JP7492120B2 - Consumption state learning device, trained model generation method and program - Google Patents

Description

The present invention relates to a consumption state learning device, a method for generating a trained model, and a program.

Patent document 1 describes a tire assembly that includes a tire pressure monitoring sensor (TPMS) and an RFID (Radio Frequency Identification) tag, in which a unique tire identification code is shared between the TPMS and the RFID tag.

Patent Document 2 describes a technology in which a card reader installed at a gas station reads a reference ID code, and identifies the vehicle based on the result of matching the ID code in the signal transmitted from the tire sensor unit of each vehicle with the reference ID code. The technology described in Patent Document 2 displays information on the tire condition corresponding to the ID code, which is indicated by data contained in the signal transmitted from the identified vehicle.

Patent document 3 describes a technology in which a terminal reads an RFID tag attached to the inside surface of a tire, which has a unique number written on each tire, and transmits tire mounting/removal record information relating to the mounting/removal of the tire to a central tire management device. In the technology described in patent document 3, tire usage history data that records events that have occurred in the past regarding the tire is accumulated in the central tire management device.

Patent document 4 describes a machine learning database creation system that outputs virtual sensor data and training data based on environmental information automatically recognized from three-dimensional shape information of terrain and buildings and sensor parameters of sensors.

Patent document 5 describes an anomalous data detection method that uses a machine learning algorithm to detect abnormal data from data related to tire usage history.

JP 2017-114483 A JP 2013-252777 A JP 2005-138738 A International Publication No. 2018/020954 JP 2019-74927 A

The technologies described in Patent Documents 1 to 3 make it possible to collect information about the condition of each tire after it has been in use, but the collected information alone does not provide sufficient knowledge for designing new tires.

Furthermore, even if the techniques described in Patent Documents 4 and 5 are used, it is not possible to obtain knowledge for designing new tires.

This is true not only when designing new tires, but also when designing new products in general.

The present invention has been made in consideration of the above-mentioned circumstances, and one of its objectives is to provide a consumption state learning device, a method for generating a trained model, and a program that can obtain knowledge useful for designing new products.

In order to solve the above problem, the consumption state learning device according to the present invention is a consumption state learning device that executes learning of a machine learning model used to estimate the consumption state of an individual product, and includes: an identification code reading means that reads an identification code unique to the individual product recorded on an information recording medium provided in the individual product; a training data generating means that generates training data indicating the consumption state of the individual product based on an image of the individual product; an acquisition means that accesses a storage means that stores design data of a plurality of individual products including the individual product, and acquires the design data of the individual product identified by the identification code; and a learning means that executes learning of the machine learning model using an output when learning input data including the acquired design data is input to the machine learning model, and the training data.

In one aspect of the present invention, the storage means further stores production data indicating at least one of the factory where the individual product was produced, the production equipment used to produce the individual product, or the production date and time of the individual product, the acquisition means further acquires the production data of the individual product identified by the identification code, and the learning means performs learning of the machine learning model using the output when the learning input data further including the production data is input to the machine learning model.

In one aspect of the present invention, the learning means performs learning of the machine learning model using an output when the learning input data, which further includes usage status data indicating the usage status of the individual product in which the information recording medium is provided, is input to the machine learning model.

In this aspect, the method may further include a usage data generation means for generating the usage data based on the image, and the learning means may execute learning of the machine learning model using an output when the learning input data, which further includes the generated usage data, is input to the machine learning model.

In one aspect of the present invention, the machine learning model is used to estimate the wear state of an individual tire, the identification code reading means reads an identification code unique to the individual tire recorded on an information recording medium provided on the individual tire, the teacher data generating means generates the teacher data indicating the wear state of the tire based on an image of the tire, and the acquisition means accesses the storage means that stores design data of multiple tires including the tire, and acquires the design data of the tire identified by the identification code.

In this aspect, the method further includes a position data generating means for generating position data indicating the position of the tire on a vehicle equipped with the tire, and the learning means may perform learning of the machine learning model using an output when the learning input data further including the position data is input to the machine learning model.

The system may further include a vehicle model data generation means for generating vehicle model data indicating the type of the vehicle, and the learning means may perform learning of the machine learning model using an output when the learning input data further including the vehicle model data is input to the machine learning model.

Alternatively, the method may further include a tire number data generation means for generating tire number data indicating the number of tires equipped on the vehicle, and the learning means may perform learning of the machine learning model using an output when the learning input data further including the tire number data is input to the machine learning model.

The method may further include a usage data generation means for generating usage data indicating the usage status of the tire based on the remaining tread depth of the tire on which the information recording medium is provided, the usage data being estimated based on the image, and the learning means may execute learning of the machine learning model using an output when the learning input data further including the generated usage data is input to the machine learning model.

The machine learning model may be used to estimate the wear condition of an individual tire, and the training data generation means may generate the training data indicating the wear condition of the tire.

Alternatively, the machine learning model may be used to estimate the presence or absence of uneven wear on an individual tire, and the training data generating means may generate the training data indicating the presence or absence of uneven wear on the tire.

The method for generating a trained model according to the present invention is a method for generating a trained model that executes training of a machine learning model used to estimate the wear state of an individual product, and includes the steps of reading an identification code unique to the individual product recorded on an information recording medium provided in the individual product, generating training data indicating the wear state of the individual product based on an image of the individual product, accessing a storage means that stores design data of multiple individual products including the individual product, and acquiring the design data of the individual product identified by the identification code, and executing training of the machine learning model using an output when training input data including the acquired design data is input to the machine learning model and the training data.

