An aircraft semantic knowledge structuring method and system

Abstract

The application belongs to the technical field of aircraft design, and particularly relates to an aircraft semantic knowledge structuring method and system, wherein the aircraft semantic knowledge structuring method comprises the following steps: step one, carrying out paragraph splitting on an aircraft design document; the aircraft design document comprises a task description, a general scheme, a system analysis and an experimental data document; step two, predicting a design stage corresponding to the split paragraph, and constructing a design stage label alignment structure; the design stage comprises a task definition, a general scheme, an aerodynamic design, a structural design, a system analysis and a test verification stage; step three, carrying out semantic and design parameter extraction on the design stage label alignment structure; and step four, constructing a knowledge graph based on the extracted semantics and design parameters.

Description

Technical Field

[0001] This application belongs to the field of aircraft design technology, specifically relating to a method and system for structuring semantic knowledge of aircraft. Background Technology

[0002] With the deep integration of artificial intelligence technology and aerospace engineering, knowledge-driven intelligent design is becoming an important trend in aircraft system development. Especially in the development of modern aircraft characterized by strong mission adaptability, rapid design iteration, and high structural complexity, fully leveraging the wealth of engineering experience and semantic knowledge contained in the design phase and constructing a knowledge structure system that supports automatic reasoning and optimization decision-making has become an important foundation for intelligent aircraft design.

[0003] Aircraft design knowledge is largely scattered across unstructured documents, including design specifications, simulation reports, and test records. This textual content is often verbose, loosely structured, and semantically inconsistent, severely hindering the efficiency of automated knowledge acquisition, structured management, and reuse. Traditional design knowledge modeling methods rely heavily on manual annotation or expert rules, resulting in low extraction efficiency, insufficient semantic coverage, and missing knowledge evolution paths. These methods fail to meet the demands for efficient modeling and intelligent collaboration in complex design tasks.

[0004] Some studies have attempted to introduce natural language processing methods for keyword extraction and entity recognition in design texts. However, in the context of aircraft design, there are problems such as vague expressions of technical terms, vague parameter descriptions, and semantic distribution across paragraphs, which often lead to coarse-grained knowledge extraction and insufficient structural expression capabilities. At the same time, existing methods often ignore the phased characteristics of the design process and the evolutionary path of knowledge formed by the iteration of schemes, and cannot truly reflect the structural logic and evolutionary relationship of aircraft design knowledge.

[0005] In view of the aforementioned technical deficiencies, this application is hereby filed. Summary of the Invention

[0006] The purpose of this application is to provide a method and system for structuring semantic knowledge of aircraft, in order to overcome or mitigate at least one of the known technical deficiencies.

[0007] The technical solution of this application is:

[0008] A method for structuring semantic knowledge of aircraft, comprising:

[0009] Step 1: Break the aircraft design document into paragraphs;

[0010] The aircraft design document includes mission description, overall scheme, system analysis and experimental data documents;

[0011] Step 2: Predict the design stages corresponding to the split paragraphs and construct a design stage marker alignment structure;

[0012] The design phase includes task definition, overall scheme, aerodynamic design, structural design, system analysis, and experimental verification.

[0013] Step 3: Extract semantics and design parameters from the design phase marker alignment structure;

[0014] Step 4: Construct a knowledge graph based on the extracted semantics and design parameters.

[0015] Optionally, in the above-mentioned aircraft semantic knowledge structuring method, in step two, the set of split paragraphs is input into the semantic stage structure alignment model. The semantic stage structure alignment model adopts the RoBERTa variant structure to predict the design stage corresponding to the split paragraphs, and introduces common words and key prompts in the design stage of the aircraft development process for supervision.

[0016] Optionally, in the above-mentioned aircraft semantic knowledge structuring method, in step two, the semantic stage structure alignment model, the segmented paragraphs are encoded with BERT, and then a classifier with Softmax added on top of RoBERTa predicts the corresponding design stage, thus constructing a design stage label alignment structure.

[0017] Optionally, in the above-mentioned aircraft semantic knowledge structuring method, in step two, a contextual semantic boundary detection mechanism is introduced into the semantic stage structure alignment model. This mechanism, combined with syntactic dependency analysis and BERT embedding local consistency scoring, improves the accuracy of identifying the boundaries of different design stages of the segmented paragraphs.

[0018] Optionally, in the above-mentioned aircraft semantic knowledge structuring method, in step three, the design phase marker alignment structure is input into the semantic processing model to extract semantics and design parameters.

[0019] The semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, identifies the content that conforms to the "design premise-design entity-response solution" template, and extracts it into structured triples.

[0020] When the semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, it adopts a BERT encoding + BiLSTM + CRF structure and introduces a structure guidance token to improve the boundary recognition capability.

[0021] Optionally, in the above-mentioned aircraft semantic knowledge structuring method, in step three, the semantic processing model extracts design parameters from the design stage marker alignment structure, identifies the content expressed by the numerical values ​​of the design parameters, extracts the minimum value vmin and the maximum value vmax of the design parameters, and generates a confidence score p to perform design parameter interval modeling.

[0022] When the semantic processing model extracts design parameters from the marker-aligned structure in the design phase, for single-valued design parameter values, the minimum value vmin and maximum value vmax of the extracted design parameter are equal. For fuzzy-described design parameter values, the minimum value vmin and maximum value vmax of the design parameter are calculated based on the typical redundancy coefficient learned from the context.

[0023] When the semantic processing model extracts design parameters for the marker-aligned structure in the design phase, it performs BERT encoding on the text expressing the numerical content of the design parameters to obtain the embedding vector H, which is then input into two independent fully connected networks for predicting the minimum value vmin and the maximum value vmax, and into a Sigmoid network for predicting the confidence level p.

