Intelligent robot prototype aided design method based on knowledge graph

Abstract

The invention discloses an intelligent robot prototype aided design method based on a knowledge graph, and the method comprises the steps: 1, data acquisition: data sources are structured data and unstructured data stored in a third party and an enterprise; step 2, knowledge extraction: inputting the data obtained in the step 1 into a knowledge extraction layer, performing knowledge extraction through a vertical field large language model, extracting structured information and knowledge from a text, including entity extraction, relationship extraction and attribute extraction, forming a relationship triple, and extracting the structured information and knowledge from the text; carrying out complex relation modeling among mechanical design knowledge graph entities in combination with a graph neural network so as to improve the accuracy of entity recognition and relation extraction; 3, performing knowledge fusion processing; and 4, data storage. According to the method, complex relation modeling between entities is carried out in combination with GNN, so that the accuracy of entity recognition and relation extraction is improved, more intelligent decision support functions such as an inference engine based on a knowledge graph are added, and the intelligent level of design is further improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing, and in particular to an intelligent robot prototype auxiliary design system. Background Art

[0002] With the development of new generation information technology, manufacturing enterprises are accelerating the technological upgrading and replacement of production technology from mechanization, automation, informatization to intelligence, and industrial manufacturing is developing towards digital manufacturing, network manufacturing, and intelligent manufacturing. Mechanical engineering is the technical pillar of the manufacturing industry and a comprehensive reflection of the country's scientific and technological level. Mechanical design is a complex engineering process involving multiple disciplines and their interrelationships. Designers need to comprehensively consider many factors such as the working principle, structure, movement mode, force and energy transmission mode, material and size of parts of the machine. However, these professional knowledge and experience are usually scattered in different books, documents, patents and practices, which are difficult to directly obtain and integrate.

[0003] With the development of basic theories and new disciplines in mechanical engineering, as well as the widespread application of computer technology, mechanical design is gradually shifting from reliance on subjective judgment and experience to reliance on scientific calculations and optimized design.

[0004] In recent years, knowledge graph technology has developed rapidly. The technology has been successfully applied in the medical, financial, legal, fault diagnosis and other industries, but there are few application cases in the field of prototype design. With the continuous development of prototype design, a large amount of design schemes, drawings, processes, documents and other data are generated in the process of prototype development and manufacturing, and high performance, high reliability and high efficiency are required for prototype design. The automation level, work efficiency and manufacturing quality of prototype development and manufacturing systems are constantly improving. In this transformation, efficient acquisition, integration and application of relevant information have become the key to improving the efficiency and quality of mechanical design. In the process of prototype development, a large amount of standard specifications, design schemes, drawings, processes and other information are involved, and this information is usually scattered in various documents, databases and internal knowledge bases. At the same time, the types of data are complex, including structured data (fixed format such as BOM table), semi-structured data (with relevant tags such as design documents) and unstructured data (no pre-organization method such as design drawings), lacking a unified data storage form and organization and management method. This leads to the inability to effectively store knowledge, and the objectively existing valuable information between knowledge cannot be deeply mined and utilized.

[0005] Therefore, knowledge graphs, as a technology that can quickly describe concepts and their relationships in the physical world, provide new solutions for the field of prototype design. By structuring and abstracting the knowledge in the field of mechanical design, knowledge graphs can integrate existing information, data, and link relationships on the Internet into knowledge, making information resources easier to calculate, understand, and evaluate. At the same time, knowledge graphs can also achieve rapid response and reasoning of knowledge, providing designers with comprehensive and accurate reference information and optimization suggestions, thereby improving design efficiency and quality.

[0006] The Chinese patent application with publication number CN116933528A discloses an intake duct auxiliary design method and system based on case knowledge graph. It uses Neo4j to build a knowledge base, builds a knowledge graph through analysis and modeling of intake duct data, and integrates the use of graph database, which can make the management of intake duct data more complete and more convenient. This system is mainly aimed at the graph database built for intake duct design cases of inhaled aircraft, and has certain limitations for the design and development of prototypes such as robots and mechanical arms.

