Computational model generation system, computational model generation method, and tire wear estimation system

Abstract

This technology provides an improved accuracy in estimating the wear condition of tires. [Solution] The vehicle information acquisition unit 12 of the calculation model generation system 110 acquires vehicle information, including the mileage measured by the vehicle. The wear estimation unit 13 estimates the wear state of the tire 7 using a learning-type calculation model 13a that takes the vehicle information as input data. The dataset generation unit 22 extracts a limited number of second datasets from a plurality of datasets composed of vehicle information and wear information as training data, where the amount of wear progressed is less than the second threshold R2 (less than the first threshold R1), according to the quantity of first datasets where the amount of wear progressed is greater than or equal to the first threshold R1, and generates a training dataset including the first dataset and the extracted second dataset. The learning processing unit 23 trains the calculation model 13a based on the training dataset generated by the dataset generation unit 22.

Description

Technical Field

[0001] The present invention relates to an arithmetic model generation system for generating an arithmetic model for estimating the wear state of a tire mounted on a vehicle, an arithmetic model generation method, and a tire wear estimation system.

Background Art

[0002] Generally, the wear of a tire progresses according to the driving state, driving distance, etc. In recent years, the development of a technique for estimating tire wear by using an arithmetic model with information measured by a vehicle as input data has been advanced.

[0003] Patent Document 1 describes a conventional arithmetic model generation system. In the arithmetic model generation system, a vehicle information acquisition unit acquires vehicle information including the driving distance measured by the vehicle. A wear information acquisition unit acquires information on the wear state measured by the tire. A wear estimation unit estimates the wear state of the tire by using a learning-type arithmetic model with the vehicle information as input data. A data set generation unit acquires a first data set composed of the vehicle information and the information on the wear state acquired by the wear information acquisition unit as teacher data, and generates a second data set based on a plurality of first data sets. A learning processing unit learns an arithmetic model based on the first data set and the second data set.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] By the way, in order to continuously measure the wear state of a tire from the initial wear stage when it is new to the final wear stage when the remaining tread depth approaches its limit and obtain a data set, it is necessary to manage the tire over a long period of time. In reality, there are cases where used tires with unknown wear history are used, or where tires are discarded before reaching the final wear stage, and the wear state at the final wear stage is not measured.

[0006] Due to the cases described above, the amount of data in the dataset representing wear that has progressed over a long period is small, while the amount of data representing wear that has progressed over a short period is excessive, resulting in an imbalance in the training dataset. The inventors of this invention have realized that when estimating the wear state of a tire based on a learning-type computational model, there is a problem that the accuracy of estimating the wear state at the end of wear may deteriorate compared to the initial wear state due to this imbalance in the dataset.

[0007] This invention has been made in view of the above circumstances, and its objective is to provide a computational model generation system, a computational model generation method, and a tire wear estimation system that can improve the accuracy of estimating the wear state of tires. [Means for solving the problem]

[0008] A computational model generation system according to one aspect of the present invention includes: a vehicle information acquisition unit that acquires vehicle information including mileage measured by a vehicle; a wear estimation unit that estimates the tire wear state using a learning-type computational model that takes the vehicle information as input data; a dataset generation unit that extracts from a plurality of datasets composed of the vehicle information and wear information as training data, limiting the number of second datasets where the amount of wear progressed is less than a second threshold (less than the first threshold) according to the number of first datasets where the amount of wear progressed is greater than or equal to a first threshold, and generates a learning dataset including the first dataset and the extracted second dataset; and a learning processing unit that trains the computational model based on the learning dataset generated by the dataset generation unit.

[0009] Another aspect of the present invention is a method for generating a computational model. The method for generating a computational model includes: a vehicle information acquisition step of acquiring vehicle information including the mileage measured by the vehicle; a wear estimation step of estimating the tire wear state using a learning-type computational model that takes the vehicle information as input data; a dataset generation step of extracting from a plurality of datasets composed of the vehicle information and wear information as training data, limiting the number of second datasets whose advanced wear amount is less than a second threshold (where the advanced wear amount is less than the first threshold) according to the number of first datasets whose advanced wear amount is greater than or equal to a first threshold, and generating a training dataset including the first dataset and the extracted second dataset; and a learning process step of training the computational model based on the training dataset generated by the dataset generation step.

[0010] Another aspect of the present invention is a tire wear estimation system. The tire wear estimation system comprises a vehicle information acquisition unit that acquires vehicle information including the mileage measured by the vehicle, and a wear estimation unit that estimates the tire wear state using a learning-type computation model that takes the vehicle information as input data. The computation model is trained based on a learning dataset which includes the first dataset and the extracted second dataset, and is selected from a dataset consisting of the vehicle information and wear information as training data, such that the number of second datasets whose advanced wear amount is less than a second threshold is limited according to the number of first datasets whose advanced wear amount is equal to or greater than a first threshold, and the number of second datasets whose advanced wear amount is less than a second threshold, which is less than the first threshold.

