Device, method, and program for estimating friction performance of tire

Abstract

To provide a technology of estimating friction performance of a tire.SOLUTION: A friction performance estimation device for a tire includes a derivation part for inputting the input data including a contact pressure distribution of a ground surface of an objective tire to be an estimation object, a groove depth of the objective tire, and viscoelasticity of a rubber member constituting the objective tire to a learned machine learning model to derive output data from the learned machine learning model. The output data corresponds to a frictional coefficient of the objective tire on a wet road surface.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a technique for estimating the frictional performance of a tire. [Background technology]

[0002] Patent Document 1 discloses a method for accurately evaluating the friction performance of a rubber member. Patent Document 1 verifies that the evaluation results of the rubber member obtained by this method correlate with the results of a running test of an actual vehicle equipped with a tire made of the same rubber member with high accuracy. Therefore, the method described in Patent Document 1 makes it possible to clearly evaluate the grip performance of a tire on an actual road surface without actually manufacturing the tire. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-169933 Summary of the Invention [Problem to be solved by the invention]

[0004] Incidentally, one aspect of a tire's grip performance is its coefficient of friction on wet roads. Generally, tests to measure a tire's coefficient of friction on wet roads require pre-processing, which alone requires a large amount of work. In this regard, Patent Document 1 does not verify the correlation between the evaluation results of the rubber member and the tire's coefficient of friction on wet roads. In other words, no method has been proposed to estimate a tire's coefficient of friction on wet roads.

[0005] An object of the present invention is to provide a technique for estimating the coefficient of friction of a tire on a wet road surface. [Means for solving the problem]

[0006] A tire friction performance estimation device according to a first aspect includes a derivation unit that inputs input data, including a contact pressure distribution on the contact patch of a target tire to be estimated, a groove depth of the target tire, and viscoelasticity of a rubber member constituting the target tire, into a trained machine learning model, and derives output data from the trained machine learning model. The output data corresponds to the friction coefficient of the target tire on a wet road surface.

[0007] A tire frictional performance estimation device according to a second aspect is the tire frictional performance estimation device according to the first aspect, wherein the contact pressure distribution is configured as image data.

[0008] A tire frictional performance estimation device according to a third aspect is the tire frictional performance estimation device according to the first or second aspect, wherein the viscoelasticity of the rubber member is a tangent loss measured in accordance with ASTM D4065 and ASTM D4473 under conditions of a static strain of 10%, a dynamic strain of 2.5%, a frequency of 10 Hz, and a temperature of 0°C.

[0009] A tire friction performance estimation method according to a fourth aspect is a tire friction performance estimation method executed by one or more computers, and includes inputting input data, including a contact pressure distribution of a contact patch of a target tire to be estimated, a groove depth of the target tire, and viscoelasticity of a rubber member constituting the target tire, into a trained machine learning model, and deriving output data from the trained machine learning model. The output data corresponds to the friction coefficient of the target tire on a wet road surface.

[0010] A tire friction performance estimation program according to a fifth aspect causes one or more computers to input input data, including the contact pressure distribution of the contact patch of a target tire to be estimated, the groove depth of the target tire, and the viscoelasticity of a rubber member constituting the target tire, into a trained machine learning model, and derive output data from the trained machine learning model, the output data corresponding to the friction coefficient of the target tire on a wet road surface.

[0011] A trained model generation device according to a sixth aspect includes a storage unit and a learning unit. The storage unit stores a training dataset including a plurality of datasets each combining first data including a contact pressure distribution of a tire's contact patch, a groove depth of the tire, and viscoelasticity of a rubber member constituting the tire, and second data including an actual measurement value of the friction coefficient of the tire on a wet road surface. The learning unit inputs the first data to a machine learning model, derives output data from the machine learning model, and adjusts parameters defining the machine learning model so as to reduce an error between the output data and the second data combined with the first data.

[0012] A method for generating a trained model according to a seventh aspect is a method for generating a trained model executed by one or more computers, and includes: storing a training dataset including a plurality of datasets combining first data including the contact pressure distribution of the contact surface of a tire, the groove depth of the tire, and the viscoelasticity of a rubber component constituting the tire, and second data including an actual measured value of the friction coefficient of the tire on a wet road surface; inputting the first data into a machine learning model, deriving output data from the machine learning model, and adjusting parameters defining the machine learning model so as to reduce an error between the output data and the second data combined with the first data.

