Knowledge graph construction method and vehicle fault diagnosis method

Abstract

The invention provides a method for constructing a knowledge graph for vehicle fault diagnosis. The method comprises the following steps: establishing a database of vehicle parts; establishing a database of abnormal signals; creating a fault knowledge triple from known unstructured fault diagnosis data and semi-structured fault diagnosis data; and associating the abnormal signal, the vehicle component and the maintenance suggestion together based on the fault knowledge triad so as to construct a knowledge graph for vehicle fault diagnosis.

Description

Technical Field

[0001] The present invention relates to automated fault diagnosis of vehicles, and more specifically, to a method for constructing a knowledge graph for vehicle fault diagnosis and a vehicle fault diagnosis method and system based on the knowledge graph. Background Art

[0002] The vehicle fault diagnosis method based on knowledge graph is a method that uses big data and artificial intelligence technology to build a knowledge graph by analyzing maintenance history data, and uses the knowledge graph to automatically identify and classify vehicle faults, provide maintenance suggestions, and greatly improve maintenance efficiency and accuracy.

[0003] The vehicle fault diagnosis method based on knowledge graph first extracts valuable information from massive maintenance data through entity extraction and relationship recognition technology, and then organizes this information into a knowledge graph. On this basis, the system uses complex algorithm models (such as the XGBoost fault classification model) to diagnose vehicle faults, determine possible faulty parts, and provide maintenance guidance for faulty parts. Summary of the invention

[0004] The first aspect of the present disclosure provides a method for constructing a knowledge graph for vehicle fault diagnosis. The method may include: establishing a database of vehicle components, the vehicle components including basic units that can be replaced or repaired; establishing a database of abnormal signals, the abnormal signals including vehicle information related to vehicle faults, the vehicle information related to vehicle faults including the vehicle's operating mode, status, sensor signals, and driving data; creating fault knowledge triples from known unstructured fault diagnosis data and semi-structured fault diagnosis data, the fault knowledge triples including: <abnormal signal-fault name-faulty vehicle component> and <faulty vehicle component-repair method-repair suggestion>; and based on the fault knowledge triples, associating abnormal signals, vehicle components, and repair suggestions together to construct a knowledge graph for vehicle fault diagnosis. In some instances, the probability P that abnormal signal i indicates that vehicle component k is faulty may be i / k As the weight E of the edge between the node corresponding to the abnormal signal i and the node corresponding to the vehicle component k in the knowledge graph i / k .

[0005] In some embodiments, the knowledge graph includes an abnormal signal node, a faulty component node, and a maintenance suggestion node. The abnormal signal node is associated with one or more faulty component nodes that can generate the abnormal signal. The faulty component node is associated with one or more maintenance suggestion nodes that can repair the faulty component.

[0006] In some embodiments, the probability P can be calculated using Equation 1 i / k :

[0007] P i / k =C(i,k) / C(i) (Formula 1)

[0008] Here, C(i,k) is the number of times abnormal signal i and vehicle component k appear simultaneously in the same fault scenario, and C(i) is the number of times abnormal signal i appears in all fault scenarios.

[0009] In some embodiments, the method for constructing a knowledge graph for vehicle fault diagnosis may further include: updating the weight E using new data through a machine learning algorithm. i / k .

[0010] In some embodiments, the method of constructing a knowledge graph for vehicle fault diagnosis may also include: adding the probabilities corresponding to multiple abnormal signals associated with vehicle component k to indicate the probability of a fault in vehicle component k.

[0011] In some embodiments, the method of constructing a knowledge graph for vehicle fault diagnosis may also include: using fault knowledge triples to train a machine learning model so that the machine learning model can predict possible future faults and recommend maintenance measures.

[0012] In some embodiments, the method of constructing a knowledge graph for vehicle fault diagnosis may also include: identifying associations between multiple fault knowledge triples, thereby linking the multiple fault knowledge triples to each other.

[0013] The second aspect of the present disclosure provides a method for diagnosing vehicle faults using a knowledge graph. The method may include: when a vehicle fault monitoring module of a vehicle terminal detects a vehicle fault, a fault snapshot image is generated; the vehicle terminal sends the fault snapshot image to a cloud server; the cloud server parses vehicle information from the fault snapshot image, and performs fault diagnosis in the fault knowledge graph in the cloud server based on the vehicle information to determine a target maintenance suggestion node; and the cloud server sends the information of the target maintenance suggestion node as diagnostic information to the vehicle terminal. The fault knowledge graph may include an abnormal signal node, a fault component node, and a maintenance suggestion node, wherein the abnormal signal node is associated with one or more fault component nodes that can generate the abnormal signal, and the fault component node is associated with one or more maintenance suggestion nodes that can repair the fault component. In some instances, determining the target maintenance suggestion node includes determining the fault component node with the highest fault probability from multiple fault component nodes as the target fault component node, and determining the maintenance suggestion node associated with the target fault component node as the target maintenance suggestion node.

[0014] In some embodiments, the failure probability P of each failed component node is k Calculated by formula 2:

[0015] P k =∑ i P i / k (Formula 2)

[0016] Here, P i / k =C(i,k) / C(i), C(i,k) is the number of times abnormal signal i and vehicle component k appear simultaneously in the same fault scenario, and C(i) is the number of times abnormal signal i appears in all fault scenarios.