[0024] Optionally, in the above-mentioned aircraft semantic knowledge structuring method, in step three, a back-translation model or a rewriting model is used to perform structure-preserving semantic enhancement on the split paragraphs, generating equivalent but different domain descriptions, generating 2-3 semantically equivalent but different sentence variations for each split paragraph, obtaining diverse but structurally consistent expressions, and constructing an enhanced corpus set;

[0025] When enhancing the contextual diversity of the marker alignment structure in the design phase, a term freezing mechanism is introduced;

[0026] A pre-trained language model based on the Transformer architecture is selected, and an enhanced corpus is used to fine-tune the semantic processing model for multiple tasks in the field of aircraft design. The goal of the fine-tuning is to minimize the comprehensive loss function S.

[0027] S=LMLM+λ×Ldomain+μ×Lstage;

[0028] in,

[0029] LMLM is the loss for masked language modeling;

[0030] Ldomain represents the design entity recognition loss;

[0031] Lstage represents the label constraint loss during the design phase;

[0032] λ and μ are the task weight coefficients.

[0033] Optionally, in the above-mentioned aircraft semantic knowledge structuring method, in step four, the extracted semantics and design parameters are input into the knowledge graph construction model to construct the knowledge graph;

[0034] Knowledge graphs consist of entity sets E, relationships between entities R, and evolution path graphs P, stored in the form of graph databases or nested key-value structures, supporting retrieval and reasoning.

[0035] Optionally, in the above-mentioned aircraft semantic knowledge structuring method, in step four, a semantic embedding mechanism is used to convert all identified design elements into nodes of a knowledge graph.

[0036] Calculate the semantic consistency score between knowledge graph nodes:

[0037] Sim(i,j)=cos(i_vec, j_vec)×δ_structure×γ_context×ω_attribute;

[0038] Sim(i,j) is the semantic consistency score between knowledge graph nodes i and j;

[0039] i_vec and j_vec are the embedding vectors generated by BERT encoding of knowledge graph nodes i and j;

[0040] δ_structure is a Boolean term indicating whether the structure categories of knowledge graph nodes are consistent;

[0041] γ_context represents the similarity of knowledge graph nodes within their context.

[0042] ω_attribute represents the consistency rate of the attribute fields of the knowledge graph nodes;

[0043] If Sim(i,j) exceeds the consistency threshold, then the knowledge graph nodes i and j are fused.

[0044] The evolutionary path graph P is embedded into the knowledge graph in the form of a directed subgraph.

[0045] Optionally, in the above-mentioned aircraft semantic knowledge structuring method, step four involves calculating a knowledge graph quality score to evaluate the quality of the knowledge graph.

[0046] Q(G)=α×C(G)+β×P(G)+γ×Con(G);

[0047] in,

[0048] Q(G) represents the quality score of the knowledge graph;

[0049] C(G) represents the knowledge graph coverage;

[0050] P(G) represents the accuracy of the knowledge graph;

[0051] Con(G) represents the consistency of the knowledge graph;

[0052] α, β, and γ are weights.

[0053] A semantic knowledge structuring system for aircraft, comprising:

[0054] The paragraph splitting module is used to split aircraft design documents into paragraphs;

[0055] The aircraft design document includes mission description, overall scheme, system analysis and experimental data documents;

[0056] The design phase marker alignment module is used to predict the design phase corresponding to the split paragraphs and construct the design phase marker alignment structure.

[0057] The design phase includes task definition, overall scheme, aerodynamic design, structural design, system analysis, and experimental verification.

[0058] The semantic and design parameter extraction module is used to extract semantic and design parameters from the marker alignment structure in the design phase.

[0059] The knowledge graph construction module is used to build knowledge graphs based on extracted semantics and design parameters.

[0060] Optionally, in the above-mentioned aircraft semantic knowledge structuring system, the design phase marker alignment module inputs the set of split paragraphs into the semantic phase structure alignment model. The semantic phase structure alignment model adopts the RoBERTa variant structure to predict the design phase corresponding to the split paragraphs, and introduces common words and key prompts in the design phase of the aircraft development process for supervision.

[0061] Optionally, in the above-mentioned aircraft semantic knowledge structuring system, in the design phase label alignment module, in the semantic phase structure alignment model, the split paragraphs are encoded with BERT, and then a classifier with Softmax added on top of RoBERTa predicts the corresponding design phase, thus constructing the design phase label alignment structure.

[0062] Optionally, in the above-mentioned aircraft semantic knowledge structuring system, the semantic stage structure alignment model in the design stage marker alignment module introduces a contextual semantic boundary detection mechanism, which, combined with syntactic dependency analysis and BERT embedding local consistency scoring, improves the accuracy of identifying the boundaries of different design stages of the segmented paragraphs.

[0063] Optionally, in the above-mentioned aircraft semantic knowledge structuring system, the semantic and design parameter extraction module inputs the design phase marker alignment structure into the semantic processing model to extract semantics and design parameters.

[0064] The semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, identifies the content that conforms to the "design premise-design entity-response solution" template, and extracts it into structured triples.

[0065] When the semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, it adopts a BERT encoding + BiLSTM + CRF structure and introduces a structure guidance token to improve the boundary recognition capability.

[0066] Optionally, in the above-mentioned aircraft semantic knowledge structuring system, in the semantic and design parameter extraction module, the semantic processing model extracts design parameters from the design stage marker alignment structure, identifies the content expressed by the numerical values ​​of the design parameters, extracts the minimum value vmin and the maximum value vmax of the design parameters, and generates a confidence level p to perform design parameter interval modeling.

