Knowledge Graph-Based Assembly Resource Knowledge Reuse towards Complex Product Assembly Process

Abstract

Assembly process designers typically confront the challenge of seeking information out of large volumes of non-structured files with a view to supporting the decision-making to be made. It is a leading concern that embedding data in text documents can hardly be retrieved semantically in order to facilitate decision-making with timely support. For tackling this gap, we propose in this paper a knowledge graph-based approach used to merge and retrieve information decided to be relevant within an engineering context. The proposed approach is to establish a multidimensional integrated assembly resource knowledge graph (ARKG) based on the structure of function-structure-assembly procedure-assembly resource, and this multidimensional integrated structure can well accomplish the retrieval of related knowledge. The upper semantic framework of ARKG is established by the assembly resource ontology model, which is a semantic-type framework involving multiple domains of knowledge to create instantiated data reflecting the full profile of the assembly resource for obtaining structured data of ARKG while avoiding the data redundancy problem. The ARKG method is validated through assembly scenario of the aircraft, and the results show the effectiveness and accuracy of the ARKG used by the assembly process designer in the assembly process design phase for retrieving the target knowledge of the assembly resources.

1. Introduction

Due to rapid changes in technology, iterative variations in market demand, etc., manufacturing enterprises are faced with scenarios wherein they are frequently changing their production process plans and reconfiguring/building assembly lines to manufacture new products [1,2]. Manufacturing enterprises are therefore expected to have the ability to adapt to new changes. The technical personnel involved need a certain amount of domain-specific information or knowledge to understand the production of the new product [3]. Li et al. [4] conclude that information is useful (1) as a “memorable expansion” for specific engineers and to facilitate the sharing of information among them, (2) to probe for substitutes that already exist in the database, (3) to draw on primitive data to study and analyse the fundamentals behind decisions, and (4) to seek out past cases to reuse knowledge when dealing with similar products or problems. However, the utilisation of existing knowledge from manufacturing enterprises in the manufacturing of new products is severely inadequate [5]. The majority of available information sources for engineering are in the form of non-structured, non-formal, text-based documents [4], and relevant personnel often devote considerable amounts of time to retrieving related information from databases for similar projects [5].

In order to accommodate product variations, it is often necessary to manually search for the required manufacturing processes and associated resource constraints. Due to the fact that datasets of product, process and resource information are severally available in specific data structures without effective coupling among them [6], the reconfiguration of manufacturing resources remains subjective and relies heavily on the knowledge and experience of engineers. Manufacturing process and resource modelling remains a competitive and sophisticated mission in view of the dependency on knowledge sharing, integration and reusability in engineering processes. Within a framework of a potential solution to this task being the ability to effectively associate digital descriptions of the product structure underlying user requirements for product functionality and the required assembly processes to manufacturing resources, the knowledge base resulting from such integrated descriptions could potentially boost the efficiency of the engineering process.

The focus of this paper is on a coupling of the “function-structure-process-resource” data model for mapping data across different concepts by combining an ontology-based knowledge modelling framework with knowledge graph. The engineering knowledge representation based on knowledge graph allows the dynamic modelling of products and the mapping to the required assembly processes and resources. Being framed as a network of entities and their semantic relationships [7,8], knowledge graphs serve a wide range of purposes in expressing knowledge in different domains [9,10,11,12,13,14]. Being anchored in graph theory, knowledge graphs have the merit of being able to incorporate more complicated relationships than in conventional relational databases, navigate through correlated content with greater speed, and offer users the ability to visualize the links between different items of information by making sense of them. The potential for employing knowledge graphs to serve as a means of representation of text-based knowledge for decision support in engineering design is promising.

The rest of this paper is organized according to the following structure. Section 2 presents some investigative studies related to knowledge graphs that illustrate the possibility of applying knowledge graph theoretical techniques to the knowledge reuse of assembly resources. In Section 3 the proposed approach is described in detail. A knowledge graph ARKG based on the assembly resource ontology model is established with Section 4. A case study of an aircraft is presented in Section 5 to demonstrate the proposed approach. In Section 6, this work is concluded and some potential further efforts are proposed.