[0017] The third aspect of the present disclosure provides a vehicle fault diagnosis system based on a knowledge graph. The system may include: a vehicle terminal; and a cloud server. The vehicle terminal may include: a vehicle fault monitoring module, which is configured to generate a fault snapshot image when a vehicle fault is detected; and a vehicle fault sending module, which is configured to send the fault snapshot image to the cloud server. The cloud server may include: a fault knowledge graph; a fault snapshot receiving module; and a fault diagnosis module. The knowledge graph includes an abnormal signal node, a fault component node, and a maintenance suggestion node. The abnormal signal node is associated with one or more fault component nodes that can generate the abnormal signal. The fault component node is associated with one or more maintenance suggestion nodes that can repair the fault component. The fault snapshot receiving module can be configured to receive a fault snapshot image sent from the vehicle terminal. The fault diagnosis module can be configured to parse the vehicle information from the snapshot image, perform fault diagnosis in the fault knowledge graph in the cloud server based on the vehicle information to determine the target maintenance suggestion node, and send the information of the target maintenance suggestion node as diagnostic information to the vehicle terminal. In some examples, determining the target maintenance suggestion node may include: determining a faulty component node with a maximum fault probability from among multiple faulty component nodes as the target faulty component node, and determining a maintenance suggestion node associated with the target faulty component node as the target maintenance suggestion node.

[0018] In some embodiments, the failure probability P of each failed component node is k It can be calculated by formula 2:

[0019] P k =∑ i P i / k (Formula 2)

[0020] Here, P i / k =C(i,k) / C(i), C(i,k) is the number of times abnormal signal i and vehicle component k appear simultaneously in the same fault scenario, and C(i) is the number of times abnormal signal i appears in all fault scenarios.

[0021] The fourth aspect of the present disclosure provides a system including one or more computer processors and a computer-readable memory, the computer-readable memory including machine executable code, which, when executed by one or more computer processors, implements the vehicle fault diagnosis method based on the knowledge graph of the present disclosure or the method of diagnosing vehicle faults using the knowledge graph of the present disclosure.

[0022] It should be understood that the present disclosure is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The novel features of the invention are set forth with particularity in the appended claims.A better understanding of the features and advantages of the invention will be obtained by referring to the following detailed description and accompanying drawings which set forth illustrative embodiments utilizing the principles of the invention.

[0024] Figure 1 The structure of an exemplary knowledge graph for vehicle fault diagnosis according to an embodiment of the present disclosure is shown.

[0025] Figure 2 The structure of another exemplary knowledge graph for vehicle fault diagnosis according to an embodiment of the present disclosure is shown.

[0026] Figure 3 It is a flowchart of a method for constructing a knowledge graph for vehicle fault diagnosis according to an embodiment of the present disclosure.

[0027] Figure 4 It is a flowchart of a method for diagnosing vehicle faults using a knowledge graph according to an embodiment of the present disclosure.

[0028] Figure 5 It is a structural block diagram of a vehicle fault analysis device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] Traditional vehicle fault diagnosis methods mainly rely on standard fault codes generated by the vehicle computer (ECU) after a fault occurs. By analyzing these fault codes and related log information, technicians can determine the cause of the fault and repair it. However, this diagnostic method that relies on fault codes has limitations in application due to the limited types of faults it covers.

[0030] Current fault diagnosis methods face several challenges: first, these methods rely too much on the personal experience of maintenance personnel, and these experience levels vary; second, analyzing the cause of the fault is often time-consuming and costly; finally, for frequently occurring faults, the experience of maintenance experts has not been effectively integrated and accumulated, resulting in the inability to fully utilize existing maintenance knowledge and solutions.

[0031] In view of the above problems in the prior art, the present disclosure proposes a method for constructing a knowledge graph for vehicle fault diagnosis and a method for diagnosing vehicle faults based on the knowledge graph. The knowledge graph constructed in the present disclosure can provide detailed information about the faulty vehicle parts and their repair methods, so that the vehicle fault diagnosis method based on the knowledge graph can accurately locate the faulty parts, facilitating maintenance personnel to quickly perform repairs.

[0032] like Figure 1 As shown, the present disclosure provides a knowledge graph for vehicle fault diagnosis. The knowledge graph constructs a semantic association network. In some instances, the knowledge graph of the present disclosure is a data structure based on a graph, comprising nodes (Point) and edges (Edge) connecting the nodes. In some instances, the knowledge graph structure of the present disclosure can integrate multiple types of nodes, such as abnormal signal nodes, faulty component nodes, and maintenance suggestion nodes, thereby providing a comprehensive information framework for vehicle fault diagnosis and maintenance. In some instances, the abnormal signal node represents an abnormal signal indicating a problem or fault that occurs in the system (for example, a signal indicating a low battery charge), the faulty component node represents one or more vehicle components (for example, a battery) that may cause the abnormal signal node, and the maintenance suggestion node provides one or more maintenance suggestions for the faulty component (for example, checking the connection or charge of the battery).

