JP7807638B2 - Tire characteristic estimation device, tire characteristic estimation method, estimation model generation method, estimation model generation device, and program - Google Patents

Description

This disclosure relates to a tire characteristic estimation device, a tire characteristic estimation method, an estimation model generation method, an estimation model generation device, and a program.

In tire design and development, the tire structure is designed, including the number, dimensions, and materials of the components (e.g., belts) that make up the tire, and the tire characteristics (e.g., cornering power, rolling resistance, etc.) that will be obtained from that designed structure are predicted. The predicted characteristics are then evaluated to determine whether they meet the requirements for use. Patent Document 1 listed below discloses an estimation model for estimating tire characteristics (physical values) from multiple rubber materials, fillers, etc.

Japanese Patent Application Laid-Open No. 2020-149423

Tire structures vary greatly depending on the type. For example, the number of belts varies depending on the type of tire. For tires with significantly different structures, estimating tire characteristics using only one estimation model will not provide high estimation accuracy.

(1) The tire characteristic estimation device proposed in this disclosure includes storage means in which a first estimation model and a second estimation model are stored. Each of the first estimation model and the second estimation model is an estimation model for estimating characteristic data representing tire characteristics from specification data including data on multiple structural items that are multiple items representing the tire structure. The estimation device also includes data acquisition means for acquiring specification data for a tire to be estimated, and estimation means for selecting one of the first estimation model and the second estimation model based on the multiple structural items for which the acquired specification data is provided, inputting the acquired specification data into the selected estimation model, and outputting characteristic data for the tire to be estimated. This estimation device can improve the accuracy of estimating the characteristics of multiple tires with different structural items.

(2) In the estimation device of (1), the first estimation model and the second estimation model may each include a coefficient estimation model that outputs one or more coefficients according to the specification data. The estimation means inputs the acquired specification data into the coefficient estimation model of the selected estimation model, and calculates characteristic data of the tire to be estimated using a relational equation defined by the one or more coefficients output from the coefficient estimation model. This makes it possible to estimate tire characteristics using, for example, an empirically obtained relational equation.

(3) In the estimation device of (1), the first estimation model and the second estimation model may each be an estimation model for estimating the characteristic data from the specification data and evaluation condition data indicating evaluation conditions for tire characteristics. The estimation device may include means for acquiring evaluation condition data for the characteristics to be estimated for the estimation target tire. The estimation means may estimate the characteristic data of the estimation target tire using the acquired specification data and the acquired evaluation condition data. This enables estimation that takes into account the evaluation conditions.

(4) In the estimation device of (1), the first estimation model and the second estimation model may each include a coefficient estimation model that outputs one or more coefficients according to the specification data, and may be estimation models for estimating the characteristic data from the specification data and evaluation condition data that indicates evaluation conditions for tire characteristics. The estimation device may include means for acquiring evaluation condition data for the characteristics to be estimated for the tire to be estimated. The estimation means may input the acquired specification data into the coefficient estimation model of the selected estimation model, and calculate the characteristic data of the tire to be estimated using a relational equation defined by the one or more coefficients output from the selected estimation model and the acquired evaluation condition data. This makes it possible to estimate tire characteristics using the relational equation, for example, when the relationship between evaluation conditions and tire characteristics has been empirically obtained.

(5) In the estimation device of (1), the storage means may store a plurality of first estimation models corresponding to a plurality of evaluation conditions for tire characteristics, and a plurality of second estimation models corresponding to the plurality of evaluation conditions. The estimation means may select one or more estimation models from the plurality of first estimation models and the plurality of second estimation models based on a plurality of structural items for which the acquired specification data is provided. This makes it possible to estimate characteristics according to the evaluation conditions.

(6) In the estimation device of (1), each of the first estimation model and the second estimation model may be a model generated from a tire data group including a plurality of individual tire data. Each individual tire data may include the specification data and the characteristic data. The tire data group may include a first data group and a second data group, each including the plurality of individual tire data. The individual tire data of the first data group and the individual tire data of the second data group may differ in at least some of the plurality of structural items for which the specification data is provided. The first estimation model may be a model generated using the plurality of individual tire data included in the first data group as training data, and the second estimation model may be a model generated using the plurality of individual tire data included in the second data group as training data. This estimation device can improve the accuracy of estimating the characteristics of a plurality of tires with different structural items.

(7) The tire characteristic estimation method proposed in this disclosure is an estimation method that estimates tire characteristics using a first estimation model and a second estimation model. Each of the first estimation model and the second estimation model is an estimation model for estimating characteristic data representing tire characteristics from specification data including data on multiple structural items that are multiple items representing the tire structure. The estimation method includes a data acquisition step of acquiring specification data for a tire to be estimated, and an estimation step of selecting one of the first estimation model and the second estimation model based on the multiple structural items for which the acquired specification data is provided, inputting the acquired specification data into the selected estimation model, and outputting characteristic data for the tire to be estimated. This estimation method can improve the accuracy of estimating the characteristics of multiple tires with different structural items.

(8) The program proposed in this disclosure is a program that causes a computer to function as an estimation device that estimates tire characteristics using a first estimation model and a second estimation model. Each of the first estimation model and the second estimation model is an estimation model for estimating characteristic data representing tire characteristics from specification data including data on multiple structural items that are multiple items that describe the tire structure. The program causes a computer to function as data acquisition means that acquires specification data for a tire to be estimated, and estimation means that selects one of the first estimation model and the second estimation model based on the multiple structural items for which the acquired specification data is provided, inputs the acquired specification data into the selected estimation model, and outputs characteristic data for the tire to be estimated.

(9) The estimation model generation method proposed in this disclosure includes a data acquisition step of acquiring a tire data group including multiple individual tire data. Each individual tire data includes characteristic data representing tire characteristics and specification data including data on multiple structural items representing the tire structure. The generation method also includes a model generation step of generating an estimation model using the multiple individual tire data to estimate the characteristic data based on the specification data of a tire to be estimated. The tire data group includes a first data group and a second data group, each including the multiple individual tire data. The first data group and the second data group differ in at least some of the multiple structural items. In the model generation step, a first estimation model is generated using the multiple individual tire data included in the first data group as training data, and a second estimation model is generated using the multiple individual tire data included in the second data group as training data. Using the estimation model generated by this method, it is possible to improve the accuracy of estimating the characteristics of multiple tires with different structural items.

(10) The generation method of (9) may further include a data extraction step. The first data group and the second data group may include, as part of the plurality of structural items, structural items common to the first data group and the second data group. The data extraction step may extract data including specification data of the common structural items and the characteristic data from the tire data group. The model generation step may include a source model generation step of generating a source model based on the specification data of the common structural items using the extracted data as training data, and a target model generation step of generating the first estimation model using the source model and the plurality of individual tire data in the first data group, and generating the second estimation model using the source model and the plurality of individual tire data in the second data group. This allows for improved estimation accuracy of the estimation model obtained from one of the two data groups, even when the number of individual tire data included in that one data group is small.

