A Method for Constructing a Case Library for Automotive Fault Detection and Diagnosis Based on Rules and Fault Trees

Abstract

This invention relates to the field of automotive fault detection technology, specifically a method for constructing an automotive fault detection and diagnosis case library based on rules and fault trees. The method includes: enhancing the context of original fault data with external environmental data to obtain environment-enhanced fault association data; classifying automotive fault features to obtain fault classification labels; performing rule matching on the fault classification labels according to a preset expert rule base to obtain matching results; performing case correlation analysis based on the fault propagation chain and a preset historical diagnostic case library to obtain case association features; constructing the case library according to a structured case model to obtain a diagnostic case library; and sending the diagnostic case library to a cloud platform to perform automotive fault detection and diagnosis. This invention, through rule matching of fault classification labels, can quickly and effectively identify known fault patterns based on expert experience and provide corresponding solutions for known faults.

Description

Technical Field

[0001] This invention relates to the field of automotive fault detection technology, and more specifically to a method for constructing an automotive fault detection and diagnosis case library based on rules and fault trees. Background Technology

[0002] Traditional methods often rely solely on raw fault data for fault diagnosis, neglecting the influence of external environmental factors. This results in a lack of comprehensive understanding of the background conditions under which the fault occurred, leading to low accuracy in fault diagnosis. This is particularly true for complex fault types or those heavily influenced by environmental factors, potentially resulting in misdiagnosis or missed diagnosis. Furthermore, fault identification and classification in traditional methods are typically coarse, failing to finely differentiate between different fault types, making it difficult to accurately label fault types and diagnose fault modes. This makes diagnosing complex fault modes difficult, and subsequent fault solutions are often imprecise, requiring multiple attempts or changes in diagnostic strategies. Additionally, traditional methods may rely heavily on pre-set rule bases or empirical rules, lacking in-depth integration with expert experience and a systematic rule matching mechanism. This results in slow identification speeds for known faults, low processing efficiency, and the inability to provide targeted solutions for different types of faults. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a method for constructing a case library for automobile fault detection and diagnosis based on rules and fault trees.

[0004] The technical solution adopted to solve the above-mentioned technical problems is: a method for constructing a case library for automotive fault detection and diagnosis based on rules and fault trees, including:

[0005] Obtain raw fault data and external environment data, and perform context enhancement on the raw fault data using the external environment data to obtain environment-enhanced fault association data;

[0006] Fault features are extracted from the environmental enhancement fault association data to obtain vehicle fault features; the vehicle fault features are then classified to obtain fault classification labels.

[0007] The fault classification labels are matched according to a preset expert rule base to obtain matching results; the matching results are analyzed according to the fault tree to obtain the fault cause; and the fault cause is propagated to obtain the fault propagation chain.

[0008] Based on the fault propagation chain and the preset historical diagnostic case library, case correlation analysis is performed to obtain case correlation features; the case correlation feature data is fused with rules and fault trees to obtain a structured case model;

[0009] The case library is constructed based on the structured case model to obtain a diagnostic case library; the diagnostic case library is sent to the cloud platform to perform vehicle fault detection and diagnosis.

[0010] Preferably, the original fault data and external environment data are acquired, and the external environment data is used to enhance the context of the original fault data to obtain environment-enhanced fault association data, including:

[0011] Signal acquisition is performed using a multimodal sensor array to obtain raw fault data; wherein, the multimodal sensor array includes a vibration sensor, an acoustic sensor, an infrared thermal imager, and an inertial measurement unit;

[0012] Feature extraction is performed on the original fault data to obtain vehicle operating characteristics;

[0013] Pattern matching is performed based on the vehicle operating characteristics and the original fault data to obtain preliminary fault associations; external environment data is acquired, and multi-dimensional context enhancement is performed on the preliminary fault associations based on the external environment data to obtain environment-enhanced fault association data.

[0014] Preferably, fault feature extraction is performed on the environmental enhancement fault association data to obtain vehicle fault features, including:

[0015] Time-domain waveform analysis is performed on the vibration signals in the environmental enhancement fault association data to obtain time-domain features; wherein, the time-domain features include extracting peak value, mean, and variance; spectral analysis is performed on the acoustic signals in the environmental enhancement fault association data to identify abnormal frequency components to obtain frequency-domain features;

[0016] A thermal map of the temperature distribution of vehicle components is constructed based on the infrared thermal imager. Overheated areas are located using image segmentation technology to obtain thermal imaging features. The thermal imaging features, vibration signals, and acoustic signals are spatiotemporally aligned to obtain a multimodal feature matrix.

[0017] Obtain road condition information, construct a dynamic load feature model based on the road condition information to obtain environmental adaptation features; integrate the environmental adaptation features into the multimodal feature matrix to obtain an enhanced feature set;

[0018] The enhanced feature set is subjected to nonlinear feature mapping to extract high-order abstract features to obtain deep learning features; the deep learning features, the time-domain features and the frequency-domain features are fused to obtain fused feature labels;

[0019] Semantic association is performed on the fused feature labels to identify causal relationships between features, thereby obtaining a structured fault feature map; the structured fault feature map is fed back to the edge computing node to obtain vehicle fault features.

[0020] Preferably, the fault classification labels are matched according to a preset expert rule base to obtain matching results, including:

[0021] The real-time operating status of the vehicle is obtained, and a subset of rules strongly correlated with the current operating condition is dynamically activated from the preset expert rule base based on the real-time operating status of the vehicle to obtain an environment adaptation set; wherein, the rule subset is pre-mapped through an operating condition-rule association matrix.

[0022] The structured fault feature map is matched according to the environmental adaptation set and the causal relationship paths in the structured fault feature map to obtain preliminary matching results;

[0023] The initial matching results are then conflict-resolved, and the matching conclusions of the high-priority rules are retained to obtain the final matching conclusion.

[0024] The matching conclusions are transformed into fault diagnosis suggestions described in natural language, and a visualization of the rule trigger path is added to obtain a matching report;

[0025] The weights of the corresponding rules in the preset expert rule base are dynamically adjusted based on the confidence data in the matching report. The optimized rule weights are then fed back to the dynamic rule activation module to form an adaptive rule matching mechanism to obtain the matching results.

