Sysml model generation method based on large language model in conceptual design

Abstract

The application provides a concept design SysML model generation method based on a large language model, and the method comprises the following steps: guiding an LLM by using a thinking chain prompt word template, converting user input into a concept design scheme described in pure text form; guiding the LLM by using a prompt word template 1 to split the concept design scheme into a sentence list containing complete semantics; guiding the LLM by using a prompt word template 2 to extract entities and their relationships from each sentence and output triples; creating a SysML model hierarchical structure according to a predefined rule based on the triples; and creating an XMI file in accordance with an XMI format specification to obtain a complete SysML model. The method significantly improves the automation and generalization of SysML model generation, can effectively process less text input, and generates a concept design scheme model that can be directly used for actual system design and adapts to products, and is particularly suitable for highly complex large system design.

Description

Technical Field

[0001] The present invention relates to the technical field of model-based system engineering, and in particular to a SysML model generation method based on a large language model in conceptual design. Background Art

[0002] Conceptual design consists of a series of orderly, organized, and goal-oriented design activities from analyzing user needs to generating conceptual products. It manifests as a continuous evolutionary process from coarse to fine, from vague to clear, and from abstract to concrete. As a critical stage in the design process, the quality of conceptual design directly impacts the innovation and feasibility of the final solution. However, current conceptual design faces numerous challenges. On the one hand, in traditional conceptual design, designers are limited by their personal experience and knowledge when exploring solutions, making it difficult to efficiently process massive amounts of fragmented data and cross-domain product concept designs. On the other hand, document-based systems engineering design, due to the ambiguity and complexity of natural language descriptions, makes it difficult to communicate design solutions among stakeholders, and it also makes it difficult to maintain consistency in documentation during iterative design modifications.

[0003] With the in-depth development of modern systems engineering practice, system models are becoming increasingly complex. For example, highly complex large-scale systems such as aircraft systems, automotive systems, and aerospace systems not only contain a large number of components and subsystems, but also the interaction logic between components is intricate, resulting in increasingly diverse product variability. How to efficiently design complex systems and generate solutions that adapt to conceptual designs is an urgent problem that needs to be solved.

[0004] Model-based systems engineering improves design efficiency by replacing documentation with modeling languages ​​like SysML. However, the modeling process still relies heavily on manual labor, resulting in significant time wasted on repetitive tasks. Therefore, automating the conceptual design and modeling processes, specifically automatically generating SysML models, can significantly save significant human resources and time.

[0005] Currently, there are two main approaches to automatically generating SysML models: one is based on traditional natural language processing (NLP) technology, which uses rule templates to automatically generate SysML models. This method has a low degree of automation and high input requirements. The other is based on traditional deep learning technology, which uses a model plus fine-tuning approach to automatically generate SysML models. This method has a high degree of automation but relies on professional domain datasets. Specifically, current methods for automatically generating SysML models have the following shortcomings:

[0006] 1. Additional professional domain dataset is needed. Traditional methods need to prepare specific professional domain dataset to fine-tune the model, and different datasets need to be prepared for different fields to have good performance on specific field tasks. Or a domain knowledge graph needs to be constructed according to the professional dataset, and a SysML model is generated based on the knowledge graph. The process of preparing the dataset and constructing the knowledge graph is very time-consuming and requires a lot of human resources. Without additional professional domain dataset support, the method has poor generalization.

[0007] 2. Automatically generated modeling results are difficult to apply in practice. Traditional methods use drawing software to visualize the modeling results, and the generated results cannot be directly applied to actual modeling software, making it inconvenient for designers to modify the preliminary modeling scheme.

[0008] 3. The input requirements are relatively strict. The input of traditional methods is not natural language text, but structured text with a certain format; or the input text must meet certain rules, templates and specifications; or the input needs to include existing models as input assistance. It is difficult to handle cases where only natural language expressions are used as input, and even more difficult to handle cases where only product names and other input text are used.

[0009] 4. The automation level is not high enough. Traditional NLP technology methods use custom heuristic rules or rule algorithms, and add traditional technical tools to realize SysML model automatic generation. In the main step of extracting SysML model elements from concept design scheme text, the limited custom rules are difficult to apply to an infinite number of natural language texts, the cost of maintaining the rules is high, and the generalization of the rules is poor; custom rules rely on manual operation and require more professional knowledge, and the automation level is low. Traditional deep learning methods also require human intervention in processes such as dataset construction and model fine-tuning, and the automation level needs to be further improved.

[0010] Therefore, there is an urgent need for a SysML model generation method and system that has a higher automation level, a stronger generalization, can handle less text input, and the modeling results can be applied in practice. With the development of deep learning technology, especially the breakthrough progress made by large language models in natural language processing, new ideas and technical means are provided to solve the above technical problems. SUMMARY

[0011] The purpose of the present application is to provide a SysML model generation method based on large language models in concept design, which realizes a SysML model generation method with a higher automation level, a stronger generalization, can handle less text input, and the modeling results can be applied in practice, by utilizing the powerful knowledge reserve, semantic understanding and text generation ability of large language models.

[0012] The embodiment provides a method for generating a SysML model based on a large language model in a conceptual design, comprising the following steps:

[0013] Input user requirements related to conceptual design into the large language model, and use the preset thought chain prompt word template to guide the large language model to output the conceptual design plan described in plain text form;

[0014] Using the preset prompt word template to guide the large language model, the conceptual design scheme described in plain text is divided into sentences and a list of sentences with complete semantics is output;

[0015] Using the preset prompt word template 2 to guide the large language model, entity relationship extraction is performed on each sentence in the sentence list containing complete semantics, and a triple containing entities and their relationships is output;

[0016] Create SysML model hierarchies based on triples containing entities and their relationships according to predefined rules;

[0017] According to the SysML model hierarchical structure, an XMI file is created according to the XMI format specification for representing SysML model information to obtain a complete SysML model.

[0018] In one embodiment, the user demand related to the conceptual design is a product name in a highly complex large-scale system, or a product name and a simple demand description of the product.

[0019] Furthermore, the highly complex large-scale system includes one or more of an aircraft system, an automobile system, and an aerospace system.

[0020] In one embodiment, the parameter scale of the large language model is greater than 100 B; the large language model is DeepSeek-V3.

[0021] In one embodiment, the preset thought chain prompt word template is a small sample thought chain prompt word template, including a thought chain sample prompt template and a practical problem prompt template;

[0022] The thought chain sample prompt template includes 2-10 thought chain samples. The prompt template of each thought chain sample includes: a question prompt template to guide the large language model to generate a product concept design plan, a fixed prompt to guide the large language model to perform thought chain reasoning, a sample concept design plan thought chain reasoning steps to guide the large language model to perform thought chain reasoning, and a fixed prompt to guide the large language model to summarize the concept design plan;

[0023] The actual question prompt template includes: a question prompt template for guiding the large language model to generate a product concept design plan and a fixed prompt for guiding the large language model to perform thought chain reasoning.

