Tire design support device, design method for tire, generation device for learned model, and generation method for learned model

Abstract

To provide a technique capable of designing a target tire further efficiently.SOLUTION: A tire design support device has a derivation unit which accepts input of a value, corresponding to an evaluation value obtained by making a driver's function evaluation of at least one of riding comfortability and steering stability of a vehicle equipped with tires, to a learned first machine-learning model, and then derives an output from the first machine-learning model. The output from the first machine-learning model is at least one of a value corresponding to measured values of the tires and a value corresponding to design parameters of the tires.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a tire design assistance device, a tire design method, a trained model generation device, and a trained model generation method. [Background technology]

[0002] Patent Document 1 discloses a tire evaluation method that improves the correlation between bench tests and sensory evaluations in actual vehicle evaluations. In Patent Document 1, a plurality of tires to be evaluated are subjected to bench tests to obtain predetermined measurement results. The measurement results are plotted on a graph and are associated with the sensory evaluation results of specific items in an actual vehicle to analyze how the measurement results affect the sensory evaluation results of the items, and design elements of the evaluated tires that should be reviewed are identified. [Prior art documents] [Patent documents]

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

[0004] As disclosed in Patent Document 1, a certain degree of correlation can be found between tire bench test values ​​and sensory evaluation values ​​in vehicle tests. However, this alone does not enable designing a tire that achieves a target sensory evaluation value. In order to achieve the target sensory evaluation value, it is necessary to review the tire design based on the correlation, and to conduct bench tests and vehicle tests again for verification. For this reason, a technology that allows for more efficient tire design has been desired.

[0005] An object of the present invention is to provide a technique that enables a targeted tire to be designed more efficiently. [Means for solving the problem]

[0006] A tire design assistance device according to a first aspect includes a derivation unit that inputs a value corresponding to a sensory evaluation by a driver of at least one of ride comfort and handling stability of a vehicle equipped with a tire to a trained first machine learning model, and derives an output from the first machine learning model. The output from the first machine learning model is at least one of a value corresponding to a measurement value of the tire and a value corresponding to a design parameter of the tire.

[0007] A tire design support device according to a second aspect is the tire design support device according to the first aspect, wherein the first machine learning model has been trained for one of the drivers.

[0008] A tire design support device according to a third aspect is the tire design support device according to the first or second aspect, wherein an output from the first machine learning model is a value corresponding to a measurement value of the tire, and the tire design support device further includes an estimation unit that estimates design parameters of the tire based on the derived value corresponding to the measurement value of the tire.

[0009] A tire design support device according to a fourth aspect is a tire design support device according to any one of the first aspect to the third aspect, wherein the estimation unit estimates design parameters of the tire by inputting a value corresponding to the evaluation value and a value corresponding to a measurement value of the tire into a trained second machine learning model and deriving an output from the second machine learning model, and the output from the second machine learning model is a value corresponding to the design parameters of the tire.

[0010] A tire design method according to a fifth aspect includes inputting a value corresponding to a sensory evaluation by a driver of at least one of ride comfort and handling stability of a vehicle equipped with the tire into a trained first machine learning model, and deriving an output from the first machine learning model. The output from the first machine learning model is at least one of a value corresponding to a measurement value of the tire and a value corresponding to a design parameter of the tire.

[0011] A trained model generation device according to a sixth aspect includes a learning unit that trains a machine learning model using a training dataset that combines an evaluation value obtained by a driver's sensory evaluation of at least one of ride comfort and handling stability of a vehicle equipped with a tire and correct answer data, the correct answer data being at least one of measurement values ​​of the tire and design parameters of the tire.