2. Related Work

The Knowledge Graph arose within Google’s user interface in 2012 with the aim of upgrading the capacity of the searching engine and customer satisfaction. It is, in nature, a structured semantic knowledge base intended for symbolic representation of real-world concepts and their relationships [15]. Knowledge graphs are a subject of researchers’ attention and investigation owing to the graphs’ effectiveness and efficiency in organizing and representing enormous, heterogeneous, and multiple attributes of data and knowledge reasoning.

Within engineering practice, knowledge graphs typically are deployed to make it possible for various partners or platforms to share and reuse knowledge throughout the product lifecycle, representing the complex logical relationships between elements of knowledge. An example in this context is Li et al. [16] who put forward a process knowledge graph to structuralise the multitude of heterogeneous CAM models derived from and deposited in the enterprise to promote the shareability and reuse of these models in NC process planning. Zhang et al. [17] draw on graph-based theory to construct a framework for knowledge reuse that has a positive effect on the decision-making aspects associated with the newly evolved product development process. With the establishment of an ontology to formulate a concept graph that collects and classifies knowledge resources that are acquired at successive phases of the product life cycle, they subsequently made use of the concept graph to deduce an algorithm with which to navigate the knowledge related to newly developed products. Sarica et al. [18] search for available engineering knowledge in a domain by applying a technology KG, which encloses semantic aspect knowledge of the entire technological domain being defined in the International Patent Classification System, and locates the engineering concepts surrounding the domain for future design consideration and innovation. Such an approach is adopted to foster interdisciplinary knowledge findings with respect to the context of flying car design. Liang et al. [19] put forward a framework with knowledge graphs with an entity-based relationship model as a means of representing facts in interior scene design. Under their proposed framework, both spatial relationships and relationships of reliance amongst objects are arranged in a knowledge graph for the purpose of enhancing interior scene design. Zhang et al. [20] propose an approach to build metallic materials knowledge graph (MMKG) based on DBpedia and Wikipedia, using a combination of semantic distance algorithm based on direct link subgraph, semantic distance and string similarity algorithm to extract metallic material entities, and an ontology-based approach to extract attribute knowledge from the HTML tables of the corresponding Wikipedia Web pages to enrich the MMKG. Chen et al. [21] propose an ontology and Natural Language Processing (NLP)-based framework for automatic modelling of literature knowledge graphs and inference networks to facilitate efficient mining of abstract knowledge from the literature. The framework proposes a representation ontology that characterises literature’s abstract data into four knowledge elements: context, goal, solution and outcome, and automatically extracts ontology instances from the abstracts using NLP techniques. On the basis of the representation ontology, a four-space integrated knowledge graph is constructed using NLP techniques. Inference networks are then generated based on the inference mechanisms defined in the ontology model. Hao et al. [22] put forward a knowledge graph-based approach that is adopted to integrate and navigate decision-relevant information regarding engineering design. The proposed approach is derived from a decisive knowledge graph-based metamodel (mDKG), which is rooted in a compromise decision support problem (cDSP) structure’s being adopted by designers in order to enact design decisions both linguistically and mathematically. Kwon et al. [23] pose an automated method which yields knowledge graphs on behalf of STEP and QIF data, a mapping realization that consolidates STEP and QIF knowledge graphs, and rules and queries exhibiting the consolidation to make improved decisions in terms of a product’s quality guarantee. Liu et al. [24] introduce an approach to make a data representation of cognitive manufacturing using a knowledge graph with application to the scene of cyber-physical production system (CPPS). With digital thread embedded in manufactured processual data, a multi-layered knowledge graph about manufacturing is erected, involving equipment sensed data, manufacturing processual oriented information and commercial logical related information. Taking advantage of the developed knowledge graph, a cognition driven approach, i.e., a