A Novel Kind of Knowledge Graph Construction Method for Intelligent Machine as a Service Modeling

Abstract

With the development of Intelligent Machine as a Service (IMaaS), devices increasingly require personalization, intelligence, and service orientation, making resource modeling a key challenge. Knowledge graph (KG) technology, known for unifying heterogeneous data, has become an essential tool for modeling and analyzing manufacturing resources. On this basis, this study proposes a novel resource KG construction method for IMaaS. First, an E-R diagram is used to divide the constant and variable entities and set the constant attributes and the constant relationships. Then, the triplets are named, the value space is set, and the schema layer is constructed. Finally, the related information about devices is used to fill the data layer, and then, the knowledge graph is generated. Meanwhile, this study utilizes desktop FDM 3D printing devices as a case example for validation. The method proposed in this study can enhance the accuracy and maintainability of equipment resource management in the manufacturing sector, effectively promoting subsequent activities such as management, analysis, and decision-making.

1. Introduction

Intelligent Machine as a Service (IMaaS) is a model that operates within SocialM, abstracting intelligent equipment into an intelligent service system. By integrating intelligent equipment with the community-based manufacturing model, IMaaS treats intelligent equipment as production nodes at the device level, providing more intelligent, efficient, and personalized manufacturing services.

With the proposal of Intelligent Machine as a Service (IMaaS), the need for a device is shifted to the service provided and value creation from the device itself. The need for human–machine interaction and the cost-effectiveness of devices is higher in factories, and factories also have higher demands for the fine-tuning, maintenance, and management of knowledge graphs [1]. In the IMaaS architecture, intelligent machines are the most critical production resources within the socialized manufacturing resources. Resource modeling of these machines is a crucial component.

In device resource modeling, data typically comprise both structured and unstructured types. Structured data, such as data found in databases, often lack contextual readability, whereas unstructured data, such as documents, frequently lack inter-data relationships. For Intelligent Machine as a Service (IMaaS), which encompasses a wide range of AI scenarios and applications, it is crucial to organize these data into a format that can be interpreted by AI models. Knowledge graphs provide enhanced flexibility in knowledge utilization through their use of nodes and edges [2], making them an ideal choice for the resource modeling of IMaaS.

In recent years, knowledge graphs (KGs) have played a significant role in advancing productivity in the manufacturing industry. And KGs are also regarded as a core element of next-generation industrial systems [3]. Regarding knowledge management in manufacturing, KGs can save human resources and time and provide domain knowledge retrieval for requirement analysis, solution design, and operation and maintenance management [4,5]. In terms of analysis and reasoning, based on multidisciplinary knowledge and relevant algorithms, KGs can further analyze and support decision-making [6,7]. KGs provide significant support for industrial products and services, including design, manufacturing, logistics, maintenance, recycling, and other complete lifecycle activities.

However, the primary task of using KGs to drive IMaaS requires an efficient and accurate method for the resource modeling of industrial devices. At present, there is still little research on IMaaS. For example, Ren et al. [8] proposed a hybrid knowledge graph approach to abstract devices as smart services. This method maps the process and related resources into a twin space organized with a KG, translating the physical running of IMaaS into KG interaction. Currently, the method of extracting the schema layer and data layer from data by computer-related methods lacks structure division and hierarchy combing, including manufacturing domain knowledge [9,10]. Based on this, this study proposed a novel resource KG construction method for IMaaS.

The paper is organized into five parts: Section 2: Related Work, including resource modeling methods for IMaaS and methods for the construction of IMaaS knowledge graphs; Section 3: Method, including modeling based on an E-R diagram, the construction of a schema layer, and the generation and interconnection of a data layer; Section 4: Case Validation, applying the method to desktop FDM-3D devices as an example to illustrate; Section 5: Discussion, and Section 6: Conclusion.

2. Related Work

2.1. Resource Modeling Methods for IMaaS

In the field of IMaaS, the main resource modeling methods include cases, ontology, knowledge graphs, etc. Currently, much research focuses on resource modeling using ontologies and knowledge graphs.