The program according to the present invention causes a computer that executes the learning of a machine learning model used to estimate the wear state of an individual product to execute the steps of reading an identification code unique to the individual product recorded on an information recording medium provided on the individual product, generating training data indicating the wear state of the individual product based on an image of the individual product, accessing a storage means that stores design data of multiple individual products including the individual product, and acquiring the design data of the individual product identified by the identification code, and executing learning of the machine learning model using the training data and an output when learning input data including the acquired design data is input to the machine learning model.

1 is a diagram illustrating an example of a configuration of an information processing device according to an embodiment of the present invention. FIG. 4 is a diagram illustrating an example of product management data. FIG. 2 is a diagram illustrating an example of the upper surface of a vehicle. FIG. 2 is a diagram illustrating an example of the upper surface of a vehicle. FIG. 13 is a diagram illustrating an example of learning data. FIG. 2 is a functional block diagram showing an example of functions of an information processing device according to an embodiment of the present invention. 1 is a flow diagram showing an example of a flow of processing performed by an information processing device according to an embodiment of the present invention. 1 is a flow diagram showing an example of a flow of processing performed by an information processing device according to an embodiment of the present invention. 1 is a flow diagram showing an example of a flow of processing performed by an information processing device according to an embodiment of the present invention.

One embodiment of the present invention will be described below with reference to the drawings.

Figure 1 is a diagram showing an example of the configuration of an information processing device 10 according to one embodiment of the present invention. The information processing device 10 according to this embodiment is a computer such as a personal computer. As shown in Figure 1, the information processing device 10 includes, for example, a processor 12, a memory unit 14, a communication unit 16, a display unit 18, an operation unit 20, a photographing unit 22, and an RFID reader unit 24.

The processor 12 is, for example, a program-controlled device such as a CPU that operates according to a program installed in the information processing device 10.

The memory unit 14 is a memory element such as a ROM or RAM, or a hard disk drive. The memory unit 14 stores programs executed by the processor 12, etc.

The communication unit 16 is a communication interface, such as a network board.

The display unit 18 is a display device such as an LCD display, and displays various images according to instructions from the processor 12.

The operation unit 20 is a user interface such as a keyboard and mouse that accepts user input and outputs a signal indicating the content of the input to the processor 12.

The imaging unit 22 is an imaging device such as a digital camera, and outputs an image generated by imaging to the processor 12. The imaging unit 22 may also be an imaging device such as a stereo camera capable of capturing three-dimensional images or a depth camera capable of capturing depth images, and output the three-dimensional image or depth image generated by imaging to the processor 12. The information processing device 10 according to this embodiment may include multiple imaging units 22. The imaging unit 22 may also be a video camera capable of capturing three-dimensional moving images including a series of three-dimensional images, or depth moving images including a series of depth images.

In this embodiment, the RFID reader unit 24 is a device that reads an identification code recorded on an information recording medium such as an RFID tag by short-range wireless communication using, for example, an electromagnetic field or radio waves. In this embodiment, for example, an information recording medium such as an RFID tag is provided on an individual product such as an individual tire. The RFID reader unit 24 according to this embodiment reads an identification code recorded on the information recording medium and unique to the individual product in which the information recording medium is provided. The information processing device 10 according to this embodiment may be equipped with multiple RFID reader units 24. The antenna provided in the RFID reader unit 24 according to this embodiment has directionality, and the RFID reader unit 24 is capable of detecting RFID tags present within a predetermined range based on the position and orientation of the RFID reader unit 24.

The information processing device 10 may also include an optical disc drive that reads optical discs such as DVD-ROMs and Blu-ray (registered trademark) discs, a USB (Universal Serial Bus) port, etc.

In the information processing device 10 according to this embodiment, learning of a machine learning model used to estimate the wear state of an individual product is performed. As an example, learning of a machine learning model used to estimate the wear state of an individual tire is performed.

When training the machine learning model, the information processing device 10 according to this embodiment accesses, via the communication unit 16, a product management database 30, which is, for example, a database provided within a company and which manages information on multiple individual tires.

Figure 2 is a diagram showing an example of product management data stored in the product management database 30. The product management data according to this embodiment is, for example, data associated with an individual tire.

As shown in FIG. 2, the product management data in this embodiment includes, for example, identification code data, product data, design data, production data, etc.

In this embodiment, the identification code data is, for example, data that indicates an identification code that is assigned in advance to an individual tire and is unique to that tire. In this embodiment, for example, an information recording medium such as an RFID tag on which the identification code assigned to the tire is recorded is attached to the tire in advance.

In this embodiment, the product data is data that indicates information about the product, such as the brand name of the tire.

In this embodiment, the design data is, for example, data that indicates information related to the design of the tire. The design data shown in FIG. 2 includes belt data that indicates the type of belt used in the tire. The design data also includes rubber property data that indicates the rubber properties of the tire.

In this embodiment, the production data is, for example, data indicating information related to the production of the tire. The production data shown in FIG. 2 includes a factory number indicating the identification number of the factory where the tire was produced. The production data also includes a molding machine number corresponding to the type of molding machine used to produce the tire. The production data also includes production date and time data indicating the date and time when the tire was produced.

In addition, when learning the machine learning model according to this embodiment, the RFID reader unit 24 reads an identification code unique to the tire, which is recorded on an information recording medium such as an RFID tag attached to the tire. The image capturing unit 22 captures an image of the tire (e.g., an image of the tread surface or side). Then, by using a publicly known image recognition technology, the remaining tread depth and the presence or absence of uneven wear of the tire are estimated based on the image. Here, the remaining tread depth and the presence or absence of uneven wear of the tire may be estimated by using a publicly known trained machine learning model.