[0011] Another aspect of the present invention, a tire wear estimation method, comprises a vehicle information acquisition unit that acquires vehicle information including the mileage measured by the vehicle, and a wear estimation unit that estimates the tire wear state using a learning-type calculation model that takes the vehicle information, the groove depth of the tire in a new state, and the starting remaining groove amount as input data. [Effects of the Invention]

[0012] According to the present invention, the accuracy of estimating the wear state of tires can be improved. [Brief explanation of the drawing]

[0013] [Figure 1] This is a block diagram showing the functional configuration of the tire wear estimation system according to the embodiment. [Figure 2] This is a block diagram showing the functional configuration of an in-vehicle measurement device. [Figure 3] This is a schematic diagram illustrating wear estimation and learning of the computational model. [Figure 4] This block diagram shows the functional configuration of the computational model generation system. [Figure 5] This is a diagram illustrating an example of classifying the acquired dataset. [Figure 6] This figure shows an example of a dataset extracted by the dataset generation unit. [Figure 7] This flowchart shows the procedure for tire wear estimation processing by the tire wear estimation system. [Figure 8] This flowchart shows the procedure for generating computational models using the computational model generation system. [Figure 9] Figure 5 shows an example of the wear estimation accuracy of a computational model trained using all the datasets shown. [Figure 10] Figure 6 shows an example of the wear estimation accuracy of a computational model trained using the extracted dataset. [Modes for carrying out the invention]

[0014] The present invention will be described below with reference to Figures 1 to 10, based on preferred embodiments. The same or equivalent components and members shown in each drawing will be denoted by the same reference numerals, and redundant explanations will be omitted as appropriate. Furthermore, the dimensions of the members in each drawing will be enlarged or reduced as appropriate for ease of understanding. Additionally, some members that are not important for explaining the embodiments will be omitted from the drawings.

[0015] (Embodiment) FIG. 1 is a block diagram showing the functional configuration of a tire wear estimation system 100 according to an embodiment. The tire wear estimation system 100 includes an in-vehicle measurement device 70 mounted on a vehicle, a weather information server device 80, and a wear estimation device 10 that estimates the wear state of each tire 7 mounted on the vehicle.

[0016] The wear estimation device 10 acquires vehicle measurement information such as the speed, acceleration, and position information of the vehicle, and tire measurement information measured by the tire 7, from the in-vehicle measurement device 70 mounted on the vehicle via a communication network 9 such as the Internet. The wear estimation device 10 also acquires weather information from the weather information server device 80. The wear estimation device 10 performs calculations using a learning-based calculation model based on the acquired information to estimate the wear state of each tire 7. The wear state of the tire 7 estimated by the wear estimation device 10 is represented by information such as the wear amount and wear rate of the tire 7. The wear state of the tire 7 may be, for example, the amount of wear (a value such as 1 mm), or the ratio of the amount of wear to the initial groove depth in the tire groove (a value such as 10%). Wear estimation means estimating the wear state of the tire 7, that is, information such as the wear amount and wear rate of the tire 7.

[0017] FIG. 2 is a block diagram showing the functional configuration of the in-vehicle measurement device 70. The in-vehicle measurement device 70 includes a vehicle measurement unit 71, a tire measurement unit 72, an information acquisition unit 73, and a communication unit 74. Each part in the in-vehicle measurement device 70 can be realized hardware-wise by electronic elements such as a computer's CPU and mechanical parts, and software-wise by a computer program etc. Here, however, functional blocks realized by their cooperation are depicted. Therefore, it is understood by those skilled in the art that these functional blocks can be realized in various forms by combinations of hardware and software.

[0018] The vehicle measurement unit 71 includes a speedometer 71a, a GPS receiver 71b, and an acceleration sensor 71c mounted on the vehicle. The speedometer 71a measures the running speed of the vehicle. The GPS receiver 71b measures the current position information (latitude, longitude, and altitude) of the vehicle. The acceleration sensor 71c measures the acceleration of the vehicle in three axial directions. The three axial directions are, for example, the front-rear direction, the left-right direction, and the up-down direction of the vehicle.

[0019] The tire measurement unit 72 includes a temperature sensor 72a and a pressure sensor 72b. The temperature sensor 72a and the pressure sensor 72b are arranged at the air valve or the like of the tire 7 mounted on the vehicle, or are firmly wound around and fixed to the wheel with a belt or the like, and measure the temperature and air pressure of the tire 7. The temperature sensor 72a may be arranged at the inner liner or the like of the tire 7. Incidentally, the acceleration sensor 71c may be arranged at the inner liner of the tire 7.

[0020] The information acquisition unit 73 acquires vehicle measurement information (running speed, position information, acceleration, etc.) measured by the vehicle measurement unit 71, tire measurement information (tire temperature and air pressure, etc.) measured by the tire measurement unit 72, and tire identification information described later. The information acquisition unit 73 associates the measured time information or the acquired time information with each measurement data included in the vehicle measurement information and the tire measurement information. The information acquisition unit 73 transmits the vehicle measurement information and the tire measurement information together with the time information associated with each measurement data to the wear estimation device 10 via the communication unit 74.