In the area of knowledge modeling for machining processes, He et al. [11] proposed an ontology-based rapid remanufacturing process planning (RPP) knowledge modeling method. They used the case-based reasoning (CBR) method to quickly generate the model. Eum et al. [12] proposed a modeling method for constructing process planning knowledge for machining operations and process planning of multi-axis machining features. Ye et al. [13] proposed a method to develop a CNC machining process knowledge base based on cloud technology. They used the general standard STEP-NC to map to the ontology web language (OWL) to collect descriptive and logical knowledge to realize the standardized description and large-scale processing of CNC machining process-related knowledge. There are also some other resource modeling methods. For example, Wu and Liu [14] proposed a multi-source information representation method based on a genetic algorithm, which realized the reuse of fine-grained information in remanufacturing assembly. Using the concepts of holon and attractor, Ostrosl et al. [15] integrated uncertainty into the modeling of a part design and part manufacturing network and proposed a method to solve the formation of intelligent virtual manufacturing cell in CBDM, which overcomes the discontinuous and continuous problem of traditional cell design formation. Yang et al. [16] established an expert system that is realized with Function–Behavior–Structure-oriented assembly structure characteristic analysis, ontology instance-based assembly structure characteristic modeling, and modified RETE algorithm-based design decision-making.

For the overall architecture for manufacturing resource modeling, Lin et al. [17] proposed a manufacturing architecture that integrates manufacturing system resources and applied SPARQL to query or reason machines that meet order requirements. Zhao et al. [18] proposed an implementation framework of networked collaborative processing equipment resource modeling based on graph database technology (Neo4j). The framework solved the difficulties of the geographical dispersion and functional heterogeneity of manufacturing resources in networked collaborative processing. Based on ontology web language, Yuan et al. [19] proposed a new manufacturing resource ontology model that encompasses a resource multi-layer model and includes physical, virtual, and cloud resource data layers, as well as a cloud end layer.

However, no well-established implementation method for physical equipment resource modeling exists. For example, Kjellberg et al. [20] proposed a machine tool modeling method based on ontology language according to general standards and discussed the mapping mechanism between the ontology model and existing standard concepts in the industry in detail. Based on the principle of model-based system engineering (MBSE), Li et al. [21] proposed a unified information model containing the whole process of production equipment, which solved the problems of many types of information, many forms of expression, and complex heterogeneous models in the field of production equipment. In addition, the model supports the integration of multidimensional models. Currently, physical resource modeling predominantly uses ontologies, but there are shortcomings in aspects such as hierarchical classification, maintainability, and ease of construction.

2.2. Knowledge Graph Modeling Methods for the Manufacturing Domain

In manufacturing, existing knowledge graphs can be broadly categorized into three types [22]: (1) traditional static knowledge graphs, such as a text-based 3D printing KG [23], a device KG from device manuals [24], a KG for equipment maintenance [25,26], and so on; (2) temporal knowledge graphs, such as a multi-layer manufacturing KG based on the digital thread of manufacturing process data [27], a temporal knowledge graph and its relative applications based on the IoT [28], and so on; and (3) multi-modal knowledge graphs, such as a multi-modal industrial knowledge graph [29,30]. In constructing knowledge graphs in these manufacturing domains, top–down or bottom–up approaches are commonly employed [31].