[0007] The Chinese patent application with publication number CN116340530A discloses an intelligent design method based on mechanical knowledge graph, which builds entity recognition and relationship extraction models in professional fields, combines the front-end framework to build an auxiliary design platform, and provides multiple functions such as entity recognition, sentence question and answer, and entity query. This system and method is based on a complete knowledge graph that has been constructed. As the design work progresses and the number of design cases increases, it is difficult to optimize and update the graph data. Summary of the invention

[0008] The technical problem to be solved by the present invention is how to further improve the intelligent level of design.

[0009] The present invention solves the above technical problems through the following technical means: an intelligent robot prototype auxiliary design method based on knowledge graph, comprising the following steps:

[0010] Step 1: Data acquisition, the data source is structured data and unstructured data stored by third parties and within the enterprise;

[0011] Step 2: Knowledge extraction: Input the data obtained in step 1 into the knowledge extraction layer, and extract knowledge through the vertical domain large language model to extract structured information and knowledge from the text, including entity extraction, relationship extraction, and attribute extraction to form relationship triples. Combined with the graph neural network, the complex relationship modeling between entities in the mechanical design knowledge graph is performed to improve the accuracy of entity recognition and relationship extraction.

[0012] Step 3: Knowledge fusion processing: input the relationship triples formed in step 2 into the knowledge fusion processing layer to perform entity relationship fusion and entity attribute fusion;

[0013] Step 4: Data storage: Use database and triple data to persistently store the knowledge base, and use document database and relational database to store documents and operation records.

[0014] As a technical solution for further optimization, the data mainly comes from design data of operating manipulators and mobile robots, including design plans, CAD drawings, BOM tables, three-dimensional models, product manuals, process documents, and robot design standards and specifications.

[0015] As a further optimized technical solution, the step 2 specifically includes:

[0016] Step 2.1 Entity Extraction: Automatically identify and extract entities with specific meanings from text data. Entity extraction includes two main tasks: entity boundary identification and entity category determination;

[0017] Step 2.2 Relationship extraction: Identify and extract semantic relationships between entities from the text. Relationship extraction methods include rule-based methods, template-based methods, and machine learning-based methods.

[0018] Step 2.3 Attribute extraction: Extract and identify the attribute values ​​of entities from the text. Attribute extraction methods include rule-based methods, template-based methods, and machine learning-based methods.

[0019] As a further optimized technical solution in the field of robotic arm and robot prototype design:

[0020] The entities include component names, materials, dimensions, and process parameters. Entity boundary identification involves identifying the start and end positions of specific components or parameters mentioned in the text, and entity category determination involves classifying these entities into specific types.

[0021] The relationship includes the assembly relationship between parts, size and tolerance requirements, and material selection basis;

[0022] The properties include the size, weight, color, and material properties of the component.

[0023] As a further optimized technical solution, the complex relationship modeling between entities in the mechanical design knowledge graph is carried out in combination with graph neural networks, including:

[0024] In GNN, the graph G consists of the following elements: node set V: all nodes on the way, edge set E: all edges in the graph, representing the relationship between nodes, node feature x: feature vector x of each nodei , the feature update of each node is determined by the information of its neighboring nodes. Assume that the feature vector of node i is x i , the feature vector of its neighbor node is x j (j∈N(i)), the new feature of the node is calculated by the following formula:

[0025]

[0026] in: is the feature representation of node i at layer l; AGGREGATE() is an aggregation function used to aggregate the features of neighboring nodes; W (l) is the weight matrix of the lth layer, and the updated eigenvector Processed through a non-linear activation function;

[0027] In GAT, the attention mechanism is used to dynamically adjust the influence of neighbor nodes. The attention coefficient of each node to its neighbors is calculated by the following formula:

[0028]

[0029] Where: a is the weight vector of the attention mechanism, || represents the vector concatenation operation, ReLU is the activation function, and the attention coefficient α ij The attention weight of node i to neighbor node j is calculated by the softmax function:

[0030]

[0031] The new feature of node i is updated as:

[0032]

[0033] As a further optimized technical solution, step three specifically includes:

[0034] Step 3.1 Data integration: Collect, clean, transform and load data from different sources;

[0035] Step 3.2 Entity alignment: Match and correspond entities in knowledge graphs from different sources. The main steps of entity alignment include: determining the entity type to be aligned, constructing an entity similarity measurement function, finding matching entities, and evaluating the alignment results.