Figure 3 is a schematic diagram showing an example of the top surface of a vehicle 40 equipped with the above-mentioned tires. The vehicle 40 shown in Figure 3 is equipped with four tires (a right front tire 42, a right rear tire 44, a left front tire 46, and a left rear tire 48). Each tire equipped on the vehicle 40 is provided with an RFID tag on which an identification code unique to that tire is recorded. The vehicle 40 shown in Figure 3 is traveling at a low speed to the left in Figure 3.

In this embodiment, for example, as shown in FIG. 3, two image capturing units 22 (22a and 22b) are provided beside the lane on which the vehicle 40 is traveling, with the image capturing direction being oblique to the lane.

In addition, in this embodiment, for example, as shown in FIG. 3, two RFID reader units 24 (24a and 24b) are provided beside the lane on which the vehicle 40 is traveling, so that the detection direction of the RFID tag is diagonal to the lane.

The RFID reader unit 24a is arranged alongside the image capturing unit 22a, to the left of the image capturing unit 22a in FIG. 3 (forward in the direction of travel of the vehicle 40). The RFID reader unit 24b is arranged alongside the image capturing unit 22b, to the left of the image capturing unit 22b in FIG. 3 (forward in the direction of travel of the vehicle 40).

The photographing unit 22a photographs an image showing the tread surface and side of the right front tire 42 of the vehicle 40. Based on the image, the remaining groove depth and the presence or absence of uneven wear of the right front tire 42 are estimated. The photographing unit 22a also photographs an image showing the tread surface and side of the right rear tire 44 of the vehicle 40. Based on the image, the remaining groove depth and the presence or absence of uneven wear of the right rear tire 44 are estimated.

The RFID reader unit 24a reads the identification code of the right front tire 42 recorded on an RFID tag provided on the right front tire 42 of the vehicle 40. The RFID reader unit 24a also reads the identification code of the right rear tire 44 recorded on an RFID tag provided on the right rear tire 44 of the vehicle 40.

The photographing unit 22b photographs an image showing the tread surface and side of the left front tire 46 of the vehicle 40. Based on the image, the remaining groove depth and the presence or absence of uneven wear of the left front tire 46 are estimated. The photographing unit 22b also photographs an image showing the tread surface and side of the left rear tire 48 of the vehicle 40. Based on the image, the remaining groove depth and the presence or absence of uneven wear of the left rear tire 48 are estimated.

The RFID reader unit 24b reads the identification code of the left front tire 46 recorded on an RFID tag provided on the left front tire 46 of the vehicle 40. The RFID reader unit 24b also reads the identification code of the left rear tire 48 recorded on an RFID tag provided on the left rear tire 48 of the vehicle 40.

Then, the information processing device 10 accesses the product management database 30 and acquires product management data including identification code data indicating each of the four identification codes read in the above manner. In this manner, the product management data for the right front tire 42, the right rear tire 44, the left front tire 46, and the left rear tire 48 is acquired.

In this embodiment, for example, the detection of an RFID tag by the RFID reader unit 24a may be used as a trigger for the image capture unit 22a to start capturing moving images. Then, when it is confirmed that the image captured by the image capture unit 22a does not show the vehicle 40 or tires, the image capture by the image capture unit 22a may be terminated.

Then, one or more frame images showing the right front tire 42 and one or more frame images showing the right rear tire 44 may be extracted from the captured video. As described above, in the video, the right front tire 42 is captured first, followed by the right rear tire 44, so it is possible to identify the image showing the right front tire 42 and the image showing the right rear tire 44 based on the timing at which the frame images were captured. Then, based on the extracted frame images, the remaining tread depth and the presence or absence of uneven wear may be estimated for each of the right front tire 42 and the right rear tire 44.

In addition, during the period in which the video is captured, the identification code read first by the RFID reader unit 24a may be identified as the identification code of the right front tire 42, and the identification code read later may be identified as the identification code of the right rear tire 44.

Similarly, the detection of an RFID tag by the RFID reader unit 24b may be used as a trigger for the image capture unit 22b to start capturing moving images. Then, when it is confirmed that the image captured by the image capture unit 22b does not show the vehicle 40 or tires, the image capture by the image capture unit 22b may be terminated.

Then, as described above, the remaining tread depth and the presence or absence of uneven wear may be estimated for each of the left front tire 46 and the left rear tire 48 based on the video images captured by the imaging unit 22b.

In addition, during the period in which the video is captured, the identification code read first by the RFID reader unit 24b may be identified as the identification code of the left front tire 46, and the identification code read later may be identified as the identification code of the left rear tire 48.

Figure 4 is a schematic diagram showing an example of the top surface of a vehicle 50 that is a different type of vehicle from the vehicle 40 shown in Figure 3. The vehicle 50 shown in Figure 4 has six tires (a right front tire 52, a right outer rear tire 54, a right inner rear tire 56, a left front tire 58, a left outer rear tire 60, and a left inner rear tire 62).

In the example of FIG. 4, similar to the example of FIG. 3, the remaining groove depth and the presence or absence of uneven wear of the right front tire 52 are estimated based on an image of the tread surface and side of the right front tire 52 of the vehicle 50 captured by the image capture unit 22a.

