Anomaly identification support device and learning method

Abstract

To improve anomaly detail identification support performance of an anomaly identification support device for providing assistance in identifying vehicle anomaly details.SOLUTION: An anomaly identification support device is provided, comprising an AI processing unit configured to receive an input of a failure code generated by a vehicle and infer a part and type of anomaly of the vehicle, the anomaly identification support device being configured to input a failure code to the AI processing unit to cause the AI processing unit to infer a part and type of anomaly corresponding to the failure code.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present technology relates to an abnormality identification support device that supports the task of identifying the nature of an abnormality in a vehicle, and a learning method for AI (artificial intelligence) used in the abnormality identification support device. [Background technology]

[0002] Generally, commercially available vehicles are equipped with a fault diagnosis function called OBD (On-Board Diagnostics), and when a fault is detected, a fault code called DTC (Diagnostic Trouble Code) defined according to the nature of the fault is recorded.

[0003] Such fault codes as DTCs are used to identify vehicle abnormalities at vehicle maintenance facilities such as dealers, vehicle assembly plants, and the like. For example, at a vehicle maintenance facility, a vehicle that has been brought in with a malfunction is connected to a malfunction diagnostic device, which outputs a malfunction code, and an operator identifies the nature of the malfunction based on the output malfunction code.

[0004] In addition, in vehicle assembly plants, workers install various parts such as ECUs (Electronic Control Units) and harnesses to finally complete a vehicle, but there are cases where the completed vehicle does not operate properly due to human error or a malfunction of the part itself. Vehicles with such problems are moved to the repair line within the plant, where the specific nature of the abnormality is identified and repaired. In this process of identifying the abnormality, fault codes such as DTCs are used.

[0005] Currently, the task of identifying the nature of the abnormality is performed by an operator referencing information about the fault codes and inferring the nature of the abnormality from the combination of the codes. However, because a vehicle is equipped with a large number of ECUs and there are many types of fault codes that may be detected by each ECU, the number of combinations of fault codes that may be recorded in the vehicle when an abnormality occurs is enormous. This makes it difficult for an operator to directly identify the nature of the abnormality from the combination of fault codes, and the current situation is that the operator relies on their experience and intuition.

[0006] The following Patent Document 1 discloses a technology that stores an abnormal part estimation table that associates a combination of diagnostic codes (fault codes) with an abnormal part that is estimated to be the cause when multiple diagnostic codes included in the combination are detected simultaneously, and that, when multiple diagnostic codes are detected simultaneously, identifies the abnormal part that is estimated to be the cause based on the abnormal part estimation table. This makes it possible to identify the nature of an abnormality without relying on the experience or intuition of the worker. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2006-226805 Summary of the Invention [Problem to be solved by the invention]

[0008] However, it is practically difficult to create an abnormal part estimation table using the technology of Patent Document 1. As described above, the number of combinations of failure codes is enormous, and in order to be able to handle all possible abnormalities, the amount of information in the abnormal part estimation table would be enormous, and the task of creating the table would be extremely difficult and unrealistic.

[0009] One possible method for assisting in identifying the nature of a vehicle abnormality is to create historical information showing the history of correspondence between fault codes output from the vehicle and information suggesting the nature of the abnormality, such as actually confirmed abnormal parts, for vehicle abnormalities that have occurred in the past, and then perform the identification based on this historical information.

[0010] However, with this method using history information, if a fault code with no past history is input, it is not possible to identify the corresponding abnormality.

[0011] This technology was developed in consideration of the above circumstances, and aims to improve the support performance of an abnormality identification support device that assists in the task of identifying abnormalities in a vehicle by making it possible to derive suggestive information about the abnormality even when a fault code with no past history is input. [Means for solving the problem]

[0012] The abnormality identification support device according to the present technology includes an AI processing unit that receives a fault code output by a vehicle as input and infers an abnormal part and an abnormality type of the vehicle, as well as one or more processors and a storage medium that stores a program executed by the one or more processors, the program including one or more instructions that cause the one or more processors to input a fault code to the AI ​​processing unit and execute a process to infer an abnormal part and an abnormality type corresponding to the fault code.

