A conceptual design method and equipment for industrial machine tools based on knowledge graphs and MBSE.

Abstract

This invention discloses a conceptual design method and device for industrial mother machines based on knowledge graphs and MBSE, belonging to the field of industrial design technology. This method utilizes the powerful semantic understanding and generation capabilities of RAG (Rich Language Aggregator) to combine the precise and reliable knowledge of knowledge graphs and vector databases, achieving in-depth analysis of complex requirements, generation of high-quality model elements, and more intelligent interaction. The SysML modeling tool serves as the final implementation and visualization platform for the model. The method includes: Stage 1: Collecting and preprocessing industrial mother machine data, constructing a core knowledge graph, building a vector knowledge base, and configuring the RAG system; Stage 2: Generating an initial model, dynamically modifying requirements, and enabling intelligent interaction; Stage 3: Automated traceability verification and structural and parameter consistency verification. This invention can rapidly generate digital prototype conceptual designs for industrial mother machines, improving the development efficiency of industrial mother machines.

Description

Technical Field

[0001] This invention belongs to the field of industrial design technology, specifically relating to a conceptual design method and equipment for industrial mother machines based on knowledge graphs and MBSE. Background Technology

[0002] SysML, as a general-purpose graphical modeling language, is specifically designed for the needs of systems engineering and is very suitable for the early conceptual design stage of complex systems. It has applications in aerospace, automotive, medical, and energy sectors, but its application still relies on engineers manually modeling using the platform.

[0003] As system complexity increases, traditional modeling methods struggle to meet the demands of multidisciplinary model integration. To enhance the automation of system modeling, some scholars have introduced AI technology into the modeling process. Yang et al., exploring autonomous system architecture modeling, combined traditional system modeling methods with machine learning, achieving automatic generation of requirement classification prediction and SysML architecture. Zhong et al., aiming to automatically generate system diagrams from unstructured natural language text, utilized NLP technology to extract entities and their relationships from the natural language text and transform them into SysML diagrams, promoting the standardization and automation of the system modeling process. López et al., addressing the issue of significant discrepancies between the generated results of system model generators and actual conditions, proposed a novel system model generator architecture capable of generating models with structures similar to real models using real model datasets. To address the error-prone and inefficient nature of traditional manual modeling methods, Akundi et al. proposed a method for automatically generating SysML class and block definition graphs based on NLP technology. To further improve the efficiency of MBSE system modeling, Zhang et al. proposed an automatic generation method for system model graphs based on knowledge graphs, which can automatically map SysML views. To address the shortage of human modeling experts in system design, Zhang et al. proposed a system modeling recommendation method based on NLP, which can automatically recommend suitable development processes to system engineers based on domain knowledge. The above studies all utilized AI technology to automatically or semi-automatically generate MBSE system models, demonstrating that AI technology can significantly improve the efficiency and quality of system modeling. However, because the process of AI generating system models is a "black box" and lacks guidance from physical knowledge, the generated models have poor interpretability and sometimes even produce system models that do not conform to reality. Therefore, further research is needed on how AI can empower the system modeling stage of MBSE. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the aforementioned related technologies.

[0005] Therefore, the purpose of this invention is to provide a conceptual design method and device for industrial mother machines based on knowledge graphs and MBSE, which can quickly generate digital prototype conceptual designs for industrial mother machines and improve the development efficiency of industrial mother machines.

[0006] To solve the above-mentioned technical problems, the present invention is implemented as follows: This invention provides a conceptual design method for industrial mother machines based on knowledge graphs and MBSE. The method utilizes RAG to combine the powerful semantic understanding and generation capabilities of large language models with the accurate and reliable knowledge of knowledge graphs and vector databases, thereby achieving in-depth analysis of complex requirements, generation of high-quality model elements, and more intelligent interaction; the SysML modeling tool serves as the final implementation and visualization platform for the model.

[0007] Furthermore, the industrial machine conceptual design method based on knowledge graphs and MBSE according to the present invention may also have the following additional technical features: In some of these implementations, the knowledge graph serves as a precise knowledge base for storing structured information in the industrial machine tool field, including core entities and semantic relationships. It is built on a graph database and supports precise querying and tracing. The vector database, serving as a semantic knowledge base, converts unstructured text into high-dimensional vectors for storage through a pre-trained embedding model, supporting efficient retrieval based on semantic similarity and compensating for the shortcomings of knowledge graphs in processing unstructured knowledge. The RAG system integrates a large language model, a knowledge graph, and a vector database. It first retrieves relevant text fragments from the vector database and entity relationships from the knowledge graph based on user needs, and then inputs the retrieval results as context into the large language model to achieve in-depth analysis of needs and generation of high-quality content, thereby avoiding the unreliability of the black-box output of the large language model. The SysML modeling tool: through automated scripts, it coordinates the RAG process, knowledge graph query, large language model call and SysML modeling tool API operation to realize the automatic creation, updating and management of SysML models, while supporting manual review and adjustment.

[0008] In some of these implementations, the core entity includes components, parameters, constraints, and requirements; The semantic relationships include: electric spindle - possessing - maximum speed, cooling system - satisfying - flow constraints.

[0009] In some implementations, the unstructured text includes historical project reports, supplier manuals, and papers.

