Vehicle-mounted detection device data processing method and device applied to fault detection

Abstract

The application provides a kind of vehicle-mounted detection equipment data processing method and device applied to fault detection, by the time sequence scene correlation modeling of original operation data collected by vehicle-mounted detection equipment, generate time sequence scene correlation mapping body and abnormal starting time sequence marker cluster, and based on both, construct fault correlation semantic topology, generate fault semantic correlation graph and scene-semantic correlation mapping table, then carry out fault mode qualitative clustering, generate fault mode classification label set and fault-scene correlation binding cluster, then carry out fault root cause topology reasoning, construct the logical correlation topology chain of fault explicit semantic representation and potential cause, generate fault root cause reasoning topology chain and fault propagation topology path set, based on both, carry out fault risk trend topology deduction, generate fault risk early warning semantic package and risk propagation trend topology graph.The application can improve the accuracy and system of vehicle fault detection as a whole, reduce analysis bias.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more specifically, to a data processing method and apparatus for vehicle-mounted detection equipment used in fault detection. Background Technology

[0002] Vehicle-mounted fault detection is a core component of ensuring vehicle operational safety. Existing vehicle-mounted detection equipment typically processes data by first performing continuous time-domain segmentation and marking of abnormal segments on the raw operational data collected by the equipment. Then, it matches and identifies each marked abnormal segment against a pre-defined fault rule base. Some solutions separate the temporal analysis and scenario association of the data, first completing the temporal feature analysis of the data, and then performing post-hoc correlation and matching of the analysis results with the vehicle system scenario information to finally output fault identification results and preliminary warning information. However, this approach has significant limitations. In the data preprocessing stage, the disconnect between temporal analysis and scenario association can easily lead to deviations in the scenario attribution of abnormal segments, resulting in errors in the basic information for subsequent fault identification. During the fault identification process, the matching method relying on the pre-defined rule base is difficult to cover fault types not included in the rules, and it cannot present the inherent correlation logic of fault-related information. In the fault root cause and risk prediction stages, it can only deduce based on the identified fault manifestations, making it difficult to trace the complete transmission path of the fault or effectively analyze the development logic of the fault causes. Ultimately, the accuracy and foresight of fault detection cannot meet the operational requirements of complex vehicle systems. Summary of the Invention

[0003] In view of this, the embodiments of this application provide at least one data processing method and apparatus for vehicle-mounted detection equipment applied to fault detection.

[0004] According to one aspect of the present invention, a data processing method for an on-board testing device applied to fault detection is provided. The method includes: performing time-series scene association modeling on the raw operating data collected by the on-board testing device; combining the continuous acquisition time-domain association relationship of the raw operating data and the scene switching startup logic of the on-board system to generate a time-series scene association mapping body and anomaly start time-series marker cluster for the raw operating data; based on the time-series scene association mapping body and anomaly start time-series marker cluster, constructing a fault association semantic topology; sorting out the semantic association links related to the abnormal performance of the on-board system in the raw operating data to generate a fault semantic association graph and a scene-semantic association mapping table; and importing the fault semantic association graph into a preset fault model. The system uses a semantic matching graph to perform qualitative clustering of fault patterns, matching fault pattern semantic entries in the fault semantic association graph with those in the fault pattern semantic matching graph to generate a fault pattern classification label set and a fault-scenario association binding cluster. Based on the fault pattern classification label set and the fault-scenario association binding cluster, it performs fault root cause topological reasoning, constructing a logical association topological chain between the explicit semantic representation of the fault and potential causes, generating a fault root cause reasoning topological chain and a fault propagation topological path set. Based on the fault root cause reasoning topological chain and the fault propagation topological path set, it performs fault risk trend topological deduction, analyzing the evolutionary topological logic of fault causes and the expansion topological direction of propagation paths, generating a fault risk warning semantic package and a risk propagation trend topological map.

[0005] According to another aspect of the present invention, a data processing apparatus is provided, comprising: a data modeling module, configured to perform time-series scene association modeling on raw operating data collected by an on-board detection device, and generate a time-series scene association mapping body and anomaly start time-series marker cluster based on the continuous acquisition time-domain association relationship of the raw operating data and the scene switching startup logic of the on-board system; a topology construction module, configured to perform fault association semantic topology construction based on the time-series scene association mapping body and the anomaly start time-series marker cluster, sort out the semantic association links related to the abnormal performance of the on-board system in the raw operating data, and generate a fault semantic association graph and a scene-semantic association mapping table; and a graph clustering module, configured to import the fault semantic association graph into a preset fault mode semantics. The system employs a matching graph to perform qualitative clustering of fault modes, matching fault mode semantic entries in the fault semantic association graph with those in the fault mode semantic matching graph, generating a fault mode classification label set and a fault-scenario association binding cluster. The root cause reasoning module performs topological reasoning based on the fault mode classification label set and the fault-scenario association binding cluster, constructing a logical association topological chain between explicit fault semantic representations and potential causes, generating a fault root cause reasoning topological chain and a fault propagation topological path set. The fault inference module performs topological inference of fault risk trends based on the fault root cause reasoning topological chain and the fault propagation topological path set, analyzing the evolutionary topological logic of fault causes and the expansion topological direction of propagation paths, generating a fault risk warning semantic package and a risk propagation trend topological map.

[0006] This invention models the continuous acquisition of vehicle-mounted raw operating data in a temporal domain, linking it to scene switching and initiation logic. This generates a temporal scene association mapping and anomaly initiation temporal marker clusters, ensuring a precise correlation between the temporal sequence of raw data and scene attributes. Based on these mappings and marker clusters, a fault association semantic topology is constructed. Semantic association links related to anomalies are identified, generating graphs and mapping tables. The system presents the fault association logic, matching and clustering the fault semantic association graph with a preset fault pattern graph to generate classification label sets and fault-scene binding clusters. Structured graph units are used to complete fault pattern recognition, improving the accuracy of pattern classification. Relying on the label sets and binding clusters, a logical association topology chain between explicit fault representations and potential causes is constructed, generating root cause reasoning topology chains and propagation path sets. The system identifies the origin and transmission logic of faults, accurately tracing fault causes. Based on the root cause topology chains and propagation path sets, fault risk trends are inferred, analyzing the cause evolution logic and path expansion direction, generating early warning semantic packages and trend topology maps. This accurately maps the fault risk development logic, improving the foresight of early warnings. The various steps of the method of this invention are closely linked, and the structured correlation information is used to promote fault detection throughout the entire chain, thereby improving the accuracy and systematicness of vehicle fault detection and reducing analysis bias. Attached Figure Description

[0007] Figure 1This is a schematic diagram illustrating the application scenarios provided in the embodiments of this application.

[0008] Figure 2 This is a schematic diagram illustrating the implementation process of a data processing method for vehicle-mounted detection equipment used in fault detection, as provided in an embodiment of this application.

[0009] Figure 3 This is a schematic diagram of the composition structure of a data processing device provided in an embodiment of this application.

[0010] Figure 4 This is a schematic diagram of the hardware entity of a computer device provided in an embodiment of this application. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application are further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] The data processing method for vehicle-mounted detection equipment used in fault detection provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, the vehicle-mounted terminal 102 communicates with the server 104 via a network. The server 104 can be a standalone server or a server cluster consisting of multiple servers. A data storage system can store the data that the server 104 needs to process. The data storage system can be integrated onto the server 104 or located in the cloud or on other network servers. The data collected by the vehicle-mounted terminal can be stored in the local storage of the vehicle-mounted terminal 102, or uploaded to the data storage system or cloud storage associated with the server 104. When the method of this embodiment needs to be executed, the server 104 can obtain relevant vehicle-mounted data from the local storage of the vehicle-mounted terminal 102, the data storage system, or the cloud storage.

[0013] Please refer to Figure 2 The data processing method for vehicle-mounted detection equipment used in fault detection provided in this application specifically includes the following steps:

[0014] Step S100: Perform time-series scene association modeling on the raw operating data collected by the vehicle-mounted detection equipment. Combine the continuous acquisition time-domain association relationship of the raw operating data and the scene switching startup logic of the vehicle system to generate the time-series scene association mapping body and the abnormal start time-series marker cluster of the raw operating data.

[0015] Raw operational data consists of data collected by onboard detection equipment during vehicle operation, detailing the operational status of various systems and components. Examples include engine speed, temperature, and oil pressure; vehicle speed, acceleration, and braking status. This data reflects the vehicle's actual operating conditions at different times. Continuous acquisition temporal correlation represents the chronological order and interdependencies of the raw operational data acquisition, demonstrating the continuity and temporal sequence of data collection. The onboard system's scene switching initiation logic defines the conditions and rules for switching between different operational scenarios, such as switching from a startup scenario to a driving scenario, or from a driving scenario to a braking scenario. Each scenario switch has corresponding triggering conditions. The temporal scene correlation mapping is the result of mapping and associating the raw operational data with the onboard system's scene switching rules. It contains all corresponding entries for all temporal feature data and scene rules, clearly showing the scenario corresponding to each time period. The abnormal start temporal sequence marker cluster is a set obtained by marking segments in the raw operational data that deviate from the normal scene switching rules, containing the time range information of all abnormal temporal segments.

[0016] In one implementation, step S100 may specifically include the following steps S110 to S160:

[0017] Step S110: Perform continuous temporal segmentation on the raw operating data collected by the vehicle-mounted testing equipment. Divide the raw operating data into multiple groups of time-domain segments with fixed time spans and no overlap according to the continuous time sequence of collection. Perform unified feature extraction on the multi-source heterogeneous raw data in each time-domain segment to generate standardized data feature representations of the corresponding time-domain segments. Obtain multiple groups of time-domain feature data with a continuous collection order. Each group of time-domain feature data contains standardized data feature representations of the corresponding time-domain segments.

[0018] Temporal continuous slicing divides the raw operational data chronologically into multiple non-overlapping time segments with fixed time spans; each time segment is a temporal fragment. Multi-source heterogeneous raw data refers to raw operational data from different data sources with different data types and formats, such as analog signal data from engines and digital signal data from sensors. Unified feature extraction processes the multi-source heterogeneous raw data within each temporal fragment to extract key information that represents the data characteristics of that temporal fragment. Temporal feature data is the dataset obtained after temporal slicing and feature extraction; each set of temporal feature data contains standardized data feature representations of the corresponding temporal fragment.

[0019] For example, first, a fixed time span is determined, such as 1 second, and the raw operational data is divided into multiple time-domain segments according to time sequence. For the multi-source heterogeneous raw data within each time-domain segment, different feature extraction methods are used. For example, for engine speed data, features such as average, maximum, and minimum values ​​can be extracted; for vehicle acceleration data, features such as rate of change can be extracted. Then, the extracted features are standardized to have the same dimensions and range. For example, the Z-score standardization method is used, subtracting the mean from each feature value and then dividing by its standard deviation to obtain the standardized feature value. Finally, the standardized data feature representations of each time-domain segment are combined to obtain multiple sets of time-domain feature data with a continuous acquisition sequence.

[0020] Step S120: Perform time-series correlation analysis on multiple sets of time-domain feature data, connect and correlate each set of time-domain feature data according to the order of acquisition, and obtain the acquisition sequence dependency relationship between each set of time-domain feature data and adjacent time-domain feature data to generate the continuous acquisition time-domain correlation relationship of the original running data.

[0021] The acquisition sequence dependency refers to the order and mutual influence between each set of time-domain feature data and adjacent time-domain feature data in terms of acquisition time. For example, the acquisition of a later set of data may depend on the acquisition results of a previous set of data. The continuous acquisition time-domain correlation is the result obtained by sorting out the acquisition time sequence of multiple sets of time-domain feature data, which reflects the continuity and temporal sequence of the original running data in terms of acquisition time.

[0022] For example, multiple sets of time-domain feature data are first arranged in chronological order of acquisition time. Then, the relationship between each set of time-domain feature data and adjacent time-domain feature data is analyzed. For example, for engine speed data, the speed in a later time-domain segment may be affected by the speed in the previous time-domain segment and other factors. By analyzing these relationships, the acquisition sequence dependency of each set of time-domain feature data and adjacent time-domain feature data is determined. Finally, these dependencies are organized and recorded to generate the continuous acquisition time-domain correlation of the original operating data.

