Estimation system, estimation method, and trained model
The estimation system uses a trained model to specify failure causes and repair methods from fault diagnosis data, addressing the limitations of existing systems by providing rapid and accurate diagnostics.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- TOYOTA JIDOSHA KK
- Filing Date
- 2024-11-12
- Publication Date
- 2026-05-22
Smart Images
Figure 2026085097000001_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to an estimation system, an estimation method, and a learned model.
Background Art
[0002] Patent Document 1 describes a failure diagnosis apparatus that supports reproducing a failure according to a failure code in vehicle failure diagnosis.
Prior Art Documents
Patent Documents
[0003]
Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0004] However, in the technique described in Patent Document 1, although it is possible to reproduce a failure according to a failure code, even if the situation is reproduced, the cause of the failure or the repair method has not been specified. Therefore, development of a technique for easily and quickly specifying the cause of a failure or the repair method from failure diagnosis data obtained by diagnosing the failure is desired.
Means for Solving the Problems
[0005] The estimation system according to the present disclosure inputs, as input data, first data including failure diagnosis data that is data obtained by diagnosing a plurality of types of failures related to a device, into a learned model obtained by performing machine learning on a learning data set including the first data and second data including history data indicating a repair history for the failure in natural language, and obtains, as output data from the learned model, data indicating at least one of a cause of occurrence and a repair method for the detected failure indicated by the input data in natural language.
[0006] The estimation method relating to this disclosure involves a computer that has been trained to use a training dataset, which includes first data containing fault diagnosis data, which is data diagnosing each of several types of failures relating to the equipment, and second data containing history data, which is a history of repairs for the failures expressed in natural language. The first data concerning detected failures, which are failures detected in the equipment, is input as input data to the trained model, and as output data from the trained model, data is obtained that expresses in natural language at least one of the causes of the detected failures indicated by the input data and the repair methods.
[0007] The trained model relating to this disclosure is a trained model that causes a computer to function by taking input data and outputting output data, and uses a training dataset that includes first data which includes fault diagnosis data which is data that diagnoses each of several types of failures relating to equipment and second data which includes history data which shows the repair history for the failure in natural language, to input the first data which is a detected failure which is a failure detected by the equipment as input data, and has been trained to obtain data which shows at least one of the causes of the detected failure indicated by the input data in natural language as output data. [Effects of the Invention]
[0008] According to this disclosure, the cause of the failure or the repair method can be easily and quickly identified from the fault diagnosis data. [Brief explanation of the drawing]
[0009] [Figure 1] This is a functional block diagram showing one example configuration of a learning system that generates a trained model according to an embodiment. [Figure 2] This is a schematic diagram showing an example of the architecture of a learning system that generates a trained model according to the embodiment. [Figure 3] This is a flowchart illustrating an example of processing in a learning system that generates a trained model according to an embodiment. [Figure 4] This is a functional block diagram showing one example configuration of the estimation system according to the embodiment. [Figure 5] This is a flowchart illustrating an example of processing in the estimation system according to the embodiment. [Modes for carrying out the invention]
[0010] The present invention will be described below through embodiments of the invention, but the invention as claimed is not limited to the following embodiments. Furthermore, not all of the configurations described in the embodiments are necessarily essential as means for solving the problem.
[0011] (Embodiment) The estimation system according to this embodiment is a system that estimates at least one of the causes of a malfunction detected by equipment and its repair method using a trained model, and outputs the estimation result as output data in natural language. The malfunction may include what is called a defect. The repair method may be called a countermeasure or repair procedure. When the estimation system according to this embodiment obtains data showing the causes of the malfunction in natural language as output data, it may be called a malfunction cause estimation system or a cause estimation system. When the estimation system according to this embodiment obtains data showing the repair method in natural language as output data, it may be called a repair method estimation system. Furthermore, the learning system according to this embodiment generates the above-mentioned trained model. Below, first, an example of the learning system will be described as an explanation of the learning stage, and then an example of the estimation system will be described as an explanation of the operation stage.