[0067] When the semantic processing model extracts design parameters from the marker-aligned structure in the design phase, for single-valued design parameter values, the minimum value vmin and maximum value vmax of the extracted design parameter are equal. For fuzzy-described design parameter values, the minimum value vmin and maximum value vmax of the design parameter are calculated based on the typical redundancy coefficient learned from the context.

[0068] When the semantic processing model extracts design parameters for the marker-aligned structure in the design phase, it performs BERT encoding on the text expressing the numerical content of the design parameters to obtain the embedding vector H, which is then input into two independent fully connected networks for predicting the minimum value vmin and the maximum value vmax, and into a Sigmoid network for predicting the confidence level p.

[0069] Optionally, in the above-mentioned aircraft semantic knowledge structuring system, the semantic and design parameter extraction module uses a back-translation model or a rewriting model to perform structure-preserving semantic enhancement on the split paragraphs, generating equivalent but different domain descriptions, generating 2-3 semantically equivalent but different sentence variations for each split paragraph, obtaining diverse but structurally consistent expressions, and constructing an enhanced corpus set;

[0070] When enhancing the contextual diversity of the marker alignment structure in the design phase, a term freezing mechanism is introduced;

[0071] A pre-trained language model based on the Transformer architecture is selected, and an enhanced corpus is used to fine-tune the semantic processing model for multiple tasks in the field of aircraft design. The goal of the fine-tuning is to minimize the comprehensive loss function S.

[0072] S=LMLM+λ×Ldomain+μ×Lstage;

[0073] in,

[0074] LMLM is the loss for masked language modeling;

[0075] Ldomain represents the design entity recognition loss;

[0076] Lstage represents the label constraint loss during the design phase;

[0077] λ and μ are the task weight coefficients.

[0078] Optionally, in the above-mentioned aircraft semantic knowledge structuring system, the extracted semantics and design parameters are input into the knowledge graph construction model in the knowledge graph construction module to construct the knowledge graph.

[0079] Knowledge graphs consist of entity sets E, relationships between entities R, and evolution path graphs P, stored in the form of graph databases or nested key-value structures, supporting retrieval and reasoning.

[0080] Optionally, in the above-mentioned aircraft semantic knowledge structuring system, the knowledge graph construction module uses a semantic embedding mechanism to convert all identified design elements into nodes of the knowledge graph;

[0081] Calculate the semantic consistency score between knowledge graph nodes:

[0082] Sim(i,j)=cos(i_vec, j_vec)×δ_structure×γ_context×ω_attribute;

[0083] Sim(i,j) is the semantic consistency score between knowledge graph nodes i and j;

[0084] i_vec and j_vec are the embedding vectors generated by BERT encoding of knowledge graph nodes i and j;

[0085] δ_structure is a Boolean term indicating whether the structure categories of knowledge graph nodes are consistent;

[0086] γ_context represents the similarity of knowledge graph nodes within their context.

[0087] ω_attribute represents the consistency rate of the attribute fields of the knowledge graph nodes;

[0088] If Sim(i,j) exceeds the consistency threshold, then the knowledge graph nodes i and j are fused.

[0089] The evolutionary path graph P is embedded into the knowledge graph in the form of a directed subgraph.

[0090] Optionally, in the aforementioned aircraft semantic knowledge structuring system, the knowledge graph construction module calculates a knowledge graph quality score to evaluate the quality of the knowledge graph.

[0091] Q(G)=α×C(G)+β×P(G)+γ×Con(G);

[0092] in,

[0093] Q(G) represents the quality score of the knowledge graph;

[0094] C(G) represents the knowledge graph coverage;

[0095] P(G) represents the accuracy of the knowledge graph;

[0096] Con(G) represents the consistency of the knowledge graph;

[0097] α, β, and γ are weights.

[0098] This application has at least the following beneficial technical effects:

[0099] This paper presents a method and system for structuring semantic knowledge of aircraft. It integrates semantic segmentation, semantic template extraction, parameter range modeling, evolution path construction, and knowledge graph construction to achieve automatic identification, structured expression, and dynamic evolution modeling of implicit semantic knowledge in aircraft design documents. This improves the organization efficiency, expression accuracy, and reasoning support capabilities of design knowledge, significantly enhancing the accuracy, maintainability, and intelligent reasoning capabilities of aircraft design knowledge modeling. It can be widely applied to scenarios such as aircraft design process management, intelligent design assistance systems, engineering knowledge base construction, and task-oriented design scheme derivation. While improving the engineering expression capabilities of design knowledge, it provides intelligent semantic support and knowledge reuse capabilities for the design of complex aircraft systems, demonstrating significant engineering application value and promotion potential. Attached Figure Description

[0100] Figure 1 This is a schematic diagram of the aircraft semantic knowledge structuring method provided in the embodiments of this application;

[0101] Figure 2 This is a schematic diagram of the aircraft semantic knowledge structuring system provided in the embodiments of this application.

[0102] To better illustrate this embodiment, some content in the accompanying drawings may be omitted, enlarged, or reduced. They are for illustrative purposes only and should not be construed as limiting the scope of this application. Detailed Implementation

[0103] To make the technical solution and advantages of this application clearer, the technical solution of this application will be described in a clearer and more complete manner below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only some embodiments of this application, and are only used to explain this application, not to limit this application. It should be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, and other related parts can be referred to the general design.

