Intelligent product service system conceptual design method based on cue word engineering

Abstract

The invention provides an intelligent product service system concept design method based on cue word engineering. The method comprises the following steps: constructing a knowledge base, including data collection, integration and knowledge graph construction; then, a cue word project is constructed, a reasoning mode is determined, and an optimization cue template is created; and then, gradually designing from a demand layer to a physical layer, enabling LLMs in each layer to play different roles for cooperation, realizing demand collection, arrangement and conversion, function determination and decomposition, behavior and logic architecture construction, physical architecture determination and related detection analysis and the like, integrating various technologies to improve the design quality and efficiency, and meeting the market demand.

Description

Technical Field

[0001] The present invention relates to the field of concept design of intelligent product service system (SPSS), and in particular to a concept design method of intelligent product service system based on prompt word engineering. Background Art

[0002] With the continuous advancement of science and technology, Smart Product-Service System (SPSS) has gradually become a research hotspot. It is a system that closely combines products and services and uses digital and intelligent technologies to provide users with more comprehensive solutions.

[0003] In the development process of intelligent PSS, existing research has deficiencies in many aspects:

[0004] A) Problems in the conceptual design stage

[0005] Insufficient consideration of the entire life cycle: Many studies fail to fully cover the entire life cycle of smart product service systems from design, production, use to disposal, resulting in deficiencies in the sustainability of the system and the realization of long-term value.

[0006] Incomplete initial system construction: Some studies did not establish an effective overall framework in the early stages of system design and lacked a comprehensive plan for system functions and structures, resulting in a lack of clear direction for subsequent design and development work.

[0007] Difficulty in handling the coupling characteristics of functional modules: Intelligent product service systems usually contain multiple complex functional modules, which are tightly coupled with each other. Existing research often lacks effective methods to handle such coupling characteristics, which leads to challenges in system integration and optimization.

[0008] Insufficient interdisciplinary integration and innovation in design methods: Intelligent product service systems involve multiple disciplines, such as computer science, mechanical engineering, electronic engineering, management, etc. However, existing research has obvious gaps in interdisciplinary integration and fails to fully integrate knowledge and methods from different disciplines, limiting the innovation of design methods and the improvement of system performance.

[0009] B) Problems with the application of prompt words in engineering

[0010] Although cue word engineering has attracted attention with the development of language models, it faces many challenges in the design and application of intelligent product service systems.

[0011] Limited ability to guide models: When applying prompt word engineering to the design of intelligent product service systems, it is often difficult to effectively guide language models to generate results that fully meet the requirements. Due to the complexity of intelligent product service systems, more precise and detailed prompt words are needed to guide the output of the model, but current methods are insufficient in this regard.

[0012] Lack of systematic application methods: Existing prompt word engineering applications lack a complete and systematic approach to the design of intelligent product service systems. Different design tasks may require different prompt word strategies, but there is currently no universal and effective method to determine and apply these prompt words.

[0013] C) Problems with generative concept design methods based on large language models

[0014] The generative concept design method based on large language model (LLM) also has some defects.

[0015] Difficulty in reasoning design requirements: These methods often have difficulty in effectively reasoning about design requirements and are unable to accurately extract the key elements required for system design from the initial information provided by the user, resulting in the generated conceptual solutions that may not meet actual requirements.

[0016] Models lack transparency and controllability: Large language models themselves have certain complexity and black-box characteristics, and their internal operating mechanisms and decision-making processes are not transparent. This makes it difficult for designers to effectively control and adjust these models when using them, and it is impossible to ensure that the generated results meet expectations.

[0017] There is a hallucination problem: large language models may have hallucinations when generating content, that is, generating content that is inconsistent with or irrelevant to the facts. This is very disadvantageous in the design of intelligent product service systems and may lead to errors and infeasibility of design solutions.

[0018] In summary, existing research has many shortcomings in the conceptual design of intelligent product service systems. Cue word engineering and generative concept design methods based on large language models also face a series of problems. A new method is needed to address these challenges. Summary of the invention

[0019] The present invention proposes a conceptual design method for intelligent product service system based on prompt word engineering. By integrating the large language model (LLM) and the extended requirement-function-behavior-logic-physical (RFBLP) model, the prompt word engineering is used to decompose the design task into five levels to guide the LLM-assisted design, thereby solving the complex characteristics and multidisciplinary barriers of the intelligent product service system. Specifically, the steps include:

[0020] S1. Knowledge base construction

[0021] This step lays the foundation for the entire design method, aiming to build a comprehensive and accurate knowledge base for the design of intelligent product service systems, so that the subsequent design process can make full use of relevant information and ensure the scientificity and rationality of the design.

