Processor system and fault diagnosis method

Abstract

To achieve a more accurate failure diagnosis by obtaining explicit knowledge having high comprehensiveness that can be applied to a failure diagnosis based on knowledge across a plurality of areas.SOLUTION: A processor system includes one or more processors and one or more memory resources. The memory resource is configured to store a diagnostic program configured to diagnose an electronic system including a mobile body or equipment, and a plurality of knowledge graphs for use in the diagnostic program and including at least a first knowledge graph and a second knowledge graph created based on domain knowledge different from that of the first knowledge graph. The processor is configured to execute the diagnostic program to (1) evaluate a semantic similarity between a first knowledge item included in the first knowledge graph and a second knowledge item included in the second knowledge graph, and (2) generate, when the similarity satisfies a predetermined condition, an integrated graph in which the first knowledge graph and the second knowledge graph are integrated.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a processor system and a fault diagnosis method. [Background technology]

[0002] In electronic systems related to mobile objects such as automobiles and equipment such as robots and processing machines, the causes of malfunctions are diversifying due to the increase in system components accompanying advances in autonomy, connectivity, and the Internet of Things (IoT). For example, in the automotive field, with the advancement of connectivity and autonomous driving, malfunctions caused by the environment surrounding the vehicle and connected systems are on the rise. However, because knowledge about malfunctions (e.g., Fault Tree Analysis (FTA) and service manuals) is vertically segmented and dispersed across various specialized fields related to electronic systems (e.g., equipment, environment, and connected systems), it is extremely difficult to diagnose the causes of malfunctions that span such diverse fields.

[0003] In order to diagnose the causes of these increasingly diverse defects, it is thought that technology that can consolidate knowledge about defects in each area is necessary.

[0004] Patent Document 1 discloses a distributed system that analyzes abnormality factors of edge devices, which are mobile objects or facilities. Specifically, Patent Document 1 states that "the distributed system has an edge device, which is a mobile object or facility that can operate automatically, and a diagnostic data calculator. The edge device has a moving mechanism or an operating mechanism for automatic operation, and an intra-edge controller that controls the moving mechanism or the operating mechanism. Here, the diagnostic data calculator: receives diagnostic data that indicates the internal state of the intra-edge controller." [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent Publication No. 2021-196678 Summary of the Invention [Problem to be solved by the invention]

[0006] For example, when trying to diagnose the causes of a malfunction that spans multiple domains using knowledge graphs from each domain, there are many knowledge graphs due to the complexity of the entire system surrounding the automobile, and it is necessary to integrate these many knowledge graphs for diagnosis.

[0007] However, each knowledge graph is created based on domain knowledge specific to a particular field. Therefore, knowledge graphs created based on different domain knowledge often have different representations of knowledge items with the same meaning due to differences in the domain of the underlying expertise or errors in the machine learning algorithms used to estimate the knowledge graph. Therefore, integrating such multiple knowledge graphs presents a significant challenge.

[0008] Patent Document 1 discloses that data generated from edge devices is analyzed by fault tree analysis using a knowledge graph to identify abnormality factors and their range of impact. However, the technology in Patent Document 1 performs analysis using a knowledge graph for each domain. In other words, the technology in Patent Document 1 does not take into consideration integrating knowledge about malfunctions across domains and diagnosing the causes of the malfunctions based on the integrated knowledge.

[0009] The present invention has been made in consideration of the above-mentioned problems, and aims to achieve more accurate fault diagnosis by obtaining comprehensive explicit knowledge that can be applied to fault diagnosis based on knowledge spanning multiple domains. [Means for solving the problem]

[0010] The present application includes multiple means for solving at least part of the above-mentioned problems, examples of which are as follows: A processor system according to one aspect of the present invention for solving the above-mentioned problems is a processor system having one or more processors and one or more memory resources, wherein the memory resources store a diagnostic program for diagnosing an electronic system including a mobile object or equipment, and a plurality of knowledge graphs used in the diagnostic program, the plurality of knowledge graphs including at least a first knowledge graph and a second knowledge graph created based on domain knowledge different from the first knowledge graph, and the processor executes the diagnostic program to: (1) evaluate the semantic similarity between a first knowledge item included in the first knowledge graph and a second knowledge item included in the second knowledge graph, and (2) generate an integrated graph by integrating the first knowledge graph and the second knowledge graph if the similarity satisfies a predetermined condition. [Effects of the Invention]

[0011] According to the present invention, by obtaining comprehensive explicit knowledge that can be applied to fault diagnosis based on knowledge spanning multiple domains, it is possible to realize more accurate fault diagnosis. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a diagram illustrating an example of a schematic configuration of a processor system. [Figure 2] FIG. 1 is a diagram illustrating an example of a knowledge graph. [Figure 3] FIG. 2 illustrates an example of a functional configuration of a processor system. [Figure 4] FIG. 10 is a flowchart showing an example of an integrated knowledge generation process, a learning process, and a diagnosis process. [Figure 5] FIG. 10 is a diagram illustrating an example of similarity information. [Figure 6] FIG. 10 is a diagram illustrating an example of generating an integrated graph. [Figure 7] FIG. 10 is an explanatory diagram regarding the definition of an input data format using an integrated graph. [Figure 8]1 is a diagram illustrating an example of an overview of a service system to which a processor system according to an embodiment is applied. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, each embodiment of the present invention will be described with reference to the drawings.