[0033] In some instances, an abnormal signal node may be associated with one or more faulty component nodes that may generate the abnormal signal. This association relationship may be represented by an edge between the abnormal signal node and the faulty component node. A faulty component node may be associated with one or more maintenance suggestion nodes that provide recommended maintenance measures for the faulty component. This association relationship may be represented by an edge between a faulty component node and a maintenance suggestion node. In some instances, the value of the edge between the faulty component node and the maintenance suggestion node may be 1. In some instances, the value of the edge between the faulty component node and the maintenance suggestion node may be an association probability value obtained based on statistics.

[0034] In a non-limiting example, for the fault of "unable to start the vehicle", the knowledge graph of the present disclosure may include (1) abnormal signal nodes, including "low battery abnormal signal node", "fuel system abnormal signal node" and "ignition system abnormal signal node"; (2) fault component nodes, including "battery node", "fuel pump node", "injector node", "fuel filter node", "spark plug node" and "ignition coil node"; and (3) maintenance suggestion nodes, including "battery repair node", "fuel pump repair node", "injector repair node", "fuel filter repair node", "spark plug repair node" and "ignition coil repair node". These maintenance suggestion nodes respectively include inspection / repair suggestions for the battery, fuel pump, injector, fuel filter, spark plug and ignition coil, such as text descriptions, audio and / or video tutorials. An abnormal signal node can be associated with one or more fault component nodes. For example, "low battery abnormal signal node" can be associated with "battery node". For example, a "fuel system abnormal signal node" can be associated with three fault component nodes, namely, a "fuel pump node", a "fuel injector node" and a "fuel filter node".

[0035] In another non-limiting example, for the fault of "engine temperature is too high", the knowledge graph of the present disclosure may include: (1) abnormal signal nodes related to the fault of "engine temperature is too high", including "temperature sensor abnormal signal node" and "cooling system abnormal signal node"; (2) fault component nodes associated with "temperature sensor abnormal signal node" and "cooling system abnormal signal node", including "temperature sensor node", "water pump node", "radiator node" and "coolant tank node"; and (3) maintenance suggestion nodes associated with the fault component nodes, including "temperature sensor repair node", "water pump repair node", "radiator repair node" and "coolant tank repair node". These maintenance suggestion nodes respectively include inspection / repair suggestions for the temperature sensor, water pump, radiator and coolant tank, such as text descriptions, audio and / or video tutorials.

[0036] In a car, for a fault, there may be multiple abnormal signal nodes, which may appear at the same time or interact with each other. In some examples, in the knowledge graph of the present disclosure, these multiple abnormal signal nodes associated with the same vehicle fault can be associated with each other. This association relationship can be represented by the edges between multiple abnormal signal nodes. In a non-limiting example, for the fault of "brake system abnormality", there may be multiple abnormal signal nodes, such as "brake pad wear abnormal signal node", "brake fluid volume abnormal signal node", "brake fluid pipeline blockage abnormal signal node" and "brake fluid cleanliness abnormal signal node", and these abnormal signal nodes are usually associated with each other, affect each other or appear at the same time. For example, brake pad wear will reduce the braking force, and brake fluid contamination may further affect the braking effect; brake fluid leakage will cause the brake fluid to be unable to be normally transmitted to the brake device, and brake pipeline blockage will prevent the brake fluid from flowing. In another non-limiting example, for the fault of "unable to start the vehicle", there may be multiple abnormal signal nodes that are associated with each other, affect each other or appear at the same time, such as "battery power shortage abnormal signal node", "fuel system abnormal signal node" and "ignition system abnormal signal node".

[0037] The lines (edges) between abnormal signal nodes can be assigned values. In some instances, the lines (edges) between abnormal signal nodes can be assigned values ​​based on the probability of two abnormal signals occurring simultaneously. For example, based on prior statistical information, the probability of the fuel system abnormality and the ignition system abnormality occurring simultaneously is 0.4, so the line (edge) between the "fuel system abnormal signal node" and the "ignition system abnormal signal node" can be assigned a value of 0.4.

[0038] Figure 2 FIG. 2 shows the structure of another exemplary knowledge graph for vehicle fault diagnosis according to an embodiment of the present disclosure. For the fault of "battery power failure", Figure 2 The knowledge graph shown includes: abnormal signal nodes "low battery", "low battery voltage" and "abnormal load level"; the fault component node "battery" associated with these three abnormal signal nodes; and a maintenance suggestion node that provides one or more maintenance suggestions for the fault component "battery" (for example, check the connection or power of the battery). In this exemplary embodiment, the association relationship between the three abnormal signal nodes and the fault component node "battery" can be represented by lines E11, E21 and E31 respectively. The association relationship between the fault component node "battery" and the associated maintenance suggestion node can also be represented by a line.

[0039] In some embodiments, the edges between the abnormal signal node and the associated faulty component node and the edges between the faulty component node and the associated maintenance suggestion node in the knowledge graph of the present disclosure can be assigned values. This assignment can represent the strength, importance or credibility of the relationship between the nodes. Assigning values ​​to the edges between the nodes in the knowledge graph can further enhance the vehicle fault diagnosis system's ability to handle complex information. For example, through assignment, the vehicle fault diagnosis system can more accurately determine the degree of association between the abnormal signal node and the faulty component node, as well as the correlation between the faulty component node and the maintenance suggestion node. This assignment can be based on historical data, expert knowledge or real-time feedback to ensure its accuracy and reliability.

