Ontologies in Digital Twins: A Systematic Literature Review

1. Introduction

• Digital Twins have a real-time connection between physical and virtual systems, allowing the digital model to update the physical environment on the fly.The paper notes that DTs are used across varied industries and that no single definition or engineering architecture is commonly accepted.Source: §1 · p. 1
• Ontologies provide machine-processable domain conceptualizations that support interoperability and automatic reasoning in DTs.Ontologically enriched Digital Twins are often called Cognitive Twins.Source: §1 · p. 1
• The review asks how ontologies are used in Digital Twins, where they benefit DT architecture, and what barriers and limitations remain.The authors note that established best practices and a deeper account of application trends and architectural relationships are missing.Source: §1 · p. 1
• A systematic literature review screened 460 papers and extracted 82 directly relevant studies for analysis against a reference DT architecture.The review also considers knowledge graph implementations and the domains where ontologies are used.Source: §1 · pp. 1–2
• The review examines ontologies across DT architectural layers and analyzes their inter-layer roles and application domains.The paper also identifies discussion points and possible future research directions.Source: §1 · pp. 1–2

2. Background Concepts

• The background section establishes concise context on Digital Twins and ontologies for the review that follows.Source: §2 · p. 2

2.1. Digital Twins

• Digital Twins link physical and virtual spaces bidirectionally, and their definitions have evolved across domains and applications.They can support monitoring, decision optimization, and predictive maintenance; the review also notes DTs for abstracted entities such as organizations.Source: §2.1 · p. 2
• Cognitive Twins incorporate Semantic Web technologies such as ontologies and knowledge graphs into Digital Twins.Digital Shadows lack communication from the digital environment back to the physical environment.Source: §2.1 · p. 2

2.2. Knowledge Representation and Ontology

• Ontologies are formal, machine-processable specifications of a domain, supporting semantic interoperability, data reuse, inference, and automatic reasoning.Languages such as RDF and OWL support these capabilities.Source: §2.2 · pp. 2–3
• Ontologies commonly operate behind the scenes to support machine-to-machine interaction in distributed environments, and can also support human-computer interaction.Source: §2.2 · p. 3

3. Related Work

• D’Amico et al. reviewed 59 maintenance-context Cognitive Twin articles, reporting explicit ontology use in 28 and top-level ontologies in five.Their review also examined knowledge representation, databases, and knowledge-graph data formats.Source: §3 · p. 3
• Correia et al. treated ontologies as a means of semantic interoperability and data integration in Digital Twins.Their review covered data, semantic, and communication-level interoperability and found Industry 4.0, Smart Cities, and Healthcare among common DT domains.Source: §3 · p. 3
• A construction-focused SLR found ontology prominent but reported no ontologies developed for diverse, multi-context construction workflows.The present review also did not identify concrete ontology applications to address construction workflows.Source: §3 · p. 3
• An earlier review proposed a top-level-ontology-based DT conceptual model for maintenance, while this review also examines ontologies spanning multiple DT layers.The inter-layer analysis investigates their role as semantic interfaces between layers.Source: §3 · p. 3
• The authors report no prior SLR that exhaustively analyzes ontology adoption in Digital Twins, and distinguish their review by focusing on ontology use in existing DT solutions.They identify ontology roles across layers using a reference DT architecture.Source: §3 · p. 3

4. Methodology and Approach

• The review follows Kitchenham et al.’s SLR guideline, and its intermediary and end results are publicly available.Source: §4 · p. 4

4.1. Identification and Initial Selection of Research

• The search used five research databases and three search tools, with queries combining “digital twin” and “ontology” collected in March/April 2023.The initial 833 retrieved papers were deduplicated to 460.Source: §4.1 · p. 4
• After two relevance-screening rounds, 82 English-language research articles explicitly proposing ontology use in a DT were included.The review excluded surveys and papers that discussed the topic without actually using or proposing ontology use in DTs.Source: §4.1 · p. 4

4.2. Overview of the selected research and analysis

• Relevant publications increased almost 3 times over the period, a trend also seen for semantic technologies and knowledge graphs.Figure 3 presents the number of relevant publications per year.Source: Figure 3 · p. 4, §4.2 · p. 4
• The analysis maps ontology patterns to logical layers in a reference architecture and identifies domains for a domain-based analysis.The authors also analyze inter-layer roles and knowledge graphs used alongside ontologies.Source: §4.2 · pp. 4–5

