JP2020131813A - Output device, generation method for learning model, computer program, and storage medium - Google Patents

Abstract

An output device, a learning model generation method, a computer program, and a storage medium are provided. An output device includes a processing unit that outputs output information regarding the mounting state of parts of a human-powered vehicle in accordance with input information regarding the mounting state of parts of the manpowered vehicle, using a learning model learned by a learning algorithm. [Selection drawing] Fig. 5

Description

The present invention relates to an information output device, a learning model generation method, a computer program, and a storage medium for proposing an appropriate mounting state for a human-powered vehicle component.

There are human-powered vehicles that use human power at least partially, such as bicycles, electric-assisted bicycles, and electric bicycles called E-bikes. A man-powered vehicle has multiple components such as a front derailleur, a rear derailleur, a seatpost, or a suspension.

A technique is known in which a diagnostic device is connected to a plurality of components to perform a diagnosis including the presence or absence of a failure (see Patent Documents 1 and 2). A conventional diagnostic device communicates with a component and diagnoses a failure based on whether or not the operation is completed when a predetermined operation is instructed to the component.

U.S. Pat. No. 7819032 U.S. Pat. No. 9227697

Even in a situation where each component has not failed, it may not be appropriate as a combination. Depending on the relationship with other components and the individuality of the rider, a more appropriate assembly condition may need to be proposed. Appropriate assembly conditions based on a number of conditions should be proposed by comprehensively judging the conditions of each of the multiple components.

An object of the present invention is to provide an output device, a learning model generation method, a computer program, and a storage medium for outputting information for proposing an appropriate mounting state of parts including components of a human-powered vehicle.

(1) The output device according to the first aspect of the present invention uses a learning model learned by a learning algorithm to output output information regarding the mounting state according to input information regarding the mounting state of parts of a human-powered vehicle. It has a part.
Therefore, output information corresponding to various information regarding the mounting state is output from the learning model, and it is possible to propose an appropriate mounting state of the component based on the output information.

(2) In the output device according to the second aspect of the present invention, the input information includes the identification information of the component.
Therefore, output information corresponding to the combination of parts to be mounted on the human-powered vehicle is output from the learning model, and it is possible to propose an appropriate mounting state of the parts based on the output information.

(3) In the output device according to the third aspect of the present invention, the input information includes information about the vehicle body of the human-powered vehicle.
Therefore, output information corresponding to information such as the size or type of the vehicle body of the human-powered vehicle is output from the learning model, and it is possible to propose an appropriate mounting state of parts based on the output information.

(4) In the output device according to the fourth aspect of the present invention, the input information includes the physical information of the user of the human-powered vehicle.
Therefore, regarding the parts of the human-powered vehicle, it is possible to propose an appropriate mounting state of the parts according to the weight, height, leg length, etc. of the user riding the human-powered vehicle.

(5) In the output device according to the fifth aspect of the present invention, the input information includes at least one of the still image data of the human-powered vehicle and the moving image data of the human-powered vehicle.
Therefore, regarding the parts of the human-powered vehicle, it is possible to input the still image data or the moving image data obtained by photographing the human-powered vehicle and propose an appropriate mounting state of the parts according to the observation result of the entire human-powered vehicle.

(6) In the output device according to the sixth aspect of the present invention, the input information includes data regarding vibration emitted from the human-powered vehicle.
Therefore, regarding the parts of the human-powered vehicle, it is possible to propose an appropriate mounting state of the parts based on the vibration during driving of the human-powered vehicle.

(7) In the output device according to the seventh aspect of the present invention, the data related to the vibration includes audio data.
For this reason, regarding the parts of a human-powered vehicle, it is possible to input the voice data collected during the driving of the human-powered vehicle and propose an appropriate mounting state of the parts according to various generated sounds including abnormal noise. ..

(8) In the output device according to the eighth aspect of the present invention, the output information includes the suitability information regarding the suitability of the mounting state.
Therefore, regarding the parts of the human-powered vehicle, the suitability of the mounting state of the parts is output from the learning model according to various input information, and it is possible to present the suitability of the mounting state of the parts.

(9) In the output device according to the ninth aspect of the present invention, the output information includes the proposal information regarding the proposal for the mounting state.
For this reason, regarding the parts of a human-powered vehicle, the learning model outputs a proposal of an appropriate other mounting state for the mounting state of the part corresponding to various input information, and therefore, the appropriate mounting state of the part is output from the output information. Can be proposed.

(10) In the output device according to the tenth aspect of the present invention, the proposed information includes at least one of information on how to attach the component, information on adjusting the component, and information on replacement of the component. ..
Therefore, regarding the parts of the human-powered vehicle, it is possible to propose an appropriate mounting method, adjustment amount, adjustment method, and replacement of the parts according to the mounting state of the parts corresponding to various input information.

(11) The method of generating the learning model according to the eleventh aspect of the present invention uses a neural network configured to output output information regarding the mounting state according to input information regarding the mounting state of parts of a human-powered vehicle. This is a learning model generation method for generating a learning model, in which parameters in the intermediate layer of the neural network are adjusted by a computer according to the teacher data labeled with the evaluation for the input information.
Therefore, a learning model is generated based on various input information regarding the mounting state of the parts of the human-powered vehicle and known teacher data corresponding to the appropriate mounting state corresponding to the input information. By using the output information output from the learning model, it is possible to propose an appropriate mounting state of the parts.

(12) In the method of generating a learning model according to the twelfth aspect of the present invention, the learning model is generated based on additional teacher data labeled with an evaluation for the output information output by giving the input information to the neural network. Updated by computer.
Therefore, the neural network is sequentially updated based on the evaluation of the output information output from the neural network being learned for the parts of the human-powered vehicle, and the proposal of the mounting state of the parts becomes more appropriate.

(13) The learning model generation method according to the thirteenth aspect of the present invention uses a neural network configured to output output information regarding the mounting state according to input information regarding the mounting state of parts of a human-powered vehicle. A method of generating a learning model, in which the training model is generated in the intermediate layer of the neural network according to the teacher data labeled with the evaluation of the output information output by giving the input information to the neural network. Adjust the parameters by computer.
Therefore, the neural network is sequentially updated based on the evaluation of the output information output from the neural network being trained without using the teacher data in advance, and the proposal of the mounting state of the component becomes more appropriate.

(14) A computer program according to the fourteenth aspect of the present invention applies to a learning model trained by a learning algorithm so as to output output information regarding the mounting state in response to input information regarding the mounting state of parts of a human-powered vehicle. Given the acquired input information, based on the output information output from the learning model, information on the mounting state of the parts in the human-powered vehicle is superimposed and arranged on the display unit for displaying the image of the human-powered vehicle. Have the computer perform the display process.
Therefore, the output information for proposing an appropriate mounting state based on various input information regarding the mounting state of the parts of the human-powered vehicle is output from the learning model according to various conditions such as vehicle body information and rider information. Will be done. Based on the information output from the learning model, an image showing the appropriate mounting state of the parts corresponding to the image of the human-powered vehicle is displayed, and it is possible to make a visually easy-to-understand proposal.