[0036] Step 3.3 Knowledge Reasoning: Knowledge reasoning involves rule-based design optimization, case-based reasoning, and model-based reasoning;

[0037] Step 3.4 Ontology construction: Ontology construction is the process of establishing a domain ontology, which is a structured description of concepts, attributes, and relationship knowledge in a specific domain;

[0038] Step 3.5 Quality assessment: It is the process of quality inspection and assessment of knowledge fusion results.

[0039] As a further optimized technical solution, the intelligent robot prototype auxiliary design method based on knowledge graph also includes:

[0040] Step 5: Knowledge recommendation. An intelligent recommendation algorithm based on collaborative filtering is used to recommend relevant design knowledge, components, materials or design processes by analyzing the designer's historical design behavior, preferences, choices and similar needs between designers. The collaborative filtering algorithm includes: a user-based collaborative filtering algorithm finds other designers similar to the target designer by analyzing the similarities between designers, and then recommends the design knowledge, components or materials commonly used by these similar designers to the target designer; a project-based collaborative filtering algorithm finds design elements similar to the target designer's current selection by analyzing the similarities between different design knowledge, components or materials, and then recommends these similar elements to the target designer.

[0041] As a further optimized technical solution, step five specifically includes:

[0042] Step 5.1 Data preparation: Collect user behavior data based on the user's operation records on the knowledge graph in the system; construct a user-item matrix to represent the user's interactive behavior on different design resources;

[0043] Step 5.2: Data preprocessing: After collecting relevant user data, remove outliers and noise data, and then standardize the relevant data;

[0044] Step 5.3 Similarity calculation: Use the user-based collaborative filtering algorithm and the item-based collaborative filtering algorithm to calculate the similarity between users or items. The cosine similarity calculation formula is as follows:

[0045]

[0046] Among them, r u,i and r v,i Represents the ratings of user u and user v on item i, I uv represents the set of items rated by both user u and user v;

[0047] Step 5.4 Recommendation generation: Based on the similarity of users or items, the recommendation score is calculated. The recommendation score of a user or item is expressed as:

[0048]

[0049] Among them, N xRepresents a similar set of users or projects. Based on the recommendation score, a recommendation list is generated to recommend the highest-rated design resources to the user.

[0050] As a further optimized technical solution, the intelligent robot prototype assisted design method based on knowledge graph also includes: Step 6: Knowledge update. For various files generated during the development of new prototypes, users upload these structured or unstructured process files on the web page, and the system automatically extracts and integrates relevant knowledge. The acquired structured knowledge is reviewed and verified by experts and stored in the knowledge base through the interface, realizing the ability of the knowledge base to be continuously updated during use.

[0051] As a further optimized technical solution, the intelligent robot prototype auxiliary design method based on knowledge graph also includes step seven: intelligent question answering. The specific steps are as follows:

[0052] Step 7.1 The user asks questions or requests to the system through natural language;

[0053] Step 7.2 The NLU module receives the user's natural language input and performs pre-processing of the data, including noise removal, word segmentation, part-of-speech tagging, and named entity recognition, to convert the user's original question into structured data that the system can parse;

[0054] Step 7.3 The system identifies the type of user question and determines the user's specific intention;

[0055] Step 7.4 Based on the parsing results, the system constructs the corresponding query statement and executes the query in the knowledge graph to obtain relevant information;

[0056] Step 7.5 The large model performs semantic understanding and reasoning based on the user's question and the knowledge information in the existing knowledge graph to find out the knowledge points and information related to the question;

[0057] After generating the preliminary answer in step 7.6, the system optimizes the answer.