[0021] When an electronic control device of the vehicle or a device such as a digital tachometer is mounted on the vehicle, the information acquisition unit 73 may acquire the running speed, acceleration, position information, etc. of the vehicle collected by the device. The communication unit 74 communicatively connects to the communication network 9 by wireless communication such as WiFi (registered trademark), and transmits the vehicle measurement information, the tire measurement information, and the time information acquired by the information acquisition unit 73 to the wear estimation device 10 via the communication network 9.

[0022] Returning to Figure 1, the weather information server device 80 provides weather information for each location. The weather information provided by the weather information server device 80 includes information such as precipitation, snow depth, snowfall, temperature, and sunshine duration for each location. The wear estimation device 10 obtains weather information for the location where the vehicle is traveling from the weather information server device 80.

[0023] The wear estimation device 10 comprises a communication unit 11, a vehicle information acquisition unit 12, a wear estimation unit 13, and a storage unit 14. Each part of the wear estimation device 10 can be implemented in hardware terms using electronic elements such as a computer CPU and mechanical parts, and in software terms using computer programs, etc. However, here we are describing the functional blocks that are realized through the cooperation of these components. Therefore, it will be understood by those skilled in the art that these functional blocks can be implemented in various forms by combining hardware and software.

[0024] The communication unit 11 connects to the communication network 9 via wireless or wired communication and communicates with the communication unit 74 of the in-vehicle measuring device 70. The communication unit 11 also communicates with the weather information server device 80 via the communication network 9.

[0025] The vehicle information acquisition unit 12 acquires vehicle measurement information (driving speed, position information, acceleration, etc.) and tire measurement information (tire temperature and air pressure, etc.) transmitted from the on-board measuring device 70 mounted on the vehicle. Based on the vehicle measurement information, the vehicle information acquisition unit 12 calculates and acquires the vehicle's mileage.

[0026] The vehicle information acquisition unit 12 can calculate and acquire the mileage based on the location information of the vehicle measurement information. Alternatively, the mileage of the vehicle may be calculated based on the speed data in the vehicle measurement information and the time data associated with that data. That is, the mileage of the vehicle can be calculated by multiplying the chronologically arranged speed data by the time difference until the next point in time. The vehicle's speed may be calculated from the mileage of the vehicle based on the chronologically arranged location information and the location information acquisition interval.

[0027] The vehicle information acquisition unit 12 does not need to calculate the mileage itself if information regarding the vehicle's mileage is provided by the vehicle or an external device for vehicle management, and may acquire the mileage information from the vehicle or an external device.

[0028] The vehicle information acquisition unit 12 outputs the acquired mileage to the wear estimation unit 13. The vehicle information acquisition unit 12 outputs the acquired tire measurement information (tire temperature and air pressure, etc.) to the wear estimation unit 13. The vehicle information acquisition unit 12 outputs the acceleration information from the vehicle measurement information to the wear estimation unit 13.

[0029] The vehicle information acquisition unit 12 acquires data from the storage unit 14 that is used to estimate the wear state of the tire 7, from among the vehicle specification data 14a, tire specification data 14b, and tire position data 14c, and outputs it to the wear estimation unit 13. The storage unit 14 is a storage device composed of, for example, an SSD (Solid State Drive), a hard disk, a CD-ROM, a DVD, etc., and stores data that has been provided in advance regarding the specifications of various vehicles and tires 7.

[0030] Vehicle specification data 14a includes information about the vehicle's performance, such as manufacturer, vehicle name, vehicle model, vehicle weight, drivetrain, overall length, vehicle width, vehicle height, and maximum load capacity. Tire specification data 14b includes information about the tire 7's performance, such as manufacturer, product name, tire size, tire width, aspect ratio, groove depth in new condition, wear resistance, tire strength, static stiffness, dynamic stiffness, tire outer diameter, load index, and manufacturing date. Tire position data 14c includes the position of the tire to be worn on the vehicle, tire identification information, and information about the axle to which it is mounted. Tire identification information is a serial number, such as a manufacturing number, assigned to each tire to identify it. The tire identification information, tire position, and axle information can be stored in the storage unit 14, for example, by an operator inputting the information when mounting the tires on the vehicle or by reading an RFID tag. In this embodiment, tire identification information is used, but a wear amount dataset can be generated without using tire identification information by measuring the starting tread depth.

[0031] The wear estimation unit 13 has a calculation model 13a and estimates the wear state of the tire 7. The calculation model 13a is a learning model that calculates the wear state of the tire 7 (information such as wear amount and wear rate) based on the input information. Figure 3 is a schematic diagram to explain the wear estimation and learning of the calculation model 13a. The input data to the calculation model 13a is generally classified into vehicle measurement information, tire measurement information, and other information systems.

[0032] The input data related to vehicle measurement information includes the vehicle's acceleration and mileage. The mileage is acquired by the vehicle information acquisition unit 12 as described above. The input data related to tire measurement information includes the temperature and air pressure of the tire 7.

