A design diagram construction method based on RFPC conceptual design framework

Abstract

A method for constructing a design graph based on the RFPC conceptual design framework belongs to the technical field of design graph construction. It obtains design knowledge data from text data, then maps the design knowledge to the RFPC model of requirements-function-principle-feature markers for the design knowledge-assisted product conceptual design. It uses natural language processing technology and dependency syntax analysis to characterize the design knowledge elements and the relationships between them, and uses a graph database to provide a knowledge storage and management solution for the design knowledge. This method can more intuitively realize the retrieval and reuse of knowledge in the conceptual design process and support the generation of conceptual design solutions. The present invention solves the limitations of traditional data analysis technology and overcomes the drawbacks of difficulty in effective storage and analysis. The construction of a product design knowledge graph makes it possible to intelligently generate innovative product design solutions, which will facilitate the application of design knowledge and technological innovation in the early stages of product development.

Description

Technical Field

[0001] The present invention belongs to the technical field of design graph construction, and in particular relates to a design graph construction method based on the RFPC conceptual design framework. Background Art

[0002] With the rapid advancement of science and technology and the growing demand for personalized products, product functions and structures are becoming increasingly complex, product upgrades are accelerating, and market competition is intensifying. To survive and develop, companies must continuously launch new products that meet user needs and are competitive in the market. Therefore, improving the level of innovative product design solutions has become a priority for companies.

[0003] Product innovation design is a creative intellectual process that requires deep background knowledge and extensive design experience. It is a knowledge-based creative process with innovative thinking at its core. A set of knowledge and data are used at every stage of product design, including product planning, concept design, implementation design, and detailed design. During the design process, engineers typically spend more than half of their time organizing design knowledge and data.

[0004] Therefore, effectively managing design knowledge and data is a key technology for companies to maintain competitiveness and shorten product development time. Effectively utilizing this massive amount of data to promote data-driven innovative product design methods has become a research hotspot. The conceptual design stage is a crucial phase in the product lifecycle, determining subsequent product design. Lack of innovation in conceptual design solutions directly leads to product innovation failure. Generally speaking, conceptual design can be described as the stages of defining design requirements, specifying functions, generating conceptual understandings, and evaluating and selecting concepts. Conceptual design is a time-consuming, labor-intensive, and step-by-step learning process. A systematic and comprehensive conceptual design approach, powered by knowledge graphs, can help designers thoroughly explore the entire space of options in the early stages of product development, significantly reducing the likelihood of design failure or being outperformed by competing products later in the development process.

[0005] Product concept design is a series of iterative and complex engineering processes guided by design requirements. By establishing functional behavior associations to seek the correct combination mechanism, determine the basic solution path, and generate design solutions, the success of new product development depends on the generation of design concepts in the concept design stage. Companies need to quickly produce new products that meet the diverse and personalized needs of consumers without increasing production costs and product development cycles. Product concept design is one of the key steps to solving these problems, and the efficiency of product data usage is the main factor affecting the efficiency of product concept design. As an important factor in product design and development, data has occupied an irreplaceable position throughout the product life cycle. In the process of interaction between the product and the outside world (such as users and the environment), a large amount of data can be generated. Product data contains rich design knowledge, which can improve the efficiency of concept design and the innovation of design solutions. These data represent the characteristics of the product's connection with the outside world. The product concept design framework mainly refers to the process of modeling and analyzing a large amount of design data, mining the correlations and hidden patterns of things, and clearly expressing the design intention. The generation process of the product concept design scheme is a mapping process from fuzzy requirements to specific structures. It establishes a data-based product concept design model, combines the concept design process to obtain relevant knowledge from the data, and after establishing the functional structure and finding the appropriate principle solution, combines the solution into the data involved in the concept design, including product function data, product structure data, design alternative data, etc., to assist product design.