[0104] Furthermore, unless otherwise defined, the technical or scientific terms used in this application description shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The word "comprising" as used in this application description indicates that the concept preceding the word encompasses the concepts listed following the word and their equivalents, without excluding other related concepts.

[0105] A method for structuring semantic knowledge of aircraft, such as Figure 1 As shown.

[0106] Step 1: Split the aircraft design document into paragraphs.

[0107] The aircraft design documents underwent document-level grammar cleaning, paragraph splitting, and the resulting collection of paragraphs was then processed. ,in, For the first Each paragraph is split.

[0108] The aircraft design document includes mission description, overall scheme, system analysis and experimental data, etc. It does not have clear design stage labels, does not use structured syntax, and contains a large amount of engineering language and technical terms.

[0109] Step 2: Predict the design stages corresponding to the split paragraphs and construct a design stage marker alignment structure.

[0110] The set of split paragraphs is input into the semantic stage structure alignment model, which adopts a RoBERTa variant structure to predict the design stage corresponding to the split paragraphs and introduces common vocabulary and key prompts in the design stage of the aircraft development process for supervision.

[0111] In the semantic stage structure alignment model, the split segments are encoded using BERT, and then a classifier with Softmax added to RoBERTa predicts the corresponding design stage, thus constructing a design stage label alignment structure. This ensures that subsequent semantic processing tasks occur within the same context window, where... For the first The design phase markers for each segment are aligned with the structure. For the first Each segment is split into paragraphs. The design phase includes task definition, overall scheme, aerodynamic design, structural design, system analysis, and experimental verification.

[0112] The semantic stage structure alignment model introduces a contextual semantic boundary detection mechanism, which combines syntactic dependency analysis and BERT embedding local consistency scoring to improve the accuracy of identifying the boundaries of different design stages of the segmented paragraphs and avoid semantic processing deviations caused by cross-stage interference of design behavior.

[0113] Step 3: Extract semantics and design parameters from the marker alignment structure in the design phase.

[0114] The design phase marker alignment structure is input into the semantic processing model for semantic and design parameter extraction.

[0115] The semantic processing model extracts semantics from the marker-aligned structure of the design phase, identifies the content that conforms to the "design premise-design entity-response scheme" template, and extracts it into structured triples. For example, for the content "in order to reduce drag, the main wing adopts a swept-angle configuration", "reducing drag" is identified as the design premise, "main wing" is the design entity, and "swept-angle configuration" is the response scheme, and then the triples are constructed.

[0116] When the semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, it adopts a BERT encoding + BiLSTM + CRF structure and introduces a structure guidance token to improve the boundary recognition capability.

[0117] The semantic processing model extracts design parameters from the marker-aligned structure in the design phase, identifies the content expressed by the numerical values ​​of the design parameters, such as "the main wing span should be controlled between 10.2 and 10.8 meters", extracts the minimum value vmin and the maximum value vmax of the design parameters, and generates a confidence score p to quantify the confidence in the accuracy of the design parameter extraction, and performs range modeling of the design parameters.

[0118] When the semantic processing model extracts design parameters for the marker-aligned structure in the design phase, for single-valued design parameters, such as "web thickness is 5cm", the minimum value vmin and maximum value vmax of the extracted design parameters are equal. For fuzzy design parameters, such as "partition length is not less than 7.5m", the minimum value vmin and maximum value vmax of the design parameters are calculated based on the typical redundancy coefficient learned from the context.

[0119] When the semantic processing model extracts design parameters for the marker-aligned structure in the design phase, it performs BERT encoding on the text expressing the numerical content of the design parameters to obtain the embedding vector H, which is then input into two independent fully connected networks for predicting the minimum value vmin and the maximum value vmax, and into a Sigmoid network for predicting the confidence level p.

[0120] The semantic processing model is trained to extract design parameters from the marker alignment structure in the design phase. The loss function adopts the Huber function to reduce the sensitivity to annotation errors in the design phase.

[0121] To improve the robustness of the semantic processing model, contextual diversity enhancement is performed on the marker alignment structure in the design phase. Back-translation or rewriting models are used to perform structure-preserving semantic enhancement on the split paragraphs, generating equivalent but different domain descriptions. For each split paragraph, 2-3 semantically equivalent but different sentence variations are generated to obtain diverse but structurally consistent expressions, and an enhanced corpus is constructed.

[0122] When enhancing the contextual diversity of the marker alignment structure in the design phase, a term freezing mechanism is introduced to prevent the proprietary expressions of key parameters, such as lift-to-drag ratio and wingspan, from being modified when generating equivalent but different domain descriptions, thus ensuring semantic consistency.

[0123] A pre-trained language model based on the Transformer architecture was selected, and an enhanced corpus was used to fine-tune the semantic processing model for multiple tasks in the field of aircraft design. The goal of the fine-tuning was to minimize the comprehensive loss function S.

[0124] S=LMLM+λ×Ldomain+μ×Lstage;

[0125] in,

[0126] LMLM is the loss for masked language modeling;

[0127] Ldomain represents the design entity recognition loss;

[0128] Lstage represents the label constraint loss during the design phase;

[0129] λ and μ are the task weight coefficients.

[0130] Step 4: Construct a knowledge graph based on the extracted semantics and design parameters.

[0131] The extracted semantics and design parameters are input into the knowledge graph construction model to construct the knowledge graph.

[0132] Knowledge graphs consist of entity sets E, relationships between entities R, and evolution path graphs P, stored in the form of graph databases or nested key-value structures, supporting retrieval and reasoning.

[0133] All identified design elements are converted into knowledge graph nodes using a semantic embedding mechanism. These knowledge graph nodes contain attribute fields, including entity source document number, context paragraph number, confidence level, and version tag.