[0012] [Learning Stage] Next, using Figure 1, we will describe an example configuration of the learning system that generates the trained model according to this embodiment. Figure 1 is a functional block diagram showing an example configuration of this learning system.
[0013] The learning system 1p shown in Figure 1 comprises a vehicle 10, a server 20p, and a client 30p. Figure 1 shows an example in which the learning system 1p includes a data acquisition system that obtains the learning dataset, but after data acquisition, learning can be performed using only the server 20p.
[0014] Server 20p is, for example, a server computer used by a vehicle manufacturer. While Server 20p is described as a single device, it may be configured as a distributed system with functions spread across multiple devices. Server 20p may also be a cloud server.
[0015] Client 30p is a client computer used, for example, by a vehicle dealership or repair shop, and may be a general-purpose PC (Personal Computer). Client 30p may also be a server that provides an application program for generating files such as repair reports 35 (described later) for multiple clients, or a server that manages the generated files. Client 30p can connect to server 20p via a network such as a LAN (Local Area Network) or WAN (Wide Area Network). This network may include a wireless communication network.
[0016] The vehicle 10 can be configured, for example, with a first ECU 11, a communication unit 12, a second ECU 13, and an OBD port 14, all of which are connected by an in-vehicle network. ECU stands for Electronic Control Unit. OBD stands for On-Board Diagnostics.
[0017] The first ECU 11 is an ECU that comprehensively controls the vehicle 10, and its control targets include the second ECU 13. The first ECU 11 may include, for example, a microcomputer 11a, a memory 11b, and a communication circuit 11c. The communication circuit 11c is a circuit that communicates with other parts of the vehicle 10.
[0018] Memory 11b stores one or more types of data. In memory 11b, as one type of data, failure diagnosis data 15 can be stored as a result of performing failure diagnosis including determination of whether it is normal or not. The failure diagnosis data 15 is data for diagnosing a failure related to the vehicle 10. The failure diagnosis data 15 is, for example, failure diagnosis data generated by performing failure diagnosis on the second ECU 13 as a result of the first ECU 11 executing self-diagnosis corresponding to on-board diagnosis in the vehicle 10. Alternatively, the failure diagnosis data 15 is data obtained by the second ECU 13 acquiring failure diagnosis data generated by self-diagnosis via the communication circuit 11c.
[0019] The second ECU 13 is a unit that controls the engine. Although not shown, the second ECU 13 may also be provided with a microcomputer, a memory, and a communication circuit in the same manner as the first ECU 11. The microcomputer incorporated in the second ECU 13 controls the engine and performs failure diagnosis including determination of whether the engine is normal or not, and generates failure diagnosis data as a result of the diagnosis. The control of the engine can include, for example, control of fuel injection, ignition timing, idling speed, etc. The second ECU 13 can also perform its own failure diagnosis and failure diagnosis on sensors, actuators, etc. connected to itself. The generated failure diagnosis data is transmitted to the first ECU 11 by the incorporated communication circuit and stored as the failure diagnosis data 15, but may be stored in the memory incorporated in the second ECU 13 at least until it is transmitted. Note that the second ECU 13 may be a unit that controls a part other than the engine, such as a unit that controls the lighting system.
[0020] The fault diagnosis data 15 may be, for example, a DTC (Diagnostic Trouble Code). The DTC is a code for identifying a specific problem detected by the engine or other parts. As exemplified by the DTC, the fault diagnosis data 15 may include encoded data. Further, the fault diagnosis data 15 may include, for example, FFD (Freeze Frame Data). The FFD is related data that records the driving conditions and states of the vehicle at the moment when the DTC occurs and can be time-series data. The FFD can include, for example, engine speed, vehicle speed, temperature, and the like. The DTC and FFD may be recorded when an abnormality is detected during a regular inspection.
[0021] The communication unit 12 includes a communication interface for the vehicle 10 to perform wireless communication with the outside via a network. The OBD port 14 is a communication port for connecting to a terminal device such as an external dedicated diagnostic tester to obtain the fault diagnosis data 15 from the terminal device. Hereinafter, this terminal device is simply referred to as a tester. When obtaining the fault diagnosis data 15 using this tester, since off-board diagnosis instead of on-board diagnosis is sufficient, the fault diagnosis data 15 does not have to be stored in the memory 11b. The first ECU 11 and the OBD port 14 can function as an in-vehicle fault diagnosis device.
