Domain knowledge graph construction method based on FBS conceptual design model

Abstract

The invention discloses a domain knowledge graph construction method based on an FBS conceptual design model, and relates to the technical field of knowledge graph construction. The method comprises the following steps: backtracking and analyzing design structures, design behaviors and design functions related to product concept design and feature attributes related to the design structures from a document, and constructing a design knowledge graph by pre-defining relationship types among the design elements; in order to solve the problem that a single entity corresponds to multiple entities in a document, a traditional Casre l model structure is improved, and an attention module is added. Searching a design structure set capable of realizing the function according to the mapping relation between the function and the structure; and for the two functions with interaction, retrieving structures with connection paths in the two corresponding design structure sets, and taking a set of the design structures. According to the method, the knowledge graph providing stable and systematic design knowledge can be constructed, and the problem of single entity relation overlapping can be effectively solved through the improved Casre model.

Description

Technical Field

[0001] The present invention belongs to the technical field of knowledge graph construction, and in particular, relates to a method for constructing a domain knowledge graph based on an FBS conceptual design model. Background Art

[0002] The product design process is generally divided into three stages: concept design, detailed design, and trial production. Concept development and design mainly rely on the experience of designers and abstract and uncertain conceptual influences such as consumer market demand, so it has high flexibility. However, the subsequent detailed design and manufacturing trial production stage involves the specific structure of the product. Changes in one design link will lead to corresponding changes in other links, making the product's concept development and design have an important impact on the entire production process.

[0003] The survey shows that conceptual design accounts for 70%-80% of the entire product manufacturing cost. Product concept development and design innovation can be divided into two categories. One is to conduct innovative design based on the basic principles of the product in order to design products with different working principles and more effective; the other is to upgrade and optimize products based on existing market products to improve the product's adaptability to demand. For the current common practice in the manufacturing industry, most products are optimized and improved on the basis of existing products and markets, matching consumer demand under the conditions of low product cost and mass production.

[0004] At present, there are problems in product concept development and design, such as low product design efficiency, lack of design solutions, and poor design effects. Designers lack a complete and systematic design knowledge support in the product development stage, which causes them to spend a lot of time and energy to integrate the relevant design knowledge required for product design. For example, Chinese patent CN118171726A provides a method, system, storage medium and computer for constructing a knowledge graph for the entire project process. The construction method includes: dividing the entire project process into business stages and designing a knowledge system, constructing a knowledge graph model layer according to the entity attributes and entity relationships of the knowledge system; obtaining business-related data for each business stage and preprocessing the relevant data; defining the entity type and relationship type of the relevant data according to the knowledge graph model layer and acquiring knowledge to construct the corresponding knowledge graph; obtaining the entity features of the knowledge graph, and performing entity alignment operations on all knowledge graphs according to the entity features to achieve the fusion of knowledge graphs; completing the fused knowledge graph to construct a complete knowledge graph for the entire project process. Chinese patents such as CN116204660A, CN117668248A, CN115098691A, etc., all provide a method for establishing a knowledge graph.

[0005] However, the current dilemma faced by manufacturing enterprises is the rapid changes in the consumer market and the personalization of consumer demand. This makes the innovative design capabilities of enterprises face higher challenges when facing the complex consumer demands of the current market. And it is obviously difficult for the model that mainly relies on the existing knowledge graph to keep up with the development and changes of the market. Therefore, this paper proposes a design knowledge graph construction method based on the FBS concept design model. Summary of the invention

[0006] The purpose of the present invention is to provide a method for constructing a domain knowledge graph based on the FBS concept design model, which forms a database about FBS by crawling, preprocessing, screening and classifying target information, including a structure database, a behavior database and a function database. The three databases are sequentially mapped through structure-behavior and behavior-function to form function-structure mapping nodes, thereby constructing a knowledge graph of the design model, solving the problem of insufficient objectivity of the design scheme in the existing traditional concept design process, overcoming the problem of low efficiency in reuse of design knowledge, and constructing a complete design knowledge base by collecting and combing the design knowledge, which can effectively improve the efficiency of the design, reduce R&D costs, and shorten the design and development cycle of the product.