(11) In the generation method of (9), the first estimation model and the second estimation model may each include a coefficient estimation model that outputs one or more coefficients that define a relational equation for calculating the characteristic data. Using the estimation model generated by this method, tire characteristics can be estimated, for example, using an empirically obtained relational equation.

(12) In the generation method of (9), each of the plurality of individual tire data may include evaluation condition data indicating evaluation conditions for tire characteristics. The first estimation model and the second estimation model may be models for estimating the characteristic data using the specification data and the evaluation condition data. Using the estimation model generated by this method enables estimation that takes into account the evaluation conditions.

(13) In the generation method of (9), each of the plurality of individual tire data may include evaluation condition data indicating evaluation conditions for tire characteristics. The evaluation condition data may be data that can selectively take on a plurality of condition values. The first estimation model and the second estimation model may each include a coefficient estimation model that outputs one or more coefficients that define a relational expression. The relational expression may be a relational expression defined by the condition value and the one or more coefficients, and used to calculate the characteristic data. By using the estimation model generated by this method, for example, when the relationship between evaluation conditions and tire characteristics has been empirically obtained, the tire characteristics can be estimated using that relational expression.

(14) In the generation method of (9), each of the plurality of individual tire data may include evaluation condition data indicating evaluation conditions for tire characteristics. In the model generation step, a plurality of first estimation models corresponding to the plurality of evaluation conditions may be generated, and a plurality of second estimation models corresponding to the plurality of evaluation conditions may be generated. This allows characteristics to be estimated according to the evaluation conditions.

(15) The estimation model generation device proposed in this disclosure includes a data acquisition means for acquiring a tire data group including multiple individual tire data. Each individual tire data includes characteristic data representing tire characteristics and specification data including data on multiple structural items representing the tire structure. The generation device includes a model generation means for generating an estimation model using the multiple individual tire data to estimate characteristic data based on the tire specification data. The tire data group includes a first data group and a second data group, each including the multiple individual tire data. The first data group and the second data group differ in at least some of the multiple structural items for which the specification data is provided. The model generation means generates a first estimation model using the multiple individual tire data included in the first data group as training data, and generates a second estimation model using the multiple individual tire data included in the second data group as training data. Using the estimation model generated by this device, the accuracy of estimating the characteristics of multiple tires with different structural items can be improved.

(16) A program proposed in this disclosure causes a computer to function as data acquisition means for acquiring a tire data group including multiple individual tire data. Each individual tire data includes characteristic data representing tire characteristics and specification data including data on multiple structural items that represent the tire structure. The program also causes a computer to function as model generation means for generating an estimation model using the multiple individual tire data to estimate characteristic data based on the tire specification data. The tire data group includes a first data group and a second data group, each including the multiple individual tire data. The first data group and the second data group differ in at least some of the multiple structural items for which the specification data is provided. The model generation means generates a first estimation model using the multiple individual tire data included in the first data group as training data, and generates a second estimation model using the multiple individual tire data included in the second data group as training data. Using the estimation model generated by this program can improve the accuracy of estimating the characteristics of multiple tires with different structural items.

1 is a block diagram showing hardware of a tire design support system that functions as a tire characteristic estimation device and an estimation model generation device proposed in the present disclosure. FIG. FIG. 2 is a block diagram showing functions of the design support system shown in FIG. 1 . FIG. 2 is a diagram showing an example of a tire data group stored in a storage device installed in the design support system. FIG. 2 is a block diagram showing functions of a learning unit that functions as a generation device. FIG. 4 is a diagram showing an example of data from which the tire data group shown in FIG. 3 is extracted. FIG. 4 is a diagram showing an example of data from which the tire data group shown in FIG. 3 is extracted. FIG. 4 is a diagram showing a modified example of the tire data group shown in FIG. 3 . FIG. 6B is a diagram showing an example of data extracted from the tire data group of FIG. 6A. 5 is a diagram illustrating an example of the structure of an estimation model generated by a first generation unit of the source model generation unit and the target model generation unit illustrated in FIG. 4 . FIG. FIG. 10 is a flowchart illustrating an example of processing executed by a first generating unit. 10 is an example of data used in a process executed by a first generation unit. 5 is a diagram illustrating an example of the structure of an estimation model generated by a second generation unit of the source model generation unit and the target model generation unit illustrated in FIG. 4 . FIG. 5 is a diagram illustrating an example of the structure of an estimation model generated by a third generation unit of the source model generation unit and the target model generation unit illustrated in FIG. 4 . FIG. FIG. 10 is a flowchart illustrating an example of processing executed by a learning unit. FIG. 10 is a diagram illustrating an example of an estimation model table generated by a learning unit. FIG. 3 is a flowchart illustrating an example of processing executed by an estimation unit illustrated in FIG. 2 .

The tire characteristic estimation device and tire characteristic estimation model generation device proposed in this disclosure are described below.

Figure 1 is a block diagram showing the hardware of a tire design support system 1 that functions as a tire characteristic estimation device and estimation model generation device proposed in this disclosure.

As shown in FIG. 1, the design support system 1 has a processing device 11, a storage device 12, a display device 13, and an input device 14.

The processing device 11 includes, for example, a central processing unit (CPU) and a graphics processing unit (GPU), and operates according to a program stored in the storage device 12. A field programmable gate array (FPGA) may be used as the processing device 11. The storage device 12 includes, for example, read-only memory (ROM), random access memory (RAM), hard disk drive (HDD), solid state drive (SSD), and stores programs executed by the processing device 11, tire data for generating an estimation model, and the like. The processing device 11 and storage device 12 may be one or more personal computers, server computers, or the like.

The display device 13 is, for example, a liquid crystal display or an organic EL display, and displays data stored in the memory device 12, output results of the estimation model, etc.

The input device 14 is a user interface such as a keyboard or mouse, which accepts operational inputs from the operator and inputs signals indicating the content of those inputs to the processing device 11.

Figure 2 is a block diagram showing the functions of the design support system 1. Figure 3 is a diagram showing an example of a tire data group stored in the storage device 12. As shown in Figure 2, the design support system 1 has a learning unit 20, an estimation unit 30, and an estimation model group 40.

The estimation model group 40 includes multiple estimation models for estimating tire characteristics from tire specifications. Tire specifications include, for example, the tire structure and manufacturing method, and tire structure includes, for example, the type, dimensions, number, and materials of the components that make up the tire, as well as their properties (e.g., hardness).

The learning unit 20 trains the estimation models constituting the estimation model group 40 on the tire data group stored in the storage device 12. The estimation unit 30 inputs data about the structure of the tire to be estimated into one or more estimation models constituting the estimation model group 40, and obtains characteristic data of the tire to be estimated as output from the estimation model.

Below, we will first explain the learning unit 20, followed by the estimation unit 30.

[Tire data group]
As shown in Fig. 3, the learning tire data group includes a plurality of individual tire data. Each individual tire data is data for one tire. In this figure, each row represents one individual tire data. Each individual tire data includes characteristic data representing the tire characteristics, evaluation condition data which are conditions for evaluating the tire characteristics, and specification data.