[0012] A method for generating a trained model according to a seventh aspect includes training a machine learning model using a training dataset in which an evaluation value obtained by a driver's sensory evaluation of at least one of ride comfort and handling stability of a vehicle equipped with the tire is combined with correct answer data, the correct answer data being at least one of measurement values ​​of the tire and design parameters of the tire. Effect of the Invention

[0013] According to the present invention, it is possible to estimate at least one of the tire measurement values ​​and the tire design parameters from the target evaluation value of the sensory evaluation using a machine learning model, thereby supporting efficient tire design that achieves the target evaluation value. [Brief description of the drawings]

[0014] [Figure 1] FIG. 2 is a block diagram showing the electrical configuration of the tire design assistance device according to the embodiment. [Diagram 2] A diagram showing an example configuration of a first machine learning model. [Diagram 3] 4 is a flowchart showing an example of a tire design method using the design support device. [Figure 4] FIG. 1 is a block diagram showing the electrical configuration of a trained model generation device according to an embodiment. [Diagram 5] 1 is a flowchart showing an example of a method for generating a trained model using a generation device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, a tire design assistance device, a tire design method, a trained model generation device, and a trained model generation method according to embodiments of the present invention will be described.

[0016] <1. Overview> FIG. 1 is a block diagram showing an electrical configuration of a tire design support device 1 (hereinafter, also simply referred to as "support device 1") according to one embodiment of the present invention. The support device 1 is mainly used in tire design, and supports tire design to realize a target ride comfort and handling stability in particular. Here, the evaluation of the ride comfort and handling stability is performed by an actual vehicle test by one or more professional testers (hereinafter, also referred to as "drivers"). The driver performs a sensory evaluation to compare the sensation obtained by driving a vehicle equipped with a reference tire, also referred to as a control tire, with the sensation obtained by driving the same vehicle equipped with a prototype tire. That is, the driver evaluates at least one of the ride comfort and handling stability of a vehicle equipped with a prototype tire based on the result of evaluating at least one of the ride comfort and handling stability of a vehicle equipped with a control tire.

[0017] The evaluation values ​​of the ride comfort and steering stability by the sensory evaluation are considered to have some correlation with the measured values ​​obtained by subjecting the tire itself to a bench test. The measured values ​​can be controlled by the design parameters of the tire. However, it is considered that there are measured values ​​that are correlated with multiple items of the sensory evaluation, and it is not easy to identify the degree of correlation between each measured value for all items of the sensory evaluation. For this reason, although the measured values ​​of the tire can be a guide to some extent in considering the design parameters of the tire, they cannot be said to be a sufficient guide to achieve the target evaluation value, and it is still necessary to repeat the prototype and the actual vehicle test. In addition, according to the study by the inventors, there are cases where multiple measured values ​​between different tires are approximately the same value, but even in such cases, the evaluation values ​​of the sensory test by the same driver do not necessarily match. From this, it is considered that the design parameters of the tire also have a correlation with the evaluation values ​​of the sensory test. In view of such a current situation, the support device 1 supports efficient design of a tire to achieve the target ride comfort and steering stability.

[0018] <2. Configuration of the support device> The support device 1 is a general-purpose computer in terms of hardware, and is realized as, for example, a desktop personal computer, a laptop personal computer, a tablet, or a smartphone. The support device 1 is manufactured by installing a program 132 into a general-purpose computer from a computer-readable storage medium 133 such as a CD-ROM or a USB memory, or via a network. The program 132 causes the support device 1 to execute the operations described below.

[0019] The support 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 each other via a bus line and can communicate with each other. 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 10D described later. The input unit 12 can be configured with a mouse, a keyboard, a touch panel, etc., and accepts operations on the support device 1. The display unit 11 and the input unit 12 may both be configured with the same touch panel display.

[0020] The storage unit 13 can be composed of 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 defining the first machine learning model 16 and the second machine learning model 17. The first machine learning model 16 and the second machine learning model 17 have been trained by a generating device 1A described later.