[0033] Other input data includes road surface conditions estimated based on weather information, temperature and precipitation, the maximum load capacity of the vehicle included in the vehicle specification data 14a, and the groove depth and wear resistance of the new tire 7 included in the tire specification data 14b. The wear resistance of the tire 7 is determined using, for example, a tire wear index value that quantifies the wear resistance of various tread compounds based on the Lambourn wear test, with the standard compound set to 100. Other input data includes the position of the tire 7 included in the tire position data 14c, tire identification information, and axle information.

[0034] The computational model 13a uses a learning model such as a neural network. The computational model 13a is constructed using methods such as a Deep Neural Network (DNN) or a decision tree. Alternatively, the computational model 13a may be a multilinear regression model on input information, and the model may be generated through learning.

[0035] Figure 4 is a block diagram showing the functional configuration of the computational model generation system 110. In addition to the configuration of the tire wear estimation system 100, the computational model generation system 110 includes a tire wear measuring device 60 and a computational model generation device 20 having a learning processing unit 23, etc.

[0036] The tire wear measuring device 60 directly measures the depth of the grooves in the tread of the tire 7 and acquires information on the wear condition of the tire 7. Alternatively, the operator may measure the depth of each groove using measuring instruments, cameras, or visual inspection, and the tire wear measuring device 60 may store the measurement data entered by the operator. In addition, the tire wear measuring device 60 may be a dedicated device that measures the depth of the grooves by mechanical or optical methods and stores information on the wear condition.

[0037] Specifically, the tire wear measuring device 60 measures at four points in the width direction if the tire has four grooves, and also measures at three points in the circumferential direction of the same groove, for example, at 120° intervals. This allows the tire wear measuring device 60 to store uneven wear data in the width direction or circumferential direction of the tire. Furthermore, since the diameter changes as the tire wears down, the tire wear measuring device 60 may indirectly measure the groove depth by calculation from the mileage and tire rotation speed / velocity information. In addition, a method that directly measures the groove depth may be used in combination with a method that predicts the groove depth by calculation from the mileage and tire rotation speed / velocity.

[0038] The computational model generation device 20 includes, in addition to the components of the wear estimation device 10, a wear information acquisition unit 21, a dataset generation unit 22, and a learning processing unit 23. The parts of the computational model generation device 20 that correspond to the components of the wear estimation device 10 have the same functions as those of the wear estimation device 10, but the computational model 13a is either pre-training or in the process of training.

[0039] The wear information acquisition unit 21 acquires information on the wear status of each tire 7 mounted on the vehicle from the tire wear measuring device 60 via the communication unit 11 and outputs it to the data set generation unit 22. The information on the wear status of the tires 7 includes data such as the amount of wear and wear rate of the tire grooves measured for each tire 7.

[0040] The dataset generation unit 22 acquires vehicle information from the vehicle information acquisition unit 12 as input data for the calculation model 13a, and acquires information on the wear status of the tires 7 from the wear information acquisition unit 21. The vehicle information acquired from the vehicle information acquisition unit 12, as described above, consists of vehicle measurement information, tire measurement information, and other information as input data for the calculation model 13a. The information on the wear status of the tires 7 acquired by the dataset generation unit 22 is used as training data when training the calculation model 13a.

[0041] The dataset generation unit 22 acquires a group of datasets including vehicle measurement information, tire measurement information, other information, and wear state information for each tire 7 mounted on the vehicle. Each dataset included in the group of datasets is acquired for each tire. For example, if a total of 120 tires 7 are mounted on 10 truck vehicles and one dataset is generated for each tire 7, the group of datasets will be composed of 120 datasets.

[0042] For the tire 7, it is desirable to acquire a dataset by continuously measuring the wear state from the initial wear state of a new tire to the final wear state where the remaining groove amount approaches the limit value. In reality, there are cases where the tire cannot be tracked by changing summer and winter tires when the tire identification information is not registered, cases where used tires with unknown wear state transitions are used, and cases where the tire is discarded due to the discovery of external damage visually before reaching the final wear state and the wear state at the final wear state is not measured. Therefore, the datasets acquired by the dataset generation unit 22 include datasets with a large amount of wear that has progressed over a long period and datasets with only a small amount of wear that has progressed over a short period.

[0043] The dataset generation unit 22 divides the acquired multiple datasets into a first dataset with a wear amount that has progressed beyond a first threshold value R1, a second dataset with a wear amount less than a second threshold value R2 (<R1), and a third dataset with a wear amount that is equal to or greater than the second threshold value R2 and less than the first threshold value R1.

[0044] For example, when the groove depth of the new tire state of the tire 7 is 18 mm, the first threshold value R1 is set to 11 mm, and the second threshold value R2 is set to 5 mm, etc. The first dataset is a dataset with a wear progression of the first threshold value R1 (=11 mm) or more, the second dataset is a dataset with a wear progression of less than the second threshold value R2 (=5 mm), and the third dataset is a dataset with a wear progression of the second threshold value R2 (=5 mm) or more and less than the first threshold value R1 (=11 mm). Note that the values of the first threshold value R1 and the second threshold value R2 are not limited to the above example.

