In some embodiments, a probability P_i/kthat the abnormal signal i indicating the vehicle component k is faulty can be used as a weight E_i/kfor 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. In an example, if it is established based on the statistical analysis of historical data that, a probability of an abnormal battery power signal occurring in the event of a battery failure is 50%, a probability of an abnormal battery voltage signal occurring in the event of a battery failure is 10%, and a probability of an abnormal vehicle load level occurring in the event of a battery failure is 26%, then the edge E11 can be assigned a value of 0.5, the edge E21 can be assigned a value of 0.1, and the edge E31 can be assigned a value of 0.26.

In the knowledge graph of this disclosure, which incorporates the value assignments, the probability of an association between an abnormal signal and a fault of a specific vehicle component can be represented by the weight of the edge. This enables the vehicle fault diagnosis system to rapidly assess the probability of a fault of an associated vehicle component upon the detection of an abnormal signal. For example, in the event that the battery power signal is identified as abnormal, the vehicle fault diagnosis system can be capable of determining that there is a 50% probability of a fault in the battery, based on the weight of 0.5 for edge E11. The weight assignment approach of this disclosure not only improves the accuracy of fault diagnosis, but also enhances the real-time response to the vehicle status of the vehicle fault diagnosis system. Moreover, the knowledge graph of this disclosure can be optimized through the application of a machine learning algorithm. The weight of the edge can be modified on an ongoing basis in response to new data, allowing the system to adapt to the vehicle's operation in different environments. This enhances the accuracy of fault diagnosis and the reliability of the vehicle fault diagnosis system.

As discussed above, for one faulty component, there may be multiple abnormal signal nodes associated, which may occur simultaneously or interact with each other. Therefore, if a plurality of abnormal signal nodes are identified simultaneously, then there is a greater probability that the same faulty component node, to which those abnormal signal nodes being associated, is faulty. The probability or possibility of a fault of that same vehicle component can be indicated by summing a plurality of probabilities, each corresponding to a respective one of the plurality of abnormal signals each indicating a fault of that same vehicle component. For example, the battery system of a vehicle may contain a plurality of monitoring signals, such as a power, a voltage, a temperature, and the like. If one of those monitoring signals is abnormal, the vehicle fault diagnosis system can determine it as a minor fault or a normal fluctuation. However, if both the power and voltage monitoring points are abnormal at the same time, the vehicle fault diagnosis system can determine the possibility of a battery fault to be greatly increased. In this case, the vehicle fault diagnosis system can calculate a comprehensive abnormality index of those two signals, for example, by adding up the abnormality probabilities of the two signals, so as to obtain a more accurate fault prediction. In the above example, if both an abnormal battery charge signal and an abnormal battery voltage signal are present, the probability of the battery faulty is 50%+10%=60%. The edge weight-based analysis approach disclosed in this disclosure utilizes the correlation between signal nodes to improve the accuracy and reliability of fault detection by integrating the anomalies of a plurality of signal nodes.

The approach of assigning values to the edges between nodes in the knowledge graph and performing fault diagnosis based on the assigned values, as disclosed in this disclosure, can be applied to other sub-systems of the vehicle, including the power system, the braking system, the steering system, and so forth. By constructing a signal node network for each system and analyzing the interactions between the nodes, a comprehensive monitoring of the entire vehicle status can be achieved. In the event that a plurality of signal nodes from a plurality of sub-systems are identified as abnormal simultaneously, the fault diagnosis system is capable of rapidly locating the potential source of the fault and predicting the trajectory of its development.

FIG. 3 illustrates a flowchart of a method for constructing a knowledge graph for vehicle fault diagnosis according to an embodiment of this disclosure. In some embodiments, the method can comprise steps S301-S304. In step S301, a database of vehicle components can be constructed. As defined in this disclosure, a “vehicle components” refers to a fundamental unit that can be replaced or repaired individually during the course of vehicle repair and repair. The term “vehicle component” is not synonymous with a subsystem of a vehicle, such as an “air conditioning subsystem”.