[0010] In some implementations, the large language model is GPT4 or Claude.

[0011] In some implementations, the method includes the following steps: Phase 1. Knowledge Preparation and System Configuration: This includes collecting and preprocessing data on industrial mother machines, constructing a core knowledge graph, building a vector knowledge base, and configuring the RAG system. Phase 2. RAG-driven conceptual design and SysML model generation: including initial model generation, dynamic modification of requirements, and intelligent interaction; Phase 3. Knowledge Graph-Driven Automated Verification: This includes automated traceability verification, structural and parameter consistency verification, and the generation of final results.

[0012] In some of these implementations, during Phase 1, the collected industrial machine data includes historical project experience with similar industrial machine tools, component technical data, analysis and verification reports, industry standards, and external literature. Preprocessing includes: extracting text from the collected data, segmenting long documents into retrieval-friendly fragments, and supplementing metadata to provide a high-quality data source for vector databases and knowledge graphs; The content of constructing the core knowledge graph includes: Ontology design: Define the ontology structure in the industrial mother machine domain, and clarify the core entity types and relationship types; Knowledge extraction and fusion: Information is directly imported from structured data, and entities and relationships are extracted from unstructured text using NLP technology; entity alignment and conflict resolution are performed to ensure knowledge consistency. The construction of the vector knowledge base includes: using a pre-trained embedding model to convert pre-processed text blocks into high-dimensional vectors, storing them in a vector database to support subsequent semantic retrieval; The configuration of the RAG system includes: selecting a suitable large language model and developing a retrieval-enhancement-generation workflow; in the retrieval stage, matching vector library text with knowledge graph entities according to user needs; in the enhancement stage, integrating the retrieval results with the original requirements into a context; and in the generation stage, the large language model outputs content including a requirements analysis report and component selection suggestions.

[0013] In some implementations, stage 2, the generation of the initial model includes: Requirements input: Engineers submit design requirements in natural language; RAG Requirements Analysis and Knowledge Retrieval: The RAG system retrieves relevant papers on spindle thermal stability and historical project reports from the vector library, and simultaneously queries entities and related parameters in the knowledge graph, including spindle and cooling system. The large language model combines this information to generate a requirements analysis report, clarifying key requirements, sub-requirement decomposition schemes, and the basis for selecting core components. Script-driven modeling: Automated scripts parse reports generated by large language models, extract structured information, then use knowledge graphs to accurately query and supplement complete data, and finally call the SysML modeling tool API to automatically create SysML model elements and establish traceability relationships between requirements and components; The content of dynamic modification of requirements and intelligent interaction includes: Change Analysis: When requirements change, the RAG system re-retrieves knowledge, generates an impact analysis report, and identifies the affected components, potential risks, and response strategies. Intelligent question answering and human review: Engineers ask questions through the system, RAG retrieves relevant knowledge and generates answers using a large language model; at the same time, domain experts review the automatically generated model and correct complex scenarios not covered by the automation tools. Model Update: Based on RAG recommendations and expert feedback, the script calls the SysML modeling tool API to synchronously update the SysML model, ensuring that the model is consistent with the changed requirements.

[0014] In some implementations, in Phase 3, the automated traceability verification includes: comparing the knowledge graph and the SysML model using scripts to check whether there is a valid fulfillment relationship for each requirement, and whether the component pointed to by the relationship has a corresponding entity in the knowledge graph, so as to avoid the requirement and component being disconnected; Structure and parameter consistency verification: Based on the constraint rules in the knowledge graph, the script automatically verifies whether the parameter values ​​and structural connections of the components in the SysML model conform to the rules; for complex verification results, RAG assists in generating comprehensive evaluation opinions to provide a basis for optimization. The final deliverables include: after successful verification, the system outputs a complete SysML digital prototype conceptual model and a comprehensive evaluation report, including model structure, requirement traceability relationships, verification results and optimization suggestions, providing a unified benchmark for subsequent detailed design and simulation verification.