[0134] Determining whether semantically similar nodes exist in a knowledge graph involves assessing whether two entities are similar in meaning, structure, context, and whether they belong to the same type of parameters. If similar, the two nodes are merged to avoid duplication and reduce redundancy. See the following for details:

[0135] Calculate the semantic consistency score between knowledge graph nodes:

[0136] Sim(i,j)=cos(i_vec, j_vec)×δ_structure×γ_context×ω_attribute;

[0137] Sim(i,j) is the semantic consistency score between knowledge graph nodes i and j;

[0138] i_vec and j_vec are the embedding vectors generated by BERT encoding of knowledge graph nodes i and j;

[0139] δ_structure is a Boolean term indicating whether the structure categories of knowledge graph nodes are consistent;

[0140] γ_context represents the similarity of knowledge graph nodes within their context.

[0141] ω_attribute represents the consistency rate of the attribute fields of the knowledge graph nodes.

[0142] If Sim(i,j) exceeds the consistency threshold of 0.85, then the knowledge graph nodes i and j are merged to improve the simplicity and consistency of the knowledge graph and reduce inference conflicts.

[0143] The evolution path graph P represents the ordered change relationship between design schemes. Its construction depends on the identification of design change statements. For example, when it is identified that "aerodynamic optimization has been performed on scheme Sk in scheme Sk+1", a directed edge rk is created from node scheme Sk to node scheme Sk+1, thereby constructing the evolution path graph P={(S1,r1,S2),…,(Sk,rk,Sk+1)}.

[0144] The evolution path graph P is embedded into the knowledge graph in the form of a directed subgraph, embedding meta-information such as the direction of scheme evolution, design change description, change motivation and result description, etc. It supports path query and visual backtracking based on timestamp and design logic, and is suitable for large-scale multi-scheme parallel optimization and change control management scenarios.

[0145] Calculate a quality score for the knowledge graph to evaluate its quality.

[0146] Q(G)=α×C(G)+β×P(G)+γ×Con(G);

[0147] in,

[0148] Q(G) represents the quality score of the knowledge graph;

[0149] C(G) represents the knowledge graph coverage, indicating the degree to which the knowledge graph covers the information in the aircraft design documents;

[0150] P(G) is the precision of the knowledge graph, representing the accuracy of the information in the knowledge graph;

[0151] Con(G) represents the consistency of a knowledge graph, indicating the degree to which there are no semantic conflicts or structural contradictions in the knowledge graph.

[0152] α, β, and γ are weights. Typically, α can be 0.4, β can be 0.4, and γ can be 0.2. These values ​​can be dynamically adjusted based on feedback from domain experts.

[0153] When the knowledge graph quality score is too low, such as below 90%, it will automatically identify which nodes or semantics the problem is concentrated on, generate an error analysis log and submit it to the administrator, or trigger a self-correction process to fine-tune the relevant model.

[0154] In a specific example, using the aircraft semantic knowledge structuring method disclosed in the above embodiments, a 59-page PDF overall design document of a certain type of high-speed aircraft was processed to construct a knowledge graph involving 8267 entities and 20340 edges, with a coverage of 88.6%, an accuracy of 92.3%, a consistency of 95.1%, and a quality score of 91.4%. Compared with the parsing process of unstructured extraction schemes, the model graph construction speed was increased by 3.7 times, and the manual revision cost was reduced by 62%.

[0155] The aircraft semantic knowledge structuring method disclosed in the above embodiments enables the extraction of knowledge from aircraft design documents with contextual constraints through paragraph splitting and design phase alignment, avoiding semantic confusion between different design phases and enhancing the understanding and expression of complex logical relationships in multi-stage design tasks.

[0156] By identifying and extracting the content of design premises, design entities, and response schemes from aircraft design documents, a complete semantic triple system is established, enabling structured expression and centralized management of knowledge. This significantly improves the automation level and semantic accuracy of knowledge acquisition.

[0157] By identifying the content expressed by the numerical values ​​of design parameters, predicting and extracting the minimum value vmin, maximum value vmax and their confidence level P of the design parameters, we can model the description of a large number of parameters with engineering uncertainties or semantic ambiguity in aircraft design, which can effectively improve the quantitative ability of knowledge expression and engineering adaptability.

[0158] By constructing an evolution path diagram, the ability to visualize and trace the design change process is enhanced, making it easier to identify the evolution logic and change relationships of the design scheme. It can support reasoning analysis based on the causal logic of the design and provide important support for parallel design, comparison and optimization iteration of multiple schemes.

[0159] A knowledge graph quality scoring system has been built, which can support the systematic detection and structural optimization of knowledge redundancy, conflict and missing information in multi-source documents, and ensure the uniformity and reliability of knowledge graphs under multi-data source conditions.

[0160] A structured system for semantic knowledge of aircraft, such as Figure 2 As shown, it includes:

[0161] The paragraph splitting module is used to split aircraft design documents into paragraphs;

[0162] The aircraft design document includes mission description, overall scheme, system analysis and experimental data documents;

[0163] The design phase marker alignment module is used to predict the design phase corresponding to the split paragraphs and construct the design phase marker alignment structure.

[0164] The design phase includes task definition, overall scheme, aerodynamic design, structural design, system analysis, and experimental verification.

[0165] The semantic and design parameter extraction module is used to extract semantic and design parameters from the marker alignment structure in the design phase.

[0166] The knowledge graph construction module is used to build knowledge graphs based on extracted semantics and design parameters.