The bottom–up method first extracts many entities from unstructured or semi-structured data sources and then identifies the relationships between these entities to form a triple structure, and finally summarizes the schema of the knowledge graph based on the extracted entities and relationships. This approach is suitable for building general domain knowledge graphs, such as commonsense knowledge graphs, scenarios with large amounts of data, complex businesses, and difficulty pre-defining explicit data schema. Liu and Lu [32] proposed a new task-centric knowledge graph (TCKG) model centered on maintenance task components (MTCs) to extract equipment maintenance knowledge from unstructured PDF manuals to construct a KG and proved that it is superior to traditional entity-centric knowledge representation methods. Yan et al. [33] obtained a large amount of heterogeneous data containing massive equipment information from Internet-extracted entities using the conditional random field algorithm after preprocessing. Then, they adopted an analytical method to identify the relationship between equipment entities to build a KG-based manufacturing equipment information query system. Based on the characteristics of power equipment defect records, Wang and Liu [34] combined the entities, attributes, and relationships extracted from unstructured data with the triplets contained in the specification. They proposed a method to construct a knowledge graph of power equipment defects. Haruna et al. [35] proposed the Bidirectional Encoder Representations from the Transformers (BERT) model to build an additive manufacturing knowledge graph.

The top–down approach, which enjoys higher popularity in manufacturing, first defines the top-level concepts of the knowledge graph, then models and describes the domain problems, and finally obtains the data schema based on the problem model and adds entities to the corresponding concepts. The domain problems can be modeled and described in four ways: framework, ontology, relation model, and “flow-node” model. This method is suitable for building vertical field knowledge graphs, such as financial, medical, and other industry knowledge graphs and scenarios with relatively small data volumes, transparent businesses, and the ability to pre-define explicit data schemas. He et al. [36] proposed a machine learning (ML)-assisted knowledge graph (ML-KG) design to achieve the spatiotemporal fusion optimization of resource allocation by combining spatiotemporal data after industrial logistics resource mapping with ontology-based IL resource knowledge. Meng et al. [37] identified and extracted power equipment entities from preprocessed Chinese technical literature and then combined the relational classification method based on dependency parsing to extract the semantic relations between entities to form knowledge triplets. Zeng and Hou [38] proposed a hierarchical relationship construction method based on electromechanical equipment, which extracts various model information data from electromechanical equipment and structures the data hierarchically, thereby obtaining stable and hierarchical knowledge graph features of electromechanical equipment to complete the KG construction further. He and Jiang [39] constructed an ontology-based meta-knowledge graph (MKG), used a unified meta-knowledge filter to collect and integrate multi-factor and multi-level meta-knowledge, and proposed a graph-oriented meta-knowledge model (MKM) to represent the relationships between knowledge entities.

In practical applications of the manufacturing industry, the two methods are often combined to construct a knowledge graph. Lou et al. [40] designed the knowledge ontology of maintenance manuals and analysis reports in the maintenance field by combining top–down and bottom–up methods. They used a joint extraction model based on BERT-Bi-LSTM-CRF to extract entities and relationships, eliminating knowledge redundancy and building high-quality KGs. Wang et al. [41] proposed a method to construct an equipment maintenance knowledge graph (IDS-KG) by considering the causal relationship between faults in the equipment maintenance corpus. They constructed the ontology model from top to bottom, constructed the knowledge graph from bottom to top, and finally constructed the IDS-KG.

In the process of modeling basic devices for IMaaS, a top–down approach to constructing knowledge graphs is currently the best choice. However, for designing the schema layer, there is still a lack of efficient, comprehensive, and maintainable modeling design methods. An ideal method should incorporate ontology-based modeling characteristics while also integrating with existing enterprise resource frameworks for rapid modeling, thereby meeting the efficiency requirements for IMaaS basic equipment modeling.

3. Method

In the context of IMaaS, device resource modeling for enterprises is one of the primary tasks. The entire process begins with enterprise device resource information and ultimately constructs a knowledge graph of device resources. The equipment information for enterprises can include both structured and unstructured data, such as basic information and operational data. These data involve various types of information and data centered around the equipment, including people, machines, methods, materials, environments, and tests. Before the modeling process, it is necessary to determine the hierarchy and granularity of the model, as this will determine the scale and scope of the final knowledge graph. This resource KG is designed to enable real-time maintenance and management in line with factory operations.