[0022] S1.1 Data collection

[0023] Collect the following four types of documents:

[0024] a) Intelligent product service system definition document, clarifying the basic concepts, components and operating principles of the intelligent product service system;

[0025] b) Related information of the RFBLP (Requirement-Function-Behavior-Logic-Physical) model, including the model structure, the meaning of each layer and the conversion relationship information, to provide a theoretical basis for the design analysis based on the model;

[0026] c) Specific task analysis documents, which record in detail the specific tasks and operation processes of each link from demand analysis to physical implementation, and provide practical guidance for designers;

[0027] d) Stakeholder requirements document, which includes the system requirements of different stakeholders such as users, manufacturers, and service providers, reflecting the actual expectations of the market and users.

[0028] S1.2 Data Integration

[0029] The documents collected in S1.1 are systematically integrated to build a knowledge base. The specific steps include:

[0030] a) Text segmentation: Use CharacterTextSplitter to segment the document;

[0031] b) Vectorization: Use OpenAI’s embedding model to convert text slices into vectors;

[0032] c) Storage: Store the vectorized text information in a vector database.

[0033] S1.3 Knowledge Graph Construction

[0034] Based on the integrated data, we build a knowledge graph in the field of intelligent product service system:

[0035] a) Entity Recognition: Use named entity recognition (NER) technology to identify key concepts and entities in the text;

[0036] b) Relation extraction: Use remote supervision methods to automatically extract semantic relations between entities;

[0037] c) Knowledge fusion: Using embedding-based entity alignment methods to fuse knowledge from different sources;

[0038] d) Storage and indexing: Use Neo4j graph database to store knowledge graphs and establish multi-dimensional indexes to improve query efficiency.

[0039] S2. Construction of prompt word project

[0040] This step is the key to guiding LLM to generate results that meet design requirements. By determining the appropriate reasoning mode and creating accurate prompt templates, LLM can better understand the complex logic and detailed requirements of the design task, and improve the accuracy and reliability of the design results.

[0041] S2.1 Reasoning mode determination

[0042] The "question-answer + RAG (retrieval-augmented generation)" reasoning mode is adopted, and GPT-4 is selected as the large language model.

[0043] S2.2 Prompt template creation

[0044] a) Construct the LLM operation mechanism diagram, which includes four modules: input processing, knowledge retrieval, reasoning generation and output optimization;

[0045] b) Design task prompts according to the five levels of the RFBLP model, and design 3-5 core prompt templates for each level;

[0046] c) Normalize each sub-prompt template, including:

[0047] Role definition: clarify the role played by LLM in the current task (requirements extraction engineer, mapping engineer, designer, confirmation engineer);

[0048] Task description: Detail the goals and requirements of the current step;

[0049] Input format: specifies the structure and keywords of input information;

[0050] Output format: defines the structure of the expected output, including necessary tags and separators;

[0051] Example description: Provide 1-2 input and output examples to guide LLM to understand the task requirements.

[0052] S2.3 Prompt word optimization

[0053] a) Use prompt tuning technology to fine-tune the prompt word embedding using a small amount of labeled data;

[0054] b) Realize dynamic prompt word generation and automatically adjust the prompt word content according to the current design context;

[0055] c) Establish a performance evaluation mechanism for prompt words, including quantitative indicators in the three dimensions of relevance, consistency and innovation.

[0056] S3. Demand layer design

[0057] Through LLM, requirements extraction engineers, designers, and confirmation engineers collaborate to comprehensively and accurately collect and organize requirements, and convert them into a form that can be recognized and processed by the system, providing direction and basis for subsequent functional design.

[0058] S3.1 Requirements collection and organization

[0059] a) Stakeholders raise demands;

[0060] b) LLM plays the role of requirement extraction engineer to collect requirements and use active learning strategies to guide users to express their requirements in detail;

[0061] c) LLM plays the role of a designer to decompose and analyze requirements and use the analytic hierarchy process (AHP) to prioritize requirements;

[0062] d) LLM plays the role of confirmation engineer to detect the completeness of requirements, using rule-based requirements consistency checking algorithm;

[0063] e) Iterate requirements and use genetic algorithms to optimize the requirements combination until a complete requirements document is formed.