[0058] The advantages of the present invention are: the present invention combines graph neural network (GNN) to model complex relationships between entities to improve the accuracy of entity recognition and relationship extraction, and adds more intelligent decision support functions, such as reasoning engine based on knowledge graph, automatic design suggestion generation, design optimization algorithm, etc., to further improve the intelligent level of design, specifically including:

[0059] 1. By integrating information from multiple data sources, a systematic and comprehensive knowledge graph of robot and robotic arm prototypes is constructed, covering a multi-level knowledge system from component to system level.

[0060] 2. Display the knowledge graph through a graphical interface, so that users can intuitively understand the structure and function of robots and robotic arms; at the same time, provide a variety of query methods, so that users can easily obtain the required information.

[0061] 3. Use graph algorithms and reasoning engines to realize the reasoning function of knowledge graphs, provide decision support for the design, development, application and maintenance of robots and robotic arms, and greatly improve the work efficiency of designers.

[0062] 4. This system gives users the ability to update the graph database independently, allowing new prototype development case data to be added to the graph database, thereby achieving continuous optimization of the graph database. If the node information changes, such as changes in the design model or features, and adjustments to the design technical parameters, the relevant knowledge must first be reviewed. After the review is passed, the corresponding nodes in the knowledge graph need to be updated and the associations re-established. The present invention can provide designers with auxiliary design support based on the data in the knowledge base and recommend reliable design solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of the intelligent robot prototype auxiliary design system based on knowledge graph of the present invention;

[0064] Figure 2 is a flow chart of knowledge recommendation in the present invention;

[0065] Figure 3 It is a flowchart of user autonomous knowledge update in the present invention;

[0066] Figure 4 It is the intelligent reasoning flow chart of the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] like Figure 1 As shown, the intelligent robot prototype auxiliary design method based on knowledge graph of the present invention extracts and saves a large amount of structured data and unstructured data stored by third parties and within the enterprise, and specifically divides the internal knowledge base, entities and entity relationships in the field, and uses the vertical field large language model method to build a specific extraction model, and finally stores the extraction results in the Neo4j graph database for later use.

[0069] Step 1: Data Acquisition

[0070] Upload prototype-related materials through the front-end page. The data source for uploading prototype-related materials is a large amount of structured data and unstructured data stored by third parties and within the enterprise, mainly from design materials such as operating manipulators and mobile robots, including design plans, CAD drawings, BOM tables, 3D models, product manuals, process documents, robot design standards and specifications, etc.

[0071] Step 2: Knowledge Extraction

[0072] The data obtained in step one is input into the knowledge extraction layer, and knowledge is extracted through the vertical domain large language model to extract structured information and knowledge from the text, including entity extraction, relationship extraction and attribute extraction, to form relationship triples.

[0073] Step 2.1 Entity Extraction: Automatically identify and extract entities with specific meanings from text data. In the field of robotic arm and robot prototype design, these entities include component names, materials, dimensions, process parameters, etc. The purpose of entity extraction is to convert unstructured data in the text into structured data, providing a basis for subsequent relationship extraction and attribute extraction. Entity extraction includes two main tasks: entity boundary recognition and entity category determination. In robotic arm and robot prototype design, entity boundary recognition may involve identifying the start and end positions of specific components or parameters mentioned in the text; while entity category determination is to classify these entities into specific types, such as "motor", "bearing", "aluminum alloy", etc.

[0074] Step 2.2 Relationship extraction: Identify and extract semantic relationships between entities from the text. In the design of robotic arms and robot prototypes, these relationships include assembly relationships between parts, size and tolerance requirements, material selection basis, etc. The goal of relationship extraction is to convert unstructured information in the text into structured relational data to enrich the connections between entities in the knowledge graph. Relationship extraction methods generally include rule-based methods, template-based methods, and machine learning-based methods. In the field of robotic arm and robot prototype design, it may be necessary to combine domain expertise and text mining technology to build a relationship extraction model suitable for this field.

