Vehicle fault diagnosis method and system based on environment adaptive Bayesian network

Abstract

The invention provides a vehicle fault diagnosis method based on an environment adaptive Bayesian network. The method may include receiving, in a vehicle fault detection system, an input of a fault performance of a vehicle; according to the fault performance, selecting an optimal weighted correlation fault tree model from a library comprising a plurality of weighted correlation fault tree models; mapping the weighted correlation fault tree model to an environment adaptive Bayesian network; calculating the credibility of each leaf node in the environment adaptive Bayesian network; and generating a fault sequence and fault state information according to the credibility, and carrying out multimedia fault alarm. The vehicle fault diagnosis method disclosed by the invention not only shows excellent environmental adaptability in a changeable scene, but also can realize rapid and accurate diagnosis of vehicle faults, and powerfully supports reliable operation and preventive maintenance of the vehicle.

Description

Technical Field

[0001] The present invention relates to automatic fault diagnosis of vehicles, and more particularly to a vehicle fault diagnosis method and system based on a fault tree model and an environment-adaptive Bayesian network. Background Art

[0002] With the rapid advancement of automotive technology, modern cars have become more intelligent, but also more complex. They rely on a large number of electronic control units and sensors to manage key systems such as engines and brakes. Unfortunately, this complexity increases the possibility of failure, especially when multiple systems interact with each other. Traditional fault detection methods, such as rule-based expert systems and decision tree analysis, often seem powerless in the face of complex faults and cannot effectively handle the interactions between multiple components, resulting in inaccurate diagnosis. Even neural network models encounter problems such as long training time and falling into local minima when faced with real-time complex data. Therefore, a new diagnostic method is needed that combines system analysis and advanced intelligent algorithms to improve the accuracy and efficiency of fault detection in complex environments. Summary of the invention

[0003] One aspect of the present disclosure provides a vehicle fault diagnosis method based on an environment-adaptive Bayesian network. In some embodiments, the method may include (S1) receiving an input of a vehicle fault manifestation in a vehicle fault detection system; (S2) selecting an optimal weighted association fault tree model from a library containing multiple weighted association fault tree models according to the fault manifestation; (S3) mapping the weighted association fault tree model to an environment-adaptive Bayesian network; (S4) calculating the credibility of each leaf node in the environment-adaptive Bayesian network; and (S5) generating fault sorting and fault status information according to the credibility, and performing a multimedia fault alarm.

[0004] Another aspect of the present disclosure provides a system including one or more computer processors and a computer-readable memory, wherein the computer-readable memory includes machine-executable codes, and when the machine-executable codes are executed by the one or more computer processors, the vehicle fault diagnosis method based on the environment adaptive Bayesian network of the present disclosure is implemented.

[0005] The vehicle fault diagnosis method disclosed in the present invention innovatively combines a weighted correlation fault tree model and an environmentally adaptive Bayesian network. By introducing the correlation between fault severity weights and events in the weighted correlation fault tree model, the present method can accurately evaluate the impact of each potential fault on the system as a whole, thereby enhancing the ability to identify and analyze complex fault chains. In addition, the environmentally adaptive Bayesian network dynamically integrates environmental factors (such as temperature, humidity, road conditions, etc.) so that the diagnostic system can adjust the fault probability in real time according to actual working conditions. This adaptive adjustment mechanism significantly improves the fault diagnosis accuracy and response efficiency in complex environments, ensuring that the diagnostic results are more reliable. The vehicle fault diagnosis method disclosed in the present invention not only exhibits superior environmental adaptability in changing scenarios, but also can achieve rapid and accurate diagnosis of vehicle faults, strongly supporting the reliable operation and preventive maintenance of vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] 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.

[0007] Figure 1 It is a flow chart of a vehicle fault diagnosis method based on an environment adaptive Bayesian network according to an embodiment of the present application.

[0008] Figure 2 It is a flowchart of the credibility of each leaf node in the computing environment adaptive Bayesian network according to an embodiment of the present application. DETAILED DESCRIPTION

[0009] 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, due to the limited types of faults it covers, this diagnostic method that relies on fault codes is far from being able to adapt to the complex fault diagnosis of modern intelligent vehicles.

[0010] like Figure 1The flowchart shown shows a vehicle fault diagnosis method based on an environment-adaptive Bayesian network according to an embodiment of the present application. In some embodiments, the vehicle fault diagnosis method based on an environment-adaptive Bayesian network disclosed in the present invention may include steps S1-S5. In step S1, an input of a vehicle fault performance can be received in a vehicle fault detection system. In some instances, the vehicle fault detection system may be an industrial computer in a car repair center, a handheld detector used by a vehicle maintenance technician, or an intelligent device with a remote monitoring function. The vehicle fault detection system can obtain sensor readings directly from the vehicle's sensor system. These sensors include, but are not limited to, engine sensors, transmission sensors, brake system sensors, electrical system sensors, and suspension system sensors, etc., to provide accurate real-time data support. The vehicle fault detection system can also receive fault performance manually input by a technician or user, such as an abnormality perceived by observation, hearing, or other senses. Examples of fault performance include, but are not limited to, engine jitter, vehicle startup difficulty, transmission shift jam, transmission abnormal noise, longer braking distance, electrical system fault light on, abnormal suspension system bumps, etc.

