Aviation equipment reliability evaluation method and system based on knowledge graph and model inference

Abstract

Disclosed in the present invention are an aviation equipment reliability evaluation method and system based on a knowledge graph and model inference. The method comprises: acquiring data of human factors, equipment systems, and a working environment of aviation equipment; carrying out preprocessing and text labeling on the acquired data; inputting the labeled text information into a constructed entity relationship joint extraction model to form a high-quality structured triple of the knowledge graph; constructing an elastic knowledge graph for the aviation equipment, wherein the elastic knowledge graph comprises an online knowledge graph and an offline knowledge graph which has aviation equipment reliability; and extracting semantic features, and analyzing the similarity between the extracted features to realize indirect inference of the aviation equipment reliability. The present invention fully fuses expert experience and knowledge data, and exerts respective advantages of a human brain and machine intelligence, so as to achieve accurate analysis and prediction of aviation equipment reliability, thereby providing intelligent risk analysis, early warning and optimization suggestions for command and control personnel, and reducing a fault occurrence rate.

Description

A method and system for reliability assessment of aviation equipment based on knowledge graphs and model reasoning Technical Field

[0001] This invention belongs to the field of artificial intelligence and relates to the reliability assessment of aviation equipment, specifically to a method and system for the reliability assessment of aviation equipment based on knowledge graphs and model reasoning. Background Technology

[0002] In recent years, aviation equipment has been continuously developing towards digitalization, intelligence, and unmanned operation. Due to the high integration of new technologies, the complexity of its operating conditions has continued to increase, resulting in a significant increase in unreliable factors. Low reliability and frequent failures have become major challenges restricting the high-quality development of aviation equipment. Therefore, it is of great significance to efficiently and accurately assess the reliability of aviation equipment and provide proactive and timely early warning services.

[0003] Existing reliability assessment methods typically employ deep learning-based neural network analysis and particle swarm optimization. These methods require the collection of sample data, followed by iterative training of the deep learning model using the sample dataset. Only a well-trained model can then be used for reliability assessment. Therefore, the accuracy of reliability assessment largely depends on the reliable establishment of sample data and the training set. However, on the one hand, aviation equipment systems operate under numerous conditions and contain various mechanical and electronic components, making them complex electromechanical coupling systems. Accurately obtaining sample fault characteristic datasets is difficult, resulting in low precision in reliability analysis. On the other hand, the flight environment is complex, diverse, and dynamic. Furthermore, aviation equipment systems exhibit multi-state characteristics of performance degradation or state evolution during failure processes. The distribution domain of data collected on-site often differs from that of the training data, leading to performance degradation of the trained model in practical applications. This results in low environmental adaptability and poor adaptability of reliability assessment algorithms.

[0004] Summary of the Invention

[0005] Purpose of the invention: In order to overcome the shortcomings of the existing technology, this invention provides a method and system for reliability assessment of aviation equipment based on knowledge graphs and model reasoning. It fully integrates expert experience and knowledge data, leverages the respective advantages of human brain and machine intelligence, and realizes accurate analysis and prediction of aviation equipment reliability. This helps to provide command and control personnel with intelligent risk analysis, early warning and optimization suggestions, and reduce the failure rate.

[0006] Technical Solution: To achieve the above objectives, this invention provides a method for reliability assessment of aviation equipment based on knowledge graphs and model reasoning, comprising the following steps:

[0007] S1: Acquire three types of data for aviation equipment: human factors, equipment systems, and working environment;

[0008] S2: Preprocess and annotate the data obtained in step S1;

[0009] S3: Input the annotated text information from step S2 into the constructed entity relation joint extraction model to form high-quality structured triples of the knowledge graph;

[0010] S4: Based on the triples, construct a flexible knowledge graph for aviation equipment, including an online knowledge graph and an offline knowledge graph with aviation equipment reliability features;

[0011] S5: Extract semantic features from online knowledge graphs and offline knowledge graphs with aviation equipment reliability information, analyze the similarity of extracted features, and realize indirect inference of aviation equipment reliability.

[0012] Furthermore, the human factors in step S1 include vital signs, electroencephalogram (EEG), electromyography (EMG) signals, manual operation, voice interaction, and facial expressions; the equipment system includes a navigation system, auxiliary power, air supply system, landing gear, anti-icing and rain drainage, cabin doors, communication system, and air conditioning system; and the working environment includes temperature, humidity, rainfall, air pressure, and wind speed.

[0013] Further, step S2 specifically includes:

[0014] The aviation equipment data is preprocessed by data cleaning, data transformation, and standardization. The preprocessed text is stored in .csv format to obtain a text corpus of natural language description of the system's working status. The triple joint annotation tool LAnn is used to annotate the text with entity relations. The annotated data is converted into SPO triples (Subject-Predicate-Object) in the form of SPO triples. Each data point contains the original text and the corresponding triple sequence.

[0015] Furthermore, in step S3, the entity relation joint extraction model uses a BiLSTM+CRF algorithm-based entity relation joint extraction method to extract triples from the text corpus; wherein BiLSTM integrates two sets of LSTM layers with opposite learning directions, and can achieve fast and efficient entity relation extraction through a large amount of labeled data and continuous model iteration.