[0007] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0008] The present invention is a method for constructing a domain knowledge graph based on the FBS conceptual design model, which is used to systematically express the design knowledge about the product. Entities and entity relationships are the basic components for establishing a knowledge graph. The ontology mentioned in the present invention is a category expression and summary of entities. Entities are specific expressions of the concept of ontology, and entity relationships express the connection between entities. In the subsequent tasks of the present invention, structures, behaviors, and functions are listed as the ontology of related knowledge entities. Knowledge entities are expressed in natural language, and different entity relationship concepts are defined to express design knowledge. A method for constructing a domain knowledge graph based on the FBS conceptual design model includes the following steps:

[0009] Step 1: Build a design knowledge carrier acquisition module, obtain relevant design documents through web crawlers, pre-process the document data, and establish a top-level design knowledge carrier database, that is, sort out and analyze the design field documents of related products, extract texts and store them in the product design text library;

[0010] Step 2: Use the neural network model to extract the structural entities of the products in the relevant text, classify the product structure, extract the relevant design structure knowledge, and store it in the product design structure database, that is, build the expression module of design knowledge and preliminarily establish the design knowledge graph; the specific process is as follows:

[0011] Step S021: Based on the FBS design model, a top-down hierarchical analysis of the design function, design behavior, and design structure of the design product is performed, and the behavior is used as a constraint between the design function representation and the structure of the design structure;

[0012] Step S022: The SRO-based expression structure is in the form of a triple, namely: (subject, relationship, object) as a knowledge representation framework, where both the subject and the object are knowledge ontology;

[0013] Step S023: Based on the knowledge ontology representation framework, the ontology type and relationship type of knowledge are defined in the design knowledge graph. The ontology type is the type based on FBS, and the relationship type is the inclusion relationship and connection relationship between design structures, and the input and output relationship of the behavior flow with respect to the design structure.

[0014] Step 3: Build a knowledge extraction module to extract design knowledge entities from relevant product documents and map them to corresponding ontologies, classify the attributes of knowledge representation for the entities, and then save the classified entities in the form of triples in the graph database. The specific process is as follows:

[0015] Step S331: performing named entity recognition on the patent text, and extracting the design knowledge entities including the structural entities in the text and the corresponding attribute information and behavior information of the structural entities;

[0016] Step S332: Perform relationship judgment on all identified structural entities, and based on the relationship type defined in step 2, determine all structural entities corresponding to the subject and object under a specified relationship type, thereby extracting corresponding triples;

[0017] Step S333: The same structural entities are merged based on the similarity of the character vectors, the text content and attribute information of the structural entities are vectorized, and the cosine similarity between the two is calculated. If the similarity value is higher than a certain threshold, they are regarded as the same structural entity;

[0018] Step S334: Based on the triples, the nodes of the knowledge graph are taken as structural entities, and the edges are taken as the relationship types of the structural entities, and are saved in the graph database;

[0019] Step 4: Construct a knowledge generation module to generate the functional description of the parts contained in the product from the relevant description of the product and obtain the functional entity. The specific process is as follows:

[0020] Step S41: receiving the document text through the encoder-decoder model to output the functional description of each component, and obtaining the relationship triple of function and structure;

[0021] Step S42: merging the functional entity and the structural entity respectively;

[0022] Step S43: Retrieve the corresponding structural entity in the graph database, and save the relationship triples between the function and the structure into the graph database;

[0023] Step 5: Construct a knowledge application module to realize knowledge retrieval and reuse in the product concept design process. The specific process is as follows:

[0024] Step S51: Retrieve the function entity and find the required function;

[0025] Step S52: according to the mapping relationship between the function and the structure, searching for a set of design structures that can realize the function;

[0026] Step S53: For two interactive functions, search for structures with connection paths in the corresponding two design structure sets, and obtain a collection of design structures.

[0027] Furthermore, the ontology types defined in the step 2 include design function ontology, design behavior ontology, and design structure ontology, and the relationship type definition includes the inclusion relationship and constraint relationship between design structures, and the input and output relationship between design structures and behavior flows; wherein behavior flow is a general term for material flow, energy flow, and signal flow, that is, if there is functional interaction between two design structures, then there will be an input and output relationship of behavior flow between them.

[0028] Furthermore, the named entity recognition part in step three is based on the BERT model pre-trained word vector encoder, Bi LSTM is used to encode long text, and the head and tail pointer network is used for decoding.