[0026] Preferably, fault analysis is performed on the matching results based on the fault tree to obtain the fault causation; fault propagation is then performed on the fault causation to obtain the fault propagation chain, including:

[0027] The fault diagnosis suggestions in the matching results are semantically aligned with the basic event library of the fault tree. Key entities in the fault description are extracted and mapped to the corresponding basic event nodes in the fault tree to obtain the set of triggering events.

[0028] Starting from the set of triggering events, the path is expanded along the fault propagation direction, and all causal paths are generated by a breadth-first search algorithm; each causal path includes the logical relationship between the basic event, intermediate events and the top event;

[0029] The logic gates in the causal path are dynamically calibrated based on the real-time operating status of the vehicle to obtain a personalized fault tree.

[0030] The personalized fault tree is subjected to probabilistic risk assessment using the Monte Carlo simulation method. The probability of occurrence and criticality index of each causal path are calculated, high-risk causal paths are screened out and their weak links are marked.

[0031] The high-risk causal path is fed back to the expert rule base, the description of the fault propagation logic in the rule antecedent is dynamically updated, and synchronized to the dynamic rule activation module to form a causal path-driven rule optimization mechanism to obtain the fault causality.

[0032] Extract the basic events, intermediate events, and top event nodes of the high-risk causal path from the fault causality, and map the event nodes to actual physical components according to the vehicle system topology to obtain an initial node set;

[0033] The dynamic dependencies of each component in the initial node set are analyzed based on the real-time operating status of the vehicle. The strong and weak correlations of fault propagation between components are identified through correlation analysis algorithms to obtain a dynamic dependency graph.

[0034] The dynamic dependency graph is pruned and optimized to retain critical propagation paths and merge redundant nodes. At the same time, the closed loop of fault propagation is marked by a strongly connected component identification algorithm to obtain the fault propagation chain.

[0035] Preferably, a case correlation analysis is performed based on the fault propagation chain and a preset historical diagnostic case database to obtain case correlation features, including:

[0036] The dominant propagation chain node in the fault propagation chain is mapped to the case features in the historical diagnostic case library to obtain a case-propagation chain feature matrix; wherein, the case features include fault mode, component name and propagation path;

[0037] Based on the vehicle system topology, a correlation graph is constructed between the fault propagation chain and historical cases; wherein, the nodes in the correlation graph represent faulty components or cases, the edges in the correlation graph represent the similarity of propagation paths or the strength of case association, and the weight of the edges is calculated by the probability of the propagation chain and the frequency of case occurrence.

[0038] Preferably, the method further includes performing case correlation analysis based on the fault propagation chain and a preset historical diagnostic case database to obtain case correlation features, and also includes:

[0039] Node embedding learning is performed on the association graph, and neighbor node information is aggregated through a message passing mechanism to obtain high-dimensional features of cases and propagation chains; wherein, the high-dimensional features include case feature vectors and propagation chain feature vectors; the cosine similarity between the case feature vectors and the propagation chain feature vectors is calculated, and a set of associated cases with similarity exceeding a preset threshold is selected;

[0040] Based on the real-time operating status of the vehicle, the associated case set is context-adapted, and the solutions of historical cases are mapped to the current propagation chain scenario to obtain case adaptation suggestions;

[0041] The case adaptation suggestions are fed back to the association graph, and the case features in the historical diagnostic case library are updated through an incremental learning algorithm to form a propagation chain-driven case library adaptive optimization mechanism to obtain case association features.

[0042] Preferably, the case association features are fused with rules and fault trees to obtain a structured case model, including:

[0043] Key features are extracted from the case association features, and the key features are semantically aligned with the rules in the expert rule base to obtain a case-rule mapping table; wherein, the case-rule mapping table includes the matching degree score of rule triggering conditions and case fault features;

[0044] The fault propagation chain in the case is associated with the basic events and intermediate events of the fault tree, and the matching degree between the case propagation chain and the fault tree path is calculated to obtain the case-fitted fault tree; wherein, the weight of the events in the fault tree is adjusted according to the case propagation probability;

[0045] Based on the case-rule mapping table and the case-adapted fault tree, a hybrid graph structure model of rules and fault trees is constructed to obtain an initial structured model; wherein, rules are embedded as nodes in the logic gates of the fault tree, and intermediate events of the fault tree are the conclusion parts of the rules.

[0046] The initial structured model is fed back to the rule base and fault tree to update the condition descriptions of the rules and the event parameters of the fault tree, so as to obtain a structured case model.

[0047] Preferably, the case library is constructed based on the structured case model to obtain a diagnostic case library, including:

[0048] The rule-fault tree hybrid graph structure is parsed from the structured case model, and the correlation information is extracted to obtain a set of triples; wherein, the correlation information includes case features, rule nodes, and fault tree events;

[0049] The case library is categorized according to the vehicle system topology, and the triplet sets are stored hierarchically to obtain the initial case library; wherein, the hierarchical storage includes system modules, fault types, and propagation stages;

[0050] The initial case library is transformed into a knowledge graph using graph database technology. In the knowledge graph, nodes represent cases, rules, or fault tree events, and edges represent the triggering relationship between cases and rules, and the propagation path association between cases and fault trees. The attributes of the edges include confidence and occurrence frequency.

[0051] Preferably, constructing a case library based on the structured case model to obtain a diagnostic case library further includes:

[0052] The knowledge graph is populated with real cases from the historical diagnostic case library, and the historical cases are mapped to the corresponding nodes of the knowledge graph to obtain a hybrid case library.

[0053] The hybrid case library is fed back to the structured case model, and the generation strategy of the triple set and the storage structure of the knowledge graph are adjusted to obtain the diagnostic case library.

[0054] The beneficial effects of this invention are as follows:

[0055] (1) By enhancing the context of external environment data and original fault data, this invention can more comprehensively understand the background conditions of fault occurrence, which helps to improve the accuracy of fault diagnosis; and by extracting fault features from the environment-enhanced fault association data and classifying them, it can more precisely identify different types of faults and mark the fault type, which helps to transform complex fault patterns into operable classification labels, which facilitates subsequent fault diagnosis.

[0056] (2) By performing rule matching on the fault classification labels, the present invention can quickly and effectively identify known fault modes based on expert experience and provide corresponding solutions for known faults. Furthermore, by applying fault trees to further analyze the matching results, the causal relationship of the faults can be clarified, helping diagnostic personnel to understand how the faults propagate and clarifying the dependencies and interactions between the faults.