[0064] S3.2 Transformation of requirements into system requirements

[0065] Design targeted prompts to guide LLM to transform stakeholder needs into system requirements. Specific steps:

[0066] a) Requirements classification: Use a multi-label text classification model to classify requirements into categories such as function, performance, and constraints;

[0067] b) Requirements formalization: Using ontology-based requirements formalization methods to convert natural language requirements into structured representations;

[0068] c) Requirements verification: Use model-based verification methods to verify the consistency and completeness of formal requirements.

[0069] S4. Functional layer design

[0070] The functional layer design is determined based on the requirement layer. Through LLM, the functional mapping engineer, designer, and confirmation engineer are used to further clarify the system functional architecture, including function determination and decomposition as well as key functional information determination, providing a functional basis for the behavioral layer design.

[0071] S4.1 Function determination and decomposition

[0072] a) LLM plays the role of functional mapping engineer to conduct system use case analysis and quantify the functional scale using COSMIC function point method;

[0073] b) LLM plays the role of a designer to decompose functions into sub-functions, using a graph-based function decomposition algorithm to ensure the completeness of the decomposition;

[0074] c) LLM plays the role of confirmation engineer to detect functional completeness and accuracy and use functional dependency diagrams to analyze the relationship between functions.

[0075] S4.2 Determination of key functional information

[0076] Design special prompts and use LLM to determine the goals, scope and priority information for each key function. Introduce the following techniques:

[0077] a) Goal modeling: Use the Goal Graph to represent the hierarchical relationship between functional goals;

[0078] b) Functional scope definition: Use boundary value analysis to determine the input and output scope of the function;

[0079] c) Priority assessment: The function priority is determined by combining the analytical network process (ANP) and fuzzy comprehensive evaluation method.

[0080] S5. Behavior layer design

[0081] The behavioral layer design is based on the functional architecture determined by the functional layer. It maps the functions to the behavioral layer, constructs the behavioral architecture, describes the specific behaviors and reactions that must be completed to achieve the functions under specific conditions, and provides a behavioral basis for the logical layer design.

[0082] S5.1 Function to behavior mapping

[0083] a) LLM plays the role of mapping engineer to map functions to the behavior layer, using the ontology-based function-behavior mapping algorithm;

[0084] b) Mapping engineers check and verify the mapping results and use cross-validation methods to evaluate the mapping quality.

[0085] S5.2 Behavioral architecture construction

[0086] a) LLM plays the role of a designer to define the properties of system behavior elements and use state diagrams and activity diagrams to describe complex behaviors;

[0087] b) LLM analyzes the collaboration and impact between behaviors and uses discrete event simulation technology to evaluate the interaction effects of behaviors;

[0088] c) LLM plays the role of confirmation engineer to detect the completeness and accuracy of the behavioral architecture and use formal verification methods to check the correctness of the behavioral model.

[0089] S6.Logical layer design steps

[0090] The logical layer design determines the logical architecture according to the behavioral architecture, including function to logic, and then continues to analyze the system logic in depth, including function to logic mapping and the construction of the logical architecture, describing how the system works, defining the system's behavior and the logical relationship between behaviors, and providing a logical foundation for the physical layer design.

[0091] S6.1 Function to logic mapping

[0092] a) LLM acts as a mapping engineer to map functions to logical subsystems, using reasoning methods based on knowledge graphs to assist in mapping;

[0093] b) Mapping engineers check and verify the mapping results and use multi-criteria decision analysis (MCDA) to evaluate the mapping solutions.

[0094] S6.2 Logical Architecture Construction

[0095] a) LLM plays the role of a designer to analyze the input, output and interaction relationships of the logical subsystem, and uses data flow diagrams (DFDs) and IDEF0 diagrams to represent the system logical structure;

[0096] b) LLM plays the role of confirmation engineer to detect the completeness and accuracy of the logical architecture, using a constraint-based architecture evaluation method;

[0097] c) Build a logical architecture relationship model within the EA software and use the Model Driven Architecture (MDA) approach to ensure the consistency and traceability of the model.

[0098] S7. Physical layer design

[0099] The physical layer design determines the physical architecture based on the logical architecture, including the mapping from logical to physical and the construction of the physical architecture, to achieve the specific implementation of the system in the physical world, to ensure that the system can be effectively manufactured, produced and deployed, while considering the relationship between physical components and the overall performance of the system.