(15) The computer program of (14) is stored in the storage medium according to the fifteenth aspect of the present invention.
Therefore, based on the computer program read from the storage medium by the computer, the computer displays an image showing an appropriate mounting state of the parts corresponding to the image of the human-powered vehicle based on the information output from the learning model. , A visually easy-to-understand proposal is possible.

According to the output device for outputting the information on the mounting state of the parts of the human-powered vehicle according to the present invention, it is possible to output the information for proposing an appropriate mounting state of the parts including the components of the human-powered vehicle.

It is a block diagram which shows the structure of the output device and the terminal device of 1st Embodiment. It is a flowchart which shows an example of the generation method of a learning model. It is a flowchart which shows another example of the generation method of a learning model. It is a figure which shows the outline of the learning model. It is a flowchart which shows an example of the processing procedure about the proposal of the mounting state. It is a figure which shows the screen example displayed on the terminal apparatus. It is a figure which shows the screen example displayed on the terminal apparatus. It is a block diagram which shows the structure of the output device and the terminal device of 2nd Embodiment. It is a flowchart which shows an example of the generation method of the learning model in 2nd Embodiment. It is a figure which shows the outline of the learning model of 2nd Embodiment. It is a flowchart which shows an example of the processing procedure about the proposal of the mounting state. It is a flowchart which shows an example of the processing procedure about the proposal of the mounting state. It is a figure which shows the screen example displayed by the terminal apparatus of 2nd Embodiment. It is a figure which shows the screen example displayed by the terminal apparatus of 2nd Embodiment. It is a block diagram which shows the structure of the terminal apparatus in 3rd Embodiment. It is a side view of the human-powered vehicle to which the output device of 4th Embodiment is applied. It is a block diagram which shows the structure of the output device in 4th Embodiment. It is a flowchart which shows an example of the generation method of the learning model in 4th Embodiment. It is a figure which shows the outline of the learning model in 4th Embodiment. It is a flowchart which shows an example of the processing procedure concerning the determination of the suitability of the attachment state in 4th Embodiment.

The following description of each embodiment is an example of possible embodiments of the output device according to the present invention and is not intended to limit the embodiments. The output device according to the present invention may take a form different from each embodiment, such as a modification of each embodiment and a combination of at least two modifications that do not contradict each other.

In the following description of each embodiment, the terms front, rear, front, rear, left, right, sideways, up, and down are the directions in which the user is seated in the saddle of a human-powered vehicle. Used as a reference.

(First Embodiment)
In the first embodiment, the output device 1 managed by the parts manufacturer of the human-powered vehicle and the terminal used for the maintenance of the human-powered vehicle based on the information regarding the mounting state of the parts of the human-powered vehicle provided by the output device 1. A maintenance method using the device 2 will be described.

FIG. 1 is a block diagram showing the configurations of the output device 1 and the terminal device 2 of the first embodiment. In the first embodiment, the output device 1 uses a server computer. The output device 1 may be a personal computer. The output device 1 includes a processing unit 10, a storage unit 12, and a communication unit 14. In the first embodiment, the output device 1 will be described as one server computer, but the functions or processes may be distributed among a plurality of server computers, and the output device 1 is virtually generated by one large computer. It may be one of a plurality of instances.

The processing unit 10 is a processor using a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The processing unit 10 uses a built-in memory such as a ROM (Read Only Memory) and a RAM (Random Access Memory) to control each component to execute processing. The processing unit 10 generates a learning model 1M based on the learning program 1P including the learning model 1M stored in the storage unit 12, and uses the generated learning model 1M to output information regarding the assembly of the human-powered vehicle to the terminal device. Output toward 2.

The storage unit 12 uses a hard disk or an SSD (Solid State Drive) to store programs and data referenced by the processing unit 10 including the learning program 1P. The storage unit 12 stores the learning model 1M generated based on the learning program 1P. The storage unit 12 stores the data of the manual explaining how to attach the parts handled by the parts manufacturer. The learning program 1P and the learning model 1M may be those in which the learning program 30P and the learning model 30M stored in the storage medium 3 are read out and duplicated in the storage unit 12.

The communication unit 14 is a communication module that realizes a communication connection with the terminal device 2 via the network N including the public communication network N1. The communication unit 14 is, for example, a wireless communication device that supports Wi-Fi. In the second example, the communication unit 14 is a carrier communication module that realizes a communication connection via the carrier network N2 included in the network N. In the third example, the communication unit 14 is a network card for wired connection.

The terminal device 2 is a communication terminal such as a tablet owned by a user of a human-powered vehicle for maintenance or a seller of the human-powered vehicle. The terminal device 2 may be a so-called smartphone. The terminal device 2 may be a laptop personal computer. The terminal device 2 includes a processing unit 20, a storage unit 22, a communication unit 24, a display unit 26, and an operation unit 28.

The processing unit 20 is a processor using a CPU or GPU. The processing unit 20 uses a built-in ROM and a memory such as a RAM to control each component to execute processing. The processing unit 20 may be configured as one piece of hardware, a SoC (System On a Chip), in which a processor, a memory, a storage unit 22, and a communication unit 24 are integrated. The processing unit 20 displays information on the assembly of the human-powered vehicle on the display unit 26 based on the maintenance application program 2P stored in the storage unit 22.

The storage unit 22 uses a flash memory to store programs and data referred to by the processing unit 20 including the maintenance application program 2P. The maintenance application program 2P stored in the storage unit 22 is provided by a parts manufacturer of a human-powered vehicle, distributed from an arbitrary distribution server, and installed in a terminal device 2 which is a general-purpose computer. The maintenance application program 2P may be one that reads out the maintenance application program 50P stored in the storage medium 5 and duplicates it in the storage unit 22.

The communication unit 24 is a communication module that realizes a communication connection to the public communication network N1. The communication unit 24 is, for example, a wireless communication device that supports Wi-Fi. In the second example, the communication unit 24 is a carrier communication module that realizes communication via the carrier network N2. The communication unit 24 is a network card in the third example.

The display unit 26 uses a display device such as a liquid crystal panel or an organic EL display. The display unit 26 displays information on the mounting state of parts in the human-powered vehicle together with an image of the human-powered vehicle.

The operation unit 28 is an interface that accepts user operations, and uses physical buttons, a touch panel device with a built-in display, a speaker, a microphone, and the like. The operation unit 28 may accept an operation on the screen displayed on the display unit 26 by a physical button or a touch panel, or recognizes the operation content from the input voice by the microphone and interacts with the voice output by the speaker. The operation may be accepted in the form.

The output device 1 of the first embodiment having the above-described configuration uses a learning model 1M learned by a learning algorithm described later, and uses a learning model 1M, which has a component mounting state according to input information regarding the mounting state of the parts of the human-powered vehicle. Output information about. The output information is transmitted to the terminal device 2, the terminal device 2 receives this based on the maintenance application program 2P, and based on the received output information, the information regarding the mounting state of the parts in the human-powered vehicle is displayed on the display unit 26. indicate.

First, the learning model 1M generated by the output device 1 will be described. The learning model 1M is generated in advance by the output device 1 or an external device. FIG. 2 is a flowchart showing an example of a method of generating the learning model 1M. The output device 1 generates a learning model 1M that outputs output information for input information regarding the mounting state of parts of a human-powered vehicle by a supervised deep learning algorithm using a neural network (hereinafter referred to as NN: Neural Network). The learning algorithm may be an unsupervised learning algorithm or a recurrent neural network. The learning algorithm may be reinforcement learning.