The main steps of the resource KG construction method for IMaaS now include three steps to accommodate both structured and unstructured data: (1) The acquisition or modeling of resource E-R diagrams for structured data and text mining or data extraction for unstructured data, (2) the construction of the schema layer, and (3) the generation and interconnection of the data layer. As shown in Figure 1, the process begins by inputting both structured and unstructured information about industrial devices. The triple principle serves as the control condition, and the E-R diagram (for structured data) and data extraction techniques (for unstructured data) act as enabling mechanisms.

Section 1 involves two parallel processes: acquiring the structured resource data E-R diagram from the enterprise or constructing it and utilizing natural language processing (NLP) techniques, computer vision, or sensor data processing to collect unstructured data from the manufacturing environment. For structured data, modeling the E-R diagram considers hierarchical and structural relationships of equipment data. For unstructured data (such as text logs, images, or audio recordings), data must be preprocessed and categorized based on context, relevance, and content type.

Once structured data are converted into a two-dimensional table and unstructured data are parsed and classified, both datasets are used as input for Section 2. In this section, value assignment rules and domain space conditions serve as control conditions. The E-R diagram acts as the enabling mechanism for structured data, while unstructured data are mapped to relevant entities and attributes within the KG. The process involves triplet naming, value domain setting, and schema layer construction for both types of data. Triplet naming for unstructured data may require entity recognition, relationship extraction, and sentiment analysis to ensure it accurately represents physical facts.

In Section 3, the schema layer constructed in Section 2 serves as the control condition for generating both structured and unstructured data layers. Information from industrial devices during operation—whether structured data from sensors or unstructured data from maintenance logs, reports, or images—is used as input. This section performs schema layer filling and data layer generation and outputs the combined knowledge graph. During the operation and service of the IMaaS equipment knowledge graph, the data layer is continually expanded by structured and unstructured data. The schema layer is refined based on the actual factory conditions and the evolving nature of the unstructured data collection methods.

3.1. E-R Diagrams for IMaaS Modeling

3.1.1. E-R Diagram Device Resource Modelling

As a conceptual model, the E-R model is usually used to describe the needs of the real world and the types of information that need to be stored. An E-R diagram, also known as an entity–relationship diagram, is a tool for describing an E-R model including entities, attributes, and connections, which can clearly and intuitively express demand information to the audience. Resource modeling for industrial devices using an E-R diagram is the basis for constructing the schema layer for the knowledge graph. In the resource modeling process for IMaaS, we need to extract and model the relevant device resource information within the factory. This information can include details about operators, machines, materials, methods, environments, and measurements, which exist in the form of structured or unstructured data.

During the modeling process, some enterprises already manage resource data for certain equipment, and these data are often stored in relational databases. These structured databases typically have existing E-R diagrams. If the target factory already has basic data management in place, we can leverage these existing E-R diagrams for modeling. At the same time, we need to ensure that the modeling process meets the enterprise’s data management requirements. Then, we can further refine or construct the E-R diagram model by integrating existing enterprise data management E-R diagrams. It is essential to establish reasonable criteria and rules for defining entities, relationships, and attributes. Based on this, we can extract entities, relationships, and attributes and model the E-R diagram. For example, in a 3D printing factory, various related information centered around the printer needs to be extracted, including data on machines, materials, and measurements. By analyzing the structural relationships, we can create a simplified E-R diagram representation.

As shown in Figure 2, according to the principles of the triplets, “entity–relationship–entity” and “entity–attribute–attribute value” are used to construct the E-R diagram. In the diagram, rectangles represent entities, diamonds represent relationships, and ellipses represent attributes. Relationships can also have attributes. These two types of triplets can satisfy the requirements for equipment resource modeling. For example, entity_1 has attribute_1, attribute_2, and ID_1, and there is relationship_2 between entity_1 and entity_4.

The information on industrial devices will be abstracted into the E-R diagram. For example, “Desktop-FDM-Printer” and “Z-axis controller” have a hierarchical relationship; “Desktop-FDM-Printer” and “Wire Reel” have a usage relationship, whereas “Desktop-FDM-Printer” uses “Wire Reel” and “Wire Reel” is used by “Desktop-FDM-Printer”; “Desktop-FDM-Printer” has attributes such as power and precision.