[0040] In some embodiments, the edge between the faulty component node and the associated maintenance suggestion node may be assigned a value of 1. In some embodiments, multiple edges between the faulty component node and multiple associated maintenance suggestion nodes may be assigned values ​​based on previously statistically obtained probability values. In some embodiments, multiple edges between the faulty component node and multiple associated maintenance suggestion nodes may be assigned values ​​based on the quality of the maintenance suggestion.

[0041] In some embodiments, the probability P of the abnormal signal i indicating that the vehicle component k is faulty may be i / k As the weight E of the edge between the abnormal signal node corresponding to the abnormal signal i and the faulty component node corresponding to the vehicle component k in the knowledge graph i / k In one example, if it is known based on historical data statistics that among all battery failures, the probability / proportion of abnormal battery power signals is 50%, the probability / proportion of abnormal battery voltage signals is 10%, and the probability / proportion of abnormal vehicle load levels is 26%, then edge E11 can be assigned a value of 0.5, edge E21 can be assigned a value of 0.1, and edge E31 can be assigned a value of 0.26.

[0042] In the assigned knowledge graph of the present disclosure, the probability of association between abnormal signals and specific vehicle component failures is represented by edge weights. This allows the vehicle fault diagnosis system to quickly evaluate the possibility of related component failures when a specific abnormal signal is detected. For example, if the battery charge signal is abnormal, the vehicle fault diagnosis system will determine that there is a 50% probability of a battery failure based on the weight of edge E11 of 0.5. The weight allocation method of the present disclosure not only improves the accuracy of fault diagnosis, but also enhances the real-time response capability of the vehicle fault diagnosis system to the vehicle status. In addition, the knowledge graph of the present disclosure can be optimized through a machine learning algorithm, and the weights of the edges are continuously adjusted according to new data to adapt to the operation of the vehicle in different environments, thereby improving the accuracy of fault diagnosis and the reliability of the vehicle fault diagnosis system.

[0043] As mentioned above, for a component failure, there may be multiple abnormal signal nodes, and these abnormal signal nodes may appear at the same time or interact with each other. Therefore, if multiple abnormal signal nodes appear at the same time, the probability of the same faulty component node associated with these abnormal signal nodes failing is greater. The probability that multiple abnormal signals appearing at the same time indicate that the same vehicle component has failed can be summed to indicate the probability or possibility of the vehicle component failing. For example, the battery system of a car may contain multiple monitoring signals, such as power, voltage, temperature, etc. If a monitoring signal is abnormal, the vehicle fault diagnosis system may only regard it as a minor fault or normal fluctuation. However, if the two monitoring points of power and voltage are abnormal at the same time, the vehicle fault diagnosis system will think that the possibility of battery failure is greatly increased. At this time, the vehicle fault diagnosis system can calculate the comprehensive abnormal index of the two signals, for example, add the abnormal probabilities of the two signals, so as to obtain a more accurate fault prediction. In the above example, if the abnormal battery power signal and the abnormal battery voltage signal appear at the same time, the probability of the battery failing is 50%+10%=60%. The edge weight-based analysis method disclosed in the present invention utilizes the correlation between signal nodes and improves the accuracy and reliability of fault detection by comprehensively considering the abnormal conditions of multiple signal nodes.

[0044] The disclosed method of assigning values ​​to the edges between nodes in the knowledge graph and performing fault diagnosis based on the values ​​can also be extended to other systems of the vehicle, such as the power system, the braking system, the steering system, etc. By constructing a signal node network for each system and analyzing the interactions between the nodes, comprehensive monitoring of the entire vehicle status can be achieved. When multiple signal nodes of multiple systems are abnormal at the same time, the fault diagnosis system can quickly locate the possible source of the fault and predict the trend of the fault development.

[0045] Figure 3 It is a flow chart of a method for constructing a knowledge graph for vehicle fault diagnosis according to an embodiment of the present disclosure. In some embodiments, the method may include steps S301-S304. In step S301, a database of vehicle parts is established. As defined in the present disclosure, "vehicle parts" refer to basic units that can be replaced or repaired separately during vehicle maintenance and repair. "Vehicle parts" do not refer to a certain subsystem of the vehicle (for example, "air conditioning subsystem"), but more specific components, such as air outlet temperature sensors, cooling fans, cooling fan controllers, compressors, etc. that constitute the air conditioning subsystem. Examples of "vehicle parts" include not only common components such as batteries, brake pads, and water temperature sensors, but also pistons, connecting rods, crankshafts, valves, etc. inside the engine. In the creation of the knowledge graph of the present disclosure, by defining "vehicle parts" as basic units in maintenance and repair, faults can be more accurately identified and located, thereby improving maintenance efficiency and vehicle reliability.

[0046] In step S302, a database of abnormal signals is established. The database of abnormal signals can be a comprehensive information base containing various vehicle fault indicators. These abnormal signals are not limited to the basic mode and state of the vehicle, but also include signals collected from various sensors and the driving data of the vehicle. In some instances, the abnormal signals may include vehicle information related to the vehicle fault. These signals usually come from multiple monitoring points inside the vehicle, including but not limited to the vehicle's operating mode, current state, data collected by sensors, and real-time driving information. The vehicle's operating mode can include not only the driving mode, but also the switching state between automatic driving and manual driving. The current state of the vehicle is a dynamically changing set, including various situations such as starting, accelerating, changing lanes, decelerating, braking, and stopping. Sensor signals are a direct reflection of the health of the vehicle, and they can provide key indicators such as battery temperature, power, temperature inside and outside the vehicle, oil pressure, tire pressure, etc. Driving data records the vehicle's speed, acceleration, mileage, and possible environmental factors such as road conditions and weather conditions.