In addition, based on images captured by the image capture unit 22a showing the tread surface and side of the right rear outer tire 54 of the vehicle 50 and the tread surface of the right rear inner tire 56, the remaining groove amount and the presence or absence of uneven wear for each of the right rear outer tire 54 and the right rear inner tire 56 are estimated. For example, the tire captured on the left of the captured image may be identified as the right rear outer tire 54, and the tire captured on the right may be identified as the right rear inner tire 56.

In addition, the remaining groove depth and the presence or absence of uneven wear in the left front tire 58 are estimated based on an image of the tread surface and side of the left front tire 58 of the vehicle 50 captured by the image capture unit 22b.

In addition, based on images captured by the image capture unit 22b showing the tread surface and side of the left rear outer tire 60 of the vehicle 50 and the tread surface of the left rear inner tire 62, the remaining groove amount and the presence or absence of uneven wear for each of the left rear outer tire 60 and the left rear inner tire 62 are estimated. For example, the tire captured on the left of the image may be identified as the left rear inner tire 62, and the tire captured on the right may be identified as the left rear outer tire 60.

In the example of FIG. 4, similar to the example of FIG. 3, the RFID reader unit 24a reads the identification code of the right front tire 52 recorded on the RFID tag provided on the right front tire 52 of the vehicle 50.

In addition, the RFID reader unit 24a reads the identification codes of the right rear outer tire 54 and the right rear inner tire 56 recorded on the RFID tags provided on the right rear outer tire 54 and the right rear inner tire 56 of the vehicle 50.

The RFID reader unit 24b reads the identification code of the left front tire 58 recorded on the RFID tag attached to the left front tire 58 of the vehicle 50.

In addition, the RFID reader unit 24b reads the identification codes of the left rear outer tire 60 and the left rear inner tire 62 recorded on the RFID tags provided on the left rear outer tire 60 and the left rear inner tire 62 of the vehicle 50.

Then, the information processing device 10 accesses the product management database 30 and obtains product management data including the identification code data corresponding to each of the six identification codes read in the above manner.

In the example of FIG. 4, the image capturing unit 22a may start capturing moving images when the RFID reader unit 24a detects an RFID tag. Then, when it is confirmed that the image captured by the image capturing unit 22a does not show the vehicle 50 or tires, the image capturing by the image capturing unit 22a may end. Also, the image capturing unit 22b may start capturing moving images when the RFID reader unit 24b detects an RFID tag. Then, when it is confirmed that the image captured by the image capturing unit 22b does not show the vehicle 50 or tires, the image capturing by the image capturing unit 22b may end.

Then, when the video capture period for one vehicle ends, the information processing device 10 according to this embodiment generates learning data as shown in FIG. 5, which is associated with each tire of the vehicle. For example, for a vehicle with four tires as shown in FIG. 3, four learning data are generated, and for a vehicle with six tires as shown in FIG. 4, six learning data are generated.

As shown in FIG. 5, the learning data according to this embodiment includes, for example, an index number, learning input data, and teacher data. The learning input data according to this embodiment includes product data, design data, production data, location data, vehicle model data, and usage status data. The teacher data according to this embodiment includes consumption status data.

As described above, in this embodiment, learning data is generated that includes new index numbers corresponding to each tire equipped on the vehicle.

The values of the product data, design data, and production data of the learning data are set to the values of the product data, design data, and production data contained in the product management data identified by the identification code read from the RFID tag attached to the tire.

The value of the position data in the learning data is set to a value that indicates the position of the tire on the vehicle that has that tire.

For example, if the tire is a right front tire, the position data value is set to "right front." If the tire is a right rear tire, the position data value is set to "right rear." If the tire is a left front tire, the position data value is set to "left front." If the tire is a left rear tire, the position data value is set to "left rear."

If the tire is the outer right rear tire, the position data value is set to "outer right rear." If the tire is the inner right rear tire, the position data value is set to "inner right rear." If the tire is the outer left rear tire, the position data value is set to "outer left rear." If the tire is the inner left rear tire, the position data value is set to "inner left rear."

The value of the vehicle model data of the learning data is set to a value indicating the type of vehicle that has the tire. In the example of FIG. 5, if the vehicle that has the tire has four tires, the value of the vehicle model data is set to "1." Also, if the vehicle that has the tire has six tires, the value of the vehicle model data is set to "2."

The value of the usage data of the learning data is set to a value corresponding to the estimated remaining tread depth for the tire. For example, if the value indicating the remaining tread depth is smaller than a predetermined value d1, the value of the usage data is set to "1", if it is equal to or greater than d1 and less than d2, the value is set to "2", and if it is equal to or greater than d2, the value is set to "3". In this embodiment, for example, the longer the period of use (driving period) of the tire, the smaller the value of the usage data included in the learning input data corresponding to the tire.

The value of the wear state data of the learning data is set to a value according to the presence or absence of uneven wear estimated for the tire. For example, the wear state data value is set to "1" for tires estimated to have uneven wear, and "0" for tires estimated to not have uneven wear.

Then, using the learning data generated for multiple tires in the above manner, learning of a machine learning model used to estimate the wear state of an individual tire is performed. For example, a difference may be identified between an output when the learning input data included in the learning data is input to the machine learning model and the teacher data included in the learning data. Then, supervised learning may be performed in which the parameter values of the machine learning model are updated based on the identified difference.

In this embodiment, for example, the trained machine learning model generated in this way is used to design new tires.

For example, a tire designer inputs estimated input data including values of product data, design data, production data, location data, vehicle model data, and usage data into a trained machine learning model. Then, based on the output corresponding to the input, it becomes possible to estimate the probability of uneven wear occurring when a tire conforming to the product, design, production, location, and vehicle model conditions corresponding to the values indicated by the estimated input data is used until the remaining tread depth corresponds to the value of the usage data included in the estimated input data.