[0016] Furthermore, the specific process of extracting triples from the text corpus using the entity relation joint extraction method based on the BiLSTM+CRF algorithm in step S3 is as follows:

[0017] Input text sequence X = {x1, x2, ..., x} t}, single time step input x t After being represented by a word-level embedding layer, it is input into the forward LSTM unit for computation. The inverted LSTM unit calculates The output h′ of the hidden layer t yes and The splicing; BiLSTM is represented as:

[0018] h′ t Input is fed into a linearly linked CRF layer; given input and the label set y = {l1, l2, l3, l4, ..., l t The probability of sequence label Y is:

[0019] In equation (2), Z′ is the normalization factor, and y′ represents any label sequence. It is for l t A dedicated model parameter matrix, the same It is for l t A proprietary model bias parameter matrix;

[0020] For the decoding process, the Viterbi algorithm is used to obtain the Viterbi result sequence and Viterbi result value, and the position of the predicted entity word is obtained based on the Viterbi result sequence.

[0021] Furthermore, the flexible knowledge graph of aviation equipment in step S4 includes a schema layer and a data layer, and the specific construction method is as follows:

[0022] A1: The schema layer for constructing a flexible knowledge graph of aviation equipment; the schema layer describes the entity concepts in the aviation equipment working state scenario domain and the relationships between them, and is responsible for guiding the entire knowledge structure;

[0023] A2: Construct an online knowledge graph for the aviation equipment data layer;

[0024] Based on the pattern layer, the text corresponding to the three types of data in the real-time scene of aviation equipment—human factors, equipment system, and working environment—is extracted by the trained extraction model, followed by triple extraction, knowledge fusion, and quality assessment. The extracted text is then stored in the Neo4j graph database to complete the construction of an online knowledge graph.

[0025] A3: Construct an offline knowledge graph for the aviation equipment data layer;

[0026] The offline knowledge graph mainly targets multiple different aviation equipment information, using historical human factors, equipment systems, working environment, fault investigation reports, maintenance history, and fault repair reports as the original data sources. It mainly selects aviation equipment text corpora corresponding to different fault types in multiple working scenarios, and follows the construction process of online knowledge graphs to obtain an offline knowledge graph of aviation equipment with reliability and fault data.

[0027] Furthermore, in step S5, the GraphSAGE model is used to extract semantic features from the online and offline knowledge graphs of aviation equipment, and the semantic similarity of the graphs is analyzed, specifically as follows:

[0028] After extracting global feature vectors from the online knowledge graph and scene graph using the GraphSAGE model, two feature vectors V1 and V2 are obtained. The cosine value of the two vectors is calculated to transform the nonlinear difference into a difference in vector value and direction, thereby obtaining the similarity value. The similarity calculation formula is shown in the following formula:

[0029] In equation (3), V1 represents the global semantic feature vector corresponding to the sub-scene graph in the offline knowledge graph of aviation equipment; V2 represents the global semantic feature vector corresponding to the online knowledge graph; cosθ represents the cosine value of the two vectors;

[0030] A similarity threshold of T is set. When the Simc similarity value is greater than the threshold T, it indicates that the two maps match. When the Simc similarity value is less than the threshold T, it indicates that the two maps do not match.

[0031] Furthermore, in step S5, if the graph matching is successful, the reliability of the scenario with the highest matching degree is selected as the reliability of the real-time scenario; if the matching is unsuccessful, the real-time aviation equipment working state is treated as a new scenario, and the reliability of the scenario is inferred using a Bayesian network model; finally, the accuracy of the estimated aviation equipment system reliability is analyzed in conjunction with expert experience. If the estimation is incorrect, the online knowledge graph of the aviation equipment for the new scenario is discarded; if the estimation is correct, the online knowledge graph of the new scenario with aviation equipment reliability is numbered and integrated into the offline knowledge graph for use in the next scenario graph matching, thereby updating the offline knowledge graph.

[0032] Furthermore, the training and operation method of the Bayesian network model is as follows:

[0033] B1: Acquisition of Bayesian network training data;

[0034] Before learning the structure and parameters of the Bayesian network, the training data must first be obtained. The selected training data mainly comes from the offline knowledge graph of aviation equipment system reliability, which contains 9 types of faults, 3 working scenarios for each type, and a total of 27 aviation equipment knowledge graphs. The information of different graphs is mapped to the corresponding input variables of the Bayesian network, and the variable values ​​of the corresponding variables are obtained. The label of the target variable, the fault type, is added to the end of the variable value to obtain the raw data corresponding to different fault scenarios. In order to facilitate the import of data during the subsequent learning of the structure and parameters of the Bayesian network, it is necessary to discretize each variable.

[0035] B2: Before training the Bayesian network model, to improve the model's generalization ability and interpretability while avoiding overfitting, a random forest algorithm is used to optimize the feature dimensionality of the training data; specifically as follows:

[0036] Suppose we have sample data k = 1, 2, 3, ..., K, where K represents the number of samples, and each sample has N-dimensional features. The steps for calculating the ranking of feature importance are as follows:

[0037] 1) Initialize k=1 and create decision tree T k ;

[0038] 2) Train T on the k-th sample k Calculate the classification accuracy for each data point.