The characteristic data includes, for example, cornering power, noise generated between the road surface and the tire, rolling resistance, friction coefficient, and durability. Figure 3 shows cornering power and friction coefficient as examples. The evaluation conditions for cornering power are, for example, the load acting on the tire; generally, the higher the load, the greater the cornering power. The evaluation conditions for friction coefficient are, for example, the road surface condition, such as icy road, gravel road, or asphalt. The characteristics recorded in the individual tire data are data measured for actual tires.

Specification data includes data on structural items that describe the structure of a tire. Structural items include, for example, tire width, tread type (tread shape), tread width, tire height, rubber material, belt material, and belt hardness. Figure 3 shows tire width, tread type, tread width, structural value A, and structural value B as examples. Specification data items may also include tire manufacturing methods. Individual tire data includes not only numerical data such as tire width, but also non-numeric data such as rubber material, belt material, and manufacturing method, recorded as numerical values.

[Study Department]
4 is a block diagram showing the functions of the learning unit 20. As shown in the figure, the learning unit 20 has a data acquisition unit 21, a data extraction unit 22, and a model generation unit 23.

The data acquisition unit 21 acquires a group of tire data (e.g., the data shown in Figure 3) stored in the storage device 12. If the design support system 1 has a communication device, the data acquisition unit 21 may receive the group of tire data from an external device via the communication device.

[Extraction part]
The data extraction unit 22 extracts data to be used for learning from the acquired tire data group. That is, the data extraction unit 22 extracts a plurality of individual tire data corresponding to the characteristics for which an estimation model is to be generated from the tire data group.

In the example shown in Figure 3, all of the individual tire data includes cornering power and data on its evaluation conditions, while some of the individual tire data (data surrounded by dashed line area A) does not include friction coefficient, with Null entered in the friction coefficient cell. When generating an estimation model for cornering power, the data extraction unit 22 extracts from the tire data group multiple individual tire data items that have significant values (non-null values) in the cornering power cell. When generating an estimation model for friction coefficient, the data extraction unit 22 extracts from the tire data group multiple individual tire data items that have significant values (non-null values) in the friction coefficient cell.

The data extraction unit 22 includes a first extraction unit 22A and a second extraction unit 22B. As shown in FIG. 4, the estimation model group 40 includes a source estimation model and a target estimation model. The first extraction unit 22A extracts data for generating a source estimation model from the tire data group, and the second extraction unit 22B extracts data for generating a target estimation model from the tire data group. As described below, the target estimation model is generated using the source estimation model as well.

Also, depending on the tire structure, there may be structural items for which no data is available. For example, if there are tires with one belt and tires with two belts, all individual tire data will include data on the hardness of the first belt, while data for tires with only one belt will not include data on the hardness of the second belt. In this case, in the specification data for tires with only one belt, the cell for the structural item for the second belt will be null.

In the example shown in Figure 3, both structural item A (e.g., the hardness of the first belt) and structural item B (e.g., the hardness of the second belt) are items that affect cornering power. The tire data group has a first data group and a second data group. Each data group is composed of multiple individual tire data. Both the individual tire data of the first data group and the individual tire data of the second data group have a significant value for structural item A. In contrast, Null is entered for structural item B in the individual tire data of the first data group, and a significant value is also recorded for structural item B (e.g., the hardness of the second belt) in the individual tire data of the second data group. In this case, as shown in Figure 5A, the first extraction unit 22A extracts the specification data included in the first data group, the characteristic data to be estimated, and the evaluation condition data from the tire data group. The extracted data is then used to generate a source estimation model in the model generation unit 23. Additionally, as shown in FIG. 5B, the second extraction unit 22B extracts the specification data included in the second data group, the characteristic data to be estimated, and the evaluation condition data from the tire data group. The extracted data is used by the model generation unit 23 to generate a target estimation model.

[Variations of extraction]
The processing performed by the extraction units 22A and 22B is not limited to the example described with reference to Figs. 5A and 5B. Fig. 6A is a diagram showing a modified example of a tire data group. In the example shown in Fig. 6A, in the individual tire data of the first data group, a significant value is recorded in structure item A (see dashed line area M1), but Null is recorded in structure item B. Conversely, in the individual tire data of the second data group, a significant value is recorded in structure item B (see dashed line area M2), but Null is recorded in structure item A.

In the tire data group shown in FIG. 3 above, all of the structural items contained in the data of the first data group were included in the structural items contained in the data of the second data group. In contrast, in the tire data group shown in FIG. 6, only some of the structural items contained in the data of the first data group are common to some of the structural items contained in the data of the second data group. When the tire data group is as shown in FIG. 6, the first extraction unit 22A, which extracts data for the source estimation model, extracts specification data, characteristic data (cornering power in FIG. 6B), and evaluation condition data (load in FIG. 6B) of structural items common to the structural items of the first data group and the second data group from the tire data group, as shown in FIG. 6B.

As explained with reference to Figures 6A and 6B, when structural items common to the first data group and the second data group are used to generate a source estimation model, a target estimation model (first estimation model in the claims) using the individual tire data (data including structural item A) of the first data group and the source estimation model is generated in the model generation unit 23, which will be described later. Also, a target estimation model (second estimation model in the claims) using the individual tire data (data including structural item B) of the second data group and the source estimation model is generated in the model generation unit 23, which will be described later.

On the other hand, as explained with reference to Figures 3, 5A, and 5B, when the data of the first data group is used as data for a source estimation model, this source estimation model also becomes one of multiple target estimation models. That is, the source estimation model generated based on the data of the first data group is used as a target estimation model (the first estimation model in the claims) for estimating the characteristics of a tire having the same structural items as the first data group. Furthermore, the model generated based on the data of the second data group is used as a target estimation model (the second estimation model in the claims) for estimating the characteristics of a tire having the same structural items as the second data group.

[Model generation section]
The model generation unit 23 generates an estimation model for estimating tire characteristics based on tire specification data, using the data extracted by the model extraction unit 22. As shown in Fig. 4, the model generation unit 23 has a source model generation unit 23A and a target model generation unit 23B. The source model generation unit 23A has a first generation unit 23Aa, a second generation unit 23Ab, and a third generation unit 23Ac. Similarly, the target model generation unit 23B has a first generation unit 23Ba, a second generation unit 23Bb, and a third generation unit 23Bc.

The target model generation unit 23B performs so-called transfer learning using the source estimation model generated by the source model generation unit 23A. That is, the first generation unit 23Ba of the target model generation unit 23B generates a target estimation model using the source estimation model generated by the first generation unit 23Aa of the source model generation unit 23A and the data extracted by the second extraction unit 22A. Similarly, the second generation unit 23Bb of the target model generation unit 23B generates a target estimation model using the source estimation model generated by the second generation unit 23Ab of the source model generation unit 23A and the data extracted by the second extraction unit 22A. The third generation unit 23Bc of the target model generation unit 23B generates a target estimation model using the source estimation model generated by the third generation unit 23Ac of the source model generation unit 23A and the data extracted by the second extraction unit 22A.

