A Domain Knowledge Graph Framework for Product Concept Design and Its Construction Method

Abstract

The present invention discloses a domain knowledge graph framework for product conceptual design and a construction method thereof. The domain knowledge graph framework is constructed in a top-down manner. Centering on the design unit class, design feature classes, design constraint classes, design activity classes, etc. that are closely related to product conceptual design are defined, and the relationships between various classes are defined to form a domain knowledge ontology for product conceptual design, that is, the schema layer of the knowledge graph. Then, in a bottom-up manner, after operations such as entity recognition, relationship extraction, knowledge fusion, data storage, and visualization on multi-source heterogeneous design data, the construction of the data layer is completed to form a domain knowledge graph for product conceptual design. The present invention not only integrates design principle knowledge and design process knowledge, solves the problem of insufficient knowledge coverage in existing research, but also overcomes the technical limitation that only natural language technology cannot mine implicit knowledge, and at the same time ensures the quality of the knowledge graph.

Description

Technical Field

[0001] The present invention belongs to the technical field of knowledge graphs, and particularly relates to a domain knowledge graph framework for product conceptual design and a construction method thereof. Background Art

[0002] Product conceptual design is a knowledge-intensive activity, and designers' rich design knowledge is required as support in all stages of product design, such as requirement analysis, conceptual solution generation, and detailed design. Since the knowledge related to product design is scattered in design documents, design logs, designs, and 3D models, these multi-source heterogeneous design data increase the difficulty of design knowledge management and cannot provide sufficient support for product conceptual design.

[0003] How to effectively manage and utilize design knowledge is one of the important ways and key technologies to improve the efficiency of product conceptual design, and it is also a current research hotspot. Chinese invention patent CN103995886B (a method for constructing a design atlas based on an RFPC conceptual design framework) discloses a method for constructing a design atlas based on an RFPC conceptual design framework to effectively store and analyze design knowledge. It obtains design knowledge data from text data, and then maps the design knowledge to the requirement - function - principle solution - feature mark RFPC model that uses the design knowledge to assist product conceptual design. Through natural language processing technology and dependency syntactic analysis, the association relationships between the design knowledge elements are characterized. A knowledge storage and management solution for the design knowledge is provided using a graph database, and the retrieval, reuse of knowledge during the conceptual design process, and support for the generation of conceptual design solutions can be realized more intuitively. Although this method can achieve the representation and management of design knowledge, there are still the following problems:

[0004] (1) The coverage of design knowledge is insufficient. The above research mainly uses knowledge graphs to manage design principle knowledge such as product functions and requirements, lacking a complete representation of the entire design process. Information such as optimization iteration and design intent is included in the design process, but a large amount of design process knowledge has not been effectively managed and utilized, resulting in waste of data resources and thus unable to effectively support product conceptual design.

[0005] (2) It is impossible to mine implicit design knowledge. In the process of constructing a design domain knowledge graph, the above research overemphasizes the use of natural language processing technology and ignores the implicit knowledge contained in design cases and designers' experiences, resulting in low-quality and impractical knowledge graphs. Summary of the Invention

[0006] To overcome the deficiencies of the prior art, the present invention provides a domain knowledge graph framework for product conceptual design and a construction method thereof. The domain knowledge graph framework is constructed in a top-down manner, with the design unit class as the center, defining design feature classes, design constraint classes, design activity classes, etc. that are closely related to product conceptual design, and defining the relationships between various classes to form a domain knowledge ontology for product conceptual design, that is, the schema layer of the knowledge graph. Then, in a bottom-up manner, after performing operations such as entity recognition, relation extraction, knowledge fusion, data storage, and visualization on multi-source heterogeneous design data, the construction of the data layer is completed, forming a domain knowledge graph for product conceptual design. The present invention not only integrates design principle knowledge and design process knowledge, solves the problem of insufficient knowledge coverage in existing research, but also uses FBS and DM to perform structured representation and relation extraction on implicit knowledge, overcomes the technical limitation that only natural language technology cannot mine implicit knowledge, and at the same time ensures the quality of the knowledge graph. A design knowledge retrieval and recommendation system can be developed based on the present invention to support product conceptual design, thereby improving the efficiency of design knowledge reuse.