[0021] The control unit 10 may be configured with a central processing unit (CPU), a graphics processing unit (GPU), a ROM, a RAM, and the like. 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, an estimation unit 10C, and a screen generation unit 10D. The acquisition unit 10A acquires an input value to the first machine learning model 16 via the input unit 12, the communication unit 14, and the like. The derivation unit 10B inputs the acquired input value to the first machine learning model 16 and derives an output from the first machine learning model 16. The estimation unit 10C estimates design parameters of the tire based on the output derived by the derivation unit 10B. The screen generation unit 10D generates a screen that displays various information to the user, such as the output derived by the derivation unit 10B and the estimation result by the estimation unit 10C. The communication unit 14 functions as a communication interface that transmits and receives data to and from devices external to the support device 1.

[0022] <3. Configuration of the first machine learning model> 2 is a diagram showing an example configuration of the first machine learning model 16. Below, a description will be given of the first machine learning model 16. The first machine learning model 16 according to this embodiment is a model constructed (trained) for one driver.

[0023] The input to the first machine learning model 16 is a value corresponding to an evaluation value of at least one of the ride comfort and the driving stability of a vehicle equipped with the tire, which is evaluated by the driver, and is typically a target value of the sensory evaluation that the driver wants to obtain when performing an actual vehicle test. This target value is set for each predetermined item related to the ride comfort and driving stability of the vehicle. For the ride comfort of the vehicle, the predetermined items include, for example, bodily sensations expressed as "bumpy," "shaky," "lumpy," "tingly," etc., and the impact (shock) felt when going over a protrusion or a foreign object. For the driving stability of the vehicle, the predetermined items include, for example, convergence, response, responsiveness, grip, and hardness.

[0024] The evaluation value (or target value) can be expressed, for example, by a graded index. For example, if the index for each item of the sensory evaluation of the control tire is "3," the upper limit of each item is "5," and the lower limit is "1," the evaluation of each item compared to the control tire can be expressed as a number from 1 to 5.

[0025] The first machine learning model 16 has already learned, by a learning method described below, the correspondence between the evaluation values ​​obtained by the driver's sensory evaluation of predetermined items in the actual vehicle test and the measurement values ​​of the tires mounted on the vehicle used in the actual vehicle test. Therefore, the output from the first machine learning model 16 is the measurement values ​​of the tires predicted to achieve the target values ​​of the sensory evaluation.

[0026] The tire measurement values ​​are not particularly limited as long as they are data obtained by measuring the tire alone, and examples thereof include lateral spring constant, vertical spring constant, dynamic spring constant, resonance frequency of the tire alone, protrusion reaction force, cornering power (CP), maximum cornering force, self-aligning torque (SAT), CTP(N), H(α), G(α), etc. Among these, the lateral spring constant, vertical spring constant, dynamic spring constant, resonance frequency of the tire alone, and protrusion reaction force are considered to have a relatively large effect on the sensory evaluation of ride comfort. In addition, the dynamic spring constant, CP, maximum cornering force, and self-aligning torque are considered to have a relatively large effect on the sensory evaluation of steering stability. For example, the ratio of the CP of the rear wheels to the CP of the front wheels, or the ratio of the CP of the front wheels to the CP of the rear wheels, has a strong correlation with the item of steady yaw rate gain in the sensory evaluation. These measurement values ​​can be measured using a known tire performance tester or the like that reproduces and measures the behavior of a vehicle equipped with the tire to be measured on a stand.

[0027] The first machine learning model 16 according to this embodiment is a model based on a neural network (NN), and includes an input layer 160 having a predetermined number of nodes, an intermediate layer 161, and an output layer 162. The input layer 160 is a layer for reading input target values, and has the same number of nodes as the number of input target values. The input values ​​are output as they are from each node of the input layer 160, and are input to each node of the first intermediate layer 161, which is connected immediately after the input layer 160.

[0028] Each node in the first hidden layer 161 calculates a value by multiplying the input by a weight w and adding a bias b. Then, the calculated value is converted by an activation function and output to the node in the subsequent hidden layer 161. The input here is one or more values ​​read in the input layer 160. The weight w has been adjusted for each value passed from each node in the input layer 160 to each node in the first hidden layer 161 by a learning process described later. That is, for each node in the first hidden layer, weights w corresponding to the number of values ​​input thereto are optimized. In addition, biases b corresponding to the number of nodes are also optimized by a learning process described later.