In this embodiment, for example, learning may be performed using multiple machine learning models depending on the application. For example, learning may be performed using a machine learning model that inputs a portion of the learning input data shown in FIG. 5, in addition to a machine learning model that inputs all of the learning input data shown in FIG. 5. For example, learning may be performed using another machine learning model in which the learning input data includes design data, location data, vehicle model data, and usage data.

In this case, the tire designer inputs estimated input data including values of design data, position data, vehicle model data, and usage data into the trained machine learning model. Then, based on the output corresponding to the input, it becomes possible to estimate the probability of uneven wear occurring when a tire conforming to the conditions related to design, position, and vehicle model that correspond to the values indicated by the estimated input data is used until the remaining tread depth corresponds to the value of the usage data included in the estimated input data. In this case, it is possible to estimate the probability of uneven wear occurring that is not dependent on the values of product data or production data.

For example, learning of multiple machine learning models that differ for each type of vehicle equipped with tires may be performed. For example, learning of a machine learning model that uses learning input data for a four-wheeled vehicle as input, and learning of a machine learning model that uses learning input data for a six-wheeled vehicle as input may be performed. Then, a trained machine learning model that has been trained using the learning input data for a four-wheeled vehicle may be used for designing a four-wheeled vehicle, and a trained machine learning model that has been trained using the learning input data for a six-wheeled vehicle may be used for designing a six-wheeled vehicle. In this case, the learning input data does not need to include vehicle type data.

In addition, in this embodiment, various values of estimated input data may be input to a trained machine learning model using a genetic algorithm or the like to identify inputs that are less likely to cause uneven wear. Then, a design plan corresponding to the input identified in this way may be adopted as a design plan for a new tire.

In this embodiment, there is a possibility that the RFID reader unit 24a cannot determine whether the identification code it reads is for the right rear outer tire 54 or the right rear inner tire 56. In light of this, for example, if the value of the product management data corresponding to the identification code of the right rear outer tire 54 is different from the value of the product management data corresponding to the identification code of the right rear inner tire 56, generation of learning data may not be performed. Similarly, if the value of the product management data corresponding to the identification code of the left rear outer tire 60 is different from the value of the product management data corresponding to the identification code of the left rear inner tire 62, generation of learning data may not be performed.

In addition, in this embodiment, the product management data does not have to include product data. Then, by performing character recognition processing on an image showing the side of the tire, information about the tire, such as the brand name of the tire, may be identified. Then, learning input data may be generated that includes product data indicating the identified information.

The functions implemented in the information processing device 10 and the processing performed by the information processing device 10 are further described below.

Figure 6 is a functional block diagram showing an example of functions implemented in the information processing device 10 according to this embodiment. Note that it is not necessary for all of the functions shown in Figure 6 to be implemented in the information processing device 10 according to this embodiment, and functions other than the functions shown in Figure 6 may also be implemented.

As shown in FIG. 6, the information processing device 10 functionally includes, for example, a machine learning model 70, an identification code reading unit 72, a product management data acquisition unit 74, an image acquisition unit 76, a learning input data generation unit 78, a teacher data generation unit 80, a learning data generation unit 82, a learning data storage unit 84, a learning unit 86, an estimated input acceptance unit 88, and an estimation unit 90.

The machine learning model 70 is implemented primarily using the processor 12 and the memory unit 14. The identification code reading unit 72 is implemented primarily using the processor 12 and the RFID reader unit 24. The product management data acquisition unit 74 is implemented primarily using the communication unit 16. The image acquisition unit 76 is implemented primarily using the processor 12 and the photographing unit 22. The learning input data generation unit 78, the teacher data generation unit 80, the learning data generation unit 82, the learning unit 86, and the estimation unit 90 are implemented primarily using the processor 12. The learning data memory unit 84 is implemented primarily using the memory unit 14. The estimated input reception unit 88 is implemented primarily using the processor 12 and the operation unit 20.

The above functions may be implemented by executing a program including instructions corresponding to the above functions, which is installed in the information processing device 10, which is a computer, by the processor 12. This program may be supplied to the information processing device 10 via a computer-readable information storage medium, such as an optical disk, a magnetic disk, a magnetic tape, a magneto-optical disk, or a flash memory, or via the Internet, for example.

The functions of the machine learning model 70, identification code reading unit 72, product management data acquisition unit 74, image acquisition unit 76, learning input data generation unit 78, teacher data generation unit 80, learning data generation unit 82, learning data storage unit 84, and learning unit 86 in the information processing device 10 correspond to the functions of a consumption state learning device that performs learning of a machine learning model used to estimate the consumption state of individual products.

In addition, the functions of the machine learning model 70, the estimated input receiving unit 88, and the estimation unit 90 in the information processing device 10 correspond to the function of a consumption state estimation device that estimates the consumption state of an individual product according to conditions corresponding to the input to the trained machine learning model 70.

In this embodiment, the machine learning model 70 is a machine learning model that implements machine learning such as AdaBoost, random forest, neural network, support vector machine (SVM), nearest neighbor classifier, etc. As described above, the machine learning model 70 according to this embodiment is used to estimate the wear state of individual products. The machine learning model 70 is used, for example, to estimate the wear state of individual tires (for example, the wear state of the tires, such as the presence or absence of uneven wear).