[0007] The technical solution adopted by the present invention to solve its technical problems includes the following steps:

[0008] Step 1: Construct the schema layer of the knowledge graph;

[0009] Adopt a top-down approach to hierarchically analyze relevant design units, design features, design constraints, and design activity concepts in the field of product conceptual design, define concepts, attributes, and relationships, form a domain knowledge ontology for product conceptual design, and complete the construction of the schema layer of the knowledge graph;

[0010] The domain knowledge ontology for product conceptual design is composed of a triple description form, denoted as ProductDesign Ontology = {Entity, Attribute, Relation}, where: Entity is the set of entity-related concepts, used to represent the set of objective entities in the field of product conceptual design; Attribute is the set of attribute-related concepts; Relation represents the association relationship between entities and attributes;

[0011] Step 2: Construct the data layer of the knowledge graph;

[0012] Adopt a bottom-up approach. After performing operations such as entity recognition, relation extraction, knowledge fusion, data storage, and visualization on multi-source heterogeneous design data, the construction of the data layer is completed, forming a domain knowledge graph for product conceptual design. The specific process is as follows:

[0013] (1) Entity recognition: Use the BERT-BiLSTM-CRF model to recognize entities in text information to form an entity dictionary;

[0014] (2) Relationship extraction: Adopt a template-based method to extract relationships in text information;

[0015] Use the "Function - Behavior - Structure" model FBS to model the principle knowledge in product cases, analyze the mutual correlation relationships among product function - behavior - structure, and form <entity, relationship, entity> triples;

[0016] If there is an interaction between substances, energy, and signals between two components of a product, there is a functional correlation relationship between these two components; if there is a logical causal correlation between the behaviors of two components of a product, there is a behavioral correlation relationship between these two components; if there is a spatial correlation or surface contact between the structures of two components of a product, there is a structural correlation relationship between these two components;

[0017] Use the Design Matrix DM to model the design process knowledge, describe the serial, parallel, and coupling relationships between design tasks; extract the correlation relationships between the "Function - Behavior - Structure" dimension and design tasks in product cases, and form <entity, relationship, entity> triples;

[0018] (3) Knowledge fusion: Fuse knowledge entities according to string similarity and ontology similarity;

[0019] The calculation formula for string similarity is:

[0020]

[0021] Among them, λ is the similarity adjustment coefficient; A represents the number of identical characters in strings i and j; B represents the characters that exist in string i but not in string j; C represents the characters that do not exist in string i but exist in string j; D represents the total number of characters in strings i and j;

[0022] When determining the similarity of two entities through strings, the following judgment rules are used to determine the knowledge fusion result: (a) When the parent nodes and child nodes of two entities are the same, the two entities are the same; (b) When the parent nodes of two entities are the same and the child nodes are different, the two entities are different;

[0023] (4) Data storage and visualization: Use a graph database to store product conceptual design knowledge triples and the constructed knowledge ontology, and form a one-to-many "concept - example" relationship.

[0024] A knowledge graph framework for the field of product concept design, including four ontology classes: design unit class, design feature class, design constraint class and design activity class; with the design unit class ontology as the center, the relationship between the design feature class, design constraint class and design activity class and the design unit is defined to form a knowledge graph framework structure in the field of product concept design;

[0025] The design unit ontology is the basis for constructing the knowledge graph in the field of product concept design. It describes the hierarchical characteristics of the product, including product class, system class and part class; the design feature ontology includes functional characteristics, behavioral characteristics, and structural characteristics; the design constraint ontology includes functional constraints, behavioral constraints, structural constraints, rule constraints, spatial constraints and environmental constraints; the design activity ontology is used to describe the product design process, and the design activity class includes design cycle, design problem and design iteration.

[0026] Preferably, the graph database is Neo4j.

[0027] Preferably, the domain knowledge ontology is constructed using protégé software, and the Entity in the ontology is mapped to a node in a Neo4j graph database, the Attribute is mapped to a node property, and the Relation is mapped to an edge in Neo4j.

[0028] The beneficial effects of the present invention are as follows:

[0029] (1) The present invention defines relevant concepts, attributes and their relationships from top to bottom through hierarchical analysis of design units, design attributes, design constraints and design activities that are closely related to product concept design, and constructs a knowledge ontology that can comprehensively describe the product concept design process, namely, a design knowledge graph framework. The constructed knowledge graph includes not only design principle knowledge such as design constraints and design attributes, but also design process knowledge, and the knowledge coverage is more comprehensive.

[0030] (2) The present invention utilizes the implicit knowledge contained in the product cases of the FBS model and the DM model for structured modeling, which can mine the correlation between the "function-behavior-structure" dimension in the product cases and the design tasks. At the same time, combined with expert annotation, the quality of the knowledge graph is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flow chart of the method of the present invention.

[0032] Figure 2 It is a schematic diagram of the product concept design knowledge ontology expression model in the method of the present invention.

[0033] Figure 3 It is a schematic diagram of constructing a knowledge ontology in a protégé in the method of the present invention.

[0034] Figure 4 It is a schematic diagram of the network structure of the BERT-BiLSTM-CRF model in the method of the present invention.