The following describes the processing performed by the first generation units 23Aa and 23Ba, the second generation units 23Ab and 23Bb, and the third generation units 23Ac and 23Bc in order.

[First generation section]
FIG. 7 is a diagram showing an example of the structure of an estimation model (source estimation model) generated by the first generation unit 23Aa of the source model generation unit 23A. Among tire characteristics, there are some for which a relational expression for calculating the characteristic has been empirically obtained. For example, as shown in FIG. 7, it is known that cornering power can be calculated from the load, which is an evaluation condition. Specifically, cornering power can be calculated using the following relational expression:
CP=a×W+b×W^2+c×W^3
CP: Cornering power W: Load (evaluation conditions)
a, b, and c: coefficients

In this case, the estimation model includes a coefficient estimation model for estimating the coefficients a, b, and c from the specification data, and a relational equation (empirical regression model). The coefficient estimation model is a machine learning model, such as a neural network. The coefficient estimation model may be a model such as a random forest or a support vector machine that outputs multiple coefficients a, b, and c. Alternatively, the coefficient estimation model may be a regression model (for example, LASSO regression) that outputs multiple coefficients a, b, and c.

When the estimation model has the structure shown in Figure 7, the first generation unit 23Aa executes the processing illustrated in Figures 8A and 8B. Figure 8A is a flow diagram showing an example of the processing executed by the first generation unit 23Aa, and Figure 8B is an example of data used in the processing executed by the first generation unit 23Aa.

The first generation unit 23Aa acquires multiple individual tire data having common specification data from the multiple individual tire data extracted by the first extraction unit 22A (S101). That is, the first generation unit 23Aa acquires multiple individual tire data for tires having the same structure (e.g., one tire). This multiple individual tire data has different evaluation conditions for the characteristic data. The multiple individual tire data illustrated in FIG. 8B is for one tire and has the same data (significant values) for all structural items. In contrast, the load, which is an evaluation condition, is different from each other, and as a result, the characteristic data is also different. The first generation unit 23Aa calculates multiple coefficients a, b, and c using such evaluation conditions (e.g., load) and characteristic data (e.g., cornering power) (S102). The first generation unit 23Aa can determine the coefficients a, b, and c, for example, by using a polynomial regression model. Then, the first generation unit 23Aa trains a coefficient estimation model using the specification data of the multiple individual tire data acquired in S101 and the coefficients a, b, and c calculated in S102 as training data (S103). At this time, the coefficients a, b, and c are used as correct labels.

The first generation unit 23Aa acquires multiple individual tire data sets having common specification data, as illustrated in FIG. 8B, from the data extracted by the first extraction unit 22A (data for the source estimation model). The first generation unit 23Aa then performs the processes shown in S101 to S103 in FIG. 8A for multiple tires. This allows the source estimation model to progress in learning.

The estimation model (target estimation model) generated by the first generation unit 23Ba of the target model generation unit 23B also has a structure similar to that shown in Figure 7. The first generation unit 23Ba of the target model generation unit 23B performs transfer learning using, for example, the source estimation model generated by the first generation unit 23Aa of the source model generation unit 23A and the data extracted by the second extraction unit 22B (for example, the data in Figure 5B).

The processing of the first generation unit 23Ba may be similar to the processing of the first generation unit 23Aa of the source model generation unit 23A. Specifically, the second generation unit 23Ba acquires multiple individual tire data sets having common specification data, as illustrated in FIG. 8B, from the data extracted by the second extraction unit 22B (S101 in FIG. 8A). The second generation unit 23Ba then calculates coefficients a, b, and c from the data obtained in S101 (S102 in FIG. 8A). Next, the second generation unit 23Ba uses these coefficients a, b, and c as correct labels and the specification data of the individual tire data obtained in S101 as explanatory variables to proceed with training of the estimation model (target estimation model) (S103 in FIG. 8A). In S103, the second generation unit 23Ba may use the source estimation model. For example, the second generation unit 23Ba may adjust (fine-tune) the weight parameters of the source estimation model, reuse an intermediate layer of the source estimation model, or add a new layer to the source estimation model.

[Modification of the first generation unit]
The processing performed by the first generators 23Aa and 23Ba and the structure of the estimation model are not limited to the examples shown in Figures 7, 8A, and 8B. For example, even if a relational expression for calculating tire characteristic data is known, the estimation model (neural network) generated by the first generators 23Aa and 23Ba may directly calculate tire characteristic data. In this case, the difference between the cornering power calculated from the estimation model and the actually measured cornering power (characteristic data included in the individual tire data) may be used as a loss function to train the estimation model.

Note that the estimation model of the structure shown in Figure 7 may also be used to estimate characteristics other than cornering power, as long as a relational expression is obtained for the characteristics.

[Second generation section]
FIG. 9 is a diagram showing an example of the structure of an estimation model (source estimation model) generated by the second generation unit 23Ab of the source model generation unit 23A. The second generation unit 23Ab generates a plurality of source estimation models corresponding to a plurality of evaluation conditions. The estimation model illustrated in FIG. 9 is a model for estimating a friction coefficient, which is one of the tire characteristics. As described above, evaluation conditions for the friction coefficient include road surface conditions such as icy roads, gravel roads, and asphalt. In this case, the second generation unit 23Ab generates a model for estimating the friction coefficient on icy roads, a model for estimating the friction coefficient on gravel roads, and a model for estimating the friction coefficient on asphalt.

The second generation unit 23Ab acquires data with the same evaluation condition from the multiple individual tire data extracted by the first extraction unit 22A (see Figure 4). For example, the second generation unit 23Ab acquires multiple individual tire data with "frozen road" as the evaluation condition data. The second generation unit 23Ab then trains an estimation model that estimates the friction coefficient on frozen roads, using the specification data of the acquired individual tire data as explanatory variables and the friction coefficient on asphalt as the correct label. The second generation unit 23Ab performs processing similar to that for frozen roads for other evaluation conditions (e.g., gravel roads, asphalt, etc.) to generate estimation models for the other evaluation conditions.

The estimation model (target estimation model) generated by the second generation unit 23Bb of the target model generation unit 23B also has a structure similar to that shown in FIG. 9. The second generation unit 23Bb of the target model generation unit 23B performs transfer learning, for example, using the source estimation model generated by the second generation unit 23Ab of the source model generation unit 23A and the data extracted by the second extraction unit 22B. For example, the second generation unit 23Bb may adjust (fine-tune) the weight parameters of the source estimation model, reuse an intermediate layer of the source estimation model, or add a new layer to the source estimation model.

The estimation model of the structure shown in FIG. 9 may also be used to estimate characteristics other than the friction coefficient. For example, the estimation model of the structure shown in FIG. 9 may be used to estimate cornering power.

