattern information D1 for three different types of tires.

[0069] Each piece of pattern information D1 is an image of a portion (two pitches) of the tread pattern, and is information on a rectangular image with a first side in the tread width direction and a second side in the tread circumferential direction. Of the multiple rectangular images (pattern information D1) acquired from multiple tires, the first side of the image with the longest first side is defined as the "reference first side," and the second side of the image with the longest second side is defined as the "reference second side."

[0070] In the example shown in Fig. 8, the rectangular image on the left (pattern information D1) has the longest first side, so the first side of that rectangular image is defined as the "reference first side." The rectangular image on the center (pattern information D1) has the longest second side, so the second side of that rectangular image is defined as the "reference second side."

[0071] A rectangular frame formed by the first reference edge and the second reference edge is defined as a "reference rectangular frame F." 8, rectangular images that are three types of pattern information D1 are set within the reference rectangular frame F. That is, each of the three types of rectangular images (pattern information D1) is set within the reference rectangular frame F without being enlarged or reduced, at the original scale.

[0072] Furthermore, in the reference rectangular frame F, the portions other than the rectangular image (pattern information D1) are painted black. In other words, the portions other than a portion (two pitches) of the tread pattern are painted black. The image parameters (brightness values) of the portions other than the rectangular image in the reference rectangular frame F are invalidated. In this embodiment, the brightness values of the portions other than the rectangular image (two pitches of the tread pattern) are set to zero.

[0073] The reference rectangular frame F containing each rectangular image (each piece of pattern information D1) is compressed to an image size of, for example, a predetermined size (256 x 128). Although the rectangular images (pattern information D1) vary in size for each tire, they become data of a common size when fitted into the reference rectangular frame F. Therefore, the aspect ratio of each rectangular image (each piece of pattern information D1) remains the same even when the image is compressed. Based on this rectangular image (pattern information D1), the brightness value for each divided area is calculated.

[0074] 〔others〕 In the estimation method of this embodiment (see FIG. 3), the cornering power value is calculated by combining the pattern information of the target tire, which is image information, with numerical information such as the specifications and load of the target tire. In other words, the estimation method of this embodiment uses multimodal processing. The cornering power of the target tire may be estimated by other methods, provided that the method uses a trained model to output the cornering power corresponding to tire information, including the specifications of the target tire, the load applied to the target tire, and pattern information related to the tread pattern of the target tire.

[0075] The following method, for example, may be used as a method different from the estimation method of this embodiment. That is, the pattern information of the target tire, which is image information, is used as first tire information, and a first cornering power is calculated using this as input by a first trained model. Information on the specifications and load of the target tire is used as second tire information, and a second cornering power is calculated using this as input by a second trained model. The average value of the first cornering power and the second cornering power is the output result, which is the cornering power. Cornering power may be estimated using this method. [Industrial Applicability]

[0076] The tire performance value estimation method (estimation device) and learning model learning method (learning device) described above are used for the development of various tires.

[0077] [Note] The present invention includes the following aspects. (1) A method for estimating a tire performance value is a method for estimating a tire performance value by a computer, the method comprising: a preprocessing step of acquiring tire information of a target tire whose performance value is to be estimated; and an estimation step of outputting a performance value corresponding to the tire information using a trained model; The tire information includes specifications of the target tire, a load applied to the target tire, and pattern information relating to a tread pattern of the target tire.

[0078] (2) The method for estimating a tire performance value according to (1), wherein the performance value is cornering power. (3) The estimation method according to (1) or (2), wherein the pattern information is information based on image information for two pitches of the tread pattern.

[0079] (4) The pre-processing step includes a first process of determining image parameters of the target tire based on the pattern information of the target tire, and a second process of determining pattern feature quantities of the target tire by calculation using the image parameters as input, The estimation method according to any one of (1) to (3), wherein in the estimation step, the pattern feature, the specifications of the target tire, and the load are input, and the performance value is output by the trained model.

[0080] (5) The first processing is a processing of acquiring primary image parameters for each of a plurality of regions obtained by dividing the tread pattern based on the pattern information of the target tire, and calculating secondary image parameters for each of the regions by a first calculation of each of the primary image parameters for each of the regions and the load applied to the target tire, The estimation method according to (4), wherein the second processing is processing for obtaining the pattern feature amount of the target tire by a second calculation using the secondary image parameters as input.

[0081] (6) The estimation method according to (5), wherein the second calculation is a calculation using a trained model for a tread pattern that has been machine-learned to output the pattern feature when the secondary image parameters are input.

[0082] (7) A learning method for a learning model is a method for making a computer learn a learning model for estimating tire performance values, and includes a preparation step of acquiring, for each of a plurality of tires, correspondence data that associates data based on tire information of the tire with the performance values, and a learning step of machine learning a learning model using the correspondence data so that the learning model outputs the performance values when the tire information is input, The tire information includes the specifications of the tire, the load applied to the tire, and pattern information related to the tread pattern of the tire.