[0011] In step S2, the detection system can select the optimal weighted associated fault tree model from a library containing multiple weighted associated fault tree models based on the fault manifestation input in S1. Specifically, the vehicle fault detection system first identifies key features related to the fault manifestation based on fault manifestation information (such as sensor data or manually entered fault description). These features may include fault type, frequency of occurrence, fault severity, and correlation with other faults. For example, if the input fault manifestation is "engine shaking", the system will extract key features related to this manifestation, such as engine speed, temperature, and fault code. Subsequently, these features are used as input to select the optimal weighted associated fault tree model from a pre-established library containing multiple weighted associated fault tree models. The model library can be pre-developed based on different types of vehicles, different fault modes, or different operating conditions. Each model should have its specific top events, intermediate events, and bottom events, as well as corresponding weights and fault correlations.

[0012] In the weighted associated fault tree model, each fault mode has a different weight, indicating its degree of impact on the overall system reliability. Through the matched fault tree model, the system can identify the key nodes and associated fault modes corresponding to the input fault manifestation. For example, for the fault manifestation of "engine jitter", the system can match a fault tree model centered on engine-related nodes to further analyze its possible causes, such as fuel system, ignition system or sensor abnormalities.

[0013] In some examples, step S2 may include: extracting and analyzing the characteristics of the fault manifestation; and using similarity calculation, selecting a weighted associated fault tree model having the highest similarity with the characteristics of the fault manifestation from a library containing multiple weighted associated fault tree models as the optimal weighted associated fault tree model. The similarity calculation may include using cosine similarity or Euclidean distance.

[0014] Through this mechanism of selecting the optimal weighted correlation fault tree model based on the characteristics of the fault manifestation, the method disclosed in the present invention can select the most suitable weighted correlation fault tree model according to different fault manifestations and environmental conditions, so that the system can better adapt to various complex situations. The rapid screening of the weighted correlation fault tree model enables the detection system to quickly determine the key nodes and potential fault paths for priority investigation, so as to more efficiently utilize resources and focus on possible root causes in subsequent diagnostic steps.

[0015] In step S3, the weighted association fault tree model can be mapped to an environmentally adaptive Bayesian network. In some instances, the top event, intermediate event, and bottom event of the weighted association fault tree model can be mapped to the root node, intermediate node, and leaf node of the environmentally adaptive Bayesian network, respectively, so as to construct a more refined fault probability model. The top event of the weighted association fault tree model can represent a system failure. The bottom event of the weighted association fault tree model can represent a specific fault that causes a system failure.

[0016] The weighted correlation fault tree model can include top events, intermediate events and bottom events. Taking engine failure as an example, the top event can be "engine failure", the intermediate events can include "insufficient fuel supply", "ignition failure", "overtemperature", and the bottom events can include "fuel pump failure", "spark plug damage", and "coolant leakage".

[0017] The top event in the weighted association fault tree model usually represents the overall failure of the system, and also represents the final manifestation or result of the system failure, which is used to aggregate the comprehensive impact of the bottom events and intermediate events on the system failure. The top event in the fault tree model can be mapped to the root node in the Bayesian network. The intermediate events in the fault tree model usually represent the failure or functional failure of a specific subsystem, which is an aggregation of the bottom events. The intermediate events in the fault tree model are mapped to the intermediate nodes in the Bayesian network, which helps to determine the failure impact path of the subsystem or subcomponent under environmental adaptive conditions, and provides middle-level analysis support. The bottom events in the fault tree model represent the specific faults that cause system failures (for example, sensor failures, motor failures, etc.). The bottom events in the fault tree model can be mapped to leaf nodes in the environmental adaptive Bayesian network. The leaf nodes reflect the dynamic performance of the underlying fault factors under different environmental conditions. These leaf nodes are combined with environmental factors θ iTo adjust the probability of failure, ensure that the failure risk of each leaf node can be accurately reflected under different working conditions.

[0018] In an environmentally adaptive Bayesian network, the root node, the intermediate node, and the leaf node have specific meanings respectively. The root node usually represents the most important result of the target event or system of the entire Bayesian network, such as an engine failure. The value of the root node reflects the probability of the environmental state (such as temperature, humidity, vibration, etc.) that causes the target event to occur under specific conditions. These values ​​provide the basic conditions for environmental dependence for subsequent calculations. The intermediate node connects the root node and the leaf node, and is used to represent different event combinations or hierarchical states in the system (such as the comprehensive state of different subsystems). The value of the intermediate node is the probability calculation result based on the root node condition, which represents the probability of occurrence of the intermediate event under the current environment and conditions, and is usually derived from the value propagation of the root node through the conditional probability distribution (i.e., Bayesian condition). The leaf node is the terminal node in the Bayesian network, which usually represents the state of a specific component or module (such as whether it is faulty). The value of the leaf node represents the probability of failure of the component under a specific environment, and these values ​​are calculated based on the intermediate node state and adjusted in combination with the environmental conditions. In the description of the present disclosure, "the probability of failure of a leaf node" refers to the probability of failure of the corresponding component represented by the value of the leaf node.