[0100] S7.1 Logical to Physical Mapping

[0101] a) LLM acts as a mapping engineer to map logical subsystems to physical components using a pattern-based mapping strategy;

[0102] b) Mapping engineers check and verify the mapping results and use sensitivity analysis methods to evaluate the stability of the mapping.

[0103] S7.2 physical architecture construction

[0104] a) LLM plays the role of designer to define physical interfaces, using the UML-based interface modeling approach;

[0105] b) LLM establishes the interaction relationship between physical entities and uses system dynamics model to analyze the dynamic interaction between components;

[0106] c) LLM analyzes the output of physical entities and uses Monte Carlo simulation method to evaluate system performance;

[0107] d) LLM acts as a confirmation engineer to detect the integrity and accuracy of the physical architecture and evaluate the design reliability using failure mode and effects analysis (FMEA);

[0108] e) Build a physical architecture relationship model within the EA software to achieve automatic conversion from logical model to physical model.

[0109] The beneficial effects of the present invention include:

[0110] 1. Solve complex characteristics and multidisciplinary barriers

[0111] By integrating the Large Language Model (LLM) and the extended RFBLP model, and using prompt word engineering to decompose the design task into five levels, it is possible to effectively deal with the complex characteristics of intelligent product service systems and issues involving multiple disciplines. This combination breaks down the barriers between disciplines, allowing knowledge from different disciplinary backgrounds to be better integrated into the design process, providing designers with a systematic and comprehensive design method.

[0112] 2. Improve the accuracy of demand collection and conversion

[0113] In the design of the demand layer, LLM plays different roles and collaborates with stakeholders to collect and organize requirements more comprehensively and accurately. The demand extraction engineer has in-depth dialogues with stakeholders to explore the requirements, the designer conducts decomposition analysis, confirms the integrity of the engineer's inspection, and forms a complete demand document through iteration. At the same time, the prompt word engineering is used to guide LLM to transform vague and abstract requirements into specific system requirements, which improves the accuracy of demand conversion and provides a reliable foundation for subsequent design.

[0114] 3. Optimize the rationality and completeness of functional design

[0115] In the functional layer design, LLM is used to assist in determining the functional path, decomposing functions, and determining key functional information, taking into account the synergy between sub-functions and various factors, such as user needs, system performance requirements, and actual application scenarios. This makes the design of the system functional architecture more reasonable and complete, and can better meet user needs and system requirements.

[0116] 4. Enhance the accuracy and logic of behavior and logic design

[0117] The behavioral layer design maps functions to the behavioral layer based on the functional architecture and builds a behavioral architecture to describe the specific behavior and response of the system. The logical layer design further determines the logical architecture based on the behavioral architecture, including the mapping of functions to logic and the construction of the logical architecture. The design of these two levels improves the accuracy and logic of the behavioral and logical design through the assistance of LLM and the corresponding verification mechanism, ensuring that the system can work as expected.

[0118] 5. Ensure the feasibility and efficiency of physical implementation

[0119] The physical layer design determines the physical architecture based on the logical architecture, including the mapping from logic to physical and the construction of the physical architecture. In this process, the interrelationships between physical components and the overall performance of the system are taken into account, making the implementation of the system in the physical world more feasible and efficient, and enabling effective manufacturing, production and deployment.

[0120] 6. Innovative reasoning mode and prompt template design

[0121] The "question-answer + RAG" reasoning mode is adopted and GPT-4 is selected as the model. The prompt word engineering is created by combining text segmentation, knowledge vectorization, and Rerank technology, which improves the adaptability of the model to the design task and the reliability of the generated content. At the same time, the specification is created for the sub-prompt template, which reduces the difficulty of LLM understanding prompts based on zero-shot learning and improves the reliability of generated content. This innovative reasoning mode and prompt template design can better guide LLM to generate results that meet the design requirements.

[0122] 7. Role prompt word template improves design efficiency and accuracy

[0123] The role prompt word templates corresponding to the five levels of RFBLP were designed. LLM completed the design of each level by playing different roles and collaborating, which improved the efficiency and accuracy of the design process. This role setting and template design made the design process more standardized and systematic, reducing human errors and uncertainties. BRIEF DESCRIPTION OF THE DRAWINGS

[0124] Figure 1 The steps of the method of the present invention are shown. DETAILED DESCRIPTION

[0125] The present invention provides a conceptual design method for an intelligent product service system based on prompt word engineering, which can automatically and intelligently handle the entire process from demand acquisition, function decomposition, behavior analysis, logical architecture design to physical implementation.

