Tire performance prediction model learning method, tire performance prediction method, system and program

Abstract

To provide a method for learning a tire performance prediction model, a method for predicting tire performance, a system and a program which predict a tire performance value based on a tire ground surface image.SOLUTION: A method for learning a tire performance prediction model executed by one or a plurality of processors includes: a feature amount extraction step of inputting an input image based on a ground surface image indicating a tire ground surface shape to a feature extraction part, and extracting a feature amount of the image; and a learning step of performing machine learning on a prediction model so as to output a tire performance value with the extracted feature amount of the image and a different parameter relating to the tire of the ground surface image as explanatory variables.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a tire performance prediction model learning method, a tire performance prediction method, a system, and a program. [Background technology]

[0002] The use of machine learning is spreading to various industrial fields, and there are an increasing number of cases where other companies in the same industry are applying it to tires. For example, Patent Document 1 describes a method of inputting an image of a tire tread into a neural network to estimate the state or amount of wear.

[0003] Since tire designers often discuss tire performance based on the tire contact patch, the inventors of the present disclosure believe that tire performance is expressed in the contact patch. However, no specific method has been proposed for predicting tire performance from a tire contact patch image. In addition, there is a demand for predicting tire performance with high accuracy. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2019-35626 A Summary of the Invention [Problem to be solved by the invention]

[0005] The present disclosure provides a tire performance prediction model learning method, a tire performance prediction method, a system, and a program for predicting tire performance values ​​based on a tire contact patch image. [Means for solving the problem]

[0006] The training method for a tire performance prediction model disclosed herein is a method executed by one or more processors, and includes a feature extraction step of inputting an input image based on a contact patch image representing the tire contact patch shape into a feature extraction unit to extract image features, and a training step of machine learning a prediction model so as to output tire performance values ​​using the extracted image features and another tire-related parameter of the contact patch image as explanatory variables. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram showing a usage manner of a tire performance prediction model learning system and a tire performance prediction system according to a first embodiment. [Diagram 2] 1 is a block diagram showing a tire performance prediction model learning system and a tire performance prediction system according to a first embodiment. [Diagram 3] 3 is a flowchart executed by the tire performance prediction model learning system of the first embodiment. [Figure 4] 3 is a flowchart executed by the tire performance prediction system of the first embodiment. [Diagram 5] FIG. 13 is an explanatory diagram relating to a process of generating an input image by trimming a ground plane image. [Figure 6] FIG. 11 is a block diagram showing a tire performance prediction model learning system and a tire performance prediction system according to a second embodiment. [Figure 7] 10 is a flowchart executed by a tire performance prediction model learning system according to a second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] First Embodiment Hereinafter, a first embodiment of the present disclosure will be described with reference to the drawings.

[0009] FIG. 1 is a diagram showing how a tire performance prediction model learning system 1 and a tire performance prediction system 2 are used. As shown in FIG. 1, a contact patch image 5 is obtained from a test device 3 or a tire contact simulation system 4. The tire performance prediction model learning system 1 uses training data that associates the contact patch image 5 and other parameters 7 as inputs (explanatory variables) to the prediction model with correct tire performance values ​​as outputs from the prediction model, to perform machine learning of a prediction model. The input image used for the training data is based on the contact patch image. The tire performance prediction system 2 uses the prediction model constructed by the tire performance prediction model learning system 1 to calculate (predict) and output tire performance values ​​based on the contact patch image 5 to be predicted.

[0010] The tire performance values ​​to be predicted are actual measured values. The tire performance values ​​used in the first embodiment are CP (cornering power), SAP (self-aligning power), CFmax (maximum cornering force), and SAT (self-aligning torque). Of course, any tire performance other than these can be used as the tire performance values.

[0011] The ground contact patch image 5 represents the shape of the tire ground contact patch. The ground contact patch image 5 of the first embodiment represents the shape of the tire ground contact patch, and also represents the ground contact pressure Pz in a direction perpendicular to the road surface by color in the case of a color image, or by brightness in the case of a grayscale image. Since the ground contact pressure is displayed together with the ground contact patch shape, the ground contact patch image 5 represents the ground contact pressure distribution. The ground contact pressure distribution is associated with a pressure value for each area of ​​the ground contact patch. The ground contact patch image 5 of the first embodiment is an image of the tire in a stationary state where the tire is not rolling, but is not limited to this, and may be an image of the tire rolling. Regardless of whether the tire is in a rolling state or stationary state, the camber angle of the tire can be set to any angle. The ground contact patch image 5 of the first embodiment is an image of the tire in a stationary state, but when the tire is in a rolling state, the slip angle with respect to the traveling direction may be 0 degrees or may be an angle other than 0 degrees. The ground contact patch image 5 of the first embodiment represents only the ground contact pressure Pz in a direction perpendicular to the road surface in a stationary state, but is not limited to this. For example, the pressure Px along the front-rear direction perpendicular to both the tire axial direction and the direction perpendicular to the road surface in a stationary state, or the pressure Py along the tire axial direction may be expressed. In addition, when the tire is in a rolling state, the pressure Px along the tire traveling direction or the pressure Py along the direction perpendicular to the tire traveling direction may be expressed. Note that these coordinate systems are only examples and can be changed as appropriate. Note that, as long as accuracy can be ensured, the contact patch image 5 may represent only the tire contact patch shape. The test device 3 grounds the test target tire 30 on the test road surface 31 under a predetermined load, and photographs the contact patch shape with the camera 32 through the transparent road surface of the test road surface 31. The contact pressure is measured by an optical method using a transparent plate or the like, or by using a pressure sensor. The test device 3 generates the contact patch image 5 by the above test. In addition, the contact patch image 5 may be acquired from a simulation result obtained by the tire contact simulation system 4 shown in FIG. 1.