[0014] 1 is a diagram showing an example of a schematic configuration of a processor system 100 according to this embodiment. The processor system (hereinafter sometimes referred to as "this system") 100 is a device that diagnoses malfunctions that occur in electronic systems including equipment and mobile objects. Specifically, this system 100 integrates knowledge graphs related to malfunctions in each specialized field (hereinafter sometimes referred to as "field") related to equipment and mobile objects equipped with electronic systems, based on the semantic similarity of the knowledge items included in the knowledge graphs.

[0015] In addition, the present system 100 uses the integrated knowledge graph (hereinafter sometimes referred to as the "integrated graph") to generate learning data to be used for machine learning of an information model (hereinafter sometimes referred to as the "diagnostic model") for identifying the cause of a defect in the object to be diagnosed.

[0016] Furthermore, the system 100 inputs information indicating the state of the malfunction that has occurred into the diagnostic model, thereby obtaining the cause of the malfunction identified as the diagnostic result.

[0017] In this way, the system 100 can generate an integrated graph that summarizes knowledge about defects across a variety of domains.

[0018] Furthermore, according to the present system 100, it is possible to generate learning data to be used in machine learning of a diagnostic model using an integrated graph generated across domains.

[0019] Furthermore, according to the present system 100, it is possible to acquire, as a diagnostic result, a precisely identified cause of a defect based on a diagnostic model that has been machine-learned from knowledge of cross-domain defects.

[0020] Although there is no particular limitation on the equipment or mobile object in which the electronic system is installed, in this embodiment, an example will be described in which the processor system 100 diagnoses malfunctions in an electronic system installed in an automobile, which is a mobile object.

[0021] <Configuration of Processor System 100> 1 is a diagram showing an example of a schematic configuration of a processor system 100. As shown in the figure, the system 100 is connected to an external device 10 so as to be able to communicate with each other via, for example, a communication cable or a predetermined communication network N (e.g., the Internet, a LAN (Local Area Network), or a WAN (Wide Area Network)).

[0022] <<External device 10>> The external device 10 is a device that transmits input information to the processor system 100. The external device 10 is also a device that acquires information generated by the processor system 100. Specifically, the external device 10 transmits fault-related information (graphic information or non-graphic information) to the processor system 100. In addition to this information, the external device 10 also provides (transmits) to the processor system 100 various types of information used in processing executed by the processor system 100. The external device 10 also acquires from the processor system 100 diagnostic results identified by the processor system 100.

[0023] The external device 10 includes computers used by businesses that receive diagnostic services provided by the system 100, such as product manufacturing and maintenance businesses such as automobile manufacturers and infrastructure management businesses, and computers used by businesses that provide various useful information used in processing executed by the processor system 100. In other words, the external device 10 may be a plurality of types of computers depending on the type of information provided to the processor system 100.

[0024] <<Details of the Processor System 100>> The processor system 100 executes an integrated knowledge generation process for generating an integrated graph, a learning process for performing machine learning on a diagnostic model, and a diagnostic process for diagnosing a malfunction by having the processor 30 read various programs and information stored in the memory resource 40. Details of each process will be described later.

[0025] The processor system 100 is a computer such as a personal computer, a tablet terminal, a smartphone, a server computer, or a cloud server, and is a system including at least one of these computers.

[0026] Specifically, the processor system 100 includes a processor 30, a memory resource 40, an NI (Network Interface Device) 50, and a UI (User Interface Device) 60.

[0027] The processor 30 is an arithmetic device that reads various programs stored in the memory resource 40 and executes processing corresponding to each program. Examples of the processor 30 include a microprocessor, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), or other semiconductor devices capable of performing calculations.

[0028] The memory resource 40 is a storage device that stores various types of information. Specifically, the memory resource 40 is a non-volatile or volatile storage medium such as a random access memory (RAM) or a read-only memory (ROM). Note that the memory resource 40 may also be a rewritable storage medium such as a flash memory, a hard disk, or a solid state drive (SSD), or a universal serial bus (USB) memory, a memory card, or a hard disk.

[0029] The NI 50 is a communication device that communicates information with the external device 10. The NI 50 communicates information with the external device 10 via a predetermined communication network N, such as a LAN or the Internet. Unless otherwise specified below, it is assumed that information communication between the processor system 100 and the external device 10 is performed via the NI 50.

[0030] The UI 60 is an input device that inputs instructions from a user (operator) to the processor system 100, and an output device that outputs information generated by the processor system 100. Examples of input devices include a keyboard, a touch panel, a pointing device such as a mouse, and an audio input device such as a microphone.

[0031] The output device may be, for example, a display, a printer, a voice synthesizer, etc. Unless otherwise specified below, it is assumed that user operations on the processor system 100 (for example, inputting and outputting information, issuing instructions to execute processing, etc.) are performed via the UI 60.

[0032] Furthermore, the configurations, functions, processing means, etc. of the present system 100 may be partially or entirely implemented in hardware, for example, by designing them as integrated circuits. Furthermore, the present system 100 may also implement partially or entirely each function in software, or through a combination of software and hardware. Furthermore, the present system 100 may use hardware having fixed circuits, or may use hardware having at least some of its circuits changeable.