[0057] (3) By analyzing the fault propagation chain and combining it with the historical diagnostic case library to conduct case correlation analysis, this invention can extract similar fault cases, which helps to summarize experience from historical data; and by integrating rules and fault trees to construct a structured case model, it can effectively transform complex fault information into diagnostic cases that are easy to manage and find, making the subsequent case library construction more systematic and convenient for classifying and retrieving fault cases. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the overall method steps in one embodiment of the present invention. Detailed Implementation

[0059] Example 1, as Figure 1 As shown, the method for constructing a case library for automotive fault detection and diagnosis based on rules and fault trees proposed in this invention includes:

[0060] S1. Obtain the original fault data and external environment data, and perform context enhancement on the original fault data using the external environment data to obtain environment-enhanced fault-related data.

[0061] S2. Extract fault features from the environmental enhancement fault association data to obtain vehicle fault features; classify the vehicle fault features to obtain fault classification labels.

[0062] S3. Perform rule matching on the fault classification labels according to the preset expert rule base to obtain the matching results; perform fault analysis on the matching results according to the fault tree to obtain the fault cause; perform fault propagation on the fault cause to obtain the fault propagation chain.

[0063] S4. Perform case correlation analysis based on the fault propagation chain and the preset historical diagnostic case library to obtain case correlation characteristics; fuse the case correlation characteristics with rules and fault trees to obtain a structured case model;

[0064] S5. Construct a case library based on the structured case model to obtain a diagnostic case library; send the diagnostic case library to the cloud platform to perform vehicle fault detection and diagnosis.

[0065] In this invention, the final diagnostic case library is uploaded to a cloud platform, making it remotely accessible. The cloud platform provides powerful computing capabilities, supporting large-scale data processing and real-time fault diagnosis. Once the diagnostic case library is uploaded to the cloud platform, the system can automatically perform vehicle fault detection and diagnosis. By analyzing the vehicle's operating data in real time, the system can quickly identify fault types and provide repair suggestions.

[0066] In an optional embodiment, raw fault data and external environment data are acquired, and the external environment data is used to enhance the context of the raw fault data to obtain environment-enhanced fault-related data, including:

[0067] A1. Signal acquisition is performed using a multimodal sensor array to obtain raw fault data; the multimodal sensor array includes a vibration sensor, an acoustic sensor, an infrared thermal imager, and an inertial measurement unit.

[0068] A2. Extract features from the original fault data to obtain vehicle operating characteristics;

[0069] A3. Perform pattern matching based on vehicle operating characteristics and original fault data to obtain preliminary fault associations; acquire external environment data, and perform multi-dimensional context enhancement on the preliminary fault associations based on the external environment data to obtain environment-enhanced fault association data.

[0070] It should be noted that signal acquisition uses an array of multiple sensors (such as vibration sensors, acoustic sensors, infrared thermal imagers, and inertial measurement units) to collect signals generated by the vehicle during operation. These signals can include vehicle vibration, sound waves, heat changes, and inertial information, providing multi-dimensional data about the vehicle's state. The collected raw data is processed to extract features that characterize the vehicle's operating state. These features are obtained through data analysis and help identify key characteristics of vehicle operation, such as vibration frequency, sound wave intensity, and temperature changes. These features can provide effective information for subsequent fault analysis. The extracted vehicle operating features are then matched with known fault modes, with the aim of comparing existing fault data with the vehicle's current operating state. The system first assesses the vehicle's operational status to identify any preliminary fault correlations. It then acquires external environmental data related to vehicle operation, such as temperature, humidity, road conditions, and climate, as these factors affect vehicle performance. By combining this external environmental data, the preliminary fault correlation results are enhanced through multi-dimensional analysis. The aim is to increase the accuracy of the analysis by incorporating environmental factors. For example, the same vibration signal may indicate different faults at different ambient temperatures, or operation on slippery surfaces may lead to different fault modes than under normal conditions. This process, combined with the enhanced external environmental data, improves the accuracy and reliability of fault detection through more comprehensive contextual information, enabling fault diagnosis to more accurately reflect the actual situation.

[0071] In an optional embodiment, fault feature extraction is performed on the environmentally enhanced fault association data to obtain vehicle fault features, including:

[0072] B1. Perform time-domain waveform analysis on the vibration signals in the environmental enhancement fault association data to obtain time-domain features; the time-domain features include extracting peak value, mean, and variance; perform spectral analysis on the acoustic signals in the environmental enhancement fault association data to identify abnormal frequency components to obtain frequency-domain features.

[0073] B2. Construct a thermal map of vehicle component temperature distribution based on infrared thermal imager, locate overheated areas using image segmentation technology to obtain thermal imaging features; perform spatiotemporal alignment of thermal imaging features, vibration signals, and acoustic signals to obtain a multimodal feature matrix;

[0074] B3. Obtain road condition information, construct a dynamic load feature model based on the road condition information to obtain environmental adaptation features; integrate the environmental adaptation features into a multimodal feature matrix to obtain an enhanced feature set;

[0075] B4. Perform nonlinear feature mapping on the enhanced feature set to extract high-order abstract features to obtain deep learning features; fuse deep learning features, time-domain features, and frequency-domain features to obtain fused feature labels;

[0076] B5. Semantically associate the fused feature labels to identify causal relationships between features and obtain a structured fault feature map; feed the structured fault feature map back to the edge computing node to obtain vehicle fault features.