Before starting the generation, the processing unit 10 collects information to be teacher data and stores it in the storage unit 12. The teacher data is a group of information in which information on the mounting state of parts of a human-powered vehicle and a judgment result of suitability for the mounting state are given. These pieces of information are information on a man-powered vehicle in a mounted state that has been collected by the manufacturer of the man-powered vehicle or component and has been determined to be appropriate or inappropriate. Information gathered at human-powered vehicle dealers, assembly plants or test sites. The main body of the suitability judgment is an assembly engineer, a manufacturer engineer, and the like. The subject of the determination may be the user. The input information is, for example, component identification information, such as a manufacturer name, a model number, a product number, and a product name. The input information includes vehicle body information. Body information includes the size, type and part number of each frame, front fork, handlebar, stem and seatpost. The input information includes physical information of a user riding a human-powered vehicle corresponding to component identification information and vehicle body information. The physical information includes at least one of the user's weight, height, leg length, arm length, and the like.

The processing unit 10 executes the following processing using the collected information group.
The processing unit 10 acquires identification information of the parts to be attached to the human-powered vehicle with respect to the attachment state of the parts of the human-powered vehicle stored in the storage unit 12 (step S101). The processing unit 10 acquires information about the vehicle body of the human-powered vehicle to which one or more parts are attached, which was acquired in step S101 (step S103). The processing unit acquires the physical information of the user of the human-powered vehicle to which the one or more parts acquired in step S101 are attached (step S105).

The processing unit 10 creates teacher data by adding a label of a suitability determination result given in advance to the input information acquired in steps S101 to S105 (step S107).

The processing unit 10 determines whether or not the teacher data is created from all the information groups that satisfy the predetermined conditions among the information groups stored in the storage unit 12 (step S109). The predetermined condition is, for example, the number of pieces of information that becomes teacher data with respect to acquiring input information for the same mounting state. When it is determined in step S109 that the teacher data has not been created from all the information groups (S109: NO), the processing unit 10 returns the processing to step S101.

When it is determined in step S109 that the teacher data has been created from all the information groups (S109: YES), the processing unit 10 gives the created teacher data to the NN (step S111). The processing unit 10 learns a parameter including at least one of the weight and the bias in the NN mesosphere from the given teacher data (step S113), and ends the processing.

As a result, the learning model 1M is generated by adjusting the parameters in the NN mesosphere by the computer according to the teacher data in which the evaluation for the input information is labeled in advance.

The generation of the learning model 1M is not limited to the above-labeled supervised learning method. FIG. 3 is a flowchart showing another example of the method of generating the learning model 1M. Among the processing procedures shown in the flowchart of FIG. 3, the procedures common to the flowchart of FIG. 2 are given the same step numbers and detailed description thereof will be omitted. In the flowchart of FIG. 3, the processing unit 10 executes the processing based on the information regarding one mounting state or one human-powered vehicle. The processing unit 10 executes the following processing for each of various human-powered vehicles or various mounting states.

The processing unit 10 gives the input information acquired in steps S103 to S105 to the learning NN for one mounting state (step S121). The processing unit 10 specifies the suitability information as a whole of the mounting state output from the NN when the input information is given to the NN, for example, the score (step S123). The processing unit 10 compares the specified suitability information with the determination result given in advance and derives an evaluation value (step S125). The evaluation value is, for example, an error between the score specified in step S123 and the score corresponding to the determination result. The parameters of the error are, for example, angle, distance, dimension, bearing, or weight.

The processing unit 10 gives the derived evaluation value to the intermediate layer, updates the parameters in the intermediate layer (step S127), and ends the processing. In step S127, the parameters are updated so that the derived error is minimized. The learning model 1M is generated by adjusting with the teacher data of the evaluation for the output information output when the input information acquired in step S101 to step S105 is input to the NN in step S127.

The processing unit 10 may generate the learning model 1M by either the processing procedure shown in the flowchart of FIG. 2 or the processing procedure shown in the flowchart of FIG. 3, or may execute any of the processing procedures. Good. When any of the processing procedures is executed, for example, the processing unit 10 initially generates the learning model 1M by the procedure shown in the flowchart of FIG. 2, and sets the parameters in the intermediate layer of the generated learning model 1M in FIG. It is advisable to adjust and re-update according to the processing procedure shown in the flowchart of.

FIG. 4 is a diagram showing an outline of the learning model 1M. The learning model 1M includes an input layer 31 for inputting input information regarding parts of the human-powered vehicle, and an output layer 32 for outputting output information regarding suitability for the mounting state of the parts of the human-powered vehicle. The learning model 1M includes an intermediate layer 33 having parameters trained by teacher data so as to output suitability when input information is input. When the input information regarding the component is input to the input layer 31, the intermediate layer 33 performs the calculation according to the learned parameters, and the output information regarding the suitability of the component input from the output layer 32 is output.

In the specific example of FIG. 4, the input layer 31 is given identification information for each of different types of components regarding the suitability of the component mounting state. In the specific example of FIG. 4, the identification information of the component is given the identification information of the electric component. A front derailleur, a rear derailleur, and a disc rotor are exemplified.

Vehicle body information is given to the input layer 31. The vehicle body information includes the size, type, and part number of each of the frame, front fork, handlebar, stem, and seatpost as described above.

The physical information of the user is given to the input layer 31. Physical information includes height and weight. Physical information may also be leg length, hand length, or gender.

A specific example of FIG. 4 is a configuration in which all of the component identification information, the vehicle body information, and the physical information are given to the input layer 31. However, the processing in the output device 1 is not limited to the configuration of giving everything. It may be only a part of the identification information, the vehicle body information, and the physical information of the part. The information to be input may be distinguished according to the object of suitability. For example, in the case of the learning model 1M that outputs the suitability information regarding the mounting of the rear derailleur, the frame type, the length, and the part number of the front derailleur may be input.

The learning model 1M generated in this way is used for proposing the mounting state of the following parts. FIG. 5 is a flowchart showing an example of a processing procedure related to the proposal of the mounting state. The proposal of the mounting state is started when the terminal device 2 used by the user executes the maintenance application program 2P and starts the maintenance.

The processing unit 20 of the terminal device 2 accepts the input on the input screen of the information about the user's human-powered vehicle based on the maintenance application program 2P (step S201). Based on the input to the input screen, the processing unit 20 receives identification information, vehicle body information, and physical information of the user of the parts of the human-powered vehicle that are attached to or will be assembled from the human-powered vehicle to be maintained ( Step S203).

The processing unit 20 transmits a determination request including the received component identification information, vehicle body information, and user's physical information from the communication unit 24 to the output device 1 (step S205).

The output device 1 receives a determination request by the communication unit 14 (step S131), and the processing unit 10 acquires identification information of a component attached to the human-powered vehicle from the received determination request (step S133). The processing unit 10 acquires information on the vehicle body of the human-powered vehicle from the determination request (step S135). Body information is one of the size, type, and part number of each of the frame, front forks, handles, stems, and seatposts of a man-powered vehicle. The processing unit 10 acquires the physical information of the user of the human-powered vehicle from the determination request (step S137). In step S137, the physical information includes at least one of the user's weight, height, leg length, arm length, and the like.