combined perceptual-cognitive dual system, is put forward to implement perceptual analysis and cognitive decision regarding the assignment of resources in the production procedure. Bharadwaj et al. [25] present a methodology for the construction of multiple relationships and multiple levels of knowledge graphs to abstract out information embedded in three-dimensional product model data in order to establish assembly-subassembly-part and shape similarity relationships. Such a method relies on a conjunction of the usage of three-dimensional model metadata and the utilization of assembly-part hierarchies and three-dimensional shape-based clustering to construct its graphs. Lyu et al. [26] present a Knowledge as a Service (KaaS) model with an emphasis on cognitive abilities through proactive exploitation of vast amounts of knowledge, wherein the Industrial Knowledge Graph (IKG) as a core yields situation-aware and initiative-serving returns with the aim of optimizing service-ability and productive ability. In addition, a generalized crowd-sourcing method is put forward for the ongoing exploitation and enrichment of the IKG. Zhou et al. [27] put forward a solution for the generation and evaluation of complex component assembly processes powered by knowledge graphs. On the basis of an APKG (assembly process knowledge graph) model, a sequential knowledge graph convolutional network SKGCN model is devised for the generation of assembly process planning, which is based on decentralized graph embedding. In addition, a combination of model- and knowledge- driven assembly sequence evaluation method is to offer assembly-specific knowledge supporting the assessment approach of assembly sequence interference inspection based on point cloud assembly feature recognition. Zhou et al. [28] put forward a novel end-to-end ITIESF approach to merge table information and construct a cause-and-effect event evolving knowledge graph that is oriented to table information. Then, a novel NSGCN (Neighbourhood Sample Graph Convolutional Network)-based entity aligning method is put forward to converge intersecting knowledge graphs into a uniform knowledge repository. In addition, a translate-based Q & A (question and answer) method motivated by graph structure is conceived to cope with reason analysis and problem tracking. A multi-subject reinforcement learning (MARL) approach based on industrial knowledge graphs (IKGs) for implementing self-X cognitive manufacturing networks was introduced by Zheng et al. [29]. The IKG is developed by leveraging a large amount of artificially generated and mechanically controlled multimodal data; a graph neural network-based embedding algorithm is put forward in order to implement semantic-based self-graph solution search and task decomposition on the basis of the IKG. Li et al. [30] proposed an evolving designing method by employing knowledge graph (KG) technique and concept-knowledge (C-K) model in order to effectively support intelligent product–service system (PSS) development. The diagram-based query model and computational linguistic algorithms are carried out with four knowledge graph-assisted C-K operators under the triggering of personalized requirements to generate an innovative solution for the evolution of intelligent PSS. Yoo et al. [31] exploited and fine-tuned a pre-trained multi-linguistic BERT model to construct knowledge graphs that are free of language dependencies, and used BERT-based relationship mining model to draw new relationships to be integrated into the knowledge graph. Zhou et al. [32] put forward a Workshop Resource Knowledge Graph (WRKG) model for the integration of engineered semantic information in machining shops. An algorithm for distributive knowledge representation was formulated to extract the implied resource information available for real-time updating of the WRKG. Guo et al. [33] presented a framework for automated process knowledge base construction on the basis of a knowledge graph (KG) in the machining domain for enhancing the intelligence of process knowledge base construction. Guo et al. [34] established a KG-based process reasoning framework by adopting knowledge graph (KG) into the automatic machining process decision system, and constructed a mixed reasoning algorithm based on semantic analysis and attribute weighting to solve the problem of heterogeneity of process knowledge in decision making. Li et al. [35] conducted an in-depth study of the knowledge graph matching problem for service discovery, and converted the service discovery mechanism into a knowledge graph knowledge query with the design of matching templates. In conjunction with the characteristics of technical documents, Zhao et al. [36] put forward a TextCNN-based topic information mining model and structured the domain knowledge graph of technical documents.