[0006] A knowledge graph is a structured semantic knowledge base composed of entity-relationship-entity and entity-attribute-attribute-value triplets. Essentially, it represents a large semantic network. As scholars delve deeper into textual information, natural language processing technology has gradually developed and matured, enabling applications such as text data mining, semantic analysis, knowledge discovery, information retrieval, and artificial intelligence. Integrating knowledge graph technology can better describe heterogeneous information at the design information data and model levels. Knowledge graphs allow for intuitive visualization of relationships between design information, enabling the identification of required functions based on design goals, identifying appropriate primitives, and seeking appropriate innovative design solutions. Decision makers can analyze and mine the specific design information within the design graph to discover new combinations to aid product design, thereby improving product and system solutions and effectively reducing the time and financial costs of early product design investments. Summary of the Invention

[0007] In view of the above problems existing in the prior art, the purpose of the present invention is to provide a design map construction method based on the RFPC conceptual design framework to solve the limitations of traditional data analysis technology and overcome the disadvantages of difficulty in effective storage and analysis.

[0008] The present invention provides the following technical solutions:

[0009] A design map construction method based on the RFPC conceptual design framework includes the following steps:

[0010] S1. Construct a knowledge acquisition module to acquire design knowledge data from text data and preprocess the acquired design knowledge data;

[0011] S2. Construct a design knowledge framework module, build an RFPC design knowledge ontology framework, and map the pre-processed design knowledge data into the framework model;

[0012] S3. Build a knowledge representation module, establish a design knowledge graph model, and use deep learning and semantic-based dependency syntax analysis combined with trigger words to extract design knowledge elements and their corresponding relationships to form design knowledge triples;

[0013] S4. Build a graph construction module to map the design knowledge triples into the RFPC design knowledge ontology framework and store them in the graph database;

[0014] S5. Build a graph application module to provide knowledge retrieval and reuse during the concept design process of product design and support the generation of design solutions.

[0015] Furthermore, the specific process of step S1 is as follows:

[0016] S1.1. Obtaining raw design knowledge from text data;

[0017] S1.2. Perform design knowledge data integration on unprocessed design knowledge: Integrate the acquired design knowledge into a unified type through multivariate data integration;

[0018] S1.3. Remove stop words: that is, remove irrelevant words from unstructured design knowledge text;

[0019] S1.4. Feature selection: that is, selecting design vocabulary from the design knowledge text as training input data for the entity recognition model.

[0020] Furthermore, the RFPC design knowledge ontology framework model includes four first-level design knowledge ontology layers, namely, the requirement layer, the function layer, the principle analysis layer and the feature mark layer; each first-level design knowledge ontology layer includes a group of lower-level design knowledge ontology classifications.

[0021] Furthermore, the requirement layer describes the design requirements or design tasks, determines the goals of product design, determines the overall functions by the overall design requirements, and decomposes and refines the user's design requirements step by step according to the hierarchical relationship; the function layer is used to complete the decomposition of the overall function, thereby obtaining the functional structure of the product; the principle understanding layer is a hierarchical solution to the requirements, which is a specific implementation method of one or more functions of the functional layer, and is also a combination and decomposition of multiple principle understandings to achieve specific functions of product design; the feature mark layer is used to describe the intuitive effect of one or more principle understandings mapped to achieve a certain specific function in the product design process.

[0022] Furthermore, the specific process of step S3 is as follows:

[0023] 3.1) Input pre-processed unstructured design knowledge text data;

[0024] 3.2) Collect technical vocabulary and design knowledge attributes from design knowledge texts to build a basic corpus of the design knowledge graph;

[0025] 3.3) Train the BERT model to generate word vectors;

[0026] 3.4) For named entity recognition of design knowledge, we use a natural language processing algorithm model to extract entities corresponding to design knowledge categories at the requirement and principle understanding levels in the design knowledge graph from the design knowledge text, thereby improving the accuracy and efficiency of the design knowledge entity recognition of the training model.

[0027] 3.5) Extract the functional relationships between design knowledge entities, generate triples of the design knowledge graph, generalize the extracted relationship templates, and standardize the functional relationships;

[0028] 3.6) Construct a design knowledge graph: Based on triples, design knowledge entities are represented by graph nodes, and the functional relationships between design knowledge entities are represented by edges in the graph database to construct a design knowledge graph.

[0029] Furthermore, the knowledge representation module includes a design knowledge extraction part S31, a design knowledge relationship extraction part S32, a design knowledge disambiguation part S33 and a design knowledge merging part S34.

[0030] Furthermore, the steps 3.1-3.4 are implemented based on the design knowledge extraction part S31, and the design knowledge extraction part S31 uses the Bert-BiLSTM-CRF entity recognition model to train the design text.