[0019] In step S4, the credibility of each leaf node in the environment adaptive Bayesian network is calculated. In some examples, calculating the credibility of each leaf node in the environment adaptive Bayesian network may include steps S401-S403.

[0020] like Figure 2 As shown in the flowchart of , in step S401, leaf nodes in the environment adaptive Bayesian network can be defined. In the environment adaptive Bayesian network, leaf nodes represent specific failure modes or failure sources. These leaf nodes are closely related to the overall failure state and environmental factors of the system. The credibility of each leaf node reflects the possibility of failure of the leaf node under specific environmental conditions.

[0021] In step S402, input of environmental factors may be received. When performing credibility calculation, environmental factors (such as real-time environmental parameters such as temperature, humidity, vibration, etc.) need to be taken into account, which will affect the failure probability of leaf nodes. The system should acquire and process these environmental data in real time to ensure the accuracy of credibility calculation.

[0022] In step S403, the following formula 1 can be used to calculate each leaf node e in the environment adaptive Bayesian network: i The credibility P k (T):

[0023]

[0024] Among them, F1(e i / T) represents the leaf node e i The probability of the first failure, and F2(e i / T) represents the leaf node e i The second failure probability F(T) represents the system failure probability calculated based on the failure probability of multiple leaf nodes. By introducing parameters related to the failure mode, the first failure probability F1(e i / T) can more concisely reflect the leaf node e i The probability of failure occurrence is described in detail in the following formula 2. i / T) By introducing dynamic real-time environmental factors, dynamic adjustment of the fault probability based on actual environmental data is achieved. With this method, the fault diagnosis method disclosed in the present invention can respond to environmental changes more flexibly and adaptably, thereby improving the accuracy and effectiveness of diagnosis.

[0025] In Formula 1, by multiplying the first fault probability and the second fault probability, the numerator reflects the probability of a specific leaf node failing under a specific environment and state. This can effectively quantify the risk of leaf node failure, especially considering the impact of environmental factors on leaf node failure. In Formula 1, by dividing the result of the numerator by F(T), the calculated numerator value can be standardized to a probability value. The probability ranges from 0 to 1, indicating the relative probability of a specific leaf node failing under given environmental conditions.

[0026] In some examples, the first failure occurrence probability F1(e i / T) can be calculated by the following formula 2:

[0027]

[0028] Formula 2 represents the cumulative probability of the normal and faulty states of a leaf node under all fault modes, reflecting the basic state distribution independent of environmental conditions. In Formula 2, under the jth fault mode, the leaf node e i Status of γ j By F(ω j ) weighted. In formula 2, all fault modes are accumulated for leaf node e i The influence of fault or normal state is determined by γ j The value of each fault mode will be i The influence of is distinguished and weighted. In Formula 2, ni is the leaf node e i The total number of relevant failure modes; γ j Indicates that leaf node e under the jth failure mode i The state of γj =1 indicates that leaf node e is in the jth fault mode i Normal, γ j =0 means that leaf node e is in the jth fault mode i Failure; j represents the state of the system under the jth fault mode, ω j =1 means the system is normal in the jth fault mode, ω j =0 indicates system failure in the jth fault mode; F(ω j ) represents the probability of system failure in the jth failure mode.

[0029] In some examples, the second failure occurrence probability F2(e i / T) can be calculated by the following formula 3:

[0030]

[0031] Formula 3 represents the joint impact of the states of all leaf nodes in the system on the potential failure of a leaf node under real-time environmental conditions. Formula 3 adjusts the interaction effect between leaf nodes by environmental factors and the total failure probability of the system. The product operation of Formula 3 Represents the state of all leaf nodes in the system for leaf node e i The combined impact of potential failures. i The value combination of can reflect the interaction between different leaf nodes on a certain leaf node e i The comprehensive impact of the fault state. In Formula 3, n represents the total number of leaf nodes in the environment adaptive Bayesian network; θ i is an environmental factor used to adjust the leaf node e under environmental changes i The probability of failure; i Represents the i-th leaf node e i The state of γ i =1 indicates leaf node e i Normal, γ i =0 indicates leaf node e i Fault. Formula 3 considers the system global fault F(T) and environmental factor θ i The comprehensive effect of the system on the leaf node e i Dynamic adjustment of the state is suitable for the adaptive response of the system under different environmental conditions.

[0032] Substituting equation 2 and equation 3 into equation 1, we get equation 4:

[0033]

[0034] In formula 4, (1-γ i) is a state adjustment item: when the leaf node e i Normal (i.e. γ i =1), the item is 0, indicating that the contribution of the normal state to the failure probability is zero; and when the leaf node e i Fault (i.e. γ i =0), this term is 1, indicating that the contribution of the fault state to the fault probability is 1. In formula 4, It is used to adjust the impact of each leaf node in the context of system failure. i Normal (i.e. γ i =1), The value of is 1, which is minus the system failure and environmental factors F(T)θ i The influence of leaf node e i When a fault occurs (i.e. γ i =0), The value is -1, that is, the system failure and environmental factors F(T)θ are added i Therefore, the numerator in Formula 4 Describes the situation under which the environment changes and considers a leaf node e under all failure modes i The cumulative sum of the probabilities of normal and faulty states, and considering the states of all leaf nodes in the system for a leaf node e i The comprehensive impact of potential failures on a leaf node e i The numerator in Equation 4 converts the leaf node e i The first failure probability (the failure probability of introducing parameters related to the failure mode) and the leaf node e i The second fault occurrence probability (fault occurrence probability with the introduction of dynamic environmental factors) is integrated together to achieve the effect of environmental adaptive fault diagnosis.

