Unmanned aerial vehicle fault traceability analysis method, device and equipment and storage medium

Abstract

The invention relates to an unmanned aerial vehicle fault traceability analysis method and device, equipment and a storage medium. The method comprises the steps of defining entity types and relationship types among entities based on a predefined fault ontology model to construct a mode layer of an unmanned aerial vehicle fault knowledge graph; based on the mode layer, extracting a fault triple from the multi-source operation data of the unmanned aerial vehicle by using a mixed extraction model, and constructing a fault knowledge graph containing instance data; endowing a dynamic weight probability representing confidence to a relation edge in the fault knowledge graph, and generating a probabilistic fault knowledge graph; and mapping to-be-analyzed fault information to the probabilistic fault knowledge graph, performing traceability analysis by using a hybrid inference engine, and outputting a fault reason and a transmission path. According to the method, structured deep fusion of domain knowledge and data value is realized, and the traceability conclusion is improved from qualitative judgment to quantitative decision support with confidence measurement.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault analysis, and in particular to a method, device and equipment for unmanned aerial vehicle (UAV) fault trace analysis and a storage medium. BACKGROUND

[0002] With the wide application of UAVs in the fields of inspection, surveying and mapping, and logistics, the system complexity and task load of the UAVs are increasing, leading to diversified and concealed fault modes. Ensuring flight safety and achieving rapid and accurate fault trace analysis have become key challenges for improving the reliability and operation efficiency of UAV systems. The traditional post-analysis method based on human experience cannot cope with massive flight data and complex system correlations, and therefore, it is of great practical significance to study intelligent fault trace analysis methods.

[0003] Currently, the research and practice of UAV fault analysis mainly follow two technical paths. One is a rule-based method that relies on domain experts to pre-summarize fault logic and form a rule base for reasoning in the form of "if... then...". The other is a data-driven method that usually uses deep learning models to learn fault features and patterns from historical data for diagnosis or prediction. Existing technical solutions focus on either using predefined static knowledge or mining statistical rules from data.

[0004] However, the above existing technologies have obvious defects. First, the rule-based method has high cost for knowledge base construction and update and maintenance, poor flexibility, and is difficult to adapt to new or complex correlated faults. Second, the pure data-driven method lacks interpretability and relies heavily on a large amount of high-quality labeled data. The conclusion is often a "black box" judgment, which is difficult to trace to specific fault components and propagation paths. The more core common problem is that existing technologies cannot effectively realize the deep fusion of domain expert knowledge and historical data value, and lack the ability to quantitatively reason about fault propagation uncertainty, resulting in an opaque trace analysis process and low reliability of conclusions when facing complex fault causal chains. SUMMARY

[0005] Therefore, it is necessary to provide a UAV fault trace analysis method, device, equipment and storage medium to solve the problem that existing technologies cannot effectively realize the deep fusion of domain expert knowledge and historical data value, and lack the ability to quantitatively reason about fault propagation uncertainty, resulting in an opaque trace analysis process and low reliability of conclusions when facing complex fault causal chains.

[0006] To achieve the above technical purposes, the present application adopts the following technical solutions: In a first aspect, the present application provides a UAV fault trace analysis method, comprising: Based on the pre-defined fault ontology model, define the entity type and the relationship type between entities to build the mode layer of the unmanned aerial vehicle fault knowledge graph; Based on the mode layer, use the hybrid extraction model to extract fault triples from the multi-source operation data of the unmanned aerial vehicle, and build the fault knowledge graph containing instance data; Assign a dynamic weight probability representing confidence to the relationship edge in the fault knowledge graph to generate a probabilistic fault knowledge graph; Map the fault information to be analyzed to the probabilistic fault knowledge graph, use the hybrid reasoning engine for traceability analysis, and output the fault cause and transmission path.

[0007] In one possible implementation, the pre-defined fault ontology model is constructed by using a hybrid modeling method, specifically including: Based on the knowledge of domain experts, define the core concepts and the relationships between concepts for describing fault causal relationships to form an initial ontology framework; Extract specific fault instances from historical fault reports, sensor data and maintenance records of the unmanned aerial vehicle; Align, integrate and dynamically extend the fault instances extracted in the bottom-up phase with the initial ontology framework defined in the top-down phase to obtain the fault ontology model.

[0008] In one possible implementation, based on the mode layer, the hybrid extraction model is used to extract fault triples from the multi-source operation data of the unmanned aerial vehicle to build the fault knowledge graph containing instance data, including: Call the rule model, the deep learning model and the large language model respectively to extract candidate triples from the multi-source operation data; Perform confidence evaluation and weighted fusion on the candidate triples, and perform knowledge graph logical consistency verification on the fused results to obtain fault triples; Incrementally update the fault triples to the unmanned aerial vehicle fault knowledge graph to build the fault knowledge graph containing instance data.

[0009] In one possible implementation, the weighted fusion specifically includes the following steps: According to the type and quality of the data source, assign dynamic weights to the triples output by different models; Among them, for the text with standard format, the output weight of the rule model is the highest; for the descriptive text with complex semantics, the output weight of the large language model is the highest.

[0010] In one possible implementation, a dynamic weight probability representing confidence is assigned to the relationship edge in the fault knowledge graph to generate a probabilistic fault knowledge graph, including: Based on the frequency of the fault triples in historical data, set the prior probability for them; Logical rules describing fault transmissibility are introduced as soft constraints, and each rule is assigned a weight; Joint probability inference is performed in the probabilistic graph model, and the posterior probability of the relationship edge is calculated as its dynamic weight to generate a probabilistic fault knowledge graph.