[0029] The activation function is a function for realizing a nonlinear transformation. There are no particular limitations on the activation function, 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.

[0030] The values ​​output from each node of the first intermediate layer 161 are input to each node of the subsequent intermediate layer 161. The subsequent intermediate layer 161 also has a predetermined number of nodes, and the weight w and bias b of each node are optimized by a learning process described later. Like the first intermediate layer 161, each node outputs a value calculated using the corresponding weight w and bias b and an activation function for an input value. That is, in the intermediate layer 161, the same process as the first intermediate layer 161 is repeated in principle. However, in the last intermediate layer 161, a linear transformation is performed on the input using the optimized weight w and bias b, but a nonlinear transformation using an activation function is not performed. That is, the linearly transformed values ​​output from each node of the last intermediate layer 161 are input to each node of the output layer 162. The number of intermediate layers 161 is preferably 2 or more, but is not particularly limited and can be set appropriately.

[0031] The output layer 162 has nodes in the same number as the tire measurement values, and each node typically outputs the input value as is. The output from each node is a value corresponding to the tire measurement value, and represents the tire measurement value predicted to achieve the input target value of the sensory evaluation. This output becomes the output of the first machine learning model 16.

[0032] In the support device 1 according to this embodiment, a second machine learning model 17 is used to further estimate design parameters of the tire based on the target value of the sensory evaluation input to the first machine learning model 16 and the measurement values ​​of the tire output from the first machine learning model 16. The second machine learning model 17 will be described below.

[0033] <4. Configuration of the second machine learning model> The second machine learning model 17, like the first machine learning model 16, is constructed for one driver based on a neural network. The second machine learning model 17 includes an input layer, one or more intermediate layers, and an output layer. The input to the second machine learning model 17 is the target value of the sensory evaluation and the predicted tire measurement values, and the input layer has the same number of nodes as the number of these values. The number of intermediate layers connected to the input layer and the number of nodes each intermediate layer has can be set appropriately. The same applies to the processing performed in each intermediate layer as the description of the first machine learning model 16, so a description thereof will be omitted.

[0034] The output from the second machine learning model 17 is a predetermined number of tire design parameters predicted to achieve the above-mentioned target value of the sensory evaluation and the predicted tire measurement value, and the output layer has nodes of the same number as the number of these values. The tire design parameters can be appropriately selected from the following exemplary values. Design parameters particularly related to the rigidity of the tread portion include the hardness of the material of the cap portion constituting the tread portion, the hardness of the material of the base portion constituting the tread portion, the thickness of the tread portion (the thickness at a predetermined position or the maximum thickness), the ratio of the thickness of the cap portion to the thickness of the base portion, the structure and material of the reinforcing material of the tread portion, and the like. Examples of the reinforcing material of the tread portion include a breaker and a band. In radial tires, a breaker is a member also called a belt, which tightens the carcass in the circumferential direction to increase the rigidity of the tread portion. Examples of the structure of the breaker include the width of the breaker and the angle it makes with respect to the circumferential direction of the tire. The band is a cloth-like member that is incorporated into the tread portion as necessary to increase the rigidity of the tread portion, and examples of the structure of the band include the number of band layers, the position in the tire width direction where the band is incorporated, and the like. The material of the reinforcing material can be quantified by a number that is assigned in advance according to the type of material, for example. The angle of the breaker relative to the circumferential direction of the tire is strongly correlated with the ratio of the CP of the rear wheels to the CP of the front wheels, or the ratio of the CP of the front wheels to the CP of the rear wheels. In other words, the steady yaw rate gain, the ratio of the CP of the rear wheels to the CP of the front wheels (or the ratio of the CP of the front wheels to the CP of the rear wheels), and the angle of the breaker are examples of sensory evaluation items, bench values, and design parameters that are correlated with each other.

