A SysML Model Automatic Generation System and Method Based on Hybrid AI and Domain Knowledge

Abstract

This invention discloses an automatic SysML model generation system based on hybrid AI and domain knowledge, comprising: a preprocessing module for text preprocessing and structure enhancement of PDF or Word versions of engineering documents; an NLP extraction module for identifying six core entities by fine-tuning BERT using aviation corpus, constructing a document-level relationship graph using GNN, modeling cross-paragraph dependencies, and generating "thought chains" by calling LLM for semantically ambiguous sentences to extract inference paths and resolve ambiguities; a rule transformation engine module for mapping entity relationship graphs to SysML in-memory object trees; and a controllable generation module for performing restricted decoding of LLM using the Guidance framework. An automatic SysML model generation method based on hybrid AI and domain knowledge is also disclosed. This invention solves the problems of low efficiency and poor semantic consistency in traditional MBSE implementations involving manual modeling.

Description

Technical Field

[0001] This invention belongs to the field of intelligent modeling technology in systems engineering, specifically a SysML model automatic generation system and method based on hybrid AI and domain knowledge. Background Technology

[0002] Currently, in complex systems engineering fields such as aircraft manufacturing, Model-Based Systems Engineering (MBSE) has become the mainstream industry practice. Traditional MBSE implementation relies on manually converting natural language-described requirements documents (such as requirements specifications and design reports) into SysML models. The specific process typically includes the following steps:

[0003] 1. Manual analysis and annotation: Engineers manually identify entities such as requirements, functions, and components and their relationships by reading documents, and draw SysML diagrams using modeling tools (such as MagicDraw and EnterpriseArchitect).

[0004] 2. Rule-Driven Transformation Tools: Some enterprises use semi-automated tools based on regular expressions or template matching (such as IBM DOORS's DXL scripts) to map text paragraphs with specific formats to SysML elements. For example, sentences containing "shall" are marked as requirements through keyword matching.

[0005] 3. Traditional NLP-assisted extraction: Early studies attempted to use classic natural language processing models such as Bi-LSTM and CRF for entity recognition, but the extraction accuracy was limited by small-scale labeled data.

[0006] The existing technology has the following key defects, which seriously restrict the implementation efficiency and model quality of MBSE:

[0007] 1. Manual modeling is inefficient:

[0008] Aeronautical engineering documents typically contain tens of thousands of requirements, and manually parsing each one is extremely time-consuming (according to statistics, manual modeling accounts for more than 60% of the total working hours of MBSE).

[0009] The mapping relationship between text and model elements relies on the engineer's experience and is prone to semantic bias (such as misclassifying "navigation system" as an interface instead of a functional module).

[0010] 2. Limitations of traditional NLP techniques:

[0011] Insufficient entity recognition accuracy: Traditional models (such as Bi-LSTM) have difficulty handling complex terms (such as “flight control computer redundant channel”) and synonyms (such as “cockpit crew” and “pilot”) unique to the aviation field, leading to misclassification of entity types.

[0012] Long-distance dependency missing: Requirements and verification conditions, functions and constraints are often scattered across different chapters, and existing models cannot effectively capture cross-paragraph relationships.

[0013] 3. The rigidity of rule-driven tools:

[0014] Conversion tools based on fixed rules (such as regular expression matching) can only handle structured text (such as a numbered list of requirements) and cannot adapt to the diverse expressions of free text (such as the implicit function switching logic of "when A fails, B should take over control").

[0015] High maintenance costs for rule bases: Aviation terminology and standards are frequently updated (such as upgrades to airworthiness provisions), requiring continuous manual adjustments to rule templates.

[0016] 4. Poor compliance of model generation:

[0017] Existing automated tools often generate SysML models with syntax errors (such as missing namespaces in XMI files) or logical contradictions (such as requirements not being covered by use cases), requiring manual correction item by item.

[0018] Lack of dynamic verification mechanism: The generated model cannot check SysML metamodel constraints (such as "Block must be connected via port") in real time, resulting in rework in the downstream design stage.

[0019] 5. Insufficient integration of domain knowledge:

[0020] Existing methods do not effectively integrate aviation domain ontology (such as the SAEARP4754 standard terminology library), resulting in model element mapping deviating from industry standards (such as incorrectly classifying "airworthiness requirements" as general functional requirements).