[0045] Figure 5 is a diagram showing an example of classifying acquired datasets. In Figure 5, there are 346 datasets classified as the first dataset, for example, with a wear progression of 11 mm. In contrast, there are 43,627 datasets classified as the second dataset, for example, with a wear progression of 1 mm. As shown in Figure 5, the quantity of the second dataset is significantly larger than the quantity of the first dataset, resulting in an imbalance in the datasets. In the example shown in Figure 5, regardless of whether the tire is new or used, the wear progression from the start of measurement is divided into units of wear (e.g., 1 mm), and the datasets are classified accordingly. For new tires, tire identification information is acquired and the groove depth in the new state is used as the starting remaining groove amount. For used tires, tire identification information is acquired and the current remaining groove amount of the tire is measured as the starting remaining groove amount.

[0046] The first dataset, which includes a wear progression of 11 mm, contains both a dataset showing a decrease in tread depth from 18 mm for new tires to 7 mm (starting tread depth of 18 mm) and a dataset showing a decrease in tread depth from 14 mm for used tires to 3 mm (starting tread depth of 14 mm). The second dataset, which includes a wear progression of 1 mm, also contains both a dataset showing a decrease in tread depth from 18 mm for new tires to 17 mm and a dataset showing a decrease in tread depth from 4 mm for used tires to 3 mm.

[0047] The dataset generation unit 22 extracts the second dataset with a limited quantity according to the quantity of the first dataset. The dataset generation unit 22 randomly extracts the second dataset within the limited quantity range.

[0048] Figure 6 is a diagram showing an example of a dataset extracted by the dataset generation unit 22. In the example shown in Figure 6, the number of datasets classified as the first dataset, with a wear progression of 11 mm, is 346 sets. The limit on the number of datasets included in each classification for unit wear amount (1 mm) is set to 500 sets, and the second and third datasets are randomly extracted.

[0049] Looking at the entire first dataset, the total number of sets classified as having a wear progression between 11 mm and 15 mm is 645. Looking at the entire second dataset, the number of sets classified as having a wear progression between 1 mm and 4 mm is limited to 2000 sets. And looking at the entire third dataset, the number of sets classified as having a wear progression between 5 mm and 10 mm is limited to 3000 sets.

[0050] For example, if the second threshold R2 is set to 6 mm, the total number of sets in the second dataset is limited to 2500 sets where the wear progression is classified from 1 mm to 5 mm, and the total number of sets in the third dataset is limited to 2500 sets where the wear progression is classified from 6 mm to 10 mm. The limits on the number of sets in the second and third datasets can be specified by the user, or preset values ​​can be used.

[0051] The learning processing unit 23 trains the computational model 13a based on the training dataset, which includes the first dataset, the second dataset, and the third dataset obtained from the dataset generation unit 22. For each set of the first, second, and third datasets, the learning processing unit 23 compares the wear state of the tire 7 estimated by the computational model 13a based on the input data with the wear state information used as training data.

[0052] The learning processing unit 23 performs learning by repeatedly updating the model, setting various coefficients in the calculation process, such as weighting, in the calculation model 13a based on the comparison result between the estimated wear state and the training data. The tire wear estimation system 100 estimates the wear state of the tire 7 using the calculation model 13a that has been trained by the calculation model generation system 110. In the learning process of the calculation model 13a, known learning methods such as gradient boosting can be used. In addition, known validation methods such as random data sampling and cross-validation can be used to validate the calculation model 13a.

[0053] Next, the operations of the tire wear estimation system 100 and the calculation model generation system 110 will be described. FIG. 7 is a flowchart showing the procedure of the wear estimation process by the tire wear estimation system 100. The vehicle information acquisition unit 12 starts acquiring vehicle information such as vehicle measurement information and tire measurement information (S1). Also, in step S1, the vehicle information acquisition unit 12 reads out necessary information such as vehicle specifications, tire specifications, tire position, maximum load of the vehicle, and wear resistance performance of the tire from the storage unit 14 as other information. The vehicle information acquisition unit 12 starts calculating the travel distance (S2).

[0054] The wear estimation unit 13 acquires the input data from the vehicle information acquisition unit 12, estimates the wear state of the tire 7 by the calculation model 13a (S3), and ends the process. The calculation model 13a uses the learned calculation model generated by the calculation model generation system 110.

[0055] FIG. 8 is a flowchart showing the procedure of the generation process of the calculation model 13a by the calculation model generation system 110. The processes from step S11 to step S12 shown in FIG. 8 are equivalent to the processes from step S1 to step S2 shown in FIG. 7, and the description is omitted for the sake of brevity. The wear information acquisition unit 21 of the calculation model generation device 20 acquires information on the wear state of each tire 7 from the tire wear measurement device 60 (S13).

[0056] The dataset generation unit 22 acquires a group of datasets including vehicle measurement information, tire measurement information, other information, and wear state information for each tire 7 mounted on the vehicle (S14). The dataset generation unit 22 classifies the acquired multiple datasets into a first dataset with a wear amount progressed being equal to or greater than the first threshold value R1, a second dataset with a wear amount less than the second threshold value R2 (<R1), and a third dataset with a wear amount equal to or greater than the second threshold value R2 and less than the first threshold value R1 (S15).