3.1.2. Mapping and Conversion of a Two-Dimensional Table

Currently, most manufacturing companies use relational databases, and a few use graph databases. Considering the storage method of the traditional manufacturing company, after the construction of an E-R diagram, E-R diagrams can be converted to two-dimensional tables.

Mapping serves to convert both entity types and relationship types into relational schemas. For the conversion of entity types, each entity type can be transformed into a relational schema, where the attributes of the entity become the attributes of the relational schema, and the entity identifier becomes the key of the relational schema. The conversion of relationship types primarily involves the transformation of unary, binary, and ternary relationships. The E-R diagram is more advantageous for the intuitive analysis and judgment of resource modeling, while the tabular format is more suitable for storage and filling the schema layer.

As shown in Figure 2, entity_1 and its attributes and its ID can be converted into a two-dimensional table named entity_1, entities with relationship_1 between them can be converted into a two-dimensional table named relationship_1, and entities with the same attribute can be converted into a two-dimensional table along with their attribute values. In that case, information on the knowledge graph can be stored in graph databases or relational databases, which is beneficial for the transition of the databases used by companies.

3.2. Construction of the Schema Layer Based on E-R Diagram

3.2.1. Triplet Naming and Value Space Setting

Once the E-R diagram is completed, the schema layer can be constructed based on the intuitive E-R diagram or the E-R diagram in the form of a two-dimensional table. As shown in Figure 3, when constructing the schema layer, we need to set the naming rules for the schema layer’s triplet and set the attributes’ value space.

The naming conventions must comprehensively consider the needs of enterprise data management, the rationality, and scalability of modeling, as well as the structural characteristics of the machines. The value domain settings should be based on the characteristics of relationships, the meanings of attributes, and other factors, ensuring they align with the physical meanings of the equipment data.

In this study, a triplet includes nodes and edges; nodes are divided into constant and variable entities, and we can set the constant entity types. For example, “device”, “sensor”, etc., are set to constant entities. The variable entities are named based on these constant entities. “device_id”, “sensor_id”, etc., are set to the variable entities. Meanwhile, the value space of these variable entities is set.

Edges are divided into constant relations and constant attributes. For example, “uses”, “includes”, etc., are set to the relationship. And “Attribute_1”, “Attribute_1”, etc., are set to the constant attributes. Meanwhile, the value space of these constant attributes is set.

3.2.2. The Method of Construction for the Schema Layer

After the triplet naming and value space setting, the schema layer can be constructed based on the triplets.

Based on the E-R diagram designed in Section 3.1, which provides the relationships between nodes and edges, and combined with the naming requirements and value domain space settings from Section 3.2.1, the schema layer is designed using the triplet forms of <entity − relation − entity> and <entity − attribute − attribute value>. According to the defined minimum granularity, the information related to entities, relations, attributes, and attribute values is extracted from the knowledge involved with the equipment in the factory. This extracted information is then populated into constant entities, variable entities, constant relations, or variable relations in the nodes and edges. Finally, the schema layer is transformed into n sets of schema layer structure languages.

As shown in Figure 3, according to the entities and the relationship between them, we can construct pieces of schema layer data: <Device, include, Device_id: Device_id_a>, etc. According to the attributes and their value spaces, we can construct pieces of schema layer data: <Device_id: Device_id_a, attribute: Attribute_1, Attribute_value: Value>, etc. A schema layer can be constructed based on the triplet model that includes nodes and edges.

3.3. Construction of the Data Layer Based on the Schema Layer

3.3.1. The Generation of the Data Layer

The data layer can be generated based on the schema layer constructed by the method described in Section 3.2.

Once the schema layer is constructed, the data layer can initially be populated with basic information about factory equipment, such as equipment details and consumables used. During the factory’s operational process, the data layer can be further enriched with operational data, including equipment changes and data generated during equipment operation. Additionally, the schema layer can be continuously refined and improved based on ongoing feedback throughout the data layer construction process.