[0012] Depending on the implementation method, the tire performance prediction model learning system 1 and the tire performance prediction system 2 may not be built on the same computer system, but may be operated independently. In other words, only the tire performance prediction model learning system 1 may be implemented, or only the tire performance prediction system 2 may be implemented.

[0013] [Tire performance prediction model learning system 1] Fig. 2 is a block diagram showing a tire performance prediction model learning system 1 and a tire performance prediction system 2. As shown in Fig. 2, the tire performance prediction model learning system 1 includes a conversion unit 10, a feature extraction unit 11, a learning unit 12, and an acquisition unit 14. The conversion unit 10 is optional.

[0014] These units 10 to 12, 14 are realized by software and hardware working together as the processor 1a executes a processing routine shown in FIG. 3, which is stored in advance in a computer equipped with the processor 1a, memory 1b, various interfaces, etc. In this embodiment, the processor 1a in one device realizes each unit, but this is not limited to this. For example, the units may be distributed using a network, and multiple processors may execute the processing of each unit. In other words, one or multiple processors execute the processing.

[0015] When an input image 6 based on the ground plane image 5 is input, the feature extraction unit 11 extracts a feature amount 6' of the image. In a configuration in which the conversion unit 10 is not provided, the ground plane image 5 is input to the feature extraction unit 11 as the input image 6. In a configuration in which the conversion unit 10 is provided, the feature extraction unit 11 inputs the input image 6 output by the conversion unit 10. The feature extraction unit 11 may use any algorithm as long as it can extract a feature amount 6' from an image. Examples of such algorithms include neural networks (e.g., SqueezeNet, Alexnet, GoogleNet, ResNet101), discrete cosine transform processing, AutoEncoder, and a configuration in which discrete cosine transform processing and AutoEncoder are used in combination.

[0016] The acquisition unit 14 acquires a teacher data set D1 used for machine learning. The teacher data set D1 is data in which the contact patch image 5, the separate parameters 7 related to the tire of the contact patch image 5, and tire performance are associated with each other. The separate parameters 7 are parameters related to the tire other than the contact patch image 5. The separate parameters 7 are preferably values ​​that cannot be derived from the contact patch shape or contact pressure. By the separate parameters 7 having information different from the information possessed by the contact patch image 5, the prediction accuracy of the tire performance value by the prediction model 13 can be improved. Examples of values ​​that can be derived from the contact patch shape or contact pressure include the area of ​​the contact patch, the aspect ratio of the contact patch, the width direction dimension or circumferential direction dimension of the contact patch, the maximum contact pressure, the minimum contact pressure, the number of longitudinal grooves, the number of lateral grooves, and the number of blocks partitioned by the grooves. The acquisition unit 14 may acquire a teacher data set D2 in which the input image 6 based on the contact patch image 5, the separate parameters 7 relating to the tire of the contact patch image 5, and the tire performance are associated with each other.

[0017] Specifically, the other parameters 7 may include at least one of the tire specifications, compounding identification information of the tread rubber forming the contact surface, physical properties of the tread rubber, and static characteristics of the tire. It is possible to select one of these, or any two or more parameters from these.

[0018] Tire specifications refer to values ​​related to the shape of the tire, such as dimensions. For example, tire specifications include at least one of the following: tire outer diameter, total tire width, pitch number, tire height, and aspect ratio. Tire outer diameter is the outer diameter when the tire is viewed parallel to the axis of rotation. Pitch number is the number of repeating elements that make up the tire in one circumference of the tire. Tire height is the height of the tire cross section, and can be calculated as (tire outer diameter - rim diameter) / 2. Aspect ratio is a value that represents the ratio of tire height to tire cross section width as a percentage.