Taking the above into account, it can be recognized that the advantages and applications of the knowledge graph are embodied in the product developing phase, designing phase, process designing phase, manufacturing phase and other phases. Regarding the product assembly stage, it involves the effective utilization of assembly resources, which embrace a variety of attribute information (such as structure information, process information, management information, etc.), and the selection of assembly resources is associated with a variety of information, such as the structure to be assembled, the planning of assembly process, and the management of assembly resources. Therefore, in combination with the strengths of knowledge graphs in knowledge representation, reasoning and retrieval, knowledge graphs open up a promising perspective for the allocation of assembly resources during the product assembly stage.

3. Overview of the Proposed Approach

Assembly resources owned by manufacturing enterprises serve the assembly process of products determined by product structure, which is closely related to the functional indexes of products. Starting from product functions, this paper determines the assembly resources needed in the assembly process of products through the mapping relationship among function, structure, procedure, and resource, as shown in Figure 1. Based on the refined sub-functional knowledge extracted from product functional knowledge and the granularity level of functional knowledge, the important product function classes in the ontology-based assembly resource upper pattern framework are established, which provide a common understanding for the retrieval and reuse of assembly resource knowledge in the assembly process. Then, according to the different granularity levels of functions, the relationships between product function classes and product structure classes are established, and the transition from product concept planning to concrete design is realized.

According to the product structure at different hierarchies, the assembly process of the corresponding product structure is formed by adding assembly process knowledge such as assembly sequence and subordination between components, so as to obtain the whole assembly process entity domain of the product, which serves as a bridge for the mapping relationship between product structure and assembly resources. Based on the product assembly process entity domain, the assembly procedure entity construction is completed. As the retrieval index of assembly resource knowledge, assembly procedure entities match corresponding assembly resource entities by combining structural parameters, assembly resource capabilities, materials and other attributes, and the mapping relationship between assembly procedures and assembly resources is established. The model also allows resources in the assembly resource layer to be clustered into different clusters, thus generating features of different resource clusters in the semantic space. Clustering features can be used as clustering indicators in the process of resource matching.

4. Ontology-Based Knowledge Graph

The knowledge graph is a mechanism to bond discrete entities captured across manufacturing processes with specific semantic associations and is expected to function as a vital contributor to cognitive intelligence within a manufacturing setting. The multiple-hop semantic links and intricate networks generated by knowledge graphs are the foundations of cognitive reasoning.

4.1. Assembly Resource Ontology Construction

The assembly resource ontology is the foundation for constructing the assembly resource knowledge graph (ARKG). This ontology defines the concepts related to the assembly resource domain and erects a complete and accurate concept platform. It is essentially a top-level semantic-type framework that reasonably categorizes the concepts and knowledge resources of the assembly resource domain and delineates the semantic relatedness among the domain concepts. Each concept of the assembly resource domain is a structured collection with a large number of instances, as shown in Figure 2.

Assembly resources are tools and devices that provide functions such as conveying, positioning, clamping, connecting, adjusting and inspecting in the product assembly process. Assembly resources generally encompass three aspects of information, which are resource structure (Str), resource process (Pro) and resource management (Man). As shown in Figure 3, assuming that the assembly resource ontology model is AROnto, then AROnto is represented as

$$AROnto=\left\{Str,Pro,Man\right\}$$

• Resource structure (Str)

$$Str=\{Topol,Mat,Area,Dimen\}$$

where Str is primarily intended to provide a depiction of the basic attributes of the assembly resource model, encompassing six main parameters, such as structural topology information (Topol), material information (Mat), factory area information (Area), and working part dimension information (Dimen).

2.

Resource process (Pro)

$$Pro=\left\{App,Rang,Clamp,Datum,Accuracy\right\}$$

where Pro is primarily intended to provide a depiction of the process properties of assembly resource model, encompassing five main parameters, namely, functional application (App), range of application (Rang), positioning clamping mode (Clamp), positioning datum information (Datum), assembly accuracy information (Accuracy).