[0035] In some examples, the numerator F(T) in Formula 1 and Formula 4 can be calculated according to Formula 5 below:

[0036]

[0037] In Formula 5, F(e1∪e2…∪e i ) represents multiple leaf nodes e1, e2, …, e in the environment-adaptive Bayesian network n The failure probability of the root node (i.e., the entire system) in the adaptive Bayesian network is calculated. In Formula 5, q i It is a leaf node e i The probability of failure under the current environmental conditions. This is the probability value calculated based on the leaf node's own state and environmental factors, representing the leaf node e iThe probability that the corresponding component or module will fail under certain circumstances. i It is a leaf node e i The severity weight of the fault is in the range of [0,1]. The higher the weight value, the more likely the leaf node e i The greater the impact of the failure on the entire system or the root node. This weight enables the model to distinguish the importance of failures of different leaf nodes, ensuring that the failure of important components has a more obvious impact on the overall system. j ∣e i ) is an event association item, indicating that in the leaf node e i In case of failure, another leaf node e j The probability of failure. This term takes into account the interrelationships between different leaf nodes. For example, leaf node e i With leaf node e j The correlation is strong, then the leaf node e j At the leaf node e i The probability of failure is higher when a failure occurs, and vice versa. m represents the probability of failure at a given leaf node e i The number of other leaf nodes affected when a failure occurs. This value is used to describe the complexity of interaction between leaf nodes, that is, the number of leaf nodes that may be affected after a leaf node fails.

[0038] F(T) in Formula 5 is obtained by calculating the joint effect of the failure probability of each leaf node. Specifically, Formula 5 first calculates the joint effect of the failure probability of each leaf node e i The failure probability q i Weighted (i.e. multiplied by the severity weight ρ i ), and obtain the impact of leaf node failure on the overall failure probability. Then, consider the event correlation term P(e j ∣e i ) to reflect the correlation between leaf nodes, and these values ​​multiplied together reflect the fault transmission relationship between different leaf nodes. In formula 5, the overall failure probability of the system (root node) is calculated by subtracting the product of the failure probabilities of each leaf node from 1. This method ensures that in the extreme case where the leaf nodes do not affect each other (that is, there are no event correlation items), the overall failure probability of the system is the weighted sum of the failure probabilities of each leaf node. In the case where there is a correlation between leaf nodes, the overall failure probability of the system will be further adjusted according to the strength of the correlation. By combining the respective failure probabilities, severity weights, and event correlation relationships of the leaf nodes, formula 5 synthesizes the overall failure probability of the system or root node. This calculation method can dynamically adapt to different environmental conditions, accurately reflect the contribution of leaf node failures to system failures, and improve the accuracy of fault prediction and the environmental adaptability of the model.

[0039] In some embodiments, the severity weight p i Represents a leaf node e iThe importance or impact of the root node. Severity weight ρ i It measures the impact of a leaf node failure on the reliability or function of the overall system. i =0, indicating leaf node e i The impact of the system failure is completely ignored, that is, the occurrence of this bottom event will not have any impact on the reliability or function of the system. At this point, it can be considered that there is no relationship between the occurrence of this leaf node and the system failure. i =1, indicating leaf node e i The occurrence of has the greatest impact on system failure, that is, the leaf node e i The occurrence of will directly lead to system failure. i The value is between 0 and 1, indicating a leaf node e i It affects the failure probability of the system to a certain extent, but will not absolutely cause the system to fail. For example, ρi=0.5 may indicate that the leaf node e i After it occurs, the possibility of system failure increases, but the system does not necessarily fail.

[0040] In some embodiments, the event association term P(e j |e i ) is between [0, 1]. j |e i )=0, it means that at the leaf node e i In case of failure, another leaf node e j The probability of failure is zero. j |e i )=1, it means that at the leaf node e i In case of failure, another leaf node e j A failure will inevitably occur. <P(e j |e i )<1, it means that at the leaf node e i In case of failure, another leaf node e j There is a certain probability of failure, but it is not inevitable.

[0041] In the weighted correlation fault tree model, by introducing the event correlation term P(e j |e i ), can systematically analyze the temporal relationship between fault events, depict the fault propagation path, and deeply understand the root cause of the fault. For example, when a fuel pump fault occurs, the fault may increase the probability of insufficient fuel supply. The system will analyze the fuel pump fault based on the correlation between the two. At the same time, by setting the severity weight ρ i, the model can quantify the impact of each bottom event on the top event (system failure). For example, brake fluid leakage in the braking system will be given a higher severity weight than brake pad wear because brake fluid leakage directly affects the safety of the vehicle. This approach ensures that the method of the present invention can give priority to faults with greater impacts, thereby improving the accuracy and effectiveness of fault detection. When driving on slippery roads, abnormal vibration of the suspension system and brake fluid leakage may jointly cause brake system failures. The weighted associated fault tree model analyzes the impact of suspension system failures on the braking system, combines the severity weight of brake fluid leakage, and gives priority to brake fluid problems, thereby effectively avoiding possible brake failure accidents.