[0019] The tire specifications are values ​​measured under no-load conditions when the tire is mounted on a standard rim and inflated to the standard internal pressure. A standard rim is a rim that is determined for each tire by the standard system that includes the standard on which the tire is based, for example, the standard rim for JATMA, and the "Measuring Rim" for TRA and ETRTO. The standard internal pressure is the air pressure determined for each tire by the standard system that includes the standard on which the tire is based. For truck and bus tires and light truck tires, it is the maximum air pressure for JATMA, the maximum value listed in the table "TIRE LOAD LIMITS AT VARIOUS COLD INFLATION PRESSURES" for TRA, and "INFLATION PRESSURE" for ETRTO. For passenger car tires, it is usually 180 kPa, but for tires labeled "Extra Load" or "Reinforced", it is 220 kPa.

[0020] The compounding identification information is information for identifying the compounding of the tread rubber. In the first embodiment, the compounding identification information is a compounding number which is a numerical value, but is not limited to this and can be changed in various ways. Once the rubber compounding (i.e., compounding number, compounding identification information) is determined, various physical property values ​​of the rubber (rigidity, rubber hardness, loss tangent, etc.) are determined. In other words, simply including one compounding identification information in the explanatory variables means that multiple physical property values ​​of the rubber are included in the explanatory variables, and it is possible to reduce the dimension of machine learning.

[0021] The physical property value of the tread rubber may include at least one of a storage modulus of the tread rubber, a loss modulus of the tread rubber, and a loss tangent (tan δ) of the tread rubber.

[0022] The static characteristics of the tire may include at least one of longitudinal stiffness, lateral stiffness, longitudinal stiffness, torsional stiffness, and side stiffness.

[0023] The vertical stiffness [N / mm] is calculated by measuring the vertical force and tire deflection (vertical displacement after the tire touches the road surface) measured at the contact surface when the tire is mounted on a standard rim and inflated to the standard internal pressure and pressed vertically against the road surface, and then calculating 2α / (R2-R1) based on the load F0-α, F0+α at the specified load difference α from the standard load F0 and the deflection R1, R2 at that time. Here, α is 490N for passenger car tires, ULT, LT with 8PR or less and less than 15 inches, and LT of the 85 series, 1961N for LT with 8PR or more and 15 inches or more, and 2942N for truck and bus tires.

[0024] The normal load is the load that each standard specifies for each tire in the standard system that includes the standard on which the tire is based. For JATMA, it is the "maximum load capacity," for TRA, it is the maximum value listed in the table "TIRE LOAD LIMITS AT VARIOUS COLD INFLATION PRESSURES," and for ETRTO, it is the "LOAD CAPACITY." If the tire is for passenger cars, it is a load equivalent to 88% of the above load. If the tire is for racing karts, the normal load is 392N.

[0025] Lateral stiffness [N / mm] is calculated by mounting a tire on a standard rim, inflating it to the standard internal pressure, applying a standard load, and moving the tire axially along the road surface, measuring the amount of movement and the axial force, measuring the amount of axial displacement when the axial force is 30% of the load, and dividing the axial force at that time by the lateral tire axial displacement.

[0026] The longitudinal stiffness [N / mm] is calculated by mounting a tire on a standard rim, inflating it to the standard internal pressure, applying a standard load, and measuring the amount of movement and longitudinal force when the tire is moved axially across the road surface, measuring the amount of longitudinal displacement when the longitudinal force is 30% of the load, and dividing the longitudinal force at that time by the amount of longitudinal displacement.

[0027] Torsional rigidity [N / mm] is calculated by mounting a tire on a standard rim, inflating it to the standard internal pressure, applying a standard load, and measuring the angle and torsional torque when the tire is subjected to an angular displacement (set angle) around the vertical axis, calculating the torque at a torsional angle of 2 degrees, and dividing the torsional torque at that time by the torsional angle (2 degrees).

[0028] The side stiffness [N / mm] is measured by mounting a tire on a standard rim, inflating it to the standard internal pressure, constraining the tire tread surface in the radial direction, and in this constrained state, displacing the tire support shaft relative to the tire. In the section where the absolute value of the displacement of the relative displacement increases, the tire displacement amount and the force acting on the tire support shaft in each direction are measured. The maximum load applied at that time is divided by the maximum displacement to calculate the side stiffness value. For example, the testing machine described in Patent No. 6552937 can be used.

[0029] The learning unit 12 uses the teacher data set D1 (or D2) to machine-learn a prediction model 13 so as to output a tire performance value using the image feature amount 6' extracted by the feature extraction unit 11 and the other parameter 7 as explanatory variables. The teacher data set D1 is a set of the other parameter 7 and the tire performance value (X 1 ,X 2 ,…,X N The training data set D2 is data in which the other parameters 7 and the tire performance values ​​(X 1 ,X 2 ,…,X N ) is associated with the prediction model 13. N indicates the number of pieces of teacher data. As the prediction model 13, various models such as Gaussian process regression, linear regression, classification tree, decision tree, random forest, support vector machine, and ensemble tree can be used as long as the prediction model 13 is a supervised machine learning model.