[0029] Furthermore, the relationship judgment part in step three is based on the Casel model, which extracts the subject of the triple in advance, and then constructs a multi-layer head and tail pointer network, each layer corresponds to a predefined relationship type, and decodes the object corresponding to the subject under each head and tail pointer network.

[0030] Furthermore, the graph database in step three is Neo4j.

[0031] Furthermore, the encoder-decoder model in step 4 is a Transformer.

[0032] The present invention has the following beneficial effects:

[0033] The present invention uses the FBS concept design model to trace back and analyze the design structure, design behavior, design function and characteristic attributes related to the product concept design from the design documents. By pre-defining the relationship types between these design elements, a design knowledge graph is constructed to provide stable and systematic design knowledge support for the product development process, thereby improving the efficiency of product development and reducing product development costs.

[0034] The present invention improves the Casrel model, constructs a multi-layer head-tail pointer network, adds a self-attention mechanism to fuse the head entity information and the sequence element information, can solve the problem that a single entity in a document corresponds to multiple entities, and improves the quality of knowledge graph generation.

[0035] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0037] Figure 1 It is a flow chart of the design knowledge graph construction method based on the FBS conceptual design model of the present invention;

[0038] Figure 2 It is a type diagram of the behavior chain in the design knowledge graph construction method based on the FBS conceptual design model of the present invention;

[0039] Figure 3 It is a structural diagram of an entity recognition model of a knowledge extraction module in a design knowledge graph construction method based on an FBS conceptual design model of the present invention;

[0040] Figure 4 It is a structural diagram of the entity relationship extraction model of the knowledge extraction module in the design knowledge graph construction method based on the FBS conceptual design model of the present invention;

[0041] Figure 5 It is a model structure diagram generated by the design function description of the knowledge generation module in the design knowledge graph construction method based on the FBS conceptual design model of the present invention;

[0042] Figure 6 It is an example diagram of the division of functional modules in the knowledge application module in the design knowledge graph construction method based on the FBS conceptual design model of the present invention;

[0043] Figure 7 It is an application flow chart of the knowledge application module in the design knowledge graph construction method based on the FBS conceptual design model of the present invention. DETAILED DESCRIPTION

[0044] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0045] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0046] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0047] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0048] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0049] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0050] Embodiment 1:

[0051] See also Figure 1As shown, the present invention is a method for constructing a domain knowledge graph based on the FBS conceptual design model, comprising the following steps:

[0052] S1: Sort and analyze the design documents of related products, extract the texts and store them in the product design text library;

[0053] S2: Use the neural network model to extract the structural entities of the products in the relevant text, classify the product structure, extract the relevant design structure knowledge, and store it in the product design structure database;

[0054] S3: According to the product design structure database established in step S2, a behavior chain between product structures is established, and the product behavior flow is stored in the product design behavior database;

[0055] S4: Based on the product design structure database and product design behavior database established in steps S2 and S3, using structure-behavior and behavior-function mapping, a design function is generated according to the behavior chain between product structures, and stored in the product design function database;

[0056] S5: search the design function database and extract the structural solution set corresponding to the design function using the relevant mapping;

[0057] S6: For coupled design functions, the behavior flow constraint relationship in the behavior database is used to solve the related structures. If there is a behavior chain between the corresponding structures, the corresponding design structure is extracted.

[0058] Embodiment 2:

[0059] As an embodiment provided by the present invention, preferably, in order to solve the problems of insufficient objectivity of design solutions in the field of product concept design, low efficiency of design knowledge reuse, inability to mine implicit design knowledge, etc., the present invention provides a method for constructing a design knowledge graph based on the FBS concept design model, which is specifically implemented on the patent document of the AGV car, and specifically includes the following steps:

[0060] Step 1: construct a design knowledge carrier acquisition module, obtain relevant design documents through a web crawler, pre-process the design document data, and establish a top-level design knowledge carrier database; as an embodiment provided by the present invention, preferably, step 1 specifically includes:

[0061] Crawl patent documents from major patent websites based on the Scrapy crawler framework;

[0062] Batch extract text content from patent documents in PDF format and convert the data to be processed into text type;

[0063] Remove stop words from the text based on the stop word list of Harbin Institute of Technology to eliminate the impact of noise;

[0064] Step 2: construct an expression module of design knowledge, and establish a design knowledge graph based on the FBS design model and the expression structure of SRO; as an embodiment provided by the present invention, preferably, step 2 specifically includes:

[0065] Regarding knowledge definition, there are three ontologies: design function, design behavior, and design structure;

[0066] Define the inclusion and connection relationships between design structures;

[0067] Define the behavior flow of the design structure based on the black box model;

[0068] According to the transfer of behavior flow, the behavior chain between design structures is constructed, such as Figure 2 As shown, the behavior chain is divided into sequential behavior chain, divergent behavior chain, converging behavior chain, and selective behavior chain;

[0069] Store the above definitions in a graph database;

[0070] Step 3: extracting design knowledge entities from the design document and mapping them to the corresponding ontology, classifying the attributes of knowledge representation for the entities, and saving the classified entities in the form of triples in the graph database; as an embodiment provided by the present invention, preferably, step 3 specifically includes:

[0071] Pre-train the relevant documents of the product based on Bert to obtain word vector encoding;

[0072] For the relevant documents of the product, the feature vector encoding of the long text is performed based on Bi LSTM. The model structure is as follows Figure 3 As shown;

[0073] The obtained word vector encoding and feature vector encoding are decoded based on the head and tail pointers to extract the relation triples in the text. The model structure is as follows Figure 4 As shown, the obtained structural entity triples are saved in the graph database;

[0074] Step 4: construct a knowledge generation module to generate functional descriptions of components included in the product from relevant descriptions of the product: As an embodiment provided by the present invention, preferably, step 4 specifically includes:

[0075] The model used in the knowledge generation module is Transformer, and its structure is as follows Figure 5 As shown;

[0076] Step 5: Build a knowledge application module to realize knowledge retrieval and reuse in the product concept design process;

[0077] Step 5 specifically includes:

[0078] Divide the product into functional modules, such as Figure 6As shown in the figure, the functional modules of the AGV are divided based on the behavior flow;

[0079] according to Figure 7 The flowchart shown retrieves the design structure required in the feasible design solution;

[0080] Step 5: Build a knowledge application module to realize knowledge retrieval and reuse in the product concept design process.

[0081] As an embodiment provided by the present invention, preferably, in the step 2, the specific process of constructing the expression module of design knowledge and establishing the design knowledge graph is as follows:

[0082] Based on the FBS design model, the design product is analyzed from top to bottom in terms of design function, design behavior, and design structure;

[0083] The expression structure based on SRO uses the form of triples as the representation framework of knowledge, where both the subject and the object are knowledge ontologies, and the triples are: subject, relationship, object;

[0084] The ontology-based representation framework defines ontology types and relationship types in the design knowledge graph, namely, FBS-based types.

[0085] As an embodiment provided by the present invention, preferably, in the step 2, the ontology types defined include: design function ontology, design behavior ontology, design structure ontology, and the relationship type definition includes: inclusion relationship and constraint relationship between design structures, and input and output relationship between design structure and behavior flow.

[0086] As an embodiment provided by the present invention, preferably, in step 3, the specific steps of saving the classified entities in the form of triples are:

[0087] Step S31: Perform named entity recognition on the design document, using the BERT model, Bi LSTM model, and head and tail pointer network to pass through the feature encoding layer, sequence encoding layer, and label decoding layer respectively to identify the component names and corresponding attribute information and behavior information contained in the design document;

[0088] Step S32: Perform relationship judgment on all identified entities, use the Casel model based on pre-defined relationship types to determine all entities corresponding to the subject and object under a specified relationship type, and thus extract corresponding triples;

[0089] Step S33: Based on the similarity of character vectors, the same entities are merged, the text content and attribute information of the entity are vectorized, and the cosine similarity between the two is calculated. If the similarity value is higher than a threshold, the entities are considered to be the same entity;

[0090] Step S34: Based on the triples, the nodes of the knowledge graph represent the knowledge entities, the edges represent the relationship types, and they are saved in the graph database.

[0091] As an embodiment provided by the present invention, preferably, in the step three, the method for named entity recognition is: pre-training a word vector encoder based on the BERT model, using BiLSTM to encode long texts, and using a head and tail pointer network for decoding.