[0077] It should be noted that time-domain analysis of vibration signals can reveal their maximum value, indicating the limits of vibration amplitude, which is usually related to the severity of the fault. The average value of the vibration signal reflects the overall level of vibration; a higher or lower average value may indicate an abnormal vehicle operating condition. The degree of signal fluctuation, with a large variance indicating significant vibration changes, may suggest a more severe fault or anomaly. Vehicle sound signals can be used to detect problems such as abnormal engine noises and tire wear. Spectral analysis of acoustic signals can identify abnormal components within specific frequency ranges, which may be related to specific types of faults. Infrared thermal imagers can capture temperature distribution maps of vehicle components, and temperature... Changes are used to determine the fault area; by constructing a heat map of temperature distribution, the temperature of vehicle components can be displayed intuitively; image segmentation algorithms can identify areas of abnormal temperature (such as overheated areas), further aiding in fault location; different types of sensor data are collected synchronously, but their spatiotemporal characteristics differ; to effectively fuse these data, they must be spatiotemporally aligned to ensure the correct temporal and spatial correspondence of each data type; by integrating these spatiotemporally aligned data, a multimodal feature matrix is ​​formed, containing data features from various sensors; road conditions (such as road smoothness, slope, friction coefficient, etc.) have a significant impact on the vehicle's operating load, therefore, it is necessary to analyze real-time data... The fault detection model is adjusted based on road condition information; a dynamic load feature model is constructed based on road condition information to assess the vehicle's operating load under different road conditions, thereby adjusting the fault analysis results to adapt to different environmental factors; these features reflect the impact of the vehicle's operating environment on the vehicle load, which helps to accurately diagnose vehicle faults; environmental adaptation features are integrated into the previously obtained multimodal feature matrix to obtain a more comprehensive feature set that can consider the impact of environmental factors on vehicle faults; nonlinear feature mapping is performed on the enhanced feature set, and these features are further processed using deep learning technology to extract high-order abstract features, which can capture nonlinear and complex fault modes; time-domain features, frequency-domain features, and Deep learning extracts features and fuses them to obtain a comprehensive "label" representing the possible fault types of a vehicle under different operating conditions. Further semantic analysis is performed on the fused feature label to identify the causal relationships between different features. Semantic associations can reveal that certain features may cause changes in other features; for example, an abnormal frequency component may be a precursor to a certain fault. Through these causal relationships, a structured fault feature map is formed, which can help quickly identify and locate different types of faults. The obtained structured fault feature map is fed back to edge computing nodes. Edge computing can process this data in real time on the vehicle's local computing devices to diagnose and predict vehicle faults in a timely manner.This enables vehicles to process fault data in real time without relying on cloud computing, improving the efficiency of fault diagnosis.

[0078] In an optional embodiment, rule matching is performed on the fault classification labels according to a preset expert rule base to obtain matching results, including:

[0079] C1. Obtain the real-time operating status of the vehicle, and dynamically activate a subset of rules that are strongly correlated with the current operating condition from the preset expert rule base based on the real-time operating status of the vehicle to obtain the environment adaptation set; wherein, the rule subset is pre-established with a mapping relationship through the operating condition-rule association matrix.

[0080] C2. Based on the environmental adaptation set and the causal relationship paths in the structured fault feature map, perform rule matching on the structured fault feature map to obtain preliminary matching results;

[0081] C3. Resolve conflicts in the preliminary matching results and retain the matching conclusions of high-priority rules to obtain the final matching conclusion;

[0082] C4. Transform the matching results into fault diagnosis suggestions described in natural language, and add a visualization of the rule trigger path to obtain a matching report;

[0083] C5. Based on the confidence data in the matching report, dynamically adjust the weights of the corresponding rules in the preset expert rule base, and feed the optimized rule weights back to the dynamic rule activation module to form an adaptive rule matching mechanism to obtain the matching results.

[0084] It should be noted that the real-time vehicle operating status reflects the vehicle's current operating condition, helping the system to determine if potential faults exist; the expert rule base is a database containing a large number of fault diagnosis rules, usually built by experienced engineers or experts; these rules help the system infer possible fault causes from known fault modes; the operating condition-rule association matrix is ​​a mapping table indicating which specific rules are activated under different vehicle operating conditions (such as different driving conditions, loads, road conditions, etc.); based on the vehicle's real-time operating status, the system selects and activates a set of rules closely related to the current operating condition from the expert rule base. These rules are called the "environment adaptation set," reflecting the most applicable diagnostic rules under the current operating condition; the structured fault feature map is a map containing different fault modes and their causal relationships; the map lists the potential causes, symptoms, and interrelationships of various faults; the map shows the logical path of the fault, indicating how a certain fault feature leads to other symptoms; based on this, the system performs rule matching based on the activated environment adaptation set and the causal relationship path in the map, inferring preliminary fault causes. Fault diagnosis results: Sometimes, multiple rules in the rule base may give contradictory diagnostic results, which requires a conflict resolution mechanism to resolve. Each rule has a different priority, usually determined by experience or the reliability of the rule. During conflict resolution, the system retains the matching conclusion of the rule with higher priority, thus ensuring the most reliable diagnostic result. The system converts the matching results into easily understandable natural language descriptions and provides them to users (such as car owners, repair personnel, etc.). Visual charts are used to display the rule triggering path, helping users understand the diagnostic process and basis, thereby improving the transparency and reliability of fault diagnosis. Each diagnosis is accompanied by a confidence value, indicating the reliability of the fault diagnosis result. A high confidence value means a more accurate fault diagnosis, while a low confidence value suggests the need for more data support. Based on the confidence data in the matching report of each diagnosis, the system can dynamically adjust the weight of each rule in the expert rule base. The adaptive rule matching mechanism enables the system to adapt to different operating environments. Over time, the rule base will be continuously optimized to improve the overall accuracy of fault diagnosis.

[0085] In an optional embodiment, fault analysis is performed on the matching results based on the fault tree to obtain the fault causation; fault propagation is then performed on the fault causation to obtain the fault propagation chain, including:

[0086] D1. Semantically align the fault diagnosis suggestions in the matching results with the basic event library of the fault tree, extract the key entities in the fault description, and map them to the corresponding basic event nodes in the fault tree to obtain the set of triggering events.

[0087] D2. Starting from the set of triggering events, expand the path along the direction of fault propagation and generate all causal paths using a breadth-first search algorithm; each causal path includes the logical relationship between the basic event, intermediate events, and top event;

[0088] D3. Perform dynamic parameter calibration on the logic gates in the causal path based on the real-time operating status of the vehicle to obtain a personalized fault tree;

[0089] D4. Conduct probabilistic risk assessment of personalized fault trees using the Monte Carlo simulation method, calculate the probability of occurrence and criticality index of each causal path, screen out high-risk causal paths and mark their weak links.

[0090] D5. Feed high-risk causal paths back to the expert rule base, dynamically update the description of fault propagation logic in the rule antecedents, and synchronize it to the dynamic rule activation module to form a causal path-driven rule optimization mechanism to obtain the fault causality.

[0091] D6. Extract the basic events, intermediate events, and top event nodes of high-risk causal paths from the fault causality, and map the event nodes to actual physical components according to the vehicle system topology to obtain the initial node set.