[0030] 2, the explanatory variable group composed of a plurality of explanatory variables input to the prediction model 13 includes one or more explanatory variables related to the image feature amount 6' and one or more explanatory variables related to the other parameters 7. For example, when AlexNet is used in the feature extraction unit 11, the number of explanatory variables related to the image feature amount 6' is 4096. In this case, the explanatory variables related to the other parameters 7 are input to the prediction model 13 as the 4097th and subsequent parameters.

[0031] The conversion unit 10 converts the ground plane image 5 into an input image 6 of a size that can be input to the feature extraction unit 11. In this embodiment, the ground plane image 5 of 875×656 pixels in color or grayscale (8 bit) is converted into the input image 6 of 227×227 pixels in grayscale (8 bit), but this is an example and is not limited to this.

[0032] When converting to a size that can be input to the feature extraction unit 11, it is necessary to avoid destroying the image features that appear in the original contact patch image 5. The size of the contact patch affects performance. For example, the contact patch of a tire with a large tire size appears large in the image, and conversely, the contact patch of a tire with a small tire size appears small in the image. If the image is trimmed so that the size of the contact patch occupies the same size in the trimmed image, there is a risk that the performance value of a tire with a small tire size will be overestimated to be the same as that of a tire with a large tire size, even though the performance value should actually be predicted to be small. Therefore, since the size of the contact patch shape affects the tire performance value, it is necessary to maintain the scale between multiple images.

[0033] For this purpose, the conversion unit 10 has a selection unit 10a, a trimming position determination unit 10b, a trimming unit 10c, and a size change unit 10d. As a premise, all the ground contact surface images 5 are taken under the same shooting conditions and have the same scale. The same shooting conditions means that the distance from the camera 32 to the test road surface 31 is the same, and the zoom value of the camera 32 is the same, as shown in FIG. 1.

[0034] The selection unit 10a selects the contact patch image 5, the other parameters 7, and the tire performance value (X 1~N 5, the contact surface image 50 has the largest contact surface shape.

[0035] The trimming position determination unit 10b determines a trimming position P1 in the image based on the contact patch image 50 selected by the selection unit 10a, as shown in Fig. 5. The larger the contact patch shape in the input image 6, the higher the prediction accuracy. Therefore, in this embodiment, the minimum rectangle including the contact patch shape or a range obtained by expanding the minimum rectangle by a specified number of pixels is set as the trimming position P1. The trimming position P1 determined by the trimming position determination unit 10b is stored in the memory 1b for use by the tire performance prediction system 2.

[0036] The trimming unit 10c uses the trimming positions P1 determined by the trimming position determination unit 10b to trim all of the contact patch images 5 (images 1 to N) of the teacher data set D1 to generate trimmed images.

[0037] The resizing unit 10d resizes each trimmed image generated by the trimming unit 10c to a size that can be input to the feature extraction unit 11, generating an input image 6. The resizing unit 10d resizes the trimmed image without changing the aspect ratio. In the example of FIG. 5, an input image 60 is generated from the ground plane image 50 via a trimmed image (not shown). An input image 61 is generated from the ground plane image 51 via a trimmed image (not shown). An input image 62 is generated from the ground plane image 52 via a trimmed image (not shown). By performing such processing, it is possible to prevent the scales of the ground plane shapes depicted in the input images 60 to 62 from becoming inconsistent.

[0038] The tire performance with improved prediction performance as a result of trimming by the conversion unit 10 while maintaining the aspect ratio within one image and the relative size (scale) between multiple images is as follows. Cornering: CP, SAP, CFmax (maximum cornering force), SAT Traction system: Braking performance on dry roads, braking performance on wet roads, braking performance on icy roads, braking performance on snowy roads, friction performance on ice, friction performance on snow, hydroplaning resistance Noise system: noise performance inside and outside the vehicle, sound radiation performance in individual tire tests Rolling resistance Abrasion resistance Heel and toe wear performance and uneven wear performance are influenced by the ratio and distribution of areas of high contact pressure within the contact patch, the shape of the contact patch, etc., and we believe that trimming or size changes that retain the above-mentioned tire size information are not necessary.

[0039] [Tire performance prediction system 2] 2, the tire performance prediction system 2 includes a conversion unit 20, a feature extraction unit 21, a prediction unit 22, and an acquisition unit 24. The conversion unit 20 is optional, similar to the tire performance prediction model learning system 1.

[0040] When an input image 6 based on the contact patch image 5 is input, the feature extraction unit 21 extracts a feature amount of the image. In a configuration in which the conversion unit 20 is not provided, the contact patch image 5 is input to the feature extraction unit 21 as the input image 6. In a configuration in which the conversion unit 20 is provided, the input image 6 output by the conversion unit 20 is input. The feature extraction unit 21 in the tire performance prediction system 2 has the same configuration as the feature extraction unit 11 in the tire performance prediction model learning system 1.