[0047] In step S303, valuable fault knowledge, such as fault knowledge triples, are extracted from known unstructured fault diagnosis data and semi-structured fault diagnosis data. In some instances, the fault knowledge triples may include: <abnormal signal-fault name-faulty vehicle component> and <faulty vehicle component-repair method-repair suggestion>. Unstructured fault diagnosis data and semi-structured fault diagnosis data may come from vehicle sensors, maintenance records or driver feedback. The extraction of knowledge triples is to build a more intuitive and easy-to-understand fault diagnosis model. For example, an abnormal signal may indicate a specific fault name, and this fault is associated with a specific vehicle component. Similarly, for each faulty vehicle component, the system can provide corresponding repair methods and suggestions, which not only helps technicians quickly locate the problem, but also provides effective repair solutions. In this way, the fault knowledge triples not only optimize the fault detection process, but also improve the efficiency and accuracy of maintenance work.

[0048] In some instances, unstructured fault diagnosis data may be rich media fault cases. Rich media fault cases may include descriptions of fault symptoms and handling methods, images of faulty component locations, troubleshooting maintenance videos, and any combination thereof. Semi-structured fault diagnosis data may be fault repair ticket data. As a form of unstructured data, rich media fault cases can provide multi-dimensional information to help technicians quickly locate problems. For example, a detailed description of the fault symptoms can help understand the nature of the problem, while images of the faulty components can intuitively point out where the problem is. In addition, maintenance videos not only show the specific steps of troubleshooting, but also convey precautions and techniques during the maintenance process. These rich media contents can be used alone or in combination with other data types such as semi-structured fault repair ticket data, which records key information about fault handling, such as fault codes, maintenance history, and component replacement records, in a standardized format. This combined use can provide a more comprehensive perspective for fault diagnosis and enhance the availability and value of fault data.

[0049] In some instances, the "fault name" can be extracted from the name of the rich media fault case or the preliminary fault diagnosis result in the fault repair sheet. For example, one or more parts that are repaired or replaced as described in the rich media fault case or the fault repair sheet can be determined as "faulty vehicle parts". The vehicle sensor signal information involved in the rich media fault case or the fault repair sheet can be determined as "abnormal signal". Thus, a fault knowledge triple <abnormal signal-fault name-faulty vehicle part> can be constructed. Similarly, a fault knowledge triple <faulty vehicle part-repair method-repair suggestion> can be constructed. For example, if a repair sheet indicates that "replacing brake pads" is a method to solve a brake system fault, then this "repair method" can be extracted as a knowledge triple together with the faulty vehicle part "brake system" and the corresponding "repair suggestion". The number of occurrences of each fault knowledge triple can be updated and recorded. Therefore, the frequency or number of occurrences of each fault knowledge triple can be obtained.

[0050] These triplets can be used to train machine learning models to predict possible future failures and recommended repair measures. By analyzing a large amount of repair order data, the model can identify the association between failure modes and repair effects, thereby providing more accurate repair recommendations. This not only improves maintenance efficiency, but also helps prevent potential failures in advance and ensure vehicle safety. The construction of knowledge graphs is not limited to the extraction of triplets, but also includes the identification and linking of relationships between triplets. For example, if multiple repair orders indicate that cleaning the brake disc is usually recommended after replacing the brake pads, then this pattern can be identified as an association relationship between two triplets. Through such links, the knowledge graph can reflect maintenance knowledge and experience more dynamically and comprehensively. In addition, maintenance methods and recommendations may change over time, so the knowledge graph needs to be updated regularly to reflect the latest maintenance techniques and standards.

[0051] In step S304, based on the fault knowledge triple created in step S303, the abnormal signal, vehicle components and maintenance suggestions are associated together to form a knowledge graph for vehicle fault diagnosis. The created knowledge graph may include an abnormal signal node, a fault component node and a maintenance suggestion node. The abnormal signal node is associated with one or more fault component nodes that can generate the abnormal signal. The fault component node is associated with one or more maintenance suggestion nodes that can repair the fault component.

[0052] In the knowledge graph of the present disclosure, the relationship between abnormal signal i and vehicle component k can be represented by the edge weight E i / k This weight represents the probability P of component k failing given an abnormal signal i. i / k This approach allows the system to quickly assess the risk of failure of the associated components when a specific abnormal signal is detected. To calculate this probability, a statistical method can be used, as shown in Equation 1.

[0053] P i / k =C(i,k) / C(i) (Formula 1)

[0054] Here, C(i,k) represents the number of times abnormal signal i and vehicle component k appear simultaneously in the same fault scenario, while C(i) represents the total number of times abnormal signal i appears in all fault scenarios. In this way, a quantitative fault risk assessment based on historical data can be obtained.