3.

Resource management (Man)

$$Man=\{Categ,Series,Instru,Status,Workshop\}$$

where Man is primarily intended to provide a depiction of the management information of assembly resource model, comprising five main parameters, e.g., resource category information (Categ), series information (Series), usage instructions information (Instru), current status information (Status), belong to the workshop (Workshop).

4.2. The Structure of ARKG

Due to the fact that the target of the ARKG is to aid decision makers in discovering the assembly resources required for the product assembly process, a goal which has to involve knowledge information regarding the assembly resources themselves, we define an assembly resource dimension in the ARKG. On top of that, the assembly resource knowledge must be associated with the specific assembly procedure served and the product structure to be assembled by assembly procedure, while each individual product structure has a mapping relationship with a certain sub-function to be implemented by the product, which facilitates the location of the assembly resource knowledge. Consequently, the definition of ARKG comprises four dimensions, i.e., function, structure, assembly procedure, and assembly resource. It is illustrated in Figure 4, where the ellipses indicate the entities in ARKG, the rectangles indicate the concepts in ARKG, and the arrows indicate the relationships. Within ARKG, the assembly resource dimension comprises relevant concepts representing assembly resource knowledge, such as assembly resource capacities, dimensional structures, materials, and other multi-domain elements of knowledge. The assembly procedure dimension contains the basic assembly operations and typical assembly procedure concepts composing the assembly procedure.

The construction of the assembly resource dimension is planned using the assembly resource ontology pattern layer in this theoretical research work, which covers resource structure information, resource process information, and resource management information, and that is the viewpoint from which decision makers can pursue assembly resource knowledge. The conception of resource structure information is further redefined with concepts such as structural topology information, structural material information, and structural configuration information.

Assembly procedure dimension is the arrangement of concepts in the context of specific typical assembly procedures and the availability of defined terminology to describe the assembly procedure dimension. For the industrial enterprises, the role of assembly decision makers is generally directed at identifying assembly resource parameters for the assembly process, which involves finding sources of assembly resource knowledge and determining the criteria for their selection. As such, we offer five concepts in ARKG’s assembly procedure dimension: conveying, positioning, connecting, adjusting, and inspecting. The concepts of scribing procedure, hole making procedure, screwing procedure, and riveting procedure further define the concept of joining procedure. Similarly, the concepts of adjusting and inspecting can be further defined in a specific way. The alignment process, final finishing process, and painting process are used to specifically identify the adjusting process; the horizontal measuring, static balance detecting, dynamic balance detecting, and rain testing are applied to concretely describe the inspecting concept.

Aircraft need to have the function of storing fuel during flight, and the fuel tank is the structural carrier developedto realize the function of storing fuel. In the process of fuel tank assembly, the aircraft fuel tank sealing process is used as an example to illustrate the basis for the selection of assembly resources in each assembly process. The fuel tank sealing process is divided into three assembly procedures: conveying, screwing and sealing, followed by a description of the assembly resources used for each of the three assembly procedures. The selection of assembly resources for the transport category is based on the ability of the assembly resources to carry out transport tasks within the plant and the workshop. In determining the specific conveyor series, the selection is based on the dimensions of the transport target (length/width/height), the rated load capacity of the conveyor resource, the lifting height, the lifting work area, the type of power, the workshop to which it belongs and other data. Connection class resources connect parts together to serve as a whole. When selecting connection class assembly resources, the connection part resources are selected based on material, diameter, length, maximum permissible stress, and material yield limit. The selection of resources for screwed joints is based on outline dimensions, material, air pressure of the air source, no-load speed, machine weight, type and working principle. The selection of sealing tooling resources is based on type, room temperature, relative humidity, degree of purification, process performance, curing temperature and pressure. Therefore, with the combination of the structure information, process information and management information regarding the assembly resources, the assembly resources to be utilized in each assembly procedure are reasonably determined in the process of sealing the fuel tank, such as the use of simple trolley for fuel tank transportation, the use of titanium alloy high locking bolts, hexagonal bolts and tools in the screwing procedure, and finally the use of epoxy adhesive and glue injection gun in the sealing procedure.