[0022] Note that the vehicle 10 may be configured not to include the first ECU 11. In that case, the second ECU 13 may be configured to be able to transmit the fault diagnosis data 15 to the outside via the communication unit 12 or the OBD port 14. Further, the vehicle 10 may include a plurality of second ECUs 13. Also, even when the vehicle 10 includes a plurality of second ECUs 13, it does not have to include an ECU that performs overall control like the first ECU 11. As long as each of the second ECUs 13 has a fault diagnosis function and can transmit the fault diagnosis data 15 to the outside.
[0023] Client 30p is a device that generates a repair report 35 as a report of the results of repairing vehicle 10. Client 30p is also a device that provides the repair report 35 or the repair report 35 and inspection and maintenance record book 36 to server 20p, either in accordance with instructions from server 20p or voluntarily.
[0024] The client 30p comprises, for example, a control unit 31, a communication unit 32, a UI (User Interface) unit 33, and a storage unit 34. The control unit 31 can be implemented, for example, by an integrated circuit, and can be implemented by, for example, a processor, working memory, and a non-volatile storage device. The processor is an MPU or CPU, etc. The client 30p stores a control program to be executed by the processor in this storage device or storage unit 34, and the processor reads the program into the working memory and executes it. In this way, the client 30p can perform its functions, including the function of providing data.
[0025] The communication unit 32 may be equipped with a communication interface for communicating with the server 20p via a network. The communication unit 32 may also be equipped with a communication interface for communicating with the tester via a network or by direct connection. Alternatively, in order to allow the client 30p to function as a tester, the communication unit 32 may be equipped with a communication interface for communicating with the vehicle 10 by connecting to the OBD port 14.
[0026] The storage unit 34 stores various data such as repair reports 35 or repair reports 35 and inspection and maintenance records 36 generated by the user of client 30p. The storage unit 34 also includes a diagnostic tool 37 as an application program that can be executed by the control unit 31.
[0027] Repairs to the vehicle 10 can be performed by checking the fault diagnosis data 15 using a tester or client 30p. A diagnostic tool 37 can be used as needed to check the fault diagnosis data 15. Client 30p can also obtain the fault diagnosis data 15 directly from the vehicle 10, via a tester, or by accessing server 20p. Checking the fault diagnosis data 15 can be done by running the diagnostic tool 37, which can present the diagnostic results to the user in the UI unit 33.
[0028] The repair report 35 is generated by a user of client 30p or the like when a malfunction is repaired. The repair report 35 is an example of history data that shows the repair history for malfunctions of vehicle 10 in natural language. The repair report 35 is an electronic file of a report that shows the repair history for one or more types of malfunctions that occurred in vehicle 10 in natural language. The repair report 35 may include a description of the malfunction indicated by the malfunction diagnosis data 15 in natural language.
[0029] Furthermore, the inspection and maintenance record book 36 is generated by a user of client 30p when an inspection or inspection and maintenance is performed. The inspection and maintenance record book 36 is an electronic file of a record book that describes in natural language the inspection data showing the results of the inspection of the vehicle 10, or the inspection data and the details of the maintenance performed. The inspection and maintenance record book 36 may include a description in natural language of the state of the vehicle 10 as indicated by the fault diagnosis data 15 generated at the time of inspection. Since faults may be detected during inspections that are not caused by faults, such as periodic inspections, the inspection and maintenance record book 36 may include the contents of the repair report 35, or it may include a link to the repair report 35. The state of the vehicle 10 may include driving conditions. In addition, the inspection data showing the results of the inspection of the vehicle 10 may include fault diagnosis data 15 in cases where no fault is detected by fault diagnosis, such as data showing the fact that fault diagnosis was performed.