[0035] In addition, design parameters particularly related to the rigidity of the side portion include the hardness of the material of the sidewall portion, the thickness of the sidewall portion (thickness at a predetermined position or maximum thickness), the material and structure of the carcass of the sidewall portion, the material of the reinforcing material of the sidewall portion, the hardness of the material of the apex, the height of the apex, etc. The carcass is a reinforcing material that usually extends over the entire tire, but contributes more to the rigidity of the sidewall portion than to the rigidity of the tread portion. The material of the carcass can be quantified, for example, by a number that is assigned in advance according to the type of material. Examples of the structure of the carcass include the number of layers, the thickness of the cord that constitutes the carcass, and the turn-up height. Examples of the reinforcing material of the sidewall portion include a strip apex, a steel filler, and a nylon filler. The material of the reinforcing material can be quantified, for example, by a number that is assigned in advance according to the type of material. In addition, the presence or absence of such a reinforcing material may be quantified and included in the rigidity data.

[0036] Furthermore, the design parameters may include a profile of a mold for molding the tire. Examples of the mold profile include an approximate radius of curvature (tread radius) of the inner surface of the mold corresponding to the tread surface, the width of the inner surface corresponding to the tread surface, and tread pattern rigidity. The tread radius is related to the contact width of the pneumatic tire and determines the magnitude of input from the road surface. The tread pattern rigidity represents the overall rigidity of the contact surface calculated from the pattern formed by the rubber blocks of the tread portion.

[0037] <5. Tire design method> Next, a method for designing a tire using the assisting device 1 will be described with reference to Fig. 3. Fig. 3 is a flowchart showing an example of a method for designing a tire using the assisting device 1.

[0038] First, the acquisition unit 10A acquires a target value of the sensory evaluation that the driver wants to obtain when performing an actual vehicle test (step S1). The acquisition by the acquisition unit 10A may be performed, for example, by a user of the support device 1 (typically a tire development designer) via the input unit 12, 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 target value in the RAM or storage unit 13.

[0039] Next, the derivation unit 10B inputs the acquired target value to the first machine learning model 16 and derives an output from the first machine learning model 16 (step S2). This results in the acquisition of tire measurement values ​​predicted to achieve the input target value of the sensory evaluation. In addition, in step S2, the screen generation unit 10D may generate a screen displaying a list of the acquired tire measurement values, and cause the display unit 11 to display this.

[0040] Next, the estimation unit 10C inputs the target value of the sensory evaluation acquired in step S1 and the tire measurement value derived in step S2 to the second machine learning model 17, and derives an output from the second machine learning model 17 (step S3). This allows estimation of tire design parameters that realize the input target value of the sensory evaluation and the tire measurement value.

[0041] Next, the screen generator 10D generates a result screen displaying the tire design parameters obtained in step S3, and displays this on the display unit 11 (step S4). The result screen may include the inputted target values ​​of the sensory evaluation and tire measurement values. A user who checks the result screen can easily consider the range of tire measurement values ​​to be targeted in the design. In addition, the user can easily select design parameters that are considered more suitable for achieving the tire measurement values.

[0042] <6.Generation device> FIG. 4 is a block diagram showing an electrical configuration of a trained model generating device 1A (hereinafter, also simply referred to as "generating device 1A") according to one embodiment of the present invention. The generating device 1A trains a machine learning model using first and second training data sets 18 and 19 described later, and generates first and second machine learning models 16 and 17 according to this embodiment. The generating device 1A is a general-purpose computer as hardware, and is realized as, for example, a desktop personal computer, a laptop personal computer, a tablet, or a smartphone. The generating device 1A is manufactured by installing a program 134 into a general-purpose computer from a computer-readable storage medium 136 such as a CD-ROM or a USB memory, or via a network. The program 134 causes the generating device 1A to execute operations described later.