[0013] Furthermore, the learning method according to the present technology is a learning method that performs machine learning on an AI using fault codes output by a vehicle as learning input data and abnormality types and abnormality parts regarding abnormalities that occur in the vehicle as training data, thereby generating an AI that infers the abnormality types and abnormality parts corresponding to the input fault codes. [Effects of the Invention]

[0014] According to the present technology, it is possible to improve the support performance of a vehicle abnormality identification support device in identifying the nature of an abnormality. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 1 is an explanatory diagram of an anomaly identification support environment according to an embodiment. [Figure 2] 1 is a block diagram illustrating an example of a hardware configuration of an abnormality identification support device according to an embodiment. [Figure 3] FIG. 10 is a diagram illustrating an example of history information showing the correspondence between a failure code, an abnormality portion, and an abnormality type. [Figure 4] FIG. 1 is an explanatory diagram of an AI learning method according to an embodiment. [Figure 5] FIG. 2 is an explanatory diagram of functions of an abnormality identification support device according to an embodiment. [Figure 6] FIG. 10 is a diagram showing an example of display of information showing the inference results of an abnormality portion and an abnormality type. DETAILED DESCRIPTION OF THE INVENTION

[0016] Hereinafter, embodiments of the present technology will be described with reference to the accompanying drawings. FIG. 1 is an explanatory diagram of an anomaly identification support environment according to an embodiment. In the figure, vehicle 100 is a vehicle for which the content of an abnormality is to be identified. In this embodiment, vehicle 100 is equipped with a fault diagnosis function called OBD (On-Board Diagnostics). When a fault is detected by this fault diagnosis function, a fault code called DTC (Diagnostic Trouble Code) defined according to the content of the fault is recorded in vehicle 100.

[0017] The abnormality identification support device 1 is a device that is assumed to be placed at a site where work to identify the details of an abnormality in the vehicle 100 is performed, and to be used by a worker who performs the work to identify the details of the abnormality. Here, an assembly plant for the vehicle 100 is taken as an example of a site where work to identify the details of an abnormality in the vehicle 100 is performed.

[0018] At the assembly plant, a fault diagnosis is performed on the vehicle 100 at a predetermined timing in the assembly process, such as when the assembly of the vehicle 100 is completed, and the fault diagnosis results are recorded. If the vehicle 100 is found to have a malfunction as a result of the malfunction diagnosis, it is moved to a repair line in the factory, where work is carried out to identify the specific details of the malfunction. In this example, the malfunction identification support device 1 is used in the work of identifying the specific details of the malfunction on such a repair line.

[0019] The abnormality identification support device 1 acquires the fault code recorded as the fault diagnosis result for the vehicle 100 in which a fault has been confirmed, generates information suggesting the specific abnormality occurring in the vehicle 100 based on the acquired fault code (hereinafter referred to as "abnormality suggestion information"), and performs a process to present the generated abnormality suggestion information to the worker, thereby supporting the abnormality identification work.

[0020] Specifically, the abnormality identification support device 1 in this embodiment performs a process of deriving information indicating an abnormality location and an abnormality type as abnormality content suggestion information based on a fault code acquired from the vehicle 100.

[0021] In this specification, the terms "abnormal portion," "abnormal type," and "abnormal content" are used as described above, and the definitions of these terms are as follows: "Abnormal part": A rough classification of the components of the vehicle. In this embodiment, since an operator can visually identify external abnormalities such as scratches that occur during the assembly process, the abnormalities targeted for content identification are internal abnormalities of the vehicle 100, and therefore "abnormal part" also indicates a classification of internal parts of the vehicle 100. Specific examples of "abnormal part" include classifications that at least include the type of ECU, such as engine ECU, transmission ECU, door ECU, camera ECU, etc. "Abnormality type": A classification of abnormalities that may occur in multiple parts during assembly. Examples of abnormalities that may occur during assembly include forgetting to connect a connector, a half-fitted connector, and a broken harness, which may occur regardless of the part of the vehicle 100. The abnormality types referred to here are classifications of such abnormalities that may occur regardless of the part. "Abnormality details": This refers to the specific details of the abnormality that is currently occurring. For example, the nth connector of the transmission ECU is only partially mated, or software A was written to the engine ECU but a different software was written instead.