[0013] A program for generating a trained model according to an eighth aspect causes one or more computers to execute the following steps: store a training dataset including multiple datasets combining first data including the contact pressure distribution of the contact surface of a tire, the groove depth of the tire, and the viscoelasticity of the rubber components that constitute the tire, and second data including an actual measurement value of the friction coefficient of the tire on a wet road surface; input the first data into a machine learning model, derive output data from the machine learning model, and adjust parameters that define the machine learning model so as to reduce an error between the output data and the second data combined with the first data. [Effects of the Invention]

[0014] According to the present invention, the coefficient of friction of a tire on a wet road surface can be estimated based on data that is relatively easy to obtain. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 2 is a block diagram showing the electrical configuration of the estimation device according to an embodiment. [Figure 2] FIG. 1 is a diagram showing an example of the configuration of a machine learning model. [Figure 3] 10 is a flowchart showing the flow of an estimation process executed by the estimation device. [Figure 4] FIG. 1 is a block diagram showing the electrical configuration of a trained model generation device according to an embodiment. [Figure 5] 10 is a flowchart showing the flow of a generation process executed by a generation device. [Figure 6] FIG. 1 is a diagram illustrating a method for measuring friction performance. [Figure 7] 10 shows an example of data on ground pressure distribution according to a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0016] Hereinafter, a tire friction performance estimation device, an estimation method, and an estimation program, as well as a trained model generation device, a generation method, and a generation program according to an embodiment of the present invention will be described.

[0017] <1. Estimation device> FIG. 1 is a block diagram showing the electrical configuration of an estimation device 1 according to an embodiment of the present invention. The estimation device 1 is a device that estimates the coefficient of friction (hereinafter also referred to as "WET μ") of a tire T to be estimated on a predetermined wet road surface based on measurement data of the tire T. WET μ is an important factor in evaluating the braking performance of the tire T. However, as will be described later, its actual measurement requires a large number of steps, so a method that can easily estimate it is desired. The estimation device 1 is a general-purpose computer in terms of hardware, and is realized as, for example, a desktop PC, a laptop PC, a tablet, or a smartphone. The estimation device 1 is manufactured by installing a program 132 into the general-purpose computer from a computer-readable storage medium 133, such as a CD-ROM or USB memory, or via a network. The program 132 causes the estimation device 1 to perform the operations described below.

[0018] The estimation device 1 includes a control unit 10, a display unit 11, an input unit 12, a storage unit 13, and a communication unit 14. These units 10 to 14 are connected to one another via a bus line and are capable of communicating with one another. The display unit 11 can be configured with a liquid crystal display, an organic EL display, a plasma display, a touch panel display, etc., and displays a screen generated by a screen generation unit 10C, which will be described later. The input unit 12 can be configured with a mouse, a keyboard, a touch panel, etc., and accepts operations on the estimation device 1. The display unit 11 and the input unit 12 may both be configured with the same touch panel display.

[0019] The storage unit 13 can be configured with a non-volatile memory such as a hard disk and a flash memory. In addition to storing a program 132, the storage unit 13 also stores parameters that define a trained machine learning model 2A (hereinafter also simply referred to as the "trained model 2A"). The trained model 2A has been trained by a generation device 1X, which will be described later.

[0020] The control unit 10 can be configured with, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a ROM, a RAM, etc. The control unit 10 reads and executes a program 132 in the storage unit 13, thereby virtually operating as an acquisition unit 10A, a derivation unit 10B, and a screen generation unit 10C. The acquisition unit 10A acquires input data for the trained model 2A via the input unit 12, the communication unit 14, etc. The derivation unit 10B inputs the acquired input data into the trained model 2A and derives output data from the trained model 2A. The screen generation unit 10C generates a screen that displays various information to the user, such as the output data derived by the derivation unit 10B. The communication unit 14 functions as a communication interface for transmitting and receiving data to and from devices external to the estimation device 1.

[0021] <2. Configuring the trained model> Fig. 2 is a diagram showing an example of the configuration of a trained model 2A. As shown in Fig. 2, the trained model 2A of this embodiment has a first feature extraction unit 20 and a second feature extraction unit 21 connected in this order. The first feature extraction unit 20 has a configuration based on a convolutional neural network (CNN), and the second feature extraction unit 21 has a configuration based on a neural network (NN).