[0092] D7. Analyze the dynamic dependencies of each component in the initial node set based on the real-time operating status of the vehicle, and identify the strong and weak correlations of fault propagation between components through correlation analysis algorithms to obtain a dynamic dependency graph.

[0093] D8. Prune and optimize the dynamic dependency graph, retain the critical propagation path and merge redundant nodes. At the same time, mark the closed loop of fault propagation through the strong connected component identification algorithm to obtain the fault propagation chain.

[0094] It should be noted that, based on the aforementioned matching results, the system will output a fault diagnosis report describing the fault type, cause, and impact. A fault tree is a graphical model used to describe the occurrence and propagation path of a fault; it consists of multiple basic events, logic gates, and a top event (the final fault event). Semantic alignment aims to match the key fault descriptions in the fault diagnosis recommendations with the basic events defined in the fault tree, ensuring that the fault information mentioned in the recommendations corresponds to the relevant nodes in the fault tree. This typically involves extracting key entities (such as specific components or symptoms) from the fault descriptions and mapping them to basic event nodes in the fault tree, thus forming a "trigger event set." Starting from the trigger event set, the system will expand along the fault propagation path according to the causal relationships in the fault tree. This means the system explores how a fault propagates from a basic event to other events, potentially leading to a top event. To ensure the comprehensive generation of all possible causal paths, the system uses a breadth-first search (BFS) algorithm. BFS traverses all nodes in the fault tree, expanding the paths layer by layer to form a complete causal chain from the basic event to the top event. Each causal path includes: basic events (starting events, such as component failure or sensor malfunction); intermediate events (subsequent events triggered by certain basic events); and the top event (the event that ultimately causes the system failure, usually the result of the entire fault chain). This process is based on the vehicle's real-time operating status. The system dynamically adjusts the logic gates in the causal paths; the state and behavior of these logic gates depend on the actual operating conditions of the vehicle, such as parameters like temperature and pressure. Through this dynamic calibration, the fault tree is personalized to better match the vehicle's current operating conditions and fault propagation characteristics. This ensures that the fault tree more accurately reflects the actual situation and avoids overly general diagnostic results. Monte Carlo simulation methods are used to assess the risk of the personalized fault tree through random sampling and statistical analysis. Extensive simulation experiments are conducted to calculate the probability of occurrence and criticality indicators for each causal path. These indicators help identify high-risk fault paths in the system. By calculating the probability of occurrence and criticality indicators for each causal path... The system can filter out high-risk paths based on their occurrence probability. These paths may lead to serious failures or security issues and therefore require special attention. Within high-risk paths, certain nodes or events may be particularly vulnerable, easily causing fault propagation. By identifying these weaknesses, preventative measures can be taken to reduce the probability of failure. High-risk causal paths are fed back into the expert rule base, and the rule antecedents (i.e., the preliminary conditions for fault propagation) are updated based on newly discovered fault propagation logic. Changes in the rule base are synchronized to the dynamic rule activation module, enabling the system to adjust and activate the latest fault diagnosis rules in a timely manner, thus forming a rule optimization mechanism based on causal paths.The system extracts nodes (including basic events, intermediate events, and top events) from high-risk causal paths of faults, reflecting key points in the fault propagation chain. Vehicle components, through physical and functional relationships, form a system topology. In this step, the system maps these causal path nodes to actual physical components, creating an initial node set representing the physical components involved in the fault propagation chain. Correlation analysis algorithms identify the strong and weak correlations between components in fault propagation. Strongly correlated components indicate faster fault propagation, while weakly correlated components may be relatively independent fault sources. Through this analysis and optimization, the system obtains the final fault propagation chain, providing effective information for fault prevention and repair.

[0095] In an optional embodiment, case correlation analysis is performed based on the fault propagation chain and a preset historical diagnostic case library to obtain case correlation characteristics, including:

[0096] E1. Perform feature mapping between the dominant propagation chain nodes in the fault propagation chain and the case features in the historical diagnostic case library to obtain the case-propagation chain feature matrix; where case features include fault mode, component name, and propagation path;

[0097] E2. Construct a correlation graph between the fault propagation chain and historical cases based on the vehicle system topology. In the correlation graph, nodes represent faulty components or cases, and edges represent the similarity of propagation paths or the strength of case association. The weight of the edges is calculated by the probability of the propagation chain and the frequency of case occurrence.

[0098] It's important to note that in a fault propagation chain, the dominant propagation chain node refers to those nodes that play a key role in the fault propagation process; the historical diagnostic case library contains all past fault diagnosis records; the fault mode describes the type or manifestation of the fault, such as damage or malfunction of a component; the component name is the specific component involved in the fault, such as the engine, sensor, or battery; the propagation path describes the path the fault travels from one component to another, including the mutual influence between different components; each row of the case-propagation chain feature matrix represents a historical diagnostic case, while each column represents a dominant node or specific feature in the propagation chain; through this mapping, the correlation between different fault modes and propagation paths can be analyzed, thereby more accurately predicting future faults; the vehicle system topology refers to how the vehicle's components and subsystems are interconnected and influence each other; in fault propagation chain analysis, the connections between these components are helpful... We understand how faults propagate from one component to another; the association graph connects nodes in the fault propagation chain to nodes (faulty components or cases) in the historical diagnostic case library; historical case nodes represent past fault diagnosis records; edges connecting different nodes indicate the similarity of their propagation paths or the strength of case association; the edge weights reflect the degree of propagation path similarity and case association; the edge weights are calculated using two types of information: propagation chain probability, which represents the probability of a specific fault propagation path occurring; by analyzing historical data of fault propagation chains, the probability of each propagation path can be calculated; and case occurrence frequency, which represents the frequency of a fault mode or propagation path in the historical case library, indicating the frequency of a specific case occurrence; by combining the probability of the propagation chain with the occurrence frequency of historical cases, the edge weights can reflect the relevance of the propagation path and the applicability of historical cases.

[0099] In an optional embodiment, case correlation analysis is performed based on the fault propagation chain and a preset historical diagnostic case library to obtain case correlation characteristics, and the method further includes:

[0100] E3. Perform node embedding learning on the association graph, and aggregate neighbor node information through message passing mechanism to obtain high-dimensional features of cases and propagation chains; among which, high-dimensional features include case feature vectors and propagation chain feature vectors; calculate the cosine similarity between case feature vectors and propagation chain feature vectors, and filter out the set of associated cases with similarity exceeding a preset threshold;

[0101] E4. Based on the real-time operating status of the vehicle, the relevant case set is context-adapted, and the solutions of historical cases are mapped to the current propagation chain scenario to obtain case adaptation suggestions.