[0022] For clarity, OBD fault codes are defined for each vehicle's ECU (Electric Control Unit), with various codes indicating the nature of the fault. When a fault occurs, not only a single code but a combination of multiple codes may be recorded. The abnormality identification support device 1 of this embodiment is characterized in that it derives the abnormality part and abnormality type corresponding to the fault code using AI (Artificial Intelligence), and details will be explained again later.

[0023] FIG. 2 is a block diagram showing an example of the hardware configuration of the abnormality identification support device 1. As shown in FIG. The specific device form of the abnormality identification support device 1 is not particularly limited, but it may be in the form of a general-purpose computer device that can be used for purposes other than abnormality identification support, such as a PC (personal computer), tablet terminal, smartphone, etc. Alternatively, it may be in the form of a dedicated computer device specialized for the purpose of abnormality identification support.

[0024] As shown in the figure, the abnormality identification support device 1 includes a CPU 11. The CPU 11 executes various processes in accordance with a program stored in a ROM 12 or a program loaded from a storage unit 18 into a RAM 13. The RAM 13 also stores data necessary for the CPU 11 to execute various processes as appropriate. The CPU 11, the ROM 12, and the RAM 13 are connected to one another via a bus 14.

[0025] An AI processing unit 15 is also connected to the bus 14 . The AI ​​processing unit 15 performs inference processing related to abnormality identification support using trained AI. Specifically, the AI ​​processing unit 15 receives the fault code output by the vehicle 100 as input and infers the abnormal part and type of the vehicle 100. In this example, a DNN (Deep Neural Network) architecture is adopted as the architecture of the AI ​​processing unit 15, and an AI model that realizes the above-mentioned inference is created by deep learning. Specific learning methods and configuration examples of the AI ​​processing unit 15 in this embodiment will be explained later.

[0026] In addition, an input unit 16, a display unit 17, a storage unit 18, a communication unit 19, and a media drive 20 are connected to the bus 14. The input unit 16 is made up of operators and operation devices, and various operators and operation devices such as a keyboard, mouse, keys, dial, touch panel, touch pad, and remote controller are assumed. The input unit 16 detects a user operation, and the CPU 11 interprets a signal corresponding to the input operation.

[0027] The display unit 17 is configured to have a display panel capable of displaying images, such as an LCD (Liquid Crystal Display) or an organic EL (Electro-Luminescence) panel, and is used to display various types of information. In this example, the display unit 17 is configured as a display device provided in the housing of the abnormality identification support device 1, but it can also be configured as a display device external to the abnormality identification support device 1. The display unit 17 displays various images on the display screen based on instructions from the CPU 11. In particular, in the present embodiment, the display unit 17 is used to display information indicating the inference results of the AI ​​processing unit 15. Furthermore, based on instructions from the CPU 11, the display unit 17 can also display various operation menus, icons, messages, etc., that is, images as a GUI (Graphical User Interface).

[0028] The storage unit 18 is configured with a solid-state memory such as an SSD (Solid State Drive) or a relatively large-capacity storage device such as an HDD (Hard Disc Drive), and is used to store various types of information.

[0029] The communication unit 19 is configured to be able to perform wired or wireless communication with various external devices, or communication via a network transmission path such as the Internet or a LAN (Local Area Network). In particular, in the present embodiment, the communication unit 19 is capable of performing wireless communication (or wired communication) with the vehicle 100, and the fault code recorded in the vehicle 100 is input to the abnormality identification support device 1 via the communication unit 19.

[0030] The media drive 20 is configured to allow removable media 21 such as a magnetic disk, optical disk, magneto-optical disk, or semiconductor memory to be freely attached and detached, and is configured to be able to read and write data from and to the attached removable media 21. The media drive 20 allows data files such as programs used for various processes to be read from the removable media 21. The read data files are stored in the storage unit 18, and images contained in the data files are displayed on the display unit 17. Furthermore, the computer programs and the like read from the removable media 21 are installed in the storage unit 18 as needed.