[0075] Step 2.3 Attribute extraction: Extract and identify the attribute values ​​of entities from the text. In the design of robotic arms and robot prototypes, these attributes include the size, weight, color, material properties, etc. of parts. The purpose of attribute extraction is to convert unstructured data in the text into structured attribute data to improve the attribute information of entities in the knowledge graph. Attribute extraction methods include rule-based methods, template-based methods, and machine learning-based methods. In the field of robotic arm and robot prototype design, it may be necessary to combine domain expertise and text mining technology to build an attribute extraction model suitable for this field.

[0076] Combined with graph neural network (GNN), complex relationships between entities in mechanical design knowledge graph are modeled. In GNN, graph G consists of the following elements: node set V: all nodes on the way (for example, various components in mechanical design), edge set E: all edges in the graph, representing the relationship between nodes (for example, connection, dependency), node feature x: feature vector x of each node i The key to GNN is how to update the feature representation of nodes. The feature update of each node is determined by the information of its neighboring nodes. Assume that the feature vector of node i is x i , the feature vector of its neighbor node is x j (j∈N(i)), the new feature of the node can be calculated by the following formula:

[0077]

[0078] in: is the feature representation of node i at layer l; AGGREGATE() is an aggregation function used to aggregate the features of neighboring nodes; W (l) is the weight matrix of the lth layer. The updated feature vector It can be processed by a nonlinear activation function. Common activation functions include ReLU():

[0079]

[0080] Where: b (l) is the bias term of the lth layer.

[0081] In GAT (Graph Attention Network), the attention mechanism is used to dynamically adjust the influence of neighboring nodes. The attention coefficient of each node to its neighbors is calculated by the following formula:

[0082]

[0083] Where: a is the weight vector of the attention mechanism, || represents the vector concatenation operation, and ReLU is the activation function.

[0084] Attention coefficient α ij The attention weight of node i to neighbor node j is calculated by the softmax function:

[0085]

[0086] The new feature of node i is updated as:

[0087]

[0088] This will improve the accuracy of entity recognition and relationship extraction.

[0089] Step 3: Knowledge Fusion Processing

[0090] The relation triples formed in step 2 are input into the knowledge fusion processing layer for entity relation fusion and entity attribute fusion. Knowledge fusion is the organic integration of information and data extracted by the knowledge extraction module to build a more comprehensive, accurate and valuable knowledge system. This module needs to be able to process various triple information from different data sources, analyze, integrate, deduplicate and disambiguate it with the existing data in the graph database, and finally obtain triples that can be stored in the database.

[0091] Step 3.1 Data integration: Data integration is the process of collecting, cleaning, converting and loading data from different sources to form a unified data view. In the robotic arm and robot prototype auxiliary design system, it involves integrating data on mechanical design, kinematics, dynamics, sensor technology, control algorithms, etc.

[0092] Step 3.2 Entity alignment: Entity alignment is the process of matching and corresponding entities in knowledge graphs from different sources. In the robotic arm and robot prototype auxiliary design system, entities include various parts of the robotic arm, sensors, controllers, etc. The main steps of entity alignment include: determining the entity type to be aligned, constructing an entity similarity measurement function, finding matching entities, and evaluating the alignment results.

[0093] Step 3.3 Knowledge reasoning: Knowledge reasoning involves rule-based design optimization, case-based reasoning, model-based reasoning, etc. The goal of knowledge reasoning is to help designers quickly find design inspiration, verify the feasibility of design solutions, optimize design solutions, etc.

[0094] Step 3.4 Ontology construction: Ontology construction is the process of establishing a domain ontology, which is a structured description of concepts, attributes, relationships, etc. in a specific domain. In the robotic arm and robot prototype aided design system, the ontology includes the components, functions, attributes, relationships, etc. of the robotic arm.

[0095] Step 3.5 Quality Assessment: Quality assessment is the process of checking and evaluating the quality of knowledge fusion results, specifically evaluating the accuracy, completeness, consistency, and usability of the fused knowledge graph.

[0096] Step 4: Data storage

[0097] Neo4j graph database and triple data are used for persistent storage of the knowledge base, while document and operation records are stored in document database and relational database.