[0011] In one possible implementation, the hybrid reasoning engine combines symbolic reasoning and probabilistic reasoning; The symbolic reasoning is used for logical inference and consistency checking based on predefined rules; The probabilistic reasoning is used for calculating the most likely fault transmission path based on dynamic weight probability.

[0012] In one possible implementation, the fault information to be analyzed is mapped into the probabilistic fault knowledge graph, and the hybrid reasoning engine is used for traceability analysis to output the fault cause and transmission path, including: Based on the pre-defined domain rules in the schema layer, the fault information mapped into the graph is logically deduced and consistency checked to generate fault hypothesis reasons; Based on the dynamic weight probability in the probabilistic fault knowledge graph, the comprehensive confidence of each fault hypothesis path corresponding to the fault hypothesis reason is calculated; The fault hypothesis path with the highest comprehensive confidence and the corresponding fault hypothesis reason are selected as the final traceability analysis result.

[0013] In a second aspect, the present application also provides a UAV fault traceability analysis device, comprising: A schema construction module configured to define entity types and relationship types between entities based on a pre-defined fault ontology model to construct a schema layer of a UAV fault knowledge graph; A graph construction module configured to extract fault triples from multi-source operation data of the UAV based on the schema layer using a hybrid extraction model to construct a fault knowledge graph containing instance data; A relationship assignment module configured to assign dynamic weight probabilities representing confidence to relationship edges in the fault knowledge graph to generate a probabilistic fault knowledge graph; A fault analysis module configured to map fault information to be analyzed into the probabilistic fault knowledge graph and use the hybrid reasoning engine for traceability analysis to output the fault cause and transmission path.

[0014] In a third aspect, the present application also provides an electronic device comprising a memory and a processor, wherein, The memory is used to store programs; The processor is coupled to the memory and is used to execute the programs stored in the memory to implement the steps in the UAV fault traceability analysis method in any of the above-mentioned implementation manners.

[0015] In a fourth aspect, the present application also provides a computer readable storage medium for storing computer readable programs or instructions, which can realize the steps of the unmanned aerial vehicle fault traceability analysis method in any of the above implementation manners when executed by a processor.

[0016] The unmanned aerial vehicle fault traceability analysis method provided by the present application constructs a graph pattern layer based on a fault ontology model, thereby providing a unified structured semantic framework for unmanned aerial vehicle fault knowledge, enabling expert knowledge to be systematically expressed and reused. On this basis, a hybrid extraction model is used to automatically extract fault triples from multi-source operation data, thereby realizing efficient extraction of instance knowledge from structured logs and unstructured texts. Thus, the expert-defined semantic framework and data-driven instance discovery are closely combined, and the fusion of expert knowledge and historical data is realized. Further, dynamic weight probabilities are assigned to the relationship edges in the knowledge graph, and the causal relationship of fault propagation is converted into a quantifiable probability relationship, thereby being able to represent and handle the uncertainty in the fault propagation process and improving the accuracy of analysis. Finally, traceability analysis is realized through a hybrid reasoning engine, and path search and confidence evaluation of a complex fault chain are completed under an interpretable reasoning framework, and finally a fault path with a probabilistic explanation is output. Overall, through the construction of a knowledge graph, probabilistic enhancement, and a closed loop of hybrid reasoning, the present application realizes full-process enhancement from knowledge representation to uncertainty reasoning, and improves the transparency of the traceability analysis process and the reliability of the conclusion. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of an embodiment of the unmanned aerial vehicle fault traceability analysis method provided by the present application; Figure 2 An architectural diagram of an embodiment of the pattern layer of the fault knowledge graph provided by the present application; Figure 3 A model diagram of an embodiment of the deep learning extraction submodel provided by the present application; Figure 4 A structural diagram of an embodiment of the unmanned aerial vehicle fault traceability analysis device provided by the present application; Figure 5 A structural diagram of the unmanned aerial vehicle fault traceability analysis device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0018] The preferred embodiments of the present application will be described in detail below with reference to the accompanying drawings, which form a part of this application. The accompanying drawings illustrate the principles of the present application and, together with the description, serve to explain the present application, but are not intended to limit the scope of the present application.

[0019] In the description of the present application, the meaning of "a plurality of" is two or more than two, unless otherwise explicitly and specifically limited.

[0020] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive or alternative embodiments. It is explicitly and implicitly understood that the embodiments described herein can be combined with other embodiments.

[0021] The application provides a UAV fault tracing analysis method, device, equipment and storage medium, which are described below respectively.

[0022] Please refer to Figure 1 , Figure 1 The flowchart of an embodiment of the UAV fault tracing analysis method provided by the application, a specific embodiment of the application discloses a UAV fault tracing analysis method, comprising: S101, based on a pre-defined fault ontology model, defining entity types and entity relationship types to construct a mode layer of a UAV fault knowledge graph; S102, based on the mode layer, extracting fault triples from multi-source operation data of the UAV by using a hybrid extraction model to construct a fault knowledge graph containing instance data; S103, assigning dynamic weight probabilities representing confidence to relationship edges in the fault knowledge graph to generate a probabilistic fault knowledge graph; S104, mapping fault information to be analyzed into the probabilistic fault knowledge graph, using a hybrid reasoning engine to perform tracing analysis, and outputting fault causes and transmission paths.