[0015] This invention also provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the knowledge graph and MBSE-based industrial machine conceptual design method as described in any of the preceding embodiments.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects: In this embodiment of the invention, the provided industrial mother machine conceptual design method based on knowledge graphs and MBSE clearly defines the MBSE-based SysML multi-view model system (including 4 sub-models: requirement model → SysML requirement diagram / use case diagram; process model → equipment state diagram / activity diagram; design model → block definition diagram / interface block diagram / internal block diagram; verification model → parameter diagram + indicator traceability) in the conceptual design stage of the industrial mother machine digital prototype. It constructs a unified model framework that is linked to the entire chain of "requirement-process-design-verification", clarifies the SysML modeling specifications of each link (such as the traceability relationship between requirement diagram and block diagram), and makes the model a "single source of facts", providing a reliable benchmark for subsequent multidisciplinary collaboration (such as the structural design and thermal simulation teams sharing the same model) and detailed design implementation, completely getting rid of the dilemma of "fragmented diagrams"; In this embodiment of the invention, the provided industrial machine tool conceptual design method based on knowledge graph and MBSE improves interpretability: all design decisions can be traced back to specific entities and relationships in the knowledge graph (e.g., spindle selection basis → electric spindle - maximum speed - 18000rpm, electric spindle - temperature rise constraint - ≤15℃ relationship in the knowledge graph), completely solving the "black box" problem, and engineers can clearly verify the design logic; engineering practicality is guaranteed: during the generation process, RAG calls the domain rules of the knowledge graph (e.g., oil cooling flow rate ≥5L / min) and historical engineering cases in the vector database to avoid generating models that are "syntactically correct but do not conform to manufacturing constraints and physical laws," ensuring that the model is close to industrial reality; In this embodiment of the invention, the provided industrial machine conceptual design method based on knowledge graph and MBSE establishes a closed-loop management mechanism for dynamic requirement changes (RAG impact analysis → automated script model update → expert feedback correction → knowledge graph iteration). RAG can retrieve knowledge in real time to generate impact analysis reports, saving manual analysis time. The script calls SysML tools (such as MagicDraw) APIs based on the RAG report to automatically update parameters and component relationships in the model. After combining expert feedback to correct details, the change experience is injected into the knowledge graph (such as speed 16000rpm - oil cooling flow rate 4L / min), forming a change-analysis-update-accumulation closed loop, ensuring that the model is consistent with the requirements in real time, reducing the amount of manual modification work by more than 80%. In this embodiment of the invention, the provided industrial machine conceptual design method based on knowledge graphs and MBSE integrates multi-source heterogeneous knowledge management from knowledge graphs (structured data) and vector databases (unstructured data). The knowledge graph stores structured data (component parameters, design constraints, requirement IDs), while the vector database stores unstructured text (historical project reports, spindle thermal deformation papers, FMEA reports). It also supports access to external industry standards and supplier technical documents. During the design process, multi-source knowledge can be called simultaneously (e.g., when designing a spindle, the "bearing model-rated speed" constraint can be obtained from the knowledge graph, and the "spindle thermal drift" FMEA case can be retrieved from the vector database to avoid historical risks), ensuring comprehensive design considerations and avoiding the omission of key information. In this embodiment of the invention, the provided industrial machine concept design method based on knowledge graphs and MBSE constructs a human-machine collaborative design environment of "AI automation + expert review." AI completes basic tasks such as requirement analysis and initial model generation through RAG, reducing repetitive labor for engineers (e.g., eliminating the need to manually draw basic block diagrams). Domain experts focus on innovative decisions (e.g., design of novel spindle configurations) and corrections for complex scenarios (e.g., supplementing vibration constraints under special cutting conditions), leveraging the advantages of human intelligence. Expert review feedback (e.g., adding "special working conditions - vibration thresholds") can be injected into the knowledge graph, continuously improving the accuracy of subsequent AI modeling, achieving "human-machine mutual promotion," and shortening the training cycle for new engineers (without relying on personal experience accumulation). In this embodiment of the invention, the provided industrial machine concept design method based on knowledge graph and MBSE achieves automated verification driven by knowledge graph (traceability verification: consistency of requirement-component association; structure / parameter verification: component parameters and connection relationships conform to knowledge graph constraints). The script automatically verifies based on knowledge graph rules (such as checking whether all "temperature rise requirements" have corresponding "cooling components" to meet them, and whether all "component parameters" conform to the constraint values ​​in the knowledge graph), replacing more than 90% of the manual verification work and significantly shortening the verification cycle. It avoids problems such as "missing corresponding components for requirements" and "parameter exceeding standards" (such as oil cooling flow rate being less than 5L / min) caused by human negligence, ensuring that the model structure, parameters and requirements and constraints are completely consistent, providing a reliable foundation for subsequent simulation verification.

[0017] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0018] Figure 1 This is a flowchart of an industrial machine tool conceptual design method based on knowledge graphs and MBSE, disclosed in one embodiment of the present invention. Detailed Implementation

[0019] 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, not all, of the embodiments of the present invention. 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.

[0020] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings and specific examples and application scenarios.

[0021] Current technologies lack a clear definition of conceptual design models for digital prototypes of industrial mother machines and do not place sufficient emphasis on the conceptual design phase. Introducing MBSE (Model-Based Design System) and utilizing the SysML language can significantly improve the development efficiency of industrial mother machines. MBSE emphasizes using models and simulations to represent the overall system architecture and requires teams from various disciplines to collaborate on the same model platform. This approach is more complex than traditional document management systems and demands higher professional skills from practitioners. MBSE practitioners not only need to learn modeling languages ​​such as UML and SysML but also need to shift their mindset from DBSE (Database-Based Design System) to MBSE. Systems are becoming increasingly complex and large-scale. With the increase in the number of system components, functions, and interactions, the complexity of modern products and systems is increasing dramatically. As the system scale grows larger, the number of disciplines involved broadens, and the number of participants increases, the amount of design information explodes. Such complex and large-scale systems push existing MBSE modeling methods and tools to their limits, limiting the application of MBSE in complex systems. The automation level of system modeling is low. Currently, the MBSE modeling process mainly relies on manual operation. System engineers need to gradually build system models based on system requirements, design specifications, and engineering practices. This manual modeling method is not only time-consuming, labor-intensive, and prone to errors, but also results in highly subjective MBSE models, making it difficult to guarantee their accuracy and consistency. Furthermore, as system complexity increases, the workload of manual modeling grows dramatically, further limiting the application of MBSE in complex systems. The reuse of historical MBSE models is also difficult. Although many systems have built numerous models during development, these models have extremely low reusability in subsequent projects. The specificity and coupling of these models often make them difficult to adapt to new design requirements, resulting in the historical models not playing their full role when building new models. In addition, MBSE models from different domains are difficult to interoperate and share due to differences in domain knowledge, semantic structure, and modeling standards. These limitations in model reuse not only increase the workload of modeling but also hinder the widespread application of MBSE in cross-domain collaborative design, limiting its efficiency improvement and promotion.