[0022] [Input data] The input data of the trained model 2A according to this embodiment includes the contact pressure distribution on the contact surface of the tire T, the groove depth of the tire T, and the viscoelasticity of the rubber members that make up the tire T. As will be described later, the contact pressure distribution is directly input to the first feature extraction unit 20. The groove depth of the tire T and the viscoelasticity of the rubber members that make up the tire T are directly input to the second feature extraction unit 21. Each piece of data will be explained below.

[0023] The contact pressure distribution is, for example, the contact pressure value of the tire T in a stationary state at each coordinate on a plane defined by the width direction (parallel to the rotation axis of the tire T) and the longitudinal direction (orthogonal to the rotation axis of the tire T) of the contact surface of the tire T. In other words, the contact pressure distribution has a predetermined number (H × W) of contact pressure values ​​arranged vertically and horizontally. The contact pressure values ​​can be obtained using a known tire contact pressure measuring device. The tire contact pressure measuring device is not particularly limited as long as it is a device that can measure the contact pressure (kPa) two-dimensionally. For example, there is a device in which a number of pressure sensors are arranged two-dimensionally along the width direction and the longitudinal direction in a measurement area where the tire T is pressed, and each sensor detects the contact pressure corresponding to each coordinate; a device in which the tire is pressed against a hard transparent plate, the contact area between the transparent plate and the tread is irradiated, the amount of reflected light corresponding to each coordinate is measured and converted into a contact pressure value; and a device in which the tire is pressed against a pressure-sensitive mat and the external change of the pressure-sensitive mat is measured.

[0024] The above-mentioned contact pressure distribution is, for example, zero outside the range of the contact patch of the tire T or in groove portions. Therefore, the contact pressure distribution automatically includes information about the tread pattern of the tire T. The tread pattern of the tire T is a factor that determines the friction coefficient of each rubber block and the drainage performance of the tire T, and like the static contact pressure distribution, has a strong correlation with WET μ. Note that the contact pressure distribution according to this embodiment is measured using the tire contact pressure measuring device described above for a tire T that is attached to a predetermined rim such as a regular rim, adjusted to a predetermined internal pressure, and to which a predetermined load is applied. This measurement can be performed statically. Therefore, among data indicating the characteristics of the tire T, the contact pressure distribution is relatively easy to obtain and has relatively little measurement variation.

[0025] The groove depth of the tire T is the depth (mm) of the main groove of the tire T, measured using a groove gauge in this embodiment. The groove depth of the tire T is a factor that influences the drainage performance of the tire T, and has a strong correlation with WET μ.

[0026] In this embodiment, the viscoelasticity of the rubber member constituting tire T is the loss tangent (tanδ) measured on a sample of the rubber material of a specified size under conditions of a static strain of 10%, a dynamic strain of 2.5%, a frequency of 10 Hz, and a temperature of 0°C in accordance with ASTM D4065 and ASTM D4473. tanδ is defined as (loss modulus / storage modulus). Here, the vibrations experienced by a tire due to braking on a wet road surface are high-frequency vibrations of 10,000 to 1,000,000 Hz. Therefore, it is known that braking performance on wet road surfaces improves as tanδ at high frequencies increases. Since it is known that a temperature-frequency conversion can be applied to the tanδ of a rubber member, the above can be rephrased as saying that braking performance on wet road surfaces improves as tanδ at 0°C increases. From the above, the tanδ at 0°C of the rubber member constituting tire T has a strong correlation with the wet μ of tire T.

[0027] [First feature extraction unit] The first feature extraction unit 20 is a part that extracts features specific to the tire T from the contact pressure distribution. The first feature extraction unit 20 has an input layer 200, a convolutional layer 201, a pooling layer 202, and a fully connected layer 203. The input layer 200 is a layer for reading two-dimensional data having the above-mentioned H × W values. The subsequent convolutional layer 201 applies a filter matrix to the two-dimensional data read by the input layer 200 and outputs a feature map. The size and stride of the filter matrix (the interval at which the filter matrix is ​​applied to the input data) can be set appropriately as hyperparameters. Meanwhile, the values ​​included in the filter matrix are parameters adjusted in the learning process described below. In the above process, a nonlinear transformation using an activation function such as a ReLU function may be defined. The feature map output from the convolutional layer 201 is input to the subsequent pooling layer 202.

