JP7514787B2 - Abnormal part identification device and abnormal part identification method - Google Patents

Description

The present invention relates to an abnormal part identification device and an abnormal part identification method.

Automobiles and other transport equipment have multiple electronic control units (ECUs) that control the transport equipment. When controlling the transport equipment, data is shared between these multiple ECUs. CAN (Controller Area Network) is used as the communication protocol for sharing data. Automobiles use an in-vehicle network via CAN communication to diagnose abnormalities using CAN message data (hereafter referred to as CAN data), which is time-series data that flows over CAN.

As a technology for identifying and labeling anomalies in time-series data, Patent Literature 1 discloses an apparatus for detecting and labeling corresponding intervals. This labeling apparatus includes a corresponding interval detection unit and a matching unit. The corresponding interval detection unit extracts corresponding intervals, which are intervals in which first time-series data based on a first sensor output and second time-series data based on a second sensor output are similar or co-occur. The matching unit labels information that identifies the detected corresponding intervals.

JP 2018-109882 A

However, to detect vehicle abnormalities from the vehicle's CAN data and identify the abnormal locations, it is necessary to detect similar or co-occurring sections from hundreds of types of CAN data. However, it is difficult to identify the abnormal locations because there are many combinations, nonlinear correlations exist, and there is CAN data that is not actually abnormal but changes in response to other CAN data. For this reason, even if a corresponding section is detected, the accuracy of determining that the section is abnormal is low, and the accuracy of determining that the labeled label is the correct label is also low.

The present invention was made in consideration of the above problems, and its purpose is to provide an abnormal part identification device and an abnormal part identification method that can accurately identify an abnormal part from a fault code.

In view of the above object, an abnormality part identification device according to one aspect of the present invention identifies an abnormality part of a target vehicle from a failure code of the target vehicle using at least one abnormality part identification model selected from a plurality of abnormality part identification models trained from learning data in which failure codes detected from a plurality of vehicle models different from the target vehicle are associated with abnormal parts of the plurality of vehicle models. The abnormality part identification device receives a signal indicating a failure code of the target vehicle, receives a signal indicating parts information of the plurality of vehicle models, uses the parts information to select at least one abnormality part identification model from the plurality of abnormality part identification models in which the target vehicle parts associated with the failure code are the same or similar, and identifies the abnormality part from the failure code using the at least one abnormality part identification model.

According to one aspect of the present invention, the abnormal part can be accurately identified from the fault code.

FIG. 1 is a block diagram showing the overall configuration of an abnormal part identifying system including an abnormal part identifying device according to the first embodiment. FIG. 2 is a flowchart showing an example of the operation of the abnormal part identifying device (2, 4, 5) of FIG. FIG. 3 is a block diagram showing the overall configuration of an abnormal part identifying system including an abnormal part identifying device according to the second embodiment. FIG. 4 is a flowchart showing an example of the operation of the abnormal part identifying device (2, 4, 5) of FIG. FIG. 5 is a block diagram showing the overall configuration of an abnormal part identifying system including an abnormal part identifying device according to the third embodiment. FIG. 6 is a flowchart showing an example of the operation of the abnormal portion identifying device (2, 3, 4, 5) in FIG.

The embodiment will be described with reference to the drawings. In the description of the drawings, the same parts are given the same reference numerals and the description will be omitted.

First Embodiment
An overall configuration of an abnormal part identification system including an abnormal part identification device (an example of a controller) according to the first embodiment will be described with reference to Fig. 1. The abnormal part identification system includes an abnormal part identification device (2, 4, 5) that identifies an abnormal part of a target vehicle from a fault code detected from the target vehicle, and various databases (1, 3) that store fault codes detected from the target vehicle and a plurality of vehicle models different from the target vehicle, analysis results of the fault codes, and part information of the target vehicle and the plurality of vehicle models.