[0022] While introducing artificial intelligence (AI) technology into the modeling process has shown great potential in improving system modeling efficiency, existing AI4MBSE technology has revealed a series of deep-seated systemic flaws when applied to the development of highly complex and reliable equipment such as industrial machine tools, especially in the conceptual design phase. These flaws not only limit the practical value of AI technology but also hinder the deep implementation of MBSE in the development of complex equipment. The main problems are reflected in the following seven aspects: 1. The conceptual design model for digital prototypes lacks a clear definition, resulting in an incomplete model system. Most current AI-driven modeling methods (such as generating SysML diagrams based on NLP and predicting architectures using machine learning) often equate "model generation" with "diagram drawing," producing isolated, fragmented SysML diagrams (such as class diagrams and block diagrams). They fail to construct a complete, unified, and multi-view collaborative "conceptual design model" framework. For industrial mother machines, the conceptual design phase requires a systematic expression of functional flow, physical architecture, requirement traceability, behavioral logic, and interdisciplinary interfaces, necessitating a structured model system based on MBSE. However, existing technologies lack a clear definition and modeling specifications for "digital prototypes" in the conceptual phase, rendering the generated models unreliable and unified references for subsequent detailed design, simulation verification, and multi-disciplinary collaboration. Essentially, they remain at the level of "document replacement" rather than a true prototype.

[0023] 2. A severe neglect of the strategic value of the conceptual design phase. Conceptual design is the "source" that determines the performance, cost, reliability, and development cycle of industrial mother machines. A wrong architectural choice or a mistake in a key technology route will be amplified in subsequent stages, resulting in huge waste of resources. However, existing AI4MBSE research generally focuses on execution-level tasks such as requirement extraction, model transformation, or verification, with severely insufficient investment in the decision-making stage of conceptual design. They lack the ability to support design space exploration, solution trade-off analysis, and the generation of innovative configurations. AI is used to automate existing processes rather than empower innovative decisions, making conceptual design still highly dependent on the personal experience of a few experts. Knowledge is difficult to accumulate and reuse, leading to low design efficiency, limited innovation, and long training cycles for new engineers. This neglect of the conceptual design phase prevents AI technology from reaching the most critical value-creating point in the R&D process.

[0024] 3. The model generation process is a black box, lacking interpretability and credibility. Many existing AI-driven modeling methods (such as machine learning-based requirement classification and deep learning-based graph structure generation) rely on complex model parameters and implicit learning mechanisms, making their decision-making process essentially a "black box." Engineers cannot trace "why this system architecture was chosen" or "why this connection was established." Even if a SysML diagram is generated, the underlying logic is opaque, leading to low model credibility. In industrial machine tool design, an uninterpretable model implies potentially huge engineering risks. This lack of interpretability makes it difficult for engineers to trust and adopt AI-generated results, ultimately reducing them to "formal output" that cannot truly guide design decisions.

[0025] 4. Weak integration of domain knowledge leads to models that are easily detached from engineering practice. Although some research attempts to extract information from natural language using NLP, these methods heavily rely on the surface semantics of text and lack systematic knowledge support for deep engineering physics, manufacturing constraints, material properties, and industry standards. AI models may generate grammatically correct but engineering-infeasible designs (e.g., recommending spindle configurations with excessive rotational speed or insufficient cooling capacity) or ignore key design constraints (such as thermal deformation and vibration modes). This lack of domain expertise results in models that appear complete but are actually disconnected from real-world industrial scenarios, and may even introduce misleading information.

[0026] 5. Slow response to dynamic changes in requirements and lack of closed-loop evolution capabilities. The design process of industrial machine tools is fraught with uncertainty, and frequent adjustments to requirements during the conceptual stage are common. However, most existing AI4MBSE technologies focus on generating initial models under static requirements, with extremely limited capabilities to manage requirement changes. When customers adjust performance indicators or technical routes, the system cannot automatically assess the impact of the changes (such as the cascading effects on structure, thermal management, and control systems), nor can it drive automatic model updates and synchronized adjustments to trace relationships. This forces engineers to make numerous manual modifications, resulting in limited change management capabilities and easily leading to a disconnect between the model and requirements, violating the core principle of MBSE, "single source of fact."

[0027] 6. Limited information sources hinder the integration of structured and unstructured knowledge. Existing methods often rely on a single type of data source, such as processing only structured requirement tables or analyzing only unstructured text. They lack the ability to uniformly integrate and collaboratively utilize multi-source heterogeneous information (such as historical project reports, FMEA analyses, supplier manuals, simulation results, and standards and specifications). This weak ability to integrate multi-source information means that AI can only "feel the elephant blindly," unable to fully understand the design context. For example, it cannot automatically associate lessons learned from an FMEA report on "spindle thermal drift" with current design requirements for risk warning.