Here, the machine learning model 70 may be a judgment model that judges whether or not an individual product is worn out, such as whether or not a tire is unevenly worn. In this case, the machine learning model 70 may output, in response to an input, one bit of data indicating the presence or absence of wear, similar to the wear state data shown in FIG. 5, for example.

The machine learning model 70 may also be a regression model that outputs data indicating the degree of wear and tear of individual products.

In this embodiment, for example, the identification code reading unit 72 reads an identification code unique to the individual product that is recorded on an information recording medium provided on the individual product. Here, as described above, the identification code reading unit 72 may read an identification code unique to the tire that is recorded on an information recording medium provided on the individual tire.

In this embodiment, the product management data acquisition unit 74, for example, accesses a memory unit that stores design data of multiple product individuals, including the product individual corresponding to the read identification code, and acquires the design data of the product individual identified by the identification code. In the above example, the memory unit corresponds to the product management database 30. As described above, the memory unit may store design data of multiple tires, including the tire corresponding to the read identification code.

As described above, the storage unit may further store production data indicating at least one of the factory where the individual product was produced, the production equipment (the molding machine in the above example) used to produce the individual product, or the production date and time of the individual product. The product management data acquisition unit 74 may further acquire production data of the individual product identified by the identification code read by the identification code reading unit 72.

In this embodiment, the image acquisition unit 76 acquires an image of an individual product in which the above-mentioned information recording medium is installed, for example.

In this embodiment, the learning input data generation unit 78 generates learning input data that includes, for example, design data acquired by the product management data acquisition unit 74. Here, the learning input data generation unit 78 may generate learning input data that includes production data acquired by the product management data acquisition unit 74.

The learning input data generating unit 78 may also generate learning input data including usage status data indicating the usage status of an individual product in which the above-mentioned information recording medium is provided. Here, the learning input data generating unit 78 may generate usage status data based on an image acquired by the image acquiring unit 76. For example, usage status data indicating the usage status of the tire may be generated based on the remaining tread depth of the tire estimated based on the image acquired by the image acquiring unit 76. Then, the learning input data generating unit 78 may generate learning input data including the usage status data.

The learning input data generating unit 78 may also generate position data indicating the position of the tire on a vehicle equipped with the tire. Then, the learning input data generating unit 78 may generate learning input data that includes the generated position data.

The learning input data generating unit 78 may also generate vehicle model data indicating the type of vehicle equipped with the tire. Then, the learning input data generating unit 78 may generate learning input data that includes the generated vehicle model data.

Here, the learning input data generating unit 78 may generate tire number data indicating the number of tires equipped on the vehicle instead of vehicle model data. For example, tire number data indicating "4" in the example of FIG. 3 and "6" in the example of FIG. 4 may be generated. Then, the learning input data generating unit 78 may generate learning input data including the generated tire number data.

In this embodiment, for example, the teacher data generating unit 80 generates teacher data indicating the wear state of the individual product based on an image of the individual product. Here, the teacher data generating unit 80 may generate teacher data indicating the wear state of the tire based on an image of a tire. Here, as described above, the teacher data generating unit 80 may generate teacher data indicating the wear state of the tire (for example, the presence or absence of uneven wear on the tire).

In this embodiment, the learning data generation unit 82 generates learning data including, for example, the learning input data generated by the learning input data generation unit 78 and the teacher data generated by the teacher data generation unit 80. Then, in this embodiment, the learning data generation unit 82 stores the generated learning data in the learning data storage unit 84.

In this embodiment, for example, the learning data storage unit 84 stores the learning data generated by the learning data generation unit 82.

In this embodiment, the learning unit 86 performs learning of the machine learning model 70 using, for example, the output when the learning input data included in the learning data is input to the machine learning model 70 and the teacher data included in the learning data. Here, for example, a difference may be identified between the output when the learning input data included in the learning data is input to the machine learning model 70 and the teacher data included in the learning data. Then, supervised learning may be performed in which the parameter values of the machine learning model 70 are updated based on the identified difference.

In this embodiment, the machine learning model 70 is trained using multiple pieces of training input data. The trained machine learning model 70 is then used to estimate the wear state of an individual product that is designed according to the conditions that correspond to the inputs to the machine learning model 70.

In this embodiment, for example, the estimated input receiving unit 88 receives estimated input data that is to be input to the trained machine learning model 70. Here, for example, the estimated input receiving unit 88 may receive estimated input data in which a value corresponding to a user operation is set.

In this embodiment, for example, the estimation unit 90 estimates the wear state of the individual product under conditions corresponding to the value of the estimated input data based on the output when the estimated input data is input to the trained machine learning model 70. Here, when the machine learning model 70 is a judgment model, the machine learning model 70 may output one bit of data indicating whether or not the individual product will wear out according to the condition corresponding to the value of the estimated input data (for example, whether or not uneven wear will occur in the tire according to the condition). Also, when the machine learning model 70 is a regression model, the machine learning model 70 may output an estimated value indicating the degree of wear of the individual product according to the condition corresponding to the value of the estimated input data. The estimation result by the estimation unit 90 may be displayed on the display unit 18, for example.

As described above, the estimation unit 90 may input various values of estimated input data into a trained machine learning model using a genetic algorithm or the like to identify conditions that are less likely to be worn out. Then, a screen showing the conditions that are less likely to be worn out may be displayed on the display unit 18. Then, a design proposal corresponding to the conditions identified in this way may be adopted as a design proposal for a new product.

Here, an example of the flow of the learning data generation process performed by the information processing device 10 according to this embodiment will be described with reference to the flow diagrams illustrated in FIGS. 7A and 7B.