[0023] Step S130: Semantically translate the scene switching startup logic of the vehicle system, convert the conditions for starting scene switching in the vehicle system into semantic rules that can be associated with the acquisition timing, and obtain a set of semantic scene switching rules. Each rule contains a description of the triggering conditions for scene startup.

[0024] The scene switching initiation logic of the vehicle system is semantically translated, transforming it from technical language in the form of code logic and control commands into semantic rules that can be associated with the acquisition timing, forming a set of semantic scene switching rules. Each rule contains a description of the triggering conditions for scene initiation. For example, the scene switching initiation logic of the vehicle system is first analyzed. If the vehicle system is based on CAN bus communication, the scene switching logic is stored in the firmware code of the electronic control unit (ECU). Reverse engineering can be used to convert the binary code into assembly code with the help of disassemblers. Then, combined with system documentation and comments, the key conditions and logical relationships of scene switching can be analyzed. At the same time, parameter data related to scene switching, such as real-time sensor readings and actuator status information, are extracted from the ECU using the data acquisition interface as the basis for subsequent semantic translation. Then, the parsed scene switching conditions are broken down, decomposing complex logical expressions into simple atomic conditions. For example, the condition "Switch from Eco mode to Sport mode when engine speed exceeds 2000 rpm and vehicle speed exceeds 60 km / h" can be broken down into two atomic conditions: "engine speed exceeds 2000 rpm" and "vehicle speed exceeds 60 km / h". A formal representation is then constructed for each atomic condition, using predicate logic. For instance, if P1 represents "engine speed exceeds 2000 rpm" and P2 represents "vehicle speed exceeds 60 km / h", the original condition can be represented as P1∧P2. Semantic mapping is then performed to establish the mapping relationship between the atomic conditions and natural language descriptions. Using a predefined vocabulary and grammatical rules, the formal representation is converted into a natural language description. For P1, it can be mapped to "the value collected by the engine speed sensor exceeds 2000 rpm"; for P2, it can be mapped to "the value collected by the vehicle speed sensor exceeds 60 km / h". These natural language atomic conditions are then combined according to logical relationships to form complete semantic rules, ultimately resulting in an accurate and reliable set of semantic scene switching rules.

[0025] Step S140: Match the continuous acquisition time-domain correlation with the semantic scene switching rule set, establish the scene switching rule corresponding to each set of time-domain feature data, record the acquisition time period of each set of time-domain feature data and the matching content of the corresponding scene rule, and obtain the time-domain-scene correlation correspondence.

[0026] In one implementation, step S140 may specifically include the following steps S141 to S146:

[0027] Step S141: Decompose the temporal feature nodes of the continuous acquisition temporal correlation relationship, and split each group of temporal feature data into independent temporal feature nodes. Each temporal feature node corresponds to a group of temporal feature data and its acquisition sequence dependency relationship with adjacent nodes, resulting in multiple temporal feature node sequences with sequential correlation.

[0028] Temporal feature node decomposition further breaks down each set of temporal feature data in a continuously acquired temporal correlation into independent nodes. Each node represents a temporal feature data and its relationship with neighboring nodes. A temporal feature node is an independent unit containing all the information of a set of temporal feature data, as well as its temporal order and dependencies with neighboring nodes. The sequentially associated temporal feature node sequence is the result of arranging all the decomposed temporal feature nodes in chronological order of acquisition time, reflecting the temporal sequence and continuity of the data.

[0029] For example, for each set of time-domain feature data in a continuously acquired time-domain correlation, its individual features are separated to obtain independent time-domain feature nodes. For instance, engine speed, temperature, and oil pressure data for a time-domain segment can be split into three time-domain feature nodes. Simultaneously, the acquisition sequence dependency between each node and its adjacent nodes is recorded; for example, the engine speed node is acquired before the temperature node, and the temperature node is acquired before the oil pressure node. Finally, all time-domain feature nodes are arranged in chronological order of acquisition time to obtain a sequence of time-domain feature nodes with sequential correlation.

[0030] Step S142: Split the semantic scene switching rule set into scene nodes, and split each semantic scene switching rule into an independent scene node. Each scene node corresponds to a rule content that initiates scene switching and its initiation dependency association with adjacent rules, resulting in multiple scene node sequences with initiation dependency associations.

[0031] Scene node decomposition further breaks down each rule in the semantic scene switching rule set into independent nodes. Each node represents a scene switching rule and its relationship with adjacent rules. A scene node is an independent unit containing the content of a rule that initiates a scene switch and its initiation dependency associations with adjacent rules during the scene switch process. Initiation dependency associations refer to the sequence and mutual influence between different scene switching rules; for example, the initiation of one scene switching rule may depend on the completion of another. The sequence of scene nodes with initiation dependency associations is the result of arranging all the decomposed scene nodes in the initiation order, reflecting the temporal and logical nature of scene switching rules. For example, for each rule in the semantic scene switching rule set, its various conditions and actions are separated to obtain independent scene nodes. Simultaneously, the initiation dependency associations of each node with adjacent nodes are recorded. Finally, all scene nodes are arranged in the initiation order to obtain the sequence of scene nodes with initiation dependency associations.

[0032] Step S143: Associate and adapt each temporal feature node with each scene node, analyze whether the standardized data feature representation corresponding to each temporal feature node meets the start conditions corresponding to the scene node, record the adaptation results of each temporal feature node and scene node, and obtain the temporal feature-scene node adaptation sequence.

[0033] Association adaptation involves comparing and analyzing each temporal feature node with each scene node to determine whether the standardized data feature representation corresponding to the temporal feature node meets the activation conditions corresponding to the scene node. The temporal feature-scene node adaptation sequence is a set obtained by recording the adaptation results of each temporal feature node and scene node, containing the adaptation status of all temporal feature nodes and scene nodes. For example, the standardized data feature representation of each temporal feature node is compared one by one with the activation conditions of each scene node. If the conditions are met, the adaptation result of that temporal feature node and scene node is recorded as a match; otherwise, it is recorded as a mismatch. By comparing all temporal feature nodes and scene nodes, the temporal feature-scene node adaptation sequence is finally obtained.

[0034] Step S144: Perform continuous sequence verification on the temporal feature-scene node adaptation sequence. Verify whether each temporal feature-scene node adaptation pair conforms to the continuity consistency between the acquisition order and the startup order. Remove adaptation pairs that do not conform to the continuity consistency to obtain a continuous adaptation pair sequence that conforms to the acquisition order.

[0035] Continuous sequence verification checks the sequence of temporal feature-scene node adaptations to ensure that each adaptation has continuity and consistency in terms of acquisition time and scene startup order. The continuity and consistency of acquisition and startup order means that the acquisition time order of temporal feature nodes should be consistent with the startup order of scene nodes. For example, if one temporal feature node is acquired before another, then its corresponding scene node should also be started before that scene node. The continuous adaptation sequence that conforms to the acquisition order is the result obtained after verification, eliminating adaptations that do not conform to continuity and consistency, thus guaranteeing the temporal correctness of the association between data and scenes.

[0036] Step S145: Integrate the matching results of continuous adaptation pairs into the sequence, organize all adaptation pairs that meet the continuity consistency, record the unique scene switching rule corresponding to each group of time-domain feature data, and obtain the preliminary time-domain-scene association correspondence.

[0037] For example, for each fit pair in the continuous fit pair sequence, the scene switching rule corresponding to each set of time-domain feature data is determined based on the matching relationship between time-domain feature nodes and scene nodes. If a set of time-domain feature data corresponds to multiple scene switching rules, it is necessary to filter them according to certain priorities or rules to determine a unique scene switching rule. For example, when a set of time-domain feature data satisfies two scene switching rules simultaneously, the rule can be selected based on its importance or scope of application. Finally, the unique scene switching rule corresponding to each set of time-domain feature data is recorded to obtain a preliminary time-domain-scene association correspondence.

[0038] Step S146: Cross-validate the preliminary time-domain-scene association correspondence. Cross-compare the standardized data feature representation of each group of time-domain feature data with the content of the corresponding scene rules, adjust the incorrectly matched correspondence, and generate the final time-domain-scene association correspondence.

[0039] For example, for each set of time-domain feature data and corresponding scene rules in the initial time-domain-scene association correspondence, a detailed comparison is made between the standardized data feature representation of the time-domain feature data and the content of the scene rules. For instance, for a vehicle speed feature value of a time-domain feature data and its corresponding scene rule "vehicle speed greater than 60 km / h," the system checks whether the vehicle speed feature value is indeed greater than 60 km / h. If a mismatch is found, such as the time-domain feature data not meeting the conditions of the corresponding scene rule, the correspondence needs to be adjusted. Through cross-comparison and adjustment of all time-domain feature data and scene rules, an accurate and reliable final time-domain-scene association correspondence is generated.

[0040] Step S150: Based on the time-domain-scene association correspondence, the original running data is mapped and marked. Each set of time-domain feature data is mapped and marked one by one with its corresponding scene switching rule to generate a time-series scene association mapping body of the original running data. The mapping body contains all the corresponding entries of time-domain feature data and scene rules.

[0041] For example, based on the final time-domain-scene association correspondence, each time-domain feature data is labeled with its corresponding scene switching rule. For instance, if the scene rule corresponding to a time-domain feature data is "the vehicle switches from the start-up scene to the driving scene," then that scene rule is labeled on that time-domain feature data. By labeling all time-domain feature data, a time-series scene association mapping body of the original runtime data is finally generated.

[0042] Step S160: Identify and mark segments in the original running data that deviate from the time-domain-scene association correspondence, identify time-domain segments in the original running data that do not correspond to any scene switching rules, uniformly mark all time-domain segments, and generate an abnormal start time-series marker cluster. The abnormal start time-series marker cluster contains the time range information of all abnormal time-domain segments.

[0043] For example, by traversing the original operational data and comparing it with the established time-domain-scene association, when a time-domain segment of data is found to be unable to correspond to any scene switching rule, it is judged as an abnormal segment deviating from the normal relationship. For instance, in normal vehicle operation data, according to the normal scene switching logic, when the vehicle speed is within a certain range and the engine is in the corresponding working state, it corresponds to a certain scene. However, in a certain data segment, the combination of vehicle speed and engine data does not conform to any set scene switching rule. At this time, the time-domain segment corresponding to this data segment is an abnormal segment. Next, all identified abnormal time-domain segments are uniformly marked, and the start and end times of each abnormal segment are recorded. Specifically, a special data structure can be used to store this time range information, such as a list, where each element is a tuple containing the start and end times, thus representing the time range of each abnormal time-domain segment. Finally, the time range information of all marked abnormal time-domain segments is integrated to generate an abnormal start time sequence marker cluster.

[0044] Step S200: Based on the temporal scene association mapping body and the abnormal start temporal marker cluster, construct the fault association semantic topology, sort out the semantic association links related to the abnormal performance of the vehicle system in the original operation data, and generate a fault semantic association map and a scene-semantic association mapping table.

[0045] In one implementation, step S200 may specifically include the following steps S210 to S260:

[0046] Step S210: Associate and bind the temporal scene association mapping body and the anomaly start temporal marker cluster. Bind each marker fragment in the anomaly start temporal marker cluster to the temporal feature data in its corresponding temporal scene association mapping body. Record the scene rule content corresponding to each anomaly marker to obtain the mapping body after marker binding.

[0047] The tagged segments in the anomaly initiation time-series tag cluster represent the abnormal time periods in the original runtime data, while the temporal feature data in the time-series scene association mapping body has a correspondence with the scene rules. Through association binding, the anomaly segments are mapped to specific temporal feature data in the time-series scene association mapping body, thereby determining the scene rule corresponding to the anomaly segment. The tag-bound mapping body is the result obtained after the association binding operation, which integrates anomaly tag and scene rule information.

[0048] In actual execution, each marker segment in the anomaly initiation time-series marker cluster is traversed. Based on the time range information of the marker segment, the corresponding temporal feature data is found in the temporal scene association mapping body. For example, if the time range of the anomaly marker is a certain time period, the temporal feature data within that time period is searched in the temporal scene association mapping body. Then, the scene rule content corresponding to the temporal feature data is recorded, and the anomaly marker, temporal feature data, and scene rule content are associated and stored to obtain the marker-bound mapping body.