[0083] (8) The learning method of (7), wherein in the preparation step, image parameters of the tire are determined based on the pattern information of the tire, and pattern features of the tire are determined by calculation using the image parameters as input, and in the learning step, a learning model is machine-learned to output the performance value when the pattern features, tire specifications, and the load are input.

[0084] (9) The pattern information is an image of a portion of the tread pattern, and is information of a rectangular image with a first side in the tread width direction and a second side in the tread circumferential direction, The learning method of (8) above, wherein in the preparation step, of the plurality of rectangular images acquired from a plurality of tires, the first side of the image with the longest first side is defined as a reference first side, and the second side of the image with the longest second side is defined as a reference second side, and the rectangular image that is the pattern information is set within a reference rectangular frame formed by the reference first side and the reference second side.

[0085] (10) A device for estimating a tire performance value includes a preprocessing unit that acquires tire information of a target tire whose performance value is to be estimated, and an estimation unit that outputs a performance value corresponding to the tire information using a trained model; The tire information includes specifications of the target tire, a load applied to the target tire, and pattern information relating to a tread pattern of the target tire. [Explanation of symbols]

[0086] 10... Estimation device 11. Computer 21···Preprocessing section 22...Estimation part 50 Learning Device 61. Computer D1 Pattern Information D2 Primary image parameters D3 Secondary Image Parameters D4 Pattern features F... Reference rectangular frame P···Tread pattern ST11: First treatment ST12 Secondary treatment

Claims

1. A method for estimating tire performance values by a computer, comprising: a pre-processing step of acquiring tire information of a target tire for which the performance value is to be estimated; an estimation step of outputting a performance value corresponding to the tire information using a trained model; and The tire information includes specifications of the target tire, a load applied to the target tire, and pattern information related to a tread pattern of the target tire. A method for estimating tire performance values.

2. The performance value is cornering power. The method for estimating tire performance values according to claim 1.

3. The pattern information is information based on image information for two pitches of the tread pattern. The method for estimating a tire performance value according to claim 1 or 2.

4. The pre-processing step comprises: a first process of determining image parameters of the target tire based on the pattern information of the target tire; a second process for determining a pattern feature amount of the target tire by calculation using the image parameters as input, In the estimation step, the pattern feature amount, the specifications of the target tire, and the load are input, and the performance value is output by the trained model. The method for estimating a tire performance value according to claim 1 or 2.

5. The first treatment is Based on the pattern information of the target tire, primary image parameters are acquired for each of a plurality of regions obtained by dividing the tread pattern, and a process of calculating a secondary image parameter for each of the regions by a first calculation of each of the primary image parameters for each of the regions and the load applied to the target tire, The second treatment is a process of obtaining the pattern feature amount of the target tire by a second calculation using the secondary image parameters as input; The method for estimating a tire performance value according to claim 4.

6. the second calculation is a calculation using a trained model for a tread pattern that has been machine-learned to output the pattern feature amount when the secondary image parameters are input, The method for estimating a tire performance value according to claim 5.

7. A method for training a learning model for estimating tire performance values by a computer, comprising: a preparation step of acquiring, for each of a plurality of tires, correspondence data that associates data based on tire information of the tires with the performance values; a learning step of performing machine learning on a learning model using the corresponding data so as to output the performance value when the tire information is input; and The tire information includes specifications of the tire, a load applied to the tire, and pattern information related to a tread pattern of the tire. How the learning model is trained.

8. In the preparation step, determining image parameters of the tire based on the pattern information of the tire; determining a pattern feature amount of the tire by performing a calculation using the image parameters as input; In the learning step, a learning model is machine-trained to output the performance value when the pattern feature amount, the tire specifications, and the load are input; The learning method for the learning model according to claim 7.

9. the pattern information is an image of a portion of the tread pattern, and is information of a rectangular image having a first side in the tread width direction and a second side in the tread circumferential direction, In the preparation step, Among the plurality of rectangular images acquired from the plurality of tires, the first side of an image having the longest first side is defined as a reference first side, and the second side of an image having the longest second side is defined as a reference second side, the rectangular image, which is the pattern information, is set within a reference rectangular frame formed by the reference first side and the reference second side; The learning method for a learning model according to claim 8.

10. A device for estimating a tire performance value, a pre-processing unit that acquires tire information of a target tire whose performance value is to be estimated; an estimation unit that outputs a performance value corresponding to the tire information using a trained model; and The tire information includes specifications of the target tire, a load applied to the target tire, and pattern information related to a tread pattern of the target tire. A tire performance value estimation device.

Images