[0039] 3) Randomly select features X from the dataset j Apply a perturbation to j = 1, 2, ..., N, and recalculate.

[0040] 4) Repeat steps 1) to 3) for k = 1, 2, 3, ..., K;

[0041] 5) Feature X j Importance metric P j Calculated using the following formula

[0042] In equation (4): P j Representing feature X j Importance metrics This represents the classification accuracy of data without perturbation. This represents the data classification accuracy recalculated after perturbing the j-th feature;

[0043] 6) For P j Sort the features in descending order to obtain the feature importance ranking, and select the top N features as the input parameters for training the Bayesian network;

[0044] B3: The Bayesian model structure is learned using a search scoring method, and the parameters of the Bayesian network are learned using a Bayesian estimation method.

[0045] B4: Apply the trained Bayesian model to reliability reasoning of aerospace equipment systems to obtain reliability assessment results.

[0046] This invention also provides an aviation equipment reliability assessment system based on knowledge graphs and model reasoning, comprising:

[0047] Text data processing module: From the perspective of "human, machine and environment", it acquires three types of data on human factors of aviation equipment, equipment system and working environment through the cockpit vital signs monitoring system, fault investigation form, maintenance history fault repair form and Internet of Things sensor. After preprocessing and text annotation, the data is sent to the entity relationship triplet extraction module.

[0048] Entity Relationship Triple Extraction Module: This module includes a pre-constructed entity relationship joint extraction model that integrates the feature modeling of entities and relationships, while simultaneously detecting the semantic relationships between entities. It transforms the descriptive text information corresponding to the working status of aviation equipment into high-quality structured triples that can form a knowledge graph.

[0049] The elastic knowledge graph construction module includes a knowledge graph schema layer construction unit and a knowledge graph data layer construction unit. The knowledge graph schema layer construction unit describes the entity concepts in the aviation equipment working state scenario domain and the relationships between them, and is responsible for guiding the entire knowledge structure. The knowledge graph data layer construction unit includes an aviation equipment online knowledge graph data layer and an aviation equipment offline knowledge graph data layer. Based on the definition of the knowledge graph schema layer, it extracts aviation equipment entity and its relationship data to realize the semi-automatic construction of online and offline knowledge graphs for aviation equipment.

[0050] System reliability inference module: includes a Bayesian inference model construction unit and a scene graph similarity calculation unit;

[0051] Bayesian inference model building unit: Maps triple information in the online knowledge graph of aviation equipment to the corresponding node variables of the Bayesian network model. After obtaining the values ​​of the corresponding node variables, the Bayesian model is used to infer reliability.

[0052] Scene graph similarity calculation unit: Extracts semantic features from real-time online knowledge graph and offline knowledge graph of aviation equipment respectively, analyzes the similarity of features, and realizes indirect reasoning on the reliability of aviation equipment.

[0053] This invention employs a deep learning-based triplet extraction method to construct an online knowledge graph of human factors in aviation equipment, equipment systems, and the working environment, as well as an offline knowledge graph containing fault knowledge and reliability information. A flexible knowledge graph dynamic update model based on human-machine integration is established, continuously updating the knowledge graph by learning new scenario knowledge to adapt to the complex and ever-changing operating environment of aviation equipment. Simulating the cognitive processes of the human brain, this invention combines knowledge graphs with external reasoning techniques, proposing an aviation equipment reliability assessment algorithm that integrates a data-driven reasoning method using knowledge graph mapping to Bayesian networks and a knowledge-driven analogical reasoning method using scene graph semantic similarity matching.

[0054] This invention provides a method and system for reliability assessment of aviation equipment based on knowledge graphs and model reasoning. Utilizing knowledge graphs and Bayesian network model reasoning techniques, it achieves aviation equipment reliability assessment driven by both data and knowledge. A flexible knowledge graph knowledge flow and update mechanism is established to continuously learn new knowledge about aviation equipment operating scenarios, enabling the model to adapt well to complex and ever-changing environments. A Bayesian network reasoning model is designed, introducing probabilistic concepts to allow modeling of parameter uncertainties, overcoming the bottleneck of traditional deep learning models that rely on deterministic sample datasets for training and inference, thus improving the algorithm's environmental adaptability and self-adaptability.

[0055] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0056] 1. By utilizing knowledge graphs and deep learning technologies, a dual-driven reliability assessment of aviation equipment based on data and knowledge is achieved. Compared with purely data-driven methods, the aviation equipment reliability assessment constructed in this invention can simulate human reasoning and cognition, making the assessment results more scientific and reasonable.

[0057] 2. Establish a continuous knowledge flow and update model for the elastic knowledge graph, continuously learn new knowledge about aviation equipment working scenarios, and memorize the new knowledge into the elastic knowledge graph for easy use next time. The reliability assessment system realized through this knowledge flow and update model can adapt well to complex and ever-changing environments.

[0058] 3. By incorporating abstract relationships between concepts and expert experience information, from the perspective of "human-machine-environment", an online knowledge graph of aviation equipment corresponding to human factors, equipment status, and real-time working status of the working environment, as well as an offline knowledge graph of aviation equipment with reliability knowledge, were constructed.