[0049] Step S220: Extract semantic features from the marked mapping body. Extract the semantic features corresponding to all temporal feature data related to the anomaly marked fragments from the marked mapping body to obtain multiple anomaly-related semantic feature fragments. Each anomaly-related semantic feature fragment contains a semantic description of the corresponding temporal feature.

[0050] For example, firstly, temporal feature data related to anomalous labeled segments is extracted from the mapped body after tag binding. Natural language processing and semantic analysis techniques are then used to process this temporal feature data. For instance, for engine speed temporal feature data, if its value is higher than the normal range, the semantic feature "engine speed too high" can be extracted. For vehicle speed temporal feature data, if its fluctuation is large, the semantic feature "vehicle speed unstable" can be extracted. These extracted semantic features are then organized to obtain a series of anomalous associated semantic feature segments, each segment corresponding to a temporal feature.

[0051] Step S230: Sequentially associate and sort out multiple abnormal association semantic feature fragments, connect all abnormal association semantic feature fragments end to end according to the collection time order, sort out the sequential connection relationship between each abnormal association semantic feature fragment and adjacent abnormal association semantic feature fragments, and obtain the preliminary semantic feature link.

[0052] In one implementation, step S230 may specifically include the following steps S231 to S236:

[0053] Step S231: Time stamp the collection of multiple abnormal association semantic feature segments, add a corresponding collection timestamp to each abnormal association semantic feature segment, record the start and end collection times of each segment, and obtain a set of abnormal association semantic feature segments with timestamps.

[0054] For example, based on the collection records of the original operational data, the collection time of the temporal feature data corresponding to each anomaly-related semantic feature segment is found. These collection times are recorded as the start and end collection times of the anomaly-related semantic feature segment, and a unique collection timestamp is added to them. For example, a string in the format of "year-month-day hour:minute:second" can be used to represent it. All anomaly-related semantic feature segments with timestamps are integrated together to obtain a set of anomaly-related semantic feature segments with timestamps. This set can be stored using a list or array, where each element in the list is a tuple containing the anomaly-related semantic feature segment and its corresponding time information.

[0055] Step S232: Sort the set of anomaly-related semantic feature fragments with timestamps in chronological order. Sort all anomaly-related semantic feature fragments in chronological order according to the order of their collection timestamps to obtain a sequence of semantic feature fragments arranged in chronological order.

[0056] For example, a sorting algorithm can be used to sort the set of timestamped anomaly-related semantic feature fragments. For instance, a quicksort algorithm can be used to compare and swap the positions of each fragment based on its timestamp, ultimately resulting in a sequence of semantic feature fragments arranged chronologically. After sorting, the first fragment in the sequence is the earliest collected anomaly-related semantic feature fragment, and the last fragment is the latest collected anomaly-related semantic feature fragment.

[0057] Step S233: Perform adjacent segment association identification on the sorted semantic feature segment sequence, analyze the connection relationship between the ending semantic feature of each semantic feature segment and the starting semantic feature of the next semantic feature segment, record the semantic feature continuity results of adjacent segments, and obtain the adjacent semantic feature segment association result set.

[0058] Adjacent segment association identification is the process of analyzing and judging the semantic relationships between adjacent segments in a sorted sequence of semantic feature segments. Each semantic feature segment contains a semantic description of specific temporal features. By analyzing the connection between its ending semantic feature and the starting semantic feature of the next segment, the continuity and correlation of anomalies at different stages can be understood. The semantic feature continuity result indicates the degree of semantic correlation between adjacent segments, such as whether there is a causal relationship or whether they are different manifestations of the same anomaly. The adjacent semantic feature segment association result set is the result obtained after adjacent segment association identification, which records the semantic feature continuity between all adjacent segments. For example, the sorted sequence of semantic feature segments is traversed sequentially, and for each pair of adjacent semantic feature segments, their ending and starting semantic features are analyzed. For example, if the semantic feature of the previous segment is "engine temperature too high" and the semantic feature of the next segment is "engine power decreases", it can be inferred that there may be a causal relationship between the two, that is, excessive engine temperature may lead to a decrease in engine power. Based on this analysis, the semantic feature continuity results of adjacent segments are recorded, for example, represented by "with continuity" or "without continuity", and all results are organized into a set of associated results of adjacent semantic feature segments.

[0059] Step S234: Mark the connection relationship of the adjacent semantic feature fragment association result set, add connection mark for adjacent fragments with semantic feature continuity, and add break mark for adjacent fragments without semantic feature continuity, to obtain a semantic feature fragment sequence with connection mark.

[0060] Connection labeling is the process of marking the connection between adjacent semantic feature segments based on the association result set of adjacent semantic feature segments. Connection labels are used to distinguish whether there is a semantic continuity relationship between adjacent segments. By adding labels, the structure between abnormally associated semantic feature segments can be displayed more intuitively. The semantic feature segment sequence with connection labels is the result obtained after connection labeling. It adds connection information between adjacent segments to the original sorted semantic feature segment sequence.

[0061] Step S235: Construct links for the semantic feature fragment sequence with connection markers. Logically connect the semantic feature fragments with connection markers in chronological order, retain the fragment connections with connection markers, and disconnect the fragment connections with disconnection markers to obtain multiple segmented semantic feature links.

[0062] Link construction is the process of building semantic feature links based on a sequence of semantic feature fragments marked with connectors. Logical connection according to time order involves combining the semantic feature fragments according to their acquisition time sequence and connectors to obtain a coherent link. Retaining connections with connectors means linking adjacent fragments with semantic continuity together; breaking connections with break markers means not directly connecting fragments without semantic continuity. Multiple segmented semantic feature links are the result of link construction; each link represents a local process of an anomaly's development, and these links help to analyze the development path of an anomaly in more detail.

[0063] Step S236: Merge multiple segmented semantic feature links, merge segmented semantic feature links with consistent semantic feature continuity, sort out the complete connection relationship of all segments, and generate preliminary semantic feature links. The preliminary semantic feature links contain the sequential connection order of all abnormally related semantic feature segments.

[0064] For example, multiple segmented semantic feature links are analyzed one by one to determine their semantic relationships. For instance, the segmented links "engine temperature too high - engine power decreases" and "engine power decreases - vehicle speed decreases" can be merged into "engine temperature too high - engine power decreases - vehicle speed decreases" because they have semantic continuity and a common semantic node "engine power decreases". By merging and adjusting all segmented semantic feature links, a preliminary semantic feature link containing the sequential connection order of all abnormally related semantic feature segments is finally generated. For example, a graph structure or a linked list structure can be used to represent the preliminary semantic feature link.

[0065] Step S240: Redundant segments are removed from the preliminary semantic feature links. Semantic feature segments that appear repeatedly or have no effect on semantic connection are identified and removed from the preliminary semantic feature links to obtain simplified semantic feature association links.

[0066] Specifically, each semantic feature segment in the initial semantic feature chain can be traversed to check whether it is repeated or has any effect on semantic connections. For repeated segments, the judgment can be made by recording the information of the segments that have appeared, and when the same segment is encountered again, it is marked as redundant. For segments that have no effect on semantic connections, the judgment is made by analyzing their semantic relationship with adjacent segments. If there is no logical connection between the segment and the segments before and after it, it is marked as redundant. Then, all segments marked as redundant are removed from the initial semantic feature chain to obtain the simplified semantic feature association chain. The simplified semantic feature association chain is stored in a list, and redundant segments are directly skipped during the traversal.

[0067] Step S250: Construct topology nodes for the simplified semantic feature association links, split each semantic feature fragment in the semantic feature association links into independent topology nodes, and each topology node corresponds to a semantic feature content related to the anomaly, generating a topology node sequence.

[0068] For example, each semantic feature fragment in the simplified semantic feature association link is encapsulated as an independent topology node. Each topology node contains two important pieces of information: the semantic feature content corresponding to the node and the node's position information (i.e., its order) in the link. For example, for the semantic feature association link "engine temperature too high - engine power decreases", two topology nodes can be constructed, namely "engine temperature too high" and "engine power decreases", and their order can be recorded. Arranging all the constructed topology nodes in order yields a topology node sequence, which can be stored using a list or array, where each element in the list is a topology node object.

[0069] Step S260: Label the association relationships of the topology node sequence, label the semantic feature connection relationship between each topology node and other topology nodes, and record the scene information corresponding to each topology node to generate a fault semantic association graph and a scene-semantic association mapping table.

[0070] Semantic feature connections indicate the logical relationships between the semantic features represented by different topological nodes, such as causal relationships and sequential relationships. In practical applications, the sequence of topological nodes can be traversed, and for each topological node, its semantic connections with other nodes can be analyzed. These connections can be determined based on the sequential order and semantic logic in the semantic feature association links. For example, if the semantic feature represented by one topological node is the cause of the semantic feature represented by another topological node, then a causal relationship can be marked between them. Simultaneously, based on the mapped body bound by the labels in the previous steps, the scene rule content corresponding to each topological node is found and recorded. When constructing the fault semantic association graph, graph theory methods can be used, with topological nodes as vertices and semantic feature connections as edges, and the type of connection relationship is labeled on the edges. For the scene-semantic association mapping table, a tabular format can be used, where rows represent topological nodes and columns represent scene information. The association mapping between topological nodes and scene information is achieved by filling in the elements in the table.

[0071] Step S300: Import the fault semantic association graph into the preset fault pattern semantic matching graph, perform qualitative clustering of fault patterns, match the fault pattern semantic entries in the fault semantic association graph and the fault pattern semantic matching graph, and generate a fault pattern classification tag set and a fault-scenario association binding cluster.

[0072] In one implementation, step S300 may specifically include the following steps S310~S360:

[0073] Step S310: Align the fault semantic association graph with the preset fault mode semantic matching graph. Align each topological node in the fault semantic association graph with the corresponding semantic entry in the fault mode semantic matching graph to ensure that the semantic dimensions of the two are consistent, and obtain the aligned graph pair.

[0074] For example, for each topological node in the fault semantic association graph, based on its semantic features, the semantic entry that is semantically closest to it is found in the pre-defined fault pattern semantic matching graph. For instance, a semantic similarity calculation algorithm (such as cosine similarity algorithm) is used to calculate the similarity between the semantic features of the topological nodes in the fault semantic association graph and the semantic entries in the pre-defined graph, and the entry with the highest similarity is selected as the corresponding entry. Each topological node and its corresponding semantic entry are then aligned to maintain semantic consistency. Finally, the aligned fault semantic association graph and the pre-defined fault pattern semantic matching graph are combined into a graph pair. This pair can be stored using a data structure such as a tuple, where the first element of the tuple is the fault semantic association graph and the second element is the pre-defined fault pattern semantic matching graph.

[0075] Step S320: Extract semantic feature entries from the aligned graph pairs. Extract the semantic feature content corresponding to each topological node in the fault semantic association graph and organize it to obtain independent fault semantic feature entries. Extract the content of each semantic entry in the fault pattern semantic matching graph to obtain a set of fault semantic feature entries and a set of pattern semantic entries.

[0076] For example, the fault semantic association graph in the aligned graph pair is traversed, and for each topological node, its semantic feature content is extracted and stored as an independent fault semantic feature entry. For example, for the topological node "engine power decrease", "engine power decrease" is treated as an independent fault semantic feature entry. Similarly, the preset fault pattern semantic matching graph is traversed, and the content of each semantic entry is extracted and stored in a pattern semantic entry set. All extracted fault semantic feature entries are organized into a fault semantic feature entry set. A list can be used to store these two sets, where each element in the list is either a fault semantic feature entry or a pattern semantic entry.

[0077] Step S330: Perform content matching between the set of fault semantic feature entries and the set of pattern semantic entries. Compare the content of each fault semantic feature entry with the content of each pattern semantic entry one by one, record the content matching status of each fault semantic feature entry and the pattern semantic entry, and obtain the entry matching result set.

[0078] As one implementation method, step S330 may specifically include the following steps S331 to S336:

[0079] Step S331: Perform semantic feature segmentation on the set of fault semantic feature entries, splitting each fault semantic feature entry into multiple feature sub-segments with independent semantic features. Each feature sub-segment corresponds to a core feature unit in the fault semantic feature entry, resulting in a set of multiple fault semantic feature sub-segments.