[0098] Step 5: Knowledge Recommendation

[0099] Adopting an intelligent recommendation algorithm based on collaborative filtering, by analyzing the designer's historical design behavior, preferences, choices and similar needs between designers, to recommend relevant design knowledge, components, materials or design processes. Collaborative filtering algorithms include: user-based collaborative filtering algorithms find other designers similar to the target designer by analyzing the similarities between designers, and then recommend the design knowledge, components or materials commonly used by these similar designers to the target designer; project-based collaborative filtering algorithms find design elements similar to the target designer's current choice by analyzing the similarities between different design knowledge, components or materials, and then recommend these similar elements to the target designer. The specific steps are as follows: Figure 2 ,include:

[0100] Step 5.1 Data preparation

[0101] Based on the user's operation records on the knowledge graph in the system, user behavior data is collected, including: Interaction data: collects the user's interaction behavior data in the system, such as the design documents viewed, the design specifications used, the material properties referenced, the historical cases consulted, etc.; Rating data: users can rate design resources and collect these rating data; Implicit feedback data: such as the number of clicks, browsing time, number of downloads, etc.

[0102] A user-item matrix is ​​constructed to represent the user's interaction behavior with different design resources (items). For example, rows represent users, columns represent design resources, and values ​​represent the user's rating or number of interactions with the resource.

[0103] Step 5.2 Data preprocessing

[0104] After collecting relevant user data, it is necessary to remove outliers and noise data, such as incorrect ratings, repeated interaction records, etc. Then the relevant data is standardized.

[0105] Step 5.3 Similarity calculation

[0106] The similarity between users or items is calculated using the user-based collaborative filtering algorithm and the item-based collaborative filtering algorithm. The cosine similarity calculation formula is as follows:

[0107]

[0108] Among them, r u,i and r v,i Represents the ratings of user u and user v on item i, I uv represents the set of items rated by both user u and user v.

[0109] Step 5.4 Recommendation Generation

[0110] Based on the similarity of users or items, the recommendation score is calculated. The recommendation score of a user or item can be expressed as:

[0111]

[0112] Among them, N x Represents a similar set of users or projects. Based on the recommendation score, a recommendation list is generated to recommend the highest-rated design resources to the user.

[0113] Step 6: Knowledge Update

[0114] The specific process is as follows Figure 3 As shown in the figure. For various files generated during the development of new prototypes, users can upload these structured or unstructured process files on the web page. The system will automatically extract and integrate relevant knowledge. After being reviewed and verified by experts, the acquired structured knowledge is stored in the knowledge base through the Neo4j interface, thereby realizing the ability of the knowledge base to be continuously updated during use.

[0115] Step 7: Intelligent Question and Answer

[0116] This patent builds an intelligent question-answering system that understands the natural language input by the user, then combines it with the knowledge graph for reasoning, and then uses the text generation capability of the large model to give the final answer. The specific steps are as follows: Figure 4 shown.

[0117] Step 7.1 The user asks questions or requests to the system through natural language. These inputs may include information about the robotic arm, robot prototype design, functions, parameters, etc.

[0118] Step 7.2 The NLU module receives the user's natural language input and performs pre-processing of the data. This includes steps such as noise removal, word segmentation, part-of-speech tagging, and named entity recognition, converting the user's original question into structured data that the system can parse.

[0119] Step 7.3 The system identifies the type of user question (such as technical parameter query, fault diagnosis, design suggestion, etc.) and determines the user's specific intention.

[0120] Step 7.4 Based on the parsing results, the system constructs the corresponding query statement and executes the query in the knowledge graph to obtain relevant information.

[0121] Step 7.5 The big model will perform semantic understanding and reasoning based on the user's question and the knowledge information in the existing knowledge graph to find the knowledge points and information related to the question. During the query and reasoning process, the big model will combine the information such as entities, relationships and attributes in the knowledge graph to perform complex logical reasoning and calculation to obtain a more accurate answer or solution.

[0122] After generating preliminary answers in step 7.6, the system optimizes the answers to ensure their accuracy and readability, and performs natural language polishing if necessary.