[0033] Furthermore, the system 100 can also be realized by a user (operator) performing some or all of the functions and processes realized by each program.

[0034] The DB (database) and various information in the memory resource 40 described below may be a data structure other than a file or database, as long as it is an area capable of storing data.

[0035] <<Fault-related graph information DB110>> The fault-related graph information DB110 is a database that stores fault-related graph information. The fault-related graph information is a knowledge graph that is useful for diagnosing faults, and is, for example, information resulting from manual analysis such as FMEA (Failure Mode and Effect Analysis) and FTA (Fault Tree Analysis) performed on a target electronic system (i.e., FMEA information and FTA information). Specifically, the fault-related graph information is composed of nodes corresponding to knowledge items related to faults and links that represent the relationships between them.

[0036] Figure 2 shows an example of a knowledge graph. In the illustrated knowledge graph, the boxes for processor anomalies, communication anomalies, hang-ups, and interrupt handling errors are knowledge items (nodes) related to malfunctions, and each related knowledge item is connected by a link (path). The start point of the link arrow corresponds to the malfunction cause, and the end point of the arrow corresponds to the malfunction result, and the link represents the relationship between what factors can cause a certain malfunction.

[0037] The relationships between nodes are not limited to the arrows shown in Figure 2, and for example, values ​​indicating the probability or impact of a certain defect being the result of that defect may be associated as elements indicating the relationships between nodes.Furthermore, the knowledge graph is not limited to the notation shown in the figure, and may be expressed, for example, by an array of knowledge items and an expression indicating the presence or absence of a relationship between each array element.

[0038] Such defect-related graph information is an existing knowledge graph and is acquired from the external device 10.

[0039] <<Fault-related non-graphic information DB120>> The fault-related non-graph information DB120 is a database that stores fault-related non-graph information. Note that the fault-related non-graph information is information that is useful for diagnosing faults and is not expressed by nodes and links like the fault-related graph information. Examples of the fault-related non-graph information include device internal data, probe data, user information, environmental information, infrastructure information, and product information.

[0040] The internal device data is, for example, information such as an error log output from an ECU (Electronic Control Unit). The probe data is, for example, location information while driving and video information captured by a drive recorder. The user information is, for example, medical interview information that records the user's complaints (comments) regarding malfunctions. The environmental information is, for example, information indicating the weather and road conditions while driving. The infrastructure information is, for example, information such as a server log when using a connected service. The product information is, for example, information such as product design documents and service manuals.

[0041] The processor system 100 acquires defect-related graph information and defect-related non-graph information from businesses (e.g., automobile manufacturers, infrastructure management businesses, businesses that provide environmental information) corresponding to each type of information or computers owned by automobile users (i.e., external devices 10).

[0042] <<Integrated graph storage DB130>> The integrated graph storage DB 130 is a database that stores an integrated graph in which knowledge graphs from multiple domains are integrated by the processing executed by the processor system 100.

[0043] <<Learning data storage DB140>> The learning data storage DB 140 is a database that stores learning data. The learning data is data for learning used when machine learning a diagnostic model.

[0044] <<Diagnostic Model 150>> The diagnostic model 150 is an information model for diagnosing malfunctions in an electronic system. Specifically, when information indicating the state of a malfunction is input, the diagnostic model 150 outputs a diagnosis result of the malfunction cause. The diagnostic model 150 is generated by performing machine learning using training data on a mathematical model such as a neural network or a support vector machine.

[0045] <<Diagnostic Program 210>> The diagnostic program 210 is a program for diagnosing the cause of a malfunction. The diagnostic program 210 also includes a plurality of individual programs for executing the processing steps for diagnosing the cause of a malfunction. Specifically, the diagnostic program 210 includes an integrated knowledge generation program 211, a diagnostic learning program 212, and a diagnostic execution program 213.

[0046] <<<Integrated Knowledge Generation Program 211>>> The integrated knowledge generation program 211 is a program that executes an integrated knowledge generation process for generating an integrated graph. Specifically, the integrated knowledge generation program 211 executes a process of integrating knowledge graphs of each area, which are created based on domain knowledge specialized in a specialized field, based on the semantic similarity of the knowledge items included in each knowledge graph.

[0047] <<<Diagnostic Learning Program 212>>> The diagnostic learning program 212 is a program that performs machine learning using learning data on a predetermined mathematical model when generating the diagnostic model 150. Specifically, the diagnostic learning program 212 acquires learning data from the learning data storage DB 140 and performs machine learning using the learning data on a mathematical model such as a neural network, thereby generating the diagnostic model 150 that enables the diagnosis of defects based on cross-domain knowledge.

[0048] <<<Diagnosis execution program 213>>> The diagnostic execution program 213 is a program that diagnoses malfunctions. Specifically, the diagnostic execution program 213 acquires information indicating the state of the malfunction that has occurred and inputs this information into the diagnostic model 150 to acquire a diagnosis result of the estimated cause of the malfunction.

[0049] The details of the processor system 100 have been described above.