[0028] 7. Lack of human-machine collaboration mechanisms, resulting in a disconnect between AI and engineers. Current technologies often pursue "fully automated" generation, attempting to replace engineers with AI, but neglecting the crucial role of human experts in innovation, trade-offs, and judgment during conceptual design. If AI-generated models cannot effectively connect with engineers' cognitive processes, an inefficient cycle of "AI output - human verification - massive rework" will form. This lack of a human-machine collaborative design environment prevents AI from truly becoming an "intelligence enhancer" for engineers; instead, it may increase communication costs and cognitive burdens, failing to achieve a deep integration of human and machine intelligence.

[0029] In the design process of industrial machine tools, the conceptual design phase is a crucial step in determining the system architecture, selecting key components, and setting preliminary parameters. To improve the efficiency and accuracy of the conceptual design phase, this invention proposes a SysML model intelligent generation method combining Retrieval Enhancement (RAG) and knowledge graphs. This method not only efficiently handles complex natural language requirements but also ensures that the generated model is based on a reliable knowledge foundation and supports dynamic requirement change management. The conceptual design of the industrial machine tool digital prototype based on MBSE includes a requirement model, a process model, a design model, and a verification model. The requirement model uses SysML to create requirement diagrams and use case diagrams; the process model uses SysML to create equipment state diagrams and activity diagrams for the industrial machine tool; the design model uses SysML to create mechanism block definition diagrams, interface block definition diagrams, and internal block diagrams for the industrial machine tool; and the verification model uses SysML to create parameter diagrams for the industrial machine tool and performs indicator verification and traceability analysis.

[0030] Knowledge graph: As a "precise knowledge base", it stores structured domain ontology, entities, relations, parameters and constraints.

[0031] Vector database: As a "semantic knowledge base", it stores vector representations of unstructured text and supports efficient retrieval based on semantic similarity.

[0032] RAG system: Leveraging the powerful semantic understanding and generation capabilities of Large Language Model (LLM), combined with information from knowledge graphs and vector databases, it achieves in-depth analysis of complex requirements and generation of high-quality content.

[0033] SysML modeling tools (such as MagicDraw): As a concrete implementation platform for models, they receive instructions and data generated by collaborative decision-making between RAG and knowledge graphs, and realize the concrete creation, updating and management of SysML models through their APIs.

[0034] Automation scripts: responsible for coordinating RAG processes, knowledge graph queries, LLM calls, and MagicDraw API operations to achieve end-to-end automation.

[0035] In some embodiments of the present invention, the knowledge preparation and system configuration are as follows: 1. Data Collection and Preprocessing for Industrial Machine Tools By performing the aforementioned detailed data collection steps, collect historical project and design experience data of similar industrial machine tools, component and part technical data, analysis and verification data, external knowledge and standards, etc., to obtain comprehensive design knowledge.

[0036] Perform text extraction, chunking (splitting long documents into segments suitable for retrieval), and metadata supplementation.

[0037] 2. Construct a core knowledge graph Ontology design: Design the industrial mother machine domain ontology in a graph database (such as Neo4j), including core entities and their relationships.

[0038] Knowledge extraction and fusion: Direct import from structured data (Excel, DB); extraction of entities and relationships from unstructured text using NLP techniques; entity alignment and conflict resolution.

[0039] 3. Construct a vector knowledge base The text blocks are converted into high-dimensional vectors using a pre-trained embedding model and stored in a vector database.

[0040] 4. Configure the RAG system Choose a suitable LLM (such as GPT4, Claude) and develop a RAG workflow: Retrieval: Based on the user's input requirements, search for the most relevant text blocks in the vector database and query related entities and relationships in the knowledge graph.

[0041] Enhancement: Combine the retrieved information with the original query to build a context.

[0042] Generation: Input the Context into the LLM and generate the required output.

[0043] In some embodiments of the present invention, the content generated by the RAG-driven conceptual design SysML model includes: 1. Initial Model Generation Requirements Input: Engineers input a description of their requirements in natural language (e.g., “Design a five-axis linkage machining center with high-precision positioning function”).

[0044] RAG-driven requirements analysis and knowledge retrieval include: Search: The RAG system searches for relevant technical documents, design specifications, and historical case fragments in the vector library; at the same time, it queries key entities (such as Requirement type, Component category, and Parameter name) in the knowledge graph. Generation: LLM combines the retrieved context to generate a detailed requirements analysis report, including identified key requirements, suggested sub-requirements, recommended core components and their selection criteria, etc.

[0045] Model generation (script-driven MagicDraw): Script processing of RAG output: Relying on a series of predefined rules or heuristic algorithms, scripts automatically parse the reports generated by LLM and extract structured information; Precise knowledge graph query: The script uses the extracted information to perform precise queries in the knowledge graph to obtain the complete parameters of `Component`, the specific value of `Constraint`, the ID of `Requirement`, etc. Calling the MagicDraw API: The script uses the acquired data to automatically create `Requirement`, `Block`, and `Property` through the MagicDraw API, establish connections and trace relationships, and generate the initial SysML model.