[0031] Furthermore, the S3.5 is implemented based on the design knowledge relationship extraction part S32 and the design knowledge disambiguation part S33. The specific process is as follows: the correspondence between design knowledge is jointly extracted using the LTP and trigger word methods, the design knowledge corpus is segmented into sentences, words and parts of speech are tagged, the design text is processed using the LTP tool, the syntactic structure is analyzed by analyzing the dependency relationship between words in the sentence, the triple recognition principle is constructed, each trigger word in the sentence is matched with rules, and the design knowledge relationship triples are identified.

[0032] Furthermore, the step 3.6) is implemented based on the design knowledge merging part S34, and the specific process is as follows: based on the multiple design knowledge extractions and their relationship correspondences in the design knowledge extraction part S31, the design knowledge relationship extraction part S32 and the design knowledge disambiguation part S33, the entity recognition, entity alignment and entity linking of the design knowledge are completed to form the final design knowledge merger.

[0033] Furthermore, the specific process of step 4 is as follows:

[0034] 4.1. Storing the design knowledge entities identified by the entity recognition model;

[0035] 4.2. Storing entity relationships extracted from design knowledge relationships;

[0036] 4.3. Store design knowledge triples and their corresponding design knowledge attributes;

[0037] 4.4. Map the design knowledge in steps 4.1, 4.2 and 4.3 into the PFPC conceptual design framework and store it in the graph database.

[0038] By adopting the above technology, compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] 1) The present invention obtains textual materials related to product design knowledge from a wide range of sources. Because the design knowledge is redundant, the data volume is large, and the data types are diverse and complex, the knowledge is extracted and integrated to form a complete knowledge representation system, namely the design graph model layer, and the design information is completely divided to facilitate the overall analysis of the top-down innovative design process of product design.

[0040] 2) Compared with traditional design information storage methods such as professional books and enterprise data systems, the present invention combines the design knowledge with knowledge graphs in terms of the construction of the ontology framework of design knowledge, the extraction of specific design knowledge information, and the storage method of design knowledge triples. This makes the design knowledge more scalable and visual, and can make the design knowledge more valuable. It is a powerful auxiliary means for product innovation design and can optimize the product innovation design process. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a schematic diagram of the framework of the design map construction method of the present invention;

[0042] Figure 2 This is a schematic diagram of the design knowledge text preprocessing process of the present invention;

[0043] Figure 3 This is a schematic diagram of the design knowledge ontology framework of the present invention;

[0044] Figure 4 A logical relationship diagram between the design knowledge systems of the present invention;

[0045] Figure 5 The figure is a flow chart for constructing the design map of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] On the contrary, the present invention covers any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention as defined by the claims. Furthermore, to facilitate a better understanding of the present invention, certain specific details are described in detail below in the detailed description of the present invention. Those skilled in the art will be able to fully understand the present invention without these details.

[0048] See also Figure 1-5 A design map construction method based on the RFPC conceptual design framework includes the following steps:

[0049] S1. Construct a knowledge acquisition module to acquire design knowledge data from text data and preprocess the acquired design knowledge data.

[0050] The specific process is as follows:

[0051] S1.1. Obtain unprocessed design knowledge from design books, web encyclopedias, design literature, and design reports;

[0052] S1.2. Perform design knowledge data integration on unprocessed design knowledge, i.e. integrate the multivariate data of design knowledge from web encyclopedias, design professional books, design literature and design reports into a unified type;

[0053] S1.3. Remove stop words, i.e. remove irrelevant words from unstructured design knowledge text;

[0054] S1.4, feature selection, that is, selecting design vocabulary from design knowledge text as training input data for entity recognition model, Figure 2 It is the process of designing knowledge text preprocessing;

[0055] S2. Construct a design knowledge framework module, build an RFPC design knowledge ontology framework, and map the preprocessed design knowledge data into the framework model.

[0056] Among them, in the RFPC design knowledge ontology framework, a demand-function-original understanding-feature mark design knowledge system is established. There are 32 design knowledge categories under the design knowledge system, such as Figure 3 As shown in the design knowledge ontology framework, the demand layer includes automobiles, machine tools, agricultural machinery, engineering machinery, electrical machinery, basic machinery, instruments and meters, packaging machinery, environmental protection machinery, heavy mining machinery and petrochemical general machinery; the functional layer includes separation, guidance, connection, control, conversion, supply, signal and support; the principle layer includes components, working principles and physical effects; the characteristic mark layer includes life, transportation, cost, safety, manufacturability, assembly, environmental protection, maintainability, ergonomics and recycling value.