[0126] Combination Figure 1 , the method specifically comprises the following steps:

[0127] A conceptual design method for an intelligent product service system based on prompt word engineering includes the following steps:

[0128] S1. Knowledge base construction

[0129] This step lays the foundation for the entire design method, aiming to build a comprehensive and accurate knowledge base for the design of intelligent product service systems, so that the subsequent design process can make full use of relevant information and ensure the scientificity and rationality of the design.

[0130] S1.1 Data collection

[0131] Collect the following four types of documents:

[0132] a) Intelligent product service system definition document, clarifying the basic concepts, components and operating principles of the intelligent product service system;

[0133] b) Related information of the RFBLP (Requirement-Function-Behavior-Logic-Physical) model, including the model structure, the meaning of each layer and the conversion relationship information, to provide a theoretical basis for the design analysis based on the model;

[0134] c) Specific task analysis documents, which record in detail the specific tasks and operation processes of each link from demand analysis to physical implementation, and provide practical guidance for designers;

[0135] d) Stakeholder requirements document, which includes the system requirements of different stakeholders such as users, manufacturers, and service providers, reflecting the actual expectations of the market and users.

[0136] S1.2 Data Integration

[0137] The documents collected in S1.1 are systematically integrated to build a knowledge base. The specific steps include:

[0138] a) Text segmentation: Use CharacterTextSplitter to segment the document;

[0139] b) Vectorization: Use OpenAI’s embedding model to convert text slices into vectors;

[0140] c) Storage: Store the vectorized text information in a vector database.

[0141] S1.3 Knowledge Graph Construction

[0142] Based on the integrated data, we build a knowledge graph in the field of intelligent product service system:

[0143] a) Entity Recognition: Use named entity recognition (NER) technology to identify key concepts and entities in the text;

[0144] b) Relation extraction: Use remote supervision methods to automatically extract semantic relations between entities;

[0145] c) Knowledge fusion: Using embedding-based entity alignment methods to fuse knowledge from different sources;

[0146] d) Storage and indexing: Use Neo4j graph database to store knowledge graphs and establish multi-dimensional indexes to improve query efficiency.

[0147] S2. Construction of prompt word project

[0148] This step is the key to guiding LLM to generate results that meet design requirements. By determining the appropriate reasoning mode and creating accurate prompt templates, LLM can better understand the complex logic and detailed requirements of the design task, and improve the accuracy and reliability of the design results.

[0149] S2.1 Reasoning mode determination

[0150] The "question-answer + RAG (retrieval-augmented generation)" reasoning mode is adopted, and GPT-4 is selected as the large language model.

[0151] S2.2 Prompt template creation

[0152] a) Construct the LLM operation mechanism diagram, which includes four modules: input processing, knowledge retrieval, reasoning generation and output optimization;

[0153] b) Design task prompts according to the five levels of the RFBLP model, and design 3-5 core prompt templates for each level;

[0154] c) Normalize each sub-prompt template, including:

[0155] Role definition: clarify the role played by LLM in the current task (requirements extraction engineer, mapping engineer, designer, confirmation engineer);

[0156] Task description: Detail the goals and requirements of the current step;

[0157] Input format: specifies the structure and keywords of input information;

[0158] Output format: defines the structure of the expected output, including necessary tags and separators;

[0159] Example description: Provide 1-2 input and output examples to guide LLM to understand the task requirements.

[0160] S2.3 Prompt word optimization

[0161] a) Use prompt tuning technology to fine-tune the prompt word embedding using a small amount of labeled data;

[0162] b) Realize dynamic prompt word generation and automatically adjust the prompt word content according to the current design context;

[0163] c) Establish a performance evaluation mechanism for prompt words, including quantitative indicators in the three dimensions of relevance, consistency and innovation.

[0164] S3. Demand layer design

[0165] Through LLM, requirements extraction engineers, designers, and confirmation engineers collaborate to comprehensively and accurately collect and organize requirements, and convert them into a form that can be recognized and processed by the system, providing direction and basis for subsequent functional design.