[0123] The present invention provides an intelligent robot prototype auxiliary design method based on knowledge graph, which analyzes and models various data generated in robot design, including design drawings, design documents, technical requirements, BOM tables, and various standards and specifications, through data pre-processing processes such as knowledge extraction, knowledge fusion, and knowledge representation, to form structured ontology knowledge, thereby using Neo4j to build a knowledge graph and realize the visualization of graph data. The method of the present invention gives users the ability to update the graph database autonomously, allowing new design case data to be added to the graph database. If the node information changes, such as changes in the design model or features, and adjustments to the design technical parameters, the relevant knowledge must first be reviewed. After the review is passed, the corresponding nodes in the knowledge graph need to be updated and the associations re-established. The present invention can provide designers with auxiliary design support and recommend reliable design solutions based on the data in the knowledge base.

[0124] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent robot prototype-aided design method based on knowledge graph, characterized by: The steps include: Step 1: Data acquisition, the data source is structured data and unstructured data stored by third parties and within the enterprise; Step 2: Knowledge extraction: Input the data obtained in step 1 into the knowledge extraction layer, and extract knowledge through the vertical domain large language model to extract structured information and knowledge from the text, including entity extraction, relationship extraction, and attribute extraction to form relationship triples. Combined with the graph neural network, the complex relationship modeling between entities in the mechanical design knowledge graph is performed to improve the accuracy of entity recognition and relationship extraction. Step 3: Knowledge fusion processing: input the relationship triples formed in step 2 into the knowledge fusion processing layer to perform entity relationship fusion and entity attribute fusion; Step 4: Data storage: Use database and triple data to persistently store the knowledge base, and use document database and relational database to store documents and operation records.

2. The method for intelligent robot prototype auxiliary design based on knowledge graph according to claim 1, characterized in that: The data mainly comes from design materials of operating manipulators and mobile robots, including design plans, CAD drawings, BOM tables, 3D models, product manuals, process documents, and robot design standards and specifications.

3. The method for intelligent robot prototype auxiliary design based on knowledge graph according to claim 1, characterized in that: The step 2 specifically includes: Step 2.1 Entity Extraction: Automatically identify and extract entities with specific meanings from text data. Entity extraction includes two main tasks: entity boundary identification and entity category determination; Step 2.2 Relationship extraction: Identify and extract semantic relationships between entities from the text. Relationship extraction methods include rule-based methods, template-based methods, and machine learning-based methods. Step 2.3 Attribute extraction: Extract and identify the attribute values ​​of entities from the text. Attribute extraction methods include rule-based methods, template-based methods, and machine learning-based methods.

4. The method for intelligent robot prototype auxiliary design based on knowledge graph according to claim 3, characterized in that: In the field of robotic arm and robot prototyping: The entities include component names, materials, dimensions, and process parameters. Entity boundary identification involves identifying the start and end positions of specific components or parameters mentioned in the text, and entity category determination involves classifying these entities into specific types. The relationship includes the assembly relationship between parts, size and tolerance requirements, and material selection basis; The properties include the size, weight, color, and material properties of the component.

5. The method for intelligent robot prototype auxiliary design based on knowledge graph according to claim 1, characterized in that: The complex relationship modeling between entities in the mechanical design knowledge graph using graph neural networks specifically includes: In GNN, the graph G consists of the following elements: node set V: all nodes on the way, edge set E: all edges in the graph, representing the relationship between nodes, node feature x: feature vector x of each node i , the feature update of each node is determined by the information of its neighboring nodes. Assume that the feature vector of node i is x i , the feature vector of its neighbor node is x j (j∈N(i)), the new feature of the node is calculated by the following formula: in: is the feature representation of node i at layer l; AGGREGATE() is an aggregation function used to aggregate the features of neighboring nodes; W (l) is the weight matrix of the lth layer, and the updated eigenvector Processed through a non-linear activation function; In GAT, the attention mechanism is used to dynamically adjust the influence of neighbor nodes. The attention coefficient of each node to its neighbors is calculated by the following formula: Among them: a is the weight vector of the attention mechanism, || represents the vector concatenation operation, ReLU is the activation function, Attention coefficient α ij The attention weight of node i to neighbor node j is calculated by the softmax function: The new feature of node i is updated as:

6. The method for intelligent robot prototype auxiliary design based on knowledge graph according to claim 1, characterized in that: The step three specifically includes: Step 3.1 Data integration: Collect, clean, transform and load data from different sources; Step 3.2 Entity alignment: Match and correspond entities in knowledge graphs from different sources. The main steps of entity alignment include: determining the entity type to be aligned, constructing an entity similarity measurement function, finding matching entities, and evaluating the alignment results. Step 3.3 Knowledge Reasoning: Knowledge reasoning involves rule-based design optimization, case-based reasoning, and model-based reasoning; Step 3.4 Ontology construction: Ontology construction is the process of establishing a domain ontology, which is a structured description of concepts, attributes, and relationship knowledge in a specific domain; Step 3.5 Quality assessment: It is the process of quality inspection and assessment of knowledge fusion results.

7. The method for intelligent robot prototype auxiliary design based on knowledge graph according to claim 1, characterized in that: Also includes: Step 5: Knowledge recommendation. An intelligent recommendation algorithm based on collaborative filtering is used to recommend relevant design knowledge, components, materials or design processes by analyzing the designer's historical design behavior, preferences, choices and similar needs between designers. The collaborative filtering algorithm includes: a user-based collaborative filtering algorithm finds other designers similar to the target designer by analyzing the similarities between designers, and then recommends the design knowledge, components or materials commonly used by these similar designers to the target designer; a project-based collaborative filtering algorithm finds design elements similar to the target designer's current selection by analyzing the similarities between different design knowledge, components or materials, and then recommends these similar elements to the target designer.

8. The method for intelligent robot prototype auxiliary design based on knowledge graph according to claim 7, characterized in that: Step 5 specifically includes: Step 5.1 Data preparation: Collect user behavior data based on the user's operation records on the knowledge graph in the system; construct a user-item matrix to represent the user's interactive behavior on different design resources; Step 5.2: Data preprocessing: After collecting relevant user data, remove outliers and noise data, and then standardize the relevant data; Step 5.3 Similarity calculation: Use the user-based collaborative filtering algorithm and the item-based collaborative filtering algorithm to calculate the similarity between users or items. The cosine similarity calculation formula is as follows: Among them, r u,i and r v,i Represents the ratings of user u and user v on item i, I uv represents the set of items rated by both user u and user v; Step 5.4 Recommendation generation: Based on the similarity of users or items, the recommendation score is calculated. The recommendation score of a user or item is expressed as: Among them, N x Represents a similar set of users or projects. Based on the recommendation score, a recommendation list is generated to recommend the highest-rated design resources to the user.

9. The method for intelligent robot prototype auxiliary design based on knowledge graph according to claim 1, characterized in that: Also includes: Step 6: Knowledge update. For various files generated during the development of new prototypes, users upload these structured or unstructured process files on the web page. The system automatically extracts and integrates relevant knowledge. The acquired structured knowledge is reviewed and verified by experts and then stored in the knowledge base through the interface, realizing the ability of the knowledge base to be continuously updated during use.

10. The method for intelligent robot prototype auxiliary design based on knowledge graph according to claim 1, characterized in that: It also includes step seven: intelligent question and answer. The specific steps are as follows: Step 7.1 The user asks questions or requests to the system through natural language; Step 7.2 The NLU module receives the user's natural language input and performs pre-processing of the data, including noise removal, word segmentation, part-of-speech tagging, and named entity recognition, to convert the user's original question into structured data that the system can parse; Step 7.3 The system identifies the type of user question and determines the user's specific intention; Step 7.4 Based on the parsing results, the system constructs the corresponding query statement and executes the query in the knowledge graph to obtain relevant information; Step 7.5 The large model performs semantic understanding and reasoning based on the user's question and the knowledge information in the existing knowledge graph to find out the knowledge points and information related to the question; After generating the preliminary answer in step 7.6, the system optimizes the answer.

Images