[0028] The pooling layer 202 is a layer for further reducing the size of the feature map and improving invariance to coordinate shifts and noise in the ground pressure distribution. In this embodiment, the pooling layer 202 performs max pooling, which uses the maximum value within a predetermined range of the feature map. However, instead of this, average pooling may be performed.

[0029] The output from the pooling layer 202 is further input to the subsequent convolutional layer 201. In the subsequent convolutional layer 201, a matrix filter having parameters adjusted to the input is also applied, and the same processing as described above is performed. In this way, each time the processing in the convolutional layer 201 is repeated several times, global features are gradually extracted from local features of the ground pressure distribution. The number of convolutional layers 201 and the subsequent pooling layers 202 can be set as appropriate. In addition, the pooling layer 202 following the convolutional layer 201 can also be omitted as appropriate.

[0030] The fully connected layer 203 performs processing to associate the features extracted by the convolution layer 201 and the pooling layer 202 with WETμ. There may be multiple fully connected layers 203. Each layer of the fully connected layers 203 has a predetermined number of nodes. The processing executed at each node is the same as the processing in the intermediate layer 211 described below, and the weight and bias of each node are parameters adjusted in the learning processing described below. The number of nodes in the final fully connected layer 203 is the number of factors predicted to particularly affect WETμ among the features of the ground pressure distribution. The number of nodes in each layer can be set as appropriate.

[0031] [Second feature extraction unit] The second feature extraction unit 21 has an input layer 210, followed by one or more intermediate layers 211, and an output layer 212. The input layer 210 is a layer for reading input data to the second feature extraction unit 21. Specifically, the input layer 210 reads one or more values ​​output by the first feature extraction unit 20, the tread depth of the tire T, and tan δ.

[0032] Each of the subsequent intermediate layers 211 has multiple nodes. The weights and biases of each node are parameters adjusted by a learning process described below. Each node outputs a value calculated using the corresponding weights and biases and an activation function in response to an input value. In principle, this process is repeated in the multiple intermediate layers 211. However, in the final intermediate layer 211, a linear transformation is performed on the input using the adjusted weights and biases, but a nonlinear transformation using an activation function is not performed. The activation function is not particularly limited, and any known activation function such as a tanh function, a sigmoid function, a ReLU function, a step function, an ELU function, or a Softmax function can be used.

[0033] The output layer 212 has one node. The value output from this node corresponds to the WET μ of the tire T. That is, the output from the output layer 212 becomes the output data from the trained model 2A.

[0034] <3. Estimation method> 3 is a flowchart showing the flow of a method for estimating the WET μ of the tire T using the estimation device 1. This estimation method will be described below.

[0035] First, the acquisition unit 10A acquires input data including the contact pressure distribution of the tire T, the depth of the main groove, and the viscoelasticity (step S1). As described above, these input data are actual measurements obtained using the tire T and the rubber members that constitute the tire T. The acquisition by the acquisition unit 10A may be performed, for example, by a user of the estimation device 1 (typically, a tire developer / designer) via the input unit 12, or via a storage medium such as a CD-ROM or USB memory, or by receiving data via a network. The acquisition unit 10A stores the acquired input data in RAM or the storage unit 13.

[0036] Next, the derivation unit 10B inputs the acquired input data to the trained model 2A and derives an output from the trained model 2A (step S2). As a result, the value of WET μ estimated to be possessed by the tire T is obtained.

[0037] Next, the screen generation unit 10C generates a result display screen that displays the estimated WET μ value, and causes the display unit 11 to display the result display screen (step S3). By checking the result display screen, the user can easily determine whether the tire T has the desired WET μ. In addition, the result display screen can be compared with the actual measured value of the WET μ of the tire T to examine the variation in the actual measurement.

[0038] <4. Learning Device> FIG. 4 is a block diagram showing the electrical configuration of a trained model generation device 1X (hereinafter also simply referred to as "generation device 1X") according to one embodiment of the present invention. The generation device 1X trains a machine learning model 2 using a training dataset 130 described below, and generates a trained model 2A according to this embodiment. The generation device 1X is a general-purpose computer in terms of hardware, and is realized as, for example, a desktop PC, a laptop PC, a tablet, or a smartphone. The generation device 1X is manufactured by installing a program 132X into a general-purpose computer from a computer-readable storage medium 133X such as a CD-ROM or USB memory, or via a network. The program 132X causes the generation device 1X to perform the operations described below.