[Third generation part]
10 is a diagram showing an example of the structure of an estimation model (source estimation model) generated by the third generation unit 23Ac of the source model generation unit 23A. The third generation unit 23Ac generates one source estimation model using specification data and evaluation condition data as explanatory variables. When there are a large number of evaluation conditions, the estimation model shown in FIG. 10 may be generated instead of the estimation model shown in FIG. 9. For example, when there are a large number of evaluation conditions for evaluating durability, the estimation model shown in FIG. 10 may be used as a model for estimating durability.

The estimation model (target estimation model) generated by the third generation unit 23Bc of the target model generation unit 23B also has a structure similar to that shown in FIG. 10. The third generation unit 23Bc of the target model generation unit 23B performs transfer learning, for example, using the source estimation model generated by the third generation unit 23Ac of the source model generation unit 23A and the data extracted by the second extraction unit 22B. For example, the third generation unit 23Bc may adjust (fine-tune) the weight parameters of the source estimation model, reuse an intermediate layer of the source estimation model, or add a new layer to the source estimation model.

[Learning process flow]
FIG. 11 is a flowchart showing an example of processing executed by the learning unit 20.

The data acquisition unit 21 acquires a tire data group (Figure 3 or Figure 6A) including multiple individual tire data (S201). The learning unit 20 acquires the type of characteristic data for which an estimation model is to be generated (S202). For example, the acquired type of characteristic data includes cornering power, friction coefficient, rolling resistance, etc. The operator may be able to input the type of characteristic data for which an estimation model is to be generated to the processing device 11, for example, via the input device 14.

The data extraction unit 22 extracts individual tire data including characteristic data of the type received in S202 from the tire data group acquired in S201 (S203). For example, when attempting to generate an estimation model for estimating cornering power, the data extraction unit 22 extracts individual tire data including cornering power as characteristic data. Furthermore, when attempting to generate an estimation model for estimating friction coefficient, the data extraction unit 22 extracts individual tire data including friction coefficient as characteristic data.

Next, the first extraction unit 22A extracts data for generating a source estimation model from the data extracted in S203 (S204). In the example described above, the first extraction unit 22A extracts, for example, data having significant values for all structural items (the first data group illustrated in Figures 3 and 5A). As another example, the first extraction unit 22A extracts individual tire data (the data illustrated in Figure 6B) including specification data, characteristic data, and evaluation condition data for structural items (common structural items) having significant values in all of the multiple data groups.

The data generation unit 23 determines whether a relational equation (empirical regression model) for calculating the characteristic data to be estimated has been defined (S205). For example, the data generation unit 23 determines whether such a relational equation is stored in the storage device 12. In the example described above, when attempting to generate an estimation model for cornering power, the data generation unit 23 determines whether a relational equation showing the relationship between cornering power (CP) and evaluation conditions (load at the time of CP measurement) is stored in the storage device 12.

If it is determined in S205 that the relational expression is stored in the storage device 12, the source model generation unit 23A (first generation unit 23Aa) generates a source estimation model having the structure shown in FIG. 7, for example (S206).

Specifically, as described with reference to FIGS. 8A and 8B, the source model generation unit 23A (first generation unit 23Aa) acquires, from the multiple individual tire data extracted in S204, multiple individual tire data pieces that are data pieces about tires (or a single tire) with the same structure but with different evaluation conditions (S101 in FIG. 8A). The source model generation unit 23A calculates multiple coefficients a, b, and c that define the relational equation using these evaluation conditions (e.g., load) and characteristic data (e.g., cornering power) (S102 in FIG. 8A). Then, the source model generation unit 23A trains a coefficient estimation model using the specification data of the multiple individual tire data pieces acquired in S101 and the calculated coefficients a, b, and c as training data (S103 in FIG. 8A). In S206 in FIG. 11, the source model generation unit 23A generates a source estimation model by repeatedly executing the processes (S101 to S103) illustrated in FIG. 8A.

Next, the second extraction unit 22B extracts data for generating a target estimation model from the multiple individual tire data extracted in S203 (S207). Referring to the example described above, the second extraction unit 22B extracts, for example, the individual tire data of the second data group shown in FIG. 5B. Furthermore, as explained with reference to FIG. 6A, if only some of the structural items of the first data group and only some of the structural items of the second data group are common, the second extraction unit 22B extracts, for example, each of the individual tire data of the first data group and the individual tire data of the second data group as data for the target estimation model.

The target model generation unit 23B (first generation unit 23Ba) then generates a target estimation model using the source estimation model obtained in S206 and the data obtained in S207 (S208). If multiple data groups (first data group and second data group) are extracted in S207, the target model generation unit 23B generates multiple target estimation models corresponding to the multiple data groups in S208. In contrast, if the first data group is extracted as data for a source estimation model, the target model generation unit 23B (first generation unit 23Ba) does not require transfer learning for the first data group, and therefore uses the source estimation model obtained in S206 as one of the multiple target estimation models.

If it is determined in S205 that a relational equation (empirical regression model) for calculating characteristic data has not been defined, the source model generation unit 23A (e.g., second generation unit 23Ab) generates a source estimation model (S209). Referring to the example described above, if a friction coefficient is specified as the characteristic to be estimated in S202, the source model generation unit 23A (second generation unit 23Ab) generates multiple source estimation models (e.g., a model for icy roads, a model for gravel roads, etc.) corresponding to multiple evaluation conditions, respectively, as illustrated in FIG. 9.

Next, the second extraction unit 22B extracts data for generating a target estimation model from the multiple individual tire data extracted in S203 (S210). The target model generation unit 23B (second generation unit 23Ba) generates a target estimation model using the source estimation model obtained in S209 and the data obtained in S210 (S211).

For example, when attempting to generate an estimation model for the friction coefficient, the target model generation unit 23B (second generation unit 23Bb) generates a target estimation model for gravel roads using data included in the first data group whose evaluation condition is "gravel road." The target model generation unit 23B (second generation unit 23Bb) also generates a target estimation model for icy roads using data included in the first data group whose evaluation condition is "icy road." Furthermore, it generates a target estimation model for asphalt using data included in the first data group whose evaluation condition is asphalt.

The target model generation unit 23B (second generation unit 23Bb) also performs similar processing on the second data group. That is, it generates a target estimation model for gravel roads using data included in the second data group whose evaluation condition is "gravel road." The target model generation unit 23B (second generation unit 23Bb) also generates a target estimation model for icy roads using data included in the second data group whose evaluation condition is "icy road." Furthermore, it generates a target estimation model for asphalt using data included in the second data group whose evaluation condition is asphalt.

In this way, in S211, the target model generation unit 23B (second generation unit 23Ba) generates multiple target estimation models corresponding to multiple evaluation conditions for each of the first data group and the second data group. As a result, target estimation models are generated in the number equal to the number of data groups x the number of evaluation conditions.

Note that if the first data group is extracted as data for a source estimation model in S204, the target model generation unit 23B (second generation unit 23Bb) does not require transfer learning for the first data group, and therefore the source estimation model obtained in S209 may be used as one of multiple target estimation models.