An example of a "target vehicle" is a vehicle under development or testing that has not yet been sold (hereafter referred to as a "development vehicle"). On the other hand, an example of "multiple vehicle models different from the target vehicle" is multiple vehicle models that have already been sold and traded in the market (hereafter referred to as "market vehicles"). For market vehicles, many fault codes are detected, and there is a large amount of learning data that associates fault codes with abnormal parts. On the other hand, for development vehicles, there are few fault codes and learning data. "Abnormal" is a concept that includes not only something that is not normal, that is, a fault that prevents a vehicle from performing a specified function, capability, or characteristic that it originally possesses, but also a malfunction that does not result in a breakdown and a sign of a breakdown.

An example of a "fault code" is a diagnostic trouble code (DTC) consisting of one alphabetical letter and four digits. DTCs are fault codes for the self-diagnostic function (OBD2: On-Board Diagnostics 2) programmed into multiple electronic control units (ECUs) that control vehicles and other transport equipment. DTCs are standardized by international standards, and some are commonly defined regardless of the vehicle manufacturer, while others are freely defined by each vehicle manufacturer. When a vehicle breaks down, a vehicle dealer can obtain the DTC by connecting a diagnostic device to the vehicle, and the fault can be identified from the DTC, allowing the fault to be replaced or repaired. However, multiple DTCs are often displayed on the diagnostic device, and in such cases, the dealer's engineers identify the fault by analyzing the vehicle and CAN data. In this case, identifying the fault requires a huge amount of man-hours, and the accuracy of identifying the fault depends on the person's skill, so it can be difficult to identify which DTC is the correct DTC.

As shown in FIG. 1, the abnormal part identification device (2, 4, 5) according to the first embodiment includes a first identification device (marketed vehicle abnormal part identification device) 2 that identifies abnormal parts of marketed vehicles, a second identification device (target vehicle abnormal part identification device) 4 that identifies abnormal parts of a target vehicle, and a model selection device (abnormal part identification model selection device) 5 that selects at least one abnormal part identification model to be used by the second identification device 3 from among multiple abnormal part identification models used by the first identification device 2. The abnormal part identification devices (2, 4, 5) also include a vehicle data labeling device that assigns information indicating the identified abnormal part to vehicle data (CAN data or DTC acquired from the vehicle).

The abnormal part identification device (2, 4, 5) can be realized using a microcomputer equipped with a CPU (Central Processing Unit), a memory, and an input/output unit. A computer program for causing the microcomputer to function as the abnormal part identification device (2, 4, 5) is installed and executed in the microcomputer. This allows the microcomputer to function as the multiple information processing units (9, 13, 15, 16, 17, 21) included in the abnormal part identification device (2, 4, 5). Here, an example is shown in which the abnormal part identification device (2, 4, 5) is realized by software, but it is of course possible to configure the abnormal part identification device (2, 4, 5) by preparing dedicated hardware for executing each information processing. The dedicated hardware includes devices such as application specific integrated circuits (ASICs) and conventional circuit components arranged to execute the functions described in the embodiment. The abnormal part identification device (2, 4, 5) may be a single piece of hardware, but it is also possible to configure the first identification device 2, the second identification device 4, and the model selection device 5 as separate hardware.

The various databases (1, 3) include a marketed vehicle DB (marketed vehicle database) 1 and a target vehicle DB (part identification target vehicle database) 3. The marketed vehicle DB 1 includes an abnormal part analysis result DB (abnormal part analysis result database) 6 that stores the analysis results of abnormal parts for each marketed vehicle, a marketed vehicle DTCDB (marketed vehicle DTC database) 7 that stores DTCs detected from marketed vehicles for each vehicle, and a marketed vehicle parts DB (marketed vehicle parts database) 8 that stores part data for marketed vehicles. The target vehicle DB 3 includes a target vehicle parts DB (target vehicle parts database) 11 that stores information on parts equipped in a target vehicle for which abnormal parts are required to be identified, a target vehicle CAN data DB (target vehicle CAN data database) 12 that stores CAN data output from the target vehicle, and a target vehicle DTCDB (target vehicle DTC database) 20 that stores DTCs detected from the target vehicle.