[0057] Specifically, the requirement layer (R) describes the design requirements or design tasks, determines the goal of product design, and determines the overall function by the overall design requirements. The user's design requirements are decomposed and refined step by step according to a certain hierarchical relationship, thereby breaking down vague and non-specific requirements into clear and specific sub-requirements, which is conducive to the designer's correct understanding and thus designing products that meet user needs;

[0058] Specifically, the function layer is the core design knowledge of the design knowledge ontology framework. The main task of the function layer (F) is to complete the decomposition of the overall function, thereby obtaining the functional structure of the product. According to the classification method of the function base in the defined conceptual design knowledge framework, multiple functions in the specific design scheme are described, such as the overall function, sub-function and sub-function. At the same time, the function layer based on the function base provides a feasible method for mapping to the principle layer.

[0059] Specifically, the principle layer (P) is a hierarchical solution to the needs, and is also the specific implementation of one or more functions in the functional layer. It is also the combination and decomposition of multiple principle layers to achieve a specific function of product design.

[0060] Specifically, the Characteristic layer (C) is used to describe the intuitive effects of one or more principles mapped to achieve a specific function in the product design process, such as safety, maintainability, environmental protection, stability, cost, life and other specific indicators. It is also the feedback to the needs in the design process; Figure 4 It is a logical relationship diagram between design knowledge systems.

[0061] The knowledge framework module is constructed using a bottom-up approach and is used to construct an ontology based on multiple design knowledge such as structured design knowledge, semi-structured design knowledge and unstructured design knowledge. When constructing the design knowledge framework module, it is necessary to classify the design knowledge, define the design entities, define attributes and define relationships.

[0062] S3. Build a knowledge representation module, establish a design knowledge graph, and use deep learning and semantic-based dependency syntax analysis combined with trigger words to extract design knowledge elements and their corresponding relationships to form design knowledge triples. The specific process is as follows:

[0063] 3.1) Input pre-processed unstructured design knowledge text data. The input should comprehensively consider factors such as design knowledge quality and format specifications;

[0064] 3.2) Collect proprietary technical vocabulary and design knowledge attributes from design knowledge texts to build a basic corpus of the design knowledge graph;

[0065] 3.3) Train the BERT model to generate word vectors; the BERT model can learn the contextual semantic information of words and map words into a vector space, providing support for subsequent design tasks such as knowledge entity recognition and relationship extraction.

[0066] 3.4) For named entity recognition of design knowledge, a natural language processing algorithm model is used to extract entities corresponding to 19 design knowledge categories at the "requirements" and "original understanding" levels in the design knowledge graph from the design knowledge text, improving the accuracy and efficiency of the design knowledge entity recognition of the training model;

[0067] 3.5) Extracting and standardizing design knowledge relationships. Functional relationships refer to the functions in the requirements-function-original understanding-feature mark, which are often expressed in verb form. Extracting the functional relationships between design knowledge entities generates triples of the design knowledge graph. By generalizing the extracted relationship templates and standardizing the relationships, specifically determining trigger words to improve the extraction of design knowledge relationships, these operations reduce the manual labor involved in building the design graph and improve the speed and accuracy of design graph generation.

[0068] 3.6) Construct a design knowledge graph: Based on triples, design knowledge entities are represented by graph nodes, i.e., object primitives representing the structure and function of the product. Functional relationships between design knowledge entities are represented by edges in the graph database to construct a design knowledge graph.

[0069] The knowledge representation module includes design knowledge extraction S31, design knowledge relationship extraction S32, design knowledge disambiguation S33 and design knowledge merging S34.

[0070] Specifically, steps 3.1)-3.4) are based on design knowledge extraction S31.

[0071] Design knowledge extraction S31: The BERT-BiLSTM-CRF entity recognition model is used to train the design text. The design knowledge text is mapped and encoded into an embedding vector using the BERT Chinese pre-training model. The corresponding feature vector is trained, that is, the single character feature, sentence feature, and position feature of each word in the design text are calculated. The complete vector attribute of each word in the design text is obtained by adding the feature vectors. The complete features obtained by the multi-layer Transformer in BERT are calculated to obtain the feature vector. The annotated design knowledge text passes through the BERT layer to convert each word of the design knowledge text into a low-dimensional word vector.