[0024] In one embodiment, the sample concept design scheme thinking chain reasoning steps for guiding the large language model to perform thinking chain reasoning include four mapping reasoning steps, specifically including: product-demand mapping, demand-function mapping, function-behavior mapping, and behavior-structure mapping. Each mapping reasoning step sets a corresponding mapping prompt word template.

[0025] In one embodiment, the conceptual design solution described in plain text includes a thought chain reasoning process and structural information of a SysML model, wherein the structural information of the SysML model includes module composite relationships and component attribute connector relationships.

[0026] In one embodiment, the preset prompt word template 1 includes: a large language model role definition, a sentence segmentation task goal to be achieved, a sentence segmentation task solution example and specific instructions.

[0027] In one embodiment, the preset prompt word template 2 includes: a large language model role definition, an entity relationship extraction task goal to be achieved, an example of solving the entity relationship extraction task, and specific instructions.

[0028] In one embodiment, the triples of entities and their relationships include triples of modules and their composite relationships and triples of component attributes and their connector relationships.

[0029] In one embodiment, the predefined rules include: triple connection rules, triple deduplication rules, and SysML model hierarchical structure enhancement rules;

[0030] The triple connection rule is to use the product name as the top node of the SysML model hierarchy, connect the triples according to the entity name, and add them one by one to the SysML model hierarchy; if the triple is a module and its composite relationship triple, then the tail entity is added as a child node to the next level of the head entity node of the SysML model hierarchy; if the head entity node does not exist in the SysML model hierarchy, then a new head entity node is created; if the triple is a component attribute and its connector relationship triple, then the tail entity in the triple is added to the next level of the head entity node of the SysML model hierarchy, and a connector label is added to identify the connector relationship; if the head entity node does not exist in the SysML model hierarchy, then a new head entity node is created;

[0031] The triple deduplication rules include: for module and compound relationship triples, when the head entity, relationship and tail entity of the triples are the same, it is judged as a duplicate; for component attribute and connector relationship triples, when the relationship of the triples is the same and the name set of the head entity and tail entity are the same, it is judged as a duplicate;

[0032] The rules for enhancing the SysML model hierarchical structure include: after all triples are connected, connecting the top nodes of multiple hierarchical structures to the only top node of the SysML model hierarchical structure; for all component attributes and their connector relationship triples, creating a module corresponding to the component attribute; if the module does not appear in the SysML model hierarchical structure, adding it to the SysML model hierarchical structure and connecting it to the top node; otherwise, ignoring the module.

[0033] In one embodiment, creating an XMI file includes: mapping SysML model elements to XMI tags, adding SysML model graphical information, and creating an algorithm for XMI tags based on the SysML model hierarchical structure;

[0034] The mapping of SysML model elements to XMI tags includes: mapping modules, composite relationships, component attributes, and connector relationships in the SysML model to corresponding tags in XMI;

[0035] The added SysML model graphic information includes: the name of the frame, the module to which it belongs, the display position, the frame size, the display position, explicit size and display style of the module and component attributes, and the display position, display length and display style of the connection lines of the composite relationship and connector relationship;

[0036] The algorithm for creating XMI tags based on the SysML model hierarchical structure includes: creating a module definition diagram, all modules in the module definition diagram, and the composite relationships between modules based on the SysML model hierarchical structure, then creating component attributes under the module based on the module composite relationship information, then creating an internal module diagram and connector relationships based on the component attributes and their connector relationships, and finally creating a corresponding extended graphic node under the XMI root node based on each existing module definition diagram and internal module diagram.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] (1) Reduce reliance on specialized domain datasets. This invention utilizes the rich knowledge base of the large language model to generate conceptual design solutions and automatically generates SysML models based on this knowledge base. The entire process does not require manual construction of additional specialized domain datasets. The large language model can be used to generate conceptual design solutions for product names in different fields.

[0039] (2) The automatically generated SysML model can be imported into modeling software, enhancing its practicality. The present invention uses a universal XMI format specification file to represent the SysML model, which can be imported into commonly used practical modeling software that supports the XMI format, making it easier for designers to perform subsequent modeling and modification, and making it more practical.

[0040] (3) The input requirements are few and can handle single noun input. The input of the present invention is a product name described in natural language or a product name and a simple description of its requirements. It can generate a relatively rich and complete SysML model from a single noun. There is no restriction on the input. Even if there is only one product name, a SysML model of the corresponding conceptual design scheme of the product name can be automatically generated.

[0041] (4) Significantly improve the degree of automation. The main steps of the entire process of the present invention use a large language model plus a prompt word template method, followed by rule-based methods and algorithm-assisted processing. This can be embedded in practical modeling software and can achieve a one-step generation of a SysML model of the corresponding conceptual design solution from the product name. The entire process does not require human participation, and there is no need for additional steps such as manually constructing a data set and knowledge graph, thus further improving the degree of automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 It is a flowchart of a method for generating a SysML model based on a large language model in a conceptual design provided in an embodiment.

[0044] Figure 2 It is a structural diagram of a preset thought chain prompt word template provided in an embodiment.

[0045] Figure 3 2 is a schematic structural diagram of a preset prompt word template 1 provided in an embodiment.

[0046] Figure 4 It is a structural diagram of the preset prompt word template 2 provided in the embodiment.

[0047] Figure 5 It is a schematic diagram of the structure of the predefined rules provided in the embodiment.

[0048] Figure 6 It is a structural diagram of the XMI format specification provided by the embodiment.

[0049] Figure 7 A module definition diagram of a SysML model provided in an embodiment.

[0050] Figure 8 This is an internal module diagram of the SysML model provided in the embodiment. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following is a

[0052] The present invention is further described in detail in the following examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0053] The solution of the embodiment of the present invention is as follows Figure 1 As shown in Figure 2, a SysML model generation method based on a large language model in conceptual design is Figure 1 As shown, the following steps are included:

[0054] S1. Input user requirements related to conceptual design into the large language model, and use the preset thought chain prompt word template to guide the large language model to output the conceptual design plan described in plain text.

[0055] In the embodiment, the user demand input for the conceptual design is a simple product name noun, such as sonar, drone, and other highly complex large-scale systems, such as aircraft systems, automobile systems, or aerospace systems; or a product name noun and its simple demand description, such as sonar needs to be able to detect the position and distance of underwater objects, that is, no detailed demand description is given in the input.

[0056] Traditional methods rely on detailed requirements text for subsequent work, and requirements analysis is the designer's job. The present invention uses a single product name as input, further freeing up the designer's hands and improving automation. If the input contains a very detailed requirements description, this step can be omitted and the process can begin directly at step S2.

[0057] In order to optimize the generation effect of the present invention, the parameter scale of the large language model used in this step and the subsequent S2 and S3 steps should be greater than 100 B, and the DeepSeek-V3 large language model can be used in the embodiment. The DeepSeek-V3 large language model has a total of 671 B parameters and has efficient reasoning capabilities. Its performance is better than other open source models and achieves performance comparable to that of leading closed source models. Large language models with a parameter scale of more than 100 B can better use prompt words to induce their logical reasoning ability, especially the thinking chain method used in this step is more suitable for language models with a large number of parameters. Using a model with a parameter scale greater than 100 B, the model's knowledge reserve, semantic understanding ability, and text generation ability are all at a high level, which can effectively cope with the input of a single product name and the concept design of a cross-domain product, and can ensure the quality of the generated concept design scheme, solving the problem of being limited by the designer's experience and knowledge reserve when exploring solutions, and the designer's difficulty in efficiently processing massive fragmented data and cross-domain product concept design.