[0021] Weak cross-aircraft adaptability: For different aircraft types (such as civil airliners and military transport aircraft), the model needs to be retrained or the rule base needs to be reconstructed, resulting in high migration costs.

[0022] Existing technologies have significant shortcomings in text parsing accuracy, model generation efficiency, domain knowledge fusion, and dynamic verification capabilities, resulting in long MBSE implementation cycles and unstable model quality, which severely restricts the development progress of complex systems (such as aircraft avionics systems). Therefore, this paper proposes an automatic SysML model generation system and method based on hybrid AI and domain knowledge. Summary of the Invention

[0023] To address the aforementioned problems in existing technologies, this invention provides a SysML model automatic generation system and method based on hybrid AI and domain knowledge. By constructing a hybrid AI model enhanced with domain knowledge, it solves the problems of low efficiency and poor semantic consistency in traditional MBSE implementation through manual modeling.

[0024] The technical solution to achieve the above objectives is:

[0025] One of the present inventions is an automatic SysML model generation system based on hybrid AI and domain knowledge, comprising:

[0026] The preprocessing module is used to perform text preprocessing and structural enhancement on PDF or Word versions of project documents;

[0027] The NLP extraction module is used to fine-tune BERT by using aviation corpus to identify six core entities, use GNN to build document-level relationship graphs, model cross-paragraph dependency relationships, and call LLM to generate "thinking chains" for semantically ambiguous sentences, extract reasoning paths and resolve ambiguities;

[0028] The rule transformation engine module is used to map entity relationship graphs to SysML in-memory object trees;

[0029] The controllable generation module is used to perform restricted decoding of LLM using the Guidance framework, force the output of compliant XMI fragments, and enhance the generation accuracy by fusing OMG standards or ontology definitions through the RAG mechanism.

[0030] A multi-level verification module is used to verify XMI fragments and SysML, and to provide verification data.

[0031] The human-computer interaction module is used to receive feedback verification data from the multi-level verification module, manually correct errors, and automatically label the corrected data for fine-tuning the training and optimization of BERT, GNN and LLM models using aviation corpus.

[0032] Preferably, in the preprocessing module, text preprocessing and structure enhancement include:

[0033] Input processing: Extract paragraph, table, and heading structures from project documents using a multi-format parser;

[0034] Terminology standardization: Load the aerospace terminology ontology and use the AC automaton algorithm to identify compound terms and uniformly map them to SysML standard concepts;

[0035] Semantic segmentation: The RoBERTa model is used to segment the text into blocks and label the logical structure.

[0036] Preferably, in the NLP extraction module, the Adapter or LoRA lightweight parameter fine-tuning technology is introduced into the aviation corpus fine-tuning BERT to identify six types of core entities, including: requirements, code blocks, participants, functions, constraints, and interfaces.

[0037] Preferably, the rule conversion engine module has more than 200 built-in aviation domain mapping rules to map the entity relationship graph into a SysML memory object tree.

[0038] Preferably, in the multi-level verification module, the multi-level verification includes:

[0039] Syntax validation: Use XSD Schema to check the validity of the XMI format;

[0040] Semantic validation: Call the modeling tool API to validate SysML logical constraints.

[0041] The second invention relates to an automatic SysML model generation method based on hybrid AI and domain knowledge, comprising:

[0042] Step S1: Perform text preprocessing and structural enhancement on the PDF or Word version of the project document;

[0043] Step S2 involves fine-tuning BERT using aviation corpus to identify six core entities, constructing a document-level relationship graph using GNN, modeling cross-paragraph dependency relationships, and calling LLM to generate "thought chains" for semantically ambiguous sentences, extracting reasoning paths and resolving ambiguities.

[0044] Step S3: Map the entity relationship graph to a SysML in-memory object tree;

[0045] Step S4: Use the Guidance framework to perform restricted decoding on the LLM, force the output of compliant XMI fragments, and use the RAG mechanism to fuse OMG standards or ontology definitions to enhance the generation accuracy.