[0167] In the design phase marker alignment module, the set of split paragraphs is input into the semantic phase structure alignment model. The semantic phase structure alignment model adopts a RoBERTa variant structure to predict the design phase corresponding to the split paragraphs, and introduces common design phase vocabulary and key prompts in the aircraft development process for supervision.

[0168] In the design phase marker alignment module, the semantic phase structure alignment model performs BERT encoding on the split paragraphs, and then a classifier with Softmax added on top of RoBERTa predicts the corresponding design phase, thus constructing the design phase marker alignment structure.

[0169] In the design phase marker alignment module, a contextual semantic boundary detection mechanism is introduced into the semantic phase structure alignment model. By combining syntactic dependency analysis with the local consistency score of BERT embedding, the accuracy of identifying the boundaries of different design phases of the split paragraphs is improved.

[0170] In the semantic and design parameter extraction module, the design phase marker alignment structure is input into the semantic processing model to extract semantic and design parameters.

[0171] The semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, identifies the content that conforms to the "design premise-design entity-response solution" template, and extracts it into structured triples.

[0172] When the semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, it adopts a BERT encoding + BiLSTM + CRF structure and introduces a structure guidance token to improve the boundary recognition capability.

[0173] In the semantic and design parameter extraction module, the semantic processing model extracts design parameters from the design stage marker alignment structure, identifies the content expressed by the numerical values ​​of the design parameters, extracts the minimum value vmin and the maximum value vmax of the design parameters, and generates a confidence score p to perform design parameter interval modeling.

[0174] When the semantic processing model extracts design parameters from the marker-aligned structure in the design phase, for single-valued design parameter values, the minimum value vmin and maximum value vmax of the extracted design parameter are equal. For fuzzy-described design parameter values, the minimum value vmin and maximum value vmax of the design parameter are calculated based on the typical redundancy coefficient learned from the context.

[0175] When the semantic processing model extracts design parameters for the marker-aligned structure in the design phase, it performs BERT encoding on the text expressing the numerical content of the design parameters to obtain the embedding vector H, which is then input into two independent fully connected networks for predicting the minimum value vmin and the maximum value vmax, and into a Sigmoid network for predicting the confidence level p.

[0176] In the semantic and design parameter extraction module, a back-translation model or rewriting model is used to perform structure-preserving semantic enhancement on the split paragraphs, generating equivalent but different domain descriptions. For each split paragraph, 2-3 semantically equivalent but different variants are generated to obtain diverse but structurally consistent expressions and construct an enhanced corpus.

[0177] When enhancing the contextual diversity of the marker alignment structure in the design phase, a term freezing mechanism is introduced;

[0178] A pre-trained language model based on the Transformer architecture is selected, and an enhanced corpus is used to fine-tune the semantic processing model for multiple tasks in the field of aircraft design. The goal of the fine-tuning is to minimize the comprehensive loss function S.

[0179] S=LMLM+λ×Ldomain+μ×Lstage;

[0180] in,

[0181] LMLM is the loss for masked language modeling;

[0182] Ldomain represents the design entity recognition loss;

[0183] Lstage represents the label constraint loss during the design phase;

[0184] λ and μ are the task weight coefficients.

[0185] In the knowledge graph construction module, the extracted semantics and design parameters are input into the knowledge graph construction model to construct the knowledge graph.

[0186] A knowledge graph consists of an entity set E, relationships between entities R, and an evolution path graph P, stored in the form of a graph database or a nested key-value structure, supporting subsequent retrieval and reasoning.

[0187] In the knowledge graph construction module, a semantic embedding mechanism is used to convert all identified design elements into nodes of the knowledge graph;

[0188] Calculate the semantic consistency score between knowledge graph nodes:

[0189] Sim(i,j)=cos(i_vec, j_vec)×δ_structure×γ_context×ω_attribute;

[0190] Sim(i,j) is the semantic consistency score between knowledge graph nodes i and j;

[0191] i_vec and j_vec are the embedding vectors generated by BERT encoding of knowledge graph nodes i and j;

[0192] δ_structure is a Boolean term indicating whether the structure categories of knowledge graph nodes are consistent;

[0193] γ_context represents the similarity of knowledge graph nodes within their context.

[0194] ω_attribute represents the consistency rate of the attribute fields of the knowledge graph nodes;

[0195] If Sim(i,j) exceeds the consistency threshold, then the knowledge graph nodes i and j are fused.

[0196] The evolutionary path graph P is embedded into the knowledge graph in the form of a directed subgraph.

[0197] In the knowledge graph construction module, a knowledge graph quality score is calculated to evaluate the quality of the knowledge graph.

[0198] Q(G)=α×C(G)+β×P(G)+γ×Con(G);

[0199] in,

[0200] Q(G) represents the quality score of the knowledge graph;

[0201] C(G) represents the knowledge graph coverage;

[0202] P(G) represents the accuracy of the knowledge graph;

[0203] Con(G) represents the consistency of the knowledge graph;

[0204] α, β, and γ are weights.

[0205] The aircraft semantic knowledge structuring system disclosed in the above embodiments is described in a relatively simple manner since it corresponds to the aircraft semantic knowledge structuring method disclosed in the above embodiments. For specific details, please refer to the relevant descriptions in the section on aircraft semantic knowledge structuring method. Its technical effects can also be referred to the technical effects in the relevant section on aircraft semantic knowledge structuring method, and will not be repeated here.

[0206] Furthermore, those skilled in the art should recognize that the various modules of the aircraft semantic knowledge structuring system disclosed in the embodiments of this application can be implemented in electronic hardware, computer software, or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, they are generally described in terms of function in this application. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can choose different methods to implement the described functions for each specific application and its actual constraints, but such implementation should not be considered to be beyond the scope of this application.