[0035] Figure 5 It is a schematic diagram of relation extraction in the method of the present invention.

[0036] Figure 6 It is a schematic diagram of case principle knowledge modeling in the method of the present invention.

[0037] Figure 7 It is a schematic diagram of design process knowledge modeling in the method of the present invention.

[0038] Figure 8 It is a schematic diagram of knowledge fusion in the method of the present invention.

[0039] Figure 9 It is a schematic diagram of using some nodes of the Neo4j graph database in the method of the present invention. Detailed implementation manners

[0040] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0041] Aiming at the problems of insufficient knowledge coverage and inability to mine implicit design knowledge existing in the existing research, referring to Figures 1 - 9 , the present invention provides a method for constructing a domain knowledge graph for product conceptual design, which specifically includes the following steps:

[0042] Step 1: Construct the schema layer of the knowledge graph

[0043] Adopting a top-down approach, hierarchically analyze relevant design units, design features, design constraints, and design activity concepts in the field of product conceptual design, define concepts, attributes, and relationships, form a domain knowledge ontology for product conceptual design, and complete the construction of the schema layer of the knowledge graph. The domain knowledge ontology for product conceptual design is composed of a triple description form, denoted as Product Design Ontology = {Entity, Attribute, Relation}, where: Entity is a set of entity-related concepts, used to represent the set of objective entities in the field of product conceptual design; Attribute is a set of attribute-related concepts; Relation represents the association relationship between entities and attributes. Use the protégé software to construct the ontology, map the Entity in the ontology to the nodes of Neo4j, the Attribute to the node attributes, and the Relation to the edges in Neo4j.

[0044] The ontology of the design unit class is the basis for constructing the knowledge graph in the field of product conceptual design, mainly describing the hierarchical characteristics of products, including product classes, system classes, and part classes. The design feature class mainly includes functional feature classes, behavioral feature classes, and structural feature classes. The attributes of the design constraint class mainly include functional constraints, behavioral constraints, structural constraints, rule constraints, spatial constraints, and environmental constraints, etc. The design activity class is mainly used to describe the product design process, and closely related to the design activity class are the design requirement class, design task class, designer class, design tool class, etc. The attributes of design activities mainly include design cycle, design problems, design iterations, etc. The relationships between various classes are Subclass_of, Has_feature, Acts_on, Executes, Worksfor, Supports, Makes up, Achieves, Comes up with, Makes, Has_behavior, Is based on, Produces, etc.

[0045] The meanings and examples of the above classes and relationships are specifically shown in Tables 1 and 2 as follows:

[0046] Table 1

[0047]

[0048]

[0049] Table 2

[0050] Relationship Meaning Example Subclass_of Is a subclass of Function Subclass_of Design Feature Has_feature Has a feature Part Has_feature Design Feature Acts_on Acts on Design Constraint Acts_on Part Executes Executes Designer Executes Design Task Works for Works for Designer Works for Department Supports Supports Design Tool Supports Design Task Makes up Makes up Design Task Makes up Design Activity Achieves Achieves Design Activity Achieves Design Requirement Comes up with Comes up with User Comes up with Design Requirement Makes Makes User Makes Design Evaluation Has_behavior Has a behavior Structure Has_behavior Behavior Is based on Is based on Behavior Is based on Structure Produces Produces Function Produces Behavior …… …… ……

[0051] Step 2: Construct the data layer of the knowledge graph

[0052] Adopt a bottom-up approach. After performing entity recognition, relationship extraction, knowledge fusion, data storage, and visualization on multi-source heterogeneous design data, complete the construction of the data layer to form a domain knowledge graph for product conceptual design. The specific process is as follows:

[0053] (1) Entity recognition. Use the BERT-BiLSTM-CRF model to identify entities in text information and form an entity dictionary.

[0054] (2) Relationship extraction. A template-based method is used to extract relationships in text information. At the same time, the "Function-Behavior-Structure (FBS)" model is adopted to model the principle knowledge in product cases, and analyze the mutual correlation relationships among product function, behavior, and structure. For example, if there is an interaction of matter, energy, and signals between two components of a product, there is a functional correlation relationship between these two components. If there is a logical causal correlation between the behaviors of two components of a product, there is a behavioral correlation relationship between these two components. If there is a spatial correlation, surface contact, etc. between the structures of two components of a product, there is a structural correlation relationship between these two components. The Design Matrix (DM) is used to model the design process knowledge and describe the serial, parallel, and coupling relationships between design tasks. Extract the correlation relationships between the "Function-Behavior-Structure" dimension and design tasks in product cases, and form <entity, relationship, entity> triples.