[0046] Step S5: Verify the XMI fragment and SysML, and return the verification data;

[0047] Step S6 involves manually correcting feedback to verify data errors, then automatically labeling the corrected data and using it to fine-tune the training and optimization of BERT, GNN, and LLM models using aviation corpus.

[0048] Preferably, in step S1, text preprocessing and structure enhancement include:

[0049] Input processing: Extract paragraph, table, and heading structures from project documents using a multi-format parser;

[0050] Terminology standardization: Load the aerospace terminology ontology and use the AC automaton algorithm to identify compound terms and uniformly map them to SysML standard concepts;

[0051] Semantic segmentation: The RoBERTa model is used to segment the text into blocks and label the logical structure.

[0052] Preferably, in step S2, the Adapter or LoRA lightweight parameter fine-tuning technology is introduced into the aviation corpus fine-tuning BERT to identify six types of core entities, including: requirements, code blocks, participants, functions, constraints, and interfaces.

[0053] Preferably, in step S3, more than 200 aviation domain mapping rules are built-in to map the entity relationship graph into a SysML memory object tree.

[0054] Preferably, in step S5, the verification includes:

[0055] Syntax validation: Use XSD Schema to check the validity of the XMI format;

[0056] Semantic validation: Call the modeling tool API to validate SysML logical constraints.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] This invention adopts a hybrid model collaborative architecture, combining BERT, GNN and LLM models for entity recognition and semantic understanding, reducing the overall recognition error rate to less than 8%, and significantly improving accuracy and stability;

[0059] This invention introduces graph neural networks (GNNs) to construct document-level entity relationship graphs, effectively modeling dependencies across paragraphs and pages, with a coverage rate of over 95%.

[0060] This invention employs a restricted generation mechanism and a multi-level syntax and semantic verification mechanism (including XSD verification and MagicDrawAPI verification) to ensure that the generated model achieves 100% compliance in terms of syntax structure and SysML consistency.

[0061] This invention introduces lightweight parameter fine-tuning techniques such as Adapter or LoRA, enabling rapid migration with only a minimal amount of labeled data, reducing overall migration costs by more than 70%.

[0062] This invention utilizes LLM to generate "thought chains" and infers implicit relationships through contextual understanding, achieving a success rate of over 85% in handling complex semantics. Attached Figure Description

[0063] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0064] Figure 1 This is a block diagram of the SysML model automatic generation system based on hybrid AI and domain knowledge of the present invention;

[0065] Figure 2 This is a flowchart of the SysML model automatic generation method based on hybrid AI and domain knowledge, which is the subject of this invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] like Figure 1 As shown, the SysML model automatic generation system based on hybrid AI and domain knowledge includes: preprocessing module 1, NLP extraction module 2, rule transformation engine module 3, controllable generation module 4, multi-level verification module 5, and human-computer interaction module 6.

[0068] Preprocessing module 1 is used to perform text preprocessing and structural enhancement on PDF or Word versions of project documents.

[0069] In this embodiment, text preprocessing and structure enhancement include:

[0070] Input processing: Extract paragraph, table, and heading structures from project documents using multi-format parsers (such as Apache PDFBox and Tabula);

[0071] Terminology standardization: Load the aerospace terminology ontology (such as the SAE ARP4754 terminology list), and use the AC automaton algorithm to identify compound terms (such as "flight control computer redundant channel") and uniformly map them to SysML standard concepts (such as "Redundant Flight Control System").

[0072] Semantic chunking: Using the RoBERTa model to segment text and annotate its logical structure (e.g.) <requirementblock>This provides contextual boundaries for subsequent entity extraction.

[0073] NLP extraction module 2 is used to fine-tune BERT by using aviation corpus to identify six types of core entities, and has cross-aircraft transfer capability; it uses GNN to construct document-level relationship graphs to model cross-paragraph dependency relationships (such as the traceability relationship between requirements and verification); and it calls LLM to generate "thinking chains" for semantically ambiguous sentences, extracts reasoning paths and resolves ambiguities, for example: "redundant systems should take over control" → inferred as the triggering relationship between RedundantChannel and UseCase.