As shown in Figure 4, according to a piece of schema layer data, <Device, include, Device_id: Device_id_a>, we can fill the information of industrial devices in it to form many pieces of data layer data, like <Device: FDM Device, includes, Device_id: FDM_Device_id_001>. The data in the data layer will be continuously filled with the factory running. Meanwhile, the schema layer will also be improved to meet device resource modeling requirements.

3.3.2. The Construction of a Knowledge Graph Based on the Data Layer

After obtaining the data layer, these data need to be interconnected. Based on the data layer, the data can be stored in a knowledge graph constructed by a graph database. As shown in Figure 4, we can use Cipher queries to store data in the Neo4j database. Using a graph database like neo4j can improve the query efficiency in subsequent management tasks.

In Neo4j, entity nodes and attribute value nodes are created, and edges are used to connect these nodes based on relationships and attributes. During the factory production process, the schema layer and data layer are stored separately in the enterprise database management system. As production data are generated, new data are continuously added to the data layer according to the schema layer settings, thus expanding and enriching the generated knowledge graph.

During the factory’s operation, the data layer of the knowledge graph will continuously be updated and expanded. Simultaneously, the schema layer of the knowledge will also be updated during the operation and management processes. The data generated during operation can guide the classification and hierarchical structuring of the schema layer, leading to updates in the underlying E-R diagram. By establishing a knowledge graph for the basic equipment, functionalities such as retrieval, inference, analysis, and decision-making can be achieved.

For example, using Cipher queries, we can quickly retrieve the information about the device: “FDM Device” includes a device “FDM_Device_id_001”, “FDM_Device_id_001” uses “PrintBed_Parts_id_B001_01”, and “FDM_Device_id_001” has an attribute “Max Print Precision”, the value of which is “0.1 mm”.

4. Case Validation

4.1. Case Background

In the context of IMaaS, constructing a reasonable resource model is crucial for efficiently managing resources and providing users with high-quality services. As a new manufacturing service model, 3D printing technology leverages Internet technology to integrate resources across different regions and periods, facilitating effective resource sharing. This technology exemplifies the manufacturing industry’s shift toward “decentralized resources and centralized services”. However, the scattered resources are categorized in various ways, widely distributed, and lack standardization, making model construction challenging.

This article uses FDM-3D printing devices as a case study, demonstrating the entire process and advantages of the knowledge graph (KG)-based model construction method. The first step of a construction KG is obtaining an E-R diagram from an enterprise or constructing an E-R diagram according to relative rules. This article constructs an E-R diagram according to the E-R diagram from an enterprise. As shown in Figure 5, this E-R diagram includes four types of elements: variable entity names, constant entity names, constant relation names, and constant attribute names. Variable entity names and constant entity names form the nodes in the E-R diagram, while constant relation names and constant attribute names form the edges.

4.2. Construction of the Schema Layer of Desktop FDM-3D Printing IMaaS

The construction of the schema layer needs to name triplets (node, edge, node) and set the range space. According to the E-R diagram form Section 3.2, the E-R diagram includes four types of elements: variable entity names, constant entity names, constant relation names, and constant attribute names.

For the structured data, as shown in Figure 6, the variable entity name includes the “desktop FDM device ID name”, “desktop FDM device brand name”, “desktop FDM device model name”, “extrusion head ID name”, “wire reel ID name”, “printing bed ID name”, and “hot compress paper ID name”. Each variable entity name has its corresponding range. The range of “desktop FDM device brand names” includes “Aurora Evo”, “Makebot”, etc. “Desktop FDM device models” include “Model s603” and “Model A8”. The values of the variable entity name range can be symbols, dimensionless values, dimensionless values, file names, and URLs. The different brands of desktop FDM-3D printer equipment models vary, such as “Aurora Evo brand models” S603, A6, A8, and so on; “Makebot brand models” include mb-01 and others.