[0042] In Equations 1 and 4, the calculated value can be normalized to a probability value by dividing the denominator by the numerator F(T). The probability ranges from 0 to 1 and represents the relative likelihood of a particular leaf node failing under given environmental conditions. The denominator F(T) represents the overall probability of failure of the entire system under specific environmental conditions (i.e., the baseline). Comparing the numerator (i.e., the impact of the failure of a particular leaf node) with this baseline helps to determine the importance and relative contribution of the leaf node to the overall failure probability of the system. Specifically, if the failure probability of a leaf node is low relative to the overall failure probability of the system, it may mean that the node has limited impact on the overall failure of the system. Conversely, if the failure probability of a leaf node is low, this indicates that the node plays an important role in the failure of the system.

[0043] In addition, dividing the denominator F(t) in Equations 1 and 4 enables the failure impact of a specific leaf node described by the numerator to be evaluated in the context of the overall system failure risk. This relative evaluation helps identify critical failure modes and make dynamic adjustments. In terms of identifying critical failure modes, by calculating the credibility of each leaf node, it is possible to identify which leaf nodes are more likely to cause system failures under specific environmental conditions, thereby guiding the priority of maintenance and repair.

[0044] In terms of dynamic adjustment, as environmental conditions change, f(T) will also change, which enables the credibility of each leaf node to reflect the impact of the current environment on the risk of failure in real time. In addition, by introducing F(T) as the denominator, the robustness of the fault diagnosis model can be improved. For example, in different environments and conditions, F(T) can change dynamically to ensure that the model can adapt to different working scenarios and failure modes. This design improves the system's sensitivity to environmental changes, so that fault diagnosis is not only based on static data, but also takes into account the impact of dynamically changing environments. In step S5, fault sorting and fault status information can be generated based on the calculated credibility, and a multimedia fault alarm can be performed. In some instances, fault sorting can include calculating the credibility P of each leaf node in the environmental adaptive Bayesian network according to the calculated credibility P. k(T) value, sort the leaf nodes according to their failure risk. The leaf nodes with higher credibility values ​​indicate that they are more likely to fail under given environmental conditions, so they will be prioritized. The sorting results can be displayed in the form of a list, including the name of the leaf node, the credibility value, and the description of the failure mode. This helps technicians quickly identify and locate key components that may have failures. In some instances, in addition to sorting, fault status information of each leaf node can also be output. Fault status information includes, but is not limited to: current status (normal or faulty), fault type (such as engine failure, transmission system failure, etc.), impact level (assessment of the impact on vehicle operation), possible causes of failure (based on analysis of historical data and environmental factors), maintenance suggestions (recommended maintenance measures or information on replacement parts to help technicians solve faults more efficiently) or any combination thereof.

[0045] In some instances, the method of the present disclosure can automatically trigger a multimedia fault alarm based on the fault ranking and credibility results. The multimedia fault alarm may include an audible alarm, a visual alarm, a mobile phone push notification, or any combination thereof. The multimedia fault alarm may include a multi-level alarm to provide more detailed fault warnings, reminders, and emergency notifications. This multi-level alarm can help technicians or drivers take different measures based on the severity of the fault. The alarm of the present disclosure may include dynamically adjusting the alarm level. For example, if the fault condition is monitored to deteriorate, the alarm level can be automatically upgraded from a lower level to a higher level. If the status is monitored to return to normal after the fault is restored or the problem is solved, the alarm level can be automatically downgraded or eliminated. For alarm problems that are not resolved in a timely manner, the method of the present disclosure can provide regular reminders to ensure that these problems are given priority during the next maintenance.

[0046] In the fault diagnosis method disclosed in the present invention, after the fault tree model is mapped to the environment adaptive Bayesian network, the potential fault is calculated in the environment adaptive Bayesian network, which has many technical advantages compared to direct calculation in the fault tree model. The fault diagnosis method based on the environment adaptive Bayesian network disclosed in the present invention can effectively deal with complex environmental factors, enhance dynamic adaptability, and provide more accurate diagnosis results. First, the fault diagnosis method based on the environment adaptive Bayesian network disclosed in the present invention realizes dynamic adaptation to environmental factors. The fault tree model is a static model, which is difficult to reflect the impact of external environmental changes on system failures. The environment adaptive Bayesian network can dynamically integrate environmental variables (such as temperature, humidity, vibration, etc.), automatically adjust when calculating the fault probability, and reflect the impact of actual working conditions on potential faults. After mapping to the Bayesian network, the environmental factors will propagate through each node in the network, and dynamically correct the fault probability of each leaf node and the system. Therefore, the environment adaptive Bayesian network can provide a fault analysis result that is more in line with the actual situation. Secondly, the fault diagnosis method based on the environment adaptive Bayesian network disclosed in the present invention realizes the correlation processing between nodes. In the fault tree model, the association between nodes is expressed by logic gates (such as AND and OR gates). This method is difficult to handle multiple complex associations, especially when there are probabilistic associations between nodes.