[0074] In this embodiment, the lightweight parameter fine-tuning technology of Adapter or LoRA is introduced into the BERT fine-tuning of aviation corpus to identify six types of core entities, including: requirements, code blocks, actors, functions, constraints and interfaces; this solves the problem of rule base reconstruction caused by terminology differences and standard updates in traditional methods, and reduces migration costs by 70%.

[0075] In another embodiment, for scenarios with short document lengths or relatively regular context structures (such as embedded systems or subsystem requirements), GNN can be omitted, and local dependencies can be built solely based on the short text understanding capabilities of BERT and LLM, thereby reducing deployment costs and computational complexity.

[0076] In another embodiment, the original scheme uses a DeepSeek-like LLM (supporting chain-of-thought reasoning) for complex semantic parsing and XML generation. In actual deployment, it can be replaced with other large language models that support controlled generation, such as:

[0077] OpenAI GPT-4 (combining Function Calling API and XML output);

[0078] Claude 3 Opus (Anthropic, supports structured inference chain output);

[0079] Domestic deployment versions include Qwen-72B and Yi-34B (with LoRA for domain adaptation).

[0080] Rule transformation engine module 3 is used to map entity relationship graphs to SysML in-memory object trees.

[0081] In this embodiment, more than 200 aviation domain mapping rules are built-in (such as "A executes B → UseCase +Actor") to map the entity relationship graph into a SysML in-memory object tree.

[0082] Controllable generation module 4 is used to perform restricted decoding of LLM using the Guidance framework, forcing the output of compliant XMI fragments (such as each <requirement>It includes a unique xmi:id); and enhances generation accuracy by incorporating OMG standards or ontology definitions through the RAG mechanism.

[0083] In another embodiment, the rule conversion engine module 3 and the controllable generation module 4 are integrated into a "SysML model builder" to achieve resource reuse in certain edge deployment scenarios (such as airborne development terminals); the "feedback module" can also be deployed independently as a data quality monitoring service to achieve closed-loop tracking of model quality.

[0084] The multi-level verification module 5 is used to verify the XMI fragment and SysML and to provide verification data.

[0085] In this embodiment, multi-level verification includes:

[0086] Syntax validation: Use XSD Schema to check the validity of the XMI format;

[0087] Semantic validation: Call the modeling tool API (such as MagicDraw) to validate SysML logical constraints.

[0088] In this embodiment, the generated SysML model is forced to meet the requirements of syntax compliance (100% pass rate of XMI Schema verification) and semantic consistency (≥95% bidirectional traceability coverage of requirements-functions-components) through a dynamic rule engine and a multi-level verification mechanism, thereby eliminating the risk of logical contradictions in manual modeling.

[0089] The human-computer interaction module 6 is used to receive feedback verification data from the multi-level verification module 5, manually correct errors, and automatically label the corrected data for fine-tuning the training and optimization of BERT, GNN and LLM models using aviation corpus.

[0090] like Figure 2 As shown, the SysML model automatic generation method based on hybrid AI and domain knowledge includes:

[0091] Step S1: Perform text preprocessing and structural enhancement on the PDF or Word version of the project document.

[0092] In this embodiment, text preprocessing and structure enhancement include:

[0093] Input processing: Extract paragraph, table, and heading structures from project documents using a multi-format parser;

[0094] Terminology standardization: Load the aerospace terminology ontology and use the AC automaton algorithm to identify compound terms and uniformly map them to SysML standard concepts;

[0095] Semantic segmentation: The RoBERTa model is used to segment the text into blocks and label the logical structure.

[0096] Step S2 involves fine-tuning BERT to identify six core entities using aviation corpus, constructing a document-level relationship graph using GNN, modeling cross-paragraph dependencies, and calling LLM to generate "thought chains" for semantically ambiguous sentences, extracting reasoning paths and resolving ambiguities.

[0097] In this embodiment, the lightweight parameter fine-tuning technology of Adapter or LoRA is introduced into the fine-tuning of BERT based on aviation corpus to identify six types of core entities, including: requirements, code blocks, actors, functions, constraints, and interfaces.

[0098] In this embodiment, BERT is responsible for entity accuracy, GNN models contextual relationships, and LLM handles semantic complexity. The three work together to achieve end-to-end modeling capabilities. For example, in the sentence "The flight control system needs to respond to commands within 2 seconds," BERT identifies "flight control system" as a Block, GNN extracts the "response time" constraint context, and LLM determines it to be a Requirement attribute.