[0055] The weight E of the edge between the abnormal signal i and the vehicle component k of the present disclosure is i / kOne advantage of the quantitative approach is its scalability. As more data is collected and analyzed, the knowledge graph can be continuously updated and optimized to improve the accuracy of fault diagnosis. For example, the probability calculation can be refined by introducing more parameters or using more complex statistical models. In addition, this approach can also be extended to other systems and components of the vehicle to provide support for the health management of the entire vehicle.

[0056] The edge between the faulty component node and the maintenance suggestion node can also be assigned a value. In some instances, if a faulty component node is associated with only one maintenance suggestion node, the edge between the faulty component node and the associated maintenance suggestion node can be assigned a value of 1. In some instances, if a faulty component node is associated with multiple maintenance suggestion nodes, the multiple edges between the faulty component node and the associated multiple maintenance suggestion nodes can be assigned values ​​based on the probability values ​​previously obtained. In some instances, if a faulty component node is associated with multiple maintenance suggestion nodes, the multiple edges between the faulty component node and the associated multiple maintenance suggestion nodes can be assigned values ​​based on the quality of the maintenance suggestion. For example, if a maintenance suggestion includes multi-dimensional explanations such as images, videos, and text, then the quality of the maintenance suggestion can be considered to be higher, and the value assigned to the edge between the faulty component node and the maintenance suggestion node is also higher (for example, 0.8). If a maintenance suggestion only includes images and text, then the quality of the maintenance suggestion can be considered to be lower, and the value assigned to the edge between the faulty component node and the maintenance suggestion node is also lower (for example, 0.3).

[0057] Figure 4 It is a flow chart of a method for diagnosing vehicle faults using a knowledge graph according to an embodiment of the present disclosure. The knowledge graph used in the method for diagnosing vehicle faults disclosed in the present disclosure may be a knowledge graph for vehicle fault diagnosis constructed by the method disclosed above in the present disclosure. In some embodiments, the method for diagnosing vehicle faults using a knowledge graph may include steps S401-S404. In step S401, when a vehicle fault monitoring module of a vehicle terminal detects a vehicle fault, a fault snapshot image is generated. Detecting the occurrence of a fault may include the vehicle terminal detecting a fault signal (e.g., a fault code) or an indicator light signal (e.g., a low battery indicator light is on). Detecting the occurrence of a fault may also be a fault prompt such as a report of "unable to start the car" by the car intelligent system. In some instances, the vehicle terminal may collect information through a CAN network, and collect and summarize the collected information through a fault monitoring module to generate a fault snapshot image. In step S402, the vehicle terminal sends the fault snapshot image to a cloud server.

[0058] In step S403, the cloud server parses the vehicle information from the fault snapshot image, and performs fault diagnosis in the fault knowledge graph in the cloud server based on the vehicle information to determine the target maintenance suggestion node. In some instances, parsing the vehicle information may include: the cloud server extracts the vehicle information associated with the fault from the fault snapshot image. The vehicle information may include the time of fault occurrence, vehicle driving mode, vehicle status, vehicle sensor data, vehicle driving data, or any combination thereof. The vehicle driving mode may include an automatic driving mode, an assisted driving mode, and a manual driving mode. The vehicle status may include vehicle start-up, vehicle acceleration, vehicle lane change, vehicle deceleration, vehicle braking, and vehicle stationary. The vehicle sensor data may include data collected by various components or sensors of the vehicle, including but not limited to battery temperature, battery power, coolant temperature, cabin temperature, etc. The vehicle driving data may include vehicle instantaneous speed, vehicle average speed, vehicle acceleration, etc. In step S404, the cloud server returns the information of the target maintenance suggestion node to the vehicle terminal as diagnostic information.

[0059] The fault diagnosis performed by the cloud server based on the knowledge graph may be a progressive fault diagnosis. In some embodiments, the progressive fault diagnosis performed by the cloud server may include: (i) identifying an abnormal signal in the vehicle information; (ii) locating an abnormal signal node corresponding to the abnormal signal in the knowledge graph; (iii) determining one or more faulty component nodes associated with the abnormal signal node as candidate faulty component nodes; (iv) obtaining one or more maintenance suggestion nodes associated with the candidate faulty component nodes; (v) determining a target maintenance suggestion node; and (vi) sending the target maintenance suggestion node as diagnostic information to the vehicle terminal.

[0060] In some instances, identifying an abnormal signal in vehicle information may include: the cloud server extracts vehicle information associated with the fault that occurred from the fault snapshot image. In some instances, the vehicle information may include the time of the fault, the vehicle driving mode, the vehicle status, the vehicle sensor data, the vehicle driving data, or any combination thereof. In some instances, after parsing the vehicle information such as the time of the fault, the vehicle driving mode, the vehicle status, the vehicle signal, the vehicle driving data, etc. from the fault snapshot image, the vehicle information may be compared with a normal value or value range. When one or several vehicle information exceeds the normal value range, the vehicle information that exceeds the normal data is identified as an abnormal signal. In some instances, vehicle information whose rate of change exceeds a preset threshold may be identified as an abnormal signal.