[0207] The technical solution of this application has been described in conjunction with the preferred embodiments shown in the accompanying drawings. Those skilled in the art should understand that the protection scope of this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of this application.

Claims

1. A method for structuring semantic knowledge of aircraft, characterized in that, include: Step 1: Break the aircraft design document into paragraphs; The aircraft design document includes mission description, overall scheme, system analysis and experimental data documents; Step 2: Predict the design stages corresponding to the split paragraphs and construct a design stage marker alignment structure; The design phase includes task definition, overall scheme, aerodynamic design, structural design, system analysis, and experimental verification. Step 3: Extract semantics and design parameters from the design phase marker alignment structure; Step 4: Construct a knowledge graph based on the extracted semantics and design parameters; In step two, the set of split paragraphs is input into the semantic stage structure alignment model. The semantic stage structure alignment model adopts the RoBERTa variant structure to predict the design stage corresponding to the split paragraphs, and introduces common words and key prompts in the design stage of the aircraft development process for supervision. In step two, in the semantic stage structure alignment model, the split paragraphs are encoded with BERT, and then a classifier with Softmax added on top of RoBERTa predicts the corresponding design stage, thus constructing the design stage label alignment structure. In step two, a contextual semantic boundary detection mechanism is introduced into the semantic stage structure alignment model. Combined with syntactic dependency analysis and the local consistency score of BERT embedding, the accuracy of boundary identification of different design stages of the split paragraph is improved. In step three, the design phase marker alignment structure is input into the semantic processing model to extract semantics and design parameters; The semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, identifies the content that conforms to the "design premise-design entity-response solution" template, and extracts it into structured triples; When the semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, it adopts a BERT encoding + BiLSTM + CRF structure and introduces a structure guidance token to improve the boundary recognition capability.

2. The aircraft semantic knowledge structuring method according to claim 1, characterized in that, In step three, the semantic processing model extracts design parameters from the marker-aligned structure in the design phase, identifies the content expressed by the numerical values ​​of the design parameters, extracts the minimum value vmin and the maximum value vmax of the design parameters, and generates a confidence score p to perform range modeling of the design parameters. When the semantic processing model extracts design parameters from the marker-aligned structure in the design phase, for single-valued design parameter values, the minimum value vmin and maximum value vmax of the extracted design parameter are equal. For fuzzy-described design parameter values, the minimum value vmin and maximum value vmax of the design parameter are calculated based on the typical redundancy coefficient learned from the context. When the semantic processing model extracts design parameters for the marker-aligned structure in the design phase, it performs BERT encoding on the text expressing the numerical content of the design parameters to obtain the embedding vector H, which is then input into two independent fully connected networks for predicting the minimum value vmin and the maximum value vmax, and into a Sigmoid network for predicting the confidence level p.

3. The aircraft semantic knowledge structuring method according to claim 2, characterized in that, In step three, a back-translation model or a rewriting model is used to perform structure-preserving semantic enhancement on the split paragraphs, generating equivalent but different domain descriptions. For each split paragraph, 2-3 semantically equivalent but different sentence variations are generated to obtain diverse but structurally consistent expressions and construct an enhanced corpus. When enhancing the contextual diversity of the marker alignment structure in the design phase, a term freezing mechanism is introduced; A pre-trained language model based on the Transformer architecture is selected, and an enhanced corpus is used to fine-tune the semantic processing model for multiple tasks in the field of aircraft design. The goal of the fine-tuning is to minimize the comprehensive loss function S. S=LMLM+λ⋅Ldomain+μ⋅Lstage; in, LMLM is the loss for masked language modeling; Ldomain represents the design entity recognition loss; Lstage represents the label constraint loss during the design phase; λ and μ are the task weight coefficients.

4. The aircraft semantic knowledge structuring method according to claim 3, characterized in that, In step four, the extracted semantics and design parameters are input into the knowledge graph construction model to construct the knowledge graph. Knowledge graphs consist of entity sets E, relationships between entities R, and evolution path graphs P, stored in the form of graph databases or nested key-value structures, supporting retrieval and reasoning.

5. The aircraft semantic knowledge structuring method according to claim 4, characterized in that, In step four, a semantic embedding mechanism is used to convert all identified design elements into nodes of a knowledge graph; Calculate the semantic consistency score between knowledge graph nodes: Sim(i,j)=cos(i_vec, j_vec)×δ_structure×γ_context×ω_attribute; Sim(i,j) is the semantic consistency score between knowledge graph nodes i and j; i_vec and j_vec are the embedding vectors generated by BERT encoding of knowledge graph nodes i and j; δ_structure is a Boolean term indicating whether the structure categories of knowledge graph nodes are consistent; γ_context represents the similarity of knowledge graph nodes within their context. ω_attribute represents the consistency rate of the attribute fields of the knowledge graph nodes; If Sim(i,j) exceeds the consistency threshold, then the knowledge graph nodes i and j are fused. The evolutionary path graph P is embedded into the knowledge graph in the form of a directed subgraph.

6. The aircraft semantic knowledge structuring method according to claim 5, characterized in that, In step four, a quality score for the knowledge graph is calculated to evaluate its quality. Q(G)=α⋅C(G)+β⋅P(G)+γ⋅Con(G); in, Q(G) represents the quality score of the knowledge graph; C(G) represents the knowledge graph coverage; P(G) represents the accuracy of the knowledge graph; Con(G) represents the consistency of the knowledge graph; α, β, and γ are weights.