[0039] The generation device 1X includes a control unit 10X, a display unit 11X, an input unit 12X, a storage unit 13X, and a communication unit 14X. The configurations of these units 10X to 14X are the same as the configurations of the units 10 to 14 of the estimation device 1, and therefore will not be described here. The control unit 10X may be configured with a central processing unit (CPU), a graphics processing unit (GPU), a ROM, a RAM, etc. The control unit 10X reads and executes a program 132X in the storage unit 13X, thereby virtually operating as an acquisition unit 10D and a learning unit 10E. The acquisition unit 10D acquires a training dataset 130 via the communication unit 14X, etc. The learning unit 10E trains a machine learning model 2, as will be described later.

[0040] <5. How to generate a trained model> 5 is a flowchart showing an example of a generation method using the generation device 1X. In this generation method, a training dataset 130 prepared in advance is used. Below, a method for preparing the training dataset 130 will be described, followed by a method for generating a trained model 2A that has been trained using the generation device 1X.

[0041] The training dataset 130 includes multiple datasets of first data and second data combined with the first data. The first data and second data are created using actual measurements of the same tire. The training dataset 130 is a collection of datasets of first data and second data for multiple different tires. The first data includes the contact pressure distribution of the tire's contact patch, the tire's groove depth, and the viscoelasticity (tan δ) of the rubber material that constitutes the tire. This contact pressure distribution is two-dimensional data having H × W values, similar to the contact pressure included in the input data described above. The method for acquiring the first data is the same as the method for acquiring the input data for tire T, so a description thereof will be omitted.

[0042] The second data is a measured value of WET μ of the same tire as the tire of the first data. While the method for measuring WET μ is not particularly limited, in this embodiment, the measurement is performed using a traction tester 3 as shown in FIG. 6 based on the standard to which the test tire T0 conforms. FIG. 6 is a schematic diagram of the traction tester 3 as viewed from below. The traction tester 3 is a vehicle-type tire testing machine, and includes a test wheel hub 30, a water spray nozzle 31, a tank 32, a test wheel brake 33, and a load application unit 34. The test wheel hub 30 is fitted with a test tire T0 mounted on a predetermined rim and adjusted to a predetermined air pressure. A six-component force meter (not shown) is installed on or near the test wheel hub 30 to constantly detect longitudinal forces, lateral forces, loads, and the like applied to the test tire T0.

[0043] The test wheel brake 33 applies a braking force to the shaft to which the test wheel hub 30 is connected. The load application unit 34 is composed of an air cylinder or the like installed inside the vehicle body, and is capable of applying a predetermined wheel load to the test tire T0 mounted on the test wheel hub 30. The water spray nozzle 31 is connected to a tank 32 installed inside the vehicle body, and is capable of spraying a predetermined amount of water onto the road surface. Before the test tire T0 is mounted on the test wheel hub 30, a break-in run is performed under predetermined conditions using another vehicle. Furthermore, a break-in run is also performed in advance on the test road surface on which the traction tester 3 will run, using another tire mounted thereon, under predetermined conditions.

[0044] Wet μ is measured using a traction tester 3 equipped with a test tire T0 while the tire is traveling on a test road surface. First, the traction tester 3 travels on the test road surface at a constant speed with a predetermined load applied to the test tire T0. When the traction tester 3 reaches a designated point on the test road surface during travel, the water spray nozzle 31 begins spraying water onto the test road surface. Water is sprayed so that the area where the test tire T0 is traveling is within a predetermined water depth range. After that, when the traction tester 3 reaches a designated braking point, the test wheel brake 33 applies a braking force to the test tire T0 via the shaft. From the time the test wheel brake 33 applies the braking force until the test tire T0 stops rotating, a data set is acquired of the longitudinal force applied to the test tire T0 and the normal force (load) from the road surface. The longitudinal force at a given time is divided by the normal force at that time to obtain the kinetic friction coefficient μ at that time. In this embodiment, WET μ is the maximum value of the dynamic friction coefficient μ.