[0061] In a non-limiting example, if the "engine temperature value" indicated by the "engine temperature signal" parsed from the fault snapshot image is higher than a preset threshold, the "engine temperature signal" is identified as an abnormal signal. Subsequently, the abnormal signal node "engine temperature signal node" corresponding to the abnormal signal "engine temperature signal" is located in the knowledge graph, and the faulty component nodes associated with the abnormal signal node "engine temperature signal node", including but not limited to the temperature sensor node, water pump node, radiator node and coolant tank node, are determined as candidate faulty component nodes. Then, the maintenance suggestion nodes associated with the temperature sensor node, water pump node, radiator node and coolant tank node are obtained, including but not limited to "check temperature sensor node", "check water pump node", "check radiator node" and "check coolant node".

[0062] In some instances, performing fault diagnosis in the fault knowledge graph in the cloud server to determine the target maintenance suggestion node may include: determining a fault component node with the maximum fault probability from multiple fault component nodes as the target fault component node, and determining a maintenance suggestion node associated with the target fault component node as the target maintenance suggestion node. The fault probability P of each fault component node can be calculated by Formula 2 k :

[0063] P k =∑ i P i / k (Formula 2)

[0064] Here, P i / k = C(i,k) / C(i), C(i,k) is the number of times abnormal signal i and vehicle component k appear simultaneously in the same fault scenario, and C(i) is the number of times abnormal signal i appears in all fault scenarios. k Computational methods effectively exploit statistical data from failure scenarios to predict failed parts.

[0065] In some instances, a machine learning model (e.g., a convolutional neural network model (CNN)) can be used to extract the fault feature vector in the fault snapshot and the feature vector of each maintenance suggestion node and perform vector matching, so as to obtain the most accurate target maintenance suggestion node. In some instances, the maintenance suggestion node with the highest vector matching degree with the fault feature vector in one or more maintenance suggestion nodes can be determined as the target maintenance suggestion node. In some instances, one or more maintenance suggestion nodes with a vector matching degree with the fault feature vector exceeding a preset value in one or more maintenance suggestion nodes can be determined as the target maintenance suggestion node. In some instances, vector matching can be implemented using a similarity matching algorithm. This matching process can be based on a variety of criteria, such as Euclidean distance or cosine similarity, to find the maintenance suggestion node with the highest vector matching degree with the fault feature vector in one or more maintenance suggestion nodes. In this way, when a new fault is detected by the vehicle terminal, the machine learning model can quickly provide accurate maintenance suggestions, greatly reducing the diagnosis time, improving maintenance efficiency and equipment reliability. In addition, as time goes by, this matching process can be continuously optimized, and by continuously learning new fault cases, the maintenance suggestions are made more and more accurate, thereby realizing intelligent maintenance recommendations.

[0066] Figure 5 : is a structural block diagram of a vehicle fault diagnosis system based on a knowledge graph according to an embodiment of the present disclosure. In some embodiments, the vehicle fault diagnosis system based on a knowledge graph of the present disclosure may include: a vehicle terminal 510 and a cloud server 520. The vehicle terminal 510 may include a vehicle fault monitoring module and a vehicle fault sending module. The vehicle fault monitoring module may be configured to generate a fault snapshot image when a vehicle fault is detected. The vehicle fault sending module may be configured to send the fault snapshot image to the cloud server 520.

[0067] The cloud server 520 of the present disclosure may include a fault knowledge graph (not shown), a fault snapshot receiving module 521, and a fault diagnosis module 522. In some embodiments, the fault knowledge graph may include an abnormal signal node, a fault component node, and a maintenance suggestion node. The abnormal signal node may be associated with one or more fault component nodes associated with the generation of a vehicle fault. The fault component node is associated with one or more maintenance suggestion nodes that provide maintenance suggestions for the fault component.

[0068] In some embodiments, the fault snapshot receiving module 521 can be configured to receive a fault snapshot image sent from a vehicle terminal. The fault diagnosis module 522 can be configured to parse vehicle information from the snapshot image, perform progressive fault diagnosis in the fault knowledge graph in the cloud server based on the vehicle information to determine the target maintenance suggestion node, and send the information of the target maintenance suggestion node as diagnostic information to the vehicle terminal. In some embodiments, the fault diagnosis module 522 can be further configured to extract vehicle information associated with the vehicle fault from the fault snapshot image. The vehicle information may include the time of occurrence of the fault, the vehicle driving mode, the vehicle status, the vehicle sensor data, the vehicle driving data, or any combination thereof.

[0069] In some embodiments, the fault diagnosis module 522 can be further configured to: (i) identify abnormal signals in vehicle information; (ii) locate abnormal signal nodes corresponding to abnormal signal data in the knowledge graph; (iii) determine one or more faulty component nodes associated with the abnormal signal node as candidate faulty component nodes; (iv) obtain one or more maintenance suggestion nodes associated with the candidate faulty component nodes; (v) determine the target maintenance suggestion node; and (v) send the target maintenance suggestion node as diagnostic information to the vehicle terminal.

[0070] In some embodiments, operation (v) of the fault diagnosis module 522 may include: determining a fault component node with the maximum fault probability from among the multiple fault component nodes as a target fault component node, and determining a maintenance suggestion node associated with the target fault component node as the target maintenance suggestion node. The fault probability P of each fault component node can be calculated by Formula 2: k :

[0071] P k =∑ i P i / k (Formula 2)

[0072] Here, P i / k = C(i,k) / C(i), C(i,k) is the number of times abnormal signal i and vehicle component k appear simultaneously in the same fault scenario, and C(i) is the number of times abnormal signal i appears in all fault scenarios. k Computational methods effectively exploit statistical data from failure scenarios to predict failed parts.