[0047] Again, the fault diagnosis method based on the environment-adaptive Bayesian network disclosed in the present invention realizes accurate fault reasoning and probability calculation. In a Bayesian network, the complex dependencies between leaf nodes can be modeled more flexibly and accurately through conditional probability relationships. Through the inference of conditional probabilities in the network, the system can analyze the impact of associated nodes more carefully, which helps to capture and calculate potential chain failure effects. The core advantage of the Bayesian network lies in its probability-based reasoning ability. By mapping the fault tree model to the Bayesian network, the Bayesian reasoning method can be used in the network structure to accurately calculate the failure probability of the leaf nodes, and gradually reason from the bottom nodes to the top nodes. When calculating potential faults, the Bayesian network can update the probability of unknown events according to known conditions, which means that the failure probability can be dynamically adjusted based on existing observations (such as fault manifestations), so as to more accurately infer possible potential faults.

[0048] In addition, the fault diagnosis method based on the environment adaptive Bayesian network disclosed in the present invention supports uncertainty and real-time update. In complex vehicle systems, fault data often contains certain uncertainty or ambiguity. For example, sensor noise or data loss may cause incomplete data. In the Bayesian network, the probability model itself can handle this uncertainty and provide robust fault inference. The Bayesian network allows real-time updates and can continuously adjust node probabilities as new data is input (such as real-time environmental monitoring data or system status data), providing support for real-time fault diagnosis. Further, the fault diagnosis method based on the environment adaptive Bayesian network disclosed in the present invention improves the system diagnostic accuracy and adaptability. After mapping to the environment adaptive Bayesian network, the system can dynamically adjust the fault model according to real-time conditions, improving the adaptability and diagnostic accuracy of the model in complex and dynamic environments. This is of great significance to improving fault diagnosis efficiency, reducing misjudgments and missed judgments, optimizing maintenance decisions, etc., especially in intelligent monitoring or remote monitoring scenarios, which can greatly improve the overall performance of the system.

[0049] Example

[0050] In an exemplary embodiment, it is assumed that the vehicle braking system has a fault manifestation of "increasing braking distance". The vehicle fault diagnosis method based on the environment adaptive Bayesian network disclosed in the present invention may include steps S1-S5. In step S1, the vehicle maintenance technician or the vehicle monitoring system inputs the fault manifestation of "increasing braking distance" into the vehicle fault detection system. Vehicle sensor data is also input into the system at the same time, including data such as vehicle brake pad thickness, brake fluid level, and brake master cylinder pressure. In step S2, after the vehicle fault detection system receives the fault manifestation of "increasing braking distance", it matches it with the model in the weighted associated fault tree model library. By extracting and analyzing the characteristics of the fault manifestation and using similarity calculation, it can be judged that this fault manifestation may correspond to the best match with the "brake system fault tree model". The "brake system fault tree model" includes the fault causes and event hierarchies related to braking, and weights the severity impact of each potential fault. For example, the "brake system fault tree model" may include:

[0051] Top event: Braking system failure (the overall braking effect decreases, resulting in a longer braking distance).

[0052] Intermediate events: insufficient brake fluid, worn brake pads, master cylinder failure, etc.

[0053] Bottom event: includes specific faults such as brake fluid level sensor failure and insufficient brake pad thickness.

[0054] In step S3, the "brake system fault tree model" is mapped to an environment-adaptive Bayesian network. The environment-adaptive Bayesian network may include:

[0055] Root node: represents the failure of the braking system.

[0056] Intermediate nodes: correspond to the main components in the brake system, such as brake fluid, brake pads, master cylinder, etc.

[0057] Leaf nodes: These nodes are broken down into specific detection points such as brake fluid level, brake pad thickness, and air volume in the brake system, representing different fault states of each component.

[0058] In step S4, the probability of each leaf node failing under the current environment is calculated, that is, the reliability P of a particular leaf node under specific environmental conditions (such as temperature, humidity, etc.) k (T). This calculation process can be performed according to the following formula:

[0059]

[0060] Assuming that insufficient brake pad thickness is leaf node e1, insufficient brake fluid is leaf node e2, the specific process of calculating the credibility of the leaf node is as follows:

[0061] 1. Determine the system failure probability F(ω j ) and leaf node state parameter γ j

[0062] Define leaf node state parameter γ j , to represent the leaf node e i The operating state under each fault mode. If the leaf node operates normally under fault mode j, then γ j =1; if a fault occurs, γ j = 0. For example, for the problem of "braking distance becomes longer", assuming that the fault occurs due to insufficient brake fluid in a certain fault mode, then in this fault mode, the γ of the leaf node related to brake fluid is j = 0, and the leaf node γ related to the brake pad thickness j =1.