[0080] For example, word segmentation and semantic analysis techniques from natural language processing can be used to process each entry in the set of fault semantic feature entries. Based on semantic logic and grammatical rules, the entry is split into multiple feature sub-fragments. For instance, regular expressions and part-of-speech tagging can be used to identify key semantic elements in the entry and treat them as feature sub-fragments. The feature sub-fragments from each fault semantic feature entry are then organized into a set, resulting in multiple sets of fault semantic feature sub-fragments. These sets can be stored using a list of lists, where each element of the outer list represents a set of feature sub-fragments corresponding to a fault semantic feature entry.

[0081] Step S332: Perform semantic segmentation on the pattern semantic entry set, splitting each pattern semantic entry into multiple sub-segments with independent semantics. Each sub-segment corresponds to a core content unit in the pattern semantic entry, resulting in multiple pattern semantic sub-segment sets.

[0082] This step can employ natural language processing techniques similar to those in step S331 to split each entry in the pattern semantic entry set. Based on semantic and grammatical rules, the entries are decomposed into sub-segments with independent semantics. The sub-segments of each pattern semantic entry are then organized into a set, ultimately resulting in multiple sets of pattern semantic sub-segments. These sets can be stored using a list of lists, with the same structure as the method used to store the fault semantic feature sub-segment set.

[0083] Step S333: Perform content matching between the set of fault semantic feature sub-fragments and the set of pattern semantic sub-fragments. Compare the content of each fault semantic feature sub-fragment with the content of each pattern semantic sub-fragment one by one, record the content matching status of each sub-fragment, and obtain the sub-fragment matching result set.

[0084] Sub-fragment content matching involves a finer-grained comparative analysis of fault semantic feature sub-fragments and pattern semantic sub-fragments to determine their degree of matching. For example, nested loops are used to iterate through the sets of fault semantic feature sub-fragments and pattern semantic sub-fragments. For each pair of sub-fragments, semantic similarity is calculated to determine if they match. For instance, a cosine similarity algorithm can be used to calculate the similarity between the vector representations of the sub-fragment content; a similarity score above a certain threshold indicates a match, otherwise, a mismatch. The matching status of each pair of sub-fragments is recorded, such as using a dictionary to store the sub-fragment matching result set, where the keys are the combination identifiers of the sub-fragment pairs, and the values ​​are the matching status.

[0085] Step S334: Perform matching result statistics on the sub-segment matching result set, count the number of matches of all corresponding feature sub-segments for each fault semantic feature entry, record the total number of feature sub-segments matched for each fault semantic feature entry, and obtain the entry matching statistics result set.

[0086] For example, the sub-fragment matching result set can be traversed. For each fault semantic feature entry, the set of feature sub-fragments can be used to count the number of sub-fragments that match the pattern semantic sub-fragments. Alternatively, the sub-fragment pairs related to the fault semantic feature entry can be filtered out by traversing the sub-fragment matching result set, and then the number of matching sub-fragment pairs can be counted. The total number of feature sub-fragment matches for each fault semantic feature entry can be recorded to obtain the entry matching statistics result set. For example, this result set can be stored using a dictionary, where the keys are fault semantic feature entries and the values ​​are the total number of their feature sub-fragment matches.

[0087] Step S335: Perform matching judgment on the item matching statistics result set. Determine the fault semantic feature item whose total number of matching feature sub-fragments reaches the total number of feature sub-fragments of the corresponding item as matching the content of the corresponding pattern semantic item; otherwise, determine it as not matching, and obtain the item matching judgment result set.

[0088] As one implementation method, step S335 may specifically include the following steps S3351~S3356:

[0089] Step S3351: Extract the number of sub-fragments from the item matching statistics result set. Extract the total number of feature sub-fragments contained in each fault semantic feature item, and extract the total number of sub-fragments contained in each pattern semantic item, to obtain the sub-fragment count record set for each item.

[0090] For example, for each fault semantic feature entry in the entry matching statistics result set, the number of feature sub-fragments it contains is counted based on the fault semantic feature sub-fragments obtained in the previous steps. Similarly, for each pattern semantic entry, the number of its sub-fragments is counted based on the pattern semantic sub-fragments obtained in the previous steps. The number of sub-fragments for each fault semantic feature entry and pattern semantic entry is recorded to obtain a sub-fragment count record set for each entry. This record set can be stored using a dictionary, where the key is the entry identifier (fault semantic feature entry or pattern semantic entry), and the value is the number of sub-fragments.

[0091] Step S3352: Perform a quantity consistency check on the sub-fragment quantity record set, compare the total number of sub-fragments of each fault semantic feature entry with the corresponding pattern semantic entry to ensure that the number of sub-fragments of the two are consistent, remove the entry pairs with inconsistent quantities, and obtain a set of entry pairs with consistent quantities.

[0092] Specifically, the set of sub-fragment count records can be traversed. For each fault semantic feature entry, its corresponding pattern semantic entry is found, and their total sub-fragments are compared. If the total sub-fragments of the two entries do not match, the entry pair is removed from consideration; if they match, the entry pair is retained. All entry pairs with the same total sub-fragment count are organized into a set of entry pairs with the same count. This set can be stored using a list, where each element of the list is a tuple containing a fault semantic feature entry and its corresponding pattern semantic entry.

[0093] Step S3353: Set a matching threshold for the set of entries with the same number of entries. Set the threshold for the total number of matching feature sub-fragments of each entry pair to the total number of feature sub-fragments of the corresponding entry. Record the matching threshold for each entry pair to obtain the set of entry matching thresholds.

[0094] Specifically, for each item pair in the set of identical item pairs, the threshold for the total number of matching feature sub-fragments is set to the total number of feature sub-fragments of the corresponding item. For example, if a fault semantic feature item contains 3 feature sub-fragments, and its corresponding pattern semantic item also contains 3 sub-fragments, then the matching threshold for that item pair is set to 3. The matching threshold for each item pair is recorded to obtain a set of item matching thresholds. This set can be stored using a dictionary, where the key is the combination identifier of the item pair, and the value is the matching threshold.

[0095] Step S3354: Compare the item matching statistics result set with the item matching threshold set. Compare the total number of matching sub-fragments of each fault semantic feature item with its corresponding matching threshold. Record the comparison result of the total number of matching for each item with the threshold to obtain the threshold comparison result set.

[0096] For example, the item matching statistics result set is traversed. For each fault semantic feature item, its corresponding matching threshold in the item matching threshold set is found. The total number of matching sub-fragments of the fault semantic feature item is compared with the matching threshold, and the comparison result is recorded. For example, if the total number of matches equals the threshold, it is recorded as "equal to"; if the total number of matches is greater than the threshold, it is recorded as "greater than"; if the total number of matches is less than the threshold, it is recorded as "less than". The comparison results of all items are compiled into a threshold comparison result set. This result set can be stored using a dictionary, where the keys are fault semantic feature items and the values ​​are the comparison results of the total number of matches and the threshold.

[0097] Step S3355: Perform matching judgment on the threshold comparison result set, determine the fault semantic feature entries whose total number of matches equals the matching threshold as matching with the corresponding pattern semantic entries, and determine the entries whose total number of matches does not equal the matching threshold as not matching, thus obtaining the preliminary entry matching judgment result set.

[0098] When the total number of matching sub-fragments of a fault semantic feature entry equals its corresponding matching threshold, it means that all feature sub-fragments after the entry is split match the pattern semantic sub-fragments, and therefore it is determined to be a match; conversely, if the total number of matches does not equal the matching threshold, it is determined to be a mismatch. The preliminary entry matching determination result set is a preliminary result obtained after the matching determination, which records the matching status of each fault semantic feature entry with the corresponding pattern semantic entry, but further verification is still needed.

[0099] For example, the threshold comparison result set is traversed. For each fault semantic feature entry, a judgment is made based on the comparison result of its total number of matches with the threshold. If the result is "equal to", the entry is judged as matching the content of the corresponding pattern semantic entry and recorded as "match"; if the result is "greater than" or "less than", it is judged as not matching and recorded as "not matching". All judgment results are compiled into a preliminary entry matching judgment result set.

[0100] Step S3356: Verify the preliminary item matching result set, compare the contents of the item pairs that are judged to be matched again, confirm the consistency of the contents, adjust the results of the judgment errors, and generate the final item matching result set.

[0101] For example, for item pairs initially determined to be matched in the item matching result set, a detailed comparison is then performed on the content of their fault semantic feature items and pattern semantic items. More precise semantic matching algorithms, such as semantic similarity calculation combined with semantic logic analysis, can be used to confirm their consistency. If inconsistencies are found, the determination result for that item pair is adjusted to "not matched." All the verified and adjusted determination results are then compiled into the final item matching result set.

[0102] Step S336: Organize the results of the item matching judgment set, pair all the fault semantic feature items that are judged to be matched with the pattern semantic items, record the content and matching status of each paired item, and generate the item matching result set.

[0103] For example, iterate through the final item matching result set and filter out the fault semantic feature items that are judged as "matches". For each such item, find its corresponding pattern semantic item and pair them together. Record the content of each paired item, that is, the specific content of the fault semantic feature item and the pattern semantic item, as well as the matching status (such as the degree of matching, the basis for matching, etc.). Organize all the pairing information into an item matching result set.

[0104] Step S334: Filter the matching results of the item matching result set. Filter the fault semantic feature items and pattern semantic items that match the content completely, and remove the item pairs that do not match the content to obtain the set of item pairs that match the content.

[0105] For example, for each pair of entries in the matching result set, a more stringent semantic matching criterion is used to determine whether they are content-complete matches. Semantic understanding and syntactic analysis techniques from natural language processing can be combined to perform in-depth comparisons of the content of the entry pairs. For example, each word, grammatical structure, and semantic logic in the entry pair is checked for consistency. If differences are found in the content of the entry pair, it is removed from consideration; if the content is completely identical, the entry pair is retained. All content-completely matching entry pairs are then compiled into a set of content-matching entry pairs.

[0106] Step S335: Perform qualitative clustering on the set of entries that match the content, group the entries with the same content into groups, generate a generalized label for each group, and record all fault semantic feature entries and pattern semantic entries contained in each group to obtain a fault pattern classification label set.

[0107] Qualitative clustering is the process of grouping item pairs in a set of content-matching item pairs according to content similarity. Grouping item pairs with the same content into one group can integrate similar fault semantics and fault patterns, facilitating fault classification and identification. For example, for each item pair in the set of content-matching item pairs, its content is compared with the content of other item pairs. If the content is found to be the same, they are grouped into the same group. For example, if multiple item pairs contain fault semantic feature items and pattern semantic items related to "engine overheating fault," they can be grouped into the "engine overheating fault" group. A summary label is generated for each group, such as determining the label based on the main semantic information of the item pairs within the group. All fault semantic feature items and pattern semantic items contained in each group are recorded and organized into a fault pattern classification label set.

[0108] Step S336: Associate and bind the fault mode classification tag set with the scene-semantic association mapping table, bind each fault mode classification tag with its corresponding scene information, record all scene content corresponding to each tag, and generate a fault-scene association binding cluster.

[0109] Specifically, for each fault mode classification label in the fault mode classification label set, the scene information related to it is searched in the scene-semantic association mapping table. This search can be performed by examining the association between fault semantic feature entries and pattern semantic entries and scene information. For example, if a fault semantic feature entry corresponding to a fault mode classification label matches a semantic feature in a certain scene, then that scene information is bound to that fault mode classification label, recording all scene content corresponding to each label, and organizing this information into a fault-scene association binding cluster.

[0110] Step S400: Based on the fault mode classification label set and the fault-scenario association binding cluster, perform fault root cause topological reasoning, construct a logical association topological chain between the explicit semantic representation of the fault and the potential causes, and generate a fault root cause reasoning topological chain and a fault propagation topological path set.

[0111] In one implementation, step S400 may specifically include the following steps S410 to S460:

[0112] Step S410: Merge the fault mode classification tag set and the fault-scene association binding cluster. Merge each fault mode classification tag with the scene information and related fault semantic feature entries in its corresponding fault-scene association binding cluster to obtain the fused content containing fault tags, scene information and related semantic features, and obtain the fused association body.