[0023] In the above embodiment, a core fault cause framework is defined based on domain expert knowledge, including but not limited to core concepts such as parts, fault types, causes, effects, and their associated relationships, and then specific fault instances, attributes and relationships are extracted from instance data such as historical fault reports of the UAV, sensor data and maintenance records. Through entity alignment and relationship fusion, the initial framework is dynamically expanded and refined to form a mode layer of the UAV fault knowledge graph.

[0024] Please refer to Figure 2 , Figure 2The architecture schematic diagram of an embodiment of the mode layer of the fault knowledge graph provided by the application, the first step of constructing the unmanned aerial vehicle fault diagnosis graph is to obtain entities, relationships and attributes in the monitoring record and the fault transmission relationship, so as to design the ontology mode based on the fault transmission relationship of the internal equipment components, construct the mode layer of the fault diagnosis graph, mainly including: {component, contains, input element / output element}, {output element, fault transmission, input element}, {output element, monitoring, monitoring result} and the like, wherein the monitoring result data of the element is dynamic data, which is divided into two states of abnormality and normality, at this time, the data lake function in the system can be used to update the monitoring result data in a timing manner, for example, data update is performed every day / hour. Overall, the visualization effect of the fault point is realized by using the form of nodes and edges.

[0025] Collect various sensor data (attitude, speed, temperature, voltage, etc.), task execution logs, and maintenance records during the flight of the unmanned aerial vehicle. Automatically extract actual fault instances from flight logs, maintenance records, and laboratory test results. Then identify fine-grained attributes such as "fault type, cause, impact, detection method, and repair strategy".

[0026] Organize the data of the maintenance records, and complete the mode layer constructed by the experts according to the specific data. Automatically extract actual fault instances from flight logs, maintenance records, and laboratory test results. Then identify fine-grained attributes such as "fault type, cause, impact, detection method, and repair strategy".

[0027] Dig out specific fault instances from flight data and maintenance records, such as "2025-05-13 task motor speed anomaly, cause bearing wear"; then extract the attributes of the instance: occurrence condition, environmental factors, impact range, diagnosis means, etc. Then compare the instance with the top-level framework, if there is a corresponding concept, then instantiate the class and complete the attributes; then if a new type of fault (such as "ultrasonic altimeter data mutation") appears, a new class is dynamically created and a relationship is established with the related components or functions.

[0028] The hybrid extraction model integrates rule-based models, deep learning models, and large language models. Among them, the rule-based model extracts triples from structured or semi-structured text based on predefined regular expressions and domain templates; the deep learning model uses bert-CRF and TP-LINKER models for entity recognition and combines attention mechanisms for relationship extraction, suitable for semantic analysis of semi-structured text; the large language model extracts fault attributes from semantically complex descriptive text through prompt engineering guidance. Through multi-model parallel extraction and confidence weighted fusion, a high-quality fault knowledge graph instance data is finally constructed.

[0029] The probability soft logic method is used, the prior probability is calculated based on the frequency of the triplets in the historical data, then the logic rules describing the fault transmission are introduced as soft constraints, and weights are assigned to each rule, finally, the joint probability inference is carried out in the probabilistic graph model, the posterior probability of the relationship edge is solved by the maximum posterior probability estimation, and is taken as a dynamic weight, and the traditional qualitative knowledge graph is converted into a probabilistic graph capable of representing uncertainty.

[0030] Logical deduction is carried out based on predefined domain rules through symbolic reasoning, a set of possible fault hypothesis paths is generated, then the comprehensive confidence of each path is calculated based on dynamic weight probability through probability reasoning, and finally, the path with the highest confidence is selected as the traceability result, so that the logical rigor of the reasoning process is ensured, and quantitative evaluation of uncertainty is provided.

[0031] Compared with the prior art, the unmanned aerial vehicle fault traceability analysis method provided by the embodiment provides a unified structured semantic framework for unmanned aerial vehicle fault knowledge by constructing a graph pattern layer based on a fault ontology model, so that expert knowledge can be systematically expressed and reused, on the basis of which, a hybrid extraction model is used to automatically extract fault triplets from multi-source operation data, efficient extraction of instance knowledge from structured logs and unstructured texts is realized, so that the semantic framework defined by experts is closely combined with data-driven instance discovery, and the fusion of expert knowledge and historical data is realized. Further, dynamic weight probability is assigned to the relationship edge in the knowledge graph, the causal relationship of fault propagation is converted into a quantifiable probability relationship, so that the uncertainty in the fault propagation process can be represented and handled, and the accuracy of the analysis is improved. Finally, the traceability analysis is realized through the hybrid reasoning engine, the path search and confidence evaluation of the complex fault chain are completed under the interpretable reasoning framework, and finally, the fault path with the probability explanation is output. Overall, the scheme realizes the whole-process enhancement from knowledge representation to uncertainty reasoning through the closed loop of knowledge graph construction, probabilistic enhancement and hybrid reasoning, improves the transparency of the traceability analysis process and the reliability of the conclusion.

[0032] In some embodiments of the application, the predefined fault ontology model is constructed by using a hybrid modeling method, specifically including: Based on the domain expert knowledge, the core concepts and the relationships between the concepts for describing the fault causal relationship are defined, and an initial ontology framework is formed; Specific fault instances are extracted from the historical fault reports, sensor data and maintenance records of the unmanned aerial vehicle; The fault instances extracted in the bottom-up stage are aligned, integrated and dynamically extended with the initial ontology framework defined in the top-down stage, and the fault ontology model is obtained.