7. A semantic knowledge structuring system for aircraft, characterized in that, include: The paragraph splitting module is used to split aircraft design documents into paragraphs; The aircraft design document includes mission description, overall scheme, system analysis and experimental data documents; The design phase marker alignment module is used to predict the design phase corresponding to the split paragraphs and construct the design phase marker alignment structure. The design phase includes task definition, overall scheme, aerodynamic design, structural design, system analysis, and experimental verification. The semantic and design parameter extraction module is used to extract semantic and design parameters from the marker alignment structure in the design phase. The knowledge graph construction module is used to construct knowledge graphs based on extracted semantics and design parameters. In the design phase marker alignment module, the set of split paragraphs is input into the semantic phase structure alignment model. The semantic phase structure alignment model adopts the RoBERTa variant structure to predict the design phase corresponding to the split paragraphs, and introduces common words and key prompts in the design phase of the aircraft development process for supervision. In the design phase marker alignment module, in the semantic phase structure alignment model, the split paragraphs are encoded with BERT, and then a classifier with Softmax added on top of RoBERTa predicts the corresponding design phase, thus constructing the design phase marker alignment structure. In the design phase marker alignment module, a contextual semantic boundary detection mechanism is introduced into the semantic phase structure alignment model. Combined with syntactic dependency analysis and BERT embedding local consistency scoring, the accuracy of identifying the boundaries of different design phases of the split paragraphs is improved. In the semantic and design parameter extraction module, the design phase marker alignment structure is input into the semantic processing model to extract semantic and design parameters. The semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, identifies the content that conforms to the "design premise-design entity-response solution" template, and extracts it into structured triples; When the semantic processing model performs semantic extraction on the marker-aligned structure in the design phase, it adopts a BERT encoding + BiLSTM + CRF structure and introduces a structure guidance token to improve the boundary recognition capability.

8. The aircraft semantic knowledge structured system according to claim 7, characterized in that, In the semantic and design parameter extraction module, the semantic processing model extracts design parameters from the design stage marker alignment structure, identifies the content expressed by the numerical values ​​of the design parameters, extracts the minimum value vmin and the maximum value vmax of the design parameters, and generates a confidence score p to perform design parameter interval modeling. When the semantic processing model extracts design parameters from the marker-aligned structure in the design phase, for single-valued design parameter values, the minimum value vmin and maximum value vmax of the extracted design parameter are equal. For fuzzy-described design parameter values, the minimum value vmin and maximum value vmax of the design parameter are calculated based on the typical redundancy coefficient learned from the context. When the semantic processing model extracts design parameters for the marker-aligned structure in the design phase, it performs BERT encoding on the text expressing the numerical content of the design parameters to obtain the embedding vector H, which is then input into two independent fully connected networks for predicting the minimum value vmin and the maximum value vmax, and into a Sigmoid network for predicting the confidence level p.

9. The aircraft semantic knowledge structured system according to claim 8, characterized in that, In the semantic and design parameter extraction module, a back-translation model or rewriting model is used to perform structure-preserving semantic enhancement on the split paragraphs, generating equivalent but different domain descriptions. For each split paragraph, 2-3 semantically equivalent but different variants are generated to obtain diverse but structurally consistent expressions and construct an enhanced corpus. When enhancing the contextual diversity of the marker alignment structure in the design phase, a term freezing mechanism is introduced; A pre-trained language model based on the Transformer architecture is selected, and an enhanced corpus is used to fine-tune the semantic processing model for multiple tasks in the field of aircraft design. The goal of the fine-tuning is to minimize the comprehensive loss function S. S=LMLM+λ⋅Ldomain+μ⋅Lstage; in, LMLM is the loss for masked language modeling; Ldomain represents the design entity recognition loss; Lstage represents the label constraint loss during the design phase; λ and μ are the task weight coefficients.

10. The aircraft semantic knowledge structured system according to claim 9, characterized in that, In the knowledge graph construction module, the extracted semantics and design parameters are input into the knowledge graph construction model to construct the knowledge graph. Knowledge graphs consist of entity sets E, relationships between entities R, and evolution path graphs P, stored in the form of graph databases or nested key-value structures, supporting retrieval and reasoning.

11. The aircraft semantic knowledge structured system according to claim 10, characterized in that, In the knowledge graph construction module, a semantic embedding mechanism is used to convert all identified design elements into nodes of the knowledge graph; Calculate the semantic consistency score between knowledge graph nodes: Sim(i,j)=cos(i_vec, j_vec)×δ_structure×γ_context×ω_attribute; Sim(i,j) is the semantic consistency score between knowledge graph nodes i and j; i_vec and j_vec are the embedding vectors generated by BERT encoding of knowledge graph nodes i and j; δ_structure is a Boolean term indicating whether the structure categories of knowledge graph nodes are consistent; γ_context represents the similarity of knowledge graph nodes within their context. ω_attribute represents the consistency rate of the attribute fields of the knowledge graph nodes; If Sim(i,j) exceeds the consistency threshold, then the knowledge graph nodes i and j are fused. The evolutionary path graph P is embedded into the knowledge graph in the form of a directed subgraph.

12. The aircraft semantic knowledge structured system according to claim 11, characterized in that, In the knowledge graph construction module, a knowledge graph quality score is calculated to evaluate the quality of the knowledge graph. Q(G)=α⋅C(G)+β⋅P(G)+γ⋅Con(G); in, Q(G) represents the quality score of the knowledge graph; C(G) represents the knowledge graph coverage; P(G) represents the accuracy of the knowledge graph; Con(G) represents the consistency of the knowledge graph; α, β, and γ are weights.

Images