[0031] In the abnormality identification support device 1 having the above-described hardware configuration, for example, software for the processing of this embodiment can be installed via network communication by the communication unit 19 or removable media 21. Alternatively, the software may be stored in advance in the ROM 12, the storage unit 18, etc. The CPU 11 performs processing operations based on various programs, thereby executing information processing and communication processing required by the abnormality identification support device 1.

[0032] Next, an AI learning method for inferring the abnormality location and type from the fault code will be described with reference to FIGS. 3 and 4. First, the basic concept of learning is to use historical information on the correspondence between the failure code and the abnormality location and type of abnormality as learning data.

[0033] Specifically, assuming an application to assist in identifying abnormalities in an assembly factory as in this example, the assembly factory stores history information indicating the correspondence between fault codes, abnormal parts, and abnormality types for abnormalities that have occurred in the past, for example in the form of list information. FIG. 3 shows an example of this history information.

[0034] To start learning, a learning data set is first created from this history information. Specifically, for each failure code listed in the history information, the corresponding abnormal part and abnormality type are associated as correct answer data.

[0035] Here, the process of inferring the corresponding anomaly part and type from the fault code could be realized by a single AI model, but if an attempt is made to infer the anomaly type and part using a single AI model, the difficulty of the problem that the AI ​​model must solve will increase, leading to a decrease in learning efficiency and a resulting decrease in inference accuracy, and there is a risk that this will result in an increase in the network size (i.e., an increase in required resources).

[0036] Therefore, in this embodiment, a method is adopted in which the AI ​​model is divided into a first AI model 31 that infers only the type of abnormality using the failure code as input data, and a second AI model 32 that infers the abnormal part using the failure code as well as the type of abnormality inferred by the first AI model as input data.

[0037] Dividing the AI ​​models in this way significantly reduces the difficulty of the problems that each AI model must solve. Specifically, the first AI model 31 only needs to infer the abnormal part, so the difficulty of the problems that it must solve is reduced. Furthermore, the second AI model 32 can use not only fault code information but also anomaly type information as auxiliary information when inferring the abnormal part, so the difficulty of the problems that it must solve is also reduced. As a result, the learning efficiency of each AI model can be improved, and the inference accuracy can be improved accordingly. Furthermore, the network size of each AI model can be made very small, so the network size of the AI ​​processing unit 15 as a whole can be reduced, and the required resources can be reduced.

[0038] FIG. 4 is an explanatory diagram of a learning method for creating the first AI model 31 and the second AI model 32. 4A shows a learning method for first AI model 31. As shown in the figure, a first AI learner 31b is prepared for creating first AI model 31, and the fault codes described in the history information are input to this first AI learner 31b as learning input data, and the anomaly types associated with the input fault codes as correct answer data are input as training data, to perform machine learning as deep learning. This makes it possible to create a first AI model 31 that infers the type of abnormality from the failure code.

[0039] Figure 4B shows a learning method for the second AI model 32. First, a second AI learner 32b is prepared for creating the second AI model 32. The failure code recorded in the above-mentioned history information and the abnormality type associated with the failure code as the correct answer data are input to this second AI learner 32b as learning input data. The abnormality part associated with the correct answer data for the input failure code is also input as training data, and machine learning as deep learning is performed. This makes it possible to create a second AI model 32 that infers the abnormal part from the combined information of the failure code and the abnormality type.

[0040] In this embodiment, the number of categories of abnormality types inferred by the first AI model 31 is smaller than the number of categories of abnormal parts inferred by the second AI model 32. Specifically, in this example, the number of categories of abnormality types is seven in total, namely, "disconnected harness," "forgotten connector," "partially mated connector," "connector failure," "wrong connector connection destination," "wrong software (wrong writing)," and "bent pin," whereas the categories of abnormal parts include at least the type of ECU, as mentioned above, and the specific number of categories is, for example, about several tens to several hundred.