Note that if there are a large number of evaluation conditions (for example, if there are 10 or more evaluation conditions), in S209, the source model generation unit 23A (third generation unit 23Ac) may generate one source estimation model (a model with the structure illustrated in FIG. 10) using the specification data and evaluation condition data as explanatory variables. Then, the target model generation unit 23B (third generation unit 23Bc) may generate a target estimation model with the structure illustrated in FIG. 10 for each of the multiple data groups.

[Estimated Model Table]
The learning unit 20 may store in the storage device 12 a table that associates the generated target estimation models with the structural items, evaluation conditions, and characteristic data of the specification data. Hereinafter, this map will be referred to as an estimation model table. FIG. 12 is a diagram showing an example of the estimation model table. In this figure, an ID that identifies each target estimation model is recorded in the leftmost column. Furthermore, each ID is associated with the characteristic to be estimated, the evaluation conditions to be applied, and whether the structural item is "present" or "absent."

In the example shown in this figure, the target estimation models (ID: 1 and ID: 2) are models that output cornering power (CP). Also in this figure, "1 < W <7" is recorded as the evaluation condition for the target estimation models (ID: 1 and ID: 2). This indicates that any value greater than 1 and less than 7 can be input as the load W, which is the evaluation condition, for these models. As described above, the cornering power can be calculated using the following relational expression. Therefore, as the load W, which is the evaluation condition, the operator can input the evaluation condition (load W) for which he or she wants to obtain characteristic data.
CP=a×W+b×W^2+c×W^3
CP: Cornering power W: Load (evaluation conditions)
a, b, and c: coefficients

Furthermore, for the target estimation model (ID: 1), "Yes" is recorded for the structural items "Tire width," "Tread type," and "Structural item A," and "No" is recorded for "Structural item B." This indicates that the target estimation model (ID: 1) is applied to tires that do not have structural item B. On the other hand, for the target estimation model (ID: 2), "Yes" is recorded for the structural items "Tire width," "Tread type," "Structural item A," and "Structural item B." This indicates that the target estimation model (ID: 2) is applied to tires that have structural item B.

In the table illustrated in FIG. 12, the target estimation models (ID: 11 and ID: 12) are models for estimating the friction coefficient. In this figure, "frozen road" and "gravel road" are recorded as the road surface conditions, which are evaluation conditions, for the target estimation models (ID: 11 and ID: 12), respectively. This indicates that the target estimation model (ID: 11) is a model for estimating the friction coefficient on icy roads, and the target estimation model (ID: 12) is a model for estimating the friction coefficient on gravel roads. Furthermore, for the target estimation models (ID: 11 and ID: 12), "Yes" is recorded for the structural items "Tire width," "Tread type," and "Structural item A," and "No" is recorded for "Structural item B." This indicates that the target estimation models (ID: 11 and ID: 12) are applicable to tires that do not have structural item B.

[Estimation part]
The following describes the processing executed by the estimation unit 30. FIG. 13 is a flowchart showing an example of the processing executed by the estimation unit 30.

The estimation unit 30 acquires specification data for a tire whose characteristics are to be estimated (hereinafter referred to as the estimation target tire) (S301). The specification data for the estimation target tire includes data on multiple structural items such as the dimensions, materials, and positions of tire components, as illustrated in Figure 3 or Figure 6A. Depending on the type of estimation target tire, the specification data may not have significant values for some structural items. For example, similar to the data in the first data group shown in Figure 3, the specification data for the estimation target tire may not have a significant value for structural item B. The specification data may also include the tire manufacturing method as one of its items.

The estimation unit 30 acquires the type of characteristic data to be estimated (e.g., cornering power, friction coefficient, durability, etc.) (S302). For example, the operator inputs the type of characteristic data to be estimated into the processing device 11 via the input device 14, and the estimation unit 30 accepts this input type.

Next, the estimation unit 30 acquires the evaluation conditions for the characteristic to be estimated. For example, if the operator wishes to estimate the cornering power of the tire being estimated, the load, which is an evaluation condition for cornering power, is input to the processing device 11 via the input device 14, and the estimation unit 30 accepts this input load. Also, if the operator wishes to estimate the friction coefficient of the tire being estimated, the road surface condition, which is an evaluation condition for the friction coefficient (e.g., frozen road, gravel road, etc.), is input to the processing device 11 via the input device 14, and the estimation unit 30 accepts this input road surface condition as an evaluation condition.

Next, the estimation unit 30 selects one of multiple target estimation models stored in the storage device 12 based on the structural items for which the specification data acquired in S301 is provided (S304). For example, the estimation unit 30 refers to the estimation model table (see FIG. 12) stored in the storage device 12 and selects a target estimation model that corresponds to the characteristic (characteristic to be estimated) received in S302 and the structural items of the specification data acquired in S301.

For example, if cornering power is specified as the characteristic to be estimated and the specification data acquired in S301 does not have a significant value for structural item B, the estimation unit 30 selects the target estimation model (ID: 1). On the other hand, if cornering power is specified as the characteristic to be estimated and the specification data acquired in S301 has a significant value for structural item B, the estimation unit 30 selects the target estimation model (ID: 2).

Furthermore, in the example described above, multiple target estimation models (models exemplified in FIG. 9) corresponding to multiple evaluation conditions are generated for the friction coefficient. Therefore, when a friction coefficient is specified in S302, the estimation unit 30 may select a target estimation model based on the structural item for which the specification data acquired in S301 is provided and the evaluation condition (road surface condition) specified in S303.

The estimation unit 30 inputs the specification data acquired in S301 into the target estimation model selected in S304 and estimates characteristic data (S305). As explained with reference to FIG. 7, if the target estimation model has a coefficient estimation model and a relational equation (empirical regression model), the estimation unit 30 estimates the coefficients a, b, and c using the coefficient estimation model of the selected target estimation model. Then, the estimation unit 30 inputs the evaluation condition (load W) specified in S303 into the relational equation defined by the coefficients a, b, and c, and estimates cornering power, which is characteristic data.

Furthermore, as explained with reference to Figure 10, if the target estimation model selected in S304 is a model that has evaluation conditions as explanatory variables in addition to specification data, the estimation unit 30 inputs the specification data acquired in S301 and the evaluation conditions specified in S303 into this model, and estimates characteristic data as its output.

Also, unlike the example described above, if multiple target estimation models corresponding to multiple evaluation conditions have been generated (i.e., if the target estimation models have the structure illustrated in FIG. 9), the estimation unit 30 may select one of these multiple target estimation models in S304. Then, in S305, the estimation unit 30 may input the specification data acquired in S301 to all of these multiple target estimation models and output characteristic data.

[Searching for tire specifications]
The design support system 1 may have, as its functions, a search unit that searches for tire specification data that satisfies the characteristics desired by the operator, in addition to the learning unit 20 and the estimation unit 30. This process of the search unit can be executed, for example, as follows.