[0041] The acquisition unit 24 acquires another parameter 7 corresponding to the ground plane image 5 to be predicted.

[0042] The prediction unit 22 uses the prediction model 13 constructed by the tire performance prediction model learning system 1 to input the image feature quantity 6' output by the feature extraction unit 21 and another parameter 7 acquired by the acquisition unit 24 as explanatory variables, and outputs a tire performance value.

[0043] As shown in FIG. 2, the conversion unit 20 converts the contact patch image 5 to be predicted into an input image 6 suitable for input to the feature extraction unit 21. The conversion unit 20 has a trimming unit 20c and a size change unit 20d. The trimming unit 20c trims the contact patch image 5 using a predetermined trimming position P1 to generate a trimmed image. The trimming position P1 is determined by the trimming position determination unit 10b and stored in the memory 1b. The size change unit 20d changes the size of the trimmed image generated by the trimming unit 20c to a size that can be input to the feature extraction unit 21 to generate an input image 6. The size change unit 20d changes the size without changing the aspect ratio of the trimmed image. The trimming unit 20c has the same configuration as the trimming unit 10c in the tire performance prediction model learning system 1. The size change unit 20d has the same configuration as the size change unit 10d in the tire performance prediction model learning system 1.

[0044] [Training method for tire performance prediction model] A tire performance prediction model learning method executed by one or more processors in the tire performance prediction model learning system 1 shown in FIG. 2 will be described with reference to FIG.

[0045] First, by executing steps ST1 to ST4, the conversion unit 10 generates an input image 6 based on the contact patch image 5. Specifically, in step ST1, the selection unit 10a selects the contact patch image 5 (images 1 to N), the other parameters 7, and the tire performance (X 1~N ) and the ground surface image 50 having the largest ground surface shape is selected from the teacher data set D1 associated with the ground surface image 50 having the largest ground surface shape. In the next step ST2, the trimming position determining unit 10b determines a trimming position P1 in the image based on the selected contact surface image 50. In the next step ST3, the trimming unit 10c uses the trimming position P1 determined by the trimming position determination unit 10b to trim all the contact patch images (1 to N) of the teacher data set D1 to generate trimmed images. In the next step ST4, the size change unit 10d changes the size of each trimmed image to a size that can be input to the feature extraction unit 11 without changing the aspect ratio of each trimmed image, thereby generating an input image 6 (60 to 62). As a result, the input image 6 based on the contact patch image 5, the other parameters 7, and the tire performance (X 1~N ) is associated with the training data set D2. In the next step ST5, the feature extraction unit 11 inputs an input image 6 based on the contact patch image 5 representing the tire contact patch shape to the feature extraction unit 11 and extracts a feature amount 6' of the image. In the next step ST6, the learning unit 12 trains the prediction model 13 by machine learning so as to output tire performance values ​​using the extracted image feature amount 6' and another parameter 7 related to the tire in the contact patch image 5 as explanatory variables.

[0046] [Tire performance prediction method] A tire performance prediction method executed by one or more processors in the tire performance prediction system 2 shown in FIG. 2 will be described with reference to FIG.

[0047] First, by executing steps ST101 to ST102, the conversion unit 20 generates an input image 6 based on the ground plane image 5. Specifically, in step ST101, the trimming unit 20c uses a predetermined trimming position P1 to trim the prediction target ground plane image 5 to generate a trimmed image. In the next step ST102, the size change section 20d changes the size of the trimmed image to a size that can be input to the feature extraction section 21 without changing the aspect ratio of the trimmed image, thereby generating an input image 6. In the next step ST103, the feature extraction unit 21 inputs the input image 6 based on the contact patch image 5 representing the tire contact patch shape to the feature extraction unit 21, and extracts the feature amount of the image. In the next step ST104, the acquisition unit 24 acquires the additional parameters 7 relating to the tire of the contact patch image 5. In the next step ST105, the prediction unit 22 outputs tire performance values ​​corresponding to the extracted image features 6' and the other parameters 7 using a prediction model 13 that has been machine-trained to output tire performance values ​​using the extracted image features 6' and the other parameters 7 as explanatory variables.

[0048] <Second embodiment> Hereinafter, a second embodiment of the present disclosure will be described with reference to the drawings.

[0049] As shown in FIG. 6, the tire performance prediction model learning system 1 and the tire performance prediction system 2 of the second embodiment are configured to perform learning or prediction using a smaller number of explanatory variables than the number of explanatory variables of the first embodiment.