[0058] like Figure 2 As shown, in this step, the preset thinking chain prompt word template is a small sample thinking chain prompt word template, which includes two parts: a thinking chain sample prompt template and an actual problem prompt template.

[0059] The sample prompt template for the thinking chain contains 2-10 samples. The prompt template for each sample includes: a question prompt template to guide the large language model to generate a product concept design plan, a fixed prompt to guide the large language model to perform thinking chain reasoning, a sample concept design plan thinking chain reasoning steps to guide the large language model to perform thinking chain reasoning, and a fixed prompt to guide the large language model to summarize the concept design plan.

[0060] The actual question prompt templates include: question prompt templates that guide the large language model to generate product concept design solutions, and fixed prompts that guide the large language model to perform thought chain reasoning.

[0061] The sample concept design scheme thinking chain reasoning steps include four mapping reasoning steps, including: product-demand mapping, demand-function mapping, function-behavior mapping, and behavior-structure mapping. Each step is set with a corresponding mapping prompt template.

[0062] Specifically, the preset thinking chain prompt word template in this step is as follows:

[0063] Question: Generate a conceptual design for XXX. Answer: Let's think about this step by step: Step 1: Clarify the requirements for this product: ... Step 2: To meet the requirements in Step 1, the following functions need to be implemented: ... Step 3: Describe the various functions in Step 2 in more detail: ... Step 4: To implement the various functions described in Steps 2 and 3, the product requires the following structure: ... Therefore, the answer is: ... "

[0064] In this prompt word template, "Question: Generate a conceptual design plan for ×××" is the question prompt template for guiding the large language model to generate a product conceptual design plan, "Answer: Let us think step by step:" is a fixed prompt for guiding the large language model to perform thought chain reasoning, "Step 1:..., Step 4:..." are sample conceptual design plan thought chain reasoning steps for guiding the large language model to perform thought chain reasoning, and "Therefore, the answer is:" is a fixed prompt for guiding the large language model to summarize the conceptual design plan.

[0065] To simplify the conceptual design solution, which only includes modules, composite relationships, component attributes, and connector relationships in the SysML model, the following sentence can be added after the question prompt template: "The final answer only needs to tell me which components are needed and the connection relationships between these components." This will guide the large language model to generate more expected results. In the thought chain reasoning steps, "Step 1:..." is the product-requirement mapping, "Step 2:..." is the requirement-function mapping, "Step 3:..." is the function-behavior mapping, and "Step 4:..." is the behavior-structure mapping.

[0066] This prompt template is a sample concept design proposal template. The specific sample concept design proposal requires manual design. Use this template to create 2-10 samples as needed, creating a small-sample thought chain prompt. Using this small-sample thought chain prompt minimizes manual effort while ensuring design quality. When using a large language model to generate new product concept designs, simply use the prompt "Question: Generate a concept design for XXX. Answer: Let's think about it step by step." to guide the large language model to generate an answer similar to the sample, including the thought chain reasoning steps and a concept design solution summary.

[0067] The conceptual design scheme obtained in this step is described in plain text. The content of the scheme includes the thought chain reasoning process, and the conceptual design scheme obtained by summarizing the modules, composite relationships, component attributes and connector relationships in the SysML model. The complete thought chain reasoning process can improve the interpretability of the generated conceptual design scheme, facilitate designers to adjust the generated results by adjusting the reasoning process, and bring certain design inspiration to designers. The conceptual design scheme obtained by summarizing contains the basic structural information of the SysML model, that is, the composite relationship between modules in the SysML model, and the connection relationship between component attributes. The part of the result of this step used for the subsequent steps of the present invention is mainly the part of the conceptual design scheme obtained by summarizing.

[0068] S2. Use the preset prompt word template to guide the large language model to divide the concept design scheme described in plain text into sentences and output a list of sentences containing complete semantics.

[0069] In the embodiment, Figure 3 As shown, the preset prompt word template 1 of this step includes: large language model role definition, sentence segmentation task objectives to be achieved, and sentence segmentation task solution examples and specific instructions.

[0070] Specifically, a specific example of the prompt word template is as follows: "You are an experienced SysML modeler. Now you need to divide the input text into relatively complete sentences and output a list of sentences with complete semantics. Each sentence should contain relatively complete SysML modeling information as much as possible to facilitate subsequent SysML modeling based on each sentence. For example: if the sentence "A computer includes a display, a keyboard, and a host" is divided, the modules "computer," "display," "keyboard," and "host" and the composite relationship between them can be created based on this sentence; if the sentence "The host is connected to the display and keyboard" is divided, the component properties "host" and "display," "host" and "keyboard" and the connector relationship between them can be created based on this sentence. The divided sentences may contain more than one sentence." The specific examples, specific instructions, and other special requirements in the prompt word template can be freely adjusted as needed.

[0071] Using a large language model for sentence segmentation fully leverages its knowledge and semantic understanding capabilities, achieving results similar to manual segmentation. Simply input the detailed segmentation requirements into the large language model to guide the generation of the desired results. Compared to traditional sentence segmentation methods, this approach offers better results, greater generalization, and increased efficiency.

[0072] S3. Use the preset prompt word template 2 to guide the large language model, extract entity relationships for each sentence in the sentence list containing complete semantics, and output triples containing entities and their relationships.

[0073] In the embodiment, Figure 4 As shown, the preset prompt word template 2 in this step includes: a large language model role definition, the entity relationship extraction task objectives to be achieved, an example solution for the entity relationship extraction task, and detailed instructions. In this invention, triples of entities and their relationships include two types of triples: module and composite relationship triples, and component attribute and connector relationship triples. Prompt words are primarily used to extract these two types of triples contained in sentences.

[0074] Specifically, an example of the prompt template is as follows: "You are an experienced SysML modeler. Now you need to extract entity relationships for each sentence in the complete semantic sentence list and output a list of triples containing entities and their relationships. The entity relationship triples to be extracted include modules and their composite relationship triples in the SysML model, as well as component attributes and their connector relationship triples. For example, for the sentence "A computer contains a display, a keyboard, and a host," you need to extract three module and composite relationship triples: "module: computer, relationship: composite relationship, module: display," "module: computer, relationship: composite relationship, module: keyboard," and "module: computer, relationship: composite relationship, module: host." For the sentence "The host is connected to the display and keyboard," you need to extract two component attribute and connector relationship triples: "component attribute: host, relationship: connector relationship, component attribute: display," and "component attribute: host, relationship: connector relationship, component attribute: keyboard." Please output according to the format of the given example. Each entity relationship triple must indicate the SysML model element corresponding to the entity and relationship respectively." The specific examples, instructions and other special requirements in the prompt word template can be freely adjusted as needed.