[0063] ·F(ω j ) represents the probability of system failure under fault mode j. For example, in a low temperature and high humidity environment, the viscosity of the brake fluid may increase, thereby increasing the possibility of a brake failure in the system. Assume that in this fault mode, F(ω j )=0.6, which means that the probability of system failure under this environmental condition is 60%.

[0064] 2. Calculate the leaf node e i The probability of the first failure

[0065] Leaf node e iThe first fault occurrence probability represents the cumulative probability of the normal and fault states of the leaf node under all fault modes, reflecting the basic state distribution independent of environmental conditions.

[0066] ·Sum up all failure modes for leaf node e i The influence of fault or normal state is determined by γ j The value of each fault mode will be i The impacts were differentiated and weighted.

[0067] 3. Calculate the leaf node e i The probability of the second failure

[0068] The second failure probability Represents the state of all leaf nodes in the system for leaf node e i The combined impact of potential failures.

[0069] (3.1) Determine each leaf node in the system for a specific leaf node e in a real-time environment i Status influence

[0070] (1-γ i ) is a state adjustment item: when the leaf node e i Normal (i.e. γ i =1), the item is 0, indicating that the contribution of the normal state to the failure probability is zero; and when the leaf node e i Fault (i.e. γ i =0), this term is 1, indicating that the contribution of the fault state to the fault probability is 1. In formula 4, It is used to adjust the probability response of each leaf node in the context of system failure. i Normal (i.e. γ i =1), The value of is 1, that is, the system failure and environmental factors F(T)θ are removed. i The influence of leaf node e i When a fault occurs (i.e. γ i =0), The value is -1, that is, the system failure and environmental factors F(T)θ are added i impact.

[0071] (3.2) Using the product Calculate the state of all leaf nodes in the system for leaf node e i The combined impact of potential failures

[0072] By multiplying the associated influence of each node item by item, we can get the influence of the state of all leaf nodes in the system on a specific leaf node e iThe combined impact of potential failures.

[0073] 4. Calculate molecules

[0074] Under the condition of environmental changes, considering the specific leaf node e in all failure modes i The cumulative sum of the probabilities of normal and faulty states of the leaf nodes in the system, and considering the status of all leaf nodes in the system for a specific leaf node e i The comprehensive impact of potential failures is calculated for a specific leaf node e i The overall probability of failure.

[0075] 5. Calculate the denominator F(T):

[0076] The denominator F(T) represents the failure probability of the entire system in the current environment. It is calculated by the failure probability of all leaf nodes in the system under the influence of different environmental factors. It is used to normalize the calculated value to ensure the credibility P k (T) is a probability value in the range [0,1].

[0077] The denominator can be calculated as follows:

[0078]

[0079] (5.1) Calculate the failure weight and probability of leaf nodes

[0080] In this item, q i Each leaf node e i The probability of failure, and ρ i is the fault severity weight of the node (the value range is [0,1]), which indicates the impact of the leaf node on the overall system failure. i ·q i ) calculate the node e i The probability that no failure occurs when the system is operating normally. Take the product of all leaf nodes The probability that the system remains normal under the combined state of all leaf nodes being normal or faulty can be obtained.

[0081] (5.2) Impact of fault correlation between computing nodes

[0082] P(e j ∣e i ) indicates that at node e i In case of failure, another node e j The probability of failure. By calculating (1-P(e j ∣e i )), get the value at node ei The probability that other related nodes will not fail when a failure occurs. Take the product of all related nodes It can be concluded that the probability of the system remaining normal in the event of multiple node failures.

[0083] (5.3) Reversal of overall system failure probability:

[0084] Subtract the above product term from 1, that is What is obtained is the failure probability F(T) of the system under the influence of combined failures of leaf nodes.

[0085] Through this calculation method, F(T) effectively combines the individual failure probability of the leaf node, the failure impact weight, and the failure correlation between nodes to obtain the probability of overall system failure, which is used for the subsequent calculation of the credibility of each leaf node.

[0086] 6. Calculate the credibility P k (T):

[0087] Divide the numerator by the denominator F(T) to obtain the fault occurrence credibility P of each leaf node in the current environment. k (T). For example, if P k (T) = 0.85, which means that the leaf node e i The reliability of the failure under this environmental condition is 85%, which means that the probability of failure of this leaf node under the current environment is high and needs to be checked first.

[0088] Through the above calculations, the vehicle fault diagnosis method based on the environment adaptive Bayesian network disclosed in the present invention can generate the credibility of each leaf node according to the fault impact of each leaf node under different environmental factors, and provide data support for further fault sorting and alarm.

[0089] In step S5, according to the credibility P of each leaf node k Based on the calculation result of (T), the system obtains the fault ranking under the fault manifestation of "braking distance becomes longer". For example, the ranking result may be:

[0090] 1. Insufficient brake fluid (highest reliability, priority check)

[0091] 2. Brake pad wear

[0092] 3. Brake master cylinder abnormality

[0093] The fault status information can be output according to this sequence and the driver or maintenance personnel can be reminded through multimedia. The alarm information can be presented in a graded manner:

[0094] Level 1 alarm: high priority (low brake fluid), accompanied by audio and warning light reminder, displaying "Brake system failure, it is recommended to check the brake fluid level immediately".