The first identification device 2 includes an abnormal part identification unit 9 and a display unit 10 that displays the identified abnormal part. The abnormal part identification unit 9 identifies abnormal parts of market vehicles using DTCs detected from market vehicles and stored in the market vehicle DTCDB 7. The first identification device 2 generates multiple abnormal part identification models 18 in advance using data stored in the abnormal part analysis result DB 6, the market vehicle DTCDB 7, and the market vehicle parts DB 8, and stores them in the abnormal part identification unit 9. The abnormal part identification unit 9 then identifies abnormal parts of market vehicles from the DTCs of market vehicles using the multiple abnormal part identification models 18. The abnormality analysis results are displayed on the display unit 10. There is no particular limit to the method for creating the abnormal part identification model 18, and it may be created using a known method, such as machine learning such as a Bayesian network or deep learning, or it may be created based on rules.

The multiple abnormal part identification models 18 include models classified by vehicle type, models classified by vehicle parts regardless of vehicle type, models classified by combinations of vehicle types and parts, and models classified by combinations of multiple DTCs regardless of vehicle type and parts.

The second identification device 4 includes an abnormal part identification unit 13, a display unit 14 that displays the identified abnormal part, and a labeling unit 21. The abnormal part identification unit 13 identifies the abnormal part of the target vehicle using the CAN data output from the target vehicle stored in the target vehicle CAN data DB 12. The abnormal part identification unit 13 identifies the abnormal part of the target vehicle from the CAN data of the target vehicle using at least one abnormal part identification model 19 selected by the model selection device 5. The abnormal part identification unit 13 detects the DTC of the target vehicle from the CAN data of the target vehicle and identifies the abnormal part of the target vehicle from the DTC of the target vehicle. The abnormal part identification unit 13 may identify the abnormal part of the target vehicle from the DTC of the target vehicle read from the target vehicle DTCDB 20. Examples of the display unit 10 and the display unit 14 include screen display devices such as monitors and tablets, but are not limited to these, and any known display device can be used. The labeling unit 21 assigns an abnormal part label to the DTC used to identify the abnormal part, and stores the labeled DTC in the target vehicle DTCDB 20.

The model selection device 5 includes a parts determination unit (similar parts determination unit) 15, a DTC determination unit (similar DTC determination unit) 16, and a model selection unit 17. The parts determination unit 15 determines the identity and similarity of parts (vehicle components) between the target vehicle and vehicles on the market. The DTC determination unit 16 determines the identity and similarity of DTCs between the target vehicle and vehicles on the market. The model selection unit 17 selects at least one abnormal part identification model 19 to be used to identify abnormal parts of the target vehicle from among multiple abnormal part identification models 18 based on the part determination results and the DTC determination results.

The parts determination unit 15 compares the data stored in the market vehicle parts DB 8 with the data stored in the target vehicle parts DB 11, and identifies parts or parts groups of market vehicles that are identical or similar to the parts or parts groups of the target vehicle. The similar parts or parts groups may be grouped according to part performance, or may be grouped by machine learning. The DTC determination unit 16 compares the data stored in the market vehicle DTCDB 7 with the data stored in the target vehicle DTCDB 20, and identifies DTCs of market vehicles that are identical or similar to the DTCs of the target vehicle.

The model selection unit 17 selects one abnormality part identification model 19, which has the same or similar target vehicle parts related to the DTC of the target vehicle, from among the multiple abnormality part identification models 18. "Target vehicle parts related to the DTC of the target vehicle" refers to parts of the target vehicle that may have an abnormality, and indicates parts related to the abnormality indicated in each DTC. The model selection unit 17 uses part information of market vehicles that are the same or similar to the target vehicle parts and DTC information of market vehicles that are the same or similar to the target vehicle DTC to identify a vehicle model that has the same or similar target vehicle parts related to the DTC of the target vehicle. For example, the model selection unit 17 identifies at least one vehicle model of multiple vehicle models that has parts that are the same or similar to the target vehicle parts related to the DTC of the target vehicle. The model selection unit 17 selects at least one known abnormality part identification model 19 for the identified vehicle model.