[0075] Using a large language model for entity relationship extraction can fully utilize the semantic understanding and text generation capabilities of the large language model. By using specific examples and detailed explanations, the large language model can be guided to fully understand the requirements of SysML modeling and generate expected results according to the given example format.

[0076] S4. Create a SysML model hierarchy according to predefined rules based on triples containing entities and their relationships.

[0077] In the embodiment, Figure 5As shown, the predefined rules in this step include: triple connection rules, triple deduplication rules, and rules for enhancing the SysML model hierarchy. The goal of these predefined rules is to reassemble all scattered triples into a hierarchical structure that contains the complete SysML model structure. The resulting SysML hierarchy primarily includes information about all modules in the SysML model, the hierarchical relationships between modules, and the connections between modules. This information also includes the desired model structure information for the Block Definition Diagram (BDD) and Internal Block Diagram (IBD) in the SysML model.

[0078] The triple connection rule uses the product name as the top-level node of the SysML model hierarchy, connects the triples according to the entity name, and adds them one by one to the SysML model hierarchy. If the triple is a module and its composite relationship triple, the tail entity is added as a child node to the next level of the head entity node in the SysML model hierarchy. If the head entity node does not exist in the SysML model hierarchy, a new head entity node is created. If the triple is a component attribute and its connector relationship triple, the tail entity in the triple is added to the next level of the head entity node in the SysML model hierarchy, and a connector label is added to identify the connector relationship. If the head entity node does not exist in the SysML model hierarchy, a new head entity node is created.

[0079] The triple deduplication rules include: for module and composite relationship triples, when the head entity, relationship, and tail entity of the triples are all the same, it is judged as a duplicate; for component attribute and connector relationship triples, when the relationship of the triples is the same and the name set of the head entity and tail entity are the same, it is judged as a duplicate; when constructing the SysML model hierarchical structure, duplicate triples are ignored.

[0080] The rules for enhancing the SysML model hierarchical structure include: after all triples are connected, connecting the top nodes of multiple hierarchical structures to the only top node of the SysML model hierarchical structure; for all component attributes and their connector relationship triples, creating a module corresponding to the component attribute; if the module does not appear in the SysML model hierarchical structure, adding it to the SysML model hierarchical structure and connecting it to the top node; otherwise, ignoring the module.

[0081] In the embodiment, specifically, an example of the predefined rules is as follows: first, a top-level node in the SysML model hierarchy is created based on the product name; then, triples are read in the order of the list of entity and relationship triples obtained in step S3; then, the triple connection rules and triple deduplication rules are applied to each triple, and nodes corresponding to the module and component attributes in the triple are created in sequence, and added to the SysML model hierarchy according to the relationships in the triples; finally, after all triples have been read, the SysML model hierarchy structure enhancement rules are applied to the initially generated SysML model hierarchy to obtain the final enhanced and perfect SysML model hierarchy. The numerous and complex triples are converted into a concise SysML model hierarchy, which facilitates the retrieval and utilization of required elements when subsequently creating XMI files.

[0082] S5. According to the SysML model hierarchical structure, an XMI file is created in accordance with the XMI format specification for representing SysML model information to obtain a complete SysML model.

[0083] like Figure 6 As shown, the XMI format specification consists of three parts: SysML model structure information, SysML model graphics information, and other necessary tags. SysML model structure elements must first be mapped to XMI tags, and then the graphics information and other necessary tags must be added to the XMI file. The steps involved in creating the XMI file include mapping SysML model elements to XMI tags, adding SysML model graphics information, and an algorithm for creating XMI tags based on the SysML model hierarchy.

[0084] This step requires parsing the standard XMI format that represents SysML model information. According to the order in the XMI format and the mapping between SysML models and XMI tags, the necessary XMI tags are gradually created to finally obtain a complete and standardized XMI file. The XMI file can be imported into the modeling software to obtain and display the complete SysML model.

[0085] The mapping between SysML model elements and XMI tags includes: the mapping between modules, composite relationships, component attributes, and connector relationships in the SysML model and the corresponding tags in XMI. Specifically, the mapping relationship between SysML model elements and XMI tags is shown in Table 1 below:

[0086] Table 1

[0087]

[0088] For example, the first mapping in Table 1, when creating the XMI file, each read module information, will be created in the XMI structure corresponding position label attribute "name = block" "packagedElement" tag, "block" is the module name. The tag also contains a globally unique "xmi:id", "xmi:type" and other label attributes to identify the element type, used to store the various details of the module.

[0089] Add SysML model graphical information, including: the name of the frame, the module to which it belongs, the display position, the frame size, the display position, the display size, the display style of the module and component attributes, the display position, the display length, and the display style of the connection line representing the composite relationship and the connector relationship. In SysML, two graphs, BDD and IBD, in addition to containing SysML model structure information, also need to add graphical information for visualization.

[0090] Specifically, after creating the SysML model structure information in the XMI file, the corresponding BDD and IBD need to be created according to the SysML model structure information, such as: modules and their composite relationships need to create BDD belonging to the parent module, component attributes and their connector relationships need to create IBD corresponding to the module of the component attribute. After creating BDD and IBD, the corresponding extension graph node will be created under the XMI root node, which is used to store the graphical information of BDD and IBD. BDD node and BDD extension graph node, IBD node and IBD extension graph node are connected by a globally unique "xmi:id". The extension graph node contains all the graphical information needed to draw BDD and IBD, such as: frame information, which specifies the area to be drawn; the specific coordinates of the module in the graph, the size and color of the icon representing the module; the specific arrow symbol representing the composite relationship; the coordinates of the two segments of the arrow, the style and color of the arrow line, etc.

[0091] The algorithm for creating XMI tags according to the SysML model hierarchy structure includes: first, creating the BDD of the entire model, all modules in the BDD, and the composite relationships between the modules according to the SysML model hierarchy structure, then creating the component attributes under the module according to the composite relationship information of the module, and then creating the IBD and connector relationships according to the component attributes and their connector relationships, and finally creating the corresponding extension graph node under the XMI root node according to each existing BDD and IBD, which is used to store the SysML model graphical information for displaying the graph.

[0092] The algorithm for creating XMI tags based on the SysML model hierarchy is as follows: The XMI to be created is divided into two parts: the model part, which contains the SysML model structure information, and the graphical part, which contains the visualization information. First, a root node for the entire model part is created, along with a package containing all the model structure information under this node. Next, under this package node, three parts are created: a BDD describing the entire model architecture, all modules within the BDD, and the composite relationships between these modules. Next, under each module, component attributes of the modules owned by the module are created based on the composite relationship information. If a module has component attributes, an IBD is created for it and assigned to the module; otherwise, the module has no IBD. Finally, connections are created based on the connector relationships between the component attributes. After creating the model part, all BDDs and IBDs within the model part are retrieved. Then, corresponding extended graphical nodes are added to the XMI root node in the order in which the BDDs and IBDs appear. Finally, corresponding visualization graphical information is created based on the structural information in the BDDs and IBDs. The resulting XMI file contains the complete SysML model information. The entire creation process creates the corresponding XMI tags in the order in which modules and their composite relationships, component attributes, and their connector relationships appear in the SysML model hierarchy. The retrieval of required elements during the creation process also relies on the SysML model hierarchy.