[0059] 4. A data-driven reasoning method based on knowledge graph mapping Bayesian networks was designed. The random forest algorithm was used to optimize the feature data. Based on Bayesian network structure learning and parameter learning, the Bayesian reasoning model for reliability assessment of aerospace equipment systems was constructed.

[0060] 5. Based on the establishment of a graph neural network model to extract global features of the knowledge graph, and using a knowledge graph similarity analysis algorithm, indirect reasoning of the most probable faults, fault causes, and reliability was realized, providing support for the high-quality development of the aviation equipment industry. Attached Figure Description

[0061] Figure 1 is a schematic diagram of the system of the present invention;

[0062] Figure 2 is a flowchart of the evaluation method of the present invention;

[0063] Figure 3 is a schematic diagram of the physical layer of the reliability assessment model for aviation equipment. Detailed Implementation

[0064] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0065] As shown in Figure 1, this invention provides an aviation equipment reliability assessment system based on knowledge graphs and model reasoning, comprising:

[0066] Text data processing module: From the perspective of "human, machine and environment", it acquires three types of data on human factors of aviation equipment, equipment system and working environment through the cockpit vital signs monitoring system, fault investigation form, maintenance history fault repair form and Internet of Things sensor. After preprocessing and text annotation, the data is sent to the entity relationship triplet extraction module.

[0067] Entity Relationship Triple Extraction Module: This module includes a pre-constructed entity relationship joint extraction model that integrates the feature modeling of entities and relationships, while simultaneously detecting the semantic relationships between entities. It transforms the descriptive text information corresponding to the working status of aviation equipment into high-quality structured triples that can form a knowledge graph.

[0068] The elastic knowledge graph construction module includes a knowledge graph schema layer construction unit and a knowledge graph data layer construction unit. The knowledge graph schema layer construction unit describes the entity concepts in the aviation equipment working state scenario domain and the relationships between them, and is responsible for guiding the entire knowledge structure. The knowledge graph data layer construction unit includes an aviation equipment online knowledge graph data layer and an aviation equipment offline knowledge graph data layer. Based on the definition of the knowledge graph schema layer, it extracts aviation equipment entity and its relationship data to realize the semi-automatic construction of online and offline knowledge graphs for aviation equipment.

[0069] System reliability inference module: includes a Bayesian inference model construction unit and a scene graph similarity calculation unit;

[0070] Bayesian inference model building unit: Maps triple information in the online knowledge graph of aviation equipment to the corresponding node variables of the Bayesian network model. After obtaining the values ​​of the corresponding node variables, the Bayesian model is used to infer reliability.

[0071] Scene graph similarity calculation unit: Extracts semantic features from real-time online knowledge graph and offline knowledge graph of aviation equipment respectively, analyzes the similarity of features, and realizes indirect reasoning on the reliability of aviation equipment.

[0072] Based on the aforementioned aviation equipment reliability assessment system, this invention provides an aviation equipment reliability assessment method based on knowledge graphs and model reasoning. This embodiment applies the method of this invention to a specific example, referring to Figure 2, and includes the following steps:

[0073] S1: Acquire three types of data for aviation equipment: human factors, equipment systems, and working environment; human factors include vital signs, electroencephalogram (EEG), electromyography (EMG) signals, manual operation, voice interaction, and facial expressions; equipment systems include navigation systems, auxiliary power systems, air supply systems, landing gear, anti-icing and rain drainage systems, cabin doors, communication systems, and air conditioning systems; working environment includes temperature, humidity, rainfall, air pressure, and wind speed.

[0074] In this embodiment, the pilot's ECG, body temperature, heart rate, respiratory rate, blood oxygen, brain waves, and musculoskeletal status are acquired through the cockpit vital signs monitoring system. Data from various IoT sensors is collected on equipment systems such as the navigation system, auxiliary power system, air supply system, landing gear assembly, anti-icing and rain drainage system, cabin doors, communication system, and air conditioning system, as well as environmental data including but not limited to voltage, mechanical stress, vibration, load, pollution conditions, temperature, humidity, air pressure, wind speed, and rainfall. Static data on the equipment itself (attributes, status, capabilities), and dynamic event data such as equipment inspections and maintenance are obtained from fault investigation reports, maintenance records, and other text files.

[0075] S2: Preprocess and annotate the data obtained in step S1;

[0076] In this embodiment, the aviation equipment data is preprocessed by data cleaning, data conversion, and normalization. The preprocessed text is stored in .csv format to obtain a text corpus describing the system's working status in natural language. It is necessary to annotate the entities and corresponding relationships in the text corpus to lay the foundation for constructing a knowledge graph. Based on an in-depth study of the characteristics of aviation equipment data and combined with the needs of reliability assessment, the entity categories and corresponding relationship categories in the aviation equipment dataset are determined, as shown in Table 1.

[0077] Table 1. Categories of Aviation Equipment Entities and Corresponding Categories

[0078] The triplet annotation tool LAnn was used to annotate the text with entity relations. The annotated data was converted into the form of SPO triples (Subject-Predicate-Object). Each data point contains the original text and the corresponding triple sequence. Examples of triple data are shown in Table 2.