[0045] Referring again to Figure 5, first, the acquisition unit 10D acquires the learning dataset 130 prepared as described above and stores it in the storage unit 13X (step S11). At this time, the acquisition unit 10D divides the learning dataset 130 into a training dataset and a test dataset according to a predetermined ratio and stores them.

[0046] Next, the learning unit 10E divides the stored training data set into a predetermined number of data sets, and creates multiple subsets each including the same number of data sets (step S12). The predetermined number is the number of data sets to be continuously input to the machine learning model 2 in the next step S13, and can be set as appropriate.

[0047] Next, the learning unit 10E selects one of the subsets, inputs the first data included in the selected subset to the machine learning model 2, and derives output data (step S13). The machine learning model 2 is a machine learning model that is the basis of the trained model 2A, and has a layer structure as shown in FIG.

[0048] Next, the learning unit 10E adjusts the parameters defining the machine learning model 2 so that the value of an error function representing the error between the output data derived in step S13 and the second data combined with the first data input in step S13 is minimized (step S14). More specifically, the learning unit 10E adjusts the elements of the filter matrix, weight values, and bias values ​​in each layer of the machine learning model 2 by gradient descent or the like, and updates the parameters defining the machine learning model 2 stored in the storage unit 13X.

[0049] Next, the learning unit 10E determines whether or not one epoch of learning has been completed (step S15). In this embodiment, it is determined that one epoch of learning has been completed when steps S13 and S14 have been performed once for each subset created in step S12. If it is determined that one epoch of learning has not been completed (NO), the learning unit 10E repeats steps S13 to S15 using a subset that has not yet been used. On the other hand, if it is determined that one epoch of learning has been completed (YES), step S16 is executed.

[0050] In the following step S16, the learning unit 10E determines whether learning of all epochs has been completed. The total number of epochs is not particularly limited and can be set as appropriate. If it is determined that learning of all epochs has not been completed (NO), the learning unit 10E executes step S12 again to create a subset of combination patterns different from the previous one. Then, steps S13 to S15 are repeated. On the other hand, if it is determined that learning of all epochs has been completed (YES), the learning unit 10E stores the latest parameters in the memory unit 13X and sets these as parameters that define the trained model 2A. Through the above procedure, the trained model 2A is generated.

[0051] Note that the learning unit 10E may input the first data of the test dataset into the current machine learning model 2 each time one epoch of learning is completed, or at an appropriate time, and calculate the error between the output data and the second data of the test dataset. As a result, if it is considered that the error in the output data of the machine learning model 2 has converged within a predetermined range before the completion of learning for all epochs, the learning of the machine learning model 2 may be terminated at that point.

[0052] <6. Features> The estimation device 1 can estimate the WET μ of tire T using a trained model 2A that has learned the relationship between previously acquired tire contact pressure distribution, main groove depth, and viscoelasticity of the rubber member of the tire and the WET μ of the tire. Measurement data on the contact pressure distribution, main groove depth, and viscoelasticity of the rubber member has relatively little variance and requires relatively little labor to obtain. Therefore, it is expected that the WET μ of tire T can be estimated with a certain degree of accuracy. Furthermore, by comparing the WET μ of tire T estimated by the estimation device 1 with the WET μ of tire T actually measured using the above method, it is possible to verify the variance in the WET μ of the actual measurement. Furthermore, by estimating the WET μ of worn tire T, it is possible to predict changes in friction characteristics due to tire wear.

[0053] <7. Variations> Although one embodiment of the present invention has been described above, the present invention is not limited to the above embodiment, and various modifications are possible without departing from the spirit of the present invention. The gist of the following modifications can be combined as appropriate.

[0054] (1) The configuration of the trained model 2A is not limited to that of the above embodiment. For example, other machine learning models, such as models using support vector machines (SVMs), clustering, k-means, decision trees, and logistic regression models, and models combining these, may be used. Furthermore, the layer structure and the number of nodes may be changed as appropriate. The training method of the machine learning model 2A is also not limited to that of the above embodiment and may be changed as appropriate.

[0055] (2) The contact pressure distribution in the input data may be configured as image data such as that shown in FIG. 7. The image data in FIG. 7 is an example of data output from a known tire contact pressure measuring device, in which each pixel of an image representing the tire contact patch is colored in a color corresponding to a predetermined range of contact pressure values. The image data may be in color or grayscale. The same applies to the contact pressure distribution in the first data.