[0041] When the second AI model 32 is configured to use the information on the anomaly type inferred by the first AI model 31 as auxiliary information for inference, the inference accuracy of the second AI model 32 depends on the reliability of the information on the anomaly type input from the first AI model 31. With the above configuration, the first AI model 31 only needs to make inferences for candidates with a smaller number of classifications than the second AI model 32, which simplifies the problem to be solved and, as a result, improves the inference accuracy, that is, improves the reliability of the information on the anomaly type obtained as the inference result. Therefore, the inference accuracy of the second AI model 32 is also improved, and the inference accuracy of the entire AI processing unit 15, that is, the inference accuracy of the abnormality type and abnormality location, can be improved.

[0042] FIG. 5 is an explanatory diagram of functions of the abnormality identification support device 1 according to the embodiment. FIG. 5 shows a functional block showing the functions of the CPU 11 of the abnormality identification support device 1, and also shows the AI ​​processing unit 15 and the display unit 17.

[0043] As shown in the figure, the CPU 11 includes a code acquisition unit F1, an inference control unit F2, and a display control unit F3. It has the function as. The AI ​​processing unit 15 also has a first AI model 31 and a second AI model 32.

[0044] In the CPU 11, the code acquisition unit F1 performs a process of acquiring a fault code from the target vehicle 100.

[0045] The inference control unit F2 inputs the failure code acquired by the code acquisition unit F1 to the AI ​​processing unit 15, and performs processing to infer the abnormal part and type of abnormality corresponding to the failure code. Specifically, the inference control unit F2 in this embodiment inputs a failure code as input data to the first AI model 31 in the AI ​​processing unit 15, and causes the first AI model 31 to execute inference processing for the abnormality type. Generally, depending on the AI ​​model, information indicating the inference result includes information on multiple candidates that may be correct and likelihood information for each candidate. For example, in an AI model that performs image recognition and classifies classes such as "person," "dog," and "cat," information on the classes such as "person," "dog," and "cat" and likelihood information for each class are obtained as inference result information. Therefore, the first AI model 31 also obtains a plurality of candidates for the abnormality type and likelihood information for each candidate as information indicating the inference result.

[0046] The inference control unit F2 inputs multiple candidates for abnormality type obtained as the inference result of the first AI model 31 and likelihood information for each candidate, and determines the candidate with the highest likelihood as the abnormality type information to be provided to the second AI model 32 as auxiliary information. Then, the inference control unit F2 inputs the information on the abnormality type with the highest likelihood thus determined and the fault code acquired by the code acquisition unit F1 as input data to the second AI model 32, and causes the second AI model 32 to perform inference processing on the abnormal part. To be clear, the inference result information of the second AI model 32 also provides information indicating multiple candidates and the likelihood of each candidate.

[0047] The display control unit F3 inputs information on the abnormality type inference results (candidates and likelihood) obtained by the first AI model 31 and information on the abnormality site inference results (candidates and likelihood) obtained by the second AI model 32, and performs processing to display information showing the inference results of the abnormal site and abnormality type on the display screen 17a of the display unit 17.

[0048] FIG. 6 is a diagram showing an example of displaying information showing the inference results of the abnormality location and the abnormality type. As shown in the figure, the information showing the inference result may be displayed as information showing candidates for abnormality areas and abnormality types in order of likelihood.

[0049] Here, one possible method for presenting the abnormality location and abnormality type corresponding to a failure code to the worker is to derive only the abnormality location and abnormality type with the highest likelihood for the input failure code and present that single abnormality location and abnormality type to the worker. However, in this case, if an error occurs in the derived result, it would be extremely difficult for the worker to identify the abnormality. By displaying information on multiple candidates for the abnormality location and abnormality type that may be correct as described above, the worker can identify the abnormality based on the displayed multiple candidate information for the abnormality location and abnormality type, even if an error occurs in the likelihood. For example, even if the abnormality estimated by the worker based on the candidate information for the abnormality location and abnormality type with the highest likelihood does not occur, the user can infer a new abnormality from the candidate information for the abnormality location and abnormality type with the second highest likelihood. This can lead to cases where the abnormality estimated is actually the abnormality. In this case, the worker can correctly identify the abnormality even if the likelihood is incorrect. Therefore, it is possible to improve the ease with which an operator can identify the actual details of an abnormality, and it is possible to improve the support performance for the abnormality identification work.