[0079] Table 2 Example of triplet data

[0080] In Table 2, text represents the original text data, sop_list represents the contents of the triples, each triple is represented by a bracket, predicate represents the predicted relation, subject and object represent two entities respectively, and subject_type and object_type represent the corresponding entity types respectively;

[0081] S3: Input the annotated text information from step S2 into the constructed entity relation joint extraction model to form high-quality structured triples of the knowledge graph;

[0082] A triple is composed of either <entity 1, relation, entity 2> or <entity, attribute, attribute value>. A triple consists of an entity and the relation between the entities: <entity 1 - relation - entity 2>. The basic building block of a knowledge graph is the triple, and forming triples lays the foundation for constructing a flexible knowledge graph for aerospace equipment.

[0083] Since triples are the basic building blocks of knowledge graphs, they are directly related to semantic feature extraction. Semantic feature extraction of knowledge graphs refers to extracting features such as distance, path length, and common neighbors of entities within the knowledge graph structure.

[0084] The entity relation joint extraction model uses a BiLSTM+CRF algorithm to extract triples from text corpora. The BiLSTM integrates two LSTM layers with opposite learning directions. Through a large amount of labeled data and continuous model iteration, fast and efficient entity relation extraction can be achieved.

[0085] The specific process of extracting triples from text corpora using the entity relation joint extraction method based on BiLSTM+CRF algorithm is as follows:

[0086] Input text sequence X = {x1, x2, ..., x} t}, single time step input x t After being represented by a word-level embedding layer, it is input into the forward LSTM unit for computation. The inverted LSTM unit calculates The output h′ of the hidden layer t yes and The splicing; BiLSTM is represented as:

[0087] h′ t Input is fed into a linearly linked CRF layer; given input and the label set y = {l1, l2, l3, l4, ..., l t The probability of sequence label Y is:

[0088] In equation (2), Z′ is the normalization factor, and y′ represents any label sequence. It is for l t A dedicated model parameter matrix, the same It is for l t A proprietary model bias parameter matrix;

[0089] For the decoding process, the Viterbi algorithm is used to obtain the Viterbi result sequence and Viterbi result value, and the position of the predicted entity word is obtained based on the Viterbi result sequence.

[0090] S4: Based on the triples, construct a flexible knowledge graph for aviation equipment, including an online knowledge graph and an offline knowledge graph with aviation equipment reliability features;

[0091] The aerospace equipment resilient knowledge graph comprises a schema layer and a data layer. The data layer includes an online knowledge graph and an offline knowledge graph. The offline knowledge graph is equivalent to a knowledge graph storing the reliability of the aerospace equipment system under different scenarios, much like the human brain stores knowledge. The online knowledge graph corresponds to the operational status information of the aerospace equipment during real-time reliability assessment. The reliability of the aerospace equipment system in real-time can be judged based on the information in the offline knowledge graph. The entire process is equivalent to the process of using and updating the offline knowledge graph for aerospace equipment reliability assessment, completing both aerospace equipment reliability reasoning and offline knowledge graph updates. This represents a continuous fusion of human intelligence and machine learning.

[0092] The specific construction method of the aviation equipment elastic knowledge graph is as follows:

[0093] A1: The schema layer for constructing a flexible knowledge graph of aviation equipment; the schema layer describes the entity concepts in the aviation equipment working state scenario domain and the relationships between them, and is responsible for guiding the entire knowledge structure;

[0094] As shown in Figure 3, the model layer in this embodiment includes three types of data: human factors of aviation equipment, equipment systems, and working environment. Human factors include vital signs, electroencephalogram (EEG), electromyography (EMG) signals, manual operation, voice interaction, and facial expressions. Equipment systems include navigation systems, auxiliary power systems, air supply systems, landing gear, anti-icing and rain drainage systems, cabin doors, communication systems, and air conditioning systems. The working environment includes temperature, humidity, rainfall, air pressure, and wind speed.

[0095] In this embodiment, Protégé Version 5.5 software is used to construct the aircraft equipment ontology model. The aircraft equipment entities are integrated into the constructed ontology model, and the abstract relationships of various entity concepts are added to the ontology model to form a new pattern layer, which can be updated later.

[0096] A2: Construct an online knowledge graph for the aviation equipment data layer;

[0097] Based on the pattern layer, the text corresponding to the three types of data in the real-time scene of aviation equipment—human factors, equipment system, and working environment—is extracted by the trained extraction model, followed by triple extraction, knowledge fusion, and quality assessment. The extracted text is then stored in the Neo4j graph database to complete the construction of an online knowledge graph.

[0098] A3: Construct an offline knowledge graph for the aviation equipment data layer;

[0099] The offline knowledge graph mainly targets multiple different aviation equipment information, using historical human factors, equipment systems, working environment, fault investigation reports, maintenance history, and fault repair reports as the original data sources. It mainly selects aviation equipment text corpora corresponding to different fault types in multiple working scenarios, and follows the construction process of online knowledge graphs to obtain an offline knowledge graph of aviation equipment with reliability and fault data.

[0100] S5: Extract semantic features from online knowledge graphs and offline knowledge graphs with aviation equipment reliability information, analyze the similarity of the extracted features, and realize indirect inference of aviation equipment reliability.