Here, an example is shown in which the model selection unit 17 identifies the vehicle model and selects an abnormal part identification model 19 classified for each vehicle model. However, this is not limited to the above, and the model selection unit 17 can also select an abnormal part identification model 19 classified for each part regardless of the vehicle model. In this case, the model selection unit 17 can select one abnormal part identification model 19 without identifying the vehicle model, using at least one of part information of vehicles on the market that are identical or similar to the parts of the target vehicle and DTC information of vehicles on the market that are identical or similar to the DTC of the target vehicle.

That is, as described above, the abnormal part identification models 19 include models classified by parts regardless of vehicle type, and models classified by DTC regardless of vehicle type and parts. For these models, the model selection unit 17 can directly select an abnormal part identification model 19 in which the target vehicle parts related to the target vehicle's DTC are the same or similar from the parts information and DTC information, without specifying the same or similar vehicle type.

With reference to Figure 2, an example of the operation of the abnormal part identification device (2, 4, 5) in Figure 1 will be described.

First, in step S101, the model selection device 5 analyzes the target vehicle DTCDB 20 and the target vehicle CAN data DB 12 to determine whether the DTC of the target vehicle has been detected. If the DTC of the target vehicle has been detected (YES in S102), the process proceeds to step S103. The model selection device 5 extracts the DTC of the target vehicle from the target vehicle DTCDB 20 (S103), and extracts part information of the target vehicle related to the DTC from the DTC of the target vehicle (S104). In step S105, the parts determination unit 15 compares the part information stored in the market vehicle parts DB 8 with the part information of the target vehicle related to the DTC of the target vehicle.

Proceeding to step S106, the model selection unit 17 selects an abnormality part identification model 19 having the same or similar target vehicle parts related to the DTC of the target vehicle from among the multiple abnormality part identification models 18. For example, the model selection unit 17 first identifies a vehicle model having the same or similar target vehicle parts related to the DTC of the target vehicle. Then, the model selection unit 17 selects at least one known abnormality part identification model 19 for the identified vehicle model.

In step S107, the abnormal part identification unit 13 applies the selected abnormal part identification model 19, and in step S108, the selected abnormal part identification model 19 is used to identify the abnormal part of the target vehicle from the CAN data of the target vehicle. In step S109, the second identification device 4 displays the identified abnormal part on the display unit 14. In addition, the label assignment unit 21 assigns an abnormal part label to the DTC used to identify the abnormal part, and stores the labeled DTC in the target vehicle DTCDB 20.

According to the first embodiment, the following effects can be obtained.

The abnormal part identification device (controller) identifies the abnormal part of the target vehicle from the DTC detected from the target vehicle using one abnormal part identification model 19 selected from multiple abnormal part identification models 18 trained by learning data in which DTCs (fault codes) detected from market vehicles (multiple vehicle types) different from the target vehicle are associated with abnormal parts of the market vehicles. The abnormal part identification device receives a signal indicating the DTC of the target vehicle, receives a signal indicating parts information of the market vehicles, and uses the parts information of the market vehicles to select one abnormal part identification model 19 from the multiple abnormal part identification models 18 in which the parts of the target vehicle related to the DTC are the same or similar, and identifies the abnormal part from the DTC using the selected abnormal part identification model. This makes it possible to increase the learning data (including teacher data) for identifying abnormal parts, and improve the accuracy of the label. In addition, since the selected abnormal part identification model 19 is a known abnormal part identification model and the performance of the model is quantitatively evaluated, it becomes possible to quantitatively evaluate the accuracy of labeling.

The multiple abnormal part identification models 18 are multiple abnormal part identification models classified by multiple vehicle types. The abnormal part identification device uses parts information of vehicles on the market to identify vehicle types in which the target vehicle parts related to the DTC are the same or similar, and selects the abnormal part identification model 19 for the identified vehicle type. This allows data from multiple vehicle types, not just a single vehicle type, to be used as training data, and increases the training data used for machine learning to identify abnormal parts. Instead of increasing the training data, the number of abnormal part identification models may be increased. Furthermore, new vehicle types for the abnormal part identification model may be added. Even for vehicle types that are under development or testing and have not yet been released on the market, abnormal parts can be accurately identified by using learning data related to other vehicle types on the market.