[0033] In the above embodiments, in the top-down construction phase, the core concept system describing the fault causal relationship is systematically defined based on domain expert knowledge, covering six core concept categories: fault components (such as motors, flight controls, batteries), fault phenomena (such as abnormal vibration, voltage drop), fault types (such as short circuit, wear and tear, broken wire), fault causes (such as overload, aging, collision), fault effects (such as shutdown, performance degradation) and detection methods (such as vibration analysis, infrared detection). At the same time, the semantic relationship between concepts is clearly defined, including "hasSymptom" (has symptom), "causedBy" (caused by), "affects" (affects), "locatedIn" (located in) and other semantic relationships, forming a structured initial ontology framework.

[0034] In the bottom-up construction phase, specific fault instances are extracted from the full life cycle data of the unmanned aerial vehicle. Through multi-source data collection and integration of sensor time series data in flight logs, text descriptions in maintenance reports, laboratory test records and other heterogeneous data sources, natural language processing technology is used to extract entity relationships from unstructured maintenance reports, and sensor data analysis algorithms are used to identify fault patterns from time series data, to establish a fault instance library containing specific parameters, environmental conditions, timestamps and other fine-grained attributes.

[0035] In the knowledge fusion phase, the instance knowledge is deeply integrated with the initial framework through ontology alignment technology. Specifically, a concept matching algorithm based on semantic similarity is used to map the data-driven instance concepts to the expert-defined ontology concepts. For new instances that match successfully, their attributes are filled in and enriched under the corresponding concepts; for new fault patterns that cannot be matched, the ontology framework is dynamically extended by experts to create new concept categories and relationship types. Through this two-way feedback mechanism, the ontology model not only maintains the accuracy of expert knowledge, but also has the ability to evolve to adapt to new faults.

[0036] In some embodiments of the present application, based on the pattern layer, a hybrid extraction model is used to extract fault triples from the multi-source operation data of the unmanned aerial vehicle, and a fault knowledge graph containing instance data is constructed, including: Respectively call the rule model, the deep learning model and the large language model to extract candidate triples from the multi-source operation data; Conduct confidence evaluation and weighted fusion on the candidate triples, and perform knowledge graph logical consistency verification on the fused results to obtain fault triples; Incrementally update the fault triples to the unmanned aerial vehicle fault knowledge graph, and construct a fault knowledge graph containing instance data.

[0037] In the above embodiment, based on the hybrid extraction model, the mode layer of the fault knowledge graph is relied on for information extraction, and entity, relationship and attribute extraction are performed on various sensor data (attitude, speed, temperature, voltage, etc.), task execution logs, maintenance records and other data in the flight process of the unmanned aerial vehicle. The hybrid extraction model is defined as a fusion application model of a large language model, a deep learning model and a rule model, and the extraction mode relies on the hybrid extraction model.

[0038] The rule extraction submodule designs templates and regular expressions according to domain knowledge, and adapts to common sentence patterns, such as: mode: [component] + appears + [fault description] -> extracts (component, fault type, state); mode: due to [reason] causes [fault] -> extracts (reason, causes, fault).

[0039] Please refer to Figure 3 , Figure 3 The model schematic diagram of an embodiment of the deep learning extraction submodel provided by the present application adopts a bert-CRF and TP-LINKER model to identify entity boundaries and relationships, and training data is derived from a manually annotated fault corpus.

[0040] The named entity recognition technology realized by using the bert pre-training model, BERT is short for Bidirectional Encoder Representation from Transformers, which is a pre-trained language representation model. It emphasizes that the traditional unidirectional language model or the method of shallowly splicing two unidirectional language models is no longer used for pre-training, but a new masked language model (MLM) is used, so that a deep bidirectional language representation can be generated.

[0041] The relationship extraction uses the algorithm of bert+attention. Among them, BERT is responsible for context semantic coding of the input sentence, and obtains a deep representation of each token. Attention takes the entity vector as a query, pays attention to the most relevant part of the sentence, forms a context relationship representation, and uses a classifier to splice the entity representation and the aggregated relationship vector to predict the relationship category.

[0042] Information extraction based on large language models does not rely on manually designed patterns or rules, but mainly relies on the semantic understanding ability of pre-trained large language models. The key step of pattern-free information extraction is to use large language models to automatically identify and extract useful information. Large language models can be guided to extract specific information from text through prompt engineering. The core step is prompt design: pattern-free information extraction relies on carefully designed prompts to guide the model to perform specific tasks. Through reasonable prompts, large models can understand the type of information to be extracted.

[0043] The confidence evaluation comprehensively considers factors such as the historical accuracy of each extraction model, the confidence score of this extraction, and the credibility of the data source, and establishes a multi-dimensional evaluation system. In the weighted fusion stage, the fusion algorithm based on D-S evidence theory is used to combine the high accuracy of the rule model in structured data, the stable performance of the deep learning model in semi-structured text, and the advantages of large language models in complex semantic understanding. Through an adaptive weight distribution strategy, the advantages of each model are complementary. Subsequently, a description logic-based reasoning mechanism is used to verify the logical consistency of the fusion results, checking for conflicts, contradictions, or inconsistencies in the existing knowledge graph, ensuring the logical rationality of new knowledge.

[0044] For triples with partial conflicts with existing knowledge but high confidence, a knowledge resolution process is started, and conflict resolution is performed based on evidence strength, time novelty, and source authority. At the same time, the system monitors the quality indicators of the knowledge graph in real time, including knowledge density, relationship richness, and logical consistency, and continuously improves the quality of the graph through dynamic optimization of the knowledge fusion strategy. Not only does this effectively integrate multi-source fault knowledge, but it also ensures that the knowledge graph has dynamic evolution capabilities and continuous optimization characteristics.