[0101] In this embodiment, the GraphSAGE model is used to extract semantic features from the online and offline knowledge graphs of aviation equipment, and the semantic similarity of the graphs is analyzed. Specifically:

[0102] After extracting global feature vectors from the online knowledge graph and scene graph using the GraphSAGE model, two feature vectors V1 and V2 are obtained. The cosine value of the two vectors is calculated to transform the nonlinear difference into a difference in vector value and direction, thereby obtaining the similarity value. The similarity calculation formula is shown in the following formula:

[0103] In equation (3), V1 represents the global semantic feature vector corresponding to the sub-scene graph in the offline knowledge graph of aviation equipment; V2 represents the global semantic feature vector corresponding to the online knowledge graph; cosθ represents the cosine value of the two vectors;

[0104] Based on experience, a similarity threshold of T is set. When the Simc similarity value is greater than the threshold T, it indicates that the two graphs match; when the Simc similarity value is less than the threshold T, it indicates that the two graphs do not match. If the graphs match successfully, the reliability of the scenario with the highest matching degree is selected as the reliability of the real-time scenario. If the match fails, the real-time aviation equipment operating status is treated as a new scenario, and the reliability of the scenario is inferred using a Bayesian network model. Finally, the accuracy of the estimated aviation equipment system reliability is analyzed in conjunction with expert experience. If the estimation is incorrect, the online knowledge graph of the aviation equipment for the new scenario is discarded. If the estimation is correct, the online knowledge graph of the new scenario with aviation equipment reliability is numbered and integrated into the offline knowledge graph for use in the next scenario graph matching, thereby updating the offline knowledge graph.

[0105] Bayesian networks are network reasoning models based on probability theory and graph theory. Following a human reasoning approach, they infer the target variable from relevant variables. Relying on multiple factors, they are generally used for probabilistic and uncertain events. Reliability assessment of aviation equipment falls under the category of probabilistic events, and the reliability of aviation equipment can be estimated by using the probability inferred from the Bayesian network model. In this embodiment, the training and operation method of the Bayesian network model is as follows:

[0106] B1: Acquisition of Bayesian network training data;

[0107] Before learning the structure and parameters of the Bayesian network, the training data must first be obtained. The selected training data mainly comes from the offline knowledge graph of aviation equipment system reliability, which contains 9 types of faults, 3 working scenarios for each type, and a total of 27 aviation equipment knowledge graphs. The information of different graphs is mapped to the corresponding input variables of the Bayesian network, and the variable values ​​of the corresponding variables are obtained. The label of the target variable, the fault type, is added to the end of the variable value to obtain the raw data corresponding to different fault scenarios. In order to facilitate the import of data during the subsequent learning of the structure and parameters of the Bayesian network, it is necessary to discretize each variable.

[0108] B2: Before training the Bayesian network model, to improve the model's generalization ability and interpretability while avoiding overfitting, a random forest algorithm is used to optimize the feature dimensionality of the training data; specifically as follows:

[0109] Suppose we have sample data k = 1, 2, 3, ..., K, where K represents the number of samples, and each sample has N-dimensional features. The steps for calculating the feature importance ranking are as follows:

[0110] 1) Initialize k=1 and create decision tree T k ;

[0111] 2) Train T on the k-th sample kCalculate the classification accuracy for each data point.

[0112] 3) Randomly select features X from the dataset j Apply a perturbation to j = 1, 2, ..., N, and recalculate.

[0113] 4) Repeat steps 1) to 3) for k = 1, 2, 3, ..., K;

[0114] 5) Feature X j Importance metric P j Calculated using the following formula

[0115] In equation (4): P j Representing feature X j Importance metrics This represents the classification accuracy of data without perturbation. This represents the data classification accuracy recalculated after perturbing the j-th feature;

[0116] 6) For P j Sort the features in descending order to obtain the feature importance ranking, and select the top N features as the input parameters for training the Bayesian network;

[0117] B3: The Bayesian model structure is learned using a search scoring method, and the parameters of the Bayesian network are learned using a Bayesian estimation method.

[0118] B4: Apply the trained Bayesian model to reliability inference for aerospace equipment systems to obtain reliability assessment results. The formula for reliability inference is:

[0119] In equation (5), X1,…,X N Let N be the number of nodes in a Bayesian network, and π(X) be the number of nodes. i ) is node X i The parent node, p i (X i |π(X i )) represents the conditional probability of a node in a Bayesian network model.

[0120] In this embodiment, the reliability of aviation equipment is integrated into an offline knowledge graph. Expert experience is used to analyze the reliability assessment results of aviation equipment in each sub-scenario. If the estimation is correct, the system reliability value of the aviation equipment inferred by the model is integrated into the corresponding sub-scenario in the form of a triple [System, Reliability, reliability value], thus obtaining an offline knowledge graph of aviation equipment with system reliability. This prepares the groundwork for subsequent semantic similarity calculation of the aviation equipment graph. If the reliability estimation of some sub-scenarios is incorrect, these incorrectly estimated sub-scenario graphs are discarded.

[0121] In this embodiment, the similarity calculation results are arranged from largest to smallest. If the similarity exceeds a threshold, the top-ranked matching result is taken as the reliability of the aviation equipment in that scenario. For example, Table 3 shows the top three similarity results between a certain aviation equipment's real-time online knowledge graph scenario and each sub-scenario of the offline knowledge graph. With a similarity threshold of 0.8, only sub-scenario graphs KG_13 and KG_26 are similar to the offline knowledge graph scenario.