[0093] In order to clearly demonstrate the SysML model generation method based on a large language model in a conceptual design provided by the present invention, a corresponding example is presented below.

[0094] ①. Question: Generate a conceptual design for a lunar rover. The final answer simply requires telling me what components are needed and how they are connected.

[0095] Answer: Let's think about it step by step:

[0096] Step 1: Clarify the requirements of the product: The product needs to be able to detect the composition of lunar soil, water ice distribution, geological structure and dynamic environment, and adapt to the extreme environment of the moon such as low gravity, extreme temperature difference, lunar dust, vacuum, etc. It must be able to support short-distance flight using rotor mode and long-distance jump using thruster mode, and have autonomous navigation and multi-detector collaboration functions.

[0097] Step 2: To meet the requirements of Step 1, the following functions must be implemented: Core functions include mobility and navigation, energy management, scientific exploration, and communication and collaboration. Mobility and navigation include short-distance flight, long-distance jumps, and autonomous obstacle avoidance. Energy management includes power generation, energy storage, and thermal control. Scientific exploration includes sampling, imaging, and environmental monitoring. Communication and collaboration includes data transmission and formation control. Other functions include terrain perception, power mode switching, device temperature regulation, and data compression and backhaul.

[0098] Step 3: A more detailed description of the various functions in Step 2 is provided: The mobility module will utilize rotorcraft for flight (50-200 m altitude) in illuminated areas and cold air thrusters for hops (50 km per trip) in shadowed areas. The energy module will deploy solar panels for charging during the day and utilize RTGs for power supply at night, dynamically allocating energy to various subsystems. The exploration module will scan the lunar surface in real time during flight (LiDAR + spectrometer) and, after landing, drill and analyze samples. The communications module will upload data to the orbiter via phased array antennas, synchronizing the status of multiple detectors via a local mesh network.

[0099] Step 4: To achieve the various functions detailed in Steps 2 and 3, the product requires the following structure: a power system, energy system, navigation system, detection system, control system, and communication system. The power system includes a rotor propulsion system, cold gas thrusters, and fuel tanks. The rotor propulsion system uses electric ducted rotors to provide lift for short-range flight and connects to the power management module (energy input) and the navigation controller (flight command reception). The cold gas thrusters use nitrogen fuel for propulsion and perform long-range hops. They connect to the fuel tanks, the navigation controller (ignition command), and the inertial measurement unit (IMU). The energy system includes foldable solar panels, RTG nuclear batteries, and energy storage batteries. The foldable solar panels use flexible perovskite batteries for daytime power generation and connect to the power management module (charging control) and energy storage batteries. The RTG nuclear batteries provide continuous power during the lunar night and connect to the power management module (priority switching) and the thermal control system (waste heat utilization). The navigation system includes LiDAR, a visual camera, an IMU, and a star sensor. The LiDAR and visual camera are responsible for real-time terrain modeling and obstacle avoidance, connecting to the main control computer (transmitting point cloud data) and a storage unit (map cache). The detection system includes a micro SAR radar, a spectrometer, and a modular payload interface. The micro SAR radar is responsible for detecting the distribution of underground water ice and connects to the main control computer (data preprocessing) and the communication module (transmitting results). The modular payload interface supports expansion equipment (drills, seismometers, etc.) and connects to the main control computer (control instructions) and the power module (independent power supply). The control system includes a main control computer, a power management module, and a thermal control system. The main control computer is responsible for running autonomous navigation algorithms and mission planning, and connects all sensors (LiDAR, camera, IMU), the propulsion system, and the communication module (data flow hub). The thermal control system uses phase change material (PCM) and electric heating film for temperature regulation and connects to the RTG (waste heat collection), the equipment compartment (temperature sensor), and the external heat sink (radiative heat dissipation). The communication system includes phased array antennas and Mesh network relay nodes. The phased array antennas are responsible for high-speed data transmission and relay, connecting the main control computer (command reception) and the deep space network (data uplink).

[0100] Therefore, the answer is: the lunar flight probe includes a power system, an energy system, a navigation system, a detection system, a control system, and a communication system. The power system includes a rotor propulsion system, a cold air thruster, and a fuel storage tank. The energy system includes foldable solar panels, RTG nuclear batteries, and energy storage batteries. The navigation system includes LiDAR, a visual camera, an IMU, and a star-sensitive sensor. The detection system includes a micro SAR radar, a spectrometer, and a modular payload interface. The control system includes a main control computer, a power management module, and a thermal control system. The communication system includes a phased array antenna and a Mesh network relay node. The rotor propulsion system is connected to the power management module and the navigation controller, the cold gas thrusters are connected to the fuel tank, the navigation controller (ignition command) and the inertial measurement unit (IMU), the foldable solar panels are connected to the power management module and the energy storage battery, the RTG nuclear battery is connected to the power management module and the thermal control system, the LiDAR and visual camera are connected to the main control computer and the storage unit, the micro SAR radar is connected to the main control computer and the communication module, the modular payload interface is connected to the main control computer and the power module, the main control computer is connected to the LiDAR, visual camera, IMU, propulsion system and communication module, the thermal control system is connected to the RTG, the equipment cabin and the external heat sink, and the phased array antenna is connected to the main control computer and the deep space network.

[0101] ②Problem: Generate a conceptual design for a diesel engine. The final answer only requires telling me which components are needed and how they are connected.

[0102] Answer: Let's think about it step by step:

[0103] Step 1: Clarify the product requirements: convert diesel chemical energy into mechanical energy and output stable torque and power.

[0104] Step 2: To meet the requirements of Step 1, the following functions must be implemented: Primary functions include efficient diesel combustion to release chemical energy, energy conversion (linear piston motion to crankshaft rotation), and waste heat and exhaust gas management. Sub-functions include precise fuel injection, efficient air compression, mechanical energy transfer, cooling systems, and exhaust aftertreatment.

[0105] Step 3: Describe the various functions of the second step in more detail: Through each function, the four-stroke cycle of intake → compression → combustion → exhaust is realized, and energy is transmitted through the path of piston → connecting rod → crankshaft → flywheel / output shaft. In addition, the injection amount and boost pressure need to be adjusted in real time. The fuel is atomized under high pressure. The intake volume also needs to be dynamically adjusted, and the emission valve needs to be adaptively adjusted according to sensor feedback.