[0050] Specifically, in the second embodiment, the prediction model 13 is a decision tree. In the second embodiment, a random forest, which is a decision tree, is used, but any decision tree algorithm is not limited to this. In addition, the learning unit 12 has an importance calculation unit 12a, a variable selection unit 12b, and a re-learning unit 12c. For ease of explanation, an example will be described in which the number of explanatory variables, which are composed of the image feature amount 6' extracted by the feature extraction unit 11 and the other parameters 7 acquired by the acquisition unit 14, is K (K is a natural number equal to or greater than 2).

[0051] The importance calculation unit 12a calculates the importance of each of a plurality of explanatory variables to be input to the machine-learned prediction model 13. Here, the importance is calculated for each of K explanatory variables. The importance is a value that indicates the contribution of an explanatory variable, and can be calculated if a decision tree algorithm is used.

[0052] The variable selection unit 12b selects some of the explanatory variables (L; L < K) to be used from among all the explanatory variables (K) based on the importance calculated by the importance calculation unit 12a. The variable selection unit 12b may select the top predetermined number (L) of explanatory variables with high importance, or may arrange the importance in descending order and select only the explanatory variables with importance whose difference from the next importance is equal to or greater than a threshold value. For example, if the threshold value is 0.2, the difference in importance between the first and second ranks is 0.4, the difference in importance between the second and third ranks is 0.3, and the difference in importance between the third and fourth ranks is 0.15, it may be assumed that the explanatory variables from the first to the third ranks are selected.

[0053] The relearning unit 12c retrains the prediction model 13 so as to output the tire performance value using some of the explanatory variables (L) selected by the variable selection unit 12b as input. As a result, initially, when learning was performed using K explanatory variables, when retraining in the relearning unit 12c, since the number of explanatory variables decreases from K to L, dimensionality reduction can be achieved while ensuring prediction accuracy.

[0054] The operation of the tire performance prediction model learning system 1 is as shown in FIG. 7. Steps ST1 to ST6 are the same as steps ST1 to 6 shown in FIG. 3 of the first embodiment. That is, the learning unit 12 inputs K explanatory variables to the prediction model 13 and trains the prediction model 13. In the next step ST207, after the machine learning using K explanatory variables is completed, the importance calculation unit 12a calculates the importance of each of the plurality of explanatory variables (K) input to the machine-learned prediction model 13. In the next step ST208, the variable selection unit 12b selects some of the explanatory variables (L) to be used from among all the explanatory variables (K) based on the importance calculated by the importance calculation unit 12a. In the next step ST209, the relearning unit 12c retrains the prediction model 13 so as to output the tire performance value using some of the explanatory variables (L) selected by the variable selection unit 12b as input.

[0055] The tire performance prediction system 2 predicts tire performance values ​​by inputting L explanatory variables pre-selected by the variable selection unit 12b to the prediction model 13. Here, an explanatory variable conversion unit may be provided that converts K explanatory variables including the image feature amount 6' output by the feature extraction unit 21 and the separate parameters 7 acquired by the acquisition unit 24 into L explanatory variables with reduced dimensions.

[0056] <Modification> (1) In the first and second embodiments, the tire performance prediction model learning system 1 is provided with the conversion unit 10 and the learning unit 12 uses the teacher data set D2, but the conversion unit 10 can be omitted. Similarly, the conversion unit 20 of the tire performance prediction system 2 can also be omitted.

[0057] As described above, although not particularly limited, the learning method of the tire performance prediction model of the first or second embodiment may be a method executed by one or more processors, and may include a feature extraction step of inputting an input image 6 based on a contact patch image 5 representing the tire contact patch shape to a feature extraction unit 11 to extract image features 6', and a learning step of machine learning the prediction model 13 to output tire performance values ​​using the extracted image features 6' and another parameter 7 related to the tire in the contact patch image 5 as explanatory variables.

[0058] Although not particularly limited, the tire performance prediction model learning system 1 of the first embodiment or the second embodiment includes a feature extraction unit 11 that inputs an input image 6 based on a contact patch image 5 representing a tire contact patch shape and extracts a feature amount of the image; The tire may further include a learning unit 12 that performs machine learning on a prediction model 13 so as to output tire performance values ​​using the extracted image feature amount 6' and another parameter 7 related to the tire in the contact patch image 5 as explanatory variables.

[0059] This makes it possible to provide a prediction model 13 that predicts tire performance values ​​based on feature quantities 6' extracted from the contact patch image 5, making it possible to know the tire performance values ​​from the contact patch image 5. Furthermore, since the explanatory variables input to the prediction model 13 include another parameter 7, it is possible to improve the prediction accuracy of tire performance values ​​compared to the case where only feature quantities 6' based on the contact patch image 5 are input. This is also useful as it reduces the number of prototypes and tests required, and may provide clues as to which elements of the contact patch are related to each tire's performance.