[0122] Table 3. Similarity results between real-time online knowledge graph scenarios and offline knowledge graph scenarios.

[0123] If the most similar sub-scene graph KG_13 is selected from the similar sub-scene graphs, then the reliability of the aerospace equipment is 0.7016. The most likely failure is the failure of the aircraft landing gear wheel bearing, and the cause of the failure is the lack of lubrication of the main landing gear wheel axle.

Claims

1. A reliability assessment method for aviation equipment based on knowledge graphs and model reasoning, characterized in that, Includes the following steps: S1: Acquire three types of data for aviation equipment: human factors, equipment systems, and working environment; S2: Preprocess and annotate the data obtained in step S1; S3: Input the annotated text information from step S2 into the constructed entity relation joint extraction model to form structured triples of the knowledge graph; S4: Construct a resilient knowledge graph for aviation equipment, including an online knowledge graph and an offline knowledge graph with aviation equipment reliability features; S5: Extract semantic features from online knowledge graphs and offline knowledge graphs with aviation equipment reliability information, analyze the similarity of extracted features, and realize indirect inference of aviation equipment reliability.

2. The method for reliability assessment of aviation equipment based on knowledge graphs and model reasoning according to claim 1, characterized in that, In step S1, human factors include vital signs, electroencephalogram (EEG), electromyography (EMG) signals, manual operation, voice interaction, and facial expressions; equipment systems include navigation systems, auxiliary power systems, air supply systems, landing gear, anti-icing and rain drainage systems, cabin doors, communication systems, and air conditioning systems; and the working environment includes temperature, humidity, rainfall, air pressure, and wind speed.

3. The method for reliability assessment of aviation equipment based on knowledge graphs and model reasoning according to claim 1, characterized in that, Step S2 specifically involves: The aviation equipment data is preprocessed by data cleaning, data transformation, and standardization. The preprocessed text is stored in .csv format to obtain a text corpus of natural language description of the system's working status. The triple joint annotation tool LAnn is used to annotate the text with entity relations. The annotated data is converted into the form of SPO triples, and each data point contains the original text and the corresponding triple sequence.

4. The method for reliability assessment of aviation equipment based on knowledge graphs and model reasoning according to claim 1, characterized in that, In step S3, the entity relation joint extraction model uses a BiLSTM+CRF algorithm-based entity relation joint extraction method to extract triples from the text corpus. BiLSTM integrates two sets of LSTM layers with opposite learning directions, enabling fast and efficient entity relationship extraction through continuous iteration of a large amount of labeled data and the model.

5. The method for reliability assessment of aviation equipment based on knowledge graphs and model reasoning according to claim 4, characterized in that, The specific process of extracting triples from the text corpus using the entity relation joint extraction method based on the BiLSTM+CRF algorithm in step S3 is as follows: Input text sequence X = {x1, x2, ..., x} t }, single time step input x t After being represented by a word-level embedding layer, it is input into the forward LSTM unit for computation. The inverted LSTM unit calculates The output h' of the hidden layer t yes and The splicing; BiLSTM is represented as: h' t Input is fed into a linearly linked CRF layer; given input and the label set y = {l1, l2, l3, l4, ..., l t The probability of sequence label Y is: In equation (2), Z' is the normalization factor, and y' represents any label sequence. It is for l t A dedicated model parameter matrix, the same It is for l t A proprietary model bias parameter matrix; For the decoding process, the Viterbi algorithm is used to obtain the Viterbi result sequence and Viterbi result value, and the position of the predicted entity word is obtained based on the Viterbi result sequence.

6. The method for reliability assessment of aviation equipment based on knowledge graphs and model reasoning according to claim 1, characterized in that, The aviation equipment elastic knowledge graph in step S4 includes a schema layer and a data layer, and the specific construction method is as follows: A1: The schema layer for constructing a flexible knowledge graph of aviation equipment; the schema layer describes the entity concepts in the aviation equipment working state scenario domain and the relationships between them, and is responsible for guiding the entire knowledge structure; A2: Construct an online knowledge graph for the aviation equipment data layer; Based on the pattern layer, the text corresponding to the three types of data in the real-time scene of aviation equipment—human factors, equipment system, and working environment—is extracted by the trained extraction model, followed by triple extraction, knowledge fusion, and quality assessment. The extracted text is then stored in the Neo4j graph database to complete the construction of an online knowledge graph. A3: Construct an offline knowledge graph for the aviation equipment data layer; The offline knowledge graph targets multiple different aviation equipment information sources, using historical human factors, equipment systems, working environments, fault investigation reports, maintenance records, and fault repair reports as raw data sources. It selects aviation equipment text corpora corresponding to different fault types in multiple working scenarios and follows the construction process of online knowledge graphs to obtain an offline knowledge graph of aviation equipment with reliability and fault data.