[0056] (3) In the above embodiment, the estimation device 1 is configured as a single device, but the functions of the units 10A to 10C and the storage unit 13 may be distributed across multiple devices. Similarly, the functions of the units 10D to 10E and the storage unit 13X of the generation device 1X may be distributed across multiple devices. Furthermore, the estimation device 1 and the generation device 1X may be configured as a single device.

[0057] (4) The control unit 10 of the estimation device 1 or the control unit 10X of the generation device 1X may be configured to include a CPU, a GPU, a vector processor, an FPGA (Field Programmable Gate Array), an ASIC (Application Specific Integrated Circuit), or other chips dedicated to artificial intelligence. Furthermore, the operation of the control unit 10 or the control unit 10X may be executed by one or more processors.

[0058] (5) The viscoelasticity in the input data and the first data is not limited to tan δ obtained by the above method. Any value that represents the viscoelasticity of the tire T that is correlated with the braking performance of the tire T on a wet road surface can be used in addition to or instead of the above tan δ.

[0059] 1 Estimation device 1X generator 10A Acquisition Department 10B Derivation part 10E Learning Department 2A Pre-trained model T Target tires

Claims

1. a derivation unit that inputs input data including the contact pressure distribution of the contact surface of a target tire to be estimated, the groove depth of the target tire, and the viscoelasticity of a rubber member that constitutes the target tire into a trained machine learning model, and derives output data from the trained machine learning model; Equipped with The output data corresponds to a friction coefficient of the target tire on a wet road surface. Tire friction performance estimation device.

2. The ground pressure distribution is configured as image data. The tire friction performance estimation device according to claim 1.

3. The viscoelasticity of the rubber member is a loss tangent measured according to ASTM D4065 and ASTM D4473 under the conditions of a static strain of 10%, a dynamic strain of 2.5%, a frequency of 10 Hz, and a temperature of 0°C. The tire friction performance estimation device according to claim 1 or 2.

4. 1. A method for estimating tire frictional performance executed by one or more computers, comprising: Inputting input data including the contact pressure distribution of the contact surface of a target tire to be estimated, the groove depth of the target tire, and the viscoelasticity of a rubber member constituting the target tire into a trained machine learning model, and deriving output data from the trained machine learning model. Including, The output data corresponds to a friction coefficient of the target tire on a wet road surface. A method for estimating tire friction performance.

5. Inputting input data including the contact pressure distribution of the contact surface of a target tire to be estimated, the groove depth of the target tire, and the viscoelasticity of a rubber member constituting the target tire into a trained machine learning model, and deriving output data from the trained machine learning model. on one or more computers, The output data corresponds to a friction coefficient of the target tire on a wet road surface. Tire friction performance estimation program.

6. a storage unit that stores a learning dataset including a plurality of datasets each combining first data including a contact pressure distribution of a contact surface of a tire, a groove depth of the tire, and viscoelasticity of a rubber member that constitutes the tire, and second data including an actual measurement value of a friction coefficient of the tire on a wet road surface; a learning unit that inputs the first data into a machine learning model, derives output data from the machine learning model, and adjusts parameters that define the machine learning model so that an error between the output data and the second data combined with the first data is reduced; Equipped with A device for generating trained models.

7. A method for generating a trained model, executed by one or more computers, comprising: storing a training dataset including a plurality of datasets each combining first data including a contact pressure distribution of a contact surface of a tire, a groove depth of the tire, and viscoelasticity of a rubber member constituting the tire, and second data including an actual measurement value of a friction coefficient of the tire on a wet road surface; inputting the first data into a machine learning model, deriving output data from the machine learning model, and adjusting parameters defining the machine learning model so as to reduce an error between the output data and the second data combined with the first data; Including, How to generate a trained model.

8. storing a training dataset including a plurality of datasets each combining first data including a contact pressure distribution of a contact surface of a tire, a groove depth of the tire, and viscoelasticity of a rubber member constituting the tire, and second data including an actual measurement value of a friction coefficient of the tire on a wet road surface; inputting the first data into a machine learning model, deriving output data from the machine learning model, and adjusting parameters defining the machine learning model so as to reduce an error between the output data and the second data combined with the first data; on one or more computers, A program for generating trained models.

Images