[0060] Although not particularly limited, the tire performance prediction method of the first or second embodiment may be a method executed by one or more processors, and may include a feature extraction step of inputting an input image 6 based on a contact patch image 5 representing the tire contact patch shape to a feature extraction unit 21 and extracting image features 6', a step of acquiring other parameters 7 related to the tire of the contact patch image 5, and a prediction step of outputting tire performance values ​​corresponding to the extracted image features 6' and other parameters 7 using a prediction model 13 machine-learned to output tire performance values ​​using the extracted image features 6' and other parameters 7 as explanatory variables.

[0061] Although not particularly limited, the tire performance prediction system 2 of the first or second embodiment may include a feature extraction unit 21 that inputs an input image 6 based on a contact patch image 5 representing the tire contact patch shape and extracts image features 6', an acquisition unit 24 that acquires other parameters 7 related to the tire of the contact patch image 5, and a prediction unit 22 that outputs tire performance values ​​corresponding to the extracted image features 6' and other parameters 7 using a prediction model 13 that has been machine-learned to output tire performance values ​​using the extracted image features 6' and other parameters 7 as explanatory variables.

[0062] As a result, the prediction model 13 predicts the tire performance values ​​based on the feature values ​​6' extracted based on the contact patch image 5 and the other parameters 7 related to the tire, making it possible to know the tire performance values ​​from the contact patch image 5. Furthermore, since the other parameters 7 are included in the explanatory variables input to the prediction model 13, it is possible to improve the prediction accuracy of the tire performance values ​​compared to the case where only the feature values ​​6' based on the contact patch image 5 are input.

[0063] Although not particularly limited, as in the first or second embodiment, the contact patch image 5 may represent the tire contact patch shape and contact pressure, and the other parameter 7 may be a value that cannot be derived from the tire contact patch shape or contact pressure. According to this, the separate parameters 7 input to the prediction model 13 are values ​​that cannot be derived from the contact surface shape or contact pressure appearing in the contact surface image 5, so it is possible to avoid inputting features that overlap with features appearing in the contact surface image 5 into the prediction model 13. Therefore, compared to the case where overlapping features are input as explanatory variables, it is possible to reduce the number of explanatory variables and reduce dimensions, thereby improving prediction accuracy.

[0064] Although not particularly limited, like the first or second embodiment, the other parameter 7 may include compounding identification information of the tread rubber that forms the contact surface. Once the rubber compounding is determined, various physical property values ​​of the rubber (such as rigidity, rubber hardness, and loss tangent) are determined, and various physical property values ​​are linked to the compounding identification information (such as compounding number) itself. Therefore, once the compounding identification information is determined, even if the multiple physical property values ​​determined by the compounding are not input individually as separate parameters to the prediction model 13, multiple physical property values ​​are input to the prediction model 13, and by using the same compounding identification information for learning, the tendency of the tire performance value is learned taking into account the physical property values ​​of the rubber. Therefore, it is possible to suppress an increase in the number of explanatory variables input to the prediction model 13, improve prediction accuracy, and reduce calculation costs.

[0065] Although not particularly limited, the separate parameters 7 may include at least one of the tire specifications, the physical properties of the tread rubber forming the contact surface, and the static characteristics of the tire, as in the first or second embodiment. This is a preferred example of the separate parameters 7.

[0066] Although not particularly limited, the tire specifications may include at least one of the tire outer diameter, the tire total width, the number of pitches, the tire height, and the aspect ratio.

[0067] Although not particularly limited, the physical property value of the tread rubber forming the ground contact surface may include at least one of the storage modulus of the tread rubber, the loss modulus of the tread rubber, and the loss tangent of the tread rubber. This is a preferable example of the separate parameter 7.

[0068] Although not particularly limited, as in the first or second embodiment, the static characteristics of the tire may include at least one of the longitudinal stiffness of the tire, the lateral stiffness of the tire, the longitudinal stiffness of the tire, the torsional stiffness of the tire, and the side stiffness of the tire.

[0069] Although not particularly limited, as in the learning method for a tire performance prediction model of the second embodiment, the prediction model 13 may be a decision tree, and the method may include a step of calculating the importance of each of a plurality of explanatory variables (K items) to be input to the machine-learned prediction model 13, a step of selecting a portion (L items) of explanatory variables to be used from all the explanatory variables (K items) based on the calculated importance, and a re-learning step of re-machining the prediction model 13 to output tire performance values ​​using the selected portion (L items) of explanatory variables as input.

[0070] Although not particularly limited, in the tire performance prediction model learning system 1 of the second embodiment, the prediction model 13 is a decision tree, and the system may include an importance calculation unit 12a that calculates the importance of each of a plurality of explanatory variables (K items) to be input to the machine-learned prediction model 13, a variable selection unit 12b that selects a portion of explanatory variables (L items) to be used from all the explanatory variables (K items) based on the calculated importance, and a re-learning unit 12c that re-trains the prediction model 13 by machine learning so as to output tire performance values ​​using the selected portion of explanatory variables (L items) as input.