[0106] Step 4: To achieve the various functions detailed in Steps 2 and 3, the product requires the following structure: Core components include the combustion system, intake and exhaust system, powertrain, control system, exhaust aftertreatment system, and lubrication and cooling system. The combustion system comprises the cylinder block, piston, and fuel injector. The cylinder block accommodates the piston's motion, forming the combustion chamber. The piston converts combustion pressure into linear motion. The fuel injector precisely atomizes the fuel and injects it into the combustion chamber. The intake and exhaust system includes a turbocharger, intercooler, and exhaust gas recirculation valve. The turbocharger compresses the intake air to increase oxygen density. The intercooler cools the compressed, high-temperature air to improve combustion efficiency. The EGR valve recirculates some exhaust gas back to the intake port to reduce NOx emissions. The powertrain system comprises the crankshaft, connecting rod, and flywheel. The crankshaft converts the piston's linear motion into rotational motion. The connecting rod connects the piston to the crankshaft, transferring kinetic energy. The flywheel stores rotational inertia and balances output torque fluctuations. The control system comprises the ECU and a sensor network. The ECU receives sensor signals to control fuel injection, boost pressure, and EGR. The sensor network monitors engine status in real time. The exhaust aftertreatment system includes a diesel oxidation catalyst (DOC) and a particulate filter (DPF). The DOC oxidizes CO, while the DPF filters PM particles from the exhaust. The lubrication and cooling system includes a variable-displacement oil pump and a split-type water pump. The variable-displacement oil pump supplies lubricant to friction components such as the crankshaft and pistons on demand, while the split-type water pump independently controls coolant flow to the cylinder block and cylinder head. The piston, connecting rod, crankshaft, and flywheel are mechanically connected to form a linear-rotational energy transmission chain. The turbocharger, intercooler, intake manifold, and cylinders are mechanically connected to deliver compressed air to the combustion chamber. The exhaust pipe, turbocharger, EGR valve, DOC, and DPF are also mechanically connected to form the exhaust gas treatment and energy recovery pathway. The fuel tank, high-pressure fuel pump, common rail, and injectors are fluidically connected to form the fuel supply path. The oil pump, main oil gallery, crankshaft bearing / piston pin are fluidically connected to form the lubricating oil circulation path. The water pump, cylinder block water jacket, cylinder head water jacket, and radiator are fluidically connected to form the coolant circulation path. The sensor network is connected to the ECU via control signals for data input. The ECU is connected to the injectors, EGR valve, turbocharger vane angle, and oil pump displacement via control signals to execute control outputs.

[0107] Therefore, the answer is: a diesel engine includes a combustion system, an intake and exhaust system, a power transmission system, a control system, an emission after-treatment system, and a lubrication and cooling system. The combustion system includes a cylinder block, a piston, and a fuel injector; the intake and exhaust system includes a turbocharger, an intercooler, and an EGR valve; the power transmission system includes a crankshaft, a connecting rod, and a flywheel; the control system includes an ECU and a sensor network; the emission after-treatment system includes a diesel oxidation catalyst (DOC) and a particulate filter (DPF); and the lubrication and cooling system includes a variable displacement oil pump and a split water pump. The piston, connecting rod, crankshaft, and flywheel are mechanically connected, the turbocharger, intercooler, intake pipe, and cylinder are mechanically connected, the exhaust pipe, turbocharger, EGR valve, DOC, and DPF are mechanically connected, the fuel tank, high-pressure oil pump, common rail pipe, and injector are fluidically connected, the oil pump, main oil gallery, crankshaft bearing / piston pin are fluidically connected, the water pump, cylinder water jacket, cylinder head water jacket, and radiator are fluidically connected, the sensor network is connected to the ECU through control signals, and the ECU is connected to the injector, EGR valve, turbocharger blade angle, and oil pump displacement through control signals.

[0108] ③. Question: Generate a conceptual design for a sonar. The final answer only requires telling me which components are needed and how they are connected.

[0109] Answer: Let's think about it step by step:

[0110] Step 1: Clarify the requirements of the product: The sonar system needs to be able to detect the position, distance, speed and other information of underwater targets through sound waves, adapt to the complex underwater environment, and have high precision and anti-interference capabilities.

[0111] Step 2: To meet the requirements of Step 1, the following functions must be implemented: Core functions include acoustic wave transmission and reception, signal processing, target identification and positioning, and data output. Acoustic wave transmission and reception include acoustic wave transmission, reception, and signal amplification. Signal processing includes noise filtering, signal enhancement, and target feature extraction. Target identification and positioning includes target distance, velocity, and bearing calculation. Data output includes data visualization and data transmission.

[0112] Step 3: Describe the various functions in Step 2 in more detail: The acoustic wave transmission module generates an acoustic wave signal of a specific frequency and transmits it into the water through a transducer. The acoustic wave receiving module receives and amplifies the reflected acoustic wave signal through a hydrophone. The signal processing module filters and enhances the received signal and extracts target features. The target recognition and positioning module calculates the target's range, velocity, and direction based on the processed signal. The data output module displays the processed data through a visual interface or transmits it to other systems.

[0113] Step 4: To achieve the various functions described in steps 2 and 3, the product requires the following structure: an acoustic wave transmission system, an acoustic wave reception system, a signal processing system, a target identification and positioning system, and a data output system. The acoustic wave transmission system includes a signal generator, a power amplifier, and a transducer. The signal generator generates an acoustic wave signal of a specific frequency and connects to the power amplifier (signal amplification) and the transducer (acoustic wave transmission). The acoustic wave reception system includes a hydrophone and a preamplifier. The hydrophone receives the reflected acoustic wave signal and connects to the preamplifier (signal amplification) and the signal processing system (signal input). The signal processing system includes a filter and a signal processor. The filter is used for noise filtering and connects to the preamplifier (signal input) and the signal processor (signal enhancement and feature extraction). The target identification and positioning system includes a distance calculation module, a velocity calculation module, and a bearing calculation module. The distance calculation module connects to the signal processor (signal input) and the data output system (distance data output). The velocity calculation module connects to the signal processor (signal input) and the data output system (speed data output). The bearing calculation module connects to the signal processor (signal input) and the data output system (bearing data output). The data output system includes a data visualization module and a data transmission module. The data visualization module connects the target recognition and positioning system (data input) and the display device (data output), and the data transmission module connects the target recognition and positioning system (data input) and the external system (data transmission).

[0114] Therefore, the answer is: the sonar system consists of an acoustic wave transmission system, an acoustic wave reception system, a signal processing system, a target identification and positioning system, and a data output system. The acoustic wave transmission system includes a signal generator, a power amplifier, and a transducer; the acoustic wave reception system includes a hydrophone and a preamplifier; the signal processing system includes a filter and a signal processor; the target identification and positioning system includes a distance calculation module, a velocity calculation module, and a bearing calculation module; and the data output system includes a data visualization module and a data transmission module. The signal generator connects the power amplifier to the transducer, the hydrophone to the preamplifier to the signal processing system, the filter to the preamplifier to the signal processor, the distance calculation module, the velocity calculation module, and the bearing calculation module to the signal processor and the data output system, respectively. The data visualization module and the data transmission module connect the target identification and positioning system to a display device or external system, respectively. This answer will be used in subsequent steps.

[0115] Next, you are an experienced SysML modeler. Now you are required to divide the input text into relatively complete sentences and output a list of sentences with complete semantics.