[0113] Association fusion is the process of integrating information from the fault mode classification tag set and the fault-scenario association binding cluster. By fusing each fault mode classification tag with its corresponding scenario information and related fault semantic feature entries, a more comprehensive understanding of the specific circumstances and related characteristics of each fault mode in different scenarios can be obtained. For example, the fault mode classification tag set is traversed, and for each fault mode classification tag, its corresponding scenario information in the fault-scenario association binding cluster is found. Simultaneously, based on the previous entry matching results, fault semantic feature entries related to that fault mode classification tag are found. This information is then integrated to obtain a fused content containing fault tags, scenario information, and related semantic features. For example, for the fault mode classification tag "engine overheating fault," its corresponding scenario information might be "prolonged high-speed driving," and the related fault semantic feature entry might be "engine temperature too high." These are combined, and all the fused content is organized into a fused association.

[0114] Step S420: Extract explicit semantic representations from the fused associated entities. Extract semantic content that directly reflects the fault performance from the fused associated entities, record the explicit fault performance descriptions corresponding to each fault label, and obtain a set of explicit fault semantic representations.

[0115] For example, for each fused content in the fused association, the fault semantic feature entries and scene information are analyzed to extract the semantic content that directly reflects the fault performance. For instance, for the fused content containing the fault label "engine overheating fault", the scene information "long-term high-speed driving", and the fault semantic feature entry "engine temperature too high", "engine temperature too high" is extracted as the explicit fault performance description. The explicit fault performance descriptions corresponding to each fault label are recorded and organized into a set of explicit fault semantic representations.

[0116] Step S430: Perform potential cause semantic feature mining on the fused association, extract indirect semantic features that may cause fault manifestations from the fused association, record the potential cause semantic feature descriptions corresponding to each fault label, and obtain a set of potential cause semantic features.

[0117] In one implementation, step S430 may specifically include the following steps S431 to S436:

[0118] Step S431: Perform causal semantic feature identification on the fused associated body, identify indirect semantic feature fragments that may cause fault manifestations from the fused associated body, record the content of each semantic feature fragment and the corresponding fault label, and obtain multiple causal candidate semantic feature sets.

[0119] Cause semantic feature identification is the process of searching for indirect semantic information that may lead to fault manifestations within the fused association body. Indirect semantic feature fragments that may trigger fault manifestations are those parts of the fused association body that are potentially related to the occurrence of the fault but do not directly reflect the fault phenomenon, such as "component aging" or "excessively high ambient temperature." Recording the content of each semantic feature fragment and its corresponding fault label clarifies the correspondence between each potential cause and the fault mode. Multiple sets of candidate cause semantic features are the results obtained after cause semantic feature identification; each set corresponds to a fault label and contains possible candidate cause semantic feature fragments for that fault label.

[0120] For example, the fused associated entities are traversed. For each fused content, possible indirect semantic feature fragments are identified based on the fault semantic feature entries, scenario information, and relevant professional knowledge. For instance, for the fused content of "engine overheating fault," combined with the scenario information of "prolonged high-speed driving" and the fault semantic feature entry of "engine temperature too high," possible causal semantic feature fragments such as "cooling system blockage" and "engine oil deterioration" are identified. Each semantic feature fragment and its corresponding fault label are recorded and organized into multiple sets of causal candidate semantic features.

[0121] Step S432: Perform semantic feature backtracking for each candidate semantic feature of the cause. Backtrack from the collection time point corresponding to the candidate semantic feature of the cause to determine the relevant semantic feature content fragments before that time point, record the semantic feature fragments obtained by backtracking, and obtain the set of backtracked semantic feature fragments.

[0122] For example, for each candidate semantic feature of a cause, based on its collection time point, related semantic feature content fragments are searched backward in the original operating data or relevant historical records. For instance, for the candidate semantic feature "heat dissipation system blockage," its collection time point is a certain moment, and tracing back may find related semantic feature content fragments such as "abnormal cooling fan speed" and "coolant temperature gradually increases" in the period before that moment. These backtracked semantic feature fragments are recorded and organized to obtain a set of backtracked semantic feature fragments.

[0123] Step S433: Perform association matching on the set of retrospective semantic feature fragments, and match the content of the retrospective semantic feature fragments with the content of the candidate semantic feature fragments of the cause to determine the retrospective semantic feature fragments related to the content of the candidate semantic feature fragments of the cause, and obtain the set of associated retrospective semantic feature fragments.

[0124] For example, for each retrospective semantic feature fragment in the set of retrospective semantic feature fragments, its content is compared with the content of the candidate semantic feature fragments for the cause. Semantic matching algorithms in natural language processing, such as the cosine similarity algorithm, can be used to determine their relevance. If the similarity is higher than a certain set threshold, the retrospective semantic feature fragment is considered relevant to the candidate semantic feature fragment for the cause and is retained; otherwise, it is discarded. All relevant retrospective semantic feature fragments are then organized into a set of associated retrospective semantic feature fragments.

[0125] Step S434: Sequentially sort out the set of related retrospective semantic feature fragments, arrange all related retrospective semantic feature fragments according to the order of collection time, sort out the sequential connection relationship between each related retrospective semantic feature fragment and the candidate semantic feature fragment of the cause, and obtain the sequential retrospective sequence.

[0126] In one implementation, step S434 may specifically include the following steps S4341 to S4346:

[0127] Step S4341: Extract time stamps from the set of associated backtracking semantic feature fragments, add a corresponding collection timestamp to each associated backtracking semantic feature fragment, record the start and end collection times of each fragment, and obtain a set of associated backtracking semantic feature fragments with timestamps.

[0128] In practical applications, for each associated backtracking semantic feature fragment in the associated backtracking semantic feature fragment set, its corresponding acquisition time information is retrieved. This can be obtained through the recording of the original running data or related timestamps. An acquisition timestamp is added to each fragment, and its start and end acquisition times are recorded. For example, for an associated backtracking semantic feature fragment "abnormal cooling fan speed," its acquisition timestamp might be "2024-01-01 10:30:00," the start acquisition time might be "2024-01-01 10:25:00," and the end acquisition time might be "2024-01-01 10:35:00." All such time-stamped associated backtracking semantic feature fragments are then compiled into a time-stamped associated backtracking semantic feature fragment set.

[0129] Step S4342: Sort the set of associated retrospective semantic feature fragments with timestamps by time. Sort all associated retrospective semantic feature fragments according to the order of their collection timestamps to obtain a sequence of retrospective semantic feature fragments arranged in chronological order.

[0130] For example, a sorting algorithm can be used to sort the set of timestamped associated retrospective semantic feature fragments. For instance, a bubble sort algorithm can be used to compare and swap the positions of each fragment based on its timestamp, ultimately resulting in a sequence of retrospective semantic feature fragments arranged chronologically. After sorting, the first fragment in the sequence is the earliest collected associated retrospective semantic feature fragment, and the last fragment is the latest collected associated retrospective semantic feature fragment.

[0131] Step S4343: Perform adjacent segment association identification on the sorted backtracking semantic feature segment sequence, analyze the connection relationship between the ending semantic feature content of each backtracking semantic feature segment and the starting semantic feature content of the next backtracking semantic feature segment, record the semantic feature content continuity results of adjacent segments, and obtain the adjacent backtracking semantic feature segment association result set.

[0132] For example, by sequentially traversing the sorted sequence of backtracking semantic feature fragments, for each pair of adjacent backtracking semantic feature fragments, their ending and beginning semantic feature contents are analyzed. For instance, if the semantic feature of the preceding fragment is "cooling fan speed gradually decreases" and the semantic feature of the following fragment is "coolant temperature begins to rise," it can be inferred that there may be a causal relationship between the two, i.e., the decrease in cooling fan speed may lead to an increase in coolant temperature. Based on this analysis, the continuity results of the semantic feature contents of adjacent fragments are recorded, for example, represented by "with continuity" or "without continuity," and all results are organized into a set of associated results for adjacent backtracking semantic feature fragments.

[0133] Step S4344: Mark the connection relationship of the adjacent retrospective semantic feature fragment association result set, add connection mark for adjacent fragments with semantic feature content continuity, and add break mark for adjacent fragments without semantic feature content continuity, to obtain the retrospective semantic feature fragment sequence with connection mark.

[0134] Connectivity marking is the process of labeling the connectivity between adjacent segments based on the association result set of adjacent backtracking semantic feature segments. Adding connectivity and disconnection markers can more intuitively show the semantic relationships between adjacent segments, facilitating the subsequent construction of backtracking semantic feature links. For example, by traversing the association result set of adjacent backtracking semantic feature segments, connectivity markers are added to adjacent segment pairs that show "continuity," for example, using the "-" symbol; and disconnection markers are added to adjacent segment pairs that show "no continuity," for example, using the "|" symbol. These markers are then added to the sorted sequence of backtracking semantic feature segments to obtain a sequence of backtracking semantic feature segments with connectivity markers.

[0135] Step S4345: Construct links for the sequence of backtracking semantic feature segments with connection markers. Logically connect the backtracking semantic feature segments with connection markers in chronological order, retain the segments with connection markers, and disconnect the segments with disconnect markers to obtain multiple segmented backtracking semantic feature links.

[0136] For example, based on the sequence of backtracking semantic feature segments with connector markers, starting from the first segment, adjacent segments are connected when a connector marker "-" is encountered, and the link is broken when a break marker "|" is encountered, resulting in a new segmented link. For instance, for the sequence "cooling fan speed gradually decreases - coolant temperature begins to rise | vehicle driving conditions worsen - engine load increases", two segmented backtracking semantic feature links can be constructed: "cooling fan speed gradually decreases - coolant temperature begins to rise" and "vehicle driving conditions worsen - engine load increases". All the constructed segmented backtracking semantic feature links are then combined to obtain multiple segmented backtracking semantic feature links.

[0137] Step S4346: Merge multiple segmented backtracking semantic feature links, merge segmented backtracking semantic feature links with consistent semantic feature content, sort out the complete connection relationship between all backtracking semantic feature segments and candidate cause semantic feature segments, and generate a sequential backtracking sequence. The sequential backtracking sequence contains the sequential connection order of all related backtracking semantic feature segments.

[0138] For example, multiple segmented backtracking semantic feature links are analyzed one by one to determine their semantic relationships. For instance, the segmented links "cooling fan speed gradually decreases - coolant temperature begins to rise" and "coolant temperature continues to rise - cooling system blockage" can be merged into "cooling fan speed gradually decreases - coolant temperature begins to rise - coolant temperature continues to rise - cooling system blockage" because they have semantic continuity and a common semantic node "coolant temperature rises". By merging and adjusting all segmented links, a sequential backtracking sequence containing the order in which all related backtracking semantic feature segments are connected is finally generated. For example, a graph structure or a linked list structure can be used to represent the sequential backtracking sequence.

[0139] Step S435: Perform segment verification on the sequential backtracking sequence to verify whether each semantic feature segment in the sequential backtracking sequence is indirectly related to the fault performance. Remove segments that are not indirectly related to the fault performance to obtain the effective cause tracing link.

[0140] Segment verification is the process of examining and evaluating each semantic feature segment in a sequential backtracking sequence to determine whether there is an indirect correlation between it and the fault manifestation. An indirect correlation with the fault manifestation means that although the semantic feature segment does not directly manifest as a fault phenomenon, it has a certain influence on the occurrence and development of the fault. For example, for each semantic feature segment in the sequential backtracking sequence, the relationship between it and the fault manifestation is analyzed by combining the set of explicit fault semantic representations and professional knowledge. For instance, for a semantic feature segment "unpleasant odor in the vehicle interior," if it has no obvious indirect correlation with the current fault manifestation "engine overheating," it is removed from the sequential backtracking sequence. However, for a semantic feature segment like "abnormal radiator fan speed," since it may affect engine cooling and has an indirect correlation with the "engine overheating" fault manifestation, the segment is retained. All semantic feature segments that have indirect correlation with the fault manifestation are organized into a valid cause-based tracing chain.

[0141] Step S436: Extract semantic features from the effective cause tracing link, extract the core potential cause semantic features from the effective cause tracing link, record the potential cause semantic feature description corresponding to each fault label, organize all potential cause semantic feature descriptions, and generate a potential cause semantic feature set.