[0071] In this way, the number of explanatory variables can be reduced using the importance of the decision tree algorithm, and prediction accuracy can be improved by dimensionality reduction. In addition, the number of explanatory variables can be reduced, which reduces the calculation cost.

[0072] The program according to this embodiment is a program for causing a computer to execute the above method. By executing these programs, it is possible to obtain the effects of the above-mentioned methods.

[0073] Although the embodiments of the present disclosure have been described above based on the drawings, the specific configurations should not be considered to be limited to these embodiments. The scope of the present disclosure is indicated not only by the description of the above-mentioned embodiments but also by the claims, and further includes all modifications within the meaning and scope equivalent to the claims.

[0074] The structures employed in the above-described embodiments may be employed in any other embodiment. The specific configurations of the components are not limited to the above-described embodiments, and may be modified in various ways without departing from the spirit of the present disclosure.

[0075] For example, in the claims, the specification, and the drawings, the execution order of each process such as the operations, procedures, steps, and stages in the apparatus, system, program, and method shown can be realized in any order, unless the output of the previous process is used in the subsequent process. Regarding the flow in the claims, the specification, and the drawings, even if terms such as "first" and "next" are used for convenience in the description, it does not mean that it is essential to execute in this order.

[0076] Each part shown in FIGS. 2 and 6 is realized by executing a predetermined program on one or more processors, but each part may be configured by a dedicated memory or a dedicated circuit. In the system 1 of the above embodiment, each part is implemented in the processor 1a of one computer, but each part may be distributed and implemented in a plurality of computers or in the cloud. That is, the above method may be executed by one or more processors.

[0077] The system 1 includes a processor 1a. For example, the processor 1a can be a central processing unit (CPU), a microprocessor, or any other processing unit capable of executing computer-executable instructions. The system 1 also includes a memory 1b for storing the data of the system 1. In one example, the memory 1b includes a computer storage medium, including RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, DVD or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store desired data and can be accessed by the system 1.

Description of Reference Numerals

[0078] 1... Tire performance prediction model learning system, 11... Feature extraction unit, 12... Learning unit, 13 Prediction model, 2... Tire performance prediction system, 21... Feature extraction unit, 22... Prediction unit, 24... Acquisition unit, 7... Another parameter

Claims

1. A method executed by one or more processors, comprising: a feature extraction step of inputting an input image based on a contact patch image representing a tire contact patch shape to a feature extraction unit, and extracting feature values ​​of an image constituted by a plurality of values ​​obtained by converting the input image by the feature extraction unit; a learning step of machine learning a prediction model so as to output a tire performance value using the extracted image feature amount and another parameter related to the tire of the contact patch image as explanatory variables; Includes A method for learning a tire performance prediction model, wherein the feature extraction unit includes any one of a neural network, a discrete cosine transform processor, an AutoEncoder, and a configuration in which the discrete cosine transform processor and the AutoEncoder are used in combination.

2. A method executed by one or more processors, comprising: a feature extraction step of inputting an input image based on a contact patch image representing a tire contact patch shape to a feature extraction unit, and extracting feature values ​​of an image constituted by a plurality of values ​​obtained by converting the input image by the feature extraction unit; obtaining additional parameters related to the tire of the contact patch image; a prediction step of outputting the tire performance value corresponding to the extracted image feature amount and the other parameter by using a prediction model that has been machine-learned to output a tire performance value using the extracted image feature amount and the other parameter as explanatory variables; Including, A tire performance prediction method, wherein the feature extraction unit includes any one of a neural network, a discrete cosine transform processor, an AutoEncoder, and a configuration in which the discrete cosine transform processor and the AutoEncoder are used in combination.

3. 3. The method of claim 1, wherein the contact patch image represents the tire contact patch shape and contact pressure, and the further parameter is a value that cannot be derived from the tire contact patch shape or the contact pressure.

4. The method according to any one of claims 1 to 3, wherein the separate parameter includes compounding identification information of a tread rubber forming the contact surface.

5. The method according to any one of claims 1 to 4, wherein the other parameters include at least one of tire specifications, physical properties of a tread rubber forming the contact surface, and static characteristics of the tire.

6. the predictive model is a decision tree; 2. The method according to claim 1, comprising: a step of calculating the importance of each of a plurality of explanatory variables to be input to the machine-learned prediction model; a step of selecting a portion of explanatory variables to be used from among all explanatory variables based on the calculated importance; and a re-learning step of re-machining the prediction model so as to output the tire performance value using the selected portion of explanatory variables as an input.

7. A system comprising one or more processors for carrying out the method according to any one of claims 1 to 6.

8. A program causing one or more processors to execute the method according to any one of claims 1 to 6.

Images