5. Reference Architecture

• Existing DT architectures vary in their layer structures, and the field has not yet converged on a commonly accepted reference architecture.The review notes that common architectural patterns are emerging, while convergence remains incomplete.Source: §5 · p. 5
• The organizational context is treated generically, covering business and organizational aspects as well as system engineering at different levels.The architecture's virtual layer represents or specifies context-specific aspects.Source: §5 · p. 5
• The reference architecture distinguishes Physical, Communication, Digital, and Application layers, from physical elements and interactions to application-specific aspects.The review uses this architecture to structure its analysis.Source: §5 · p. 5
• Knowledge representation is modeled as local to one layer, inter-layer between contiguous layers, or multi-layer across non-contiguous layers.The review also notes that data is often addressed across architectures without a specific data view.Source: §5 · p. 5

6. Ontologies in Digital Twins

• The review organizes its results around ontology value, structural and inter-layer analyses, domain analysis, and knowledge graphs.It identifies four interrelated objectives that generally affect a DT system as a whole.Source: §6 · p. 5, §6.1 · p. 6

6.1. Objectives

• Ontologies model DT systems and integrate heterogeneous component data; comprehensive domain models can also guide development.The reviewed examples include a workshop model with ResourceInformation, TaskInformation, and ProcessInformation classes.Source: §6.1 · p. 6
• Ontologies support semantic interoperability by clarifying data meaning across DT components or cooperating twins; domain ontologies suit shared contexts, while top-level ontologies can span domains.The review describes shared learned parameters as one possible reason for DTs to cooperate.Source: §6.1 · p. 6
• Ontologies and sensor-supported knowledge graphs can help extract implicit semantic relations from instances and their changing states.Rule-extraction algorithms can be adapted to work with knowledge graphs and sensor data.Source: §6.1 · p. 6
• Ontologies facilitate automated reasoning by providing semantic information and data from diverse sources; reasoning outputs are propagated to relevant system components.One reviewed example uses reasoning to plan industrial-robot actions toward a system goal.Source: §6.1 · p. 6

6.2. Structural analysis

• 63 of 82 reviewed publications use ontologies for physical-layer concepts, compared with 49 for DT, 15 each for organization and application, and 5 for communication.The layer counts describe publications, not mutually exclusive assignments.Source: §6.2 · p. 6
• 30 of 82 ontologies cover one layer, 35 cover two, 16 cover three, and 1 covers four; Figure 5 summarizes the layer-count distribution.Of the single-layer ontologies, 23 are matched to physical layers; the review characterizes the remaining 52 as more task- or application-oriented.Source: §6.2 · p. 6, Figure 5 · p. 6

6.2.1. Physical layer

• Physical-layer ontologies represent components, attributes, states, and relations, or model actions and processes such as robot movements and tactile events.Examples include plant parts and CNC machine concepts; one CNC ontology aggregates data from diverse sources.Source: §6.2.1 · p. 7

6.2.2. Communication layer

• Communication-layer ontologies represent protocols between far-edge, edge, and centralized units, or between parts of a machining system or production line.Examples include using SOSA's Procedure subclass to specify protocol attributes and a NetworkProtocol class in home automation.Source: §6.2.2 · p. 7

6.2.3. Digital layer (DT)

• Digital-layer ontologies include domain-independent DT concepts, digital assets and operations, and abstract concepts represented through digital data.Examples include a generic DT ontology, construction data resources, and a power-plant fault model in which sensor-stream rules detect root causes.Source: §6.2.3 · p. 7

6.2.4. Application layer

• Application-layer ontologies range from task- and domain-specific requirements to generic manufacturing concepts used to match resource capabilities with processes.Examples address aircraft assembly, construction-project modules, and manufacturing resources and processes.Source: §6.2.4 · p. 7

6.2.5. Organizational context

• Organizational-context ontologies represent production and planning concepts, support construction-project resource allocation, or connect DTs with non-DT systems such as CMMS and ERP.The reviewed examples include manufacturing production planning and demand forecasting, and semantic interoperability with organizational software.Source: §6.2.5 · pp. 7–8

6.3. Inter-layer analysis

• 52 of 82 papers use ontologies spanning multiple layers, and 31 ontologies include both physical and digital concepts.Some ontologies span three or more layers.Source: §6.3 · p. 8
• No ontology covers both communication and organization concepts.The review therefore reports no connection between these layers in the analyzed ontologies.Source: §6.3 · p. 8
• The physical layer has the most connections to other layers, while the communication layer has the fewest.Application and communication layers connect most to the physical layer; organization connects most to the digital layer.Source: §6.3 · p. 8