Rather, it refers to a specific component, such as an air outlet temperature sensor, a cooling fan, a cooling fan controller, a compressor, and others, that collectively constitute the air conditioning subsystem. Examples of a “vehicle component” can include not only common components such as a battery, a brake pad and a water temperature sensor, but also a piston, a connecting rod, a crankshaft, a valve, and the like, which are found within an engine. The creation of the knowledge graph of this disclosure is based on the definition of a “vehicle component” as a basic unit in repair and repair. This approach allows for more accurate identification and localization of faults, thereby enhancing repair efficiency and vehicle reliability.

In step S302, a database of abnormal signals can be created. The database of abnormal signals can serve as a comprehensive repository of information, containing various vehicle fault indicators. The aforementioned abnormal signals are not limited to the fundamental operational modes and states of the vehicle; they also include signals collected from various sensors, as well as the vehicle's driving data. In some instances, the abnormal signals can comprise vehicle information pertinent to vehicle faults. These signals typically originate from a plurality of monitoring points within the vehicle and include, but are not limited to, data regarding the vehicle's operational mode, current state, sensor data, and real-time driving information. The mode of operation of the vehicle can include not only a driving mode, but also a switching state between automatic driving and manual driving. The current state of the vehicle can be defined as a set of conditions that are in a state of flux, comprising such actions as starting, accelerating, changing lanes, decelerating, braking, and stopping. The sensor signals can be a direct reflection of the vehicle's health, providing key indicators such as a battery temperature, a battery charge level, an interior temperature, an exterior temperature, an oil pressure, a tire pressure, and more. The driving information can include the vehicle's speed, acceleration, mileage, and potentially environmental factors such as road and weather conditions.

In step S303, valuable fault knowledge, such as fault knowledge triples, can be extracted from existing unstructured fault diagnostic data and semi-structured fault diagnostic data. In some instances, the fault knowledge triple can comprise: <abnormal signal-fault name-faulty component> and <faulty component-repair method-repair suggestion>. The unstructured fault diagnostic data and semi-structured fault diagnostic data can be derived from vehicle sensors, repair records, or driver feedback. The extraction of knowledge triples can facilitate the construction of a fault diagnosis model that is more intuitive and accessible. For example, an abnormal signal can indicate a specific fault name, which in turn can be linked to a specific vehicle component. Similarly, for each faulty vehicle component, the system can provide appropriate repair methods and recommendations, which facilitate the rapid identification of the source of the problem and offer effective repair solutions. In this manner, the fault knowledge triple not only serves to optimize the fault detection process, but also enhances the efficiency and accuracy of the repair work.

In some instances, the unstructured fault diagnostic data can be a media-rich fault case. A media-rich fault case can include a description of faulty symptoms and the repair methods employed to address them, an image of a location of the faulty component, a troubleshooting repair video, and any combination thereof. A semi-structured fault diagnostic data can be vehicle repair order data. As an example of unstructured data, the media-rich fault case can furnish the technicians with multi-dimensional, thereby facilitating the rapid identification of the source of the problem. For instance, a comprehensive description of the faulty symptoms can assist in comprehending the nature of the problem, whereas images of the faulty components can serve to visually indicate the location of the problem. Moreover, repair videos serve not only to illustrate the precise steps for troubleshooting but also to convey supplementary information in the form of notes and tips pertaining to the repair process. Such media-rich contents can be employed alone or in combination with other data types, including semi-structured vehicle repair order data. The latter records pertinent information regarding fault handling, including fault codes, repair history, and component replacement records, in a standardized format. The combination of these approaches offers a more comprehensive view of the troubleshooting process, thereby enhancing the accessibility and value of the troubleshooting data.

In some instances, the “fault name” can be derived from the name of the media-rich fault case or an initial fault diagnosis within the vehicle repair order. For example, one or more vehicle parts that are repaired or replaced as outlined in the media-rich fault case or the vehicle repair order can be identified as the “faulty vehicle component”. The vehicle sensor signal information that is described in the media-rich fault case or the vehicle repair order can be identified as the “abnormal signal”. Consequently, a fault knowledge triple <abnormal signal-fault name-faulty vehicle component> can be constructed. Similarly, a fault knowledge triple <faulty vehicle component-repair method-repair suggestion> can be constructed. For example, if a vehicle repair order indicates that “replacing the brake pads” is a solution to a brake system fault, then this solution can be extracted and associated with the faulty vehicle component “brake pad” and the corresponding repair suggestion, thus forming a fault knowledge triple. The number of instances of each fault knowledge triple can be updated and recorded, thus allowing the frequency or number of occurrences of each fault knowledge triple to be obtained.