[0043] The generating device 1A includes a control unit 100, a display unit 11A, an input unit 12A, a storage unit 13A, and a communication unit 14A. The configurations of these units 10A to 14A are the same as the configurations of the units 10 to 14 of the support device 1, so that the description thereof will be omitted here. The control unit 100 can be configured with a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a ROM, a RAM, and the like. The control unit 100 reads and executes a program 134 in the storage unit 13A, thereby virtually operating as an acquisition unit 10E and a learning unit 10F. The acquisition unit 10E acquires a first learning data set 18 and a second learning data set 19 via the communication unit 14A and the like. The learning unit 10F trains a machine learning model, as described later.

[0044] <7. How to generate a trained model> Hereinafter, a method for generating the trained first and second machine learning models 16 and 17 using the generation device 1A will be described with reference to Fig. 5. Fig. 5 is a flowchart showing an example of the generation method.

[0045] First, the acquisition unit 10E acquires the first learning dataset 18 and the second learning dataset 19 prepared in advance and stores them in the storage unit 13A (step S11). The first learning dataset 18 is a learning dataset for generating the first machine learning model 16. The second learning dataset 19 is a learning dataset for generating the second machine learning model 17. The acquisition by the acquisition unit 10E may be performed via a storage medium such as a CD-ROM or a USB memory, or may be performed by receiving data via a network. The acquisition unit 10E divides the first learning dataset 18 and the second learning dataset 19 into a training dataset for parameter adjustment and a test dataset for accuracy verification in advance and stores them. The ratio between the two can be set appropriately.

[0046] As described above, the first and second machine learning models 16 and 17 are generated for one driver, respectively, and therefore the first and second learning datasets 18 and 19 are also prepared for the driver. The first learning dataset 18 is a dataset of evaluation values ​​obtained by the driver performing an actual vehicle test and sensory evaluation of the ride comfort and handling stability of the vehicle, and correct answer data. The actual vehicle test is performed on a vehicle equipped with a plurality of different tires including a control tire. The correct answer data is the measurement values ​​of the plurality of different tires including the control tire used in the actual vehicle test. The second learning dataset 19 is a dataset of the evaluation values ​​of the actual vehicle test of the driver, the measurement values ​​of the plurality of different tires including the control tire used in the actual vehicle test, and correct answer data. The evaluation values ​​and measurement values ​​in the second learning dataset 19 are common to the evaluation values ​​and measurement values ​​in the first learning dataset 18. The correct answer data is the design parameters of the plurality of different tires including the control tire used in the actual vehicle test.

[0047] Next, the learning unit 10F divides each of the training data sets into a predetermined number of subsets to create a plurality of subsets of the training data set (step S12). The predetermined number is the number of data to be input to the first learning-stage machine learning model 16 and the second learning-stage machine learning model 17 in the next step S13, and can be set appropriately.

[0048] Next, the learning unit 10F selects one of the subsets of the first learning dataset 18, inputs the evaluation value included in the selected subset to the first machine learning model 16 in the learning stage, and derives an output (step S13). This output is a value corresponding to the supervised data combined with the input evaluation value, and in this embodiment, a value corresponding to the measurement value of the tire. In addition, in step S13, the learning unit 10F selects one of the subsets of the second learning dataset 19, inputs the evaluation value and the measurement value included in the selected subset to the second machine learning model 17 in the learning stage, and derives an output. This output is a value corresponding to the supervised data combined with the input evaluation value and the measurement value, and in this embodiment, a value corresponding to the design parameter of the tire.

[0049] Next, the learning unit 10F adjusts the parameters that define the first machine learning model 16 and the second machine learning model 17 in the learning stage so that the value of an error function that represents the error between the output derived in step S13 and the ground truth data combined with the value input in step S13 is minimized (step S14). As a result, the weight coefficients and biases in the intermediate layers of the first machine learning model 16 and the second machine learning model 17 are updated.