[0142] The core semantic features of potential causes are the deep-seated reasons and factors that significantly influence the occurrence of faults, such as "aging cooling system" and "poor engine oil quality." Recording the semantic feature descriptions of potential causes corresponding to each fault label can clearly show the potential causes of each fault mode. The set of semantic features of potential causes is the result obtained after semantic feature extraction. It contains the semantic feature descriptions of potential causes corresponding to all fault labels, providing key data for constructing the fault root cause reasoning topology chain. For example, for each semantic feature fragment in the effective cause tracing link, its influence on fault performance is analyzed, and the core semantic features of potential causes are extracted. For example, for the effective cause tracing link "cooling fan speed gradually decreases - coolant temperature begins to rise - coolant temperature continues to rise - cooling system blockage", "cooling system blockage" is extracted as the core semantic feature of potential causes. For each fault label, the corresponding semantic feature description of potential causes is recorded according to its corresponding effective cause tracing link. All such descriptions are organized to generate the set of semantic features of potential causes.

[0143] Step S440: Logically associate the set of explicit semantic representations of faults with the set of semantic features of potential causes, and logically connect each explicit semantic representation of faults with the corresponding semantic features of potential causes to construct a topological node that reflects the logical relationship between faults and causes, thus obtaining a set of logically associated topological nodes.

[0144] For example, for each explicit fault semantic representation in the set of explicit fault semantic representations, find its corresponding potential cause semantic feature content. For instance, for the explicit fault semantic representation "engine temperature too high," its corresponding potential cause semantic feature content might be "cooling system blockage." Logically connect these to construct a topological node, which contains the explicit fault semantic representation, the potential cause semantic feature content, and the logical relationships (such as causal relationships) between them. Organize all such topological nodes into a logically related topological node set.

[0145] Step S450: Construct a topology chain for the set of logically related topology nodes. Connect all logically related topology nodes according to the logical relationship between faults and causes, sort out the complete logical path from potential causes to fault manifestations, and generate a logically related topology chain.

[0146] For example, based on the logical relationships contained in each topological node in the logically associated topological node set, potential cause nodes are connected to their corresponding fault manifestation nodes. For instance, for a topological node containing "cooling system blockage" (potential cause) and "engine overheating" (fault manifestation) with a causal relationship, the "cooling system blockage" node is connected to the "engine overheating" node with an edge. Integrating all such connections reveals the complete logical path from potential cause to fault manifestation. A graph structure can be used to represent this logically associated topological chain, where nodes are topological nodes and edges represent the logical relationships between them.

[0147] Step S460: Extract the path from the logical association topology chain, extract the complete logical path from the potential cause semantic features to the fault explicit semantic representation, as well as the fault propagation path in different scenarios, and generate the fault root cause reasoning topology chain and the fault propagation topology path set.

[0148] For example, for logically related topological chains, graph traversal algorithms, such as depth-first search, are used to find all complete logical paths leading to the nodes representing explicit fault semantics, starting from the potential cause semantic feature nodes. These paths are then organized into a fault root cause reasoning topological chain. Simultaneously, by combining scenario information from the fault-scenario association cluster, the possible propagation directions and methods of the fault under different scenarios are analyzed to determine the fault propagation paths. For example, in the "long-term high-speed driving" scenario, the "engine overheating" fault may propagate to other fault manifestations such as "vehicle power reduction." All such fault propagation paths are then organized into a fault propagation topological path set.

[0149] Step S500: Based on the fault root cause reasoning topology chain and fault propagation topology path set, perform fault risk trend topology deduction, analyze the evolution topology logic of fault causes and the expansion topology direction of propagation paths, and generate fault risk warning semantic package and risk propagation trend topology map.

[0150] In one implementation, step S500 may specifically include the following steps S510~S560:

[0151] Step S510: Perform topological fusion of the fault root cause reasoning topological chain and the fault propagation topological path set, and fuse the logical association content in the fault root cause reasoning topological chain with the propagation path content in the fault propagation topological path set to obtain a unified topological content containing root cause reasoning and propagation path, and obtain the fused topological body.

[0152] For example, for each logical association in the root cause reasoning topology chain, such as "cooling system blockage - engine overheating", and each propagation path in the fault propagation topology path set, such as "engine overheating - engine component damage", they are integrated. This fusion can be achieved using a graph structure, merging the nodes and edges of the root cause reasoning topology chain and the fault propagation topology path set into the same graph. If there are duplicate nodes or edges, appropriate handling is performed, such as merging identical nodes and adjusting edge weights. All the integrated content is then organized into a fused topology.

[0153] Step S520: Perform cause evolution logic analysis on the fused topology, analyze the logical relationship of the potential cause semantic features changing over time from the fused topology, record the evolution process of each cause semantic feature and the corresponding fault manifestation, and obtain the evolution topology logic of the fault cause.

[0154] As one implementation method, step S520 may specifically include the following steps S521 to S526:

[0155] Step S521: Extract the causal semantic feature nodes from the fused topology. Extract all topology nodes that embody potential causal semantic features from the fused topology. Record the semantic feature content of each node and the corresponding collection timestamp to obtain the causal semantic feature topology node set.

[0156] Semantic feature node extraction of potential causes is the process of selecting nodes representing the semantic features of potential causes from the fused topology and obtaining their key information. Topological nodes embodying the semantic features of potential causes are nodes in the fused topology related to the potential causes, containing specific semantic information about the potential causes. Recording the semantic feature content of each node and its corresponding collection timestamp allows for precise understanding of the state of the potential causes at different time points. The set of topological nodes embodying the semantic features of potential causes is the result after extraction, containing all topological nodes embodying the semantic features of potential causes and their related information. For example, traversing all topological nodes in the fused topology, nodes embodying the semantic features of potential causes are selected based on the semantic information and labels of the nodes. For instance, for a fused topology containing nodes such as "cooling system blockage" and "engine temperature too high," the potential cause semantic feature node "cooling system blockage" is selected. The semantic feature content of each node, such as "cooling system blockage," and its corresponding collection timestamp, such as "2024-01-01 10:30:00," are recorded. All such nodes and information are organized into a set of topological nodes representing causal semantic features.

[0157] Step S522: Sort the set of topological nodes of the causal semantic features by time. Sort all the topological nodes of the causal semantic features according to the order of the collection timestamps to obtain the sequence of causal semantic feature nodes arranged in time order.

[0158] For example, a sorting algorithm can be used to sort the set of topological nodes for causal semantic features. For instance, a quicksort algorithm can be used to compare and swap the positions of each node based on its acquisition timestamp, ultimately resulting in a sequence of causal semantic feature nodes arranged chronologically. After sorting, the first node in the sequence is the earliest acquired causal semantic feature topological node, and the last node is the latest acquired causal semantic feature topological node.

[0159] Step S523: Perform adjacent node association analysis on the sorted sequence of causal semantic feature nodes, analyze the evolution relationship between the content of each causal semantic feature node and the content of the next causal semantic feature node, record the changes in the semantic feature content of adjacent nodes, and obtain the association result set of adjacent causal semantic feature nodes.

[0160] Neighbor node association analysis is the process of analyzing and judging the semantic relationships between adjacent nodes in a sorted sequence of causal semantic feature nodes. Analyzing the evolution of the content of each causal semantic feature node with the content of the next causal semantic feature node can reveal the state changes and development trends of potential causal factors at different points in time. Recording the changes in the semantic feature content of adjacent nodes can clarify the evolution process and degree of potential causal factors. The adjacent causal semantic feature node association result set is the result obtained after adjacent node association analysis, which records the changes in the semantic feature content between all adjacent nodes. For example, the sorted sequence of causal semantic feature nodes is traversed sequentially, and for each pair of adjacent causal semantic feature nodes, their semantic feature content is analyzed. For example, if the semantic feature of the previous node is "slight blockage in the cooling system" and the semantic feature of the next node is "moderate blockage in the cooling system", it can be analyzed that the degree of blockage in the cooling system is worsening. The changes in the semantic feature content of adjacent nodes are recorded, for example, represented by "the degree of blockage is worsening" or "the degree of blockage remains unchanged", and all results are organized into the adjacent causal semantic feature node association result set.

[0161] Step S524: Mark the evolutionary relationship of the association result set of adjacent causal semantic feature nodes. Add evolutionary tags to adjacent nodes that have semantic feature content evolutionary relationships, and add independent tags to adjacent nodes that do not have semantic feature content evolutionary relationships, to obtain the sequence of causal semantic feature nodes with evolutionary tags.

[0162] For example, the association result set of adjacent causal semantic feature nodes is traversed. For adjacent node pairs where the results show an evolutionary relationship in semantic feature content, evolutionary markers are added to them, such as using "↑" to indicate increased severity and "↓" to indicate decreased severity. For adjacent node pairs where the results show no evolutionary relationship in semantic feature content, independent markers are added, such as using "=". These markers are added to the sorted sequence of causal semantic feature nodes to obtain a sequence of causal semantic feature nodes with evolutionary markers. This sequence can be represented as a list of strings, where each element is a causal semantic feature node and its evolutionary marker.

[0163] Step S525: Construct evolutionary links for the sequence of causal semantic feature nodes with evolutionary labels. Logically connect the causal semantic feature nodes with evolutionary labels in chronological order, retain the node connections with evolutionary labels, and disconnect the node connections with independent labels to obtain multiple segmented evolutionary semantic feature links.

[0164] For example, based on the sequence of causal semantic feature nodes with evolutionary markers, starting from the first node, adjacent nodes are connected when encountering evolutionary markers such as "↑" or "↓", and the link is broken when encountering an independent marker "=", resulting in a new segmented link. For example, for the sequence "Minor blockage in the cooling system ↑ Moderate blockage in the cooling system = Insufficient tire pressure ↑ Increased tire wear", two segmented evolutionary semantic feature links can be constructed: "Minor blockage in the cooling system ↑ Moderate blockage in the cooling system" and "Insufficient tire pressure ↑ Increased tire wear". All the constructed segmented evolutionary semantic feature links are then organized together to obtain multiple segmented evolutionary semantic feature links. These segmented links can be stored using a list, where each element in the list represents a segmented evolutionary semantic feature link.

[0165] Step S526: Merge multiple segmented evolutionary semantic feature links, merge segmented evolutionary semantic feature links with consistent semantic feature content evolutionary relationships, sort out the complete evolutionary relationship of all causal semantic feature nodes, record the evolutionary process of each causal semantic feature and the corresponding fault manifestation, and generate the evolutionary topology logic of the fault causation.

[0166] Consistent semantic feature evolution relationships indicate a certain semantic correlation between different segmented links. For example, they may represent different stages of the evolution of the same potential trigger. By merging these segmented links, the complete evolutionary relationship of all trigger semantic feature nodes can be identified, resulting in a more complete evolutionary trajectory of the potential trigger. Recording the evolutionary process of each trigger semantic feature and its corresponding fault manifestation clearly demonstrates the dynamic process of fault occurrence. The evolutionary topological logic of the fault trigger is the result obtained after merging, which presents the evolutionary rules of the potential trigger and its relationship with fault manifestation in the form of a topological structure. For example, multiple segmented evolutionary semantic feature links are analyzed one by one to determine their semantic correlation. For instance, for the segmented links "minor blockage in the cooling system ↑ moderate blockage in the cooling system" and "moderate blockage in the cooling system ↑ severe blockage in the cooling system," since they have semantic continuity and a common semantic node "moderate blockage in the cooling system," they can be merged into "minor blockage in the cooling system ↑ moderate blockage in the cooling system ↑ severe blockage in the cooling system." Simultaneously, by combining the information in the fused topology, the evolution process of each trigger semantic feature and its corresponding fault manifestation are recorded, such as "severe blockage of the cooling system" possibly corresponding to the fault manifestation of "engine overheating". Through merging and adjusting all segmented links, the evolutionary topology logic of the fault triggers is finally generated, which contains the complete evolutionary relationship of all trigger semantic feature nodes and their corresponding relationship with fault manifestations.

[0167] Step S530: Perform path expansion direction analysis on the fused topology, analyze the expansion direction of the fault propagation path in different scenarios from the fused topology, record the propagation path expansion content in each scenario, and obtain the expansion topology direction of the propagation path.