The values of the variable entities in the schema layer are as follows: the device is a three-digit device serial number. For example, the range of the desktop FDM device ID name includes “desktop FDM device 001” and “desktop FDM device 002”. The equipment components include the hot extrusion head, printing bed, hot compress paper, and wire reel. The value rules are as follows: the hot extrusion head “H-” followed by the three-digit equipment serial number “-” the two-digit hot extrusion head serial number. For example, the value range of the hot extrusion head ID name includes the hot extrusion head “H001-01”, “H001-02”, etc. The printing bed takes “B-” followed by the three-digit device serial number “-” the two-digit printing bed serial number; for example, the value of the printing bed ID is “B001-01”, “B00-02”, etc.; hot compress paper takes “P-” followed by the three-digit device number “-” two-digit print bed number “-” three-digit hot compress paper number; for example, the value of the hot compress paper ID name includes hot compress paper “P00-01-001” and hot compress paper “P001-01-002”; the wire reel ID value is “M-” followed by the three-digit equipment serial number “-” the three-bit reel serial number; for example, the reel ID range includes reel “M001-001”, “M001-002”, and so on.

In the example of the desktop FDM device, the constant entity names include the “desktop FDM device”, “the hot extrusion head”, “the printer”, “the X-Y axis controller”, and “the Z-axis controller”.

Constant relation names are divided into “uses” and “includes”. Constant attribute names include “attributes-power”, “attributes-max-printing-width”, “attributes-max-printing-width”, “attributes-max-printing-height”, and “attributes-max-precision”. Each attribute value has its range. For example, the value range of “attributes-power attribute” is “200w”, “500w”, etc.; the value range of “attributes- max-print-length” is “150mm”, “300mm”, etc. The value range of “attributes-max-precision attribute” is “0.05mm”, “0.1mm”, etc. Note that the attribute value of an attribute is a node.

The E-R diagram is converted into a schema-layer triplet. As shown in Figure 7, we can obtain a lot of triplets, which are composed of <node − edge − node> form. For example, <desktop FDM devices − includes − Desktop FDM device ID: value>; <Desktop FDM device ID: range − uses − desktop FDM device brand: range>; <desktop FDM device brand: range − includes − desktop FDM device model: range>; <desktop FDM device model: range − attributes-power − Attribute value: range>, and so on.

For the unstructured data, Figure 8 is a piece of text describing how the 3D printer works and its type, which is unstructured data. In the previous example, we generated an E-R graph from structured data; in this example, we can also generate an E-R graph from unstructured data. The text information is processed by an AI large model, and useful information is extracted to form an E-R graph.

The E-R graph generated by processing the unstructured data in Figure 8 is shown in Figure 9. The working principle attribute of the 3D printer entity is that 3D printing works by blending layers of material to build an object. Furthermore, 3D printers include FDM-3D printers, SLS-3D printers, SLM-3D printers, SLA-3D printers, and DLP-3D printer. The FDM-3D printer uses ABS and PLA. The SLS-3D printer uses laser and powdered plastic material. The SLM-3D printer uses high-power density laser and metallic types of powder. The DLP-3D printer uses light through a light projector screen and resin materials.

4.3. Construction of the Data Layer and Generation of KG of Desktop FDM-3D Printing IMaaS

In this section, we will use the data-layer triplets of an Aurora Evo s603 desktop FDM device as an example to introduce the generation of data-layer triplets. As shown in Figure 10, the data layer of the generated triplet corresponds to the schema layer in Figure 7. For example, <desktop FDM device − includes − Desktop FDM device 001>; <desktop FDM equipment 001 − uses − brand Aurora Evo>; <brand Aurora Evo − includes- model s603>; <model s603 − attributes-power − 200w>, etc.

Figure 11 illustrates the construction of a knowledge graph for IMaaS using desktop FDM-3D printing equipment as an example. The top–down knowledge graph construction method is adopted to establish the schema layer based on the E-R diagram, facilitating the division of variable entities and constant entities and ultimately forming the knowledge graph of desktop FDM-3D printing devices by defining entity attributes and extracting entity relationships. This approach ensures the accuracy of the knowledge graph itself and enables dynamic updates, maintenance, and expansion in later stages.