[0092] As an embodiment provided by the present invention, preferably, in the step three, the BERT model is a pre-trained model for extracting text entities, and the BERT model is formed by stacking Enocder modules in multiple Transformer model structures; BiLSTM is a bidirectional long short-term memory network, which is composed of a forget gate, an input gate, and an output gate, and the calculation formula is:

[0093] Step S301: Calculate the forget gate, multiply the input text vector by the weight matrix, add it to the bias vector, process it through the sigmoid function, and calculate a vector representing the probability:

[0094] p t =sig(W i g[l t-1 ,x t ]+a i )

[0095] Where: x t represents the input at the current time t, l t-1 represents the hidden state vector at time t-1, sig is the sigmoid function; W i and a i Represents the weight matrix and bias vector of the forget gate, vector p t Represents probability, vector p t It is 0 or 1, 0 means forget everything, 1 means keep everything;

[0096] Step S302: Calculate the input gate. The first step is to receive the current input text vector and the hidden state information of the previous moment. After processing with the sigmoid function, a vector representing the probability value is obtained. If the probability value of the probability vector is 0, the information is not updated. If the probability value is 1, the information is updated. The second step is to concatenate the received current input text vector with the hidden state information received at the previous moment, process the information with the tanh function to obtain a short-term memory unit, and then jointly update the short-term memory unit and the long-term memory unit to obtain a new memory unit. The calculation formula is:

[0097] d t =sig(W i g[l t-1 ,xt ]+a n )

[0098]

[0099] Where: x t represents the input at the current time t, W n and a n Represents the weight matrix and bias vector of the input gate; represents the short-term memory unit at time t; S t represents the long-term memory unit at time t, and th represents the tanh function;

[0100] Step S303: Calculate the output gate. The output gate receives the current input text vector and the hidden state information of the previous moment. After sigmoid processing, a vector representing the probability value is obtained, indicating which information in the long-term memory unit needs to be output; then the long-term memory unit is processed using the tanh function, and finally the vector is summed with O. t Point multiplication outputs a final result, the calculation formula is:

[0101] G t =sig(W O g[l t-1 ,x t ]+a f )

[0102] h t =G t gth(C t )

[0103] Where: W O and a f Represents the weight matrix and bias vector of the output gate; l t Represents the hidden state information at time t.

[0104] As an embodiment provided by the present invention, preferably, in step 3, the method for performing relationship judgment on all identified entities is:

[0105] Step S3001: Based on the Casrel model, a multi-layer head-tail pointer network is constructed to add a self-attention mechanism to fuse the head entity information and the sequence element information. The vector representing the head entity is mapped to two different vector spaces to generate M and N. The sequence element information is mapped to the query vector space to generate O. Then, the scaled dot product attention model is used to calculate the correlation coefficient between O and M. The calculation formula is:

[0106]

[0107] Where: represents the tth head entity identified, Represents the tth element in the sequence information;

[0108] Step S3002: Use the distribution function y to process vectors O and M to obtain an energy score b. Use the calculated energy score to characterize the correlation between the sequence element and the head entity. The calculation expression is as follows:

[0109]

[0110] Where: Indicates the dimensions of O and M;

[0111] Step S3003: The energy score b is multiplied by the vector representing the head entity and then concatenated with the vector representing the sequence element to generate a new vector Characterize the column elements, and then For projection transformation, the expression is as follows:

[0112] w=soft(b)

[0113]

[0114] Where: w represents the attention weight of the sequence element and the head entity, N sb is the vector representing the head entity, The vector representing the head entity is processed and concatenated with the sequence element vector to form a new head entity representation vector. P represents the head entity representation vector Mapping is performed to the vector space of the relational dimension, and F represents the probability value of the correlation between the sequence element vector and the head entity vector in the relational dimension.

[0115] As an embodiment provided by the present invention, preferably, in the step three, the Casel model is used based on a predefined relationship type to determine all entities corresponding to the subject and object under a specified relationship type, and after the corresponding triples are extracted, the Neo4j graph database is used to store the triples.

[0116] As an embodiment provided by the present invention, preferably, in the step 4, the specific process of generating the functional description of the parts included in the product from the relevant description of the product is:

[0117] Through the encoder-decoder model, the design document extracted in step 1 is received to output the functional description of each component, and the relationship triple of function and structure is obtained;

[0118] The functional entities and structural entities are fused separately;

[0119] Retrieve the corresponding structural entities in the graph database and save the relationship triples between function and structure into the graph database.