[0168] For example, by combining the fault propagation topology path set and the scene-semantic association mapping table information in the fused topology, the influence of different scene factors, such as ambient temperature, vehicle usage frequency, and driving habits, is considered. For each fault propagation path, possible expansion directions under different scenarios are analyzed. For example, in a high-temperature environment, an "engine overheating" fault may further propagate to "engine component damage," and the propagation speed may accelerate. The propagation path expansion content under each scenario is recorded, including the expanded nodes, edge weights (such as the probability of propagation), and other information. All such information is then organized into the expansion topology directions of the propagation path.

[0169] Step S540: Semantically package the evolutionary topological logic of the fault cause, organize the evolutionary logic content of the fault cause, and obtain a semantic content package that can completely describe the evolution process of the cause. Record the cause content and corresponding time nodes of each evolution stage to obtain a preliminary fault risk warning semantic package.

[0170] Specifically, for the evolutionary topological logic of fault causes, a graph traversal algorithm, such as breadth-first search, can be used to traverse the nodes and edges in the topological logic. For each node, its corresponding cause content and time node are recorded, for example, "the cooling system experienced slight blockage at 10:30:00 on 2024-01-01". For edges, the changes in the cause are described in natural language according to the evolutionary relationship they represent, such as "the blockage gradually worsened afterward". All such information is organized into a semantic content package. For example, "the cooling system experienced slight blockage at 10:30:00 on 2024-01-01, and the blockage gradually worsened, developing into moderate blockage at 12:00:00 on 2024-01-01". This semantic content package serves as a preliminary fault risk warning semantic package, such as storing this semantic package as a string, and connecting the descriptions of each evolutionary stage sequentially in chronological order.

[0171] Step S550: Improve the content of the preliminary fault risk warning semantic package, supplement the fault manifestation and scenario information corresponding to each evolution stage, and generate the fault risk warning semantic package.

[0172] Content enhancement is the process of further enriching the information in the initial fault risk warning semantic package, making it more comprehensively reflect the fault risk situation. Supplementing the fault manifestations and scenario information corresponding to each evolution stage allows relevant personnel to better understand the possible fault phenomena and corresponding scenario conditions at different inducing factor evolution stages. For example, the initial fault risk warning semantic package is supplemented by combining information from the fault root cause reasoning topology chain, the fault propagation topology path set, and the scenario-semantic association mapping table. For the inducing factor content of each evolution stage in the initial semantic package, the corresponding fault manifestations and scenario information are found. For example, for the inducing factor stage "a slight blockage occurred in the cooling system at 10:30:00 on 2026-01-01," according to the fault root cause reasoning topology chain, the possible corresponding fault manifestation at this time is "a slight increase in engine temperature," and combined with the scenario-semantic association mapping table, the scenario information might be "the vehicle is driving in congested urban traffic." This supplementary information is added to the initial fault risk warning semantic package to obtain a more complete description, such as "At 10:30:00 on 2026-01-01, the vehicle experienced a slight blockage in the cooling system while driving in urban congestion, resulting in a slight increase in engine temperature." In this manner, content is supplemented for each evolutionary stage of the initial semantic package, ultimately generating the fault risk warning semantic package. For example, string concatenation operations can be used to reasonably insert the supplementary fault manifestations and scenario information into the corresponding positions in the initial semantic package.

[0173] Step S560: Perform topology drawing on the expansion direction of the propagation path, present the content of the expansion direction of the fault propagation path in the form of a topology diagram, mark the scene content and path nodes corresponding to each expansion direction, and generate a risk propagation trend topology diagram.

[0174] For example, using tools such as Graphviz and Matplotlib, corresponding nodes and edges are created in the plotting tool based on the node and edge information in the expansion topology direction of the propagation path. Nodes represent fault manifestations, potential causes, or scenarios, while edges represent the expansion direction of the fault propagation path. Then, corresponding labels are added to each node and edge, including the node name (e.g., "engine overheating," "cooling system blockage," etc.) and the edge weight (e.g., the probability of propagation). For each expansion direction, the corresponding scenario content is labeled, for example, labeling "high temperature environment" on the edge connecting "engine overheating" and "engine component damage." Simultaneously, detailed labels are added to the path nodes, clarifying the role and status of each node in the fault propagation process. Finally, the layout and style of the graph are adjusted to make it clear and easy to read, generating a risk propagation trend topology map.

[0175] Based on the same inventive concept, this application also provides a data processing apparatus for implementing the above-described data processing method for vehicle-mounted testing equipment used in fault detection. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more data processing apparatus embodiments provided below can be found in the above-described limitations of the data processing method for vehicle-mounted testing equipment used in fault detection, and will not be repeated here.

[0176] In one embodiment, such as Figure 3 As shown, a data processing apparatus 300 is provided, comprising:

[0177] The data modeling module 310 is used to perform time-series scene association modeling on the raw operating data collected by the vehicle-mounted detection equipment. Combining the continuous acquisition time-domain association relationship of the raw operating data and the scene switching startup logic of the vehicle system, it generates a time-series scene association mapping body and anomaly start time-series marker cluster for the raw operating data.

[0178] The topology construction module 320 is used to construct the fault association semantic topology based on the temporal scene association mapping body and the abnormal start temporal marker cluster, sort out the semantic association links related to the abnormal performance of the vehicle system in the original operation data, and generate a fault semantic association map and a scene-semantic association mapping table.

[0179] The graph clustering module 330 is used to import the fault semantic association graph into the preset fault pattern semantic matching graph, perform qualitative clustering of fault patterns, match fault pattern semantic entries in the fault semantic association graph and the fault pattern semantic matching graph, and generate fault pattern classification tag set and fault-scene association binding cluster.

[0180] The root cause reasoning module 340 is used to perform fault root cause topological reasoning based on the fault mode classification label set and fault-scenario association binding cluster, construct the logical association topological chain between the explicit semantic representation of the fault and the potential causes, and generate the fault root cause reasoning topological chain and the fault propagation topological path set.

[0181] The fault inference module 350 is used to perform fault risk trend topology inference based on the fault root cause reasoning topology chain and fault propagation topology path set, analyze the evolutionary topology logic of fault causes and the extended topology direction of propagation path, and generate fault risk warning semantic package and risk propagation trend topology map.

[0182] Each module in the aforementioned data processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0183] In another embodiment, the present invention also provides a server, the internal structure of which can be shown in the figure below. Figure 4 As shown, the server includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The server's processor provides computing and control capabilities. The server's memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces of the data processing device are used for exchanging information between the processor and external devices. The communication interface of the data processing device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a data processing method for vehicle-mounted detection equipment applied to fault detection.

Claims

1. A data processing method for vehicle-mounted fault detection equipment, characterized in that, The method includes: The raw operating data collected by the vehicle-mounted detection equipment is modeled in a time-series scenario association. Combining the continuous acquisition time-domain association relationship of the raw operating data and the scenario switching startup logic of the vehicle system, a time-series scenario association mapping body and anomaly start time-series marker cluster of the raw operating data are generated. Based on the time-series scene association mapping body and the abnormal start time-series marker cluster, a fault association semantic topology is constructed, the semantic association links related to the abnormal performance of the vehicle system in the original operating data are sorted out, and a fault semantic association map and a scene-semantic association mapping table are generated. The fault semantic association graph is imported into a preset fault pattern semantic matching graph, and fault pattern qualitative clustering is performed. The fault semantic association graph and the fault pattern semantic matching graph are matched to generate a fault pattern classification tag set and a fault-scenario association binding cluster. Based on the fault mode classification tag set and the fault-scenario association binding cluster, fault root cause topological reasoning is performed to construct a logical association topological chain between the explicit semantic representation of the fault and the potential causes, and to generate a fault root cause reasoning topological chain and a fault propagation topological path set. Based on the fault root cause reasoning topology chain and the fault propagation topology path set, fault risk trend topology deduction is performed to analyze the evolutionary topology logic of fault causes and the expansion topology direction of propagation paths, and to generate fault risk warning semantic packages and risk propagation trend topology diagrams.

2. The method as described in claim 1, characterized in that, The process involves performing time-series scene association modeling on the raw operating data collected by the vehicle-mounted detection equipment. Combining the continuous acquisition time-domain correlation of the raw operating data with the scene switching and startup logic of the vehicle system, a time-series scene association mapping body and anomaly initiation time-series marker cluster for the raw operating data are generated, including: The raw operational data collected by the vehicle-mounted testing equipment is divided into continuous time-domain segments. According to the continuous time sequence of the collection, the raw operational data is divided into multiple time-domain segments with fixed time spans and no overlap. Unified feature extraction is performed on the multi-source heterogeneous raw data in each time-domain segment to generate standardized data feature representations of the corresponding time-domain segments. Multiple sets of time-domain feature data with continuous collection sequence are obtained, and each set of time-domain feature data contains standardized data feature representations of the corresponding time-domain segments. The acquisition time sequence correlation of multiple sets of time-domain feature data is sorted out. Each set of time-domain feature data is connected and correlated according to the acquisition sequence. The acquisition sequence dependency relationship between each set of time-domain feature data and adjacent time-domain feature data is obtained, and the continuous acquisition time-domain correlation relationship of the original running data is generated. The scene switching startup logic of the vehicle system is semantically translated, and the conditions for starting scene switching in the vehicle system are converted into semantic rules that can be associated with the collection time sequence, resulting in a set of semantic scene switching rules. Each rule contains a description of the triggering conditions for scene startup. The continuous acquisition time-domain correlation relationship is matched with the semantic scene switching rule set to establish the scene switching rule corresponding to each group of time-domain feature data. The acquisition time period of each group of time-domain feature data and the matching content of the corresponding scene rule are recorded to obtain the time-domain-scene correlation correspondence. Based on the time-domain-scene association correspondence, the original running data is mapped and marked. Each set of time-domain feature data is mapped and marked one by one with its corresponding scene switching rule, generating a time-series scene association mapping body of the original running data. The mapping body contains all the corresponding entries of time-domain feature data and scene rules. The segments in the original running data that deviate from the time-domain-scene association correspondence are identified and marked. Time-domain segments in the original running data that do not correspond to any scene switching rules are identified. All time-domain segments are uniformly marked to generate an abnormal start time-series marker cluster. The abnormal start time-series marker cluster contains the time range information of all abnormal time-domain segments.

3. The method as described in claim 2, characterized in that, The step involves associating and matching the continuously acquired temporal correlation with the semantic scene switching rule set to establish scene switching rules corresponding to each set of temporal feature data. This process records the matching content between the acquisition period of each set of temporal feature data and the corresponding scene rule, resulting in a temporal-scene correlation correspondence, including: The temporal correlation of continuous acquisition is split into temporal feature nodes. Each set of temporal feature data is split into independent temporal feature nodes. Each temporal feature node corresponds to a set of temporal feature data and its acquisition sequence dependency with adjacent nodes, resulting in multiple temporal feature node sequences with sequential correlation. The semantic scene switching rule set is split into scene nodes, and each semantic scene switching rule is split into an independent scene node. Each scene node corresponds to a rule content that initiates scene switching and its initiation dependency association with adjacent rules, resulting in multiple scene node sequences with initiation dependency associations. Each temporal feature node is associated and adapted with each scene node. The standardized data feature representation corresponding to each temporal feature node is analyzed to see if it meets the start-up conditions of the scene node. The adaptation results of each temporal feature node and scene node are recorded to obtain the temporal feature-scene node adaptation sequence. The temporal feature-scene node adaptation sequence is continuously verified to check whether each temporal feature-scene node adaptation conforms to the continuity consistency between the acquisition order and the startup order. Adaptation pairs that do not conform to the continuity consistency are removed to obtain a continuous adaptation sequence that conforms to the acquisition order. The matching results of continuous adaptation pairs are integrated, all adaptation pairs that meet the continuity consistency are sorted out, and the unique scene switching rule corresponding to each group of time-domain feature data is recorded to obtain the preliminary time-domain-scene association correspondence. Cross-validation of the preliminary time-domain-scene association correspondence is performed by cross-comparing the standardized data feature representation of each group of time-domain feature data with the content of the corresponding scene rules, adjusting the mismatched correspondence, and generating the final time-domain-scene association correspondence.