[0046] 2. Supports dynamic modification of requirements and intelligent interaction. Change Analysis: When requirements change, the RAG system can generate a more in-depth impact analysis report, listing the affected components, potential risks, and response strategies.

[0047] Intelligent Question Answering and Suggestions: Engineers can ask questions within the system, and the RAG system retrieves relevant knowledge, generating comprehensive answers and suggestions from the LLM (Local Management Model) to aid decision-making. Initial suggestions from the RAG system can be reviewed manually. Although much of the work can be automated, in practice, domain experts often need to review and adjust the automatically generated model. This is because automated tools may not fully understand the nuances of all complex situations, requiring expert knowledge and judgment for optimization and refinement.

[0048] Requirements refinement and confirmation: During the requirements analysis phase, experienced engineers are often needed to ensure that the proposed requirements are reasonable and feasible, and can be effectively transformed into technical solutions.

[0049] Model update: The update process is the same as before, but the decision-making basis may include intelligent suggestions generated by RAG.

[0050] In some embodiments of the present invention, knowledge graph-driven automated verification includes: 1. Automated traceability verification The script compares the knowledge graph with the MagicDraw model to ensure that every "Requirement" element has a valid "satisfy" relationship, and that the "blocks" to which these relationships point have corresponding entities in the knowledge graph.

[0051] 2. Verification of structural and parameter consistency The core parameter comparison, constraint check, and structural consistency verification are still performed by the script based on precise data from the knowledge graph.

[0052] RAG Assist: For complex validation results, RAG can be used to generate comprehensive evaluation opinions.

[0053] This method systematically introduces the MBSE methodology and uses SysML as the medium. It clearly defines the SysML multi-view model system for digital prototypes of industrial mother machines in the conceptual design phase, laying the foundation for full lifecycle collaboration. By combining Retrieval Enhanced Generation (RAG) and knowledge graphs, this method achieves efficient parsing of complex requirements and intelligent generation of SysML models. It not only improves the accuracy and depth of requirement understanding but also ensures the precision and reliability of the model. Even when precise information cannot be retrieved, the design process can be effectively advanced through multi-source information integration, rule-based and heuristic algorithms, expert intervention mechanisms, and template-based methods. This method is particularly suitable for the conceptual design phase of complex equipment such as industrial mother machines, significantly improving design efficiency and quality, and enabling AI to empower core aspects of industrial mother machine R&D.

[0054] Currently, AI technology has been applied in various ways in system modeling, but it is still in its early stages. These applications mainly include demand classification and prediction, automatically generating system diagrams from natural language text, generating models with similar structures using real model datasets, automatically generating SysML class and block definition graphs based on NLP technology, and automatically mapping SysML views based on knowledge graphs. However, when compared with the method proposed in this invention that combines Retrieval Augmentation (RAG) and knowledge graphs, these technologies have some obvious limitations: 1. Insufficient explainability and transparency Limitations of existing technologies: Many AI-driven system modeling methods rely on machine learning models or deep learning frameworks, which are often considered "black boxes" and whose decision-making processes are difficult to explain. This means that even if high-quality models are generated, engineers may struggle to understand why certain design choices were made.

[0055] The advantages of this invention's method are: by using a knowledge graph as the core, all decisions can be traced back to specific entities and relationships, providing a high degree of transparency and interpretability. This not only helps engineers understand and validate the model but also promotes knowledge sharing across teams.

[0056] 2. Lack of domain expertise support Limitations of existing technologies: Although some methods attempt to extract information from natural language text using NLP, they often lack a deep understanding of specific domains, resulting in models that may not conform to reality or industry standards.

[0057] The advantages of this invention's method: This invention not only utilizes NLP to process natural language needs, but also incorporates domain knowledge graphs, ensuring that all generated content is based on reliable professional knowledge. This makes the generated models more accurate and closer to real-world application scenarios.

[0058] 3. Limited ability to manage dynamic changes. Limitations of existing technologies: Most existing AI4MBSE solutions focus on generating initial models and offer little support for subsequent requirement change management, making it difficult to cope with frequent changes throughout the project lifecycle.

[0059] The advantages of this invention's method are: It places particular emphasis on dynamic requirements management and intelligent response mechanisms. By rapidly assessing the impact of requirement changes through the RAG system and automatically generating corresponding adjustment suggestions, it greatly improves the system's flexibility and adaptability.

[0060] 4. Weak ability to integrate multi-source information. Limitations of existing technologies: Some technologies may rely on only a single data source or format, which limits their ability to handle diverse information needs.

[0061] The advantages of this invention are that it not only uses knowledge graphs to store structured data but also integrates vector databases to manage unstructured text information. Furthermore, this invention allows for the integration of data from more sources, such as external literature databases and industry standard documents, thereby enhancing the system's coverage and usability.

[0062] 5. Lack of a human-machine collaborative design environment Limitations of existing technologies: Although some technologies have achieved highly automated system modeling processes, when faced with complex problems, relying entirely on automated tools may ignore the role of human experts.

[0063] The advantages of this invention's method are: it designs a collaborative human-machine design environment, allowing domain experts to intervene and provide supplementary information when necessary, while simultaneously feeding this feedback back to the system to continuously enrich the knowledge graph. This approach leverages the advantages of artificial intelligence while fully utilizing the experience of human experts.