7. The method for reliability assessment of aviation equipment based on knowledge graphs and model reasoning according to claim 1, characterized in that, In step S5, the GraphSAGE model is used to extract semantic features from the online and offline knowledge graphs of aviation equipment, and the semantic similarity of the graphs is analyzed. Specifically: After extracting global feature vectors from the online knowledge graph and scene graph using the GraphSAGE model, two feature vectors V1 and V2 are obtained. The cosine value of the two vectors is calculated to transform the nonlinear difference into a difference in vector value and direction, thereby obtaining the similarity value. The similarity calculation formula is shown in the following formula: In equation (3), V1 represents the global semantic feature vector corresponding to the sub-scene graph in the offline knowledge graph of aviation equipment; V2 represents the global semantic feature vector corresponding to the online knowledge graph; cosθ represents the cosine value of the two vectors; A similarity threshold of T is set. When the Simc similarity value is greater than the threshold T, it indicates that the two maps match. When the Simc similarity value is less than the threshold T, it indicates that the two maps do not match.

8. The method for reliability assessment of aviation equipment based on knowledge graphs and model reasoning according to claim 7, characterized in that, In step S5, if the graph matching is successful, the reliability of the scenario with the highest matching degree is selected as the reliability of the real-time scenario; if the matching is unsuccessful, the real-time aviation equipment working state is treated as a new scenario, and the reliability of the scenario is inferred using a Bayesian network model; finally, the accuracy of the estimated aviation equipment system reliability is analyzed in combination with expert experience, and if the estimation is wrong, the aviation equipment online knowledge graph of the new scenario is discarded. If the estimation is correct, the online knowledge graph of new scenarios with aviation equipment reliability will be numbered and integrated into the offline knowledge graph so that it can be used in the next scenario graph matching to update the offline knowledge graph.

9. The method for reliability assessment of aviation equipment based on knowledge graphs and model reasoning according to claim 8, characterized in that, The training and running method of the Bayesian network model is as follows: B1: Acquisition of Bayesian network training data; The selected training data comes from the offline knowledge graph of aviation equipment system reliability, which contains 9 types of faults, 3 working scenarios for each type, and a total of 27 aviation equipment knowledge graphs. The information of different graphs is mapped to the corresponding Bayesian network input variables to obtain the variable values ​​of the corresponding variables. The label of the target variable, the fault type, is added to the end of the variable value to obtain the original data corresponding to different fault scenarios. B2: Before training the Bayesian network model, to improve the model's generalization ability and interpretability while avoiding overfitting, a random forest algorithm is used to optimize the feature dimensionality of the training data; specifically as follows: Suppose we have sample data k = 1, 2, 3, ..., K, where K represents the number of samples, and each sample has N-dimensional features. The steps for calculating the ranking of feature importance are as follows: 1) Initialize k=1 and create decision tree T k ; 2) Train T on the k-th sample k Calculate the classification accuracy for each data point. 3) Randomly select features X from the dataset j Apply a perturbation to j = 1, 2, ..., N, and recalculate. 4) Repeat steps 1) to 3) for k = 1, 2, 3, ..., K; 5) Feature X j Importance metric P j Calculated using the following formula In equation (4): P j Representing feature X j Importance metrics This represents the classification accuracy of data without perturbation. This represents the data classification accuracy recalculated after perturbing the j-th feature; 6) For P j Sort the features in descending order to obtain the feature importance ranking, and select the top N features as the input parameters for training the Bayesian network; B3: The Bayesian model structure is learned using a search scoring method, and the parameters of the Bayesian network are learned using a Bayesian estimation method. B4: Apply the trained Bayesian model to reliability reasoning of aerospace equipment systems to obtain reliability assessment results.

10. A reliability assessment system for aviation equipment based on knowledge graphs and model reasoning, characterized in that, include: Text data processing module: From the perspective of "human-machine-environment", it acquires three types of data on human factors of aviation equipment, equipment system and working environment through the cockpit vital signs monitoring system, fault investigation form, maintenance history fault repair form and Internet of Things sensor. After preprocessing and text annotation, the data is sent to the entity relationship triplet extraction module. Entity Relationship Triple Extraction Module: This module includes a pre-constructed entity relationship joint extraction model that integrates the feature modeling of entities and relationships, while simultaneously detecting the semantic relationships between entities. It transforms the descriptive text information corresponding to the working status of aviation equipment into high-quality structured triples that can form a knowledge graph. The elastic knowledge graph construction module includes a knowledge graph schema layer construction unit and a knowledge graph data layer construction unit. The knowledge graph schema layer construction unit describes the entity concepts in the aviation equipment working state scenario domain and the relationships between them, and is responsible for guiding the entire knowledge structure. The knowledge graph data layer construction unit includes an aviation equipment online knowledge graph data layer and an aviation equipment offline knowledge graph data layer. Based on the definition of the knowledge graph schema layer, it extracts aviation equipment entity and its relationship data to realize the semi-automatic construction of online and offline knowledge graphs for aviation equipment. System reliability inference module: includes a Bayesian inference model construction unit and a scene graph similarity calculation unit; Bayesian inference model building unit: Maps triple information in the online knowledge graph of aviation equipment to the corresponding node variables of the Bayesian network model. After obtaining the values ​​of the corresponding node variables, the Bayesian model is used to infer reliability. Scene graph similarity calculation unit: Extracts semantic features from real-time online knowledge graph and offline knowledge graph of aviation equipment respectively, analyzes the similarity of features, and realizes indirect reasoning on the reliability of aviation equipment.

Images