[0050] <Functional Configuration of Processor System 100> 3 is a diagram showing an example of the functional configuration of the processor system 100. The functional units shown in the figure are categorized according to the main processing content in order to facilitate understanding of the functions realized by the processor 30 reading each program stored in the memory resource 40. Therefore, the present invention is not limited by the way the functions are categorized or the names of the functional units.

[0051] The integrated knowledge generation processing unit 300 is a functional unit classified for easy understanding of the functions realized by the processor 30 reading the integrated knowledge generation program 211. The integrated knowledge generation processing unit 300 is further classified into a graph acquisition unit, a graph estimation unit, a similarity calculation unit, an integrated graph generation unit, an integrated graph correction unit, and an integrated graph output unit, depending on the processing content.

[0052] The diagnostic learning unit 310 is a functional unit classified to facilitate understanding of the functions realized by the processor 30 reading the diagnostic learning program 212. The diagnostic learning unit 310 is further classified into a data format definition unit, a learning data generation unit, and a diagnostic model generation unit according to the processing content.

[0053] The diagnostic execution unit 320 is a functional unit classified for easy understanding of the functions realized by the processor 30 reading the diagnostic execution program 213. The diagnostic execution unit 320 is further classified into a data input unit, a diagnostic processing unit, and a diagnostic result output unit according to the processing content.

[0054] Note that some of the functional units may be constructed using hardware (such as an integrated circuit such as an ASIC) implemented in a computer. Furthermore, the processing of each functional unit may be executed by a single piece of hardware, or may be executed by multiple pieces of hardware.

[0055] <Processing Description> 4 is a flow diagram showing an example of the integrated knowledge generation process, learning process, and diagnostic process. The following describes each process in detail. The subject of each process is the processor 30 that has read a program stored in the memory resource 40, but in the explanation of the processes, the subject of each function unit realized by the program will be described.

[0056] <<Integrated Knowledge Generation Processing>> The integrated knowledge generation process is started when the processor system 100 receives an execution instruction from a user (operator) and the processor 30 reads the integrated knowledge generation program 211.

[0057] When the process starts, the graph acquisition unit acquires defect-related graph information that has been acquired from the external device 10 and stored in the defect-related graph information DB 110 (step S001). Specifically, the graph acquisition unit acquires defect-related graph information that is expressed in a knowledge graph, such as FMEA information or FTA information, from the defect-related graph information DB 110.

[0058] The graph acquisition unit outputs a group of graphs made up of the acquired plurality of pieces of defect-related graph information to the similarity calculation unit and the integrated graph generation unit.

[0059] Next, the graph estimation unit estimates a knowledge graph (step S002). Specifically, the graph estimation unit acquires the defect-related non-graph information stored in the defect-related non-graph information DB 120 after being acquired from the external device 10. More specifically, the graph estimation unit acquires the defect-related non-graph information that is not expressed as a knowledge graph, such as device internal data, probe data, user information, environmental information, infrastructure information, and product information, from the defect-related non-graph information DB 120.

[0060] Furthermore, the graph estimation unit estimates a knowledge graph from the acquired defect-related non-graph information. Specifically, the graph estimation unit estimates a knowledge graph corresponding to the acquired defect-related non-graph information by processing the defect-related non-graph information using a predetermined mathematical model such as a neural network. That is, the graph estimation unit generates a knowledge graph estimated from the acquired defect-related non-graph information using a mathematical model. Note that the following methods for estimating a knowledge graph using a mathematical model are possible, but are not limited to these: *Based on machine learning, the knowledge graph is estimated using a mathematical model created by machine learning from defect-related non-graph information. *When the defect-related non-graph information is text information, a mathematical model is applied to extract words related to the defect, and a knowledge graph with the extracted words as knowledge items is generated based on predetermined rules.

[0061] The graph estimation unit outputs a group of graphs consisting of the estimated knowledge graphs to the similarity calculation unit and the integrated graph generation unit. The graph estimation unit also stores the knowledge graphs estimated (generated) from the defect-related non-graph information in the defect-related graph information DB 110.

[0062] Next, the similarity calculation unit calculates the similarity of the knowledge items (step S003). Specifically, the similarity calculation unit identifies knowledge graphs included in the group of graphs acquired from the graph acquisition unit and the graph estimation unit. The similarity calculation unit also calculates the similarity between knowledge items in knowledge graphs generated (estimated) based on mutually different domain knowledge.

[0063] Specifically, the system 100 stores in the memory resource 40 a dictionary of equivalent words in which semantic equivalents for the notation of knowledge items (e.g., names of defect causes) are registered, and the similarity calculation unit uses the dictionary to calculate the similarity between knowledge items that are semantically equivalent, rather than being notationally equivalent.

[0064] The method of calculating the similarity is not limited to this, and the similarity calculation unit may, for example, use an information model that performs natural language processing (e.g., word2vec) to learn the semantic similarity in the notation of knowledge items through machine learning, and then calculate the similarity using the information model.

[0065] The similarity calculation unit generates similarity information in which the calculated similarity is registered. Specifically, the similarity calculation unit assigns predetermined identification information (e.g., graph ID and node ID) to the knowledge graph and knowledge item, and generates similarity information in which the identification information consisting of the graph ID of the knowledge graph and the node ID of the knowledge item is associated with the calculated similarity and registered.