[0102] E5. Feedback the case adaptation suggestions to the association graph, update the case features in the historical diagnostic case library through incremental learning algorithm, and form a case library adaptive optimization mechanism driven by the propagation chain to obtain case association features.

[0103] It's important to note that node embedding transforms each component node in the fault propagation chain and historical case nodes into a vector. This vector contains the node's features and its association information with neighboring nodes. In Graph Neural Networks (GNNs), the message passing mechanism involves each node exchanging information with its neighbors to update its embedding vector. Each node aggregates the information obtained from its neighbors and updates its own embedding representation. Historical diagnostic cases learn a vector representation through node embedding, containing the case's features. Each dominant node in the propagation chain also learns a vector through node embedding, representing its role in fault propagation and its relationship with neighboring nodes. Cosine similarity is a method for measuring the similarity between two vectors; the closer the value is to 1, the more similar the two vectors are. For case feature vectors and propagation chain feature vectors, the cosine similarity between them is calculated. Cosine similarity helps determine the degree of similarity between a historical case and the current propagation chain. By calculating cosine similarity, cases with high similarity to the current propagation chain can be selected from the historical case library, forming a set of related cases. Through context adaptation, solutions in historical cases can be mapped to the current propagation chain scenario. During the process of feeding case adaptation suggestions back to the association graph, the edge weights between nodes in the graph can be updated, making the information in the graph richer and more accurately reflecting the current fault propagation chain. Incremental learning is a continuous learning method that can update the model as new data arrives. Through incremental learning, the case features in the historical diagnostic case library can be dynamically updated. Whenever a new case adaptation suggestion is proposed, the model can update the node embeddings (i.e., case features) in the case library through the incremental learning algorithm, so that the case library can continuously adapt to new fault modes and propagation chains.

[0104] In an optional embodiment, rule-based and fault tree fusion is performed on case association features to obtain a structured case model, including:

[0105] F1. Extract key features from case association features, and semantically align the key features with the rules in the expert rule base to obtain a case-rule mapping table; wherein, the case-rule mapping table includes the matching degree score of rule triggering conditions and case fault features;

[0106] F2. Associate the fault propagation chain in the case with the basic events and intermediate events of the fault tree, and calculate the matching degree between the case propagation chain and the fault tree path to obtain the case-fitted fault tree; wherein, the weight of the events in the fault tree is adjusted according to the case propagation probability;

[0107] F3. Construct a hybrid graph structure model of rules and fault trees based on the case-rule mapping table and the case-adapted fault tree to obtain the initial structured model; wherein, rules are embedded as nodes in the logic gates of the fault tree, and intermediate events of the fault tree are the conclusion parts of the rules.

[0108] F4. Feed the initial structured model back to the rule base and fault tree, update the condition descriptions of the rules and the event parameters of the fault tree to obtain the structured case model.

[0109] It's important to note that case-related features refer to key attributes extracted from historical failure cases, such as failure type, occurrence conditions, and propagation path. These key features are the core of the set of features, effectively aiding in failure diagnosis and rule matching. The case-rule mapping table, through an alignment process, displays a score indicating the degree of match between the triggering conditions of each rule and the failure features in the case. This score measures the applicability of the rule to a specific failure case; a high score indicates that the rule is suitable for the current case, while a low score may indicate a lack of relevance. The case propagation chain refers to the path from one node to other nodes, typically related to different failure modes in the system. When matching the failure propagation chain with the fault tree path, the goal is to find the correspondence between the failure path propagating in the case and the events in the fault tree. The system calculates the matching degree between the case propagation chain and the fault tree path to determine which fault tree path is closest to the current case propagation chain; the weight of the event is adjusted according to the probability of case propagation; a new graph structure model is constructed by combining the rule base and the fault tree; rules may be logical judgments that trigger certain fault conditions, while the logic gates of the fault tree represent the results of these judgments; the occurrence of intermediate events in the fault tree may be affected by the rule triggering conditions; by constructing a hybrid graph structure model, the generated initial structured model is fed back into the rule base and the fault tree; as the number of cases increases, the condition descriptions of the rule base and the event parameters of the fault tree are updated according to the actual situation; the collaborative optimization mechanism combines the mutual influence between historical cases, rules, and the fault tree, enabling the entire fault diagnosis system to be continuously optimized.

[0110] In an optional embodiment, a case library is constructed based on a structured case model to obtain a diagnostic case library, including:

[0111] G1. Parse the rule-fault tree hybrid graph structure from the structured case model and extract the correlation information to obtain a set of triples; where the correlation information includes case features, rule nodes and fault tree events;

[0112] G2. Classify the case library according to the vehicle system topology and store the triple set in layers to obtain the initial case library; the layered storage includes system modules, fault types and propagation stages;

[0113] G3. The initial case library is transformed into a knowledge graph using graph database technology. In the knowledge graph, nodes represent cases, rules, or fault tree events, and edges represent the triggering relationship between cases and rules, and the propagation path association between cases and fault trees. The attributes of the edges include confidence and occurrence frequency.

[0114] It should be noted that the rule-fault tree hybrid graph structure is a graphical model that combines fault trees and rule bases, describing the occurrence, propagation, and rule triggering conditions of faults. Cases are categorized according to system modules, fault types, and fault propagation stages. Specific categorization is based on system modules, such as engines, braking systems, and transmission systems; each module may have specific fault types; fault types include powertrain faults, battery faults, and sensor faults; propagation stages include initial stages, expansion stages, and system crashes. Each case is stored in a database according to the above hierarchical structure, making subsequent queries and analysis more efficient. Knowledge graphs typically use graph databases (such as Neo4j) for storage and management because graph databases excel at handling complex relationships and associations. In this step, the initial case base (composed of triples) is... The set (composition) will be transformed into a graph structure; edges connect a case to the rule it triggers; that is, a case's features satisfy the triggering condition of a rule; this edge connects the case to the path in the fault tree, indicating that the fault in the case is associated with the event in the fault tree, and the fault propagates through this path; the occurrence frequency indicates the frequency of a specific rule or fault tree event in historical cases; high-frequency rules or events indicate that the problem is more common, which helps to further optimize the system's diagnostic model; the knowledge graph is a graphical data structure that displays the relationships between cases, rules, and fault tree events in the form of nodes and edges; it can: find rules and fault events related to a specific case; identify the commonalities and differences between different fault modes, helping to discover common fault types and their causes; and provide intelligent recommendations or decision support to guide maintenance personnel to prioritize high-risk faults.