[0050] It should be noted that when displaying information showing the inference results, it is not essential to display candidates for anomaly location and anomaly type in order of likelihood as in the above example. To solve the problem of displaying only the candidate with the highest likelihood as described above, it is important to display at least multiple candidates, and in that sense, displaying them in order of likelihood is not essential. For example, it is possible to narrow down the candidates to a small number ranked up to second or third in likelihood and display them in a non-order of likelihood.

[0051] It should be noted that this embodiment is not limited to the specific example described above, and various modified configurations can be adopted. For example, in the above example, an inference device that infers the abnormal part and type of abnormality based on a fault code is illustrated as having a function of acquiring the fault code from vehicle 100 and a function of displaying the inference results, but a configuration in which at least one of these functions is implemented in a device separate from the inference device is also conceivable.

[0052] Furthermore, in the above example, the processing of the inference control unit F2 is executed by a single processor as the CPU 11, but it is also possible that the processing is shared and executed by a plurality of processors.

[0053] Furthermore, although the above provides an example of applying the abnormality identification support device according to the present technology to supporting the abnormality identification work at a vehicle assembly plant, the abnormality identification support device according to the present technology can also be applied to supporting the abnormality identification work at vehicle maintenance facilities such as dealerships, and can be widely and preferably applied to supporting the abnormality identification work that is performed based on fault codes obtained by the vehicle's fault diagnosis function.

[0054] As described above, the abnormality identification support device (same as above, 1) as an embodiment includes an AI processing unit (same as above, 15) that receives a fault code output by a vehicle (same as above, 100) as input and infers the abnormal part and type of abnormality of the vehicle, as well as one or more processors and a storage medium storing a program executed by the one or more processors, the program including one or more instructions that cause the one or more processors to input the fault code to the AI ​​processing unit and execute a process to infer the abnormal part and type of abnormality corresponding to the fault code. By using the AI ​​processing unit described above, even if a fault code with no past history is entered, the AI's inference function can derive the corresponding abnormal part and type of abnormality. Therefore, the vehicle abnormality identification support device can improve the support performance for identifying the abnormality content.

[0055] In addition, in the anomaly identification support device as an embodiment, the above instruction causes one or more processors to execute an inference result display process that causes a display unit to display information indicating the inference result by the AI ​​processing unit. This allows the worker to visually recognize information indicating the inference results regarding the abnormal location and type of abnormality. Therefore, it is possible to allow the user to recognize the inference results even in a noisy environment such as a vehicle assembly factory, and to properly support the task of identifying the abnormality.

[0056] Furthermore, in the anomaly identification support device according to the embodiment, the inference result display process displays information indicating multiple candidates for each of the anomaly location and anomaly type obtained as an inference result of the AI ​​processing unit on the display unit. By displaying multiple candidates for the abnormal area and abnormality type, even if the abnormality estimated from one set of candidate information for the abnormal area and abnormality type has not occurred, it is possible to identify the correct abnormality content from other sets of candidate information for the abnormal area and abnormality type. Therefore, it is possible to improve the ease with which an operator can identify the actual details of an abnormality, and it is possible to improve the support performance for the abnormality identification work.

[0057] Furthermore, in the anomaly identification support device as an embodiment, the AI ​​processing unit has a first AI model (31) that has been trained to infer an anomaly type from a failure code by performing machine learning using a failure code as input data and an anomaly type as training data, and a second AI model (32) that has been trained to infer an anomaly type from combined information of the failure code and anomaly type by performing machine learning using a failure code and an anomaly type as input data and an anomaly site as training data, the first AI model infers an anomaly type corresponding to an input failure code that is a failure code given as input data, and the second AI model infers an anomaly site using the input failure code and the anomaly type inferred by the first AI model as input data. As described above, by dividing the AI ​​model into one that infers anomaly types and one that infers anomaly locations, the difficulty of the problems that each AI model must solve is significantly reduced, resulting in improved learning efficiency and improved inference accuracy.In addition, the network size of each AI model can be made very small, which reduces the network size of the entire AI processing unit and reduces the required resources.