[0045] In some embodiments of the present application, the weighted fusion specifically includes the following steps: According to the type and quality of the data source, dynamic weights are assigned to the triples output by different models; Among them, for texts with standard formats, the output weight of the rule model is the highest; for complex descriptive texts, the output weight of the large language model is the highest.

[0046] In the above embodiments, establishing a weight evaluation system requires comprehensive consideration from three dimensions: data source characteristics, text complexity, and model characteristics. The data source dimension mainly evaluates the structured degree, noise level, and information integrity of the data; the text complexity dimension is quantified through syntax analysis, semantic density calculation, and domain term proportion; and the model characteristic dimension is evaluated based on the accuracy, recall rate, and F1 value of each model in historical similar tasks.

[0047] For the standardization of the text of the format specification, such as the parameter record in the flight log, the standard description in the maintenance report, etc., the rule model is given a higher weight of 0.6-0.8 due to its accurate pattern matching capability, and the deep learning model and the large language model are configured with auxiliary weights of 0.2-0.3 and 0.1-0.2, respectively. For the semantically complex descriptive text, such as the handwritten records of maintenance personnel, the free description of fault phenomena, etc., the large language model is given a core weight of 0.5-0.7 due to its strong semantic understanding capability, the deep learning model is configured with a weight of 0.3-0.4 as a supplement, and the rule model is reduced to a reference weight of less than 0.1.

[0048] In the weight distribution process, a real-time feedback mechanism is introduced to dynamically adjust the weight parameters by continuously monitoring the performance of each model in actual application. For mixed text of structured data and unstructured data, a hierarchical processing strategy is adopted: structured information is extracted by the rule model first, then the large language model is used to analyze the free text, and finally the results are integrated through a weighted voting mechanism. Thus, the optimal information extraction effect can be achieved under different data characteristics, and the accuracy and robustness of knowledge extraction are improved.

[0049] In some embodiments of the present application, a dynamic weight probability representing confidence is assigned to the relationship edge in the fault knowledge graph to generate a probabilistic fault knowledge graph, comprising: Based on the frequency of the fault triplets appearing in the historical data, a prior probability is set for each fault triplet; Logical rules describing fault transitivity are introduced as soft constraints, and each rule is given a weight; Joint probability inference is performed in the probabilistic graph model to calculate the posterior probability of the relationship edge as its dynamic weight, and a probabilistic fault knowledge graph is generated.

[0050] In the above embodiments, Bayesian statistical method is used to set the prior probability, and by analyzing the operation and maintenance data of the whole life cycle of the unmanned aerial vehicle, the occurrence frequency of each type of fault mode, the number of fault transmissions between components, and the triggering probability of faults in specific environments are counted. For frequent fault relationships such as "motor overheating causing abnormal speed", a higher initial probability value (such as 0.85) is assigned; and for rare but existing fault associations such as "sensor drift causing control system failure", a proper baseline probability (such as 0.15) is given based on Laplace smoothing technique to ensure the rationality and completeness of the probability distribution. The specific settings can be made according to the actual needs, and the present application does not make further limitation.

[0051] The logical rules describing fault transitivity include spatiotemporal characteristic rules of fault conduction (such as "if component A is located upstream of component B, and A fails earlier than B, then there is a fault conduction relationship from A to B"), causal strength rules (such as "the confidence that a voltage drop causes system restart is higher than that of performance degradation"), and environment-dependent rules (such as "the probability of battery failure causing abnormality of the power system is increased in a high-temperature environment"). Each rule is verified for reliability through historical data, and its weight coefficient is determined by domain experts, with a value range usually set between 0.5 and 1.0.

[0052] A Markov logic network is adopted as the modeling basis to unify prior probabilities and weighted logical rules in a probabilistic graph model. Iterative reasoning is performed through a belief propagation algorithm, and the posterior probability of each relationship edge under a given evidence condition is calculated by comprehensively considering historical statistical rules and domain knowledge constraints. This process combines the advantages of data-driven and knowledge-driven methods, and the finally generated probabilistic fault knowledge graph not only contains the traditional entity relationship topology, but also quantifies the confidence degree of each fault path.

[0053] To realize quantitative analysis of fault propagation uncertainty, the present application adopts a Bayesian and logic network hybrid method based on probabilistic soft logic (PSL) to assign dynamic weight probabilities to relationship edges in the fault knowledge graph. The specific implementation steps are as follows: (1) Define the random variable and prior probability of the triple Each candidate triple X(h, r, t) in the knowledge graph is regarded as a Boolean random variable, which takes the value true (1) or false (0) to represent whether the fact is true. First, a prior probability is calculated for each triple based on historical observation data. This calculation is based on the Bernoulli distribution, and the Beta distribution is used as the conjugate prior for smoothing, as follows: ; Where c(x) is the number of times the triple appears in historical observations, N(x) is the total number of candidate triple samples, and a and β are smoothing hyperparameters. This prior probability comprehensively considers the historical frequency of the fact and the quality of the data source.

[0054] (2) Introduce domain logic rules as soft constraints Domain expert knowledge is formalized as logical rules, which are introduced as soft constraints into the probabilistic model. For example, the transitivity rule is defined as: "if component A is a component of component B, and component B is a component of component C, then component A is a component of component C". Each such logical rule is assigned a weight w i , which represents the relative importance and credibility of the rule in the reasoning process.