[0030] Furthermore, both the repair report 35 and the inspection and maintenance record book 36 may be document files containing text or text and images, but they may also be image files created by scanning or photographing paper documents.
[0031] Server 20p is a device that functions as a data acquisition device and a learning device. Server 20p comprises, for example, a control unit 21, a communication unit 22, and a storage unit 24p. The control unit 21 can be implemented by, for example, an integrated circuit, and can be implemented by, for example, a processor, working memory, and a non-volatile storage device. The processor is an MPU (Micro Processor Unit) or a CPU (Central Processing Unit), etc. Server 20p stores a control program executed by the processor in this storage device or storage unit 24p, and the processor reads the program into working memory and executes it. As a result, Server 20p can perform functions including a data acquisition function to collect training datasets and a generation function to generate trained models.
[0032] The communication unit 22 may be equipped with a communication interface that communicates with the vehicle 10 and the client 30p via a network. The storage unit 24p stores fault diagnosis data 15 and the like acquired from one or more vehicles 10 that are the target of the learning process. The fault diagnosis data 15 can be acquired, for example, by the control unit 21 communicating with the communication unit 12 of the vehicle 10 or the communication unit 32 of the client 30p via the communication unit 22. The fault diagnosis data 15 may also include data read by a tester connected via the OBD port 14. The storage unit 24 also stores repair reports 35 or repair reports 35 and inspection and maintenance records 36 acquired by the control unit 21 communicating with the communication unit 32 of the client 30p via the communication unit 22. The storage unit 24 also stores the unlearned models 25p, which will be described later and used in the learning process.
[0033] In the configuration described above, the control unit 21 collects a training dataset and stores it in the storage unit 24p, and executes a training process to train the untrained model 25p. The training dataset may have necessary information added to it by the builder of the trained model. The training dataset is a dataset that includes a first data set containing multiple fault diagnosis data 15, and a second data set containing historical data for each fault, or historical data and inspection data.
[0034] Here, the multiple fault diagnosis data 15 included in the training dataset are data that diagnoses faults for each of several types of faults related to the vehicle 10. Of course, the training dataset includes multiple fault diagnosis data 15 for each type of fault in order to perform accurate estimation processing. The fault diagnosis data 15 includes data showing faults that were actually obtained when the vehicle 10 was used before training, but it may also include data showing faults that were known in advance to be possible to occur in the vehicle 10. Furthermore, the fault diagnosis data 15 included in the training dataset should be the fault diagnosis data obtained when a fault was detected by fault diagnosis. However, as mentioned above regarding inspection data, faults may not be detected by fault diagnosis, so the fault diagnosis data 15 included in the training dataset may also include fault diagnosis data when no fault was detected.
[0035] Furthermore, as mentioned above, an example of historical data included in the second data set is the repair report 35, and an example of inspection data that can be included in the second data set is the inspection and maintenance record book 36. By including inspection data in the training dataset in this way, it is possible to determine whether or not a failure has occurred or is in the way of signs of failure at the inspected location at the time of inspection, and whether or not there is a history of replacement of parts at the inspected location.
[0036] In the learning process, the control unit 21 generates a trained model by performing machine learning on the untrained model 25p using the above-mentioned training dataset. Hereafter, this trained model will be referred to as the trained model 25. The untrained model 25p can use models such as deep learning, but the algorithm is not restricted. The untrained model 25p only needs to be a model that can output output data for input data in the estimation process described later. If the untrained model 25p uses time series data as the first data, it is advisable to use an algorithm that corresponds to time series data. Also, if the untrained model 25p uses data including images as the second data, it is advisable to use an algorithm that corresponds to images. Furthermore, since the untrained model 25p includes the second data in the training dataset, it should be equipped with a language model such as a large-scale language model.
[0037] The input data for the estimation process is the first data set, which includes fault diagnosis data 15 for the detected fault, that is, the fault to be estimated, which is a fault detected in the vehicle 10. The output data for this input data in the estimation process is data that shows the cause and repair method for this detected fault in natural language. The cause and repair method will be learned by machine learning from the repair report 35 or the repair report 35 and the inspection and maintenance record book 36 for the fault diagnosis data 15 in the training dataset. The cause to be included in the output data may be only one or more of the main causes, that is, one or more primary causes.