4.3. Knowledge Graph Generation

Given that a knowledge graph is a collection of semantically associated concepts, the most fundamental step in constructing a knowledge graph involves yielding a glossary of triples. A triplet denotes a connection between a pair of concepts, as in the example of triplet {individual, relation, individual}. In this case, individuals and relations are defined in the ARKG. To accomplish this task, a widely available graph database, Neo4J, is applied to construct and archive ARKGs. Neo4J furnishes a high-grade query language that empowers us to manipulate the knowledge graphs stored in Neo4J. The knowledge graph is constructed in two steps: node generation and edge generation.

A.

Node generation

Nodes denote concepts in the ARKG and hold information about multiple attributes of the assembly resources. The nodes of ARKG are constructed under the specification of the top-level framework of the assembly resource ontology, and are related to the individuals in the triplet with corresponding one-to-one association. The first step is to screen the individuals and eliminate the repetitive individuals to generate a list of individuals. The below code is executed to establish a new node for each individual in the list.

Pseudocode: Node Generation

CreateNode Input: IndividualList (the list consists of the total amount of concepts involved in constructing ARKG), Relationship foreach individual in IndividualList do create Node end foreach relationship in Relationship do create NodeID-relationship-NodeID end

B.

Edge generation

Edges denote the association relationships available among all nodes in the ARKG, which are established by the code of constructing edges (NodeID -relationship -NodeID). In combination with the construction of nodes and edges, an ARKG is generated in which all the deciding knowledge is attached in order to make further exploration.

5. Case Study

The effectiveness of the proposed approach is proven with accurate and quickly accessed information on different types of assembly resources during the assembly process of an aircraft. The product manufacturing process in a broad sense is a giant network that extends from the product concept to the final product. During the process of product flows across this huge network of manufacturing, product manufacturing participants increase certain quantities of value, each of which poses a node within the production manufacturing network. It is therefore essential to empower the different participants involved within the network for sharing and reusing knowledge and providing better decision making.

5.1. The Process of Building ARKG

The functional domain of aircraft is obtained based on the decomposition of the total functions of aircraft layer by layer, which contains six level functions. The process follows the principle that each functional module is dependent on a structural module to carry out the mapping between function and structure, and gain the corresponding six level structural decomposition, as shown in Figure 5. For instance, the total product function as a primary function corresponds to the primary structure aircraft. One of the secondary function modules is the manoeuvring flight module corresponding to the tail of the secondary structure, which can be functionally decomposed to obtain the two corresponding tertiary function modules of lifting and steering relating to the horizontal and vertical tail of the tertiary structure. The tertiary functional module can be further decomposed into fourth level functional modules lifting operation, steering operation and connecting fuselage, with corresponding fourth level structures elevation rudder, steering rudder, horizontal stability surface and vertical stability surface.

On the basis of product structure decomposition, the primary assembly principle is parts → segments → components → whole machine. According to the primary principle, the assembly sequence of parts is arranged and the process domain is constructed under the principles of precision level of parts, material characteristics of parts, physical properties of parts and the structure of the previous sequence not affecting the connection path of the later sequence, as shown in Figure 6. With the five basic assembly operations of conveying, positioning, connecting, adjusting and inspecting, each assembly process is decomposed into assembly procedures related to these five basic assembly operations, and each of these assembly procedures is associated with specific assembly resources considering the relevant geometric information and PMI information of the structure. For example, Section 4.2 elaborates on the three assembly procedures of delivery, screwing, and sealing, which are broken down by the tank seal assembly process, and the basis for selecting the assembly resources to be associated with each assembly procedure.