[0055] (3) Constructing joint probability distribution and performing posterior inference Under the PSL framework, the aforementioned prior probability and weighted logical rules are jointly modeled as a probabilistic graphical model. The joint probability distribution of the model is defined as: ; where, n i (X) represents the number of times the logical rule i is satisfied under the current variable assignment X, Z is the normalization constant.

[0056] In order to seamlessly integrate the prior probability into the model, it is converted into an equivalent virtual formula weight: ; Subsequently, based on this joint probability distribution, probabilistic inference algorithms such as Gibbs Sampling are used to calculate the posterior probability of each triple being true under the condition that all logical rules and prior knowledge are given: ; (4) Generating probabilistic graph and decision output The posterior probability obtained by the above calculation is the final dynamic weight probability of the corresponding relationship edge. Assigning this probability value to the edge in the knowledge graph generates a probabilistic fault knowledge graph. In practical applications, a confidence threshold (e.g., p>0.8) can be set to determine whether to adopt the relationship, thereby supporting subsequent accurate tracing and reliable decision-making.

[0057] In some embodiments of the present application, the hybrid inference engine combines symbolic reasoning and probabilistic reasoning; wherein the symbolic reasoning is used for logical inference and consistency checking based on predefined rules; and the probabilistic reasoning is used for calculating the most likely fault propagation path based on dynamic weight probability.

[0058] In the above embodiments, at the symbol reasoning level, a complete reasoning mechanism is built based on description logic and rule engine, relying on the pre-defined domain rules in the mode layer of the fault knowledge graph, including component hierarchical relationship (such as "the motor is a component of the power system"), fault propagation constraints (such as "power failure may cause all dependent components to be abnormal"), and timing logic rules (such as "the appearance time of the fault phenomenon is not earlier than the occurrence time of the root cause"). The reasoning process first performs concept classification and consistency check through the description logic reasoning machine, ensuring that the fault hypothesis conforms to the basic logic of the domain knowledge; then performs forward chain reasoning based on the SWRL rules to deduce the possible root cause set from the known fault phenomena; and finally generates fault propagation hypotheses that conform to all constraint conditions through the abductive reasoning.

[0059] At the probability reasoning level, a reasoning algorithm based on a probabilistic graph model is used to fully utilize the dynamic weight information in the probabilistic fault knowledge graph. Specifically, first, the candidate fault paths generated by the symbol reasoning are mapped to the reasoning paths in the Bayesian network, where the nodes represent fault states and the edge weights correspond to the dynamic probabilities in the knowledge graph. Then, a likelihood weighting algorithm is used to calculate the joint probability of each path in combination with real-time observation evidence. For complex multi-fault concurrent scenarios, the Markov Chain Monte Carlo method is introduced for approximate reasoning to obtain a stable estimate of the probability distribution through random sampling. Finally, the most likely fault transmission path is determined by maximum a posteriori probability estimation, and a quantitative confidence evaluation is provided for each path.

[0060] The symbol reasoning provides a candidate solution space that conforms to the domain constraints for the probability reasoning, effectively reducing the computational complexity; the probability reasoning provides a quantitative evaluation for the conclusions of the symbol reasoning, assisting in determining the optimal option among multiple logically reasonable solutions, which not only ensures that the reasoning process conforms to the logical judgment of domain experts, but also fully utilizes the data-driven probability information, showing excellent accuracy and practicality in complex fault scenarios.

[0061] In some embodiments of the present application, the fault information to be analyzed is mapped to the probabilistic fault knowledge graph, and a hybrid reasoning engine is used for traceability analysis to output the fault causes and transmission paths, including: Based on the pre-defined domain rules in the mode layer, the fault information mapped to the graph is logically deduced and consistency checked to generate fault hypothesis causes; Based on the dynamic weight probabilities in the probabilistic fault knowledge graph, the comprehensive confidence of each fault hypothesis path corresponding to the fault hypothesis cause is calculated; The fault hypothesis path with the highest comprehensive confidence and the corresponding fault hypothesis cause are selected as the final traceability analysis result.

[0062] In the above embodiment, the real-time monitored fault phenomena (such as "abnormal motor speed", "voltage fluctuation") are mapped to the corresponding nodes of the knowledge graph through the entity linking technology, and then a description logic-based inference engine is started to perform a symbolic reasoning process containing component topology constraints (such as "fly-by-wire system failure will not directly cause power battery damage"), fault propagation timing (such as "sensor data anomaly should precede control system alarm") and physical law constraints (such as "current over-limit may cause temperature rise"). The initial fault hypothesis is generated by forward chain reasoning, and the hypothesis with logical conflict is excluded by ontology consistency detection, and finally all reasonable fault hypothesis reasons conforming to the field knowledge are output to form a hypothesis set to be verified.

[0063] Each fault hypothesis is converted into a query path in the knowledge graph, and the dynamic weights of each relationship edge on the path are aggregated through an attention mechanism, while considering the first-order adjacency relationship and multi-hop propagation characteristics between nodes. For complex scenarios containing multiple fault phenomena, a Bayesian network is used to model the conditional independence between phenomena, and a belief propagation algorithm is used to calculate the joint probability distribution. In particular, a time decay factor is also introduced to give higher weight to recently occurring fault relationships, ensuring that the reasoning result conforms to the actual running state of the system.