[0099] Regarding entity recognition accuracy, traditional methods rely on engineers' experience, with recognition error rates typically between 15% and 20%. Rule-based tools, on the other hand, can mostly only handle structured fields, resulting in even higher errors, reaching around 25%. This invention employs a hybrid model collaborative architecture, combining BERT, GNN, and LLM models for entity recognition and semantic understanding, reducing the overall recognition error rate to below 8%, significantly improving accuracy and stability.

[0100] Regarding cross-domain adaptation costs, traditional methods require retraining engineers each time they are applied to a new device or system; rule-based tools often require a complete reconstruction of the rule base, with adaptation cycles lasting 3 to 6 months. This invention introduces lightweight parameter fine-tuning techniques such as Adapter or LoRA, enabling rapid migration with only a minimal amount of labeled data, reducing overall migration costs by more than 70%.

[0101] In handling long-distance dependencies, traditional manual methods typically trace requirements by manually comparing them, which is inefficient and prone to omissions. Rule-based tools usually lack the ability to model contexts at the paragraph level or above, making it impossible to handle dependencies between chapters. This invention introduces a graph neural network (GNN) to construct a document-level entity relationship graph, effectively modeling dependencies across paragraphs and pages, with a coverage rate of over 95%.

[0102] In terms of complex semantic parsing capabilities, traditional manual or rule-based methods often cannot handle implicit logic, ambiguous referential meanings, and other phenomena, resulting in a failure rate as high as 40%. However, this invention utilizes LLM to generate "thought chains" and infers implicit relationships through contextual understanding, achieving a success rate of over 85% in handling complex semantics.

[0103] Step S3: Map the entity relationship graph to a SysML in-memory object tree.

[0104] In this embodiment, more than 200 aviation domain mapping rules are built-in to map the entity relationship graph into a SysML in-memory object tree.

[0105] Step S4: Use the Guidance framework to perform restricted decoding on the LLM, force the output of compliant XMI fragments, and use the RAG mechanism to fuse OMG standards or ontology definitions to enhance generation accuracy.

[0106] In another embodiment, an initial XMI draft is first generated directly using LLM, and then NER / GNN is used to backtrack and identify the core entities in the corresponding paragraphs and bind them to the original text, which is suitable for high-throughput rapid review tasks.

[0107] Step S5: Verify the XMI fragment and SysML, and return the verification data.

[0108] In this embodiment, verification includes:

[0109] Syntax validation: Use XSD Schema to check the validity of the XMI format;

[0110] Semantic validation: Call the modeling tool API to validate SysML logical constraints.

[0111] Regarding compliance in model generation, traditional methods rely on manual format and semantic verification, resulting in a rework rate as high as 30%. Rule-driven tools have limited syntax verification capabilities, with a compliance pass rate of approximately 70%. In contrast, this invention employs a restricted generation mechanism and a multi-level syntax and semantic verification mechanism (including XSD verification and MagicDraw API verification) to ensure that the generated model achieves 100% compliance in terms of both syntax structure and SysML metamodel consistency.

[0112] Step S6 involves manually correcting feedback to verify data errors, then automatically labeling the corrected data and using it to fine-tune the training and optimization of BERT, GNN, and LLM models using aviation corpus.

[0113] In another embodiment, "manual correction" is replaced by "human-machine co-annotation": during the cold start phase with a small amount of data, engineers can work with the model to complete the initial annotation and then drive subsequent training; this process can be efficiently interacted through drag-and-drop components in WebGME, reducing the barrier to first-time use.

[0114] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.< / requirement> < / requirementblock>

Claims

1. A SysML model automatic generation system based on hybrid AI and domain knowledge, characterized in that, include: The preprocessing module is used to perform text preprocessing and structural enhancement on PDF or Word versions of project documents; The NLP extraction module is used to fine-tune BERT by using aviation corpus to identify six core entities, use GNN to build document-level relationship graphs, model cross-paragraph dependency relationships, and call LLM to generate "thought chains" for semantically ambiguous sentences, extract reasoning paths and resolve ambiguities; The rule transformation engine module is used to map entity relationship graphs to SysML in-memory object trees; The controllable generation module is used to perform restricted decoding of LLM using the Guidance framework, force the output of compliant XMI fragments, and enhance the generation accuracy by fusing OMG standards or ontology definitions through the RAG mechanism. A multi-level verification module is used to verify XMI fragments and SysML, and to provide verification data. The human-computer interaction module is used to receive feedback verification data from the multi-level verification module, manually correct errors, and automatically label the corrected data for fine-tuning the training and optimization of BERT, GNN and LLM models using aviation corpus.