[0057] The dataset generation unit 22 sets limits on the quantities of the second and third datasets according to the quantity of the first dataset (S16). The limits on the quantities of the second and third datasets may be specified by the user, or preset values ​​may be used.

[0058] The dataset generation unit 22 randomly selects the second and third datasets according to the limit set in step S16 (S17). The dataset generation unit 22 generates a training dataset that includes the first dataset and the second and third datasets selected in step S17 (S18).

[0059] The wear estimation unit 13 inputs vehicle information, such as vehicle measurement information and tire measurement information, for each dataset included in the training dataset generated in step S18 into the calculation model 13a and estimates the wear state of the tire 7 (S19). The learning processing unit 23 compares the wear state of the tire 7 estimated by the calculation model 13a with the measured wear state of the tire 7 as training data (S20). The learning processing unit 23 updates the calculation model 13a based on the comparison result from step S20 (S21).

[0060] The learning processing unit 23 determines whether or not training has been performed on all training datasets (S22). If it is determined in step S22 that training has not been performed on all training datasets (S22: NO), the process returns to step S19 and is repeated. If it is determined in step S22 that training has been performed on all training datasets (S22: YES), the process terminates.

[0061] Figure 9 is a diagram showing an example of the wear estimation accuracy of the computational model 13a trained on all the datasets shown in Figure 5. In the example shown in Figure 9, as mentioned above, there are 346 datasets with a wear progression of 11 mm and 43,627 datasets with a wear progression of 1 mm, resulting in an imbalance in the number of datasets. Therefore, it can be seen that the wear estimation accuracy is best from 1 mm to 4 mm of wear, and deteriorates as the wear progresses from 5 mm to 10 mm, and further from 11 mm to 15 mm.

[0062] Figure 10 is a diagram showing an example of the wear estimation accuracy of the computational model 13a trained using the dataset extracted in Figure 6. In the example shown in Figure 10, it can be seen that the wear estimation accuracy deteriorates slightly from 1 mm to 4 mm compared to the example shown in Figure 9, but becomes equivalent or better from 5 mm to 10 mm, and further improves as the wear progresses to 11 mm to 15 mm.

[0063] Specifically, the estimation error when wear has progressed to 2 mm is 0.34 mm in the example shown in Figure 9, while it deteriorates to 0.54 mm in the example shown in Figure 10. When wear has progressed to 7 mm, the estimation error is 0.75 mm in the example shown in Figure 9, while it improves to 0.62 mm in the example shown in Figure 10. When wear has progressed to 12 mm, the estimation error is 1.09 mm in the example shown in Figure 9, while it improves to 0.61 mm in the example shown in Figure 10. Furthermore, compared to the example shown in Figure 9, the variation in estimation error throughout the entire wear progression is smaller in the example shown in Figure 10.

[0064] Considering that tire wear estimation is performed to determine when to replace the tires, it is preferable that the estimation accuracy is good when the tire is in the final stages of wear, for example, when the wear has progressed from 11 mm to 15 mm.

[0065] In the examples shown in Figures 6 and 10, the first threshold R1 is set to 11 mm, given that the groove depth of the new tire 7 is 18 mm. Assuming that the tire is replaced when the remaining groove depth of tire 7 is about 3 mm, the allowable wear amount for tire 7 is up to 15 mm. The first threshold R1 (=11 mm) is located at a point where approximately 73% of the wear progression has progressed to 15 mm. By setting the first threshold R1 to a value greater than 70% of the allowable wear amount before tire replacement, it is possible to improve the estimation error in the final stages of wear when tire replacement is imminent.

[0066] Furthermore, in order to further improve the estimation error in the final stages of wear when tire replacement is imminent, it is desirable to set the first threshold R1 to a value of 80% or higher, which is even greater than the 70% of the amount of wear that can be tolerated before tire replacement. However, this tends to reduce the number of datasets classified as the first dataset, making it difficult to secure a sufficient amount of training dataset.

[0067] The computational model generation system 110 comprises a vehicle information acquisition unit 12, a wear estimation unit 13, a dataset generation unit 22, and a learning processing unit 23. The vehicle information acquisition unit 12 acquires vehicle information, including the mileage measured by the vehicle. The wear estimation unit 13 estimates the wear state of the tire 7 using a learning-type computational model 13a that takes the vehicle information as input data. The dataset generation unit 22 extracts data from multiple datasets, which consist of vehicle information and wear information as training data, so as to limit the number of second datasets where the amount of wear is less than the second threshold R2 (less than the first threshold R1) according to the number of first datasets where the amount of wear is greater than or equal to the first threshold R1, and generates a learning dataset including the first dataset and the extracted second dataset. The learning processing unit 23 trains the computational model 13a based on the learning dataset generated by the dataset generation unit 22. As a result, the computational model generation system 110 can adjust the balance of the datasets and improve the accuracy of estimating the tire wear state.

[0068] The dataset generation unit 22 extracts a third dataset in which the amount of wear progressed is greater than or equal to the second threshold R2 and less than the first threshold R1, limiting the quantity of the third dataset according to the quantity of the first dataset, and generates a training dataset that includes the extracted third dataset. This allows the computational model generation system 110 to adjust the quantity of the dataset in the middle stage of wear and improve the accuracy of estimating the tire wear state.