The search unit first provides the estimation unit 30 with specification data that has been defined in advance or specification data set by the worker. The estimation unit 30 selects a target estimation model corresponding to the specification data acquired from the search unit from among multiple target estimation models stored in the storage device 12. The estimation unit 30 then inputs the specification data into the selected target estimation model, calculates characteristic data as its output, and provides the calculated characteristic data to the search unit. The search unit may generate new specification data so that the calculated characteristic data approaches the target set by the worker and provide this to the estimation unit 30. The design support system 1 searches for specification data that achieves characteristics close to the target set by the worker by repeatedly executing the processing of the search unit and the processing of the estimation unit 30.

The estimation unit 30 may estimate multiple characteristic data based on the specification data acquired from the search unit. That is, the estimation unit 30 may estimate multiple characteristics, such as cornering power, noise generated between the road surface and the tire, rolling resistance, friction coefficient, and durability. The estimation unit 30 may select a target estimation model for estimating these multiple characteristics based on the specification data acquired from the search unit. The estimation unit 30 may then provide the search unit with multiple characteristic data obtained from each of the multiple target estimation models. The search unit may then generate new specification data so that the entire set of these multiple characteristic data approaches the target set by the operator, and provide this to the estimation unit 30. The design support system 1 may search for specification data that achieves values close to the target set by the operator for the multiple characteristics by repeatedly executing the processing of the search unit and the processing of the estimation unit 30. Such a search may be achieved using a so-called genetic algorithm or gradient method.

The design support system 1 may also display the specification data obtained as a result of processing by the search unit and the estimated characteristic data on the display device 13. In this case, the number of dimensions of the specification data may be reduced, and the reduced-dimensional specification data may be displayed on the display device 13. Techniques such as self-organizing maps and principal component analysis may be used to reduce the number of dimensions.

[summary]
As described above, in the tire characteristic estimation device (design support system 1) proposed in this disclosure, the estimation unit 30 has data acquisition means (S301 in FIG. 13) that acquires specification data of a tire to be estimated. The estimation unit 30 also selects one of a first target estimation model and a second target estimation model based on a plurality of structural items for which the acquired specification data is provided (S304 in FIG. 13), inputs the acquired specification data into the selected target estimation model, and outputs characteristic data of the tire to be estimated (S305 in FIG. 13). A device including this estimation unit 30 can improve the accuracy of estimating the characteristics of a plurality of tires with different structural items.

As described above, the tire data group includes a first data group (see FIG. 3) and a second data group (see FIG. 3), each of which includes multiple individual tire data. The first data group and the second data group differ in some of the multiple structural items. The model generation unit 23 (target model generation unit 23B) generates a first target estimation model using the multiple individual tire data included in the first data group as training data, and generates a second target estimation model using the multiple individual tire data included in the second data group as training data. Using the estimation model generated in this manner can improve the accuracy of estimating the characteristics of multiple tires with different structural items.

As described above, the first data group and the second data group include, as part of the multiple structural items, structural items common to the first data group and the second data group (see Figures 3 and 6A). The data extraction unit 22 (first extraction unit 22A) extracts data including specification data and characteristic data for the common structural items from the tire data group. The model generation unit 23 includes a source model generation unit 23A that generates a source estimation model based on the specification data for the common structural items using the extracted data as training data, and a target model generation unit 23B that generates a first target estimation model using the source estimation model and multiple individual tire data from the first data group, and generates a second target estimation model using the source estimation model and multiple individual tire data from the second data group. This allows for improved estimation accuracy of the estimation model obtained from one of the two data groups, even when the number of individual tire data included in that one data group is small.

Note that the tire characteristic estimation device and tire characteristic estimation model generation device proposed in this disclosure are not limited to the examples described above, and various modifications may be made.

For example, in the above description, the tire data group including multiple individual tire data (e.g., FIG. 3) has two data groups, a first data group and a second data group, each containing data on cornering power. However, the tire data group may also include more data groups. In this case, multiple target estimation models may be generated that respectively correspond to these multiple data groups.

In addition, in the above explanation, cornering power is given as an example of a characteristic estimated using the estimation model of the structure shown in FIG. 7. However, the estimation model of the structure shown in FIG. 7 may be used to estimate other characteristics. Furthermore, friction coefficient is given as an example of a characteristic estimated using the estimation model of the structure shown in FIG. 9. However, the estimation model of the structure shown in FIG. 9 may be used to estimate other characteristics. For example, the estimation model of the structure shown in FIG. 9 may be used to estimate cornering power.

11: Processing device, 12: Storage device, 13: Display device, 14: Input device, 20: Learning unit, 21: Data acquisition unit, 22: Data extraction unit, 22A: First extraction unit, 22B: Second extraction unit, 23: Model generation unit, 23A: Source model generation unit, 23B: Target model generation unit, 23Aa and 23Ba: First generation unit, 23Ab and 23Bb: Second generation unit, 23Ac and 23Bc: Third generation unit, 30: Estimation unit, 40: Estimation model group.

Claims (10)