[0064] In summary, while existing AI-enabled MBSE system modeling techniques have improved modeling efficiency and quality to some extent, they still have limitations in interpretability, domain knowledge support, dynamic change management, multi-source information integration, and human-machine collaboration. In contrast, the method proposed in this invention, combining retrieval augmentation (RAG) and knowledge graphs, can better overcome these problems, providing a more comprehensive, flexible, and reliable system modeling solution. This method not only improves the automation level of system modeling but also ensures the accuracy, consistency, and interpretability of the generated model, making it particularly suitable for application in the research and development of complex equipment.

[0065] Example 1: Design of Spindle Unit for High-Speed ​​Vertical Machining Center A machine tool company received a new project request to design a new type of high-speed vertical machining center. One of the customer's core requirements is: "The maximum spindle speed must reach 18,000 rpm, while ensuring thermal stability under continuous high-load cutting, with the temperature rise at the front end of the spindle not exceeding 15°C within 30 minutes."

[0066] 1. Knowledge Preparation Stage A knowledge graph has been constructed, containing various types of entities and their relationships. These entities mainly include: Component: Refers to a specific physical part or subsystem within a system. For example, the "electric spindle" is the core component in a machine tool that enables high-speed rotation, and the "circulating oil cooling system" is a cooling device used to dissipate heat from heat-generating components such as the electric spindle.

[0067] Parameters: These represent the technical performance indicators of each component. For example, "maximum speed" describes the fastest rotational speed the electric spindle can achieve, and "rated power" indicates its output power during normal operation.

[0068] Constraints refer to the technical conditions or limitations that a system must meet during its design or operation. For example, "oil cooling flow rate ≥ XL / min" means that the cooling oil flow rate provided by the circulating oil cooling system must not be lower than a certain value, otherwise it may cause the equipment to overheat.

[0069] Requirement: Represents the overall functional or performance requirements put forward by the user or design goals. For example, "The temperature rise of the spindle should not exceed 15°C after 2 hours of continuous operation" is a typical performance requirement.

[0070] These entities are connected through various semantic relationships. For example, an "electric spindle" has a certain "maximum speed". This parameter may trigger a demand for the cooling capacity of the "circulating oil cooling system", and this demand is transformed into a specific constraint condition of "oil cooling flow rate ≥ 5 L / min".

[0071] Building such a knowledge graph means organizing previously scattered and unstructured engineering knowledge into a queryable, reasonable, and traceable intelligent model. It can be used in various scenarios such as intelligent equipment design, solution verification, fault diagnosis, and configuration recommendation, improving design efficiency and system reliability.

[0072] The vector database has indexed historical project reports, spindle supplier technical documents, thermal deformation analysis papers, etc.

[0073] 2. RAG-driven model generation Requirements Input and RAG Analysis. Engineers input customer requirements, and the RAG system retrieves relevant documents and knowledge graph information, generates an analysis report, and proposes specific component selection suggestions and cooling solutions.

[0074] Automated model generation. An automated script parses the RAG report, extracts action items, and creates SysML model elements and relationships using the MagicDraw API.

[0075] Changes in requirements and intelligent response. If a customer suddenly changes their requirements, needing to adjust the target speed to 16,000 rpm, the RAG system assesses the impact, generates a report, and guides a model update.

[0076] 3. Knowledge Graph-Driven Automated Validation Run the verification script regularly to check that all "Requirements" and "Blocks" are consistent and correct.

[0077] Any part of this invention not described in detail can be referred to in the prior art or in the art known to those skilled in the art. This embodiment does not limit such part and will not describe it in detail here.