[0069] The dataset generation unit 22 can improve the estimation accuracy in the final stages of wear, when tire replacement is imminent, by setting the first threshold R1 to a value greater than 70% of the amount of wear allowed before tire replacement.

[0070] The calculation model 13a may also take the groove depth and starting groove amount of the tire 7 in its new condition as input data. The starting groove amount should be the most recent data actually measured for the tire 7, or a value obtained by subtracting the amount of wear estimated in the past. By knowing the groove depth and starting groove amount in the new condition, it is possible to determine whether the wear progression is in the early, middle, or final stages of wear, and the calculation model generation system 110 can generate a calculation model 13a that estimates the wear state of the tire 7, taking into account the degree of wear progression.

[0071] The calculation model generation method comprises a vehicle information acquisition step, a wear estimation step, a dataset generation step, and a learning process step. The vehicle information acquisition step acquires vehicle information, including the mileage measured by the vehicle. The wear estimation step estimates the tire wear state using a learning-type calculation model 13a that takes the vehicle information as input data. The dataset generation step extracts data from multiple datasets, which consist of vehicle information and wear information as training data, so as to limit the number of second datasets where the amount of wear is less than the second threshold R2 (less than the first threshold R1) according to the number of first datasets where the amount of wear is greater than or equal to the first threshold R1, and generates a learning dataset that includes the first dataset and the extracted second dataset. The learning process step trains the calculation model 13a based on the learning dataset generated by the dataset generation step. This method allows for adjusting the balance of the datasets and improving the accuracy of estimating the tire wear state.

[0072] The tire wear estimation system 100 includes a vehicle information acquisition unit 12 and a wear estimation unit 13. The vehicle information acquisition unit 12 acquires vehicle information, including the mileage measured by the vehicle. The wear estimation unit 13 estimates the tire wear state using a learning-type computation model 13a that takes the vehicle information as input data. The computation model 13a of the wear estimation unit 13 is trained on a training dataset, which is a dataset consisting of vehicle information and wear information as training data. The training dataset is extracted such that the quantity of a second dataset where the amount of wear is less than a second threshold R2 (less than the first threshold R1) is limited according to the quantity of a first dataset where the amount of wear is greater than or equal to a first threshold R1. The system includes the first dataset and the extracted second dataset. As a result, the tire wear estimation system 100 can improve the accuracy of estimating the tire wear state by using a computation model 13a trained on a training dataset with a balanced dataset.

[0073] (modified version) The wear estimation unit 13 of the tire wear estimation system 100 may use a learning-type calculation model 13a that takes vehicle information, the groove depth of the tire 7 in its new state, and the starting remaining groove amount as input data. The starting remaining groove amount may be the latest data actually measured for the tire 7 or the value obtained by subtracting the estimated wear amount. This is also true in the above embodiment when the wear estimation unit 13 uses a learning-type calculation model 13a that takes vehicle information, the groove depth of the tire 7 in its new state, and the starting remaining groove amount as input data. For example, if the amount of wear is estimated every month, the remaining groove amount for the previous month is calculated based on the latest remaining groove amount measured on the tire 7 (for example, measured 3 months ago) and the wear amount estimated up to the previous month. The tire wear estimation system 100 may use the remaining groove amount for the previous month as input data for the calculation model 13a and estimate the amount of wear on the tire 7 due to vehicle driving over the one month since the previous month.

[0074] The tire wear estimation system 100 acquires vehicle information, including the mileage measured by the vehicle, using the vehicle information acquisition unit 12. The wear estimation unit 13 uses a learning-type calculation model 13a, which takes the vehicle information, the groove depth of the tire 7 in its new state, and the starting groove amount as input data, to estimate the wear state of the tire 7. As a result, the tire wear estimation system 100 can estimate the wear state of the tire 7 by taking into account the groove depth in its new state and the starting groove amount, and whether the wear progression is in the early, middle, or final stages of wear.

[0075] The technical ideas embodied in the above embodiments and variations can be generalized to include the technical ideas described in the following items.

[0076] The first item is a computational model generation system comprising: a vehicle information acquisition unit that acquires vehicle information including the mileage measured by the vehicle; a wear estimation unit that estimates the tire wear state using a learning-type computational model that takes the vehicle information as input data; a dataset generation unit that extracts from a plurality of datasets composed of the vehicle information and wear information as training data, limiting the number of second datasets where the amount of wear progressed is less than a second threshold, according to the number of first datasets where the amount of wear progressed is greater than or equal to a first threshold, and generates a learning dataset including the first dataset and the extracted second dataset; and a learning processing unit that trains the computational model based on the learning dataset generated by the dataset generation unit.

[0077] The second item is a computational model generation system as described in the first item, wherein the dataset generation unit extracts a quantity of third datasets in which the amount of further wear is greater than or equal to the second threshold and less than the first threshold, in accordance with the quantity of the first dataset, and generates the training dataset including the extracted third dataset.

[0078] The third item is the calculation model generation system described in item 1 or 2, wherein the first threshold is set to a value greater than 70% of the amount of wear allowed before tire replacement.