6.4. Domains

• Figure 7 aggregates the reviewed papers by application domain.Generic groups papers on digital twins, IoT, smart home, materials science, IT, smart city, and IT security; Agriculture includes smart farming and fisheries, while Infrastructure includes building management, construction, public infrastructure, and cultural heritage.Source: §6.4 · p. 8, Figure 7 · p. 8
• Ontologies in DTs are used most in Manufacturing, followed by Generic and Infrastructure; Governance, Medicine, and Business each have one paper.The review notes that this pattern is similar to findings from other DT literature reviews.Source: §6.4 · p. 8

6.5. Knowledge Graphs

• 17 reviewed articles build knowledge graphs using ontologies.The review notes that ontology construction is one of the steps identified for constructing a knowledge graph.Source: §6.5 · p. 8
• One use of knowledge graphs is querying node and edge properties with ontological terms to extract information about DT components.Banerjee et al. use a Path Ranking algorithm to extract semantic relations not explicit in the graph.Source: §6.5 · p. 8
• Knowledge graphs can store a DT component’s current state, including state derived from fused sensor data.In a smart farming use case, Chukkapalli et al. use the graph to detect anomalies in sensor data.Source: §6.5 · p. 9
• Ontology-based knowledge graphs are also used to integrate heterogeneous data from multiple sources.Proper et al. describe a GITM ontology and a graph-based approach to unify data from IT governance processes, IT management processes, and organizational IT assets.Source: §6.5 · p. 9

7.1. Ontologies in Different Layers of a DT

• Ontologies often span multiple DT layers and can act as semantic interfaces between them.The review suggests that physical-layer concepts are more domain-specific, which may help explain why 31 ontologies cover both physical and digital concepts.Source: §7.1 · p. 9
• For DTs covering heterogeneous entities, a domain-specific ontology may unify data within one domain, while matching domain ontologies to a top-level ontology is a popular option across domains.The discussion distinguishes multiple entities in the same context from entities in different domains.Source: §7.1 · p. 9

7.2. Application Domain

• Ontology use in DTs is most prevalent in Manufacturing, and the review finds domain distribution proportionate to DT publication volume across domains.The comparison draws on a Scopus query and recent DT reviews.Source: §7.2 · p. 9

7.3. Knowledge Graphs

• 17 of 82 reviewed papers use knowledge graphs, mainly as stores for DT metadata or current component state rather than as part of reasoning.The review expects greater involvement of knowledge graphs in future DT reasoning algorithms.Source: §7.3 · p. 9

7.4. Multiple ontologies used in a single layer

• Multiple ontologies in one layer can be integrated by matching or creating common terms and using namespaces, or by using a comprehensive top-level ontology.The discussion also notes technical storage options including RDF stores and property graphs such as Neo4j.Source: §7.4 · p. 9

7.5. Distributed Digital Twins

• Semantic integration of ontologies and knowledge graphs across co-existing distributed DTs remains unstudied in this review and is proposed as a research direction.The discussion raises the question of using a single or multiple knowledge graphs for the same or different types of physical twins.Source: §7.5 · p. 10

7.6. Knowledge engineering and ontology re-use

• 64 of 82 reviewed articles propose new ontologies, but only 19 of those reuse existing concepts or align with top-level ontologies; more than half do not report openly sharing a source file.The review suggests alignment with top-level ontologies and open sharing as ways to alleviate reuse problems.Source: §7.6 · p. 10

• Ontologies are mostly used to represent concepts in the physical layer, which the review interprets as more closely tied to the respective domain.Source: §8 · p. 10
• 30 of 82 reviewed articles use domain-specific ontologies, while 52 use application/task-oriented ontologies whose concepts span multiple layers.Source: §8 · p. 10
• Manufacturing and Infrastructure lead DT and Cognitive Twin implementations, which are often limited to those domains.Source: §8 · p. 10
• More than half of reviewed articles neither reused an existing ontology nor matched their proposed ontology to a top-level ontology.Source: §8 · p. 10
• Knowledge graphs are increasingly popular for semantic-relation expressiveness and fast queries; combined with aggregated sensor data, they can serve as state graphs that reduce processing time and facilitate reasoning.Source: §8 · p. 10
• The review calls for closer study of ontology integration into DT knowledge bases and broader Cognitive Twin adoption beyond Manufacturing and Infrastructure.Source: §8 · p. 10
• Knowledge graphs may also guide reasoning algorithms, including decisions about where reasoning should occur in a system.Source: §8 · p. 10