[0073] In some embodiments, operation (v) of the fault diagnosis module 522 may include a vector matching operation. The vector matching operation may include: extracting one or more feature vectors from one or more maintenance suggestion nodes using a convolutional neural network model, and extracting a fault feature vector from a fault snapshot using a convolutional neural network model. In one example, the maintenance suggestion node with the highest vector matching degree with the fault feature vector in one or more maintenance suggestion nodes may be used as a target maintenance suggestion node. In one example, one or more maintenance suggestion nodes whose vector matching degree with the fault feature vector exceeds a preset value in one or more maintenance suggestion nodes may be determined as a target maintenance suggestion node.

[0074] The present disclosure also provides a system including one or more computer processors and a computer readable memory. The computer readable memory may include machine executable code, which, when executed by the one or more computer processors, implements the method of constructing a knowledge graph for vehicle fault diagnosis or the method of diagnosing vehicle faults using a knowledge graph of the present disclosure.

[0075] Although preferred embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that these embodiments are provided by way of example only. Other variations, changes and substitutions may be conceived by those skilled in the art without departing from the present invention.

Claims

1. A method for constructing a knowledge graph for vehicle fault diagnosis, the method comprising: Establishing a database of vehicle components, the vehicle components comprising basic units that can be replaced or repaired; Establishing a database of abnormal signals, wherein the abnormal signals include vehicle information related to vehicle failures, wherein the vehicle information related to vehicle failures includes an operating mode, a state, a sensor signal, and driving data of the vehicle; Creating a fault knowledge triple from known unstructured fault diagnosis data and semi-structured fault diagnosis data, wherein the fault knowledge triple includes: <abnormal signal-fault name-faulty vehicle component> and <faulty vehicle component-repair method-repair suggestion>; as well as Based on the fault knowledge triples, the abnormal signal, the vehicle component and the maintenance suggestion are associated together to construct the knowledge graph for vehicle fault diagnosis, wherein the probability P of abnormal signal i indicating a fault in vehicle component k is i / k As the weight E of the edge between the node corresponding to the abnormal signal i and the node corresponding to the vehicle component k in the knowledge graph i / k .

2. The method according to claim 1, wherein the knowledge graph comprises an abnormal signal node, a faulty component node and a maintenance suggestion node, wherein the abnormal signal node is associated with one or more faulty component nodes that can generate the abnormal signal, and the faulty component node is associated with one or more maintenance suggestion nodes that can repair the faulty component.

3. The method according to claim 1, wherein the probability P is calculated using formula 1 i / k : P i / k =C(i,k) / C(i) (Formula 1) in, C(i,k) is the number of times the abnormal signal i and the vehicle component k appear simultaneously in the same fault scenario, and C(i) is the number of times the abnormal signal i appears in all fault scenarios.

4. The method according to claim 1 further comprises updating the weight E using new data through a machine learning algorithm i / k . 5 . The method according to claim 2 , further comprising adding the probabilities corresponding to the plurality of abnormal signals associated with the vehicle component k to indicate the probability of failure of the vehicle component k.

6. The method according to claim 1 also includes using the fault knowledge triples to train a machine learning model so that the machine learning model can predict possible future faults and recommend maintenance measures.

7. The method according to claim 1 further includes identifying association relationships between multiple fault knowledge triples, thereby linking multiple fault knowledge triples to each other.

8. A system comprising one or more computer processors and a computer-readable memory, wherein the computer-readable memory comprises machine executable code, and when the machine executable code is executed by the one or more computer processors, the knowledge graph-based vehicle fault diagnosis method as described in any one of claims 1-3 is implemented.

9. A method for diagnosing vehicle faults using a knowledge graph, the method comprising: (a) When the vehicle fault monitoring module of the vehicle terminal detects that a vehicle fault occurs, a fault snapshot image is generated; (b) the vehicle terminal sends the fault snapshot image to a cloud server; (c) the cloud server parses vehicle information from the fault snapshot image, and performs fault diagnosis in a fault knowledge graph in the cloud server based on the vehicle information to determine a target maintenance suggestion node; as well as (d) the cloud server sends the information of the target maintenance suggestion node as diagnostic information to the vehicle terminal, The fault knowledge graph includes an abnormal signal node, a fault component node and a maintenance suggestion node, wherein the abnormal signal node is associated with one or more fault component nodes that can generate the abnormal signal, and the fault component node is associated with one or more maintenance suggestion nodes that can repair the fault component, and Determining the target maintenance suggestion node includes determining the faulty component node with the highest fault probability from the multiple faulty component nodes as the target faulty component node, and determining the maintenance suggestion node associated with the target faulty component node as the target maintenance suggestion node.

10. According to the method of claim 9, the failure probability P of each of the faulty component nodes k Calculated by formula 2: P k =∑ i P i / k (Formula 2) in, P i / k =C(i,k) / C(i), Wherein, C(i,k) is the number of times the abnormal signal i and the vehicle component k appear simultaneously in the same fault scenario, and C(i) is the number of times the abnormal signal i appears in all fault scenarios.

Images