The triples can be used to train a machine learning model, thereby enabling the model to forecast prospective failures and suggest appropriate repair strategies. In some instances, the analysis of a large amount of vehicle repair order data can enable the machine learning model to identify the relationships between faults and repair outcomes, thus facilitating the formulation of more precise repair recommendations.

This not only enhances the efficacy of repair operations but also facilitates a prevention of potential failures in a timely manner, thereby ensuring the safety of vehicles. The construction of fault knowledge graph is not limited to the extraction of triples; it also encompasses the identification and interlinking of relationships between triples. For example, if multiple vehicle repair orders indicate that cleaning the brake disc is typically advised following the replacement of brake pads, this correlation can be identified as an association relationship between two triples. By establishing such correlations, the knowledge graph can facilitate a more dynamic and comprehensive representation of repair-related knowledge and experience. Moreover, repair methodologies and recommendations may evolve over time. Consequently, the knowledge graph must be updated on a regular basis to reflect the latest repair techniques and standards.

In step S304, the fault knowledge triple created in step S303 can be utilized to associate the abnormal signal, faulty vehicle component and repair suggestion, thereby forming a knowledge graph for vehicle fault diagnosis. The created knowledge graph can comprise an abnormal signal node, a faulty component node and a repair suggestion node. The abnormal signal node can be linked to one or more faulty component nodes that can generate the abnormal signal. The faulty component node can be linked to one or more repair suggestion nodes that can repair the faulty component.

In the knowledge graph of this disclosure, the relationship between an abnormal signal i and a vehicle component k can be quantified by a weight E_i/kfor an edge. This weight can be interpreted as the probability P_i/kof fault of the component k when the abnormal signal i is present. This approach enables the system to rapidly evaluate the likelihood of a fault in the relevant component when a specific abnormal signal is identified. The probability can be calculated using a statistical approach, as in Formula I:

P_i/k=C(i,k)/C(i) Formula I

where C(i, k) is the number of times the abnormal signal i and the vehicle component k occur simultaneously in a same fault, and C(i) is the number of times the abnormal signal i occurs in all faults. In this way, a quantitative fault risk assessment based on historical data can be obtained.

The aforementioned quantization approach, as disclosed in this disclosure, offers a scalable solution for weight E_i/kof the edge between the abnormal signal i and the vehicle component k. As more data is accumulated and analyzed, the knowledge graph can be continuously updated and optimized, thereby enhancing the precision of fault diagnosis. For example, the probability calculation can be refined by incorporating additional parameters or utilizing more sophisticated statistical models. Moreover, this approach can be extended to other subsystems and components of the vehicle, thus facilitating comprehensive vehicle health management.

An edge between the faulty component node and the repair suggestion node can also be assigned a value. In some instances, if a faulty component node is associated with only one repair suggestion node, the edge between the faulty component node and the associated repair suggestion node can be assigned a value of 1. In some instances, if a faulty component node is associated with a plurality of repair suggestion nodes, a probability value based on previous statistics can be assigned to each of the plurality of edges between the faulty component node and the associated repair suggestion nodes. In some instances, if a faulty component node is associated with a plurality of repair suggestion nodes, the plurality of edges between the faulty component node and the associated plurality of repair suggestion nodes can each be assigned values based on a quality of the repair suggestion. For example, if a repair suggestion includes multi-dimensional explanations such as images, videos, and text, then the repair suggestion can be considered to be of a higher quality, and thus a higher value (for example, 0.8) can be assigned to the edge between the faulty component node and the repair suggestion node. If a repair suggestion includes only images and text, then the repair suggestion can be considered to be of a lower quality, and thus a lower value (for example, 0.3) can be assigned to the edge between the faulty component node and the repair suggestion node.

FIG. 4 illustrates a flowchart illustrating a method for diagnosing vehicle faults using a knowledge graph according to an embodiment of this disclosure. The knowledge graph employed in the diagnostic method for vehicle faults of this disclosure can be a knowledge graph for diagnosing vehicle faults constructed using the method disclosed in this disclosure. In some embodiments, the method for diagnosing vehicle faults using a knowledge graph can comprise steps S