[0058] Furthermore, in the anomaly identification support device according to the embodiment, the number of classifications of anomaly types inferred by the first AI model is smaller than the number of classifications of anomaly parts inferred by the second AI model. When the second AI model is configured to use the information on the anomaly type inferred by the first AI model as auxiliary information for inference, the inference accuracy of the second AI model depends on the reliability of the information on the anomaly type input from the first AI model. With the above configuration, the first AI model only needs to make inferences for candidates with a smaller number of classifications than the second AI model, which simplifies the problem to be solved and, as a result, improves the inference accuracy, i.e., improves the reliability of the information on the anomaly type obtained as the inference result. Therefore, the inference accuracy of the second AI model is also improved, and the inference accuracy of the entire AI processing unit, that is, the inference accuracy of the abnormality type and abnormality location, can be improved.

[0059] Furthermore, the learning method as an embodiment is a learning method that performs machine learning of an AI using fault codes output by a vehicle as learning input data and abnormality types and abnormality parts regarding abnormalities that occur in the vehicle as training data, thereby generating an AI that infers the abnormality types and abnormality parts corresponding to the input fault codes. This learning method makes it possible to create an AI model that can derive the corresponding abnormality location and type even when a fault code with no past history is entered. Therefore, the vehicle abnormality identification support device can improve the support performance for identifying the abnormality content. [Explanation of symbols]

[0060] 1 Abnormality identification support device 100 vehicles 11 CPU 12 ROM 13 RAM 14 Bus 15 AI processing section 16 Input section 17 Display 17a Display screen 18 Memory section 19 Communications Department 20 Media Drive 21 Removable Media 31b First AI learning device 32b Second AI learning device 31 First AI Model 32 Second AI Model F1 Code Acquisition Department F2 Inference control unit F3 Display control section

Claims

1. An AI processing unit is provided that receives a fault code output by a vehicle and infers an abnormal part and an abnormality type of the vehicle, One or more processors; and a storage medium storing a program executed by the one or more processors, The program includes one or more instructions that cause the one or more processors to: The fault code is input to the AI ​​processing unit, and a process is executed to infer the abnormal part and type of the fault corresponding to the fault code. Abnormality identification support device.

2. The instructions may cause the one or more processors to: Execute an inference result display process to display information indicating the inference result by the AI ​​processing unit on a display unit. The abnormality identification support device according to claim 1 .

3. In the inference result display process, information indicating a plurality of candidates for each of the abnormality part and the abnormality type obtained as the inference result of the AI ​​processing unit is displayed on the display unit. The abnormality identification support device according to claim 2 .

4. The AI ​​processing unit a first AI model that is trained to infer the abnormality type from the failure code by performing machine learning using the failure code as input data and the abnormality type as training data; and a second AI model that is trained to infer an abnormality location from combined information of the failure code and the abnormality location by machine learning using the failure code and the abnormality location as input data and the abnormality location as training data. The first AI model infers an abnormality type corresponding to an input failure code, which is a failure code given as input data, and the second AI model infers an abnormal part using the input failure code and the abnormality type inferred by the first AI model as input data. The abnormality identification support device according to claim 1 .

5. The number of classifications of abnormality types inferred by the first AI model is smaller than the number of classifications of abnormality parts inferred by the second AI model. The abnormality identification support device according to claim 4.

6. The fault code output by the vehicle is used as learning input data, and machine learning is performed on the AI ​​using the abnormality type and abnormality part of the abnormality that occurs in the vehicle as training data, to generate an AI that can infer the abnormality type and abnormality part corresponding to the input fault code. How to learn.

Images