[0072] First, the vector generated by encoding the knowledge text is designed to pass through three different fully connected layers to obtain three vectors Q, K, and V. Then Q and K T Perform matrix multiplication to obtain the vector QK of the degree of relevance between the word and other words T , and finally the standardized QK T Put it into the softmax activation function to get the correlation vector between words, and then multiply it by V to get the final vector, as shown in the formula:

[0073]

[0074] Among them, Q is the character vector of the current design knowledge encoding, K is the attention vector of the character vector of the current design knowledge encoding, and V is the relationship information vector between each character vector of the current design knowledge text sequence. is a normal distribution with variance d.

[0075] The vectors of the design knowledge text are then concatenated through a multi-head structure. To address the problem that the attention mechanism does not extract the temporal features of the text, the Transformer adds a position encoding vector to the design knowledge text before data preprocessing.

[0076]

[0077]

[0078] Among them, pos refers to the position of the current design knowledge in the design text, i refers to the index of each value in the vector, PE refers to the position code of the design knowledge, and d model Refers to the word vector embedding dimension of design knowledge.

[0079] And perform weighted summation with the vector generated after encoding the design knowledge text to obtain the relative position of each word in the design knowledge sentence, and obtain the complete encoding vector of each word in the design text;

[0080] The features, or word vector sequences, are fed into the BiLSTM model for training. Based on feature extraction, the intermediate state of the BiLSTM output is used as input to the attention layer. The association between design knowledge is determined by calculating the degree of attention between each design knowledge entity and other statements. The output sequences of the forward and backward LSTM hidden layers are then calculated separately, and the vector output sequences in both directions are concatenated to obtain a feature matrix. This layer outputs the probability that each word in the design knowledge text belongs to a different design knowledge entity. This helps better mine the semantic information between technology-related words and other words in the design knowledge text, and better understand the design knowledge information implicit in the context.

[0081] Finally, combined with the CRF layer, this layer is used to decode and output the design knowledge entity with the highest probability to predict the label sequence, and learn the transfer rules between the named entity labels related to adjacent design knowledge in the design knowledge sentence. For example, using "I-coal mining machine" as the label of the first word of the coal mining machine entity is an illegal label, because a word can only be in two situations, one is a term, the label is "B-coal mining machine" or "I-coal mining machine", and the other is not a term, the label is "O". The conditional random field model can avoid the occurrence of this illegal situation, obtain the label type of each design knowledge entity, extract and classify the entities in the sequence, and obtain the globally optimal design knowledge word label sequence.

[0082] The specific process is as follows:

[0083] a) Unstructured design knowledge entity recognition text data is annotated using the YEEDA annotation tool, and a dataset is constructed according to the RFPC conceptual design knowledge framework;

[0084] b) Data labeling rules: Use the BIO labeling rules. For example, label "coal mining machine" with "B-demand" representing the first character in the entity, "I-demand" representing the remaining characters in the entity, and "O" representing words that do not belong to the entity. The labeled data can be used for training and testing the design knowledge entity recognition model;

[0085] c) Divide the corpus into a training set and a test set: Using a ten-fold cross-validation approach, the corpus is divided into a 9:1 ratio. Part of the design knowledge corpus is used as the training set for the BERT-BiLSTM-CRF model, and the remaining corpus is used as the test set to ensure the effectiveness of the design knowledge entity recognition model.

[0086] d) Training the BERT-BiLSTM-CRF model: Use BERT to encode the design knowledge text into word vectors, input vector data into the model, and then optimize the model parameters to determine the optimal parameter combination and train the optimal model for design knowledge entity recognition.

[0087] e) The trained design knowledge entity recognition model is used to process the entire corpus: the unlabeled unstructured design knowledge text is input into the trained design knowledge entity recognition model, i.e., the BERT-BiLSTM-CRF model, and finally the relevant design knowledge entity recognition results, i.e., the design knowledge entity fragments, are output.

[0088] Specifically, step 3.5) is based on design knowledge relationship extraction S32 and design knowledge disambiguation S33.