[0078] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A conceptual design method for industrial machine tools based on knowledge graphs and MBSE, characterized in that, The method utilizes RAG to combine the powerful semantic understanding and generation capabilities of large language models with the accurate and reliable knowledge of knowledge graphs and vector databases, thereby achieving in-depth analysis of complex requirements, generation of high-quality model elements, and more intelligent interaction. SysML modeling tools serve as the final implementation and visualization platform for models; The MBSE-based SysML multi-view model system for the conceptual design phase of the industrial machine tool digital prototype comprises four sub-models: a requirement model, a process model, a design model, and a verification model. The requirement model uses SysML to create requirement diagrams and use case diagrams; the process model uses SysML to create equipment state diagrams and activity diagrams for the industrial machine tool; the design model uses SysML to create mechanism block definition diagrams, interface block definition diagrams, and internal block diagrams for the industrial machine tool; and the verification model uses SysML to create parameter diagrams for the industrial machine tool and performs indicator verification and traceability analysis. This system constructs a unified model framework that links the entire process—requirements, process, design, and verification—and clarifies SysML modeling specifications for each stage, including the traceability relationships between requirement diagrams and block diagrams. This ensures that the model serves as a single source of fact, providing a reliable benchmark for subsequent multidisciplinary collaboration and detailed design implementation, thus overcoming the limitations of fragmented diagrams. The process of the method includes: Phase 1. Knowledge Preparation and System Configuration: This includes collecting and preprocessing data on industrial mother machines, constructing a core knowledge graph, building a vector knowledge base, and configuring the RAG system. Phase 2. RAG-driven conceptual design and SysML model generation: including initial model generation, dynamic modification of requirements, and intelligent interaction; Phase 3. Knowledge Graph-Driven Automated Verification: This includes automated traceability verification, structural and parameter consistency verification, and the generation of final results. The knowledge graph serves as a precise knowledge base for storing structured information in the field of industrial mother machines, including core entities and semantic relationships. It is built on a graph database and supports precise querying and tracing. The vector database, serving as a semantic knowledge base, converts unstructured text into high-dimensional vectors for storage through a pre-trained embedding model, supporting efficient retrieval based on semantic similarity and compensating for the shortcomings of knowledge graphs in processing unstructured knowledge. The RAG system integrates a large language model, a knowledge graph, and a vector database. It first retrieves relevant text fragments from the vector database and entity relationships from the knowledge graph based on user needs, and then inputs the retrieval results as context into the large language model to achieve in-depth analysis of needs and generation of high-quality content, thereby avoiding the unreliability of the black-box output of the large language model. The SysML modeling tool: through automated scripts, it coordinates the RAG process, knowledge graph query, large language model call and SysML modeling tool API operation to realize the automatic creation, updating and management of SysML models, while supporting manual review and adjustment; In Phase 2, the content of generating the initial model includes: Requirements input: Engineers submit design requirements in natural language; RAG Requirements Analysis and Knowledge Retrieval: The RAG system retrieves relevant papers on spindle thermal stability and historical project reports from the vector library, and simultaneously queries entities and related parameters in the knowledge graph, including spindle and cooling system. The large language model combines this information to generate a requirements analysis report, clarifying key requirements, sub-requirement decomposition schemes, and the basis for selecting core components. Script-driven modeling: Automated scripts parse reports generated by large language models, extract structured information, then use knowledge graphs to accurately query and supplement complete data, and finally call the SysML modeling tool API to automatically create SysML model elements and establish traceability relationships between requirements and components; The content of dynamic modification of requirements and intelligent interaction includes: Change Analysis: When requirements change, the RAG system re-retrieves knowledge, generates an impact analysis report, and identifies the affected components, potential risks, and response strategies. Intelligent question answering and human review: Engineers ask questions through the system, RAG retrieves relevant knowledge and generates answers using a large language model; at the same time, domain experts review the automatically generated model and correct complex scenarios not covered by the automation tools. Model Update: Based on RAG recommendations and expert feedback, the script calls the SysML modeling tool API to synchronously update the SysML model, ensuring that the model is consistent with the changed requirements; In Phase 3, the automated traceability verification includes: scripts comparing the knowledge graph and the SysML model to check whether there is a valid fulfillment relationship for each requirement, and whether the component pointed to by the relationship has a corresponding entity in the knowledge graph, so as to avoid the requirement and component being disconnected; Structure and parameter consistency verification: Based on the constraint rules in the knowledge graph, the script automatically verifies whether the parameter values ​​and structural connections of the components in the SysML model conform to the rules; for complex verification results, RAG assists in generating comprehensive evaluation opinions to provide a basis for optimization. The final deliverables include: after successful verification, the system outputs a complete SysML digital prototype conceptual model and a comprehensive evaluation report, including model structure, requirement traceability relationships, verification results and optimization suggestions, providing a unified benchmark for subsequent detailed design and simulation verification.

2. The industrial machine tool conceptual design method based on knowledge graphs and MBSE according to claim 1, characterized in that, The core entities include components, parameters, constraints, and requirements; The semantic relationships include: electric spindle - possessing - maximum speed, cooling system - satisfying - flow constraints.

3. The industrial machine tool conceptual design method based on knowledge graphs and MBSE according to claim 1, characterized in that, The unstructured texts include historical project reports, supplier manuals, and papers.

4. The industrial machine tool conceptual design method based on knowledge graphs and MBSE according to claim 1, characterized in that, In Phase 1, the collected industrial machine data includes historical project experience of similar industrial machine machines, component technical data, analysis and verification reports, industry standards and external literature; Preprocessing includes: extracting text from the collected data, segmenting long documents into retrieval-friendly fragments, and supplementing metadata to provide a high-quality data source for vector databases and knowledge graphs; The content of constructing the core knowledge graph includes: Ontology design: Define the ontology structure in the industrial mother machine domain, and clarify the core entity types and relationship types; Knowledge extraction and fusion: Information is directly imported from structured data, and entities and relationships are extracted from unstructured text using NLP technology; entity alignment and conflict resolution are performed to ensure knowledge consistency. The construction of the vector knowledge base includes: using a pre-trained embedding model to convert pre-processed text blocks into high-dimensional vectors, storing them in a vector database to support subsequent semantic retrieval; The configuration of the RAG system includes: selecting a suitable large language model and developing a retrieval-enhancement-generation workflow; in the retrieval stage, matching vector library text with knowledge graph entities according to user needs; in the enhancement stage, integrating the retrieval results with the original requirements into a context; and in the generation stage, the large language model outputs content including a requirements analysis report and component selection suggestions.

5. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the knowledge graph and MBSE-based industrial machine conceptual design method according to any one of claims 1-4.

Images