[0116] The sonar system includes a sound wave emitting system, a sound wave receiving system, a signal processing system, a target identification and positioning system, and a data output system. The sound wave emitting system includes a signal generator, a power amplifier, and a transducer. The sound wave receiving system includes a hydrophone and a preamplifier. The signal processing system includes a filter and a signal processor. The target identification and positioning system includes a distance calculation module, a speed calculation module, and a direction calculation module. The data output system includes a data visualization module and a data transmission module. The signal generator is connected to the power amplifier and the transducer. The hydrophone is connected to the preamplifier and the signal processing system. The filter is connected to the preamplifier and the signal processor. The distance calculation module, the speed calculation module, and the direction calculation module are connected to the signal processor and the data output system. The data visualization module and the data transmission module are connected to the target identification and positioning system and a display device or an external system.

[0117] The sonar system includes a sound wave emitting system, a sound wave receiving system, a signal processing system, a target identification and positioning system, and a data output system.

[0118] The sound wave emitting system includes a signal generator, a power amplifier, and a transducer.

[0119] The sound wave receiving system includes a hydrophone and a preamplifier.

[0120] The signal processing system includes a filter and a signal processor.

[0121] The target identification and positioning system includes a distance calculation module, a speed calculation module, and a direction calculation module.

[0122] The data output system includes a data visualization module and a data transmission module.

[0123] The signal generator is connected to the power amplifier and the transducer.

[0124] The hydrophone is connected to the preamplifier and the signal processing system.

[0125] The filter is connected to the preamplifier and the signal processor.

[0126] The distance calculation module, the speed calculation module, and the direction calculation module are connected to the signal processor and the data output system.

[0127] The data visualization module and the data transmission module are connected to the target identification and positioning system and a display device or an external system.

[0128] Next, you are a senior SysML modeler, and now you need to extract entity relationships from each sentence in the complete semantic sentence list. Output a list of triples containing entities and their relationships.

[0129] Input sentence list: The sonar system includes a sound wave transmitting system, a sound wave receiving system, a signal processing system, a target identification and positioning system, and a data output system.

[0130] The acoustic wave emission system includes a signal generator, a power amplifier, and a transducer.

[0131] The sound wave receiving system includes a hydrophone and a preamplifier.

[0132] The signal processing system includes filters and signal processors.

[0133] The target recognition and positioning system includes a distance calculation module, a speed calculation module, and a direction calculation module.

[0134] The data output system includes data visualization module and data transmission module.

[0135] The signal generator is connected to the power amplifier and the transducer.

[0136] The hydrophone is connected to the preamplifier and signal processing system.

[0137] The filter connects the preamplifier to the signal processor.

[0138] The distance calculation module, the speed calculation module and the direction calculation module are respectively connected to the signal processor and the data output system.

[0139] The data visualization module and the data transmission module are connected to the target recognition and positioning system and the display device or the external system respectively.

[0140] The output is a list of triples of entities and their relations, specifically:

[0141] Module: Sonar, Relationship: Composite Relationship, Module: Acoustic Emission System

[0142] Module: Sonar, Relationship: Composite Relationship, Module: Sound Wave Receiving System

[0143] Module: Sonar, Relationship: Composite Relationship, Module: Signal Processing System

[0144] Module: Sonar, Relationship: Composite Relationship, Module: Target Recognition

[0145] Module: Sonar, Relationship: Composite Relationship, Module: Positioning System

[0146] Module: Sonar, Relationship: Composite Relationship, Module: Data Output System

[0147] Module: Acoustic Wave Emission System, Relationship: Composite Relationship, Module: Signal Generator

[0148] Module: Acoustic Wave Transmission System, Relationship: Composite Relationship, Module: Power Amplifier

[0149] Module: Acoustic Wave Emission System, Relationship: Composite Relationship, Module: Transducer

[0150] Module: Acoustic Wave Receiving System, Relationship: Composite Relationship, Module: Hydrophone

[0151] Module: Sound wave receiving system, Relationship: Composite relationship, Module: Preamplifier

[0152] Module: Signal Processing System, Relation: Composite Relation, Module: Filter

[0153] Module: Signal Processing System, Relation: Composite Relation, Module: Signal Processor

[0154] Module: Positioning System, Relationship: Composite Relationship, Module: Distance Calculation Module

[0155] Module: Positioning System, Relationship: Composite Relationship, Module: Speed ​​Calculation Module

[0156] Module: Positioning System, Relationship: Composite Relationship, Module: Position Calculation Module

[0157] Module: Data Output System, Relationship: Composite Relationship, Module: Data Visualization Module

[0158] Module: Data Output System, Relationship: Composite Relationship, Module: Data Transmission Module

[0159] Component attributes: Signal generator, Relationship: Connector relationship, Component attributes: Power amplifier

[0160] Component Properties: Signal Generator, Relationship: Connector Relationship, Component Properties: Transducer

[0161] Component Properties: Hydrophone, Relationship: Connector Relationship, Component Properties: Preamplifier

[0162] Component Properties: Hydrophone, Relationship: Connector Relationship, Component Properties: Signal Processing System

[0163] Component Properties: Filter, Relationship: Connector Relationship, Component Properties: Preamplifier

[0164] Component Properties: Filter, Relationship: Connector Relationship, Component Properties: Signal Processor

[0165] Component attributes: distance calculation module, relationship: connector relationship, component attributes: signal processor

[0166] Component properties: Speed ​​calculation module, Relationship: Connector relationship, Component properties: Signal processor

[0167] Component attributes: Orientation calculation module, Relationship: Connector relationship, Component attributes: Signal processor

[0168] Component attributes: distance calculation module, relationship: connector relationship, component attributes: data output system

[0169] Component properties: Speed ​​calculation module, Relationship: Connector relationship, Component properties: Data output system

[0170] Component attributes: Orientation calculation module, Relationship: Connector relationship, Component attributes: Data output system

[0171] Component attributes: data visualization module, relationship: connector relationship, component attributes: target recognition

[0172] Component properties: Data visualization module, Relationship: Connector relationship, Component properties: Positioning system

[0173] Component properties: Data visualization module, Relationship: Connector relationship, Component properties: Display device

[0174] Component properties: Data visualization module, Relationship: Connector relationship, Component properties: External system

[0175] Component attributes: Data transmission module, Relationship: Connector relationship, Component attributes: Target recognition

[0176] Component properties: Data transmission module, Relationship: Connector relationship, Component properties: Positioning system

[0177] Component properties: Data transmission module, Relationship: Connector relationship, Component properties: Display device

[0178] Component properties: Data transfer module, Relationship: Connector relationship, Component properties: External system

[0179] Next, a SysML model hierarchy is created using the triples containing entities and their relationships as input. The SysML model hierarchy is specifically as follows:

[0180] -Sonar

[0181] -Acoustic emission system

[0182] -Signal generator

[0183] -to power amplifier

[0184] -to transducer

[0185] -Power amplifier

[0186] -Transducer

[0187] -Sound wave receiving system

[0188] hydrophone

[0189] pre-amplifier

[0190] pre-amplifier

[0191] signal processing system

[0192] filter

[0193] signal processor

[0194] signal processor

[0195] positioning system

[0196] distance calculation module

[0197] speed calculation module

[0198] azimuth calculation module

[0199] data output system

[0200] data visualization module

[0201] data transmission module

[0202] display device

[0203] target identification

[0204] external system

[0205] Finally, taking the above SysML model hierarchy as input, an XMI file is generated, and the XMI file is imported into modeling software to obtain a visual SysML model, such as Figure 7 a BDD diagram generated for a sonar instance, Figure 8 an IBD diagram of a SysML model generated for a sonar instance, wherein Figure 7 the small icons under “sonar” “sound wave emitting system” “sound wave receiving system” “signal processing system” in the internal module diagram of “sonar” correspond to the four IBD diagrams of Figure 8 , which serve as a navigation function. Figure 8 In the internal module diagram of “sonar”, the small icons under “sound wave emitting system” “sound wave receiving system” “signal processing system” correspond to the other three IBD diagrams, which also serve as a quick navigation function.