2. The SysML model automatic generation system based on hybrid AI and domain knowledge according to claim 1, characterized in that, The preprocessing module includes text preprocessing and structure enhancement, which includes: Input processing: Extract paragraph, table, and heading structures from project documents using a multi-format parser; Terminology standardization: Load the aerospace terminology ontology and use the AC automaton algorithm to identify compound terms and uniformly map them to SysML standard concepts; Semantic segmentation: The RoBERTa model is used to segment the text into blocks and label the logical structure.

3. The SysML model automatic generation system based on hybrid AI and domain knowledge according to claim 1, characterized in that, In the NLP extraction module, the Adapter or LoRA lightweight parameter fine-tuning technology is introduced into the aviation corpus to fine-tune BERT, and six types of core entities are identified. The six types of core entities include: requirements, code blocks, participants, functions, constraints, and interfaces.

4. The SysML model automatic generation system based on hybrid AI and domain knowledge according to claim 1, characterized in that, The rule conversion engine module has more than 200 built-in aviation domain mapping rules, which map entity relationship graphs into SysML memory object trees.

5. The SysML model automatic generation system based on hybrid AI and domain knowledge according to claim 1, characterized in that, The multi-level verification module includes the following multi-level verification: Syntax validation: Use XSD Schema to check the validity of the XMI format; Semantic validation: Call the modeling tool API to validate SysML logical constraints.

6. A method for automatically generating SysML models based on hybrid AI and domain knowledge, characterized in that, include: Step S1: Perform text preprocessing and structural enhancement on the PDF or Word version of the project document; Step S2 involves fine-tuning BERT using aviation corpus to identify six core entities, constructing a document-level relationship graph using GNN, modeling cross-paragraph dependency relationships, and calling LLM to generate "thought chains" for semantically ambiguous sentences to extract reasoning paths and resolve ambiguities. Step S3: Map the entity relationship graph to a SysML in-memory object tree; Step S4: Use the Guidance framework to perform restricted decoding on the LLM, force the output of compliant XMI fragments, and use the RAG mechanism to fuse OMG standards or ontology definitions to enhance the generation accuracy. Step S5: Verify the XMI fragment and SysML, and return the verification data; Step S6: Verify data errors using manual correction feedback, correct the data, automatically label it, and use it to fine-tune the training and optimization of BERT, GNN, and LLM models using aviation corpus.

7. The method for automatically generating SysML models based on hybrid AI and domain knowledge according to claim 6, characterized in that, In step S1, text preprocessing and structure enhancement include: Input processing: Extract paragraph, table, and heading structures from project documents using a multi-format parser; Terminology standardization: Load the aerospace terminology ontology and use the AC automaton algorithm to identify compound terms and uniformly map them to SysML standard concepts; Semantic segmentation: The RoBERTa model is used to segment the text into blocks and label the logical structure.

8. The method for automatically generating SysML models based on hybrid AI and domain knowledge according to claim 6, characterized in that, In step S2, the Adapter or LoRA lightweight parameter fine-tuning technology is introduced into the aviation corpus fine-tuning BERT to identify six types of core entities, including: requirements, code blocks, participants, functions, constraints, and interfaces.

9. The method for automatically generating SysML models based on hybrid AI and domain knowledge according to claim 6, characterized in that, In step S3, more than 200 aviation domain mapping rules are built-in to map the entity relationship graph into a SysML memory object tree.

10. The method for automatically generating SysML models based on hybrid AI and domain knowledge according to claim 6, characterized in that, In step S5, the verification includes: Syntax validation: Use XSD Schema to check the validity of the XMI format; Semantic validation: Call the modeling tool API to validate SysML logical constraints.

Images