The processing of step S135 and step S137 may be performed by only one of them. That is, the information to be input to the learning model 1M may be both vehicle body information and physical information, or may be only one of them.

The processing unit 10 of the output device 1 gives the input information acquired in steps S133, S135, and S137 to the learning model 1M (step S139). When the input information is given to the learning model 1M, the processing unit 10 specifies the suitability information as a whole of the mounting state output from the learning model 1M, for example, the score (step S141).

The processing unit 10 reads out the manual data corresponding to the target part from the storage unit 12 based on the component identification information acquired in step S133 (step S143), and uses it based on the vehicle body information acquired in step S135. The instruction manual data related to the vehicle body is read from the storage unit 12 (step S145). The processing unit 10 may read out the instruction manual data of the parts recommended for each physique based on the physical information.

The processing unit 10 transmits the suitability information specified in step S141 and the instruction manual data read in steps S143 and S145 from the communication unit 14 to the terminal device 2 of the determination request source (step S147).

The terminal device 2 receives the suitability information and the instruction manual data from the communication unit 24 (step S207), and the processing unit 20 causes the display unit 26 to display the received suitability information (step S209). The processing unit 20 determines suitability based on the received information (step S211), and if it is determined to be appropriate (S211: YES), the processing unit 20 ends the processing as it is.

When it is determined in step S211 that it is unsuitable (S211: NO), the processing unit 20 displays a sample of how to attach the parts on the display unit 26 based on the instruction manual data received in step S207 (step S213), and processes. To finish.

The processing unit 20 may display a sample of how to attach the parts on the display unit 26 regardless of the determination in step S211.

6 and 7 are diagrams showing an example of a screen displayed on the terminal device 2. When the user selects and activates the maintenance application program 2P using the terminal device 2, the processing unit 20 displays an input screen 261 for receiving input information regarding the mounting state of the parts based on the maintenance application program 2P. It is displayed on 26. As shown in FIG. 6, the input screen 261 includes a component-specific selection screen 263 for component identification information, a vehicle body information selection screen 265, and a physical information selection screen 267. The input screen 261 includes a transmission button 269 for transmitting a suitability determination request. When the send button 269 is selected, the processing unit 20 transmits a determination request including the information received on the selection screens 263, 265 and 267 from the communication unit 24 (S205). It may be a free input field instead of the selection screen 267.

FIG. 7 shows an example of a proposal screen 271 that displays suitability information corresponding to a determination request. As shown in FIG. 7, the proposal screen 271 includes a link 273 to instruction manual data on how to install, along with a result of suitability. When link 273 is selected, instructions, i.e. suggestions for proper mounting, are displayed on the display 26. Further, as shown in FIG. 7, the proposal screen 271 shows the recommended parts together with the result of suitability, and may include information on replacement of parts.

As described above, the trained learning model 1M is expected to be used as a program module of the maintenance application program 20P which is a part of the artificial intelligence software. As described above, the learning model 1M of the present disclosure is used in the cooperative processing with the terminal device 2 in the output device 1 including the processing unit 10 and the storage unit 12.

In this way, various information about the mounting state, information for identifying parts, information about the type or size of the vehicle body, and output information according to the body of the user who is the rider are output based on the learning model 1M, and are output information. It is possible to propose an appropriate mounting state of the parts based on it.

(Second Embodiment)
FIG. 8 is a block diagram showing the configurations of the output device 1 and the terminal device 2 of the second embodiment. The output device 1 in the second embodiment stores the learning model 12M in the storage unit 12. The learning model 12M is trained for each component to be suitable or unsuitable. The learning model 12M inputs still image data or moving image data obtained by imaging a human-powered vehicle as input information regarding the mounting state. The terminal device 2 includes an imaging unit 29 in addition to the processing unit 20, the storage unit 22, the communication unit 24, the display unit 26, and the operation unit 28. The image pickup unit 29 is a device that includes an image pickup element, captures a still image or a moving image based on an instruction from the processing unit 20, and outputs data obtained by the imaging.

Same as the first embodiment except for the configuration of the output device 1 and the terminal device 2, the method of generating the learning model 12M, the content of the learning model 12M, and the processing procedure regarding the output method of the information on the suitability of the mounting state based on the learning model 12M. Is. Therefore, among the configurations of the second embodiment, the configurations common to the first embodiment are designated by the same reference numerals and detailed description thereof will be omitted.

FIG. 9 is a flowchart showing an example of a method of generating the learning model 12M in the second embodiment. Also in the second embodiment, the output device 1 generates a learning model 12M that outputs output information regarding the mounting state of the component by a supervised deep learning algorithm using NN.

Before starting the generation, the processing unit 10 collects information to be teacher data and stores it in the storage unit 12. The teacher data is a group of information in which the input information, which is the still image data or the moving image data obtained by capturing the mounting state of the parts of the human-powered vehicle, is given the determination result of suitability for the mounted mounting state being imaged. These pieces of information are information on a man-powered vehicle in a mounted state that has been collected by the manufacturer of the man-powered vehicle or component and has been determined to be appropriate or inappropriate. Still image data or moving image data is collected at a dealer, an assembly plant or a test site of a human-powered vehicle. The still image data or the moving image data is classified into the types of target parts. In addition to the still image data or the moving image data, the input information may include the component identification information, the vehicle body information, or the user's physical information shown in the first embodiment.

The processing unit 10 executes the following generation processing for each part. For example, the processing unit 10 performs the following processing using still image data or moving image data classified as an image of the front derailleur as teacher data.

The processing unit 10 selects a target part with respect to the mounting state of the parts of the human-powered vehicle stored in the storage unit 12 (step S151). At least one of the still image data and the moving image data obtained by imaging the human-powered vehicle to which the parts are attached is acquired (step S153). The still image data or the moving image data may be data obtained by imaging the entire human-powered vehicle, but may also be data obtained by enlarging the target for determining the suitability of the mounting state.

The processing unit 10 acquires identification information of a part attached to the target human-powered vehicle for which a still image or a moving image is captured (step S155). In step S155, the processing unit 10 acquires, for example, the product number or model name of the front derailleur. The processing unit 10 acquires vehicle body information (step S157) and acquires physical information of the user (step S159). The processes of steps S155, S157, and S159 may be omitted, or any one or two of them may be executed.

The processing unit 10 creates teacher data by adding a label of a suitability determination result given in advance to the input information acquired in steps S153 to S159 (step S161).

The processing unit 10 determines whether or not the teacher data is created from all the information groups that satisfy the predetermined conditions for the target component among the information groups stored in the storage unit 12 (step S163). The predetermined condition is, for example, an upper limit on the number of information to be teacher data with respect to acquiring input information for the same mounting state. When it is determined in step S163 that the teacher data has not been created from all the information groups (S163: NO), the processing unit 10 returns the processing to step S153.

When it is determined in step S163 that the teacher data has been created from all the information groups (S163: YES), the processing unit 10 gives the created teacher data to the NN (step S165). The processing unit 10 learns a parameter including at least one of the weight and the bias in the NN mesosphere from the given teacher data (step S167), and ends the processing.