[0120] As an embodiment provided by the present invention, preferably, in step 5, the specific process of realizing knowledge retrieval and reuse in the product concept design process is:

[0121] Retrieve functional entities and find required functions;

[0122] According to the mapping relationship between function and structure, a set of design structures that can realize the corresponding functions is retrieved;

[0123] For two interacting functions, structures with connection paths in the corresponding two design structure sets are retrieved, and the collection of design structures is obtained.

[0124] A method for constructing a domain knowledge graph based on an FBS concept design model. The method uses the FBS concept design model to back-analyze design structures, design behaviors, design functions, and characteristic attributes related to product concept design from design documents. By pre-defining the relationship types between these design elements, a design knowledge graph is constructed to provide stable and systematic design knowledge support for the product development process, thereby improving product development efficiency and reducing product development costs. The method improves the Casel model, constructs a multi-layer head-tail pointer network, adds a self-attention mechanism to fuse head entity information and sequence element information, and can solve the problem of a single entity corresponding to multiple entities in a document, thereby improving the quality of knowledge graph generation.

[0125] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0126] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for constructing a domain knowledge graph based on the FBS conceptual design model, characterized in that: The following steps are involved: Step 1: Build a design knowledge carrier acquisition module, obtain relevant design documents through web crawlers, pre-process the design document data, and establish a top-level design knowledge carrier database; Step 2: Construct the expression module of design knowledge and establish the design knowledge graph based on the FBS design model and the expression structure of SRO; Step 3: Extract design knowledge entities from design documents and map them to corresponding ontologies, classify the attributes of knowledge representation for entities, and save the classified entities in the form of triples in the graph database; Step 4: Build a knowledge generation module to generate functional descriptions of the components contained in the product from the relevant descriptions of the product: Step 5: Build a knowledge application module to realize knowledge retrieval and reuse in the product concept design process.

2. According to claim 1, a method for constructing a domain knowledge graph based on the FBS conceptual design model is characterized in that: In step 2, the specific process of constructing the expression module of design knowledge and establishing the design knowledge graph is as follows: Based on the FBS design model, the design product is analyzed from top to bottom in terms of design function, design behavior, and design structure; The expression structure based on SRO uses the form of triples as the representation framework of knowledge, where both the subject and the object are knowledge ontologies, and the triples are: subject, relationship, object; The ontology-based representation framework defines ontology types and relationship types in the design knowledge graph, namely, FBS-based types.

3. According to claim 2, a method for constructing a domain knowledge graph based on the FBS conceptual design model is characterized in that: In the step 2, the ontology types defined include: design function ontology, design behavior ontology, design structure ontology, and the relationship type definitions include: inclusion relationship and constraint relationship between design structures, and input and output relationship between design structure and behavior flow.

4. According to claim 1, a method for constructing a domain knowledge graph based on the FBS conceptual design model is characterized in that: In step 3, the specific steps of saving the classified entities in the form of triples are: Step S31: Perform named entity recognition on the design document, using the BERT model, the BiLSTM model, and the head-tail pointer network to respectively pass through the feature encoding layer, the sequence encoding layer, and the label decoding layer to identify the component names and corresponding attribute information and behavior information contained in the design document; Step S32: Perform relationship judgment on all identified entities, use the Casrel model based on pre-defined relationship types, determine all entities corresponding to the subject and object under a specified relationship type, and extract corresponding triples; Step S33: Based on the similarity of character vectors, the same entities are merged, the text content and attribute information of the entity are vectorized, and the cosine similarity between the two is calculated. If the similarity value is higher than a threshold, the entities are considered to be the same entity; Step S34: Based on the triples, the nodes of the knowledge graph represent the knowledge entities, the edges represent the relationship types, and they are saved in the graph database.

5. According to claim 4, a method for constructing a domain knowledge graph based on the FBS conceptual design model is characterized in that: In the step three, the method of named entity recognition is: pre-training a word vector encoder based on the BERT model, using BiLSTM to encode long texts, and using a head-tail pointer network for decoding.