[0064] Not only the comprehensive confidence of the path is considered, but also the path length (preferably a more direct propagation path), historical verification frequency (preferably a path that has been verified in practice), and expert preference settings (priority rules in specific scenarios) are combined for comprehensive evaluation. The final output of the traceability analysis result contains the complete fault transmission chain, the confidence distribution of each link, and the support degree of key evidence, providing transparent, interpretable and quantitative basis for maintenance decision-making, and improving the accuracy and reliability of the fault traceability result.

[0065] In order to better implement the unmanned aerial vehicle fault traceability analysis method in the embodiment of the application, on the basis of the unmanned aerial vehicle fault traceability analysis method, please refer to Figure 4 , Figure 4 The structure diagram of an embodiment of the unmanned aerial vehicle fault traceability analysis device provided by the application is shown. The embodiment of the application provides an unmanned aerial vehicle fault traceability analysis device 400, which comprises: A schema construction module 410 configured to define entity types and entity relationship types based on a pre-defined fault ontology model to construct a schema layer of the unmanned aerial vehicle fault knowledge graph; A graph construction module 420 configured to extract fault triples from multi-source operation data of the unmanned aerial vehicle based on the schema layer and using a hybrid extraction model to construct a fault knowledge graph containing instance data; a relationship assigning module 430 configured to assign a dynamic weight probability representing a confidence to a relationship edge in the fault knowledge graph, to generate a probabilistic fault knowledge graph; a fault analysis module 440 configured to map the fault information to be analyzed to the probabilistic fault knowledge graph, to perform a trace analysis by using a hybrid reasoning engine, and to output a fault cause and a transmission path.

[0066] It should be noted that the unmanned aerial vehicle fault trace analysis device 400 provided by the above embodiments can implement the technical solutions described in the above method embodiments, and the principles of the implementation of the above modules or units can be referred to the corresponding content in the above method embodiments, which will not be described here.

[0067] Please refer to Figure 5 , Figure 5 A structure diagram of an unmanned aerial vehicle fault trace analysis device provided by an embodiment of the present application is shown. Based on the above unmanned aerial vehicle fault trace analysis method, the present application further provides an unmanned aerial vehicle fault trace analysis device. The unmanned aerial vehicle fault trace analysis device can be a mobile terminal, a desktop computer, a notebook computer, a palm computer, a server, or other computing devices. The unmanned aerial vehicle fault trace analysis device 500 includes a processor 510, a memory 520, and a display 530. Figure 5 Only some components of the unmanned aerial vehicle fault trace analysis device are shown, but it should be understood that all the shown components are not required, and more or fewer components can be alternatively implemented.

[0068] The memory 520 can be an internal storage unit of the unmanned aerial vehicle fault trace analysis device 500 in some embodiments, such as a hard disk or a memory of the unmanned aerial vehicle fault trace analysis device 500. The memory 520 can also be an external storage device of the unmanned aerial vehicle fault trace analysis device 500 in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, or the like. Further, the memory 520 can include both an internal storage unit and an external storage device of the unmanned aerial vehicle fault trace analysis device 500. The memory 520 is used to store application software and various data installed in the unmanned aerial vehicle fault trace analysis device 500, such as program codes installed in the unmanned aerial vehicle fault trace analysis device 500. The memory 520 can also be used to temporarily store data that has been output or will be output. In an embodiment, the memory 520 stores an unmanned aerial vehicle fault trace analysis program 540, which can be executed by the processor 510, thereby implementing the unmanned aerial vehicle fault trace analysis method of the embodiments of the present application.

[0069] The processor 510 may, in some embodiments, be a central processing unit (CPU), a microprocessor, or other data processing chip, for running program codes stored in the memory 520 or processing data, such as performing the unmanned aerial vehicle fault tracing analysis method, etc.

[0070] The display 530 may, in some embodiments, be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, etc. The display 530 is used to display information of the unmanned aerial vehicle fault tracing analysis device 500 and to display a visualized user interface. The components 510-530 of the unmanned aerial vehicle fault tracing analysis device 500 communicate with each other through a system bus.

[0071] In an embodiment, when the processor 510 executes the unmanned aerial vehicle fault tracing analysis program 540 in the memory 520, the steps in the above unmanned aerial vehicle fault tracing analysis method are implemented.

[0072] The embodiment also provides a computer readable storage medium having stored thereon an unmanned aerial vehicle fault tracing analysis program, which, when executed by a processor, implements the following steps: Based on a pre-defined fault ontology model, define entity types and entity relationship types to build a schema layer of an unmanned aerial vehicle fault knowledge graph; Based on the schema layer, use a hybrid extraction model to extract fault triples from multi-source operation data of the unmanned aerial vehicle, and build a fault knowledge graph containing instance data; Assign dynamic weight probabilities representing confidence to relationship edges in the fault knowledge graph, and generate a probabilistic fault knowledge graph; Map fault information to be analyzed to the probabilistic fault knowledge graph, use a hybrid reasoning engine to perform tracing analysis, and output fault causes and transmission paths.

[0073] The application also provides a computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the steps of the above method. The computer readable storage medium may, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro drive, and a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), or any type of medium or device suitable for storing instructions and / or data.

[0074] It should be noted that, for the foregoing method embodiments, the sequences of the described actions can be changed, and the actions can be performed in other sequences or concurrently. Additionally, it should be understood that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0075] In the above embodiments, the description of each embodiment is focused on different aspects, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0076] In the several embodiments provided by the present application, it should be understood that the disclosed apparatus can be implemented in other manners. For example, the described apparatus embodiments are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0077] The units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0078] In addition, each functional unit in the embodiments of the present application can be integrated in a processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware or software function unit.