As a result, the learning model 12M is generated by adjusting the parameters in the NN mesosphere by the computer according to the teacher data in which the evaluation for the input information is labeled in advance.

Also in the second embodiment, as shown in the flowchart of FIG. 3 of the first embodiment, the processing unit 10 inputs one still image data or moving image data and outputs information regarding suitability to be output from the NN, and A learning model 12M may be generated by obtaining a difference from a known determination result and propagating the error. The flowchart and detailed description of the corresponding processing procedure will be omitted. The processing unit 10 may adjust the parameters of the intermediate layer by error propagation after the processing procedure shown in the flowchart of FIG.

FIG. 10 is a diagram showing an outline of the learning model 12M of the second embodiment. The learning model 12M includes an input layer 31 for inputting input information regarding parts of the human-powered vehicle, and an output layer 32 for outputting output information regarding suitability for the mounting state of the parts of the human-powered vehicle. The learning model 12M includes an intermediate layer 33 having parameters trained by the teacher data so as to output suitability when input information is input. When the input information regarding the component is input to the input layer 31, the intermediate layer 33 performs the calculation according to the learned parameters, and the output information regarding the suitability of the component input from the output layer 32 is output.

In the specific example of FIG. 10, the input layer 31 is provided with still image data or moving image data obtained by imaging a human-powered vehicle with respect to the suitability of the mounting state of the parts. In the input layer 31, apart from the still image data or the moving image data, any one of the input information as shown in the first embodiment, for example, component identification information, vehicle body information, and user's physical information, or All may be given.

A specific example of FIG. 10 is a configuration in which still image data or moving image data, component identification information, and vehicle body information are input. However, the processing in the output device 1 is not limited to the configuration of giving everything. It may be only still image data or moving image data. As shown in FIG. 10, the component information and the vehicle body information may be input to the intermediate layer 33 after the feature amount of the still image data or the moving image data is output, or the still image data or the still image data or the vehicle body information may be input. The moving image data may be input to the input layer 31 after the feature amount is derived in advance using a convolutional neural network.

As the input information of the learning model 12M, the input information may be distinguished according to the object of suitability. As described above, when generating the learning model 12M that outputs the suitability information regarding the installation of the front derailleur, the still image data obtained by imaging the front derailleur and the component identification information are input, and the disc rotor is targeted. In some cases, the still image data of the disc rotor and the vehicle body information may be distinguished so as to be input.

In FIG. 10, the learning model 12M outputs the probabilities of which type of inappropriate state is appropriate or, in the case of inappropriate output information, which type is inappropriate. The processing unit 10 identifies which state is appropriate based on the probability of the output appropriate state, the type A inappropriate state, and the type B inappropriate state. The learning model 12M is not limited to this, and may output the probability of any one of a plurality of states including an appropriate type G state and an appropriate type E state.

The learning model 12M learned in this way outputs a probability regarding the suitability of the mounting state of the parts shown in the image when the still image data or the moving image data is newly input.

The learning model 12M is used for proposing the mounting state of the following parts. 11 and 12 are flowcharts showing an example of a processing procedure related to the proposal of the mounting state. The proposal of the mounting state is started when the terminal device 2 used by the user executes the maintenance application program 2P and starts the maintenance.

Based on the maintenance application program 20P, the processing unit 20 of the terminal device 2 displays a screen that prompts the user to capture a still image or a moving image of a part to be judged as appropriate on the input screen of information about the human-powered vehicle. It is displayed on the unit 26 (step S221). The processing unit 20 receives the operation of starting imaging by the operation unit 28, and the image pickup unit 29 captures a still image or a moving image of the human-powered vehicle (step S223). In step S223, the processing unit 20 may read the data of the still image or the moving image captured in advance from the storage unit 12.

The processing unit 20 receives the identification information of the parts of the human-powered vehicle attached to or will be assembled from the imaged human-powered vehicle and the vehicle body information based on the input to the input screen (step S225).

The processing unit 20 transmits a determination request including the captured still image or moving image data, the input component identification information, and the vehicle body information from the communication unit 24 to the output device 1 (step S227). ). In step S227, the processing unit 20 specifies information for identifying the type of component to be imaged and whether it is a front derailleur, a rear derailleur, or a disc rotor.

The output device 1 receives the determination request by the communication unit 14 (step S171), and the processing unit 10 selects the learning model 12M corresponding to the type of the component specified in the determination request (step S173).

The processing unit 10 acquires at least one of the still image data and the moving image data obtained by imaging the human-powered vehicle to which the parts are attached from the received determination request (step S175). From the determination request, the processing unit 10 acquires the still image data in step S173 or the identification information of the parts of the human-powered vehicle shown in the moving image data (step S177). The processing unit 10 acquires information on the vehicle body of the human-powered vehicle from the determination request (step S179).

The processing unit 10 gives the still image data or moving image data acquired in step S175, component identification information, and information on the vehicle body to the learning model 12M selected in step S173 (step S181). The processing unit 10 identifies information on the suitability of the mounting state of the component output from the learning model 12M, for example, whether it is an appropriate state, an inappropriate state of type A, or an inappropriate state of type B ( Step S183).

Based on the specified state, the processing unit 10 reads out the proposal information for setting the appropriate state from the storage unit 22 (step S185), and acquires similar image data (step S187). The proposal information in step S185 includes a message when it is identified as an appropriate state for each part type, a proposal content for making it an appropriate state when it is identified as an inappropriate state of type A, and a type B. It is stored in the storage unit 12 as the content of the proposal when it is identified as an inappropriate state. The similar image data may be still image data or moving image data.

The processing unit 20 transmits the suitability information specified in step S183, the proposal information read in step S185, and similar image data from the communication unit 14 to the terminal device 2 of the determination request source (step S189).

Acquisition of similar image data is not essential, and may be only the proposal information read in step S185. The proposed information may be a message that specifically indicates the angle related to the mounting of the component, the distance from the reference point, and the like. When the proposal information is transmitted together with the image data, the proposal information may be only a numerical value, and it is preferable that a numerical value indicating a distance or an angle is displayed together with the image data.

The terminal device 2 receives suitability information, proposal information, and image data from the communication unit 24 (step S229), and the processing unit 20 causes the display unit 26 to display the received suitability information (step S231). The processing unit 20 determines suitability based on the received information (step S233), and if it is determined to be appropriate (S233: YES), ends the process as it is.

If it is determined to be unsuitable in step S233 (S233: NO), the processing unit 20 superimposes an image of similar image data received in step S229 on the still image captured in step S223, and arranges the received proposal information. It is displayed on the display unit 26 (step S235), and the process ends. When the moving image is captured in step S223, the processing unit 20 uses the still image data obtained by extracting any frame image from the moving image data.

Regardless of the determination in step S233, the processing unit 20 may display the image based on the image data extracted as similar and the proposal information on the display unit 26 as a sample of how to attach the parts.

13 and 14 are diagrams showing an example of a screen displayed on the terminal device 2 of the second embodiment. When the user selects and activates the maintenance application program 20P using the terminal device 2, the processing unit 20 determines the suitability of the mounting state of the component based on the maintenance application program 20P. The input screen 281 for prompting imaging is displayed on the display unit 26 (S221).