First, the identification code reading unit 72 waits until the identification code is read from both the RFID reader unit 24a and the RFID reader unit 24b (S101). In this processing example, the identification code read in the processing shown in S101 is the identification code of the front tire.

When it is detected that the identification code has been read, the image acquisition unit 76 transmits an instruction to start capturing images to the image capture unit 22a and the image capture unit 22b (S102). In response to receiving the instruction to start capturing images, the image capture unit 22a and the image capture unit 22b start capturing images.

Then, the image acquisition unit 76 acquires the image captured by the image capture unit 22a and the image captured by the image capture unit 22b (S103).

Then, the learning input data generation unit 78 checks whether any of the images acquired in the process shown in S103 show tires (S104).

If it is confirmed that the subject is in the image (S104: Y), the process shown in S103 is executed.

Let us assume that it is confirmed that tires are not visible in either image acquired in the process shown in S103 (S104: N). In this case, the learning input data generating unit 78 identifies the usage status and wear state of the front tires based on the series of images acquired in the process shown in S103 (S105). Here, for example, for each of the two front tires, a value of usage status data based on the remaining tread depth and a value of wear state data indicating the presence or absence of uneven wear are identified.

Then, the product management data acquisition unit 74 acquires product management data including the identification code data indicating the identification code read in the process shown in S101 from the product management database 30 (S106). Here, for example, product management data corresponding to each of the two front tires is acquired.

Then, the identification code reading unit 72 waits until the identification code is read from both the RFID reader unit 24a and the RFID reader unit 24b (S107). In this processing example, the identification code read in the processing shown in S107 is the identification code of the rear tire.

Then, the image acquisition unit 76 acquires the image captured by the image capture unit 22a and the image captured by the image capture unit 22b (S108).

Then, the learning input data generation unit 78 checks whether any of the images acquired in the process shown in S108 show tires (S109).

If it is confirmed that the subject is in the image (S109: Y), the process shown in S108 is executed.

Let us assume that it is confirmed that tires are not visible in either image acquired in the process shown in S108 (S109: N). In this case, the learning input data generating unit 78 identifies the usage status and wear state of the rear tires based on the series of images acquired in the process shown in S108 (S110). Here, for example, for each of the two or four rear tires, the value of usage status data based on the remaining tread depth and the value of wear state data indicating the presence or absence of uneven wear are identified.

Then, the product management data acquisition unit 74 acquires product management data including the identification code data indicating the identification code read in the process shown in S107 from the product management database 30 (S111). Here, for example, product management data corresponding to each of the two or four rear tires is acquired.

Then, the learning input data generating unit 78 generates learning input data associated with each tire of the photographed vehicle (S112). Here, for the front tires, learning input data is generated that includes usage data indicating the usage status identified in the process shown in S105 and some or all of the product management data acquired in the process shown in S106. For the rear tires, learning input data is generated that includes usage data indicating the usage status identified in the process shown in S110 and some or all of the product management data acquired in the process shown in S111. In addition, for each learning input data, the value of the position data and the value of the vehicle model data corresponding to that learning input data are set.

Then, the teacher data generating unit 80 generates teacher data associated with each tire of the photographed vehicle (S113). Here, for the front tires, teacher data is generated that includes wear state data indicating the wear state identified in the process shown in S105. For the rear tires, learning input data is generated that includes wear state data indicating the wear state identified in the process shown in S110.

Then, the learning data generating unit 82 generates learning data associated with each tire of the photographed vehicle, and stores the generated learning data in the learning data storage unit 84 (S114). Here, for each tire, learning data is generated that includes the learning input data generated in the process shown in S112 and the teacher data generated in the process shown in S113.

Then, the image acquisition unit 76 transmits an instruction to end image capture to the image capture units 22a and 22b (S115). In response to receiving the instruction to end image capture, the image capture units 22a and 22b end image capture. Then, the process returns to S101.

Next, an example of the flow of the learning process performed by the information processing device 10 according to this embodiment will be described with reference to the flow diagram illustrated in FIG. 8. Here, for example, it is assumed that multiple pieces of learning data are stored in the learning data storage unit 84.

First, the learning unit 86 acquires one of the multiple pieces of learning data stored in the learning data storage unit 84 for which the processes shown in S202 and S203 have not been performed (S201).

Then, the learning unit 86 performs learning of the machine learning model 70 using the output obtained when the learning input data included in the learning data obtained in the process shown in S201 is input to the trained machine learning model 70 (S202). Here, for example, the parameter values of the machine learning model 70 may be updated based on the difference between the output and the teacher data included in the learning data obtained in the process shown in S201.

Then, the learning unit 86 checks whether the process shown in S202 has been performed for all of the multiple learning data stored in the learning data storage unit 84 (S203).

If it is confirmed that the process shown in S202 has not been performed for all of the multiple learning data stored in the learning data storage unit 84 (S203: N), the process returns to S201.

If it is confirmed that the process shown in S202 has been performed for all of the multiple learning data stored in the learning data storage unit 84 (S203: Y), the process shown in this processing example is terminated.

The present invention is not limited to the above-described embodiments.

For example, the usage status does not have to be determined based on the remaining tread depth of the tire. Also, the usage status does not have to be determined based on an image. For example, usage status data may be generated with a value corresponding to the mileage of the tire since it was new. Here, data indicating the mileage may be obtained from the vehicle via communication. Also, the value of the usage status data may be manually input by an operator who reads the mileage indicated on the vehicle's odometer. As described above, the value of the usage status data may be smaller as the identified mileage is longer.