[0066] 5 is a diagram showing an example of similarity information. As shown in the figure, the similarity calculation unit generates identification information in which a graph ID, which is an identification ID of the knowledge graph, and a node ID, which is an identification ID of the knowledge item, are assigned to each knowledge graph and knowledge item.

[0067] For example, the identification information of a first knowledge item included in a certain first knowledge graph is "1-1." Also, for example, the identification information of a second knowledge item different from the first knowledge item included in the first knowledge graph is "1-2." Also, for example, the identification information of a first knowledge item included in a second knowledge graph different from the first knowledge graph is "2-1."

[0068] Furthermore, the similarity calculation unit associates the identification information of the knowledge items with the calculated similarities between the knowledge items and registers them in the similarity information.

[0069] Returning to Fig. 4, the explanation will be given. Next, the integrated graph generation unit generates an integrated graph (step S004). Specifically, the integrated graph generation unit identifies pairs of knowledge items whose similarity registered in the similarity information satisfies a predetermined condition (for example, the similarity is equal to or greater than a predetermined threshold). Furthermore, the integrated graph generation unit integrates the knowledge items of the identified pairs to generate an integrated graph that integrates the knowledge graphs including the knowledge items of the pairs.

[0070] Fig. 6 shows an example of generating an integrated graph. The illustrated example shows that the similarity between a knowledge item (connected communication anomaly) included in a knowledge graph (e.g., first knowledge graph) related to a malfunction in vehicle settings from a smartphone app and a knowledge item (communication anomaly) included in a knowledge graph (e.g., second knowledge graph) related to a malfunction in the connected system linking the smartphone app and the vehicle is 0.95 (note that the similarity between semantically identical knowledge items is 1.0), which is above a predetermined threshold.

[0071] In this case, the integrated graph generation unit integrates the knowledge items of the first knowledge graph with the knowledge items of the second knowledge graph. Specifically, the integrated graph generation unit generates an integrated graph in which the second knowledge graph is integrated into the first knowledge graph by integrating the knowledge items so that the nodes under the knowledge item (communication abnormality) of the second knowledge graph are associated with the nodes under the knowledge item (connected communication abnormality) of the first knowledge graph.

[0072] In addition, in the example shown, the similarity between the knowledge item (vehicle internal communication abnormality) included in the first knowledge graph and the knowledge item (communication abnormality) included in a knowledge graph related to a malfunction inside the vehicle (for example, the third knowledge graph) is 0.98, which indicates that it is above a predetermined threshold.

[0073] In this case, the integrated graph generation unit integrates the knowledge items of the first knowledge graph with the knowledge items of the third knowledge graph. Specifically, the integrated graph generation unit generates an integrated graph in which the third knowledge graph is integrated into the first knowledge graph by integrating the knowledge items so that the node under the knowledge item (communication abnormality) of the third knowledge graph is associated with the node under the knowledge item (vehicle internal communication abnormality) of the first knowledge graph.

[0074] Furthermore, the integration of knowledge items is not necessarily limited to the above example (one-to-one integration between a specified pair of knowledge items). For example, for one knowledge item in a knowledge graph, multiple knowledge graphs containing multiple knowledge items that are equal to or greater than a predetermined threshold may be integrated (one-to-many relationship).

[0075] Furthermore, not all knowledge graphs are necessarily aggregated (integrated) into a single integrated graph, and it is acceptable for multiple integrated graphs to coexist.

[0076] Returning to FIG. 4, the explanation will be given. Next, the integrated graph correction unit corrects the integrated graph (step S005). Specifically, the integrated graph correction unit displays the generated integrated graph on a display provided in the present system 100, and receives correction instructions from, for example, an operator who is a holder of defect knowledge. Note that the correction instructions include, for example, instructions to change the relationships between nodes included in the integrated graph (such as deleting or moving nodes). The integrated graph correction unit also corrects the integrated graph based on the received correction instructions.

[0077] Next, the integrated graph output unit outputs the corrected integrated graph (step S006). Specifically, the integrated graph output unit stores the corrected integrated graph in the integrated graph storage DB 130 and also outputs it to the data format definition unit.

[0078] An example of the integrated knowledge generation process has been described above.

[0079] According to the present system 100, which executes such an integrated knowledge generation process, it is possible to obtain more comprehensive explicit knowledge (integrated graph) by absorbing variations in notation between knowledge items based on the semantic similarity of the knowledge items and integrating knowledge across multiple mutually different domains.

[0080] Typically, knowledge items in knowledge graphs created based on different domain knowledge are difficult to integrate due to variations in notation. In other words, even if knowledge items in different knowledge graphs refer to the same object semantically, it is often difficult to identify them as nodes referring to the same object because the notation of the knowledge items is different in knowledge graphs created as different domain knowledge. In contrast, the present system 100 calculates semantic similarity using the above method, and if the calculated similarity satisfies a predetermined condition (above a predetermined threshold), it integrates these knowledge items. This makes it possible to generate an integrated graph that is more comprehensive than a single knowledge graph.

[0081] Next, we will explain the learning process executed by the system 100. This process starts when the data format definition unit acquires the integrated graph.