As shown in FIG. 13, the input screen 261 includes a component selection screen 283 and an image pickup button 285. The input screen 281 also includes a vehicle body information selection screen 287. The input screen 281 includes a transmission button 289 for transmitting a suitability determination request. When the send button 289 is selected, the processing unit 20 transmits a determination request including the information received on the selection screens 263, 265 and 267 from the communication unit 24 (S227). A free input field may be included for any item instead of the selection screens 283 and 287.

FIG. 14 shows an example of a proposal screen 291 that displays suitability information corresponding to a determination request. As shown in FIG. 14, on the proposal screen 291, the proposal information is an image of the component imaged with respect to the mounting state of the component in the human-powered vehicle based on the output information output from the learning model 12M together with the result of suitability. It is superimposed on the image and similar image data is displayed side by side.

In this way, regarding the mounting state based on the learning model 12M, the suitability according to the image data obtained by imaging the mounting state of the component is output, and it is possible to propose the appropriate mounting state of the component according to the observation result of the component. is there.

(Third Embodiment)
FIG. 15 is a block diagram showing the configuration of the terminal device 2 according to the third embodiment. In both the first embodiment and the second embodiment, the suitability information of the mounting state based on the learning model 1M or the learning model 12M is output from the output device 1 managed by the component manufacturer and transmitted to the terminal device 2. It was. On the other hand, in the third embodiment, the maintenance application program 22P of the terminal device 2 used by the user realizes the processing based on the learning model 2M. The processing can be performed only by the terminal device 2 without communicating with the output device 1. In the terminal device 2 according to the third embodiment, the learning model 2M is stored in the storage unit 22 together with the maintenance application program 22P, and the terminal device 2 according to the first embodiment or the second embodiment except for the processing based on the maintenance application program 22P. It is the same as 2. Therefore, the same reference numerals are given to the same configurations, and detailed description thereof will be omitted.

The maintenance application program 22P and the learning model 2M stored in the storage unit 22 may be the ones in which the maintenance application program 52P and the learning model 50M stored in the storage medium 5 are read out and duplicated in the storage unit 22.

Since the processing by the terminal device 2 is common to the processing executed in cooperation with the output device 1 and the terminal device 2 in the first embodiment, detailed description thereof will be omitted.

In this way, it is possible to make a proposal based on the suitability of the mounting state based on the learning model 2M with the terminal device 2 alone.

(Fourth Embodiment)
In the fourth embodiment, the output device 4 is applied as a control unit of a component mounted on a human-powered vehicle. FIG. 16 is a side view of the human-powered vehicle A to which the output device 4 of the fourth embodiment is applied. The human-powered vehicle A is a road bike including an assist mechanism C that assists the propulsion of the human-powered vehicle A by using electric energy. The configuration of the human-powered vehicle A can be arbitrarily changed. In the first example, the human-powered vehicle A does not include the assist mechanism C. In the second example, the type of the human-powered vehicle A is a city cycle, a mountain bike, or a cross bike. In the third example, the human-powered vehicle A includes the features of the first and second examples.

The human-powered vehicle A includes a main body A1, a handlebar A2, front wheels A3, rear wheels A4, a front fork A5, and a saddle A6. The human-powered vehicle A includes a drive mechanism B, an assist mechanism C, an operation device D, a transmission E, a disc rotor G, a display H, and an output device 4. The human-powered vehicle A includes an acceleration sensor S1 and a sound collector S2. The main body A1 includes a frame A12.

The drive mechanism B transmits a human-powered driving force to the rear wheels A4 by a chain drive, a belt drive, or a shaft drive. FIG. 1 illustrates the drive mechanism B of the chain drive. The drive mechanism B includes a crank B1, a first sprocket assembly B2, a second sprocket assembly B3, a chain B4, and a pair of pedals B5.

The first sprocket assembly B2 has a first rotation center axis and is integrally rotatably connected to the crankshaft B11. The first sprocket assembly B2 includes one or more sprockets B22. The center of rotation of the crankshaft B11 and the center of rotation of the first sprocket assembly B2 are coaxial.

The second sprocket assembly B3 has a second rotation center axis and is rotatably supported by a hub (not shown) of the rear wheels A4. The second sprocket assembly B3 includes one or more sprockets B31.

The chain B4 is wound around any sprocket B22 of the first sprocket assembly B2 and any sprocket B31 of the second sprocket assembly B3. When the crank B1 is rotated forward by the human-powered driving force applied to the pair of pedals B5, the first sprocket assembly B2 is rotated forward together with the crank B1, and the rotation of the first sprocket assembly B2 is rotated through the chain B4 to the second sprocket assembly. The rear wheel A4 rotates forward by being transmitted to the solid B3.

The assist mechanism C assists the propulsion of the human-powered vehicle A. In one example, the assist mechanism C assists the propulsion of the human-powered vehicle A by transmitting torque to the first sprocket assembly B2. The assist mechanism C includes an electric motor. The assist mechanism C may include a speed reducer. The assist mechanism C runs a chain B4 that transmits a driving force to the rear wheels A4 of the human-powered vehicle A. The assist mechanism C is a part of a component that can be controlled by signal control for assisting the traveling of the chain B4.

The operation device D includes an operation unit D1 operated by the user. An example of the operation unit D1 is one or more buttons. In another example, the operation unit D1 may be a brake lever. In another example, the operation unit D1 not only accepts the brake operation, but also the operation unit D1 accepts the operation related to the transmission E. Each time the brake lever is tilted to the left or right, the number of gears or the gear ratio in the transmission E can be changed. The operation device D is communicatively connected to the component so that a signal corresponding to the operation of the operation unit D1 can be transmitted and received. In the first example, the operating device D is communicated with the component by a communication line or an electric wire capable of PLC (Power Line Communication). In the second example, the operating device D is communicated and connected to the component by a wireless communication unit capable of wireless communication.

The transmission E can take various forms. In the first example, the transmission E is an external transmission that changes the state of connection between the second sprocket assembly B3 and the chain B4. In the second example, the transmission E is an external transmission that changes the state of connection between the first sprocket assembly B2 and the chain B4. In the third example, the transmission E is an internal transmission. In the third example, the moving part of the transmission E includes at least one of the sleeve and the claw of the internal transmission. In the fourth example, the transmission E is a continuously variable transmission. In the fourth example, the movable part of the transmission E includes a ball planetary (planetary rolling element) of the continuously variable transmission. The transmission E is a part of a component that can be controlled by signal control for changing the number of gears.

The transmission E includes a front derailleur E1 and a rear derailleur E2. The front derailleur E1 is attached to the seat tube A121 of the frame A12. The front derailleur E1 includes a chain guide, a drive motor E11, and a stage number sensor. The front derailleur E1 drives the chain guide of the chain B4 hooked on the sprocket B22 of the first sprocket assembly B2 in the upshift and downshift directions by the drive motor E11. The front derailleur E1 positions the chain guide by the drive motor E11 at any of the shifting positions based on the shifting instruction from the operating device D based on the operating unit D1 or the automatic shifting instruction from the control unit (not shown).