In accordance with Section 4, a gradual implementation process is taken up to establish the ARKG. All edges and nodes are then introduced into Neo4J. As depicted in Figure 7, the ARKG includes blue nodes grouped by product function, pink nodes indicating the product structure, green nodes representing the assembly processes to be executed to assemble into the product structure, and the yellow nodes containing the decomposition of the assembly process into assembly procedures, and the red nodes of the assembly resources required to execute the assembly procedures. Figure 8 is a partially enlarged image of the knowledge graph concerning the tank sealing process in ARKG.

5.2. Knowledge Visualization

As shown in Figure 9, the knowledge graph illustrates a graphical representation of the linkages between domain concepts and delivers concise contextual information about the concepts. An interactively navigated field concept allows the user to highlight one of the concepts to explore further information or construct queries. The knowledge graph enables connections to be constructed between “knowledge isolates” and strengthens the relevance of knowledge in the assembly resource domain. By enabling users to navigate through assembly resource capability knowledge at a conceptual level and detect underlying cross-concept links, the sophistication of assembly resource knowledge queries during the assembly process can be better controlled. Figure 9 shows the visualization of a titanium alloy high locking bolt, which enables the designers to quickly obtain knowledgeable information about the part, as well as to obtain the assembly procedure, assembly process, structure and function associated with this assembly resource.

The knowledge graph dramatically boosts the performance of a knowledge searching system, making it more comprehensive, accurate, and intelligent. The knowledge graph enables the transition of knowledge retrieval models from text-centric to thing-centric. It is the knowledge graph which enables the system to distinguish relevant entities from user input and deliver knowledge about the entity, such as names, types, introductions, descriptions, related documents and semantic relationships. The knowledge graph is capable of capturing the user’s search intent and generating a clear and precise answer directly instead of a mass of search results.

6. Conclusions

It is crucial to provide assembly process designers with the ability to rapidly and accurately explore associated information in order to support their making decisions in the assembly process, particularly in the context of an environment with a large volume of unstructured textual documents faced by assembly process designers. Throughout this paper, we raise knowledge graph for representing the association between the structure to be assembled and the required assembly resources during the assembly process and for promoting fast browsing of the objective data to assist in decision making. The main contributions of this study are as follows.

• The ontology model of assembly resources is effectively established as the upper-level framework of ARKG. The attribute information of the model involves multiple domain knowledge, such as functional domain, semantic management, conformational design, features, performance evaluation, etc. Therefore, the upper-level schema framework of the assembly resource is constructed based on a comprehensive consideration of various input constraints (e.g., non-geometric structural attributes and assembly process factors) to accurately describe the semantic information of the assembly resource and accurately represent the input constraints in the product assembly process. Through ontology mapping-based data transformation, the instantiated CAD model data of the assembly resource is used as the structured data of ARKG in order to realize the effective integration of assembly resources for avoiding large amount of data redundancy.
• A multi-level function-structure mapping relationship of product-component-assembly-part and structure-assembly process-assembly procedure mapping relationship are introduced to construct an assembly resource knowledge map ARKG with a multi-dimensional integrated structure of function-structure-process-procedure-resource, which is utilized as a bridge between the product CAD model and the assembly process. The concepts and relationships of each dimension are further defined to integrate high-level semantic concepts of the CAD system and product-specific CAD model data into the assembly resource knowledge graph using affiliation relationships between classes and instances. ARKG delivers a common knowledge template with features such as framework, stability, reconfigurability and openness, which enables decision-related knowledge to be shared and reused.

The objective of our forthcoming research is to extend the existing assembly resource knowledge graph to combine tangible assembly resources (e.g., temperature, plant), intangible assembly resources (e.g., human resources) based on a larger scope, and to extend it to a wider range of assembly process design aspects, such as intelligent planning of product assembly sequences combined with assembly resources, and intelligent path planning of assembly resources based on the BERT model. Meanwhile, we use deep learning algorithms to achieve adaptive query answering in the context of knowledge graph-based retrieval queries.