[0082] When the process starts, the data format definition unit defines the data format of the learning data (step S010). Specifically, the data format definition unit determines whether or not there is data corresponding to each knowledge item in the integrated graph. Here, for example, if the knowledge item is a server processing abnormality, the data corresponding to the knowledge item corresponds to a server log that records the operation history of the server.

[0083] The presence or absence of data corresponding to a knowledge item may be predetermined for each knowledge item, for example. Also, for example, if the data is associated with a knowledge item and stored in memory resource 40, the presence or absence of the data may be identified by searching memory resource 40. Alternatively, if information indicating the correspondence between a knowledge item and the presence or absence of corresponding data is stored in memory resource 40, the presence or absence of the data may be identified using the information.

[0084] Furthermore, if corresponding data exists, the data format definition unit assigns a data ID to the corresponding knowledge item to identify the data. An example of a data ID is the unique name of the data, or, if the data is stored in a database in the memory resource 40, the column name or field name. If the notation of the knowledge item includes words such as "**error" or "**log," the data format definition unit automatically extracts these words using regular expressions or the like and assigns them as a data ID. The data ID may also be entered manually, for example.

[0085] 7 is an explanatory diagram for defining an input data format using an integrated graph. As shown in the figure, knowledge items in the integrated graph are linked to the data corresponding to the knowledge items. The data format definition unit assigns data IDs (e.g., application error records, operation logs, and communication logs) to the knowledge items corresponding to these data using the above method.

[0086] For example, a wiring abnormality (such as a broken wire) inside a vehicle cannot be directly recognized as data, and there may be no corresponding data for such a knowledge item. In such cases, the data format definition unit associates information indicating the absence of corresponding data with the knowledge item, or assigns a data ID indicating other related data to the corresponding knowledge item. Note that the other related data (for example, data on a communication error in the case of the wiring abnormality) may be predetermined according to the knowledge item.

[0087] The data format definition unit also identifies knowledge items corresponding to the cause of the malfunction to be diagnosed from the integrated graph. For example, the data format definition unit considers the knowledge items at the ends of the integrated graph as fundamental malfunction causes and identifies these knowledge items as knowledge items corresponding to the malfunction causes.

[0088] The method for identifying knowledge items corresponding to the failure factors of the diagnosis target from the integrated graph is not limited to this. For example, if there is a list that defines the correspondence between the failure factors of the diagnosis target and the corresponding knowledge items, the data format definition unit may use this list to identify knowledge items corresponding to the failure factors from the integrated graph. Note that the list may be stored in advance in the memory resource 40. Alternatively, the knowledge items may be identified manually, for example.

[0089] Note that malfunction factors that are not subject to diagnosis are excluded from the process. For example, if the analysis target is a malfunction that is known to be free of problems inside the vehicle, "wiring fault" and "signal abnormality" from the knowledge items shown in the figure are excluded from the malfunction factors.

[0090] The data format definition unit also extracts data IDs related to the identified defect factors and defines the list as an input data format. Specifically, the data format definition unit identifies the data IDs of the corresponding data assigned to the knowledge items of the identified defect factors and defines a list of the data IDs as an input data format.

[0091] In addition, when the integrated graph has a tree structure (the illustrated example is a tree structure), the data format definition unit may select a path (link) from the top event (root node) to the terminal event (leaf node) that includes the knowledge item of the determined defect cause, and define a list of comprehensively extracted data IDs of the corresponding data assigned to the knowledge items on the selected path as the input data format. Alternatively, the data IDs may be extracted manually, for example.

[0092] Through the above processing, the input data format shown in FIG. 7 is defined.

[0093] Returning to FIG. 4, the explanation will be given. Next, the learning data generation unit generates learning data (step S011). Specifically, the learning data generation unit acquires past malfunction cases, which are cases in which the knowledge item identified in the processing of step S011 has been identified as the malfunction cause and which have data matching the items defined in the input data format (in the example of FIG. 7, application error record, operation log, communication log, and server log), from malfunction case information in which a plurality of past malfunction cases are registered. Note that the malfunction case information may be acquired from, for example, the external device 10, or may be stored in advance in the memory resource 40 of the present system 100.

[0094] Furthermore, the learning data generation unit identifies, from the acquired failure case, data corresponding to each item in the input data format, i.e., state information (such as log information) corresponding to each item at the time of failure occurrence, and the failure cause identified in the case (i.e., the same failure cause as the failure cause identified in step S011). Furthermore, the learning data generation unit generates learning data by allocating the identified state information to the corresponding item defined in the input data format and associating the failure cause identified in the case with a record including each of these items.

[0095] The above process generates the training data shown in Figure 7. As shown in the figure, the training data is assigned status information at the time of the occurrence of a defect corresponding to each item in the input data format and the cause of the defect identified in the case. In other words, the training data plays the role of labeled data in supervised learning.

[0096] Returning to FIG. 4, the explanation will be given. Next, the diagnostic model generation unit generates a diagnostic model 150 that diagnoses a malfunction using the training data (step S012). Specifically, when information corresponding to each item of the training data is input as information indicating a malfunction, the diagnostic model generation unit performs machine learning using the training data on a mathematical model so as to estimate the malfunction cause associated with the item. In this way, the diagnostic model generation unit generates a diagnostic model that diagnoses the malfunction and outputs the cause as a diagnostic result.