The chain guide of the front derailleur E1 is positioned during installation by the assembly engineer or the user himself. The front derailleur E1 in the fourth embodiment can position the chain guide at an appropriate position by the drive motor E11.

The rear derailleur E2 is attached to the chain stay A17, which is a part of the frame A12 that supports the end of the crankshaft of the rear wheel A4, via the derailleur hanger. The rear derailleur E2 includes a chain guide, a drive motor E21, and a stage number sensor. The rear derailleur E2 drives the chain guide of the chain B4 hooked on the sprocket B31 of the second sprocket assembly B3 in the upshift and downshift directions by the drive motor E21. The rear derailleur E2 positions the chain guide by the drive motor E21 at one of the shifting positions based on the shifting instruction from the operating device D based on the operating unit D1 or the automatic shifting instruction from the control unit (not shown).

The chain guide of the rear derailleur E2 is positioned during installation by the assembly engineer or the user himself. In the rear derailleur E2 in the fourth embodiment, the chain guide can be set to an appropriate position by the drive motor E21.

The disc rotor G is a component that constitutes a disc brake, and is provided around the shafts of the front wheels A3 and the rear wheels A4. The disc brake operates based on a braking instruction from the operating device D based on the operating unit D1 or an automatic braking instruction from a control unit (not shown). The disc rotor G is positioned during installation by an assembly engineer or the user himself. The position of the disc rotor G may fluctuate due to tipping, contact, wear or wear.

The display H is attached to the handlebar A2 for the user who is a rider. The display H is, in one example, a liquid crystal display. The display H is, in another example, an organic EL display. The display H can display the number of gears acquired by communicating with the transmission E. The display H is connected to the output device 4 and displays and outputs the information output from the output device 1.

The acceleration sensor S1 is built in the front derailleur E1 attached to the seat tube A121 of the frame A12. The acceleration sensor S1 outputs a signal indicating the acceleration corresponding to the vibration of the frame A12. The acceleration sensor S1 may be incorporated not only in the front derailleur E1 but also in the rear derailleur E2. The acceleration sensor S1 may be attached in the vicinity of the disc rotor G, for example, in the front fork A5.

The sound collector S2 is built in the front derailleur E1. The sound collector S2 collects and outputs the surrounding sound. The sound collector S2 may include a microphone. The sound collector S2 built in the front derailleur E1 can collect the driving sound of the chain guide of the front derailleur E1, the contact sound between the chain B4 and the chain guide, the contact sound between the chain B4 and the sprocket B22, and the like.

The sound collector S2 may be incorporated not only in the front derailleur E1 but also in the rear derailleur E2. The sound collector S2 built in the rear derailleur E2 can collect the driving sound of the chain guide of the rear derailleur E2, the contact sound between the chain B4 and the chain guide, the contact sound between the chain B4 and the sprocket B31, and the like.

The output device 4 in the fourth embodiment is attached to the seat tube A121 together with the front derailleur E1. The output device 4 may also serve as a drive control unit for the front derailleur E1. FIG. 17 is a block diagram showing the configuration of the output device 4 according to the fourth embodiment. The output device 4 is connected to the front derailleur E1, the rear derailleur E2, the disc brake system including the disc rotor G, the display H, and the operating device D by wire or wirelessly. The output device 4 includes a processing unit 40, a storage unit 42, and an input / output unit 46.

The processing unit 40 is a processor using a CPU or GPU, and uses a built-in memory such as a ROM and a RAM to control a learning algorithm described later and a component provided in the human-powered vehicle A to execute processing.

The storage unit 42 includes a non-volatile memory such as a flash memory. The storage unit 42 stores the control program 40P and the learning model 40M. The control program 40P and the learning model 40M may be those in which the control program 9P and the learning model 90M stored in the storage medium 9 are read out and duplicated in the storage unit 42.

The input / output unit 44 is connected to the sensor S1, the sound collector S2, the operating device D, the drive motor E11 of the front derailleur E1 of the transmission E to be controlled, and the drive motor E21 of the rear derailleur E2 provided in the human-powered vehicle A. Will be done. The processing unit 40 acquires a signal indicating the speed from the acceleration sensor S1 by the input / output unit 44. The processing unit 40 acquires an audio signal from the sound collector S2 by the input / output unit 44. The input / output unit 44 outputs a display signal to the display H.

The output device 1 attached to the human-powered vehicle A uses the learning model 10M and is in an attached state according to the input information that can be acquired from the sensors S1 and the sound collector S2 regarding the attachment state of the parts of the human-powered vehicle. The suitability is acquired and an adjustment instruction signal is output to the drive motor E11, or an adjustment instruction is output to the display H to the user. Regarding the disc rotor G, a proposal for replacement is made to the display H for the user regarding proper installation and wear.

The learning model 40M in the fourth embodiment will be described. The learning model 40M is pre-generated by the component maker. FIG. 18 is a flowchart showing an example of a method of generating the learning model 40M in the fourth embodiment. The learning model 40M is generated by a supervised deep learning algorithm using NN to output output information for input information regarding the mounting state of parts of a human-powered vehicle.

The teacher data corresponds to the input information obtained from the acceleration sensor S1 and the sound collector S2 in advance under the test environment or while the model user is actually driving the human-powered vehicle A, and the input information. It includes information on suitability for the mounting state of parts in the human-powered vehicle A. The teacher data is separated for each component to be determined for suitability, for example, teacher data for the front derailleur E1, teacher data for the rear derailleur E2, and teacher data for the disc rotor G. The input information obtained from the acceleration sensor S1 is data related to vibration based on the signal output from the acceleration sensor S1. Amplitude, frequency, or power values for vibration are collected at predetermined sampling periods. The input information obtained from the sound collector S2 includes audio data of the sound collected by the sound collector S2. In the voice data, information on the waveform, amplitude, or frequency of the voice is collected at predetermined sampling periods. The main body of the suitability judgment is an assembly engineer, a manufacturer engineer, and the like. The subject of the determination may be the user. The input information may also include component identification information, vehicle body information, and user's physical information.

The generator managed by the component maker acquires data on vibration acquired during driving of the human-powered vehicle with respect to the collected mounting state of the parts of the human-powered vehicle (step S301). The generator acquires voice data collected in the same sampling period from the same human-powered vehicle as the vibration data acquired in step S301 (step S303).

The generation device creates teacher data by adding a label of a determination result of suitability of the mounting state of the target component given in advance to the input information acquired in steps S301 and S303 (step S305). In addition to the vibration data and the voice data, the generator may acquire any of the identification information of the attached target part, the vehicle body information, and the physical information of the user as input information.

The generator determines whether or not the teacher data has been created from all the collected information groups (step S307). If it is determined in step S307 that the teacher data has not been created from all the information groups (S307: NO), the generator returns the process to step S301.

When it is determined that the teacher data has been created from all the information groups (S307: YES), the generator gives the created teacher data to the NN (step S309). The generator learns the parameters including at least one of the weight and the bias in the mesosphere of the NN from the given teacher data (step S311), and ends the process.