[0050] Next, the learning unit 10F judges whether or not one epoch of learning has been completed (step S15). In this embodiment, when steps S13 and S14 have been completed once for each subset created in step S12, it is judged that one epoch of learning has been completed. If it is judged that one epoch of learning has not been completed (NO), the learning unit 10F repeats steps S13 to S15 using a subset that has not yet been used. On the other hand, if it is judged that one epoch of learning has been completed (YES), step S16 is executed.

[0051] In the next step S16, the learning unit 10F judges whether or not learning of all epochs is completed. The total number of epochs is not particularly limited and can be set appropriately. If it is judged that learning of all epochs is not completed (NO), the learning unit 10F repeats steps S13 to S15 while selecting subsets in the same order as the previous epoch. On the other hand, if it is judged that learning of all epochs is completed (YES), the learning unit 10F stores the latest parameters in the memory unit 13A and sets these as parameters that define the trained first machine learning model 16 and second machine learning model 17. Through the above procedure, the trained first and second machine learning models 16 and 17 are generated.

[0052] Note that the learning unit 10F may input test data to the first and second machine learning models 16 and 17 at that time each time one epoch of learning is completed, and calculate the error between the outputs of those models and the test data relative to the correct answer data. In this way, if it is considered that the error in the output of the first machine learning model 16 or the second machine learning model 17 has converged within a predetermined range before the completion of learning for all epochs, the learning of that machine learning model may be terminated at that time.

[0053] <8.Features> According to the support device 1, the first machine learning model 16, which has learned the tendency of sensory evaluation by a specific driver, estimates at least the measurement value of a tire that is likely to achieve an evaluation value (target value) by the specific driver. As will be described later, some knowledge has been obtained about the correlation between tire measurement values ​​and tire design parameters. This makes it possible to determine the range of tire measurement values ​​that are likely to achieve a target evaluation value, and makes it easy to estimate the design parameters for designing such a tire.

[0054] Moreover, according to the above-mentioned support device 1, the second machine learning model 17 estimates design parameters for designing a tire that realizes the measurement value estimated by the first machine learning model 16. As described above, according to the study by the present inventors, even tires whose tire measurement values ​​can be said to be roughly the same may have different design parameters. In such a case, according to conventional knowledge, there may be a plurality of design parameters (for the same design parameter) corresponding to the measurement value estimated by the first machine learning model 16. In this regard, the second machine learning model 17 learns the tendency of sensory evaluation by a specific driver, and therefore estimates design parameters that are more likely to achieve a target evaluation value.

[0055] The user can produce a prototype tire based on at least one of the tire measurement values ​​and design parameters estimated by the support device 1. A real vehicle test can be performed on the prototype tire by a specific driver to verify whether a target evaluation value is achieved. If the target evaluation value is not achieved, it can be verified whether the cause is due to an estimation error in the first machine learning model 16 or the second machine learning model 17, or due to the conditions under which the real vehicle test was performed (for example, the physical condition of the driver at that time, etc.).

[0056] The generation device 1A can construct a first machine learning model 16 and a second machine learning model 17 based on a data set of evaluation values ​​from actual vehicle tests by a specific driver, tire measurement values, and tire design parameters that have been accumulated to date.

[0057] <9. 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 modified examples shown below can be combined as appropriate.

[0058] (1) In the above embodiment, a model based on NN is used as the first machine learning model 16 and the second machine learning model 17, but the machine learning model is not limited to this, and other machine learning models and models combining these may be used, such as a model using a support vector machine (SVM), a convolutional neural network (CNN) model, a K-NN model, clustering, k-means, a decision tree, or a logistic regression model. Furthermore, even in the case of a model based on NN, the layer structure, etc. can be changed as appropriate. The learning methods of the first machine learning model 16 and the second machine learning model 17 are also not limited to the above embodiment, and each can be changed as appropriate.