[0206] Based on the results of the examples, it can be seen that the SysML model generation method based on the large language model in the conceptual design provided by the present invention uses a small number of sample thought chain prompt words to ensure the quality of the design plan while minimizing manual operation. In addition, the rich knowledge base of the large language model is used to generate the conceptual design plan, and based on this, the SysML model is automatically modeled and generated. The entire process does not require the manual construction of additional professional field data sets. The large language model can be used to generate conceptual design plans for product names in different fields; there is no restriction on the input, and even if there is only one product name, the SysML model corresponding to the conceptual design plan of the product name can be automatically generated; the general XMI format specification file is used to represent the SysML model, which can be imported into commonly used practical modeling software that supports the XMI format, facilitating designers to carry out subsequent further modeling and modification, making it more practical and further improving the degree of automation.

[0207] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A SysML model generation method based on a large language model in conceptual design, characterized by: The following steps are involved: Input user requirements related to conceptual design into the large language model, and use the preset thought chain prompt word template to guide the large language model to output the conceptual design plan described in plain text form; Using the preset prompt word template to guide the large language model, the conceptual design scheme described in plain text is divided into sentences and a list of sentences with complete semantics is output; Using the preset prompt word template 2 to guide the large language model, entity relationship extraction is performed on each sentence in the sentence list containing complete semantics, and triples containing entities and their relationships are output; According to the triples containing entities and their relationships, a SysML model hierarchy is created according to predefined rules, wherein the predefined rules include: rules for connecting triples, rules for removing duplicate triples, and rules for enhancing the SysML model hierarchy; the triple connection rule uses the product name as the top node of the SysML model hierarchy, connects the triples according to the entity name, and adds them one by one to the SysML model hierarchy; if the triple is a module and its composite relationship triple, the tail entity is added as a child node to the next level of the head entity node of the SysML model hierarchy; if the head entity node does not appear in the SysML model hierarchy, a new head entity node is created; if the triple is a component attribute and its connector relationship triple, the tail entity in the triple is added to the next level of the head entity node of the SysML model hierarchy, and a connection is added. The connector tag identifies the connector relationship. If the head entity node does not appear in the SysML model hierarchy, a new head entity node is created. The triple deduplication rules include: for module and composite relationship triples, when the head entity, relationship and tail entity of the triples are the same, it is judged as a duplicate; for component attributes and their connector relationship triples, when the relationships of the triples are the same and the name sets of the head entity and tail entity are the same, it is judged as a duplicate; the SysML model hierarchy enhancement rules include: after all triples are connected, the top nodes of multiple hierarchical structures are connected to the only top node of the SysML model hierarchy; for all component attributes and their connector relationship triples, a module corresponding to the component attribute is created. If the module does not appear in the SysML model hierarchy, it is added to the SysML model hierarchy and connected to the top node; otherwise, the module is ignored; According to the SysML model hierarchical structure, an XMI file is created according to the XMI format specification for representing SysML model information to obtain a complete SysML model.

2. The method for generating a SysML model based on a large language model in a conceptual design according to claim 1, characterized in that: The user requirements for the conceptual design are a product name in a highly complex large-scale system, or a product name and a simple description of the product requirements.

3. The method for generating a SysML model based on a large language model in a conceptual design according to claim 1, characterized in that: The preset thought chain prompt word template is a small sample thought chain prompt word template, including a thought chain sample prompt template and a practical problem prompt template; The thought chain sample prompt template includes 2-10 thought chain samples. The prompt template of each thought chain sample includes: a question prompt template to guide the large language model to generate a product concept design plan, a fixed prompt to guide the large language model to perform thought chain reasoning, a sample concept design plan thought chain reasoning steps to guide the large language model to perform thought chain reasoning, and a fixed prompt to guide the large language model to summarize the concept design plan; The actual question prompt template includes: a question prompt template for guiding the large language model to generate a product concept design plan and a fixed prompt for guiding the large language model to perform thought chain reasoning.

4. The method for generating a SysML model based on a large language model in a conceptual design according to claim 3, characterized in that: The sample concept design scheme that guides the large language model to perform thought chain reasoning includes four mapping reasoning steps, including: product-demand mapping, demand-function mapping, function-behavior mapping and behavior-structure mapping. Each mapping reasoning step sets a corresponding mapping prompt word template.

5. The method for generating a SysML model based on a large language model in a conceptual design according to claim 1, characterized in that: The conceptual design scheme described in plain text includes the thought chain reasoning process and the structural information of the SysML model, where the structural information of the SysML model includes modules and their composite relationships, component attributes and their connector relationships.

6. The method for generating a SysML model based on a large language model in a conceptual design according to claim 1, characterized in that: The preset prompt word template 1 includes: a large language model role definition, a sentence segmentation task goal to be achieved, a sentence segmentation task solution example and specific instructions.

7. The method for generating a SysML model based on a large language model in a conceptual design according to claim 1, characterized in that: The preset prompt word template 2 includes: a large language model role definition, an entity relationship extraction task goal to be achieved, an entity relationship extraction task solution example and specific instructions.

8. The method for generating a SysML model based on a large language model in a conceptual design according to claim 1, characterized in that: Creating XMI files includes: mapping SysML model elements to XMI tags, adding SysML model graphic information, and creating XMI tag algorithms based on the SysML model hierarchical structure; The mapping of SysML model elements to XMI tags includes: mapping modules, composite relationships, component attributes, and connector relationships in the SysML model to corresponding tags in XMI; The added SysML model graphic information includes: the name of the frame, the module to which it belongs, the display position, the frame size, the display position, explicit size and display style of the module and component attributes, and the display position, display length and display style of the connection lines of the composite relationship and connector relationship; The algorithm for creating XMI tags based on the SysML model hierarchical structure includes: creating a module definition diagram, all modules in the module definition diagram, and the composite relationships between modules based on the SysML model hierarchical structure, then creating component attributes under the module based on the module composite relationship information, then creating an internal module diagram and connector relationships based on the component attributes and their connector relationships, and finally creating a corresponding extended graphic node under the XMI root node based on each existing module definition diagram and internal module diagram.

Images