FIG. 19 is a diagram showing an outline of the learning model 40M according to the fourth embodiment. The learning model 40M includes an input layer 31 for inputting input information regarding parts of the human-powered vehicle A, and an output layer 32 for outputting output information regarding suitability for the mounting state of the parts of the human-powered vehicle A. The learning model 40M includes an intermediate layer 33 having parameters trained by teacher data so as to output suitability when input information is input. When the input information regarding the component is input to the input layer 31, the intermediate layer 33 performs the calculation according to the learned parameters, and the output information regarding the suitability of the component input from the output layer 32 is output.

In the specific example of FIG. 19, the input layer 31 is provided with data regarding vibration obtained while the human-powered vehicle A is being driven and voice data regarding the suitability of the mounting state of the target component. The input layer 31 may be provided with any or all of the input information as shown in the first embodiment, for example, component identification information, vehicle body information, and user's physical information.

The vibration and sound may be input as a graphed image of fluctuations during the sample period rather than the data itself, as shown in FIG.

In FIG. 19, the learning model 40M outputs, as output information, the probabilities of which type of inappropriate state is appropriate or, in the case of inappropriateness, which type of inappropriate state. The processing unit 40 identifies which state is appropriate based on the probability of the output appropriate state, the type A inappropriate state, and the type B inappropriate state.

The learning model 40M learned in this way attaches parts to the driven human-powered vehicle A when vibration data newly obtained from the acceleration sensor S1 and voice data obtained from the sound collector S2 are input. Outputs the probability of the state regarding the suitability of the state.

FIG. 20 is a flowchart showing an example of a processing procedure for determining the suitability of the mounting state in the fourth embodiment. The output device 4 uses the learning model 40M stored in the storage unit 42 in advance, and sequentially executes the following processing for each sampling period while the human-powered vehicle A is being driven.

The processing unit 40 of the output device 4 acquires data on vibrations emitted from the driving human-powered vehicle A from the acceleration sensor S1 (step S401). In step S401, the processing unit 40 may acquire the amplitude and frequency of the vibration, or may calculate the power value.

The processing unit 40 acquires voice data collected by the driving human-powered vehicle A from the sound collector S2 (step S403). In step S403, the processing unit 40 may perform a predetermined filter process on the sound collecting signal obtained from the sound collector S2, and acquire the sound after A / D conversion.

The processing unit 40 gives the input information acquired in steps S401 and S403 to the learning model 40M (step S405). The processing unit 40 specifies the information on the suitability of the mounting state output from the learning model 40M when the input information is given to the learning model 40M (step S407).

The processing unit 40 outputs information for adjusting the mounting state of the target component based on the suitability information specified in step S407 (step S409), and ends the process.

The output in step S409 may be the output of the adjustment signal to the drive motor E11 or the drive motor E21 which is an actuator, or may be the display output to the display H.

In this way, in the fourth embodiment, vibration data or voice data acquired while the human-powered vehicle A is being driven is input to the learning model 40M, and a proposal for an appropriate mounting state of parts according to the running state is proposed. It will be possible.

The output device 4 in the fourth embodiment displays and outputs information regarding the suitability of the mounting state to the display H for the user. The output device 4 includes a communication unit that communicates with a communication terminal device such as the terminal device 2 in the third embodiment owned by the user, and information regarding suitability for the terminal device 2 and adjustment of a component attachment state via the communication unit. It may be configured so that a proposal or a proposal for exchange can be made.

In the fourth embodiment, mainly, the output device 4 provided in the vicinity of the front derailleur E1 inputs the vibration and the voice of the front derailleur E1 into the learning model 40M, and outputs the suitability state. Of course, the object for specifying the suitability of the mounting state is not limited to the front derailleur E1. The target may be the distance related to the rear derailleur E2 and the gear, the stem attached to the head tube of the frame A12, the handlebar A2 attached to the stem, and the output device 4 has its respective attachment positions and angles. You may specify the suitability of. The target may be the position and angle of the seat post of the saddle A6, and the position of the seat portion of the saddle A6. The target may be a brake pad, and the output device 4 may specify the suitability of the shape from vibration and voice in the brake. The target may be a bottom bracket or a pedal mounting position. The subject may also include bending of parts and defects. Sensors for acquiring information input to the learning model 4M in order to identify suitability in these objects are appropriately selected and attached to appropriate places for each sensor.

1,4 ... Output device, 10,40 ... Processing unit, 11,42 ... Storage unit, 1P ... Learning program, 1M, 12M, 30M, 32M, 40M, 50M, 90M ... Learning model, 2 ... Terminal device, 20 ... Processing unit, 22 ... storage unit, 24 ... communication unit, 26 ... display unit, 28 ... operation unit, 2P ... maintenance application program, 3, 5, 9 ... storage medium, A12 ... frame, A5 ... front fork, E ... speed change Machine, E1 ... Front derailleur, E2 ... Rear derailleur, E11, E21 ... Drive motor, G ... Disc rotor, H ... Display, S1 ... Accelerometer, S2 ... Sound collector

Claims (15)

An output device including a processing unit that outputs output information related to the mounted state according to input information related to the mounted state of parts of a human-powered vehicle by using a learning model learned by a learning algorithm. The output device according to claim 1, wherein the input information includes identification information of the component. The output device according to claim 1 or 2, wherein the input information includes information about a vehicle body of the human-powered vehicle. The output device according to any one of claims 1 to 3, wherein the input information includes physical information of a user of the human-powered vehicle. The output device according to any one of claims 1 to 4, wherein the input information includes at least one of still image data of the human-powered vehicle and moving image data of the human-powered vehicle. The output device according to any one of claims 1 to 5, wherein the input information includes data on vibration generated from the human-powered vehicle. The output device according to claim 6, wherein the data related to the vibration includes voice data. The output device according to any one of claims 1 to 7, wherein the output information includes suitability information regarding suitability of the mounting state. The output device according to any one of claims 1 to 8, wherein the output information includes proposal information regarding a proposal for the mounting state. The output device according to claim 9, wherein the proposed information includes at least one of information on how to attach the component, information on adjusting the component, and information on replacement of the component. A learning model generation method in which a learning model is generated using a neural network configured to output output information regarding the mounting state according to input information regarding the mounting state of parts of a human-powered vehicle.
A method of generating a learning model in which parameters in the intermediate layer of the neural network are adjusted by a computer according to the teacher data labeled with the evaluation of the input information. The method for generating a learning model according to claim 11, wherein the learning model is updated by a computer based on additional teacher data labeled with an evaluation for the output information output by giving the input information to the neural network. A learning model generation method in which a learning model is generated using a neural network configured to output output information regarding the mounting state according to input information regarding the mounting state of parts of a human-powered vehicle.
A method of generating a learning model in which parameters in an intermediate layer of the neural network are adjusted by a computer according to teacher data labeled with an evaluation for the output information output by giving the input information to the neural network. The acquired input information is given to the learning model trained by the learning algorithm so as to output the output information regarding the mounting state according to the input information regarding the mounting state of the parts of the human-powered vehicle.
Based on the output information output from the learning model, the computer is made to execute a process of superimposing and arranging the information on the mounting state of the parts in the human-powered vehicle on the image on the display unit displaying the image of the human-powered vehicle. , Computer program. A storage medium in which the computer program according to claim 14 is stored.

Images