[0038] The first data included in the training dataset and the first data input in the estimation process may include data indicating the state of the vehicle 10 for a period prior to the occurrence of the failure indicated by the failure diagnosis data 15. For example, the first data may be time-series data that includes not only the failure diagnosis data 15 at the time of the failure, but also data indicating the results of previous failure diagnoses. In other words, the first data may include the failure diagnosis data 15 that includes at least the time of the failure. As illustrated in the FFD, the state of the vehicle 10 may include driving conditions.
[0039] In this example, during the estimation process in operation, by including data indicating the state of the vehicle 10 for the period prior to the failure in the fault diagnosis data 15 at the time of the failure, and inputting this data into the trained model 25, it is possible to produce an output that also takes into account information from before the failure occurred.
[0040] Furthermore, the data to be included as the training dataset should be data for vehicles 10 of the same make and model. However, vehicles that differ only in external appearance may be considered the same make and model. For different types of vehicles that share a common part, data for that common part may be included in the training dataset. Alternatively, if the trained model 25 is not specialized for a single specific make and model, the first data may include not only the fault diagnosis data 15, but also data indicating the make or model, or data indicating the make or model and year. Here, the year refers to the year, month, or date the vehicle 10 was manufactured.
[0041] Next, we will use Figure 2 to describe an example of the architecture of a learning system that generates a trained model 25. Figure 2 is a schematic diagram showing an example of this architecture.
[0042] In the learning system 1, the control unit 21 and the memory unit 24p function as a generator that generates a trained model 25 from an untrained model 25p using the training dataset. As illustrated in Figure 2, this generator may include an encoder 41 and a decoder 42 as the untrained model 25p.
[0043] The combination of encoder 41 and decoder 42 can be an example of an end-to-end architecture, where encoder 41 is a Transformer and decoder 42 is a Transformer. Alternatively, the combination of encoder 41 and decoder 42 can be an example of a lightweight architecture, where encoder 41 is a GRU or LSTM and decoder 42 is a GPT. Both GRU and LSTM are types of Recurrent Neural Networks (RNNs) and are models for learning long-term dependencies. GRU stands for Gated Recurrent Unit. LSTM stands for Long Short-Term Memory. Of course, the combination of encoder 41 and decoder 42 is not limited to these examples.
[0044] The encoder 41 takes FFD time-series data as an example of fault diagnosis data 15, encodes it into a code that can be processed by the decoder 42, and outputs it to the decoder 42. The decoder 42 takes this code as input, along with natural language text data contained in the repair report 35 or the repair report 35 and the inspection and maintenance record book 36. In response to these inputs, the decoder 42 autoregressively generates and outputs data in natural language that shows the cause of the fault and the repair method for the fault contained in the FFD time-series data. In Figure 2, the decoder 42 receives the output code from the encoder 41 along with the following: <s> Enter "Slot body carbon..." and then "Slot body carbon..."< / s> The diagram shows how the system outputs data such as "[...]". <s> tag,< / s> Each tag represents a start tag and an end tag.
[0045] The control unit 21 performs machine learning on the relationship between input and output data for all training datasets for the encoder 41 and decoder 42. Subsequently, the control unit 21 evaluates the performance of the machine-learned model using a separate validation dataset, and completes the generation of the trained model 25 once a performance evaluation suitable for operation is obtained.
[0046] As illustrated above, the server 20p, as a learning device, performs machine learning that links fault diagnosis data 15 with natural language as repair history based on that data. As a result, when fault diagnosis data 15 is input to the server 20p during operation, it obtains a trained model 25 that acts as an estimator, outputting the cause of the failure and the repair method in natural language. This estimator may also be called a predictor. Since the trained model 25 is a model that estimates the cause of the failure and the repair method, it may also be called a cause estimation model or a repair method estimation model.
[0047] Using Figure 3, an example of processing in the learning system 1p that generates the trained model 25 according to this embodiment will be explained. Figure 3 is a flowchart illustrating this processing example.