[0166] S3.1 Requirements collection and organization

[0167] a) Stakeholders raise demands;

[0168] b) LLM plays the role of requirement extraction engineer to collect requirements and use active learning strategies to guide users to express their requirements in detail;

[0169] c) LLM plays the role of a designer to decompose and analyze requirements and use the analytic hierarchy process (AHP) to prioritize requirements;

[0170] d) LLM plays the role of confirmation engineer to detect the completeness of requirements, using rule-based requirements consistency checking algorithm;

[0171] e) Iterate requirements and use genetic algorithms to optimize the requirements combination until a complete requirements document is formed.

[0172] S3.2 Transformation of requirements into system requirements

[0173] Design targeted prompts to guide LLM to transform stakeholder needs into system requirements. Specific steps:

[0174] a) Requirements classification: Use a multi-label text classification model to classify requirements into categories such as function, performance, and constraints;

[0175] b) Requirements formalization: Using ontology-based requirements formalization methods to convert natural language requirements into structured representations;

[0176] c) Requirements verification: Use model-based verification methods to verify the consistency and completeness of formal requirements.

[0177] S4. Functional layer design

[0178] The functional layer design is determined based on the requirement layer. Through LLM, the functional mapping engineer, designer, and confirmation engineer are used to further clarify the system functional architecture, including function determination and decomposition as well as key functional information determination, providing a functional basis for the behavioral layer design.

[0179] S4.1 Function determination and decomposition

[0180] a) LLM plays the role of functional mapping engineer to conduct system use case analysis and quantify the functional scale using COSMIC function point method;

[0181] b) LLM plays the role of a designer to decompose functions into sub-functions, using a graph-based function decomposition algorithm to ensure the completeness of the decomposition;

[0182] c) LLM plays the role of confirmation engineer to detect functional completeness and accuracy and use functional dependency diagrams to analyze the relationship between functions.

[0183] S4.2 Determination of key functional information

[0184] Design special prompts and use LLM to determine the goals, scope and priority information for each key function. Introduce the following techniques:

[0185] a) Goal modeling: Use the Goal Graph to represent the hierarchical relationship between functional goals;

[0186] b) Functional scope definition: Use boundary value analysis to determine the input and output scope of the function;

[0187] c) Priority assessment: The function priority is determined by combining the analytical network process (ANP) and fuzzy comprehensive evaluation method.

[0188] S5. Behavior layer design

[0189] The behavioral layer design is based on the functional architecture determined by the functional layer. It maps the functions to the behavioral layer, constructs the behavioral architecture, describes the specific behaviors and reactions that must be completed to achieve the functions under specific conditions, and provides a behavioral basis for the logical layer design.

[0190] S5.1 Function to behavior mapping

[0191] a) LLM plays the role of mapping engineer to map functions to the behavior layer, using the ontology-based function-behavior mapping algorithm;

[0192] b) Mapping engineers check and verify the mapping results and use cross-validation methods to evaluate the mapping quality.

[0193] S5.2 Behavioral architecture construction

[0194] a) LLM plays the role of a designer to define the properties of system behavior elements and use state diagrams and activity diagrams to describe complex behaviors;

[0195] b) LLM analyzes the collaboration and impact between behaviors and uses discrete event simulation technology to evaluate the interaction effects of behaviors;

[0196] c) LLM plays the role of confirmation engineer to detect the completeness and accuracy of the behavioral architecture and use formal verification methods to check the correctness of the behavioral model.

[0197] S6.Logical layer design steps

[0198] The logical layer design determines the logical architecture according to the behavioral architecture, including function to logic, and then continues to analyze the system logic in depth, including function to logic mapping and the construction of the logical architecture, describing how the system works, defining the system's behavior and the logical relationship between behaviors, and providing a logical foundation for the physical layer design.

[0199] S6.1 Function to logic mapping

[0200] a) LLM acts as a mapping engineer to map functions to logical subsystems, using reasoning methods based on knowledge graphs to assist in mapping;

[0201] b) Mapping engineers check and verify the mapping results and use multi-criteria decision analysis (MCDA) to evaluate the mapping solutions.

[0202] S6.2 Logical Architecture Construction

[0203] a) LLM plays the role of a designer to analyze the input, output and interaction relationships of the logical subsystem, and uses data flow diagrams (DFDs) and IDEF0 diagrams to represent the system logical structure;

[0204] b) LLM plays the role of confirmation engineer to detect the completeness and accuracy of the logical architecture, using a constraint-based architecture evaluation method;

[0205] c) Build a logical architecture relationship model within the EA software and use the Model Driven Architecture (MDA) approach to ensure the consistency and traceability of the model.