The abnormal part identification device identifies parts of the target vehicle related to the DTC (fault code) as parts of the target vehicle that may be abnormal. The model selection unit 17 identifies at least one model of a vehicle on the market that has a part that is the same as or similar to the identified part. This makes it possible to increase the amount of learning data for identifying abnormal parts and improve the accuracy of the labels.

The fault code is the DTC information detected from the target vehicle. This allows the use of learning data from multiple vehicle models across multiple automakers. Increasing the amount of learning data can improve the accuracy of the labels. Also, the accuracy of the corresponding sections of the CAN data can be improved.

Second Embodiment
The overall configuration of an abnormal part identification system including an abnormal part identification device according to the second embodiment will be described with reference to Fig. 3. The abnormal part determination system according to the second embodiment differs from the first embodiment in that the model selection device 5 further includes a model registration unit 22. However, other points are the same as the abnormal part determination system in Fig. 1, and therefore description thereof will be omitted.

The model registration unit (controller) 22 stores the learning data that associates the DTC of the target vehicle with the abnormal part identified from the DTC of the target vehicle as learning data for a vehicle model different from vehicles on the market. The first identification device 2 can use this learning data to generate an abnormal part identification model for a vehicle model different from vehicles on the market. The model registration unit 22 registers the abnormal part identification model 19 selected by the model selection unit 17 in the abnormal part identification unit 9 as an abnormal part identification model 18 for a new vehicle model. New vehicle models include newly developed vehicle models that have not yet been released to the public and vehicle models that are not included in the abnormal part identification model 19. This makes it possible to register the abnormal part identification model 19 in the market vehicle DB 1 and the abnormal part identification unit 9 for new vehicle models as well as existing vehicle models.

An example of the operation of the abnormal part identification device (2, 4, 5) in FIG. 3 will be described with reference to FIG. 4. Compared to FIG. 2, it differs in that it further includes step S110, but steps S101 to S109 are common, so their description will be omitted.

After the label is added to the CAN data of the target vehicle, the process proceeds to step S110, where the model registration unit 22 registers the abnormal part identification model 19 selected by the model selection unit 17 in the abnormal part identification unit 9 as the abnormal part identification model 18 for the new vehicle model. This improves the accuracy of the labels and makes it possible to accommodate new vehicle models.

Third Embodiment
The overall configuration of an abnormal part identification system including an abnormal part identification device according to the third embodiment will be described with reference to Fig. 5. The abnormal part determination system according to the third embodiment differs from the first embodiment in that the target vehicle DB 3 further includes a predetermined period setting unit 23 and a DTC extraction unit 24. However, other points are the same as the abnormal part determination system in Fig. 1, and therefore description thereof will be omitted.

The DTC extraction unit 24 acquires the DTC of the target vehicle from the CAN data (time series data) of the target vehicle. The DTC extraction unit 24 registers the acquired DTC in the target vehicle DTCDB 20. The acquired DTC is sent to the DTC determination unit 16 and compared with the DTCs of vehicles on the market. The acquired DTC is also sent to the abnormal part identification unit 13 and used to identify the abnormal part.

The predetermined period setting unit 23 sets a predetermined period or a predetermined mileage of the target vehicle as a period for the DTC extraction unit 24 to acquire the DTC. The DTC extraction unit 24 acquires the DTC at each period (predetermined period or predetermined mileage) set by the predetermined period setting unit 23. As a result, the DTC received by the DTC determination unit 16 and the abnormal part identification unit 13 is a DTC acquired from the time series data output from the target vehicle during the predetermined period or predetermined mileage. The predetermined period or the predetermined mileage may be determined based on the improvement of the accuracy of the abnormal part identification model, or a unique predetermined value may be determined. When a unique predetermined value is determined, it is preferable to determine the predetermined value multiple times a day. Since the DTC is extracted by connecting a fault diagnosis device at an automobile dealer and the result may differ, it is preferable that the predetermined period is one day or more. In addition, since another abnormality may occur as time passes, it is preferable that the predetermined period is three months or less. As a result, it is possible to identify the abnormal part without using the target vehicle DTCDB 20, and the accuracy of the label can be improved.