4. The method as described in claim 1, characterized in that, Based on the temporal scene association mapping body and the anomaly initiation temporal marker cluster, the fault association semantic topology is constructed, and the semantic association links related to the abnormal performance of the vehicle system in the original operating data are sorted out to generate a fault semantic association map and a scene-semantic association mapping table, including: The temporal scene association mapping body and the anomaly start temporal marker cluster are associated and bound together. Each marker fragment in the anomaly start temporal marker cluster is bound to the temporal feature data in its corresponding temporal scene association mapping body. The scene rule content corresponding to each anomaly marker is recorded to obtain the mapping body after marker binding. Semantic features are extracted from the marked mapping body. All semantic features corresponding to the temporal feature data related to the anomaly marked fragments are extracted from the marked mapping body to obtain multiple anomaly-related semantic feature fragments. Each anomaly-related semantic feature fragment contains a semantic description of the corresponding temporal feature. Multiple abnormal association semantic feature fragments are sequentially associated and sorted out. All abnormal association semantic feature fragments are connected end to end according to the collection time order. The sequential connection relationship between each abnormal association semantic feature fragment and adjacent abnormal association semantic feature fragments is sorted out to obtain the preliminary semantic feature link. Redundant segments are removed from the preliminary semantic feature links. Semantic feature segments that appear repeatedly or have no effect on semantic connection are identified and removed from the preliminary semantic feature links to obtain simplified semantic feature association links. Topology nodes are constructed on the simplified semantic feature association links. Each semantic feature fragment in the semantic feature association links is split into independent topology nodes. Each topology node corresponds to a semantic feature content related to the anomaly, and a topology node sequence is generated. The sequence of topological nodes is labeled with association relationships, and the semantic feature connection relationship between each topological node and other topological nodes is marked. At the same time, the scene information corresponding to each topological node is recorded, and the fault semantic association map and the scene-semantic association mapping table are generated.

5. The method as described in claim 4, characterized in that, The process involves sequentially associating and organizing multiple abnormally related semantic feature fragments, connecting all fragments end-to-end according to their acquisition time order. This process reveals the sequential connection relationships between each abnormally related semantic feature fragment and its adjacent fragments, resulting in a preliminary semantic feature chain, including: Multiple abnormal association semantic feature fragments are time-stamped during collection. A corresponding collection timestamp is added to each abnormal association semantic feature fragment, and the start and end collection times of each fragment are recorded to obtain a set of abnormal association semantic feature fragments with timestamps. The set of timestamped abnormal association semantic feature fragments is sorted chronologically, and all abnormal association semantic feature fragments are sorted according to the order of their collection timestamps to obtain a sequence of semantic feature fragments arranged chronologically. The sorted semantic feature fragment sequence is subjected to adjacent fragment association identification. The connection relationship between the ending semantic feature of each semantic feature fragment and the starting semantic feature of the next semantic feature fragment is analyzed. The semantic feature continuity results of adjacent fragments are recorded to obtain the adjacent semantic feature fragment association result set. The adjacent semantic feature fragment association result set is marked with connection relationship. Connection marks are added to adjacent fragments with semantic feature continuity, and disconnect marks are added to adjacent fragments without semantic feature continuity, so as to obtain a semantic feature fragment sequence with connection marks. Link construction is performed on the semantic feature segment sequence with connection markers. The semantic feature segments with connection markers are logically connected in chronological order. The segments with connection markers are retained, and the segments with disconnection markers are disconnected to obtain multiple segmented semantic feature links. The multiple segmented semantic feature links are merged, and the segmented semantic feature links with consistent semantic feature continuity are merged to obtain the complete connection relationship of all segments, generating the preliminary semantic feature links. The preliminary semantic feature links include the sequential connection order of all abnormally associated semantic feature segments.

6. The method as described in claim 1, characterized in that, The step of importing the fault semantic association graph into a preset fault pattern semantic matching graph, performing qualitative clustering of fault patterns, matching fault pattern semantic entries in the fault semantic association graph and the fault pattern semantic matching graph, and generating a fault pattern classification tag set and a fault-scenario association binding cluster includes: The fault semantic association graph is aligned with the preset fault mode semantic matching graph. Each topological node in the fault semantic association graph is aligned with the corresponding semantic entry in the fault mode semantic matching graph to ensure that the semantic dimensions of the two are consistent, thus obtaining an aligned graph pair. Semantic feature entries are extracted from the aligned graph pairs. The semantic feature content corresponding to each topological node in the fault semantic association graph is extracted and organized to obtain independent fault semantic feature entries. The content of each semantic entry in the fault pattern semantic matching graph is extracted to obtain a set of fault semantic feature entries and a set of pattern semantic entries. Content matching is performed on the set of fault semantic feature entries and the set of pattern semantic entries. The content of each fault semantic feature entry is compared with the content of each pattern semantic entry one by one. The content matching situation of each fault semantic feature entry and the pattern semantic entry is recorded to obtain the entry matching result set. The matching result set of the entries is filtered. The entries with fault semantic features that match the content completely are filtered out from the entries with pattern semantic features. The entries with mismatched content are removed to obtain the set of entries with matching content. Qualitative clustering is performed on the set of entries that match the content, and the entries with the same content are grouped together. A generalized label content is generated for each group, and all fault semantic feature entries and pattern semantic entries contained in each group are recorded to obtain the fault pattern classification label set. The fault mode classification tag set is associated and bound to the scene-semantic association mapping table. Each fault mode classification tag is bound to its corresponding scene information. All scene content corresponding to each tag is recorded to generate the fault-scene association binding cluster.

7. The method as described in claim 6, characterized in that, The content matching of the fault semantic feature item set and the pattern semantic item set involves comparing the content of each fault semantic feature item with that of each pattern semantic item one by one, recording the content matching status of each fault semantic feature item and the pattern semantic item, and obtaining an item matching result set, including: The set of fault semantic feature entries is split into semantic feature segments, and each fault semantic feature entry is split into multiple feature sub-segments with independent semantic features. Each feature sub-segment corresponds to a core feature unit in the fault semantic feature entry, resulting in a set of multiple fault semantic feature sub-segments. The set of pattern semantic entries is split into semantic segments, and each pattern semantic entry is split into multiple sub-segments with independent semantics. Each sub-segment corresponds to a core content unit in the pattern semantic entry, resulting in a set of multiple pattern semantic sub-segments. Content matching is performed on the set of fault semantic feature sub-fragments and the set of pattern semantic sub-fragments. The content of each fault semantic feature sub-fragment is compared with the content of each pattern semantic sub-fragment one by one, and the content matching status of each sub-fragment is recorded to obtain a sub-fragment matching result set. The matching results of the sub-segment matching result set are statistically analyzed. For each fault semantic feature entry, the number of matching of all corresponding feature sub-segments is counted, and the total number of matching of feature sub-segments for each fault semantic feature entry is recorded to obtain the entry matching statistical result set. The matching statistical result set of the entries is matched and determined. Fault semantic feature entries whose total number of matching feature sub-fragments reaches the total number of feature sub-fragments of the corresponding entry are determined to match the content of the corresponding pattern semantic entry; otherwise, they are determined to be mismatched, and the entry matching determination result set is obtained. The results of the item matching judgment set are sorted out, and all the fault semantic feature items that are judged to be matched are paired with the pattern semantic items. The content and matching status of each paired item are recorded to generate the item matching result set.

8. The method as described in claim 7, characterized in that, The matching determination of the item matching statistical result set involves determining that fault semantic feature items whose total number of matching feature sub-fragments reaches the total number of feature sub-fragments of the corresponding item are matched with the content of the corresponding pattern semantic item; otherwise, they are determined not to match, resulting in an item matching determination result set, including: The number of sub-fragments is extracted from the matching statistics result set of the entries. The total number of feature sub-fragments contained in each fault semantic feature entry is extracted, and the total number of sub-fragments contained in each pattern semantic entry is extracted, so as to obtain the sub-fragment number record set of each entry. Perform a consistency check on the number of sub-fragments in the record set. Compare the total number of sub-fragments of each fault semantic feature entry with the total number of sub-fragments of the corresponding pattern semantic entry to ensure that the number of sub-fragments of the two is consistent. Remove the entry pairs with inconsistent numbers to obtain a set of entry pairs with consistent numbers. Set a matching threshold for a set of entries with the same number of entries. Set the threshold for the total number of matching feature segments of each entry pair to the total number of feature segments of the corresponding entry. Record the matching threshold for each entry pair to obtain a set of entry matching thresholds. The statistical result set of item matching is compared with the set of item matching thresholds. The total number of matching sub-fragments of each fault semantic feature item is compared with its corresponding matching threshold. The comparison result of the total number of matching for each item with the threshold is recorded to obtain the threshold comparison result set. The threshold comparison result set is matched and determined. The fault semantic feature entries with a total number of matches equal to the matching threshold are determined to match the content of the corresponding pattern semantic entries, and the entries with a total number of matches not equal to the matching threshold are determined to not match, thus obtaining a preliminary entry matching determination result set. The preliminary item matching result set is validated by comparing the contents of the matched item pairs again to confirm the consistency of the contents, adjusting the incorrect results, and generating the final item matching result set.

9. The method as described in claim 1, characterized in that, The process involves performing fault root cause topological reasoning based on the fault mode classification tag set and the fault-scenario association binding cluster, constructing a logical association topological chain between the explicit semantic representation of the fault and potential causes, and generating a fault root cause reasoning topological chain and a fault propagation topological path set, including: The fault mode classification tag set and the fault-scenario association binding cluster are associated and fused. Each fault mode classification tag is fused with the scenario information and related fault semantic feature entries in its corresponding fault-scenario association binding cluster to obtain fused content containing fault tags, scenario information and related semantic features, and thus the fused association body is obtained. Explicit semantic representation extraction is performed on the fused associated body. Semantic content that directly reflects the fault performance is extracted from the fused associated body. The explicit fault performance description corresponding to each fault label is recorded to obtain the fault explicit semantic representation set. The potential cause semantic features are mined from the fused associated body. Indirect semantic features that may cause fault manifestations are extracted from the fused associated body. The potential cause semantic feature descriptions corresponding to each fault label are recorded to obtain a set of potential cause semantic features. Logically associate the set of explicit semantic representations of faults with the set of semantic features of potential causes, and logically connect each explicit semantic representation of faults with the corresponding semantic feature content of potential causes to construct a topological node that reflects the logical relationship between faults and causes, thus obtaining a set of logically associated topological nodes. A topology chain is constructed on the set of logically related topology nodes. All logically related topology nodes are connected according to the logical relationship between faults and causes, and a complete logical path from potential causes to fault manifestations is obtained, generating a logically related topology chain. Path extraction is performed on the logical association topology chain to extract the complete logical path from the semantic features of potential causes to the explicit semantic representation of the fault, as well as the propagation path of the fault in different scenarios, and to generate the fault root cause reasoning topology chain and the fault propagation topology path set.

10. A data processing apparatus, characterized in that, include: The data modeling module is used to perform time-series scene association modeling on the raw operating data collected by the vehicle-mounted detection equipment. Combining the continuous acquisition time-domain association relationship of the raw operating data and the scene switching startup logic of the vehicle system, it generates the time-series scene association mapping body and the abnormal start time-series marker cluster of the raw operating data. The topology construction module is used to construct a fault association semantic topology based on the time-series scene association mapping body and the abnormal start time-series marker cluster, sort out the semantic association links related to the abnormal performance of the vehicle system in the original operating data, and generate a fault semantic association map and a scene-semantic association mapping table. The graph clustering module is used to import the fault semantic association graph into a preset fault pattern semantic matching graph, perform qualitative clustering of fault patterns, match fault pattern semantic entries in the fault semantic association graph and the fault pattern semantic matching graph, and generate a fault pattern classification tag set and a fault-scenario association binding cluster. The root cause reasoning module is used to perform root cause topological reasoning based on the fault mode classification label set and the fault-scenario association binding cluster, construct a logical association topological chain between the explicit semantic representation of the fault and the potential causes, and generate a root cause reasoning topological chain and a fault propagation topological path set. The fault inference module is used to perform fault risk trend topology inference based on the fault root cause reasoning topology chain and the fault propagation topology path set, analyze the evolution topology logic of fault causes and the expansion topology direction of propagation paths, and generate fault risk warning semantic packages and risk propagation trend topology maps.

Images