[0079] The fourth item is a calculation model generation system described in one of the first three items, which takes the groove depth and starting groove amount of a new tire as input data.

[0080] The fifth item is a method for generating a computational model, comprising: a vehicle information acquisition step of acquiring vehicle information including the mileage measured by the vehicle; a wear estimation step of estimating the tire wear state using a learning-type computational model that takes the vehicle information as input data; a dataset generation step of extracting from a plurality of datasets composed of the vehicle information and wear information as training data, such that the quantity of second datasets whose advanced wear amount is less than a second threshold (where the advanced wear amount is less than the first threshold) is limited according to the quantity of first datasets whose advanced wear amount is greater than or equal to a first threshold, and generating a training dataset including the first dataset and the extracted second dataset; and a learning process step of training the computational model based on the training dataset generated by the dataset generation step.

[0081] The sixth item is a tire wear estimation system comprising a vehicle information acquisition unit that acquires vehicle information including mileage measured by the vehicle, and a wear estimation unit that estimates the tire wear state using a learning-type calculation model that takes the vehicle information as input data, wherein the calculation model is extracted from a dataset consisting of the vehicle information and wear information as training data, such that the number of second datasets whose advanced wear amount is less than a second threshold is limited according to the number of first datasets whose advanced wear amount is greater than or equal to a first threshold, and the system is trained based on a learning dataset that includes the first dataset and the extracted second dataset.

[0082] The seventh item is a tire wear estimation system comprising a vehicle information acquisition unit that acquires vehicle information including the mileage measured by the vehicle, and a wear estimation unit that estimates the tire wear state using a learning-type calculation model that takes the vehicle information, the groove depth of the tire in a new state, and the starting remaining groove amount as input data.

[0083] The embodiments of the present invention have been described above. These embodiments are illustrative, and it will be understood by those skilled in the art that various modifications and changes are possible within the scope of the claims of the present invention, and that such modifications and changes are also within the scope of the claims of the present invention. Accordingly, the descriptions and drawings herein should be treated as illustrative rather than limiting. [Explanation of Symbols]

[0084] 7 Tires, 12 Vehicle information acquisition unit, 13 Wear estimation unit, 13a Calculation model, 22 Dataset generation unit, 23 Learning processing unit, 100 Tire wear estimation system, 110 Computation model generation system.

Claims

1. A vehicle information acquisition unit that acquires vehicle information including the mileage measured by the vehicle, A wear estimation unit that estimates the tire wear state using a learning-type computation model that takes the aforementioned vehicle information as input data, A dataset generation unit extracts data sets from a plurality of datasets consisting of the vehicle information and wear information as training data, such that the number of second datasets with a wear amount less than a second threshold (where the wear amount is less than the first threshold) is limited according to the number of first datasets with a wear amount greater than or equal to a first threshold, and generates a training dataset including the first dataset and the extracted second dataset. A learning processing unit that trains the computation model based on the learning dataset generated by the dataset generation unit, A computational model generation system equipped with the following features.

2. The computational model generation system according to claim 1, wherein the dataset generation unit extracts a number of third datasets in which the amount of further wear is greater than or equal to the second threshold and less than the first threshold, in accordance with the number of first datasets, and generates the training dataset including the extracted third datasets.

3. The calculation model generation system according to claim 1, wherein the first threshold is set to a value greater than 70% of the amount of wear that can be tolerated before tire replacement.

4. The calculation model generation system according to claim 1, wherein the calculation model uses the groove depth and starting groove amount of a new tire as input data.

5. A vehicle information acquisition step that acquires vehicle information including the mileage measured by the vehicle, A wear estimation step in which the wear state of the tires is estimated using a learning-type computation model that takes the aforementioned vehicle information as input data, A dataset generation step in which, from among multiple datasets consisting of the vehicle information and wear information as training data, the number of second datasets where the amount of wear progressed is less than a second threshold (less than the first threshold) is limited according to the number of first datasets where the amount of wear progressed is greater than or equal to a first threshold, and a training dataset is generated that includes the first dataset and the extracted second dataset. A learning process step in which the computational model is trained based on the training dataset generated by the dataset generation step, A method for generating computational models that includes the following features.

6. A vehicle information acquisition unit that acquires vehicle information including the mileage measured by the vehicle, The system includes a wear estimation unit that estimates the tire wear state using a learning-type computation model that takes the vehicle information as input data, The calculation model is a tire wear estimation system in which, from a dataset consisting of vehicle information and wear information as training data, the quantity of a second dataset with a wear amount less than a second threshold (where the wear amount is less than the first threshold) is limited according to the quantity of a first dataset with a wear amount greater than or equal to a first threshold, and the system is trained on a training dataset including the first dataset and the extracted second dataset.

7. A vehicle information acquisition unit that acquires vehicle information including the mileage measured by the vehicle, A wear estimation unit estimates the tire wear state using a learning-type computation model that takes the aforementioned vehicle information, the groove depth of the tire in its new state, and the starting groove amount as input data. A tire wear estimation system equipped with the following features.

Images