[0095] Level 2 alarm: Medium priority (brake pad wear), the screen will display “Brake pads may be worn,

[0096] It is recommended to maintain it as soon as possible."

[0097] The present disclosure also provides a system including one or more computer processors and a computer readable memory. The computer readable memory includes machine executable code. When the machine executable code is executed by one or more computer processors, the vehicle fault diagnosis method based on the environment adaptive Bayesian network of the present disclosure is implemented.

[0098] 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 vehicle fault diagnosis method based on an environment-adaptive Bayesian network, the method comprising: (S1) receiving input of a vehicle fault manifestation in a vehicle fault detection system; (S2) selecting an optimal weighted associated fault tree model from a library comprising a plurality of weighted associated fault tree models according to the fault manifestation; (S3) mapping the weighted association fault tree model to an environmental adaptive Bayesian network; (S4) calculating the credibility of each leaf node in the environment adaptive Bayesian network; and (S5) Generate fault ranking and fault status information according to the credibility, and perform multimedia fault alarm.

2. The diagnostic method according to claim 1, wherein the step S2 comprises: (S201) extracting and analyzing the characteristics of the fault manifestation; and (S202) Using similarity calculation, a weighted associative fault tree model having the highest similarity to the feature of the fault manifestation is selected from a library including a plurality of weighted associative fault tree models as the optimal weighted associative fault tree model. 3 . The diagnostic method of claim 2 , wherein the similarity calculation comprises using cosine similarity or Euclidean distance.

4. The diagnostic method according to claim 1, wherein the step S4 comprises: (S401) defining the leaf nodes in the environment adaptive Bayesian network, wherein the leaf nodes represent specific fault modes or fault sources; (S402) receiving input of environmental factors, wherein the environmental factors include real-time environmental parameters; as well as (S403) The leaf node e in the environment adaptive Bayesian network is calculated by applying the following formula 1: i The credibility P k (T): Among them, F1(e i / T) represents the leaf node e i The probability of the first failure occurring; F2(e i / T) represents the leaf node e i The second failure probability of the system; F(T) represents the failure probability of the system, and The failure probability of the system is based on multiple leaf nodes e in the environment adaptive Bayesian network i The failure probability is calculated.

5. The diagnostic method of claim 4, wherein F1(e i / T) is calculated by the following formula 2: Formula 2 represents the leaf node e i The cumulative probability of normal and faulty states under all failure modes, in, n i is the same as the leaf node e i The total number of relevant failure modes; γ j Indicates that leaf node e under the jth failure mode i The state of γ j =1 indicates that leaf node e is in the jth fault mode i Normal, γ j =0 indicates that the leaf node e in the jth fault mode i Failure; j represents the state of the system under the jth fault mode, ω j =1 means the system is normal in the jth fault mode, ω j =0 indicates system failure in the jth fault mode; F(ω j ) represents the probability of system failure in the jth failure mode.

6. The diagnostic method of claim 5, wherein F2(e i / T) is calculated by the following formula 3: Formula 3 represents the state of all leaf nodes in the system under real-time environment for the leaf node e i The combined impact of potential failures in, n represents the total number of leaf nodes in the environment adaptive Bayesian network, where θ i Represents the environmental factor, which is used to adjust the leaf node e under environmental changes i The probability of failure; i Represents the i-th leaf node e i The state of γ i =1 indicates the leaf node e i Normal, γ i =0 indicates the leaf node e i Fault.

7. The diagnostic method according to claim 6, wherein the step S4 comprises calculating the leaf node e in the environment adaptive Bayesian network using the following formula 4: i The credibility P k (T): in, Describes the leaf node e under the condition of environmental changes, considering all failure modes i The cumulative sum of the probabilities of normal and fault states, and considering the status of all leaf nodes in the system under real-time environment, the leaf node e i The combined impact of potential failures on the leaf node e i The overall probability of failure.

8. The diagnostic method according to any one of claims 4 to 7, wherein F(T) is calculated according to the following formula 5: in, F(e1∪e2…∪e i ) represents multiple leaf nodes e1, e2, ..., e in the environment adaptive Bayesian network n The failure probability of the root node in the environment adaptive Bayesian network is calculated according to the failure probability of the root node in the environment adaptive Bayesian network; q i It is a leaf node e i The probability of failure; ρ i It is a leaf node e i The severity weight of the fault, the value range is [0,1]; P(e j ∣e i ) is an event association item, indicating that in the leaf node e i In case of failure, another leaf node e j The probability of failure; m represents the probability of failure at a given leaf node e i The number of additional leaf nodes affected when a failure occurs.

9. The diagnostic method as claimed in claim 2, wherein the step S3 comprises mapping the top event, intermediate event and bottom event of the weighted association fault tree model to the root node, intermediate node and leaf node of the environmental adaptive Bayesian network respectively, wherein the top event of the weighted association fault tree model represents a system failure, and the bottom event of the weighted association fault tree model represents a specific failure causing the system failure.

10. 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 vehicle fault diagnosis method based on an environment adaptive Bayesian network as described in any one of claims 1 to 9 is implemented.

Images