[0048] First, the control unit 21 collects fault diagnosis data 15 for multiple vehicles 10 from the vehicles 10 or clients 30p and stores it in the storage unit 24p (step S11). The fault diagnosis data 15 may include, for example, DTC and FFD. If the FFD does not include the fault location and fault date and time, these can be added separately to the fault diagnosis data 15 at the time of collection.
[0049] Next, the control unit 21 determines whether the collected fault diagnosis data 15 is sufficient (step S12). If a sufficient amount of fault diagnosis data 15 has not been collected, that is, if the result in step S12 is NO, the control unit 21 returns to step S11 and continues collection. When the result in step S12 is YES, the control unit 21 collects the repair reports 35 and inspection and maintenance records 36 generated by multiple clients 30p from each client 30p and stores them in the storage unit 24p (step S13).
[0050] The control unit 21 determines whether the collected repair reports 35 and inspection and maintenance records 36 are sufficient (step S14). If a sufficient amount of repair reports 35 and inspection and maintenance records 36 have not been collected, i.e., if the answer in step S14 is NO, the control unit 21 returns to step S13 and continues collection. When the answer in step S14 is YES, the control unit 21 completes the collection of fault diagnosis data 15, repair reports 35, and inspection and maintenance records 36 as a learning dataset.
[0051] Fault diagnosis data 15 may include, for example, a P0300 code indicating a cylinder misfire problem in the engine. For example, in the case of the first cylinder, the code would be "P0301". Misfires occur when fuel is not burned properly or when the spark plug is damaged, and in extreme cases, the catalytic converter of the vehicle 10 may also be damaged.
[0052] The repair report 35 generated and collected for a fault indicated by P0300 may include, for example, that wear was found as a result of checking the spark plug for P0300, and that the error was resolved by replacing the spark plug. In this case, the repair report 35 may also include, for example, that the error was not resolved despite checking the wear and replacing the spark plug, and that the error was finally resolved by cleaning the carbon deposits on the throttle body. Furthermore, for example, if the fault diagnosis data 15 indicates an engine misfire, not only ignition failure but also fuel failure or an internal engine malfunction may be contributing factors. Therefore, by collecting more data as a training dataset, the causes and repair methods can be learned. When training on repair methods, taking the detection of a misfire in an engine cylinder as an example, the repair report 35 may include, for example, items to check for the detection, methods for checking each item, and the actual repair locations including the actual check results.
[0053] Furthermore, the repair report 35 may include the date and time of the failure in addition to the location of the failure, as exemplified by the cylinder number. This allows for the association of the repair report 35 with the fault diagnosis data 15 during machine learning. The inspection and maintenance record book 36, which is generated and collected in relation to misfires, may include information such as which spark plug was replaced and when, or when the above-mentioned carbon was cleaned.
[0054] While only an example of cylinder misfire was given for the training dataset, data on each type of malfunction that occurred in each vehicle 10, as well as data from the repair reports 35 and maintenance records 36 generated for each vehicle 10, will be collected.
[0055] Once the collection of the training dataset is complete, the control unit 21 uses the training dataset to machine-learn the untrained model 25p (step S15).
[0056] In this machine learning-generated trained model 25, for example, by taking fault diagnosis data 15 shown as P0301 as input data, the following output data can be estimated and output as a result of machine learning from the repair report 35 for P0300. For example, this output data is data that describes in natural language that the cause of the problem is a misfire in the first cylinder, and that the repair method requires replacing the spark plug of the first cylinder and cleaning the throttle body.
[0057] After machine learning, the control unit 21 uses a validation dataset to determine whether the estimation accuracy of the generated trained model 25 for the cause of the problem and the repair method is above a threshold (step S16). If the result in step S16 is NO, the control unit 21 returns to step S15 and, if necessary, changes to hyperparameters, increases or decreases the training dataset, etc., and performs machine learning again. When the result in step S16 is YES, the trained model 25 obtained thereafter is designated as the trained model to be used in operation.