[0115] In an optional embodiment, the case library is constructed based on a structured case model to obtain a diagnostic case library, further comprising:

[0116] G4. Fill the knowledge graph with real cases from the historical diagnostic case library and map the historical cases to the corresponding nodes of the knowledge graph to obtain a hybrid case library.

[0117] G5. Feed the hybrid case library back to the structured case model, adjust the generation strategy of the triple set and the storage structure of the knowledge graph to obtain the diagnostic case library.

[0118] It should be noted that each case in the historical cases will be mapped to a case node in the knowledge graph. Through this mapping, historical diagnostic cases will provide the knowledge graph with rich practical data and experience, supplementing and enhancing the case nodes in the graph. The addition of each new case can help update the triggering conditions of rules and fault propagation paths, further improving the accuracy of the knowledge graph. Through the filling of historical cases, the original knowledge graph will be combined with the historical case library to form a hybrid case library. This hybrid case library combines multi-dimensional information from rules, fault trees, and real cases, possessing stronger applicability and practical reference value. The updated and expanded hybrid case library will be fed back to the original structured case model. During the feedback process, the structured case model will readjust the original triple set generation strategy based on the new cases and information in the hybrid case library. By analyzing the new patterns and rules of historical cases, the triple set generation strategy can be optimized. The storage structure of the knowledge graph will also be adjusted according to the expansion of the hybrid case library. This is to better support new case types, rules, and fault tree events; for example, new node types or edge attributes may need to be introduced into the graph structure; adjustments to the storage structure help improve the scalability and query efficiency of the graph, enabling new cases to be stored and retrieved quickly and accurately; the model-case collaboration mechanism refers to the dynamic interaction and collaboration between the diagnostic model and the case library; as more historical cases are introduced, the information in the case library will be continuously enriched, driving the optimization of the fault diagnosis model; by learning from this historical data, the model can better predict and diagnose future faults; finally, after multiple feedback and optimizations, the resulting diagnostic case library contains more accurate and comprehensive diagnostic information; it not only covers historical cases but also includes knowledge built from rules, fault trees, and expert experience, which can help maintenance personnel quickly locate the cause of faults, formulate maintenance strategies, and optimize system performance through intelligent querying and reasoning; continuous optimization of structure and content enables it to gradually improve the accuracy and efficiency of fault diagnosis.

[0119] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for constructing a case base of automobile fault detection and diagnosis based on rules and fault trees, characterized in that, The method comprises the following steps: obtaining original fault data and external environment data, and performing context enhancement on the original fault data by using the external environment data to obtain environment-enhanced fault correlation data; extracting fault features from the environment-enhanced fault correlation data to obtain automobile fault features; classifying the automobile fault features to obtain fault classification labels; wherein the automobile fault features are obtained by feeding back a structured fault feature graph to an edge computing node; matching the fault classification labels with a pre-set expert rule base to obtain a matching result; analyzing the matching result by using a fault tree to obtain fault causes and effects; and propagating the fault causes and effects to obtain a fault propagation chain. According to the pre-set expert rule base, the fault classification labels are matched to obtain a matching result, comprising: obtaining real-time vehicle operating conditions, and dynamically activating a rule subset strongly related to the current working condition from the pre-set expert rule base according to the real-time vehicle operating conditions to obtain an environment adaptation set; wherein the rule subset is pre-established by a working condition-rule association matrix to establish a mapping relationship; matching the structured fault feature graph according to the environment adaptation set and the cause-and-effect relationship path in the structured fault feature graph to obtain a preliminary matching result; resolving conflicts in the preliminary matching result, and retaining matching conclusions of high-priority rules to obtain matching conclusions; transforming the matching conclusions into fault diagnosis suggestions described in natural language, and adding a visual graph of a rule triggering path to obtain a matching report; dynamically adjusting the weights of corresponding rules in the pre-set expert rule base according to the confidence data in the matching report, feeding back the optimized rule weights to a dynamic rule activation module, forming a self-adaptive rule matching mechanism, and obtaining a matching result; analyzing the case correlation according to the fault propagation chain and a pre-set historical diagnosis case library to obtain case correlation features; and fusing rules and fault trees according to the case correlation feature data to obtain a structured case model; constructing a case library according to the structured case model to obtain a diagnosis case library; and sending the diagnosis case library to a cloud platform to perform automobile fault detection and diagnosis.

2. The rule and fault tree based automobile fault detection and diagnosis case base construction method according to claim 1, characterized in that, Obtaining original fault data and external environment data, and performing context enhancement on the original fault data by using the external environment data to obtain environment-enhanced fault correlation data, comprising: signal acquisition by a multi-modal sensor array to obtain original fault data; wherein the multi-modal sensor array comprises a vibration sensor, an acoustic sensor, an infrared thermal imager, and an inertial measurement unit; extracting features from the original fault data to obtain vehicle operating features; performing pattern matching according to the vehicle operating features and the original fault data to obtain preliminary fault correlations; obtaining external environment data, and performing multi-dimensional context enhancement on the preliminary fault correlations according to the external environment data to obtain environment-enhanced fault correlation data.

3. The rule and fault tree based automobile fault detection and diagnosis case base construction method according to claim 2, characterized in that, Extracting fault features from the environment-enhanced fault correlation data to obtain automobile fault features, comprising: perform time-domain waveform analysis on the vibration signal in the environment-enhanced fault correlation data to obtain time-domain features, wherein the time-domain features include extracted peak values, mean values, and variances; perform frequency spectrum analysis on the acoustic signal in the environment-enhanced fault correlation data to identify abnormal frequency components to obtain frequency-domain features; construct a vehicle component temperature distribution thermograph according to the infrared thermal imager, locate an overheating area through image segmentation technology to obtain thermal imaging features; perform time-space alignment on the thermal imaging features, the vibration signal, and the acoustic signal to obtain a multi-modal feature matrix; obtain road condition information, construct a dynamic load feature model according to the road condition information to obtain environment adaptation features; integrate the environment adaptation features into the multi-modal feature matrix to obtain an enhanced feature set; perform nonlinear feature mapping on the enhanced feature set, extract high-order abstract features to obtain deep learning features; fuse the deep learning features, the time-domain features, and the frequency-domain features to obtain a fused feature label; perform semantic correlation on the fused feature label, identify the causal relationship between features to obtain a structured fault feature map; feed the structured fault feature map back to an edge computing node to obtain an automobile fault feature.