[0097] An example of the learning process has been described above.

[0098] According to the present system 100 that executes such learning processing, by using an integrated graph that includes cross-domain knowledge items, it is possible to define an input data format that is more highly correlated with more defects. Furthermore, by generating learning data to be used for machine learning of the diagnostic model 150 based on past cases that correspond to the items defined in such an input data format, it is possible to generate a diagnostic model 150 that is comprehensive and accurate and can analyze the causes of more defects.

[0099] Next, a description will be given of the diagnostic processing executed by the present system 100. Note that this processing starts when an execution instruction is received from an operator who is, for example, a defect analyst.

[0100] When the process starts, the data input unit receives input of information indicating the state of the defect (hereinafter referred to as "defect information.") Specifically, the data input unit receives input of the defect information from the operator (step S020).

[0101] Next, the diagnostic processing unit executes a diagnosis of the malfunction using the diagnostic model 150 (step S021). Specifically, the diagnostic processing unit inputs the malfunction information, which is received from the operator by the data input unit and includes information corresponding to items defined in the input data format, to the diagnostic model 150, and acquires, as an output value, a diagnostic result in which the cause of the malfunction is identified.

[0102] Next, the diagnostic result output unit outputs the diagnostic result in which the cause of the malfunction has been identified (step S022). Specifically, the diagnostic result output unit outputs the diagnostic result including the identified cause of the malfunction to an output device such as a display provided in the system 100. The diagnostic result output unit may output (transmit) the diagnostic result to a predetermined external device 10.

[0103] An example of the diagnostic process has been described above.

[0104] According to the present system 100 that performs such diagnostic processing, it is possible to obtain comprehensive and accurate diagnostic results that can analyze the causes of a greater number of defects.

[0105] In this system 100, the functional units of the integrated knowledge generation processing unit 300, the diagnostic learning unit 310, and the diagnostic execution unit 320 do not necessarily have to be implemented in a single processor system 100 (computer). For example, each functional unit may be implemented by processors of computers owned by different businesses loading corresponding programs. Alternatively, the integrated knowledge generation processing unit 300 and the diagnostic learning unit 310 may be implemented in a single computer, or the integrated knowledge generation processing unit 300 and the diagnostic execution unit 320 may be implemented in a single computer.

[0106] Furthermore, the present invention is not limited to the above-described embodiment, and various modifications are possible. The processor system 100 according to the modified example updates an already-generated integrated graph each time new fault-related information is acquired, thereby generating an integrated graph with improved comprehensiveness. Specifically, when new fault-related information is acquired, the integrated graph generation unit acquires an already-generated integrated graph from the integrated graph storage DB 130. Furthermore, the integrated graph generation unit integrates a knowledge graph based on the newly acquired fault-related information (which may be either graph information or non-graph information) into an existing integrated graph according to the similarity of knowledge items, thereby updating the existing integrated graph and generating an integrated graph with higher comprehensiveness.

[0107] Furthermore, the diagnostic learning unit 310 generates learning data using the updated integrated graph, and updates the diagnostic model 150 using the learning data.

[0108] Through such processing, the integrated knowledge generation processing unit 300 can update the existing integrated graph to a more comprehensive integrated graph. Furthermore, the diagnostic learning unit 310 can update the diagnostic model 150 based on the updated integrated graph. As a result, the processor system 100 can further improve the accuracy of diagnosis using the diagnostic model 150.

[0109] <Service System to which Processor System 100 is Applied> 8 is a diagram showing an example of an outline of a service system to which the processor system 100 according to the above-described embodiment is applied. The illustrated service form shows an example in which the present invention is applied to a service for automobiles.

[0110] The illustrated product manufacturing and maintenance businesses and infrastructure management businesses provide existing defect-related information (including information collected from product operation businesses and product users) to the operation business of this system 100.

[0111] In addition, the operator of this system 100 uses the processor system 100 to create explicit knowledge (integrated graph) related to fault diagnosis, and then performs machine learning on the learning data generated using the integrated graph to generate a more comprehensive and accurate diagnostic model.

[0112] In addition, product manufacturing and maintenance businesses and infrastructure management businesses provide information on defects with unspecified causes to the operating business of this system 100.

[0113] In addition, the operator of this system 100 inputs the provided defect information into a diagnostic model, obtains diagnostic results that identify the cause of the defect, and responds to the product manufacturing and maintenance operator and the infrastructure management operator.

[0114] This type of service system configuration eliminates the need for manual analysis for product manufacturing and maintenance companies and infrastructure operators, and provides the benefit of faster analysis.

[0115] As a result, product purchasers and product operators can enjoy various benefits such as improved safety performance and improved availability, while manufacturers can also enjoy the benefit of increased product credibility.

[0116] In addition, product manufacturing and maintenance businesses can benefit from increased user purchasing motivation due to improved manufacturer credibility.

[0117] Furthermore, the operator of this system 100 can enjoy the benefit of receiving payment for fault diagnosis (fees shown in the figure) from product manufacturing and maintenance companies and infrastructure companies.