This case study demonstrates the extensive application prospects of knowledge graphs in IMaaS. Enterprises can leverage their product service management platforms to construct knowledge graphs for their manufacturing resources, extract multi-modal data information from production sites, and integrate other computer technologies to manage the entire lifecycle of manufacturing equipment. Furthermore, through functions such as semantic retrieval and fault analysis, these knowledge graphs can assist manufacturers in achieving comprehensive intelligent management, including automated operation and maintenance, fault prediction, and supporting decision-making processes and resource allocation within the manufacturing industry.

5. Discussion

Traditional equipment resource management usually uses a data-based relational database and uses a two-dimensional table to organize data. The data organized in this way has a high degree of consistency and reliability. In the context of IMaaS, how to manage equipment information effectively and quickly is the key to realizing services. However, the data-based relational database pays more attention to queries, and it is difficult to realize reasoning, so as to achieve the purpose of using smart devices as services. Through the research of this paper, the rapid construction of knowledge graphs containing equipment resource information can be realized, and this research can help enterprises to realize reasoning and decision-making in combination with other computer technologies. Finally, the dynamic update and automatic operation and maintenance of equipment resource information are realized to provide users with more sufficient services.

This research adopts a top–down knowledge graph construction method, which mainly includes three steps: the construction of the E-R model, the construction of the pattern layer, and the generation of the data layer. Establishing a schema layer and generating a data layer based on the information of the E-R diagram is conducive to integrating the existing information resources of the enterprise for entity abstraction and relationship extraction, which not only ensures the efficiency and accuracy of data acquisition, but also reduces the difficulty of constructing the pattern layer and provides support for the generation of knowledge graphs. The knowledge graph generated according to this method can visually display the relevant information of the resource.

The deficiency of this paper is that the specific applicable data magnitude verification of IMaaS knowledge graphs has not been completed, which is related to the use boundary and efficiency of enterprises in practical applications. With the development of enterprises and the increase in resource types and quantities, the scale of knowledge graphs will become larger and larger. How to complete effective storage and improve retrieval efficiency will become the next development direction.

6. Conclusions

Most of the existing relational databases for managing intelligent devices can only realize query functions and are difficult to reason. Furthermore, most methods for constructing knowledge graphs for manufacturing equipment use ontology modeling and fail to fully utilize database information for the construction of the schema layer. The new construction method proposed in this paper uses E-R diagrams to obtain database information to construct the schema layer, which is both accurate and maintainable. In the future, the construction of knowledge graphs for intelligent devices can consider integrating AI to extract unstructured data and automatically connecting to databases to obtain valid information.

This study proposes a novel method for constructing KGs tailored for an Intelligent Machine as a Service model. The method employs a top–down approach, utilizing E-R diagrams to establish the schema layer. This involves classifying variables and constants into distinct categories, including constant entities, variable entities, constant relationships, and constant attributes. Additionally, triplets are defined, and a value space is established. The schema layer is then constructed based on the E-R diagram. Finally, the data layer is created using the schema layer, allowing for data processing that ultimately generates the knowledge graph.

This study provides an effective method for device management services in IMaaS, using desktop FDM-3D printing devices as an example. This method can improve the efficiency and accuracy of pattern layer construction by constructing the schema layer based on an E-R diagram. The hierarchical division of the schema layer, the naming of triplets, and the setting of range space improve the maintainability of the knowledge graph. This method ensures the accuracy of the knowledge graph while providing significant flexibility for updates and maintenance, facilitating the expansion of the knowledge graph and laying the foundation for subsequent operations and maintenance activities.

This study presents several promising applications, such as developing an automated construction and maintenance platform based on this method, enabling more efficient and convenient device management. In addition, integrating this method with existing traditional data management platforms requires incorporating other data conversion methods to interface with historical device data. This method could also provide services such as process management, temporal signal processing, and predictive maintenance for device activities.