[0059] (2) The input to the first machine learning model 16 may be a value corresponding to an evaluation value obtained by a driver's sensory evaluation of only either the ride comfort or the handling stability of the vehicle. The output of the first machine learning model 16 may include a value corresponding to a design parameter of the tire instead of or in addition to a value corresponding to the measurement value of the tire. In this case, the ground truth data of the first training dataset 18 is at least one of the measurement value of the tire and the design parameter of the tire.

[0060] (3) The estimation unit 10C may estimate the design parameters of the tire without using the second machine learning model 17. That is, the estimation unit 10C may estimate the design parameters of the tire using a table showing the correspondence between the design parameters of the tire and the measured values ​​that the tire can take for the design parameters, or a regression equation between the measured values ​​and the design parameters that is constructed without using machine learning. The above table or regression equation can be created based on the results of past actual vehicle tests. Furthermore, the estimation unit 10C itself may be omitted. The generation of the second machine learning model 17 by the generation device 1A may also be omitted.

[0061] (4) In the above embodiment, the support device 1 is configured as a single device, but the functions of the units 10A-10D and the storage unit 13 may be distributed among multiple devices. Similarly, the functions of the units 10E-10F and the storage unit 13A of the generation device 1A may be distributed among multiple devices. Furthermore, the support device 1 and the generation device 1A may be configured as a single device.

[0062] (5) The control unit 10 of the support device 1 or the control unit 100 of the generation device 1A may be configured to include a CPU, a GPU, a vector processor, an FPGA (Field Programmable Gate Array), an ASIC (Application Specific Integrated Circuit), other chips dedicated to artificial intelligence, etc. Furthermore, the operation of the control unit 10 or the control unit 100 may be executed by one or more processors.

[0063] 1 Support equipment 1A generator 10 Control section 10A Acquisition Department 10B Derivation part 10C Estimation part 10F Learning Section 16 First Machine Learning Model 17 Second Machine Learning Model 18 First training dataset 19 Second training data set

Claims

1. a derivation unit that inputs a value corresponding to an evaluation value obtained by a driver's sensory evaluation of at least one of the ride comfort and the handling stability of a vehicle equipped with the tire into a trained first machine learning model, and derives an output from the first machine learning model; Equipped with an output from the first machine learning model is at least one of a value corresponding to a measurement of the tire and a value corresponding to a design parameter of the tire; Tire design support device.

2. The first machine learning model has been trained for one of the drivers. The tire design support device according to claim 1.

3. an output from the first machine learning model is a value corresponding to a measurement of the tire; an estimation unit for estimating design parameters of the tire based on the derived values ​​corresponding to the measurements of the tire; Further comprising: The tire design support device according to claim 1 or 2.

4. the estimation unit inputs a value corresponding to the evaluation value and a value corresponding to the measurement value of the tire into a trained second machine learning model, and derives an output from the second machine learning model to estimate design parameters of the tire; An output from the second machine learning model is a value corresponding to a design parameter of the tire. The tire design support device according to claim 3.

5. inputting a value corresponding to a sensory evaluation by a driver of at least one of ride comfort and handling stability of a vehicle equipped with the tire into a trained first machine learning model; Deriving an output from the first machine learning model. Including, an output from the first machine learning model is at least one of a value corresponding to a measurement of the tire and a value corresponding to a design parameter of the tire; How tires are designed.

6. A learning unit that trains a machine learning model using a learning dataset that combines a sensory evaluation value of at least one of the ride comfort and driving stability of a vehicle equipped with the tire by a driver and correct answer data. Equipped with The correct answer data is at least one of measurement values ​​of the tire and design parameters of the tire. A device for generating trained models.

7. A machine learning model is trained using a learning data set in which an evaluation value obtained by a driver's sensory evaluation of at least one of the ride comfort and the handling stability of a vehicle equipped with the tire is combined with correct answer data; and Including, The correct answer data is at least one of measurement values ​​of the tire and design parameters of the tire. How to generate a trained model.

Images