An example of the operation of the abnormal part identification device (2, 3, 4, 5) in FIG. 5 will be described with reference to FIG. 6. Compared to FIG. 2, steps S121 and S122 are performed instead of step S101, but steps S102 to S109 are the same, so their description will be omitted.

In step S121, the predetermined period setting unit 23 sets a predetermined period or a predetermined mileage as the cycle for the DTC extraction unit 24 to acquire the DTC. Proceeding to step S122, the DTC extraction unit 24 analyzes the CAN data of the target vehicle for each predetermined period or each predetermined mileage, and acquires the DTC of the target vehicle from the CAN data of the target vehicle. Then, proceeding to step S102, processing similar to that in FIG. 2 (steps S102 to S109) is performed. This makes it possible to identify abnormal parts without using the target vehicle DTCDB 20 of the target vehicle, and improves the accuracy of the labels.

As described above, an embodiment of the present invention has been described, but the descriptions and drawings that form part of this disclosure should not be understood as limiting this invention. Various alternative embodiments, examples, and operating techniques will become apparent to those skilled in the art from this disclosure.

2. First identifying device (controller)
4. Second identification device (controller)
5. Model selection device (controller)
18 models (multiple abnormality location specific models)
19 Models (at least one anomaly location specific model)

Claims (8)

a controller that identifies an abnormal part of a target vehicle from a fault code detected from the target vehicle by using at least one abnormal part identification model selected from a plurality of abnormal part identification models trained from learning data in which fault codes detected from a plurality of vehicle models different from a target vehicle are associated with abnormal parts of the plurality of vehicle models;
The controller:
receiving a signal indicative of the detected fault code from the target vehicle;
receiving a signal indicating parts information for the plurality of vehicle models;
Using the parts information, at least one abnormality part specifying model in which the part of the target vehicle related to the fault code is the same as or similar to the part of the target vehicle is selected from the plurality of abnormality part specifying models;
identifying the abnormal part from the failure code by using the at least one selected abnormal part identification model;
Anomaly location identification device. the plurality of abnormal part identification models are a plurality of abnormal part identification models classified according to a plurality of vehicle types,
The controller:
Using the parts information, identify a vehicle model of the target vehicle that is different from the target vehicle in which the part of the target vehicle that is associated with the fault code is the same or similar;
2. The abnormality part identifying device according to claim 1, wherein the at least one abnormality part identifying model for the identified vehicle type is selected from the plurality of abnormality part identifying models. The controller:
Identifying a part of the target vehicle related to the fault code as a part of the target vehicle that may be abnormal from the fault code;
Identifying at least one vehicle model among the plurality of vehicle models that has a part that is the same as or similar to the identified part;
The abnormality portion identifying device according to claim 2 . The abnormality part identification device according to any one of claims 1 to 3, wherein the fault code is DTC information detected from the target vehicle. The controller:
storing learning data, which associates the received fault code with an abnormality portion identified from the fault code, as learning data for a vehicle model different from the plurality of vehicle models;
The abnormality location identifying device according to claim 2 or 3. The abnormality part identification device according to any one of claims 1 to 5, wherein the received fault code is a fault code obtained from time series data output from the target vehicle for a specified period or a specified mileage. The abnormality part identification device according to any one of claims 1 to 6, wherein each of the plurality of abnormality part identification models is an abnormality part identification model that associates the fault code, the abnormality part, and time series data detected from the vehicle. receiving a signal indicating a detected fault code from the target vehicle;
receiving a signal indicating part information of a plurality of vehicle models different from the target vehicle;
Using the parts information, at least one abnormality part identification model in which the part of the target vehicle related to the fault code is identical or similar is selected from a plurality of abnormality part identification models trained using learning data in which the fault codes detected from the plurality of vehicle types are associated with the abnormal parts of the plurality of vehicle types;
identifying an abnormal part of the target vehicle from a fault code of the target vehicle using the at least one selected abnormal part identification model;
Method for identifying abnormal areas.

Images