[0089] Design knowledge relationship extraction S32:

[0090] LTP and trigger word methods are used to jointly extract the correspondence between design knowledge, and the design knowledge corpus is segmented into sentences, words and parts of speech. The LTP tool is used to perform part-of-speech tagging and dependency syntax analysis on the design text. The syntactic structure is analyzed by analyzing the dependency relationship between words in the sentence, and the triple recognition principle is constructed. Rule matching is performed on each trigger word in the sentence to identify the design knowledge relationship triples.

[0091] Design knowledge disambiguation S33: Link design knowledge of the same type but with slight differences in expression to the same design knowledge element, perform regularization on the design knowledge text, regularize the non-referential but pointed technical vocabulary related to design knowledge involved in the text, calculate the semantic similarity of the relevant pronouns, and replace them with the design knowledge entity with the highest calculated score. Secondly, there are many types of design knowledge relationships involved, and the relationship types need to be classified and standardized into 8 design knowledge functional categories under the functional layer. The specific method is as follows:

[0092] By using similarity calculation, the relationship words (function words) obtained from the design knowledge relationship extraction are used as trigger words for the design knowledge relationship extraction. Combined with clustering algorithms such as the K-means clustering algorithm, the function words are mapped to the eight design knowledge function categories under the function layer to complete the disambiguation of design knowledge. The following table is an example:

[0093] Relationship combination type trigger words Example Connection relationship Connect, mesh, couple, etc. Globe valve in series with regulating valve Control relationship Drive, adjust, compress, etc. Guide mechanism controls the push rod Conversion relationship Change, conversion, concentration, etc. Output shaft changes horizontal axis cutting axis Signal relationship Measuring, marking, etc. Detection device measures the body of the coal mining machine Supportive relationships Loading, installation, assembly, etc. The drilling rig slide is installed on the slide rail Supply Relationship to accommodate, collect, provide, etc. Lifting mechanism supports spray device Guiding relationship Spraying, rotating, pumping, etc. The turntable rotates with the turntable Branch relationship Separation, release, filtration, etc. The controller cuts off the electrical and oil pumps

[0094] Specifically, step 3.6) merges S34 based on design knowledge.

[0095] Design knowledge merging S34: Through the extraction of the above-mentioned multiple design knowledge and their relationship correspondence, entity recognition, entity alignment and entity linking of the design knowledge are completed to form the final merging of the design knowledge.

[0096] S4. Build a graph construction module to map the design knowledge triples into the RFPC design knowledge ontology framework and store them in the graph database. The specific process is as follows:

[0097] 4.1. Storing the design knowledge entities identified by the entity recognition model;

[0098] 4.2. Storing entity relationships extracted from design knowledge relationships;

[0099] 4.3. Store design knowledge triples and their corresponding design knowledge attributes;

[0100] 4.4. Map the design knowledge in steps 4.1, 4.2 and 4.3 into the PFPC conceptual design framework and store it in the graph database.

[0101] S5. Build a graph application module to provide knowledge retrieval and reuse during the concept design process of product design and support the generation of design solutions.

[0102] The graph application module includes intelligent retrieval of design knowledge S51, intelligent solution of design solutions S52, intelligent recommendation of design solutions S53, and innovation evaluation of design solutions S54.