4. The rule and fault tree based automobile fault detection and diagnosis case base construction method according to claim 3, characterized in that, perform fault analysis on the matching result according to a fault tree to obtain a fault cause and effect; perform fault propagation on the fault cause and effect to obtain a fault propagation chain, including: perform semantic alignment on fault diagnosis suggestions in the matching result and a basic event library of the fault tree, extract key entities in fault descriptions, and map them to corresponding basic event nodes in the fault tree to obtain a trigger event set; starting from the trigger event set, perform path expansion in the fault propagation direction, and generate all causal paths through a breadth-first search algorithm; wherein each causal path includes a logical correlation relationship among basic events, intermediate events, and top events; perform dynamic parameter calibration on logical gates in the causal path according to the real-time running state of the vehicle to obtain an individualized fault tree; perform probabilistic risk assessment on the individualized fault tree through a Monte Carlo simulation method, calculate the occurrence probability and key degree index of each causal path, and screen out high-risk causal paths and mark their weak links; feed the high-risk causal paths back to an expert rule library, dynamically update the description of fault propagation logic in the rule antecedent, and synchronize to a dynamic rule activation module to form a rule optimization mechanism driven by causal paths to obtain a fault cause and effect; extract basic events, intermediate events, and top event nodes of the high-risk causal path from the fault cause and effect, and map the event nodes to actual physical components according to the vehicle system topology to obtain an initial node set; analyze the dynamic dependency relationship of each component in the initial node set according to the real-time running state of the vehicle, identify the strong and weak correlations of fault propagation between components through a correlation analysis algorithm to obtain a dynamic dependency graph; perform pruning optimization on the dynamic dependency graph, retain key propagation paths and merge redundant nodes, and mark the closed loop of fault propagation through a strongly connected component identification algorithm to obtain a fault propagation chain.

5. The rule and fault tree based automobile fault detection and diagnosis case base construction method according to claim 4, characterized in that, According to the fault propagation chain and the preset historical diagnosis case library, case correlation analysis is performed to obtain case correlation characteristics, including: The dominant propagation chain node in the fault propagation chain and the case characteristics in the historical diagnosis case library are mapped to obtain a case-propagation chain feature matrix; wherein the case characteristics include fault mode, component name and propagation path; According to the vehicle system topology structure, an association graph between the fault propagation chain and the historical case is constructed; wherein the nodes in the association graph represent fault components or cases, the edges in the association graph represent propagation path similarity or case correlation strength, and the weight of the edge is calculated by the propagation chain probability and the case occurrence frequency.

6. The rule and fault tree based automobile fault detection and diagnosis case base construction method according to claim 5, characterized in that, According to the fault propagation chain and the preset historical diagnosis case library, case correlation analysis is performed to obtain case correlation characteristics, also including: Node embedding learning is performed on the association graph, and neighbor node information is aggregated through a message passing mechanism to obtain high-dimensional features of cases and propagation chains; wherein the high-dimensional features include case feature vectors and propagation chain feature vectors; the cosine similarity of the case feature vectors and the propagation chain feature vectors is calculated, and a correlation case set with a similarity exceeding a preset threshold is screened out; According to the real-time running state of the vehicle, the correlation case set is contextually adapted, and the solution of the historical case is mapped to the current propagation chain scene to obtain a case adaptation suggestion; The case adaptation suggestion is fed back to the association graph, and the case characteristics in the historical diagnosis case library are updated through an incremental learning algorithm to form a propagation chain driven case library self-adaptive optimization mechanism to obtain case correlation characteristics.

7. The rule and fault tree based automobile fault detection and diagnosis case base construction method according to claim 6, characterized in that, The case correlation characteristics are fused with rules and fault trees to obtain a structured case model, including: Key features are extracted from the case correlation characteristics, and the key features are semantically aligned with the rules in the expert rule library to obtain a case-rule mapping table; wherein the case-rule mapping table includes rule trigger conditions and matching degree scores of case fault characteristics; The fault propagation chain in the case is associated with the basic events and intermediate events of the fault tree, and the matching degree of the case propagation chain and the fault tree path is calculated to obtain a case adaptation fault tree; wherein the weight of the event in the fault tree is adjusted according to the case propagation probability; According to the case-rule mapping table and the case adaptation fault tree, a hybrid graph structure model of rules and fault trees is constructed to obtain an initial structured model; wherein the rules are embedded into the logic gates of the fault tree, and the intermediate events of the fault tree are used as the conclusion part of the rules. The initial structured model is fed back to the rule library and the fault tree to update the condition description of the rule and the event parameter of the fault tree to obtain a structured case model.

8. The rule and fault tree based automobile fault detection and diagnosis case base construction method according to claim 7, characterized in that, According to the structured case model, a case library is constructed to obtain a diagnosis case library, including: The rule-fault tree hybrid graph structure is parsed from the structured case model to extract association information to obtain a triple set; wherein the association information includes case characteristics, rule nodes and fault tree events; According to the vehicle system topology, the case library is classified, and the triple set is stored in layers to obtain an initial case library; wherein the hierarchical storage includes system modules, fault types and propagation stages; The initial case library is converted into a knowledge graph through a graph database technology, wherein the nodes in the knowledge graph represent cases, rules or fault tree events, the edges represent the triggering relationship between cases and rules and the propagation path association between cases and fault trees, and the attributes of the edges include confidence and occurrence frequency.

9. The rule and fault tree based automobile fault detection and diagnosis case base construction method according to claim 8, characterized in that, According to the structured case model, the case library is constructed to obtain a diagnostic case library, further comprising: According to the real cases in the historical diagnostic case library, the knowledge graph is filled, and the historical cases are mapped to the corresponding nodes of the knowledge graph to obtain a hybrid case library; The hybrid case library is fed back to the structured case model, and the generation strategy of the triple set and the storage structure of the knowledge graph are adjusted to obtain a diagnostic case library.

Images