The application of this invention is not limited to tires.

Furthermore, the specific numerical values and character strings above, as well as the specific numerical values and character strings in the drawings, are examples and are not limited to these numerical values and character strings.

10 Information processing device, 12 Processor, 14 Memory unit, 16 Communication unit, 18 Display unit, 20 Operation unit, 22, 22a, 22b Photography unit, 24, 24a, 24b RFID reader unit, 30 Product management database, 40 Vehicle, 42 Right front tire, 44 Right rear tire, 46 Left front tire, 48 Left rear tire, 50 Vehicle, 52 Right front tire, 54 Right outer rear tire, 56 Right inner rear tire, 58 Left front tire, 60 Left outer rear tire, 62 Left inner rear tire, 70 Machine learning model, 72 Identification code reading unit, 74 Product management data acquisition unit, 76 Image acquisition unit, 78 Learning input data generation unit, 80 Teacher data generation unit, 82 Learning data generation unit, 84 Learning data storage unit, 86 Learning unit, 88 Estimated input reception unit, 90 Estimation department.

Claims (11)

A consumption state learning device that executes learning of a machine learning model used to estimate a consumption state of an individual product,
an identification code reading means for reading an identification code unique to the individual product, the identification code being recorded on an information recording medium provided on the individual product;
a training data generating means for generating training data indicating a consumption state of the individual product based on an image of the individual product;
an acquisition means for accessing a storage means for storing design data of a plurality of individual products including the individual product, and acquiring the design data of the individual product identified by the identification code;
a usage status data generating means for generating usage status data indicating a usage status of the individual product in which the information recording medium is provided, based on the image;
a learning means for executing learning of the machine learning model using an output when learning input data, including the acquired design data and the generated usage data , is input to the machine learning model and the teacher data;
A consumption state learning device comprising: the storage means further stores production data indicating at least one of a factory where the individual product was produced, a production facility used in the production of the individual product, or a production date and time of the individual product;
The acquiring means further acquires the production data of the individual product identified by the identification code,
The learning means executes learning of the machine learning model by using an output when the learning input data further including the production data is input to the machine learning model.
2. The consumption state learning device according to claim 1. The machine learning model is used to estimate a wear state of an individual tire,
The identification code reading means reads an identification code unique to each individual tire, which is recorded on an information recording medium provided on the individual tire,
The training data generating means generates the training data indicating a wear state of the tire based on an image of the tire,
the acquiring means accesses the storage means that stores design data of a plurality of tires including the tire, and acquires the design data of the tire identified by the identification code.
3. The consumption state learning device according to claim 1 or 2. a position data generating means for generating position data indicative of a position of the tire on a vehicle having the tire;
The learning means executes learning of the machine learning model by using an output when the learning input data further including the position data is input to the machine learning model.
4. The consumption state learning device according to claim 3 . A vehicle type data generating means for generating vehicle type data indicating a type of the vehicle,
The learning means executes learning of the machine learning model by using an output when the learning input data further including the vehicle model data is input to the machine learning model.
5. The consumption state learning device according to claim 4 . a tire number data generating unit for generating tire number data indicating the number of tires provided on the vehicle;
The learning means executes learning of the machine learning model by using an output when the learning input data further including the tire number data is input to the machine learning model.
5. The consumption state learning device according to claim 4 . a usage status data generating means for generating usage status data indicating a usage status of the tire on the basis of a remaining groove depth of the tire on which the information recording medium is provided, the remaining groove depth being estimated on the basis of the image,
The learning means executes learning of the machine learning model by using an output when the learning input data, which further includes the generated usage data, is input to the machine learning model.
7. The consumption state learning device according to claim 3, wherein the consumption state learning device is a power consumption state learning device. The machine learning model is used to estimate a wear state of an individual tire,
The teacher data generating means generates the teacher data indicating a wear state of the tire.
8. The consumption state learning device according to claim 3, wherein the consumption state learning device is a power consumption state learning device. The machine learning model is used to estimate the presence or absence of uneven wear of an individual tire,
The teacher data generating means generates the teacher data indicating the presence or absence of uneven wear of the tire.
8. The consumption state learning device according to claim 3, wherein the consumption state learning device is a power consumption state learning device. A method for generating a trained model that performs training of a machine learning model used to estimate a consumption state of an individual product, comprising:
a step in which an identification code reading means reads an identification code unique to the individual product recorded on an information recording medium provided on the individual product;
A training data generating means generates training data indicating a consumption state of the individual product based on an image of the individual product;
an acquiring means accessing a storage means storing design data of a plurality of individual products including the individual product, and acquiring the design data of the individual product identified by the identification code;
a usage status data generating means generating usage status data indicating a usage status of the individual product in which the information recording medium is provided, based on the image;
A learning means executes learning of the machine learning model by using an output when learning input data including the acquired design data and the generated usage data is input to the machine learning model and the teacher data;
A method for generating a trained model, comprising: A computer that performs learning of a machine learning model used to estimate the consumption state of an individual product,
A step of reading an identification code unique to the individual product, the identification code being recorded on an information recording medium provided on the individual product;
A step of generating training data indicating a consumption state of the individual product based on an image of the individual product;
a step of accessing a storage means for storing design data of a plurality of individual products including the individual product, and acquiring the design data of the individual product identified by the identification code;
generating usage status data indicating a usage status of the individual product in which the information recording medium is provided, based on the image;
A step of executing learning of the machine learning model using an output when learning input data including the acquired design data and the generated usage data is input to the machine learning model and the teacher data;
A program characterized by executing the above.

Images