[0103] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A design graph construction method based on the RFPC conceptual design framework, characterized by: The following steps are involved: S1. Construct a knowledge acquisition module to acquire design knowledge data from text data and preprocess the acquired design knowledge data; S2. Construct a design knowledge framework module, build an RFPC design knowledge ontology framework, and map the pre-processed design knowledge data into the framework model; S3. Build a knowledge representation module, establish a design knowledge graph model, and use deep learning and semantic-based dependency syntax analysis combined with trigger words to extract design knowledge elements and their corresponding relationships to form design knowledge triples; S4. Build a graph construction module to map the design knowledge triples into the RFPC design knowledge ontology framework and store them in the graph database; S5. Build a graph application module to provide knowledge retrieval and reuse during the concept design process of product design and support the generation of design solutions; The RFPC design knowledge ontology framework model includes four first-level design knowledge ontology layers, namely, the requirement layer, the function layer, the principle layer and the feature mark layer; each first-level design knowledge ontology layer includes a group of lower-level design knowledge ontology classifications; The requirement layer describes the design requirements or design tasks, determines the goal of product design, and determines the overall function by the overall design requirements. The user's design requirements are decomposed and refined step by step according to the hierarchical relationship; The functional layer is used to decompose the overall function, thereby obtaining the functional structure of the product. The principle layer is a hierarchical solution to the requirements, which is the specific implementation method of one or more functions of the functional layer. It is also the combination and decomposition of multiple principle layers to achieve specific functions of the product design. The feature mark layer is used to describe the intuitive effect of one or more principle layers mapped to achieve a specific function in the product design process. The specific process of step S3 is as follows: 3.1) Input pre-processed unstructured design knowledge text data; 3.2) Collect technical vocabulary and design knowledge attributes from design knowledge texts to build a basic corpus for the design knowledge graph; 3.3) Train the BERT model to generate word vectors; 3.4) For named entity recognition of design knowledge, we leverage natural language processing algorithms and models to extract entities corresponding to design knowledge categories at the requirements and principle understanding levels within the design knowledge graph from design knowledge text, thereby improving the accuracy and efficiency of the training model's design knowledge entity recognition. 3.5) Extract the functional relationships between design knowledge entities, generate triples of the design knowledge graph, generalize the extracted relationship templates, and standardize the functional relationships; 3.6) Construct a design knowledge graph: Based on triples, design knowledge entities are represented by graph nodes, and the functional relationships between design knowledge entities are represented by edges in the graph database to construct a design knowledge graph.

2. A design diagram construction method based on the RFPC conceptual design framework according to claim 1, characterized in that The specific process of step S1 is as follows: S1.

1. Obtaining raw design knowledge from text data; S1.

2. Perform design knowledge data integration on unprocessed design knowledge: Integrate the acquired design knowledge into a unified type through multivariate data integration; S1.

3. Remove stop words: that is, remove irrelevant words from unstructured design knowledge text; S1.

4. Feature selection: that is, selecting design vocabulary from the design knowledge text as training input data for the entity recognition model.

3. A design diagram construction method based on the RFPC conceptual design framework according to claim 1, characterized in that The knowledge representation module includes a design knowledge extraction part S31, a design knowledge relationship extraction part S32, a design knowledge disambiguation part S33 and a design knowledge merging part S34.

4. A design diagram construction method based on the RFPC conceptual design framework according to claim 3, characterized in that The steps 3.1-3.4 are implemented based on the design knowledge extraction part S31, and the design knowledge extraction part S31 uses the Bert-BiLSTM-CRF entity recognition model to train the design text.

5. A design map construction method based on RFPC conceptual design framework according to claim 4, characterized in that Said S3.5 is implemented based on the design knowledge relationship extraction part S32 and the design knowledge disambiguation part S33. The specific process is as follows: using the LTP and trigger word methods to jointly extract the corresponding relationship between design knowledge, performing sentence segmentation, word segmentation and part-of-speech tagging on the design knowledge corpus, using the LTP tool to process the design text, analyzing the syntactic structure by analyzing the dependency relationship between words in the sentence, constructing the triple recognition principle, performing rule matching on each trigger word in the sentence, and identifying the design knowledge relationship triple; Design knowledge disambiguation S33: Link design knowledge of the same type but with different expressions to the same design knowledge element, regularize the design knowledge text, regularize the non-referential, directed, and design knowledge-related technical vocabulary involved in the text, calculate the semantic similarity of related pronouns, and replace them with design knowledge entities with larger calculated scores.

6. A design diagram construction method based on the RFPC conceptual design framework according to claim 5, characterized in that The step 3.6) is implemented based on the design knowledge merging part S34, and the specific process is as follows: based on the multiple design knowledge extractions and their relationship correspondences in the design knowledge extraction part S31, the design knowledge relationship extraction part S32 and the design knowledge disambiguation part S33, the entity recognition, entity alignment and entity linking of the design knowledge are completed to form the final design knowledge merging.

7. A design map construction method based on the RFPC conceptual design framework according to claim 6, characterized in that The specific process of step 4 is as follows: 4.

1. Storing the design knowledge entities identified by the entity recognition model; 4.

2. Storing entity relationships extracted from design knowledge relationships; 4.

3. Store design knowledge triples and their corresponding design knowledge attributes; 4.

4. Map the design knowledge in steps 4.1, 4.2 and 4.3 into the PFPC conceptual design framework and store it in the graph database.

Images