[0206] S7. Physical layer design

[0207] The physical layer design determines the physical architecture based on the logical architecture, including the mapping from logical to physical and the construction of the physical architecture, to achieve the specific implementation of the system in the physical world, to ensure that the system can be effectively manufactured, produced and deployed, while considering the relationship between physical components and the overall performance of the system.

[0208] S7.1 Logical to Physical Mapping

[0209] a) LLM acts as a mapping engineer to map logical subsystems to physical components using a pattern-based mapping strategy;

[0210] b) Mapping engineers check and verify the mapping results and use sensitivity analysis methods to evaluate the stability of the mapping.

[0211] S7.2 Physical Architecture Construction

[0212] a) LLM plays the role of designer to define physical interfaces, using the UML-based interface modeling approach;

[0213] b) LLM establishes the interaction relationship between physical entities and uses system dynamics model to analyze the dynamic interaction between components;

[0214] c) LLM analyzes the output of physical entities and uses Monte Carlo simulation method to evaluate system performance;

[0215] d) LLM acts as a confirmation engineer to detect the integrity and accuracy of the physical architecture and evaluate the design reliability using Failure Mode and Effects Analysis (FMEA);

[0216] e) Build a physical architecture relationship model within the EA software to achieve automatic conversion from logical model to physical model.

[0217] The conceptual design method of the intelligent product service system based on prompt word engineering provided by the present invention realizes the full process automation and intelligence of the intelligent product service system design through knowledge base construction, prompt word engineering construction and gradual design from the demand layer to the physical layer, improves the design quality and efficiency of the intelligent product service system, and meets the market and users' growing demand for the intelligent product service system.

[0218] In terms of the scope of protection of rights, the present invention is not limited to the specific embodiments described herein. It should be understood that the scope of protection of the present invention covers all reasonable changes, modifications and equivalent implementations according to the content of this document. The scope of rights of this application is intended to cover every novel point and creative step of the present invention and its potential application methods to ensure that the originality, practicality and technical foresight of the present invention are fully protected by law. We aim to protect our innovative achievements from unauthorized use and ensure the integrity and respect of intellectual property rights through these rights applications.

Claims

1. A conceptual design method for an intelligent product service system based on prompt word engineering, characterized in that: The following steps are involved: S1. Knowledge base construction steps, including data collection, data integration and knowledge graph construction; S2. The steps of constructing the prompt word project include determining the reasoning mode, creating the prompt template and optimizing the prompt word; S3. Requirements layer design steps, including requirements collection and organization, and transformation of requirements into system requirements; S4. Functional layer design steps, including function determination and decomposition, and key function information determination; S5. Behavior layer design steps, including function-to-behavior mapping and behavior architecture construction; S6.Logical layer design steps, including function-to-logic mapping and logical architecture construction; S7. Physical layer design steps, including logical to physical mapping and physical architecture construction; Among them, steps S3 to S7 utilize a large language model (LLM) to play different roles for design and verification.

2. The method according to claim 1, characterized in that The data collection in step S1 includes collecting the intelligent product service system definition document, RFBLP (Requirement-Function-Behavior-Logic-Physical) model related information, specific task analysis documents and stakeholder requirement documents.

3. The method according to claim 1, characterized in that The creation of the prompt template in step S2 includes: constructing an LLM operation mechanism diagram; decomposing and designing task prompts according to the five levels of the RFBLP model; and standardizing each sub-prompt template, including role definition, task description, input format, output format, and example description.

4. The method according to claim 1, characterized in that: In the above steps S3, S4 and S5, LLM plays multiple roles in design and verification, including: a) In the demand layer design: LLM plays the role of requirement extraction engineer to collect requirements; LLM plays the role of a designer to break down and analyze requirements; LLM plays the role of confirmation engineer to check the completeness of requirements; Iterate requirements until a complete requirements document is formed; b) In functional layer design: LLM plays the role of functional mapping engineer to conduct system use case analysis; LLM plays the role of designer to decompose functions into sub-functions; LLM plays the role of confirmation engineer to detect functional completeness and accuracy; c) In the behavioral layer design: LLM plays the role of designer to define the properties of system behavior elements; LLM analyzes collaboration and influence among behaviors; LLM acts as a validation engineer to detect the completeness and accuracy of the behavioral architecture.

Images