[0058] [Operational phase] Below, an example configuration of the estimation system according to this embodiment will be described using Figure 4. Figure 4 is a functional block diagram showing an example configuration of this estimation system. The equipment subject to the estimation regarding failures as described above can be a vehicle or a part of a vehicle, and the following explanation will use an example where the equipment is a vehicle. However, the equipment subject to estimation is not limited to a vehicle or a part of a vehicle.
[0059] The estimation system 1 shown in Figure 4 comprises a vehicle 10, a server 20, and a client 30. Note that server 20 and client 30 are computers that perform estimation processing instead of learning processing in the server 20p and client 30p configuration examples, respectively, and a description of the basic configuration example is omitted. Server 20 and client 30 may also have the functions of server 20p and client 30p, respectively. In other words, estimation system 1 may be constructed as a system incorporating learning system 1p.
[0060] The server 20 stores the aforementioned trained model 25 in the memory unit 24 so that it can be executed by the control unit 21. This trained model 25 is a trained model that makes the server 20 function as a computer, taking input data and outputting output data.
[0061] The control unit 21 will perform the following estimation process. This estimation process takes input data into the trained model 25 and outputs output data as the estimation result. As explained in the training stage, the input data in the estimation process is first data including fault diagnosis data 15 for the detected fault, that is, the fault to be estimated, which is a fault detected in the vehicle 10. The output data for this input data in the estimation process is data that shows at least one of the cause of the fault and the repair method for this detected fault in natural language. Even if the cause of the fault is not included in the output data and only the repair method is included in the output data, the dealer or repair shop of the vehicle 10 can still perform the repair by referring to the output data. Of course, the estimation process may also be performed by the control unit 31 of the client 30, with the trained model 25 stored in the storage unit 34 on the client 30 side.
[0062] Next, an example of the estimation process in the estimation system 1 according to this embodiment will be explained using Figure 5. Figure 5 is a flowchart illustrating this example of the estimation process.
[0063] When the estimation process is started, the control unit 21 first receives fault diagnosis data 15 from the vehicle 10 or client 30 via the communication unit 22, etc., and stores it in the storage unit 24 (step S21). In step S21, the fault diagnosis data 15 is stored for management or for later use in relearning the trained model 25, but if it is only used for the estimation process, the fault diagnosis data 15 only needs to be stored temporarily.
[0064] Next, the control unit 21 inputs the received fault diagnosis data 15 into the trained model 25 (step S22). In response to step S22, the control unit 21 obtains data from the trained model 25 describing the cause of the fault and the repair method in natural language (step S23). This result is presented by displaying it on the UI unit 33 or outputting it as audio (step S24), and the operation ends. Through this presentation, dealerships and repair shops for the vehicle 10 can confirm the results of their estimation of the cause of the fault and the repair method. In step S24, if the generated and stored trained model 25 is a model that outputs only either the cause of the fault or the repair method, then either the cause of the fault or the repair method will be output. The process illustrated in Figure 5 can be executed, for example, each time a fault diagnosis is performed or after a certain number of fault diagnoses have been performed.
[0065] Furthermore, after step S24, the dealership or repair shop of the vehicle 10 can use the client 30 to generate repair reports 35 and maintenance records 36 and store them in the storage unit 34. These repair reports 35 and maintenance records 36 may be sent to the server 20 for management. The server 20 may use the fault diagnosis data 15 stored in the storage unit 24 and these repair reports 35 and maintenance records 36 to retrain the trained model 25.
[0066] Before explaining the effects of this embodiment, let's look at a comparative example. In this comparative example, DTCs and FFDs are analyzed in off-board or on-board diagnostics using a tester, and inspections are performed according to the service manual. Because there are a wide range of causes that trigger problems corresponding to DTCs, identifying the cause takes time in this comparative example. For example, if fault code P0300, which corresponds to engine misfire, is observed, several different causes are possible, such as worn spark plugs, faulty fuel injectors, leaking head gaskets, or faulty camshaft sensors. In this case, dealerships and repair shops inspect according to the service manual, but identifying the cause of the failure requires eliminating each possible cause, which takes time.