a storage means in which a first estimation model and a second estimation model are stored, each of the first estimation model and the second estimation model being an estimation model for estimating characteristic data representing tire characteristics from specification data including data on a plurality of structural items that are a plurality of items representing the structure of the tire;
data acquisition means for acquiring specification data of the estimation target tire;
a means for acquiring evaluation condition data indicating evaluation conditions for the characteristics to be estimated for the estimation target tire;
an estimation means for selecting one of the first estimation model and the second estimation model based on a plurality of structural items for which the acquired specification data is provided, inputting the acquired specification data into the selected estimation model, and outputting characteristic data of the estimation target tire ,
each of the first estimation model and the second estimation model includes a coefficient estimation model that outputs one or more coefficients according to the specification data, and is an estimation model for estimating the characteristic data from the specification data and the evaluation condition data;
The estimation means inputs the acquired specification data into the coefficient estimation model of the selected estimation model, and calculates characteristic data of the estimation target tire using a relational expression defined by the one or more coefficients output from the coefficient estimation model and the acquired evaluation condition data.
Tire characteristic estimation device. the storage means stores a plurality of third estimation models corresponding to a plurality of evaluation conditions for tire characteristics, and a plurality of fourth estimation models corresponding to the plurality of evaluation conditions,
2. The tire characteristic estimation device according to claim 1, wherein the estimation means selects one or more estimation models from the first estimation model, the second estimation model, the plurality of third estimation models, and the plurality of fourth estimation models based on the plurality of structural items for which the acquired specification data is provided. each of the first estimation model and the second estimation model is a model generated from a tire data group including a plurality of individual tire data, each of the individual tire data including the specification data and the characteristic data;
the tire data group includes a first data group and a second data group, each of which includes the plurality of individual tire data;
the individual tire data of the first data group and the individual tire data of the second data group differ in at least a part of the plurality of structural items for which the specification data is provided,
the first estimation model is a model generated using the plurality of individual tire data included in the first data group as training data,
The tire characteristic estimation device according to claim 1 , wherein the second estimation model is a model generated using a plurality of individual tire data included in the second data group as training data. An estimation method for estimating a tire characteristic by using a first estimation model and a second estimation model,
each of the first estimation model and the second estimation model is an estimation model for estimating characteristic data representing tire characteristics from specification data including data on a plurality of structural items that are a plurality of items representing a tire structure;
a data acquisition step of acquiring specification data of the estimation target tire ;
acquiring evaluation condition data indicating evaluation conditions for the characteristics to be estimated for the estimation target tire;
an estimation step of selecting one of the first estimation model and the second estimation model based on a plurality of structural items for which the acquired specification data is provided, inputting the acquired specification data into the selected estimation model, and outputting characteristic data of the estimation target tire ,
each of the first estimation model and the second estimation model includes a coefficient estimation model that outputs one or more coefficients according to the specification data, and is an estimation model for estimating the characteristic data from the specification data and the evaluation condition data;
In the estimation step, the acquired specification data is input to the coefficient estimation model of the selected estimation model, and characteristic data of the estimation target tire is calculated using a relational expression defined by the one or more coefficients output from the coefficient estimation model and the acquired evaluation condition data.
Methods for estimating tire characteristics. A program that causes a computer to function as an estimation device that estimates tire characteristics by utilizing a first estimation model and a second estimation model,
each of the first estimation model and the second estimation model is an estimation model for estimating characteristic data representing tire characteristics from specification data including data on a plurality of structural items that are a plurality of items representing a tire structure;
data acquisition means for acquiring specification data of the estimation target tire;
The computer functions as: a means for acquiring evaluation condition data indicating evaluation conditions for the characteristics to be estimated for the tire to be estimated; and an estimation means for selecting one of the first estimation model and the second estimation model based on a plurality of structural items for which the acquired specification data is provided, inputting the acquired specification data into the selected estimation model, and outputting characteristic data for the tire to be estimated .
each of the first estimation model and the second estimation model includes a coefficient estimation model that outputs one or more coefficients according to the specification data, and is an estimation model for estimating the characteristic data from the specification data and the evaluation condition data;
The estimation means inputs the acquired specification data into the coefficient estimation model of the selected estimation model, and calculates characteristic data of the estimation target tire using a relational expression defined by the one or more coefficients output from the coefficient estimation model and the acquired evaluation condition data.
A program that makes a computer function like this. a data acquisition step of acquiring a tire data group including a plurality of individual tire data, each of the individual tire data including characteristic data representing the characteristics of the tire, specification data including data on a plurality of structure items representing the structure of the tire, and evaluation condition data being condition values indicating evaluation conditions for the characteristics of the tire ;
a model generation step of generating an estimation model for estimating the characteristic data based on the specification data of an estimation target tire by using the plurality of individual tire data,
the tire data group includes a first data group and a second data group, each of which includes the plurality of individual tire data;
the first data group and the second data group differ in at least a part of the plurality of structural items,
In the model generation step, a first estimation model is generated using the plurality of individual tire data included in the first data group as training data, and a second estimation model is generated using the plurality of individual tire data included in the second data group as training data,
each of the first estimation model and the second estimation model includes a coefficient estimation model that outputs one or more coefficients that define a relational expression for calculating the characteristic data;
In the model generation step,
calculating the one or more coefficients based on the characteristic data of the plurality of individual tire data included in the first data group and the evaluation condition data, and generating the coefficient estimation model of the first estimation model using the calculated one or more coefficients and the specification data as training data;
Calculating the one or more coefficients based on the characteristic data and the evaluation condition data of the plurality of individual tire data included in the second data group, and generating the coefficient estimation model of the second estimation model using the calculated one or more coefficients and the specification data as training data.
A method for generating an estimation model of tire characteristics, comprising: further comprising a data extraction step;
the first data group and the second data group include, as part of the plurality of structure items, a structure item common to the first data group and the second data group;
In the data extraction step, data including specification data of the common structural items and the characteristic data is extracted from the tire data group;
The model generation step includes:
a source model generating step of generating a source model based on specification data of the common structural items using the extracted data as training data;
7. The tire characteristic estimation model generation method according to claim 6, further comprising: a target model generation step of generating the first estimation model by using the source model and the plurality of individual tire data in the first data group, and generating the second estimation model by using the source model and the plurality of individual tire data in the second data group. 7. The tire characteristic estimation model generating method according to claim 6, wherein the model generating step generates a plurality of third estimation models corresponding to a plurality of evaluation conditions, respectively, and generates a plurality of fourth estimation models corresponding to a plurality of evaluation conditions, respectively. a data acquisition means for acquiring a tire data group including a plurality of individual tire data, wherein each individual tire data includes characteristic data representing tire characteristics, specification data including data on a plurality of structure items which are a plurality of items representing the tire structure, and evaluation condition data which is a condition value indicating an evaluation condition for the tire characteristics ;
and a model generation means for generating an estimation model for estimating characteristic data based on tire specification data by using the plurality of individual tire data,
the tire data group includes a first data group and a second data group, each of which includes the plurality of individual tire data;
the first data group and the second data group differ in at least some of the plurality of structure items for which the specification data is provided,
the model generation means generates a first estimation model using the plurality of individual tire data included in the first data group as training data, and generates a second estimation model using the plurality of individual tire data included in the second data group as training data;
each of the first estimation model and the second estimation model includes a coefficient estimation model that outputs one or more coefficients that define a relational expression for calculating the characteristic data;
The model generation means
calculating the one or more coefficients based on the characteristic data of the plurality of individual tire data included in the first data group and the evaluation condition data, and generating the coefficient estimation model of the first estimation model using the calculated one or more coefficients and the specification data as training data;
Calculating the one or more coefficients based on the characteristic data and the evaluation condition data of the plurality of individual tire data included in the second data group, and generating the coefficient estimation model of the second estimation model using the calculated one or more coefficients and the specification data as training data.
A device for generating an estimation model of tire characteristics. a data acquisition means for acquiring a tire data group including a plurality of individual tire data, wherein each individual tire data includes characteristic data representing tire characteristics, specification data including data on a plurality of structure items which are a plurality of items representing the tire structure, and evaluation condition data which is a condition value indicating an evaluation condition for the tire characteristics ;
a computer that functions as a model generation means that generates an estimation model for estimating characteristic data based on tire specification data by using the plurality of individual tire data;
the tire data group includes a first data group and a second data group, each of which includes the plurality of individual tire data;
the first data group and the second data group differ in at least some of the plurality of structure items for which the specification data is provided,
the model generation means generates a first estimation model using the plurality of individual tire data included in the first data group as training data, and generates a second estimation model using the plurality of individual tire data included in the second data group as training data;
each of the first estimation model and the second estimation model includes a coefficient estimation model that outputs one or more coefficients that define a relational expression for calculating the characteristic data;
The model generation means
calculating the one or more coefficients based on the characteristic data of the plurality of individual tire data included in the first data group and the evaluation condition data, and generating the coefficient estimation model of the first estimation model using the calculated one or more coefficients and the specification data as training data;
Calculating the one or more coefficients based on the characteristic data and the evaluation condition data of the plurality of individual tire data included in the second data group, and generating the coefficient estimation model of the second estimation model using the calculated one or more coefficients and the specification data as training data.
A program that makes a computer function like this.