[0055] (3) Knowledge fusion. Knowledge entities are fused according to string similarity and ontology similarity. The calculation formula for string similarity is:

[0056]

[0057] λ is the similarity adjustment coefficient; A represents the number of identical characters in strings i and j; B represents the characters that exist in string i but not in string j; C represents the characters that do not exist in string i but exist in string j; D represents the total number of characters in strings i and j.

[0058] When two entities are found to be similar through string calculation, the following judgment rules are also needed to determine the knowledge fusion result: (a) When the parent nodes and child nodes of the two entities are the same, the two entities are the same; (b) When the parent nodes of the two entities are the same but the child nodes are different, the two entities are different.

[0059] (4) Data storage and visualization. The Neo4j graph database is used to store the product conceptual design knowledge triples and the constructed knowledge ontology, and form a one-to-many "concept - example" relationship.

Claims

1. A method for constructing a domain knowledge graph framework for product conceptual design, characterized in that It includes the following steps: Step 1: Construct the schema layer of the knowledge graph; Adopt a top-down approach to hierarchically analyze relevant design units, design features, design constraints, and design activity concepts in the field of product conceptual design, define concepts, attributes, and relationships, form a domain knowledge ontology for product conceptual design, and complete the construction of the schema layer of the knowledge graph; The domain knowledge ontology for product conceptual design is composed of a triple description form, denoted as ProductDesign Ontology = {Entity, Attribute, Relation}, where: Entity is the set of entity-related concepts, used to represent the set of objective entities in the field of product conceptual design; Attribute is the set of attribute-related concepts; Relation represents the association relationship between entities and attributes; Step 2: Construct the data layer of the knowledge graph; Adopt a bottom-up approach. After entity recognition, relation extraction, knowledge fusion, data storage, and visualization operations on multi-source heterogeneous design data, complete the construction of the data layer to form a domain knowledge graph for product conceptual design. The specific process is as follows: (1) Entity recognition: Use the BERT-BiLSTM-CRF model to recognize entities in text information and form an entity dictionary; (2) Relation extraction: Use a template-based method to extract relationships in text information; Adopt the "Function - Behavior - Structure" model FBS to model the principle knowledge in product cases, analyze the mutual association relationships among product function - behavior - structure, and form <entity, relation, entity> triples; If there is an interaction of matter, energy, and signals between two components of a product, there is a functional association relationship between these two components; if there is a logical causal association between the behaviors of two components of a product, there is a behavioral association relationship between these two components; if there is a spatial association or surface contact between the structures of two components of a product, there is a structural association relationship between these two components; Use the design matrix DM to model the design process knowledge, describe the serial, parallel, and coupling relationships between design tasks; extract the association relationships between the "Function - Behavior - Structure" dimension and design tasks in product cases to form <entity, relation, entity> triples; (3) Knowledge fusion: Fuse knowledge entities according to string similarity and ontology similarity; The calculation formula for string similarity is: where λ is the similarity adjustment coefficient; A represents the number of identical characters in strings i and j; B represents the characters that exist in string i but not in string j; C represents the characters that do not exist in string i but exist in string j; D represents the total number of characters in strings i and j; When two entities are similar calculated by strings, the following judgment rules are used to determine the knowledge fusion result: (a) When the parent nodes and child nodes of the two entities are the same, the two entities are the same; (b) When the parent nodes of the two entities are the same and the child nodes are different, the two entities are different; (4) Data storage and visualization: Use a graph database to store the product conceptual design knowledge triples and the constructed knowledge ontology, and form a one-to-many "concept - example" relationship.

2. A knowledge graph framework constructed by using the method described in claim 1, characterized in that, It includes 4 ontology classes: design unit class, design feature class, design constraint class, and design activity class; with the design unit class ontology as the center, define the relationships between the design feature class, design constraint class, and design activity class and the design unit to form the framework structure of the product conceptual design domain knowledge graph; The design unit class ontology is the basis for constructing the product conceptual design domain knowledge graph, describing the hierarchical characteristics of the product, including product class, system class, and part class; The design feature class ontology includes functional features, behavioral features, and structural features; The design constraint class ontology includes functional constraints, behavioral constraints, structural constraints, rule constraints, spatial constraints, and environmental constraints; the design activity class ontology is used to describe the product design process, and the design activity class includes design cycle, design problem, and design iteration.

3. A knowledge graph framework according to claim 2, characterized in that, The graph database is Neo4j.

4. A knowledge graph framework according to claim 2, wherein The construction of the domain knowledge ontology uses protégé software, mapping the Entity in the ontology to the nodes of the Neo4j graph database, the Attribute to the node attributes, and the Relation to the edges in Neo4j.

Images