[0118] The operator of this system 100 may collect and aggregate defect-related information from multiple product manufacturing and maintenance companies or parts suppliers, thereby enabling the processor system 100 to create a more comprehensive and accurate integrated graph.

[0119] Furthermore, the present invention is not limited to the above-described embodiments and modifications, and includes various modifications within the scope of the same technical concept. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those including all of the described configurations. Furthermore, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

[0120] In addition, in the above explanation, the control lines and information lines are those that are considered necessary for the explanation, and do not necessarily show all the control lines and information lines in the product. In reality, it can be considered that almost all components are interconnected. [Explanation of symbols]

[0121] 100...Processor system, 30...Processor, 40...Memory resource, 50...NI (Network Interface Device), 60...UI (User Interface Device), 110...Fault-related graph information DB, 120...Fault-related non-graph information DB, 130...Integrated graph storage DB, 140...Learning data storage DB, 150...Diagnostic model, 210...Diagnostic program, 211...Integrated knowledge generation program, 212...Diagnostic learning program, 213...Diagnostic execution program, 300...Integrated knowledge generation processing unit, 310...Diagnostic learning unit, 320...Diagnostic execution unit, 10...External device, N...Network

Claims

1. A processor system having one or more processors and one or more memory resources, The memory resource is a diagnostic program for diagnosing an electronic system including a mobile object or equipment; storing a plurality of knowledge graphs used in the diagnostic program, the plurality of knowledge graphs including at least a first knowledge graph and a second knowledge graph created based on domain knowledge different from the first knowledge graph; The processor executes the diagnostic program to: (1) evaluating the semantic similarity between a first knowledge item included in the first knowledge graph and a second knowledge item included in the second knowledge graph; (2) if the similarity satisfies a predetermined condition, generating an integrated graph by integrating the first knowledge graph and the second knowledge graph; Identifying the knowledge items included in the integrated graph that correspond to the cause of the malfunction to be diagnosed, By itemizing the identification information of the data corresponding to the identified knowledge items, the input data format of the learning data for machine learning of a predetermined mathematical model for diagnosis is defined. A processor system comprising:

2. 2. The processor system of claim 1, The processor executes the diagnostic program to: Identifying data that matches the items defined in the input data format and indicates a state corresponding to each item when a defect occurs from past defect cause cases, and generating learning data by allocating the identified data to the corresponding items. A processor system comprising:

3. 3. The processor system according to claim 2, The processor executes the diagnostic program to: When information corresponding to each item of the learning data is input as information indicating a defect, machine learning is performed using the learning data on a predetermined mathematical model so as to estimate the cause of the defect associated with that item, thereby generating a diagnostic model for diagnosing the defect in the electronic system. A processor system comprising:

4. 4. The processor system according to claim 3, The processor executes the diagnostic program to: (1) Obtaining failure information whose cause is not yet identified from a terminal of a product manufacturing / maintenance business operator or an infrastructure management business operator, and inputting the failure information into the diagnostic model to obtain the identified failure cause as a diagnostic result; (2) The identified cause of the defect is transmitted to a terminal of the product manufacturing / maintenance company or the infrastructure management company. A processor system comprising:

5. A fault diagnosis method performed by a processor system having one or more processors and one or more memory resources, comprising: The memory resource is a diagnostic program for diagnosing an electronic system including a mobile object or equipment; storing a plurality of knowledge graphs used in the diagnostic program, the plurality of knowledge graphs including at least a first knowledge graph and a second knowledge graph created based on domain knowledge different from the first knowledge graph; The processor executes the diagnostic program to: (1) evaluating the semantic similarity between a first knowledge item included in the first knowledge graph and a second knowledge item included in the second knowledge graph; (2) if the similarity satisfies a predetermined condition, generating an integrated graph by integrating the first knowledge graph and the second knowledge graph; Identifying the knowledge items included in the integrated graph that correspond to the cause of the malfunction to be diagnosed, By itemizing the identification information of the data corresponding to the identified knowledge items, the input data format of the learning data for machine learning of a predetermined mathematical model for diagnosis is defined. A fault diagnosis method characterized by:

6. A fault diagnosis method according to claim 5, The processor executes the diagnostic program to: Identifying data that matches the items defined in the input data format and indicates a state corresponding to each item when a defect occurs from past defect cause cases, and generating learning data by allocating the identified data to the corresponding items. A fault diagnosis method characterized by:

7. A fault diagnosis method according to claim 6, The processor executes the diagnostic program to: When information corresponding to each item of the learning data is input as information indicating a defect, machine learning is performed using the learning data on a predetermined mathematical model so as to estimate the cause of the defect associated with that item, thereby generating a diagnostic model for diagnosing the defect in the electronic system. A fault diagnosis method characterized by:

8. A fault diagnosis method according to claim 7, The processor executes the diagnostic program to: (1) Obtaining failure information whose cause is not yet identified from a terminal of a product manufacturing / maintenance business operator or an infrastructure management business operator, and inputting the failure information into the diagnostic model to obtain the identified failure cause as a diagnostic result; (2) The identified cause of the defect is transmitted to a terminal of the product manufacturing / maintenance company or the infrastructure management company. A fault diagnosis method characterized by:

Images