[0079] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable memory. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a memory, including a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned memory includes: a U disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0080] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable memory, which can include a flash disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, etc.

[0081] The above is only exemplary embodiments of the present disclosure, which cannot limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. Those skilled in the art will easily think of embodiments of the present disclosure after considering the specification and practicing the disclosure herein. The present application is intended to cover any variations, uses or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not recorded in the present disclosure. The specification and examples are only considered as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

[0082] The technical features of the above embodiments can be combined in any way. In order to make the description simple, not all possible combinations of the technical features in the above embodiments are described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope recorded in the present disclosure.

[0083] Those skilled in the art readily understand that the above only describes preferred embodiments of the present application and does not limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0084] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for fault tracing analysis of a UAV, characterized in that, The application comprises the following steps: Based on a pre-defined fault ontology model, define entity types and entity relationship types to build a mode layer of a UAV fault knowledge graph; Based on the mode layer, use a hybrid extraction model to extract fault triples from multi-source operation data of the UAV, and build a fault knowledge graph containing instance data; Assign dynamic weight probabilities representing confidence to relationship edges in the fault knowledge graph to generate a probabilistic fault knowledge graph; Map the fault information to be analyzed to the probabilistic fault knowledge graph, use a hybrid reasoning engine for traceability analysis, and output fault causes and transmission paths. 2.The UAV fault traceability analysis method of claim 1, wherein, The pre-defined fault ontology model is built by using a hybrid modeling method, specifically including: Based on domain expert knowledge, define core concepts and concept relationships for describing fault causal relationships to form an initial ontology framework; Extract specific fault instances from historical fault reports, sensor data and maintenance records of the UAV; Align, integrate and dynamically extend the fault instances extracted in the bottom-up phase with the initial ontology framework defined in the top-down phase to obtain the fault ontology model. 3.The UAV fault traceability analysis method of claim 1, wherein, Based on the mode layer, use a hybrid extraction model to extract fault triples from multi-source operation data of the UAV, and build a fault knowledge graph containing instance data, including: Call rule models, deep learning models and large language models respectively to extract candidate triples from the multi-source operation data; Perform confidence evaluation and weighted fusion on the candidate triples, and perform knowledge graph logical consistency verification on the fused results to obtain fault triples; Incrementally update the fault triples to the UAV fault knowledge graph to build a fault knowledge graph containing instance data. 4.The UAV fault back analysis method of claim 3, wherein, The weighted fusion specifically includes the following steps: According to the type and quality of the data source, assign dynamic weights to the triples output by different models; Among them, for the text with standard format, the output weight of the rule model is the highest; for the descriptive text with complex semantics, the output weight of the large language model is the highest. 5.The UAV fault back analysis method of claim 1, wherein, Assigning dynamic weight probabilities representing confidence to relationship edges in the fault knowledge graph to generate a probabilistic fault knowledge graph, including: Based on the frequency of the fault triples in historical data, set a prior probability for them; Introduce logical rules describing fault transmission as soft constraints, and assign weights to each rule; Perform joint probability reasoning in the probabilistic graph model to calculate the posterior probability of the relationship edge as its dynamic weight, and generate a probabilistic fault knowledge graph. 6.The UAV fault back analysis method of claim 1, wherein, The hybrid reasoning engine combines symbolic reasoning and probabilistic reasoning; Among them, the symbolic reasoning is used for logical inference and consistency check based on pre-defined rules; The probabilistic reasoning is used to calculate the most likely fault transmission path based on the dynamic weight probability. 7.The UAV fault back analysis method of claim 1, wherein, Mapping the fault information to be analyzed to the probabilistic fault knowledge graph, using a hybrid reasoning engine for traceability analysis, and outputting fault causes and transmission paths, including: Based on the pre-defined domain rules in the mode layer, perform logical deduction and consistency check on the fault information mapped to the graph to generate fault hypothesis causes; Based on the dynamic weight probability in the probabilistic fault knowledge graph, a comprehensive confidence of each fault hypothesis reason corresponding to a fault hypothesis path is calculated; The fault hypothesis path with the highest comprehensive confidence and the corresponding fault hypothesis reason are selected as the final traceability analysis result.

8. An unmanned aerial vehicle failure root cause analysis apparatus, characterized by, Comprise: A schema construction module configured to define entity types and entity relationship types based on a pre-defined fault ontology model to construct a schema layer of the unmanned aerial vehicle fault knowledge graph; A graph construction module configured to extract fault triples from multi-source operation data of the unmanned aerial vehicle based on the schema layer using a hybrid extraction model to construct a fault knowledge graph containing instance data; A relationship assignment module configured to assign dynamic weight probabilities representing confidence to relationship edges in the fault knowledge graph to generate a probabilistic fault knowledge graph; A fault analysis module configured to map fault information to be analyzed into the probabilistic fault knowledge graph and use a hybrid reasoning engine to perform traceability analysis to output fault reasons and transmission paths.

9. An electronic device, comprising: Comprise a memory and a processor, wherein, The memory is used to store programs; The processor is coupled with the memory and is used to execute the programs stored in the memory to implement the steps in the unmanned aerial vehicle fault traceability analysis method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer readable program or instruction is used to store the program or instruction which can be executed by the processor to implement the steps in the unmanned aerial vehicle fault traceability analysis method of any one of claims 1 to 7.

Images