[0067] In contrast, the estimation system 1 according to this embodiment uses the trained model 25 generated as described above to estimate the cause of the failure, thus reducing the time required to identify the cause. In other words, according to this embodiment, the cause of the failure, the repair method, or both the cause and the repair method can be easily and quickly identified from the failure diagnosis data.
[0068] Furthermore, in this embodiment, repair reports 35, which are written in natural language and include DTCs and associated repair histories, can not only be stored but also utilized to generate a trained model 25 capable of estimating the causes of failures and repair methods. In addition, in this embodiment, DTCs and associated FFDs can not only be stored by the vehicle manufacturer but also utilized to generate the trained model 25 capable of the above estimations. Thus, in this embodiment, fault diagnosis data 15, exemplified by DTCs and FFDs as time-series data, and the repair history of the corresponding failure in natural language, which is a modal different from the fault diagnosis data 15, can be utilized by including them in the training dataset. This trained model 25 is generated as a trained model that includes a language model and is machine-trained on this training dataset. However, the part to be trained may be limited to the language model. In other words, an existing language model can be used to input second data to an untrained model and output output data, and only this untrained model part may be machine-trained.
[0069] Furthermore, this disclosure also includes, as illustrated in Figures 3 and 5, an estimation method in which a computer performs the estimation described above, and a learning method in which a computer performs the learning described above. This disclosure also includes the form of the trained model described above, a program in which a computer performs this estimation method, and a program in which a computer performs this learning method. For example, some or all of the processing in vehicle 10, servers 20, 20p, clients 30, 30p, etc., can be implemented as a computer program. This program, when loaded into a computer, includes a set of instructions (or software code) for causing the computer to perform one or more of the functions described in the embodiments. The program may be stored in a non-temporary computer-readable medium or a tangible storage medium. The program may be transmitted over a temporary computer-readable medium or a communication medium, such as an electrical, optical, acoustic, or other form of propagating signal.
[0070] It should be noted that the present invention is not limited to the embodiments described above, and can be modified as appropriate without departing from the spirit of the invention. For example, although the above embodiments were described using the example that the equipment is a vehicle, the equipment may be other types of mobile devices or other types of equipment that are not mobile devices. [Explanation of Symbols]
[0071] 1 Estimation system, 1p training system, 10 vehicles, 20, 20p servers, 25 trained models, 30, 30p clients, 41 encoder, 42 decoders
Claims
1. A trained model is trained using a training dataset that includes first data containing fault diagnosis data, which is data diagnosing each of several types of failures related to the equipment, and second data containing history data, which is a natural language representation of the repair history for the said failures. The first data containing detected failures, which are failures detected in the equipment, is then input as input data to the trained model. As output data from the trained model, data is obtained that expresses in natural language at least one of the causes of the detected fault and the repair method for the fault indicated by the input data. Estimation system.
2. The first data includes data indicating the state of the equipment for a period prior to the occurrence of the failure indicated by the fault diagnosis data, The estimation system according to claim 1.
3. The second data includes inspection data describing the results of the inspection of the equipment in natural language. The estimation system according to claim 1 or 2.
4. Computers A training model is trained using machine learning, and a training dataset containing first data including fault diagnosis data, which is data diagnosing each of several types of failures related to the equipment, and second data including history data showing the repair history for the said failure in natural language, is input as input data for detected failures, which are failures detected in the equipment. As output data from the trained model, data is obtained that expresses in natural language at least one of the causes of the detected fault and the repair method for the fault indicated by the input data. Estimation method.
5. A trained model that enables a computer to function by taking input data and outputting output data, Using a training dataset that includes first data containing fault diagnosis data, which is data diagnosing each of several types of failures related to the equipment, and second data containing history data, which is a natural language representation of the repair history for the said failures, the machine learning model was trained to input the first data, which is a detected failure that was detected by the equipment, as input data, and to obtain output data, which is a natural language representation of at least one of the causes of the detected failure indicated by the input data and the repair method for the said detected failure. A pre-trained model.