6. According to claim 5, a method for constructing a domain knowledge graph based on the FBS conceptual design model is characterized in that: In the step 3, the BERT model is a pre-trained model for extracting text entities. The BERT model is composed of multiple Enocder modules in the Transformer model structure stacked together; BiLSTM is a bidirectional long short-term memory network, which consists of a forget gate, an input gate, and an output gate. The calculation formula is: Step S301: Calculate the forget gate, multiply the input text vector by the weight matrix, add it to the bias vector, process it through the sigmoid function, and calculate a vector representing the probability: p t =sig(W i g[l t-1 ,x t ]+a i ) Where: x t represents the input at the current time t, l t-1 represents the hidden state vector at time t-1, sig is the sigmoid function; W i and a i Represents the weight matrix and bias vector of the forget gate, vector p t Represents probability, vector p t It is 0 or 1, 0 means forget everything, 1 means keep everything; Step S302: Calculate the input gate. The first step is to receive the current input text vector and the hidden state information of the previous moment. After processing with the sigmoid function, a vector representing the probability value is obtained. If the probability value of the probability vector is 0, the information is not updated. If the probability value is 1, the information is updated. The second step is to concatenate the received current input text vector with the hidden state information received at the previous moment, process the information with the tanh function to obtain a short-term memory unit, and then jointly update the short-term memory unit and the long-term memory unit to obtain a new memory unit. The calculation formula is: d t =sig(W i g[l t-1 ,x t ]+a n ) Where: x t represents the input at the current time t, W n and a n Represents the weight matrix and bias vector of the input gate; represents the short-term memory unit at time t; S t represents the long-term memory unit at time t, and th represents the tanh function; Step S303: Calculate the output gate. The output gate receives the current input text vector and the hidden state information of the previous moment. After sigmoid processing, a vector representing the probability value is obtained, indicating which information in the long-term memory unit needs to be output; then the long-term memory unit is processed using the tanh function, and finally the vector is summed with O. t Point multiplication outputs a final result, the calculation formula is: G t =sig(W O g[l t-1 ,x t ]+a f ) h t =G t gth(C t ) Where: W O and a f Represents the weight matrix and bias vector of the output gate; l t Represents the hidden state information at time t.

7. According to claim 4, a method for constructing a domain knowledge graph based on the FBS conceptual design model is characterized in that: In step 3, the method for determining the relationship between all identified entities is as follows: Step S3001: Based on the Casrel model, a multi-layer head-tail pointer network is constructed to add a self-attention mechanism to fuse the head entity information and the sequence element information. The vector representing the head entity is mapped to two different vector spaces to generate M and N. The sequence element information is mapped to the query vector space to generate O. Then, the scaled dot product attention model is used to calculate the correlation coefficient between O and M. The calculation formula is: Where: represents the tth head entity identified, Represents the tth element in the sequence information; Step S3002: Use the distribution function y to process vectors O and M to obtain an energy score b. Use the calculated energy score to characterize the correlation between the sequence element and the head entity. The calculation expression is as follows: Where: Indicates the dimensions of O and M; Step S3003: The energy score b is multiplied by the vector representing the head entity and then concatenated with the vector representing the sequence element to generate a new vector Characterize the column elements, and then For projection transformation, the expression is as follows: w=soft(b) Where: w represents the attention weight of the sequence element and the head entity, N sb is the vector representing the head entity, The vector representing the head entity is processed and concatenated with the sequence element vector to form a new head entity representation vector. P represents the head entity representation vector Mapping is performed to the vector space of the relational dimension, and F represents the probability value of the correlation between the sequence element vector and the head entity vector in the relational dimension.

8. According to claim 7, a method for constructing a domain knowledge graph based on the FBS conceptual design model is characterized in that: In the step three, the Casrel model is used to determine all entities corresponding to the subject and object under a specified relationship type based on the pre-defined relationship type, and after the corresponding triples are extracted, the Neo4j graph database is used to store the triples.

9. The method for constructing a domain knowledge graph based on the FBS conceptual design model according to claim 1, characterized in that: In step 4, the specific process of generating the functional description of the components included in the product from the relevant description of the product is as follows: Through the encoder-decoder model, the design document extracted in step 1 is received to output the functional description of each component, and the relationship triple of function and structure is obtained; The functional entities and structural entities are fused separately; Retrieve the corresponding structural entities in the graph database and save the relationship triples between function and structure into the graph database.

10. The method for constructing a domain knowledge graph based on the FBS conceptual design model according to claim 1, characterized in that: In step 5, the specific process of realizing knowledge retrieval and reuse in the product concept design process is as follows: Retrieve functional entities and find the required functions; According to the mapping relationship between function and structure, a set of design structures that can realize the corresponding functions is retrieved; For two interacting functions, structures with connection paths in the corresponding two design structure sets are retrieved, and the collection of design structures is obtained.

Images